diff --git a/.clang-format b/.clang-format
new file mode 100644
index 0000000..61e0467
--- /dev/null
+++ b/.clang-format
@@ -0,0 +1,14 @@
+---
+Language: Cpp
+BasedOnStyle: WebKit
+SpaceAfterTemplateKeyword: false
+AlignEscapedNewlines: Left
+AlignTrailingComments: true
+BreakBeforeInheritanceComma: true
+BreakConstructorInitializers: BeforeComma
+IndentPPDirectives: AfterHash
+BreakBeforeBraces: Custom
+BraceWrapping:
+    AfterFunction: true
+NamespaceIndentation: None
+QualifierAlignment: Right
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..95e5889
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,16 @@
+build/
+src/openlibm/*.o
+src/openlibm/*~
+src/openlibm/*.a
+src/openlibm/*.dll*
+src/openlibm/*.so*
+src/openlibm/*.dylib*
+src/openlibm/*.pc
+src/openlibm/openlibm-test
+src/openlibm/cov-html/
+src/openlibm/*.gcda
+src/openlibm/*.gcno
+src/openlibm/amd64/*.o
+src/openlibm/bsdsrc/*.o
+src/openlibm/ld80/*.o
+src/openlibm/src/*.o
diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 0000000..d24ed61
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,10 @@
+{
+    "jaktLanguageServer.compiler.executablePath": "/home/alec/cloned/jakt/build/bin/jakt",
+    "git.ignoreLimitWarning": true,
+    "pmd-cpd.language": [
+        "cpp"
+    ],
+    "pmd-cpd.onStartBehavior": "Show",
+    "pmd-cpd.sourceDirectory": "/home/alec/repos/erythros",
+    "terminal.integrated.shellIntegration.history": 0
+}
diff --git a/.vscode/tasks.json b/.vscode/tasks.json
new file mode 100644
index 0000000..7f1b6a9
--- /dev/null
+++ b/.vscode/tasks.json
@@ -0,0 +1 @@
+{"version":"2.0.0","tasks":[{"label":"Build All","type":"shell","command":"${workspaceFolder}/scripts/build-all ${workspaceFolder}","group":{"kind":"build","isDefault":true},"runOptions":{"instanceLimit":1,"instancePolicy":"terminateOldest"},"problemMatcher":[]}]}
\ No newline at end of file
diff --git a/NeinTerm.HC b/NeinTerm.HC
new file mode 100644
index 0000000..b37796f
--- /dev/null
+++ b/NeinTerm.HC
@@ -0,0 +1,254 @@
+#define @nt_process_debug Print
+
+Bool @nt_window_is_hovered(CTask* task)
+{
+    return (task == sys_focus_task && ms.pos.x >= task->pix_left && ms.pos.x <= task->pix_right && ms.pos.y >= task->pix_top && ms.pos.y <= task->pix_bottom);
+}
+
+U0 @nt_draw_it(CTask* task, CDC* dc)
+{
+    GrBlot(dc, 0, 0, task->user_data(@nt*)->screen);
+    @nt* nt = task->user_data;
+    if (@nt_window_is_hovered(task)) {
+        gr.fp_draw_ms = NULL;
+        GrBlot(dc, ms.pos.x - task->pix_left, ms.pos.y - task->pix_top, nt->cursor);
+    } else {
+        gr.fp_draw_ms = &DrawStdMs;
+    }
+}
+
+U0 @nt_rx_loop(@nt* nt)
+{
+    // Make tcp connection to cpu server and send the connect script
+    nt->s = @tcp_socket_create(nt->host, nt->port);
+    while (nt->s->state != TCP_SOCKET_STATE_ESTABLISHED) {
+        Sleep(1);
+    }
+    nt->s->send(nt->connect_script, StrLen(nt->connect_script));
+
+    U8* request = NULL;
+    @9p_msg * msg = NULL;
+    U8* buf = CAlloc(NT_BUF_SIZE, nt->main_task);
+    I64 len = 0;
+    I64 pos = 0;
+    I64 base = 0;
+    U32 request_length = 0;
+
+    // Export 9P filesystem, loop and receive Trequests
+    while (1) {
+        pos = 0;
+        base = 0;
+        len = nt->s->receive(buf + base, 1024);
+    nt_rx_queue_next_request:
+        msg = buf + base;
+        // If request length > data received length, keep receiving
+        while (msg->length > len) {
+            pos = len;
+            len += nt->s->receive(buf + base + pos, 1024);
+        }
+        // Insert request into rx_fifo
+        request = CAlloc(msg->length, nt->main_task);
+        // @nt_rx_dbg(nt, "fifo ins: len: %04d\ttype: %s\ttag: 0x%04x", msg->length, @nt_msg_type_string(msg->type), msg->tag);
+        MemCpy(request, buf + base, msg->length);
+        FifoI64Ins(nt->rx_fifo, request);
+        // If we have more data, queue next request
+        if (msg->length < len) {
+            base += msg->length;
+            len -= msg->length;
+            pos = 0;
+            goto nt_rx_queue_next_request;
+        }
+
+        Sleep(1);
+    }
+}
+
+U0 @nt_dbg_loop(@nt* nt)
+{
+    @nt_dbg_msg* msg = NULL;
+    while (1) {
+        while (FifoI64Cnt(nt->rx_dbg_fifo)) {
+            FifoI64Rem(nt->rx_dbg_fifo, &msg);
+            @nt_dbg_out(msg);
+        }
+        while (FifoI64Cnt(nt->tx_dbg_fifo)) {
+            FifoI64Rem(nt->tx_dbg_fifo, &msg);
+            @nt_dbg_out(msg);
+        }
+        Sleep(1);
+    }
+}
+
+U0 @nt_process_one(@nt* nt)
+{
+    if (!FifoI64Cnt(nt->rx_fifo))
+        return;
+
+    U8* request = NULL;
+    FifoI64Rem(nt->rx_fifo, &request);
+    @9p_msg * msg = request;
+    @9p_debug_dump(msg, msg->length, 0);
+
+    //@nt_tx_dbg(nt, "fifo rem: len: %04d\ttype: %s\ttag: 0x%04x", msg->length, @nt_msg_type_string(msg->type), msg->tag);
+
+    switch (msg->type) {
+    case TVERSION:
+        @nt_Tversion(nt, msg);
+        break;
+    case TATTACH:
+        @nt_Tattach(nt, msg);
+        break;
+    case TWALK:
+        @nt_Twalk(nt, msg);
+        break;
+    case TOPEN:
+        @nt_Topen(nt, msg);
+        break;
+    case TCLUNK:
+        @nt_Tclunk(nt, msg);
+        break;
+    case TREAD:
+        @nt_Tread(nt, msg);
+        break;
+    case TWRITE:
+        @nt_Twrite(nt, msg);
+        break;
+    case TFLUSH:
+        @nt_Tflush(nt, msg);
+        break;
+    case TSTAT:
+        @nt_Tstat(nt, msg);
+        break;
+    default:
+        @nt_tx_dbg(nt, "unimplemented request -");
+        PressAKey;
+        break;
+    }
+
+    Free(request);
+}
+
+U0 @nt_process_kbd(@nt* nt)
+{
+    // Process keyboard events
+
+    if (!nt->kbd_read_tag)
+        return;
+
+    if (cnts.jiffies < nt->kbd_last_jiffies + 80)
+        return;
+
+    if (!FifoI64Cnt(nt->kbd_fifo))
+        return;
+
+    U8* ev = NULL;
+    FifoI64Rem(nt->kbd_fifo, &ev);
+
+    @9p_msg * Rmsg = nt->out;
+
+    if (ev) {
+        Rmsg->length = 0x0e;
+        Rmsg->type = TREAD + 1; // RREAD
+        Rmsg->tag = nt->kbd_read_tag;
+        MemCpy(nt->out + (4 + 1 + 2), ev, 7);
+        // nt->kbd_read_tag = NULL;
+        nt->kbd_last_jiffies = cnts.jiffies;
+        nt->s->send(Rmsg, Rmsg->length);
+        Free(ev);
+    }
+}
+
+U0 @nt_tx_loop(@nt* nt)
+{
+    while (1) {
+        @nt_process_one(nt);
+        @nt_process_kbd(nt);
+        Sleep(1);
+    }
+}
+
+U0 @nt_create_qids(@nt* nt)
+{
+    @nt_qid_create(nt, "/", NT_IS_DIR);
+    @nt_qid_create(nt, "/dev", NT_IS_DIR);
+    @nt_qid_create(nt, "/dev/cons", NT_IS_FILE);
+    @nt_qid_create(nt, "/dev/draw", NT_IS_DIR);
+    @nt_qid_create(nt, "/dev/draw/new", NT_IS_FILE);
+    @nt_qid_create(nt, "/dev/kbd", NT_IS_FILE);
+    @nt_qid_create(nt, "/dev/cursor", NT_IS_FILE);
+    @nt_qid_create(nt, "/dev/mouse", NT_IS_FILE);
+    //@nt_qid_create(nt, "/dev/mousectl", NT_IS_FILE);
+    //@nt_qid_create(nt, "/dev/mousein", NT_IS_FILE);
+}
+
+@nt* @nt_new(U8* host, I64 port)
+{
+    @nt* nt = CAlloc(sizeof(@nt));
+    nt->cursor = @nt_cursor;
+    nt->screen = DCNew(GR_WIDTH, GR_HEIGHT);
+    DCFill(nt->screen, TRANSPARENT);
+    nt->main_task = Fs;
+    nt->main_task->user_data = nt;
+    nt->host = StrNew(host);
+    nt->out = CAlloc(NT_BUF_SIZE);
+    nt->rx_fifo = FifoI64New(NT_FIFO_SIZE);
+    nt->rx_dbg_fifo = FifoI64New(NT_FIFO_SIZE);
+    nt->tx_dbg_fifo = FifoI64New(NT_FIFO_SIZE);
+    nt->kbd_fifo = FifoI64New(NT_FIFO_SIZE);
+    nt->port = port;
+    nt->drawobjs = Json.CreateObject(nt->main_task);
+    nt->fids = Json.CreateObject(nt->main_task);
+    nt->qids = Json.CreateObject(nt->main_task);
+    nt->connect_script = FileRead("M:/script.rc");
+    StrCpy(nt->main_task->task_title, "NeinTerm");
+    nt->dbg_task = Spawn(&@nt_dbg_loop, nt, "NeinTerm DbgTask");
+    nt->rx_task = Spawn(&@nt_rx_loop, nt, "NeinTerm RxTask");
+    nt->tx_task = Spawn(&@nt_tx_loop, nt, "NeinTerm TxTask");
+    nt->main_task->draw_it = &@nt_draw_it;
+
+    @nt_create_qids(nt);
+
+    return nt;
+}
+
+U0 NeinTerm(U8* host, I64 port)
+{
+    @nt* nt = @nt_new(host, port);
+
+    I64 ch, sc, code;
+    U8* ev;
+
+    while (1) {
+
+        code = ScanMsg(&ch, &sc);
+        if (!ch) {
+            switch (sc & 0xff) {
+            case SC_DELETE:
+                ch = 0x7f;
+                break;
+            default:
+                break;
+            }
+        }
+        if (ch) {
+            switch (code) {
+            case MSG_KEY_DOWN:
+                ev = CAlloc(8, nt->main_task);
+                ev[0] = 3;
+                ev[4] = 0x72;
+                ev[5] = ch;
+                FifoI64Ins(nt->kbd_fifo, ev);
+                break;
+            case MSG_KEY_UP:
+                ev = CAlloc(8, nt->main_task);
+                ev[0] = 3;
+                ev[4] = 0x52;
+                ev[5] = ch;
+                FifoI64Ins(nt->kbd_fifo, ev);
+                break;
+            }
+        }
+
+        Sleep(10);
+    }
+}
\ No newline at end of file
diff --git a/Run.HC b/Run.HC
new file mode 100644
index 0000000..2b08615
--- /dev/null
+++ b/Run.HC
@@ -0,0 +1,4 @@
+// Compile System in Adam task
+Adam("Cd(\"M:/System/\");\n");
+AdamFile("M:/System/MakeSystem");
+XTalkWait(Fs, "WinMax;\nNeinTerm(\"10.20.0.254\", 9999);\n");
diff --git a/System/9P/Debug.HC b/System/9P/Debug.HC
new file mode 100644
index 0000000..466e2e4
--- /dev/null
+++ b/System/9P/Debug.HC
@@ -0,0 +1,133 @@
+// debugging drawterm
+
+#define _TVERSION 100
+#define _TAUTH 102
+#define _TATTACH 104
+#define _TERROR 106
+#define _TFLUSH 108
+#define _TWALK 110
+#define _TOPEN 112
+#define _TCREATE 114
+#define _TREAD 116
+#define _TWRITE 118
+#define _TCLUNK 120
+#define _TREMOVE 122
+#define _TSTAT 124
+#define _TWSTAT 126
+
+U0 @9p_debug_printf(U8* fmt, ...)
+{
+    U8* buf;
+    if (argc) {
+        buf = StrPrintJoin(NULL, fmt, argc, argv);
+    } else {
+        buf = StrNew(fmt);
+    }
+    U8* s = buf;
+    while (*s) {
+        OutU8(0xe9, *s++);
+    }
+    Free(buf);
+}
+
+U8* @9p_debug_get_code_str(I64 code)
+{
+    switch (code) {
+    case _TVERSION:
+        return "Tversion";
+    case _TAUTH:
+        return "Tauth";
+    case _TATTACH:
+        return "Tattach";
+    case _TERROR:
+        return "Terror";
+    case _TFLUSH:
+        return "Tflush";
+    case _TWALK:
+        return "Twalk";
+    case _TOPEN:
+        return "Topen";
+    case _TCREATE:
+        return "Tcreate";
+    case _TREAD:
+        return "Tread";
+    case _TWRITE:
+        return "Twrite";
+    case _TCLUNK:
+        return "Tclunk";
+    case _TREMOVE:
+        return "Tremove";
+    case _TSTAT:
+        return "Tstat";
+    case _TWSTAT:
+        return "Twstat";
+    case 1 + _TVERSION:
+        return "Rversion";
+    case 1 + _TAUTH:
+        return "Rauth";
+    case 1 + _TATTACH:
+        return "Rattach";
+    case 1 + _TERROR:
+        return "Rerror";
+    case 1 + _TFLUSH:
+        return "Rflush";
+    case 1 + _TWALK:
+        return "Rwalk";
+    case 1 + _TOPEN:
+        return "Ropen";
+    case 1 + _TCREATE:
+        return "Rcreate";
+    case 1 + _TREAD:
+        return "Rread";
+    case 1 + _TWRITE:
+        return "Rwrite";
+    case 1 + _TCLUNK:
+        return "Rclunk";
+    case 1 + _TREMOVE:
+        return "Rremove";
+    case 1 + _TSTAT:
+        return "Rstat";
+    case 1 + _TWSTAT:
+        return "Rwstat";
+    default:
+        return "(unimplemented)";
+    }
+}
+
+U0 @9p_debug_dump(U8* dat, U64 ndat, I64 mode)
+{
+    I64 i, j, step;
+    U8 buf[17];
+    buf[16] = 0;
+
+    @9p_debug_printf("dump: ptr=%p, length=%d, code: %s, mode=%s\n", dat, ndat, @9p_debug_get_code_str(dat[4]), @t(mode == 1, "tx", "rx"));
+    for (i = 0; i < 32; i++)
+        @9p_debug_printf("=");
+    @9p_debug_printf("\n");
+
+    i = 0;
+
+    while (i < ndat) {
+        @9p_debug_printf("%08x: ", i);
+        step = @t(ndat - i > 16, 16, ndat - i);
+        for (j = 0; j < 16; j++) {
+            if (j > step - 1) {
+                @9p_debug_printf("   ");
+                buf[j] = 0;
+            } else {
+                @9p_debug_printf("%02x ", dat[i + j]);
+                switch (dat[i + j]) {
+                case ' ' ... 'z':
+                    buf[j] = dat[i + j];
+                    break;
+                default:
+                    buf[j] = '.';
+                    break;
+                }
+            }
+        }
+        @9p_debug_printf("%s\n", buf);
+        i += step;
+    }
+    @9p_debug_printf("\n");
+}
\ No newline at end of file
diff --git a/System/9P/Defs.HC b/System/9P/Defs.HC
new file mode 100644
index 0000000..1bc2b15
--- /dev/null
+++ b/System/9P/Defs.HC
@@ -0,0 +1,293 @@
+#define TVERSION 100
+#define TAUTH 102
+#define TATTACH 104
+#define TERROR 106
+#define TFLUSH 108
+#define TWALK 110
+#define TOPEN 112
+#define TCREATE 114
+#define TREAD 116
+#define TWRITE 118
+#define TCLUNK 120
+#define TREMOVE 122
+#define TSTAT 124
+#define TWSTAT 126
+
+#define NT_BUF_SIZE 65536
+#define NT_FIFO_SIZE 128
+#define SET_REQUEST_SIZE U32 request_size = *(data)(U32*);
+
+#define NT_IS_DIR 0x80
+#define NT_IS_FILE 0x0
+
+// #define SET_RESPONSE_SIZE *(nt->out)(U32*) = response_size;
+// #define CLEAR_RESPONSE_BUFFER MemSet(nt->out, NULL, response_size);
+// #define SEND_RESPONSE_BUFFER nt->s->send(nt->out, response_size);
+#define SET_RESPONSE_CMD nt->out[4] = data[4] + 1;
+#define SET_RESPONSE_TAG  \
+    nt->out[5] = data[5]; \
+    nt->out[6] = data[6];
+
+#define SET_MSGS           \
+    @9p_msg * Tmsg = data; \
+    @9p_msg * Rmsg = nt->out;
+#define SEND_RMSG                                                            \
+    SET_RESPONSE_CMD SET_RESPONSE_TAG @9p_debug_dump(Rmsg, Rmsg->length, 1); \
+    nt->s->send(Rmsg, Rmsg->length);
+#define CLEAR_RMSG MemSet(Rmsg(U64) + 4, NULL, Rmsg->length - 4);
+
+U8* @nt_msg_type_string(I64 req)
+{
+    switch (req) {
+    case TVERSION:
+        return "Tversion";
+    case TAUTH:
+        return "Tauth";
+    case TATTACH:
+        return "Tattach";
+    case TERROR:
+        return "Terror";
+    case TFLUSH:
+        return "Tflush";
+    case TWALK:
+        return "Twalk";
+    case TOPEN:
+        return "Topen";
+    case TCREATE:
+        return "Tcreate";
+    case TREAD:
+        return "Tread";
+    case TWRITE:
+        return "Twrite";
+    case TCLUNK:
+        return "Tclunk";
+    case TREMOVE:
+        return "Tremove";
+    case TSTAT:
+        return "Tstat";
+    case TWSTAT:
+        return "Twstat";
+    default:
+        return "(unimplemented)";
+    }
+}
+
+class @9p_qid
+{
+    U8 type;
+    U32 version;
+    U64 path;
+};
+
+class @9p_str
+{
+    U8 length;
+    U8 data;
+};
+
+class @9p_msg
+{
+    U32 length;
+    U8 type;
+    U16 tag;
+    U32 fid;
+};
+
+class @9p_twalk : @9p_msg
+{
+    U32 newfid;
+    U16 nwname;
+    U8 wname; // mem location for nwname*(wname[s])
+};
+
+class @9p_stat
+{
+    U16 size;
+    U16 type;
+    U32 dev;
+    @9p_qid qid;
+    U32 mode;
+    U32 atime;
+    U32 mtime;
+    U64 length;
+    U8 name;
+};
+
+class @9p_direntry
+{
+    U16 type;
+    U32 dev;
+    @9p_qid qid;
+    U32 mode;
+    U32 atime;
+    U32 mtime;
+    U64 length;
+    U8 name;
+};
+
+// The reply will contain a machine-independent directory entry, stat, laid out as follows:
+// size[2]total byte count of the following data
+// type[2]for kernel use
+// dev[4]for kernel use
+// qid.type[1]
+// 	the type of the file (directory, etc.), represented as a bit vector corresponding to the high 8 bits of the file’s mode word.
+// qid.vers[4]
+// 	version number for given path
+// qid.path[8]
+// 	the file server’s unique identification for the file
+// mode[4]
+// 	permissions and flags
+// atime[4]
+// 	last access time
+// mtime[4]
+// 	last modification time
+// length[8]
+// 	length of file in bytes
+// name[ s ]
+// 	file name; must be / if the file is the root directory of the server
+// uid[ s ]owner name
+// gid[ s ]group name
+// muid[ s ]
+// 	name of the user who last modified the file
+
+U64 EndianU64(U64 s)
+{
+    U64 d;
+    d.u8[0] = s.u8[7];
+    d.u8[1] = s.u8[6];
+    d.u8[2] = s.u8[5];
+    d.u8[3] = s.u8[4];
+    d.u8[4] = s.u8[3];
+    d.u8[5] = s.u8[2];
+    d.u8[6] = s.u8[1];
+    d.u8[7] = s.u8[0];
+    return d;
+}
+
+// Mouse Cursor
+CDC* @nt_cursor = DCNew(16, 16);
+Free(@nt_cursor->body);
+U8 @nt_cursor_data[16 * 16] = { 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0x0f, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x0f, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x0f, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x0f, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00, 0x0f, 0xff, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x0f, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0f, 0x00, 0x0f, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0f, 0x0f, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0f, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+@nt_cursor->body = @nt_cursor_data;
+
+class @nt
+{
+    CDC* cursor;
+    CTask* dbg_task;
+    CTask* main_task;
+    CTask* rx_task;
+    CTask* tx_task;
+    TcpSocket* s;
+    U8* host;
+    I64 port;
+    U8* out;
+    CFifoI64* rx_fifo;
+    CFifoI64* rx_dbg_fifo;
+    CFifoI64* tx_dbg_fifo;
+    CFifoI64* kbd_fifo;
+    I64 kbd_last_jiffies;
+    U8* connect_script;
+    U16 kbd_read_tag;
+    // U8 last_kbd_read_request[23];
+    JsonObject* fids;
+    JsonObject* qids;
+    JsonObject* drawobjs;
+    CDC* screen;
+    I64 attach_fid;
+};
+
+class @nt_dbg_msg
+{
+    I64 ts;
+    U8* s;
+    I64 dir;
+};
+
+U0 @nt_dbg_out(@nt_dbg_msg* msg)
+{
+    U8 ts[16];
+    StrPrint(ts, "[%8d] %c ", msg->ts, @t(msg->dir, 'TX', 'RX'));
+    U8* s = ts;
+    while (*s) {
+        OutU8(0xe9, *s++);
+    }
+    s = msg->s;
+    while (*s) {
+        OutU8(0xe9, *s++);
+    }
+    OutU8(0xe9, '\n');
+    // Free(msg->s);
+    // Free(msg);
+}
+
+@9p_qid * @nt_qid_create(@nt* nt, U8* path, I64 type)
+{
+    @9p_qid * qid = CAlloc(sizeof(@9p_qid), nt->main_task);
+    qid->type = type;
+    qid->path = StrNew(path, nt->main_task);
+    nt->qids->set(path, qid, JSON_NUMBER);
+    return qid;
+}
+
+@9p_qid * @nt_lookup_qid_by_path(@nt* nt, U8* path)
+{
+    return nt->qids->@(path);
+}
+
+U0 @nt_pascal_to_c_strings(@nt* nt, U8** wname, U8* data, I64 nwname)
+{
+    I64 i, j = 0;
+    I64 len = 0;
+    for (i = 0; i < nwname; i++) {
+        len = *(data + j)(U16*);
+        j += 2;
+        wname[i] = CAlloc(len + 1, nt->main_task);
+        MemCpy(wname[i], data + j, len);
+        j += len;
+    }
+}
+
+U0 @nt_rx_dbg(@nt* nt, U8* fmt, ...)
+{
+    U8* buf;
+    if (argc) {
+        buf = StrPrintJoin(NULL, fmt, argc, argv);
+    } else {
+        buf = StrNew(fmt);
+    }
+    @nt_dbg_msg* msg = CAlloc(sizeof(@nt_dbg_msg), nt->main_task);
+    msg->ts = cnts.jiffies;
+    msg->s = buf;
+    msg->dir = 0;
+    FifoI64Ins(nt->rx_dbg_fifo, msg);
+}
+
+U0 @nt_tx_dbg(@nt* nt, U8* fmt, ...)
+{
+    U8* buf;
+    if (argc) {
+        buf = StrPrintJoin(NULL, fmt, argc, argv);
+    } else {
+        buf = StrNew(fmt);
+    }
+    @nt_dbg_msg* msg = CAlloc(sizeof(@nt_dbg_msg), nt->main_task);
+    msg->ts = cnts.jiffies;
+    msg->s = buf;
+    msg->dir = 1;
+    FifoI64Ins(nt->tx_dbg_fifo, msg);
+}
+
+U0 @nt_zrmsg(@nt* nt, I64 length)
+{
+    // Zero Rmsg to length and set length
+    @9p_msg * Rmsg = nt->out;
+    MemSet(Rmsg, NULL, length);
+    Rmsg->length = length;
+}
+
+U0 @nt_fid_set(@nt* nt, U8* fidbuf, @9p_qid * qid)
+{
+    JsonObject* fid = Json.CreateObject(nt->main_task);
+    fid->set("qid", qid, JSON_NUMBER);
+    nt->fids->set(fidbuf, fid, JSON_OBJECT);
+}
\ No newline at end of file
diff --git a/System/9P/Draw.HC b/System/9P/Draw.HC
new file mode 100644
index 0000000..9e22096
--- /dev/null
+++ b/System/9P/Draw.HC
@@ -0,0 +1,503 @@
+class @Rect
+{
+    U32 x0;
+    U32 y0;
+    U32 x1;
+    U32 y1;
+};
+
+class @Point
+{
+    U32 x;
+    U32 y;
+}
+
+class @draw_A
+{
+    U32 id;
+    U32 imageid;
+    U32 fillid;
+    U8 public;
+};
+
+class @draw_P
+{
+    U32 dstid;
+    U16 n;
+    U32 wind;
+    @Point ignore;
+    U32 srcid;
+    @Point sp;
+    // U16 sp[4];
+    //  dp[2*2*(n+1)];
+};
+
+class @draw_b
+{
+    U32 id;
+    U32 screenid;
+    U8 refresh;
+    U32 chan;
+    U8 repl;
+    @Rect r;
+    @Rect clipr;
+    U32 color;
+};
+
+class @draw_d
+{
+    U32 dstid;
+    U32 srcid;
+    U32 maskid;
+    @Rect dstr;
+    @Point srcp;
+    @Point maskp;
+};
+
+class @draw_y
+{
+    U32 id;
+    @Rect r;
+};
+
+I64 @image_cbgr24_to_4_bit(CBGR24* ptr, Bool dither_probability)
+{
+    I64 res, k;
+    if (dither_probability) {
+        k = RandU32;
+        if (SqrI64(ptr->r) + SqrI64(ptr->g) + SqrI64(ptr->b) >= 3 * SqrI64(k.u8[0]))
+            res = 8;
+        else
+            res = 0;
+        if (ptr->r >= k.u8[1])
+            res |= RED;
+        if (ptr->g >= k.u8[2])
+            res |= GREEN;
+        if (ptr->b >= k.u8[3])
+            res |= BLUE;
+    } else {
+        if (SqrI64(ptr->r) + SqrI64(ptr->g) + SqrI64(ptr->b) >= SqrI64(0x80)) {
+            res = 8;
+            if (ptr->r >= 0x80)
+                res |= RED;
+            if (ptr->g >= 0x80)
+                res |= GREEN;
+            if (ptr->b >= 0x80)
+                res |= BLUE;
+        } else {
+            res = 0;
+            if (ptr->r >= 0x40)
+                res |= RED;
+            if (ptr->g >= 0x40)
+                res |= GREEN;
+            if (ptr->b >= 0x40)
+                res |= BLUE;
+        }
+    }
+    return res;
+}
+
+I64 @nt_draw_A(@nt* nt, @draw_A* d, JsonObject* dobj)
+{
+    // FIXME: Implement this.
+    @nt_tx_dbg(nt, "id: %d, imageid: %d, fillid: %d, public: %d",
+        d->id, d->imageid, d->fillid, d->public);
+    // PressAKey;
+    return sizeof(@draw_A);
+}
+
+I64 @nt_draw_b(@nt* nt, @draw_b* d, JsonObject* dobj)
+{
+    // Allocate an image with a given id on the screen named by screenid.
+    // b id[4] screenid[4] refresh[1] chan[4] repl[1] r[4 * 4] clipr[4 * 4] color[4]
+
+    @nt_tx_dbg(nt, "id: %d, screenid: %d, refresh: %d, chan: %d, repl: %d",
+        d->id, d->screenid, d->refresh, d->chan, d->repl);
+    @nt_tx_dbg(nt, "rect: x0: %d, y0: %d, x1: %d, y1: %d",
+        d->r.x0, d->r.y0, d->r.x1, d->r.y1);
+    @nt_tx_dbg(nt, "clip rect: x0: %d, y0: %d, x1: %d, y1: %d",
+        d->clipr.x0, d->clipr.y0, d->clipr.x1, d->clipr.y1);
+    @nt_tx_dbg(nt, "color: %08x",
+        d->color);
+
+    U8 intbuf[16];
+    StrPrint(intbuf, "%d", d->id);
+
+    if (dobj->o("imgs")->@(intbuf)) {
+        @nt_tx_dbg(nt, "error: image id already exists");
+        PressAKey;
+    }
+
+    CDC* dc = NULL;
+    if (!d->repl) {
+        dc = DCNew(d->r.x1 - d->r.x0, d->r.y1 - d->r.y0);
+    }
+
+    JsonObject* iobj = Json.CreateObject(nt->main_task);
+    JsonObject* tmpr = NULL;
+
+    iobj->set("id", d->id, JSON_NUMBER);
+    iobj->set("screenid", d->screenid, JSON_NUMBER);
+    iobj->set("refresh", d->refresh, JSON_NUMBER);
+    iobj->set("chan", d->chan, JSON_NUMBER);
+    iobj->set("repl", d->repl, JSON_NUMBER);
+    iobj->set("dc", dc, JSON_NUMBER);
+    iobj->set("r", Json.CreateObject(nt->main_task), JSON_OBJECT);
+    iobj->set("clipr", Json.CreateObject(nt->main_task), JSON_OBJECT);
+
+    tmpr = iobj->o("r");
+    tmpr->set("x0", d->r.x0, JSON_NUMBER);
+    tmpr->set("y0", d->r.x1, JSON_NUMBER);
+    tmpr->set("x1", d->r.y0, JSON_NUMBER);
+    tmpr->set("y1", d->r.y1, JSON_NUMBER);
+
+    tmpr = iobj->o("clipr");
+    tmpr->set("x0", d->clipr.x0, JSON_NUMBER);
+    tmpr->set("y0", d->clipr.x1, JSON_NUMBER);
+    tmpr->set("x1", d->clipr.y0, JSON_NUMBER);
+    tmpr->set("y1", d->clipr.y1, JSON_NUMBER);
+
+    if (dc) {
+        CBGR24 bgr;
+        bgr.b = d->color.u8[0];
+        bgr.g = d->color.u8[1];
+        bgr.r = d->color.u8[2];
+        DCFill(dc, @image_cbgr24_to_4_bit(&bgr, 0));
+    }
+
+    StrPrint(intbuf, "%d", d->id);
+    dobj->o("imgs")->set(intbuf, iobj, JSON_OBJECT);
+
+    return sizeof(@draw_b);
+}
+
+U0 @nt_draw_y_mono_1bit(CDC* dc, @draw_y* d)
+{
+    U8* _c = d;
+    I64 x = 0;
+    I64 y = 0;
+    I64 sx = 0;
+    _c += sizeof(@draw_y);
+    for (y = 0; y < d->r.y1; y++) {
+        for (x = 0; x < d->r.x1; x++) {
+            dc->color = @t((reverse_bits(*_c) >> sx) & 1, 15, 0);
+            GrPlot(dc, x, y);
+            sx++;
+            if (sx > 7) {
+                sx = 0;
+                _c++;
+            }
+        }
+        if (sx) {
+            sx = 0;
+            _c++;
+        }
+    }
+}
+
+I64 @nt_draw_y(@nt* nt, @draw_y* d, JsonObject* dobj)
+{
+    // y id[4] r[4*4] buf[x*1]
+
+    // 	Replace the rectangle r of pixels in image id with the pixel data in buf.
+
+    @nt_tx_dbg(nt, "id: %d", d->id);
+    @nt_tx_dbg(nt, "rect: x0: %d, y0: %d, x1: %d, y1: %d",
+        d->r.x0, d->r.y0, d->r.x1, d->r.y1);
+
+    U32 int_w;
+    U8 intbuf[16];
+    StrPrint(intbuf, "%d", d->id);
+
+    JsonObject* iobj = dobj->o("imgs")->@(intbuf);
+
+    if (!iobj) {
+        @nt_tx_dbg(nt, "imageid %d does not exist\n", d->id);
+        PressAKey;
+    }
+
+    CDC* dc = iobj->@("dc");
+
+    if (!dc) {
+        @nt_tx_dbg(nt, "no device context for imageid %d\n", d->id);
+        PressAKey;
+    }
+
+    switch (iobj->@("chan")) {
+    case 0x31:
+        @nt_draw_y_mono_1bit(dc, d);
+        int_w = d->r.x1;
+        while (int_w % 8) {
+            int_w++;
+        }
+        return sizeof(@draw_y) + ((int_w * d->r.y1) / 8);
+    default:
+        @nt_tx_dbg(nt, "unsupported chan value: 0x%04x", iobj->@("chan"));
+        PressAKey;
+        break;
+    }
+}
+
+// drawcoord(uchar *p, uchar *maxp, int oldx, int *newx)
+U8* @nt_drawcoord(@nt* nt, U8* p, U8* maxp, U32 oldx, U32* newx)
+{
+    U32 b, x;
+
+    if (p >= maxp) {
+        @nt_tx_dbg(nt, "error: Eshortdraw1: p = 0x%08x, maxp = 0x%08x", p, maxp);
+        PressAKey;
+    }
+    b = *p++;
+    x = b & 0x7F;
+    if (b & 0x80) {
+        if (p + 1 >= maxp) {
+            @nt_tx_dbg(nt, "error: Eshortdraw2: p = 0x%08x", p);
+            PressAKey;
+        }
+        x |= *p++ << 7;
+        x |= *p++ << 15;
+        if (x & (1 << 22))
+            x |= ~0 << 23;
+    } else {
+        if (b & 0x40)
+            x |= ~0 << 7;
+        x += oldx;
+    }
+    *newx = x;
+    return p;
+}
+
+I64 @nt_draw_P(@nt* nt, @draw_P* d, JsonObject* dobj)
+{
+    // P dstid[4] n[2] wind[4] ignore[2*4] srcid[4] sp[2*4] dp[2*2*(n+1)]
+
+    // Draw a polygon as the p message, but fill it rather than outlining it.
+
+    @nt_tx_dbg(nt, "dstid: %d, n: %d, srcid: %d",
+        d->dstid, d->n, d->srcid);
+
+    // FIXME: Implement this.
+
+    JsonObject* iobj;
+    U8 idbuf[16];
+    I64 dx, dy;
+    CDC* dst = NULL;
+    CDC* src = NULL;
+
+    StrPrint(idbuf, "%d", d->dstid);
+    iobj = dobj->o("imgs")->o(idbuf);
+    if (iobj) {
+        dst = iobj->@("dc");
+    }
+
+    //    StrPrint(idbuf, "%d", d->srcid);
+    //    iobj = dobj->o("imgs")->o(idbuf);
+    //    if (iobj) {
+    //        src = iobj->@("dc");
+    //    }
+
+    if (dst) {
+        dx = 0 + d->sp.x;
+        dy = 0 + d->sp.y;
+
+        //  dy = d->sp.y;
+
+        // dst->color = BLUE;
+        //@nt_tx_dbg(nt, "plot dst pixel: dx: %d, dy: %d", dx, dy);
+        // GrPlot(dst, dx, dy);
+    }
+
+    I64 i; //, j;
+    I64 dp_length = 0;
+    U8* dp_ptr = d;
+    U8* maxp = 0x80000000000000;
+    dp_ptr += 4 + 2 + 4 + 4 + 4 + 4 + 2 * 4;
+    // d += sizeof(@draw_P);
+    I64 b = 0;
+
+    // m = 1+4+2+4 +4+4 +4 +2*4
+
+    //    U32 ox = 0;
+    //    U32 oy = 0;
+    //    p.x = d->sp.x;
+    //    p.y = d->sp.y;
+
+    U64 __p = d;
+    __p += sizeof(@draw_P);
+
+    @Point p;
+    p.x = (__p)(U32*)[0];
+    p.y = (__p)(U32*)[1];
+
+    U32 ox = p.x;
+    U32 oy = p.y;
+
+    for (i = 0; i < d->n + 1; i++) {
+        dp_ptr = @nt_drawcoord(nt, dp_ptr, maxp, ox, &p.x);
+        dp_ptr = @nt_drawcoord(nt, dp_ptr, maxp, oy, &p.y);
+
+        if (dst) {
+            dst->color = BLACK;
+            @nt_tx_dbg(nt, "plot line: x0: %d, y0: %d, x1: %d, y1: %d", ox, oy, p.x, p.y);
+            if (!(ox > dst->width || oy > dst->height || p.x > dst->width || p.y > dst->height)) {
+                GrLine(dst, ox, oy, p.x, p.y);
+            }
+            // GrPlot(dst, dx + p.x, dy + p.y);
+        }
+
+        ox = p.x;
+        oy = p.y;
+        //        for (j = 0; j < 2; j++) {
+        //            b = dp_ptr[dp_length++];
+        //            if (b & 0x80) {
+        //                dp_length += 2;
+        //            }
+        //        }
+    }
+    // Sleep(5000);
+    // PressAKey;
+    return sizeof(@draw_P) + dp_length - d(U8*);
+}
+
+I64 @nt_draw_d(@nt* nt, @draw_d* d, JsonObject* dobj)
+{
+    //  m = 1+4+4+4+4*4+2*4+2*4;
+    //  d dstid[4] srcid[4] maskid[4] dstr[4*4] srcp[2*4] maskp[2*4]
+    //  	Use the draw operator to combine the rectangle dstr of image dstid with a rectangle of image srcid ...
+    //         res = 4 + 4 + 4 + sizeof(@Rect) + sizeof(@Point) + sizeof(@Point);
+
+    I64 res = 4 + 4 + 4 + sizeof(@Rect) + sizeof(@Point) + sizeof(@Point);
+
+    @nt_tx_dbg(nt, "dstid: %d, srcid: %d, maskid: %d, dstr: (%d,%d, %d,%d), srcp: (%d,%d) maskp: (%d,%d)",
+        d->dstid, d->srcid, d->maskid, d->dstr.x0, d->dstr.y0, d->dstr.x1, d->dstr.y1,
+        d->srcp.x, d->srcp.y, d->maskp.x, d->maskp.y);
+
+    JsonObject* iobj;
+    U8 idbuf[16];
+    I64 dx, dy;
+    CDC* dst;
+    CDC* src;
+
+    StrPrint(idbuf, "%d", d->dstid);
+    iobj = dobj->o("imgs")->o(idbuf);
+    if (!iobj) {
+        return res;
+    }
+    dst = iobj->@("dc");
+    if (!dst) {
+        return res;
+    }
+
+    StrPrint(idbuf, "%d", d->srcid);
+    iobj = dobj->o("imgs")->o(idbuf);
+    if (!iobj) {
+        return res;
+    }
+    src = iobj->@("dc");
+    if (!src) {
+        return res;
+    }
+
+    dx = U32_MAX - d->srcp.x;
+    dy = U32_MAX - d->srcp.y;
+
+    GrBlot(dst, dx, dy, src);
+
+    return res;
+
+    //    if (dobj->o("imgs")->o(idbuf)) {
+    //        @nt_tx_dbg(nt, "image id %d exists for srcid", d->srcid);
+    //        src = dobj->o("imgs")->o(idbuf)->@("dc");
+    //
+    //        if (src) {
+    //            @nt_tx_dbg(nt, "src dc: width: %d, height: %d", src->width, src->height);
+    //            if (d->dstr.x1 == 640 && d->dstr.y1 == 480 && src->width != 640) {
+    //                dx = U32_MAX - d->srcp.x;
+    //                dy = U32_MAX - d->srcp.y;
+    //                GrBlot(gr.dc, dx, dy, src);
+    //            }
+    //        }
+    //    }
+    //
+    //    return 4 + 4 + 4 + sizeof(@Rect) + sizeof(@Point) + sizeof(@Point);
+}
+I64 @nt_draw_cmd(@nt* nt, U8* data, JsonObject* dobj, U8* data_ptr, I64 i)
+{
+    I64 cmd = data_ptr[i++];
+    I64 res = 0;
+    // size[4] Twrite tag[2] fid[4] offset[8] count[4]
+    @nt_tx_dbg(nt, "draw cmd: %c, offset: 0x%04x", cmd, i - 1 + (4 + 1 + 2 + 4 + 8 + 4));
+
+    JsonKey* k;
+    JsonObject* o;
+    CDC* dc;
+
+    switch (cmd) {
+    case 0:
+        @nt_tx_dbg(nt, "ERROR: draw cmd is NULL, i: 0x%04x", i - 1 + (4 + 1 + 2 + 4 + 8 + 4));
+        PressAKey;
+        break;
+    case 'A':
+        res = @nt_draw_A(nt, data_ptr + i, dobj);
+        break;
+    case 'p':
+    case 'P':
+        res = @nt_draw_P(nt, data_ptr + i, dobj);
+        break;
+    case 'b':
+        res = @nt_draw_b(nt, data_ptr + i, dobj);
+        break;
+    case 'v':
+        k = dobj->o("imgs")->keys;
+        while (k) {
+            o = k->value;
+            if (o && o->@("dc")) {
+                dc = o->@("dc");
+                if (dc->width == 640 && dc->height == 480) {
+                    GrBlot(nt->screen, 0, 0, dc);
+                }
+            }
+            k = k->next;
+        }
+        // FIXME: update display buffer
+        break;
+    case 'y':
+        res = @nt_draw_y(nt, data_ptr + i, dobj);
+        break;
+    case 'c':
+        // c dstid[4] repl[1] clipr[4*4]
+        // Change the replicate bit and clipping rectangle of the image dstid.
+        res = 4 + 1 + sizeof(@Rect);
+        break;
+    case 'd':
+        res = @nt_draw_d(nt, data_ptr + i, dobj);
+        break;
+    default:
+        // Sleep(5000);
+        @nt_tx_dbg(nt, "ERROR: unimplemented draw cmd: %c, offset: 0x%04x", cmd, i - 1 + (4 + 1 + 2 + 4 + 8 + 4));
+        // Sleep(5000);
+        PressAKey;
+        break;
+    }
+    return ++res;
+}
+
+U0 @nt_draw(@nt* nt, U8* data, @9p_qid * qid)
+{
+    SET_MSGS
+
+    // size[4] Twrite tag[2] fid[4] offset[8] count[4] data[count]
+
+    U8 dobjbuf[32];
+    StrCpy(dobjbuf, StrFind("/dev/draw/", qid->path) + 10);
+    StrFind("/data", dobjbuf)[0] = NULL;
+    JsonObject* dobj = nt->drawobjs->@(dobjbuf);
+
+    U64 draw_ptr = data + (4 + 1 + 2 + 4 + 8 + 4);
+    U32 length = *(data + (4 + 1 + 2 + 4 + 8))(U32*);
+
+    I64 i = 0;
+    while (i < length) {
+        i += @nt_draw_cmd(nt, data, dobj, draw_ptr, i);
+    }
+}
\ No newline at end of file
diff --git a/System/9P/Requests.HC b/System/9P/Requests.HC
new file mode 100644
index 0000000..2d321af
--- /dev/null
+++ b/System/9P/Requests.HC
@@ -0,0 +1,557 @@
+U0 @nt_Rerror(@nt* nt, U8* data, U8* errormsg)
+{
+    // 	size[4] Rerror tag[2] ename[s]
+
+    SET_MSGS
+
+    Rmsg->length = (4 + 1 + 2 + 2);
+    nt->out[4] = TERROR + 1; // RERROR
+
+    MemCpy(nt->out + (4 + 1 + 2 + 2), errormsg, StrLen(errormsg));
+    *(nt->out + (4 + 1 + 2))(U16*) = StrLen(errormsg);
+
+    SET_RESPONSE_TAG
+    Rmsg->length += StrLen(errormsg);
+
+    @9p_debug_dump(Rmsg, Rmsg->length, 1);
+    nt->s->send(Rmsg, Rmsg->length);
+}
+
+U0 @nt_Tversion(@nt* nt, U8* data)
+{
+    SET_MSGS
+
+    // Special case; just increment Tmsg tag to Rmsg and echo back to client
+    data[4]++;
+    @9p_debug_dump(data, Tmsg->length, 1);
+    nt->s->send(data, Tmsg->length);
+}
+
+U0 @nt_Tattach(@nt* nt, U8* data)
+{
+    SET_MSGS
+    @nt_zrmsg(nt, 20);
+    //@nt_tx_dbg(nt, "Tattach fid: %d", Tmsg->fid);
+    nt->attach_fid = Tmsg->fid;
+
+    // Map fid to qid
+    U8 fidbuf[16];
+    @9p_qid * qid = @nt_lookup_qid_by_path(nt, "/");
+    StrPrint(fidbuf, "%d", Tmsg->fid);
+    // nt->fids->set(fidbuf, qid, JSON_NUMBER);
+    @nt_fid_set(nt, fidbuf, qid);
+
+    MemCpy(nt->out + 7, qid, sizeof(@9p_qid));
+    SEND_RMSG
+}
+
+U0 @nt_Twalk(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/walk.html
+
+    // size[4] Twalk tag[2] fid[4] newfid[4] nwname[2] nwname*(wname[s])
+    // size[4] Rwalk tag[2] nwqid[2] nwqid*(qid[13])
+
+    // The fid must be valid in the current session and must not have been opened for I/O by an open or create message.
+    // If the full sequence of nwname elements is walked successfully, newfid will represent the file that results.
+    // If not, newfid (and fid) will be unaffected. However, if newfid is in use or otherwise illegal, an Rerror is returned.
+
+    // If the first element cannot be walked for any reason, Rerror is returned.
+    // Otherwise, the walk will return an Rwalk message containing:
+    //   nwqid qids corresponding, in order, to the files that are visited by the nwqid successful elementwise walks;
+    //   nwqid is therefore either nwname or the index of the first elementwise walk that failed.
+
+    // size[4] Rwalk tag[2] nwqid[2] nwqid*(qid[13])
+
+    // The value of nwqid cannot be zero unless nwname is zero.
+    // Also, nwqid will always be less than or equal to nwname.
+    // Only if it is equal, however, will newfid be affected, in which case newfid will represent:
+    //   the file reached by the final elementwise walk requested in the message.
+
+    SET_MSGS
+    @9p_twalk * Twalk = data;
+    U8** wname = NULL;
+
+    I64 i, j;
+    U16 fid = Twalk->fid;
+    U16 newfid = Twalk->newfid;
+    U16 nwname = Twalk->nwname;
+    U16 nwqid = 0;
+    @9p_qid * qid = NULL;
+
+    U8 basepath[512];
+    U8 pathbuf[512];
+    U8 fidbuf[16];
+
+    Rmsg->length = (4 + 1 + 2 + 2);
+    //@nt_tx_dbg(nt, "Twalk fid: %d, newfid: %d, nwname: %d", fid, newfid, nwname);
+
+    if (fid != nt->attach_fid) {
+        // FIXME: Lookup base path from fid
+        StrPrint(fidbuf, "%d", fid);
+        qid = nt->fids->o(fidbuf)->@("qid");
+        if (!qid) {
+            @nt_tx_dbg(nt, "ERROR: !qid");
+            PressAKey;
+        }
+        StrCpy(basepath, qid->path);
+        if (qid->type != NT_IS_DIR) {
+            StrLastOcc(basepath, "/")[0] = NULL;
+        }
+    } else {
+        StrCpy(basepath, "");
+    }
+
+    if (nwname) {
+        wname = CAlloc(sizeof(U8*) * nwname, nt->main_task);
+        @nt_pascal_to_c_strings(nt, wname, &Twalk->wname, nwname);
+
+        for (i = 0; i < nwname; i++) {
+            StrCpy(pathbuf, basepath);
+            for (j = 0; j <= i; j++) {
+                StrPrint(pathbuf + StrLen(pathbuf), "/%s", wname[j]);
+            }
+            qid = @nt_lookup_qid_by_path(nt, pathbuf);
+            if (qid) {
+                // Copy the qid to Rwalk and increment nwqid, Rwalk length
+                MemCpy(nt->out + (4 + 1 + 2 + 2 + (nwqid * sizeof(@9p_qid))), qid, sizeof(@9p_qid));
+                nwqid++;
+                Rmsg->length += sizeof(@9p_qid);
+            } else {
+
+                // Rerror
+                Rmsg->length = (4 + 1 + 2 + 2);
+                Rmsg->type = 0x6b; // Rerror
+                SET_RESPONSE_TAG
+                StrPrint(nt->out + (4 + 1 + 2 + 2), "file does not exist: '.%s'", pathbuf);
+                //@nt_tx_dbg(nt, "newfid: %d, %s", newfid, nt->out + (4 + 1 + 2 + 2));
+                *(nt->out + (4 + 1 + 2))(U16*) = StrLen(nt->out + (4 + 1 + 2 + 2));
+                Rmsg->length += StrLen(nt->out + (4 + 1 + 2 + 2));
+                @9p_debug_dump(Rmsg, Rmsg->length, 1);
+                nt->s->send(Rmsg, Rmsg->length);
+                return;
+
+                break;
+            }
+        }
+    }
+
+    // Map newfid to qid
+    StrPrint(fidbuf, "%d", newfid);
+    // nt->fids->set(fidbuf, qid, JSON_NUMBER);
+    @nt_fid_set(nt, fidbuf, qid);
+
+    // Set nwqid
+    *(nt->out + (4 + 1 + 2))(U16*) = nwqid;
+
+    //@nt_tx_dbg(nt, "newfid: %d, walked: %s", newfid, pathbuf);
+
+    SEND_RMSG
+}
+
+U0 @nt_Topen(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/open.html
+
+    // size[4] Topen tag[2] fid[4] mode[1]
+    // size[4] Ropen tag[2] qid[13] iounit[4]
+
+    SET_MSGS
+
+    U32 fid = *(data + (4 + 1 + 2))(U32*);
+    I64 mode = data[4 + 1 + 2 + 4];
+
+    U8 fidbuf[16];
+    StrPrint(fidbuf, "%d", fid);
+
+    Rmsg->length = (4 + 1 + 2 + 13 + 4);
+    //@nt_tx_dbg(nt, "Topen fid: %d, mode: %d", fid, mode);
+
+    @9p_qid * qid = nt->fids->o(fidbuf)->@("qid");
+
+    if (!qid) {
+        @nt_tx_dbg("ERROR: no qid for fid %d", fid);
+        PressAKey;
+    }
+
+    MemCpy(nt->out + 7, qid, sizeof(@9p_qid));
+    *(nt->out + (4 + 1 + 2 + 13))(U32*) = NULL; // set iounit to NULL for now
+
+    SEND_RMSG
+}
+
+U0 @nt_Tclunk(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/clunk.html
+
+    // size[4] Tclunk tag[2] fid[4]
+    // size[4] Rclunk tag[2]
+
+    SET_MSGS
+
+    U32 fid = *(data + (4 + 1 + 2))(U32*);
+
+    U8 fidbuf[16];
+    StrPrint(fidbuf, "%d", fid);
+
+    Rmsg->length = (4 + 1 + 2);
+    //@nt_tx_dbg(nt, "Tclunk fid: %d", fid);
+
+    nt->fids->unset(fidbuf);
+
+    SEND_RMSG
+}
+
+// I64 test_read_cnt = 0;
+
+I64 test_read_cnt = 0;
+// I64 test_entry_size = 0x45;
+I64 test_entry_size = 0x41;
+
+// U8 test_entry[test_entry_size] = { 0x3e, 0x00, 0x62, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x24, 0x01, 0x00, 0x00, 0x27, 0xab, 0xe2, 0x68, 0x1b, 0xab, 0xe2, 0x68, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x64, 0x72, 0x61, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63 };
+U8 test_entry[test_entry_size] = { 0x3f, 0x00, 0x69, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x24, 0x01, 0x00, 0x80, 0x27, 0xab, 0xe2, 0x68, 0x1b, 0xab, 0xe2, 0x68, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x64, 0x72, 0x61, 0x77, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63 };
+// U8 test_entry[test_entry_size] = { 0x3f, 0x00, 0x69, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x6d, 0x01, 0x00, 0x80, 0x4d, 0xb2, 0xe2, 0x68, 0x47, 0xb2, 0xe2, 0x68, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x64, 0x72, 0x61, 0x77, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x04, 0x00, 0x61, 0x6c, 0x65, 0x63, 0x42, 0x00, 0x69, 0x00 };
+
+U0 @dev_draw_new_read(@nt* nt, U8* data)
+{
+    SET_MSGS
+
+    U32 read_size = 144;
+
+    I64 i;
+    U8 newbuf[144];
+    U8 intbuf[16];
+    U8 pathbuf[128];
+    U8* chanstr = "x8r8g8b8";
+    JsonObject* tmpr;
+    I64 conn = nt->drawobjs->length + 1;
+
+    JsonObject* dobj = Json.CreateObject(nt->main_task);
+    dobj->set("conn", conn, JSON_NUMBER);
+    dobj->set("imageid", 0, JSON_NUMBER);
+    dobj->set("chan", chanstr, JSON_STRING);
+    dobj->set("r", Json.CreateObject(nt->main_task), JSON_OBJECT);
+    dobj->set("clipr", Json.CreateObject(nt->main_task), JSON_OBJECT);
+
+    tmpr = dobj->o("r");
+    tmpr->set("x0", 0, JSON_NUMBER);
+    tmpr->set("y0", 0, JSON_NUMBER);
+    tmpr->set("x1", GR_WIDTH, JSON_NUMBER);
+    tmpr->set("y1", GR_HEIGHT, JSON_NUMBER);
+
+    tmpr = dobj->o("clipr");
+    tmpr->set("x0", 0, JSON_NUMBER);
+    tmpr->set("y0", 0, JSON_NUMBER);
+    tmpr->set("x1", GR_WIDTH, JSON_NUMBER);
+    tmpr->set("y1", GR_HEIGHT, JSON_NUMBER);
+
+    dobj->set("imgs", Json.CreateObject(nt->main_task), JSON_OBJECT);
+
+    // Insert new draw object [/dev/draw/[conn#] to nt->drawobjs]
+    StrPrint(intbuf, "%d", conn);
+    nt->drawobjs->set(intbuf, dobj, JSON_OBJECT);
+
+    // Create path objects
+    StrPrint(pathbuf, "/dev/draw/%d", conn);
+    @nt_qid_create(nt, pathbuf, NT_IS_DIR);
+    StrPrint(pathbuf, "/dev/draw/%d/data", conn);
+    @nt_qid_create(nt, pathbuf, NT_IS_FILE);
+    StrPrint(pathbuf, "/dev/draw/%d/refresh", conn);
+    @nt_qid_create(nt, pathbuf, NT_IS_FILE);
+
+    // ** not yet implemented
+    // StrPrint(pathbuf, "/dev/draw/%d/colormap", conn);
+    // @nt_qid_create(nt, pathbuf, NT_IS_FILE);
+    // StrPrint(pathbuf, "/dev/draw/%d/ctl", conn);
+    // @nt_qid_create(nt, pathbuf, NT_IS_FILE);
+
+    // *** Build payload for Rread
+
+    // Default; set all fields to 0
+    MemSet(newbuf, ' ', read_size);
+    for (i = 0; i < 12; i++) {
+        newbuf[10 + (12 * i)] = '0';
+    }
+
+    // Set connection number
+    StrPrint(intbuf, "%d", conn);
+    MemCpy(newbuf + 11 - StrLen(intbuf), intbuf, StrLen(intbuf));
+
+    // Set chan string
+    MemCpy(newbuf + 11 - StrLen(chanstr) + (12 * 2), chanstr, StrLen(chanstr));
+
+    // Set max.x of display image and clipping rect
+    StrPrint(intbuf, "%d", GR_WIDTH);
+    MemCpy(newbuf + 11 - StrLen(intbuf) + (12 * 6), intbuf, StrLen(intbuf));
+    MemCpy(newbuf + 11 - StrLen(intbuf) + (12 * 10), intbuf, StrLen(intbuf));
+
+    // Set max.y of display image and clipping rect
+    StrPrint(intbuf, "%d", GR_HEIGHT);
+    MemCpy(newbuf + 11 - StrLen(intbuf) + (12 * 7), intbuf, StrLen(intbuf));
+    MemCpy(newbuf + 11 - StrLen(intbuf) + (12 * 11), intbuf, StrLen(intbuf));
+
+    Rmsg->length = (4 + 1 + 2 + 4);
+    MemCpy(nt->out + (4 + 1 + 2 + 4), newbuf, read_size);
+    *(nt->out + (4 + 1 + 2))(U32*) = read_size;
+    Rmsg->length += read_size;
+
+    SEND_RMSG
+}
+
+U0 @nt_Rread_dir(@nt* nt, U8* data, @9p_qid * qid)
+{
+    SET_MSGS
+
+    Rmsg->length = (4 + 1 + 2 + 4);
+
+    if (test_read_cnt) {
+        *(nt->out + (4 + 1 + 2))(U32*) = 0;
+        SEND_RMSG
+        return;
+    }
+
+    U32 read_size = test_entry_size;
+    MemCpy(nt->out + (4 + 1 + 2 + 4), test_entry, read_size);
+
+    *(nt->out + (4 + 1 + 2))(U32*) = read_size;
+    Rmsg->length += read_size;
+    SEND_RMSG
+
+    test_read_cnt++;
+}
+
+//    @9p_msg * Rmsg = nt->out;
+//
+//    if (ev) {
+//        Rmsg->length = 0x0e;
+//        Rmsg->type = TREAD + 1; // RREAD
+//        Rmsg->tag = nt->kbd_read_tag;
+//        MemCpy(nt->out + (4 + 1 + 2), ev, 7);
+//        // nt->kbd_read_tag = NULL;
+//        nt->kbd_last_jiffies = cnts.jiffies;
+//        nt->s->send(Rmsg, Rmsg->length);
+//        Free(ev);
+//    }
+
+U0 @nt_Tread(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/read.html
+
+    // size[4] Tread tag[2] fid[4] offset[8] count[4]
+    // size[4] Rread tag[2] count[4] data[count]
+
+    SET_MSGS
+
+    U32 _fid = *(data + (4 + 1 + 2))(U32*);
+    U64 offset = *(data + (4 + 1 + 2 + 4))(U64*);
+    U32 count = *(data + (4 + 1 + 2 + 4 + 8))(U32*);
+
+    //@nt_tx_dbg(nt, "Tread fid: %d, offset: %d, count: %d", fid, offset, count);
+
+    U8 fidbuf[16];
+    StrPrint(fidbuf, "%d", _fid);
+
+    JsonObject* fid = nt->fids->o(fidbuf);
+
+    if (!fid)
+        return;
+
+    @9p_qid * qid = fid->@("qid");
+
+    if (!qid)
+        return;
+
+    I64 read_count = fid->@("read_count");
+
+    if (qid) {
+        switch (qid->type) {
+        case NT_IS_DIR:
+            if (!StrCmp(qid->path, "/dev")) {
+                @nt_Rread_dir(nt, data, qid);
+                return;
+            }
+            //@nt_tx_dbg(nt, "got read on dir: %s", qid->path);
+            break;
+        default:
+            if (!StrCmp(qid->path, "/dev/kbd")) {
+                nt->kbd_read_tag = Tmsg->tag;
+                //@nt_tx_dbg(nt, "Set nt->kbd_read_tag: 0x%04x", Tmsg->tag);
+            }
+            if (!StrCmp(qid->path, "/dev/draw/new")) {
+                if (read_count) {
+                    @nt_Rerror(nt, data, "unknown id for draw image");
+                } else {
+                    @dev_draw_new_read(nt, data);
+                }
+            }
+            break;
+        }
+    }
+
+    read_count++;
+    fid->set("read_count", read_count, JSON_NUMBER);
+}
+
+U0 @nt_Twrite(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/read.html
+
+    // size[4] Twrite tag[2] fid[4] offset[8] count[4] data[count]
+    // size[4] Rwrite tag[2] count[4]
+
+    SET_MSGS
+
+    U32 fid = *(data + (4 + 1 + 2))(U32*);
+    U64 offset = *(data + (4 + 1 + 2 + 4))(U64*);
+    U32 count = *(data + (4 + 1 + 2 + 4 + 8))(U32*);
+    U8* p = data + (4 + 1 + 2 + 4 + 8 + 4);
+
+    Rmsg->length = (4 + 1 + 2 + 4);
+    //@nt_tx_dbg(nt, "Twrite fid: %d, offset: %d, count: %d", fid, offset, count);
+
+    U8 fidbuf[16];
+    U8 errmsgbuf[256];
+    StrPrint(fidbuf, "%d", fid);
+    @9p_qid * qid = nt->fids->o(fidbuf)->@("qid");
+
+    I64 i;
+
+    if (qid) {
+        if (String.BeginsWith("/dev/draw", qid->path) && String.EndsWith("/data", qid->path)) {
+        }
+        if (!StrCmp(qid->path, "/dev/draw/1/data")) {
+            @nt_draw(nt, data, qid);
+        }
+        if (!StrCmp(qid->path, "/dev/cons")) {
+            for (i = 0; i < count; i++) {
+                DocPrint(nt->main_task->put_doc, "%c", p[i]);
+            }
+        }
+        if (!StrCmp(qid->path, "/dev/mousectl")) {
+            p[count] = NULL;
+            StrPrint(errmsgbuf, "unknown control message \"%s\"", p);
+            @nt_Rerror(nt, data, errmsgbuf);
+            return;
+        }
+
+        //@nt_tx_dbg(nt, "got write for path: %s", qid->path);
+    }
+
+    *(nt->out + (4 + 1 + 2))(U32*) = count;
+
+    SEND_RMSG
+}
+
+U0 @nt_Tflush(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/flush.html
+
+    // size[4] Tflush tag[2] oldtag[2]
+    // size[4] Rflush tag[2]
+
+    SET_MSGS
+
+    Rmsg->length = (4 + 1 + 2);
+    SEND_RMSG
+}
+
+U0 @nt_Tstat(@nt* nt, U8* data)
+{
+    // https://9fans.github.io/plan9port/man/man9/stat.html
+
+    // size[4] Tstat tag[2] fid[4]
+    // size[4] Rstat tag[2] stat[n]
+
+    // The reply will contain a machine-independent directory entry, stat, laid out as follows:
+    // size[2]total byte count of the following data
+    // type[2]for kernel use
+    // dev[4]for kernel use
+    // qid.type[1]
+    // 	the type of the file (directory, etc.), represented as a bit vector corresponding to the high 8 bits of the file’s mode word.
+    // qid.vers[4]
+    // 	version number for given path
+    // qid.path[8]
+    // 	the file server’s unique identification for the file
+    // mode[4]
+    // 	permissions and flags
+    // atime[4]
+    // 	last access time
+    // mtime[4]
+    // 	last modification time
+    // length[8]
+    // 	length of file in bytes
+    // name[ s ]
+    // 	file name; must be / if the file is the root directory of the server
+    // uid[ s ]owner name
+    // gid[ s ]group name
+    // muid[ s ]
+    // 	name of the user who last modified the file
+
+    SET_MSGS
+
+    U32 fid = *(data + (4 + 1 + 2))(U32*);
+
+    Rmsg->length = (4 + 1 + 2 + sizeof(@9p_stat)); // set initial Rmsg length
+    //@nt_tx_dbg(nt, "Tstat fid: %d", fid);
+    // PressAKey;
+
+    U8 fidbuf[16];
+    StrPrint(fidbuf, "%d", fid);
+
+    @9p_qid * qid = nt->fids->o(fidbuf)->@("qid");
+
+    @9p_stat * stat = nt->out + (4 + 1 + 2);
+    stat->size = sizeof(@9p_stat) - 2;
+    stat->type = 0;
+    stat->dev = 0;
+    MemCpy(&stat->qid, qid, sizeof(@9p_qid));
+    stat->atime = 0;
+    stat->mtime = 0;
+    stat->length = 1024;
+
+    // add strings
+    U8* p = NULL;
+    U8* user = "alec";
+
+    // name[ s ]
+    p = nt->out + Rmsg->length;
+    U8* path = qid->path;
+    if (StrCmp(path, "/")) {
+        path = StrLastOcc(path, "/") + 1;
+    }
+    *(p)(U16*) = StrLen(path);
+    p += 2;
+    MemCpy(p, path, StrLen(path));
+    p += StrLen(path);
+    stat->size += StrLen(path) + 2;
+    Rmsg->length += StrLen(path) + 2;
+
+    // uid[ s ]owner name
+    p = nt->out + Rmsg->length;
+    *(p)(U16*) = StrLen(user);
+    p += 2;
+    MemCpy(p, user, StrLen(user));
+    p += StrLen(user);
+    stat->size += StrLen(user) + 2;
+    Rmsg->length += StrLen(user) + 2;
+
+    // uid[ s ]group name
+    p = nt->out + Rmsg->length;
+    *(p)(U16*) = StrLen(user);
+    p += 2;
+    MemCpy(p, user, StrLen(user));
+    p += StrLen(user);
+    stat->size += StrLen(user) + 2;
+    Rmsg->length += StrLen(user) + 2;
+
+    // muid[ s ]
+    p = nt->out + Rmsg->length;
+    *(p)(U16*) = StrLen(user);
+    p += 2;
+    MemCpy(p, user, StrLen(user));
+    p += StrLen(user);
+    stat->size += StrLen(user) + 2;
+    Rmsg->length += StrLen(user) + 2;
+
+    SEND_RMSG
+}
\ No newline at end of file
diff --git a/System/Api/Dns.HC b/System/Api/Dns.HC
new file mode 100644
index 0000000..79b9985
--- /dev/null
+++ b/System/Api/Dns.HC
@@ -0,0 +1,25 @@
+#define DNS_REQUEST_PTR 0x300010
+
+MemSet(DNS_REQUEST_PTR, NULL, sizeof(U64));
+
+class DnsRequest {
+    U64 host;
+    U64 pointer_to_u32;
+};
+
+U32 @dns_query(U8* host)
+{
+    U32 res = 0;
+    if (!host)
+        return U32_MAX;
+    DnsRequest* request = CAlloc(sizeof(DnsRequest), Fs->code_heap);
+    request->host = StrNew(host, adam_task);
+    request->pointer_to_u32 = &res;
+    U64* request_ptr = DNS_REQUEST_PTR;
+    while (*request_ptr)
+        Sleep(1);
+    LXchgU32(request_ptr, request);
+    while (!res)
+        Sleep(1);
+    return res;
+}
diff --git a/System/Api/Icmp.HC b/System/Api/Icmp.HC
new file mode 100644
index 0000000..e52040c
--- /dev/null
+++ b/System/Api/Icmp.HC
@@ -0,0 +1,32 @@
+#define ICMP_REQUEST_PTR 0x300020
+
+MemSet(ICMP_REQUEST_PTR, NULL, sizeof(U64));
+
+class IcmpRequest {
+    U64 addr;
+    U64 iden;
+    U64 seq;
+    U64 pointer_to_u32;
+};
+
+U32 @icmp_echo_request(U32 addr, U16 iden, U16 seq, IcmpRequest* request, I64 count)
+{
+    U32 res = 0; // low 16 = ttl, hi 16 = payload size
+    request->addr = addr;
+    request->iden = iden;
+    request->seq = seq;
+    request->pointer_to_u32 = &res;
+    I64 start_jiffies = cnts.jiffies;
+    U64* request_ptr = ICMP_REQUEST_PTR;
+    if (!count)
+        *request_ptr = NULL;
+    while (*request_ptr) {
+        if (!(cnts.jiffies < start_jiffies + 1000))
+            return res;
+        Sleep(1);
+    }
+    LXchgU32(request_ptr, request);
+    while (!res && cnts.jiffies < start_jiffies + 1000)
+        Sleep(1);
+    return res;
+}
\ No newline at end of file
diff --git a/System/Api/Ipv4.HC b/System/Api/Ipv4.HC
new file mode 100644
index 0000000..105ebe3
--- /dev/null
+++ b/System/Api/Ipv4.HC
@@ -0,0 +1,9 @@
+U32 @ipv4_address(I64 o3, I64 o2, I64 o1, I64 o0)
+{
+    U32 addr = NULL;
+    addr.u8[3] = o3;
+    addr.u8[2] = o2;
+    addr.u8[1] = o1;
+    addr.u8[0] = o0;
+    return addr;
+}
diff --git a/System/Api/MD5.HC b/System/Api/MD5.HC
new file mode 100644
index 0000000..176cc93
--- /dev/null
+++ b/System/Api/MD5.HC
@@ -0,0 +1,144 @@
+/*
+ * Simple MD5 implementation
+ *
+ * https://gist.github.com/creationix/4710780
+ */
+
+U32 md5_r[64] = { 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22,
+    5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20,
+    4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23,
+    6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21 };
+// Use binary integer part of the sines of integers (in radians) as constants//
+// Initialize variables:
+U32 md5_k[64] = {
+    0xd76aa478, 0xe8c7b756, 0x242070db, 0xc1bdceee, 0xf57c0faf, 0x4787c62a,
+    0xa8304613, 0xfd469501, 0x698098d8, 0x8b44f7af, 0xffff5bb1, 0x895cd7be,
+    0x6b901122, 0xfd987193, 0xa679438e, 0x49b40821, 0xf61e2562, 0xc040b340,
+    0x265e5a51, 0xe9b6c7aa, 0xd62f105d, 0x02441453, 0xd8a1e681, 0xe7d3fbc8,
+    0x21e1cde6, 0xc33707d6, 0xf4d50d87, 0x455a14ed, 0xa9e3e905, 0xfcefa3f8,
+    0x676f02d9, 0x8d2a4c8a, 0xfffa3942, 0x8771f681, 0x6d9d6122, 0xfde5380c,
+    0xa4beea44, 0x4bdecfa9, 0xf6bb4b60, 0xbebfbc70, 0x289b7ec6, 0xeaa127fa,
+    0xd4ef3085, 0x04881d05, 0xd9d4d039, 0xe6db99e5, 0x1fa27cf8, 0xc4ac5665,
+    0xf4292244, 0x432aff97, 0xab9423a7, 0xfc93a039, 0x655b59c3, 0x8f0ccc92,
+    0xffeff47d, 0x85845dd1, 0x6fa87e4f, 0xfe2ce6e0, 0xa3014314, 0x4e0811a1,
+    0xf7537e82, 0xbd3af235, 0x2ad7d2bb, 0xeb86d391
+};
+
+// leftrotate function
+U32 LEFTROTATE(U32 x, U32 c) { return (((x) << (c)) | ((x) >> (32 - (c)))); }
+
+U0 md5(U8* initial_msg, U32 initial_len, U32* md5_h)
+{
+
+    // These vars will contain the hash
+    U32 md5_h0, md5_h1, md5_h2, md5_h3;
+
+    // Message (to prepare)
+    U8* msg = NULL;
+
+    // Note: All variables are unsigned 32 bit and wrap modulo 2^32 when
+    // calculating
+
+    // r specifies the per-round shift amounts
+
+    md5_h0 = 0x67452301;
+    md5_h1 = 0xefcdab89;
+    md5_h2 = 0x98badcfe;
+    md5_h3 = 0x10325476;
+
+    // Pre-processing: adding a single 1 bit
+    // append "1" bit to message
+    /* Notice: the input bytes are considered as bits strings,
+       where the first bit is the most significant bit of the byte.[37] */
+
+    // Pre-processing: padding with zeros
+    // append "0" bit until message length in bit ≡ 448 (mod 512)
+    // append length mod (2 pow 64) to message
+
+    U32 new_len;
+    for (new_len = initial_len * 8 + 1; new_len % 512 != 448; new_len++)
+        ;
+    new_len /= 8;
+
+    msg = CAlloc(new_len + 64, adam_task); // also appends "0" bits
+                                           // (we alloc also 64 extra bytes...)
+    MemCpy(msg, initial_msg, initial_len);
+    msg[initial_len] = 128; // write the "1" bit
+
+    U32 bits_len = 8 * initial_len;      // note, we append the len
+    MemCpy(msg + new_len, &bits_len, 4); // in bits at the end of the buffer
+
+    // Process the message in successive 512-bit chunks:
+    // for each 512-bit chunk of message:
+    U32 offset;
+    for (offset = 0; offset < new_len; offset += (512 / 8)) {
+
+        // break chunk into sixteen 32-bit words w[j], 0 ≤ j ≤ 15
+        U32* w = (msg + offset)(U32*);
+
+        // Initialize hash value for this chunk:
+        U32 a = md5_h0;
+        U32 b = md5_h1;
+        U32 c = md5_h2;
+        U32 d = md5_h3;
+
+        // Main loop:
+        U32 i;
+        for (i = 0; i < 64; i++) {
+
+            U32 f, g;
+
+            if (i < 16) {
+                f = (b & c) | ((~b) & d);
+                g = i;
+            } else if (i < 32) {
+                f = (d & b) | ((~d) & c);
+                g = (5 * i + 1) % 16;
+            } else if (i < 48) {
+                f = b ^ c ^ d;
+                g = (3 * i + 5) % 16;
+            } else {
+                f = c ^ (b | (~d));
+                g = (7 * i) % 16;
+            }
+
+            U32 temp = d;
+            d = c;
+            c = b;
+            // printf("rotateLeft(%x + %x + %x + %x, %d)\n", a, f, k[i], w[g], r[i]);
+            b = b + LEFTROTATE((a + f + md5_k[i] + w[g]), md5_r[i]);
+            a = temp;
+        }
+
+        // Add this chunk's hash to result so far:
+
+        md5_h0 += a;
+        md5_h1 += b;
+        md5_h2 += c;
+        md5_h3 += d;
+    }
+
+    md5_h[0] = md5_h0;
+    md5_h[1] = md5_h1;
+    md5_h[2] = md5_h2;
+    md5_h[3] = md5_h3;
+
+    // cleanup
+    Free(msg);
+}
+
+U8* md5_string(U8* buf, I64 size)
+{
+    U32 md5_h[4];
+    md5(buf, size, &md5_h[0]);
+    U8* str = CAlloc(33, adam_task);
+    StrPrint(str + StrLen(str), "%02x%02x%02x%02x", md5_h[0].u8[0],
+        md5_h[0].u8[1], md5_h[0].u8[2], md5_h[0].u8[3]);
+    StrPrint(str + StrLen(str), "%02x%02x%02x%02x", md5_h[1].u8[0],
+        md5_h[1].u8[1], md5_h[1].u8[2], md5_h[1].u8[3]);
+    StrPrint(str + StrLen(str), "%02x%02x%02x%02x", md5_h[2].u8[0],
+        md5_h[2].u8[1], md5_h[2].u8[2], md5_h[2].u8[3]);
+    StrPrint(str + StrLen(str), "%02x%02x%02x%02x", md5_h[3].u8[0],
+        md5_h[3].u8[1], md5_h[3].u8[2], md5_h[3].u8[3]);
+    return str;
+}
diff --git a/System/Api/NetInfo.HC b/System/Api/NetInfo.HC
new file mode 100644
index 0000000..79f0384
--- /dev/null
+++ b/System/Api/NetInfo.HC
@@ -0,0 +1,32 @@
+#define NETINFO_REQUEST_PTR 0x300030
+
+MemSet(NETINFO_REQUEST_PTR, NULL, sizeof(U64));
+
+class NetInfoRequest {
+    U64 mac_address;
+    U64 ipv4_address;
+    U64 ipv4_netmask;
+    U64 ipv4_network;
+    U64 ipv4_gateway;
+    U64 dns_server_address;
+    U64 dns_server_port;
+    U64 rx_bytes;
+    U64 rx_frames;
+    U64 tx_bytes;
+    U64 tx_frames;
+    U64 pointer_to_u32;
+};
+
+NetInfoRequest* @net_info_request()
+{
+    U32 res = 0;
+    NetInfoRequest* req = CAlloc(sizeof(NetInfoRequest), Fs->code_heap);
+    req->pointer_to_u32 = &res;
+    U64* request_ptr = NETINFO_REQUEST_PTR;
+    while (*request_ptr)
+        Sleep(1);
+    LXchgU32(request_ptr, req);
+    while (!res)
+        Sleep(1);
+    return req;
+}
diff --git a/System/Api/Tcp.HC b/System/Api/Tcp.HC
new file mode 100644
index 0000000..0f75622
--- /dev/null
+++ b/System/Api/Tcp.HC
@@ -0,0 +1,209 @@
+#define TCP_SOCKET_REQUEST_PTR 0x300000
+#define TCP_BIND_REQUEST_PTR 0x300040
+#define TCP_ACCEPT_REQUEST_PTR 0x300050
+
+MemSet(TCP_SOCKET_REQUEST_PTR, NULL, sizeof(U64));
+
+// TcpSocket states
+
+#define TCP_SOCKET_STATE_IDLE 0
+#define TCP_SOCKET_STATE_ESTABLISHED 1
+#define TCP_SOCKET_STATE_CLOSED 2
+#define TCP_SOCKET_STATE_CONNECTING 4
+
+class TcpSocket {
+    U64 remote_addr;
+    U64 remote_port;
+    U64 state;
+    U64 receive_buffer_ptr; // Pointer to receive buffer in physical memory
+    U64 receive_buffer_size;
+    U64 receive_buffer_filled; // Number of bytes Net has put into buffer
+    U64 receive_buffer_kick;   // Net sets this to 1 when it has data available for
+                               // us, we set back to 0 when ready to receive
+    U64 send_buffer_ptr;
+    U64 send_buffer_size;
+    U64 send_buffer_filled;
+    U64 send_buffer_kick; // We set this to 1 when we have data available to net,
+                          // Net sets back to 0 when ready to receive
+    U64 close_kick;       // We set this to 1 to force close the connection
+    U0 (*close)();
+    U64 (*receive)(U64 buf, U64 length);
+    U0 (*send)(U64 buf, U64 length);
+};
+
+class TcpBind {
+    U64 port;
+    U64 function;
+    U64 response_code;
+};
+
+U8 @tcp_close_wrapper_function[16]
+    = { 0x55, 0x48, 0x8B, 0xEC, 0x68, 0x78,
+          0x56, 0x34, 0x12, 0xE8, 0x02, 0x6D,
+          0x02, 0x00, 0x5D, 0xC3 };
+
+U8 @tcp_receive_wrapper_function[32] = {
+    0x55, 0x48, 0x8B, 0xEC, 0x56, 0x57, 0x48, 0x8B, 0x75, 0x18, 0x48,
+    0x8B, 0x7D, 0x10, 0x56, 0x57, 0x68, 0x78, 0x56, 0x34, 0x12, 0xE8,
+    0x5E, 0x62, 0x02, 0x00, 0x5F, 0x5E, 0x5D, 0xC2, 0x10, 0x00
+};
+
+U8 @tcp_send_wrapper_function[32] = {
+    0x55, 0x48, 0x8B, 0xEC, 0x56, 0x57, 0x48, 0x8B, 0x75, 0x18, 0x48,
+    0x8B, 0x7D, 0x10, 0x56, 0x57, 0x68, 0x78, 0x56, 0x34, 0x12, 0xE8,
+    0x5E, 0x62, 0x02, 0x00, 0x5F, 0x5E, 0x5D, 0xC2, 0x10, 0x00
+};
+
+U0 @tcp_socket_send(TcpSocket* s, U64 buf, U64 length)
+{
+    while (s->send_buffer_kick)
+        Sleep(1);
+    U64 pos = 0;
+    U64 bytes_to_send = 0;
+    while (pos < length) {
+        if ((length - pos) > s->send_buffer_size)
+            bytes_to_send = s->send_buffer_size;
+        else
+            bytes_to_send = length - pos;
+        MemCpy(s->send_buffer_ptr, buf + pos, bytes_to_send);
+        s->send_buffer_filled = bytes_to_send;
+        s->send_buffer_kick = 1;
+        pos += bytes_to_send;
+        while (s->send_buffer_kick)
+            Sleep(1);
+    }
+}
+
+U64 @tcp_socket_receive(TcpSocket* s, U64 buf, U64 size)
+{
+    s->receive_buffer_size = size;
+    s->receive_buffer_kick = 0;
+    while (!s->receive_buffer_kick) {
+        if (s->state == TCP_SOCKET_STATE_CLOSED)
+            return NULL;
+        Sleep(1);
+    }
+    U64 bytes_received = s->receive_buffer_filled;
+    if (bytes_received > 0) {
+        MemCpy(buf, s->receive_buffer_ptr, bytes_received);
+    }
+    return bytes_received;
+}
+
+U0 @tcp_wait_for_connection_established(TcpSocket* s)
+{
+    while (s->state != TCP_SOCKET_STATE_ESTABLISHED)
+        Sleep(1);
+}
+
+U0 @tcp_socket_close(TcpSocket* s)
+{
+    if (s->close)
+        Free(s->close);
+    if (s->receive)
+        Free(s->receive);
+    if (s->send)
+        Free(s->send);
+    s->close_kick = 1;
+    s->state = TCP_SOCKET_STATE_CLOSED;
+}
+
+TcpSocket* @tcp_socket_create(U8* host, U64 port)
+{
+    U64 addr = @dns_query(host);
+    TcpSocket* s = CAlloc(sizeof(TcpSocket), adam_task->code_heap);
+    s->remote_addr = addr;
+    s->remote_port = port;
+
+    U64 a;
+
+    s->close = MAlloc(16, adam_task->code_heap);
+    MemCpy(s->close, @tcp_close_wrapper_function, 16);
+    a = s->close;
+    a += 0x05;
+    MemSetU32(a, s, 1);
+    a = s->close;
+    a += 0x09;
+    @patch_call_rel32(a, &@tcp_socket_close);
+
+    s->receive = MAlloc(25, adam_task->code_heap);
+    MemCpy(s->receive, @tcp_receive_wrapper_function, 32);
+    a = s->receive;
+    a += 0x11;
+    MemSetU32(a, s, 1);
+    a = s->receive;
+    a += 0x15;
+    @patch_call_rel32(a, &@tcp_socket_receive);
+
+    s->send = MAlloc(32, adam_task->code_heap);
+    MemCpy(s->send, @tcp_send_wrapper_function, 32);
+    a = s->send;
+    a += 0x11;
+    MemSetU32(a, s, 1);
+    a = s->send;
+    a += 0x15;
+    @patch_call_rel32(a, &@tcp_socket_send);
+
+    U64* request_ptr = TCP_SOCKET_REQUEST_PTR;
+    while (*request_ptr)
+        Sleep(1);
+    LXchgU32(request_ptr, s);
+    return s;
+}
+
+U64 @tcp_socket_bind(U64 port, U64 function)
+{
+    if (!port || !function)
+        return NULL;
+
+    TcpBind* b = CAlloc(sizeof(TcpBind), adam_task->code_heap);
+    b->port = port;
+    b->function = function; // U0 my_spawn_wrapper_function(TcpSocket* s)
+
+    U64* request_ptr = TCP_BIND_REQUEST_PTR;
+    while (*request_ptr)
+        Sleep(1);
+    LXchgU32(request_ptr, b);
+    while (*request_ptr)
+        Sleep(1);
+    U64 res = b->response_code;
+    Free(b);
+    return res;
+}
+
+TcpSocket* @tcp_socket_accept(TcpSocket* s)
+{
+    if (!s || !s->remote_addr || !s->remote_port)
+        return NULL;
+
+    U64 a;
+
+    s->close = MAlloc(16, adam_task->code_heap);
+    MemCpy(s->close, @tcp_close_wrapper_function, 16);
+    a = s->close;
+    a += 0x05;
+    MemSetU32(a, s, 1);
+    a = s->close;
+    a += 0x09;
+    @patch_call_rel32(a, &@tcp_socket_close);
+
+    s->receive = MAlloc(25, adam_task->code_heap);
+    MemCpy(s->receive, @tcp_receive_wrapper_function, 32);
+    a = s->receive;
+    a += 0x11;
+    MemSetU32(a, s, 1);
+    a = s->receive;
+    a += 0x15;
+    @patch_call_rel32(a, &@tcp_socket_receive);
+
+    s->send = MAlloc(32, adam_task->code_heap);
+    MemCpy(s->send, @tcp_send_wrapper_function, 32);
+    a = s->send;
+    a += 0x11;
+    MemSetU32(a, s, 1);
+    a = s->send;
+    a += 0x15;
+    @patch_call_rel32(a, &@tcp_socket_send);
+
+    return s;
+}
diff --git a/System/Config/Net.json b/System/Config/Net.json
new file mode 100644
index 0000000..9217064
--- /dev/null
+++ b/System/Config/Net.json
@@ -0,0 +1,10 @@
+{
+    "tcpip.ipv4_address": "10.20.0.10",
+    "tcpip.ipv4_netmask": "255.255.255.0",
+    "tcpip.ipv4_network": "10.20.0.0",
+    "tcpip.ipv4_gateway": "10.20.0.254",
+    "tcpip.ipv4_dns_server_address": "8.8.8.8",
+    "tcpip.ipv4_dns_server_port": "53",
+    "tcpip.mss_size": "1360",
+    "eof": "eof"
+}
\ No newline at end of file
diff --git a/System/FFI/Base.HC b/System/FFI/Base.HC
new file mode 100644
index 0000000..e013399
--- /dev/null
+++ b/System/FFI/Base.HC
@@ -0,0 +1,51 @@
+#define PUSH_SYSV_REGS \
+    asm {PUSH RCX PUSH RDX PUSH RBX PUSH RBP PUSH RSI PUSH RDI PUSH R8 PUSH R9 PUSH   \
+      R10 PUSH R11 PUSH R12 PUSH R13 PUSH R14 PUSH R15}
+#define POP_SYSV_REGS \
+    p0 = p0;          \
+    p1 = p1;          \
+    p2 = p2;          \
+    p3 = p3;          \
+    p4 = p4;          \
+    p5 = p5;          \
+    asm {POP R15 POP R14 POP R13 POP R12 POP R11 POP R10 POP R9 POP R8 POP RDI POP    \
+      RSI POP RBP POP RBX POP RDX POP RCX}
+#define GET_SYSV_ARGS                              \
+    asm {PUSH R9 PUSH R8 PUSH RCX PUSH RDX PUSH RSI PUSH RDI} \
+    I64 reg RDI p0;                                \
+    I64 reg RSI p1;                                \
+    I64 reg RDX p2;                                \
+    I64 reg RCX p3;                                \
+    I64 reg R8 p4;                                 \
+    I64 reg R9 p5;                                 \
+    asm {POP RDI POP RSI POP RDX POP RCX POP R8 POP R9}
+
+#define MOV_ANS_RAX asm { MOV[&ans], RAX }
+#define MOV_PARAM0_RDI asm {MOV [&param0], RDI}
+
+I64 param0;
+I64 elf_argc;
+U8** elf_argv;
+
+asm {
+_ELF_CALL::
+        PUSH    RBP
+        MOV     RBP,RSP
+        MOV     RAX,U64 SF_ARG1[RBP]
+        MOV     RDI,U64 SF_ARG2[RBP]
+        MOV     RSI,U64 SF_ARG3[RBP]
+        TEST    RAX,RAX
+        JZ      @@05
+        CALL    RAX
+@@05:   POP     RBP
+        RET1    8
+}
+
+U0 _main()
+{
+    MOV_PARAM0_RDI
+    CallInd(_ELF_CALL, param0, elf_argc, elf_argv);
+    UserTaskCont;
+}
+
+U0 _exit() { UserTaskCont; }
diff --git a/System/FFI/ELF64.HC b/System/FFI/ELF64.HC
new file mode 100644
index 0000000..b444cf7
--- /dev/null
+++ b/System/FFI/ELF64.HC
@@ -0,0 +1,300 @@
+#define EI_NIDENT 16
+#define EM_X86_64 0x3E
+#define ET_EXEC 2
+#define ET_DYN 3
+
+U0 @elf64_debug_print(U8 fmt, ...)
+{
+    // FIXME: Remove unnecessary debug_print statements and PrintErr for errors.
+    no_warn fmt, argc, argv;
+}
+
+class Elf64_Ehdr {
+    U8 e_ident[EI_NIDENT]; /* Magic number and other info */
+    U16 e_type;            /* Object file type */
+    U16 e_machine;         /* Architecture */
+    U32 e_version;         /* Object file version */
+    U64 e_entry;           /* Entry point virtual address */
+    U64 e_phoff;           /* Program header table file offset */
+    U64 e_shoff;           /* Section header table file offset */
+    U32 e_flags;           /* Processor-specific flags */
+    U16 e_ehsize;          /* ELF header size in bytes */
+    U16 e_phentsize;       /* Program header table entry size */
+    U16 e_phnum;           /* Program header table entry count */
+    U16 e_shentsize;       /* Section header table entry size */
+    U16 e_shnum;           /* Section header table entry count */
+    U16 e_shstrndx;        /* Section header string table index */
+};
+
+class Elf64_Shdr {
+    U32 sh_name;      /* Section name (string tbl index) */
+    U32 sh_type;      /* Section type */
+    U64 sh_flags;     /* Section flags */
+    U64 sh_addr;      /* Section virtual addr at execution */
+    U64 sh_offset;    /* Section file offset */
+    U64 sh_size;      /* Section size in bytes */
+    U32 sh_link;      /* Link to another section */
+    U32 sh_info;      /* Additional section information */
+    U64 sh_addralign; /* Section alignment */
+    U64 sh_entsize;   /* Entry size if section holds table */
+};
+
+class Elf64_Sym {
+    U32 st_name;  /* Symbol name (string tbl index) */
+    U8 st_info;   /* Symbol type and binding */
+    U8 st_other;  /* Symbol visibility */
+    U16 st_shndx; /* Section index */
+    U64 st_value; /* Symbol value */
+    U64 st_size;  /* Symbol size */
+};
+
+class PLT_entry {
+    U8 pad[0x10];
+};
+
+class RELA_entry {
+    U64 r_offset;
+    U64 r_info;
+    I64 r_addend;
+};
+
+class Elf {
+    union {
+        U8* u8;
+        Elf64_Ehdr* ehdr;
+    } I64 size;
+    U8* dynstr;
+    Elf64_Sym* dynsym;
+    PLT_entry* plt;
+    RELA_entry* rela_dyn;
+    RELA_entry* rela_plt;
+    Elf64_Sym* strtab;
+    Elf64_Sym* symtab;
+    I64 rela_dyn_size;
+    I64 rela_plt_size;
+    I64 strtab_size;
+    I64 symtab_size;
+};
+
+U0 (*_start)();
+
+U0 unimplemented_symbol()
+{
+    I32 s = 0xDEADF00D;
+    PrintWarn("Unimplemented symbol: %s\n", s);
+    Dbg;
+    while (1)
+        Sleep(1);
+}
+
+Bool is_valid_elf(Elf* elf)
+{
+    Bool res = TRUE;
+    if (MemCmp(elf->u8 + 1, "ELF", 3)) {
+        @elf64_debug_print("Invalid signature (not ELF).\n");
+        res = FALSE;
+    }
+    if (elf->ehdr->e_type != ET_EXEC && elf->ehdr->e_type != ET_DYN) {
+        @elf64_debug_print("Invalid object file type.\n");
+        res = FALSE;
+    }
+    if (elf->ehdr->e_machine != EM_X86_64) {
+        @elf64_debug_print("Invalid architecture.\n");
+        res = FALSE;
+    }
+    return res;
+}
+
+U0 process_elf_section_header_table(Elf* elf)
+{
+    Elf64_Shdr* shdr = elf->u8 + elf->ehdr->e_shoff;
+    Elf64_Shdr* shdr_shstrtab = shdr + elf->ehdr->e_shstrndx;
+    U8* shstrtab = elf->u8 + shdr_shstrtab->sh_offset;
+    I64 i = 0;
+    while (i < elf->ehdr->e_shnum) {
+        if (!StrCmp(shstrtab + shdr->sh_name, ".symtab")) {
+            @elf64_debug_print("found symtab at 0x%08x, size = %d\n", shdr->sh_offset,
+                shdr->sh_size);
+            elf->symtab = elf->u8 + shdr->sh_offset;
+            elf->symtab_size = shdr->sh_size;
+        }
+        if (!StrCmp(shstrtab + shdr->sh_name, ".strtab")) {
+            @elf64_debug_print("found strtab at 0x%08x, size = %d\n", shdr->sh_offset,
+                shdr->sh_size);
+            elf->strtab = elf->u8 + shdr->sh_offset;
+            elf->strtab_size = shdr->sh_size;
+        }
+        if (shdr->sh_addr) {
+            MemCpy(shdr->sh_addr, elf->u8 + shdr->sh_offset, shdr->sh_size);
+            if (!StrCmp(shstrtab + shdr->sh_name, ".dynstr"))
+                elf->dynstr = shdr->sh_addr;
+            if (!StrCmp(shstrtab + shdr->sh_name, ".dynsym"))
+                elf->dynsym = shdr->sh_addr;
+            if (!StrCmp(shstrtab + shdr->sh_name, ".plt"))
+                elf->plt = shdr->sh_addr;
+            if (!StrCmp(shstrtab + shdr->sh_name, ".rela.dyn")) {
+                elf->rela_dyn = shdr->sh_addr;
+                elf->rela_dyn_size = shdr->sh_size / shdr->sh_entsize;
+            }
+            if (!StrCmp(shstrtab + shdr->sh_name, ".rela.plt")) {
+                elf->rela_plt = shdr->sh_addr;
+                elf->rela_plt_size = shdr->sh_size / shdr->sh_entsize;
+            }
+            if (!StrCmp(shstrtab + shdr->sh_name, ".bss") || !StrCmp(shstrtab + shdr->sh_name, ".tbss")) {
+                MemSet(shdr->sh_addr, NULL, shdr->sh_size);
+                @elf64_debug_print(
+                    "Zeroed out section '%s' at physical address 0x%06x, size = %d bytes\n",
+                    shstrtab + shdr->sh_name, shdr->sh_addr, shdr->sh_size);
+            } else
+                @elf64_debug_print(
+                    "MemCpy section '%s' to physical address 0x%06x, size = %d bytes\n",
+                    shstrtab + shdr->sh_name, shdr->sh_addr, shdr->sh_size);
+            if (!StrCmp(shstrtab + shdr->sh_name, ".bss")) {
+                MemSet(shdr->sh_addr, NULL, shdr->sh_size);
+                @elf64_debug_print("MemSet section '%s' at physical address 0x%06x to NULL, "
+                                   "size = %d bytes\n",
+                    shstrtab + shdr->sh_name, shdr->sh_addr, shdr->sh_size);
+            }
+        }
+        shdr++;
+        i++;
+    }
+}
+
+U0 process_elf_rela_dyn_entries(Elf* elf)
+{
+    I64 i;
+    U8* entry_name;
+    RELA_entry* rela_dyn = elf->rela_dyn;
+    for (i = 0; i < elf->rela_dyn_size; i++) {
+        entry_name = elf->dynstr + elf->dynsym[(rela_dyn->r_info >> 32)].st_name;
+        @elf64_debug_print("rela_dyn->r_offset = %08x\n", rela_dyn->r_offset);
+        @elf64_debug_print("entry name = '%s'\n", entry_name);
+        if (!StrCmp(entry_name, "__libc_start_main")) {
+            *(rela_dyn->r_offset)(U64*) = &_main;
+            @elf64_debug_print("Set value for .rela.dyn entry '%s' to: &_main\n",
+                entry_name);
+        }
+        if (!StrCmp(entry_name, "stdin")) {
+            *(rela_dyn->r_offset)(U64*) = 0;
+            @elf64_debug_print("Set value for .rela.dyn entry '%s' to: %d\n", entry_name, 0);
+        }
+        if (!StrCmp(entry_name, "stdout")) {
+            *(rela_dyn->r_offset)(U64*) = 1;
+            @elf64_debug_print("Set value for .rela.dyn entry '%s' to: %d\n", entry_name, 1);
+        }
+        if (!StrCmp(entry_name, "stderr")) {
+            *(rela_dyn->r_offset)(U64*) = 2;
+            @elf64_debug_print("Set value for .rela.dyn entry '%s' to: %d\n", entry_name, 2);
+        }
+        rela_dyn++;
+    }
+}
+
+CHashClass* get_symbol_hash_entry(U8* entry_name)
+{
+    I64 i;
+    CHashSrcSym* sym;
+    CHashTable* tbl = Fs->hash_table;
+    while (tbl) {
+        for (i = 0; i < tbl->mask; i++) {
+            sym = tbl->body[i];
+            while (sym) {
+                if (sym->type == HTT_CLASS)
+                    if (!StrCmp(sym->str, entry_name))
+                        return sym;
+                sym = sym->next;
+            }
+        }
+        tbl = tbl->next;
+    }
+    return NULL;
+}
+
+U64 get_symbol_address(U8* entry_name)
+{
+    CHash* h = HashFind(entry_name, Fs->hash_table, Fs->hash_table->mask);
+    if (!h)
+        return NULL;
+    switch (h->type) {
+    case HTT_GLBL_VAR:
+        return h(CHashGlblVar*)->data_addr;
+        break;
+    case HTT_FUN:
+        return h(CHashFun*)->exe_addr;
+        break;
+    default:
+        return NULL;
+        break;
+    }
+    return NULL;
+}
+
+U0 process_elf_rela_plt_entries(Elf* elf)
+{
+    I64 i;
+    U32 handler;
+    U32* patch;
+    U8* entry_name;
+    Bool symbol_exists;
+    PLT_entry* plt = elf->plt;
+    RELA_entry* rela_plt = elf->rela_plt;
+    plt++;
+    for (i = 0; i < elf->rela_plt_size; i++) {
+        symbol_exists = FALSE;
+        entry_name = elf->dynstr + elf->dynsym[(rela_plt->r_info >> 32)].st_name;
+        handler = MAlloc(sizeof(unimplemented_symbol), adam_task->code_heap);
+        MemCpy(handler, &unimplemented_symbol, sizeof(unimplemented_symbol));
+        patch = handler + 0x0A;
+        *patch = entry_name;
+        @patch_jmp_rel32(plt, handler);
+        @patch_call_rel32(handler + 0x16, &PrintErr);
+        //@patch_call_rel32(handler + 0x21, &_exit);
+        if (!StrCmp(entry_name, "__libc_start_main")) {
+            symbol_exists = TRUE;
+            @patch_jmp_rel32(plt, &_main);
+            @elf64_debug_print("Set value for .rela.plt entry '%s' to &_main\n", entry_name);
+        }
+        if (get_symbol_address(entry_name)) {
+            symbol_exists = TRUE;
+            @patch_jmp_rel32(plt, get_symbol_address(entry_name));
+            @elf64_debug_print("Set value for .rela.plt entry '%s' to &%s\n", entry_name,
+                entry_name);
+        }
+        if (!symbol_exists)
+            @elf64_debug_print(
+                "Set value for .rela.plt entry '%s' to &unimplemented_symbol\n",
+                entry_name);
+        rela_plt++;
+        plt++;
+    }
+}
+
+U0 load_elf(...)
+{
+    if (argc < 1) {
+        PrintErr("Not enough arguments.\n");
+        return;
+    }
+    if (!FileFind(argv[0])) {
+        PrintErr("File not found: %s\n", argv[0]);
+        return;
+    }
+
+    Elf elf;
+    elf.u8 = FileRead(argv[0], &elf.size);
+    @elf64_debug_print("Load file '%s', size = %d bytes\n", argv[0], elf.size);
+
+    if (!is_valid_elf(&elf)) {
+        PrintErr("File is not a valid ELF x86-64 executable.\n");
+        return;
+    }
+
+    process_elf_section_header_table(&elf);
+    process_elf_rela_dyn_entries(&elf);
+    process_elf_rela_plt_entries(&elf);
+
+    _start = elf.ehdr->e_entry;
+    elf_argc = argc;
+    elf_argv = argv;
+}
\ No newline at end of file
diff --git a/System/FFI/LibC.HC b/System/FFI/LibC.HC
new file mode 100644
index 0000000..e6979ab
--- /dev/null
+++ b/System/FFI/LibC.HC
@@ -0,0 +1,327 @@
+#define stdin 0
+#define stdout 1
+#define stderr 2
+
+U0 bcmp()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    MemCmp(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U64 @calloc(I64 size)
+{
+    return CAlloc(size, adam_task->code_heap);
+}
+
+U0 calloc()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @calloc(p0 * p1);
+    POP_SYSV_REGS
+}
+
+U0 free()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Free(p0);
+    POP_SYSV_REGS
+}
+
+I64 @isatty()
+{
+    return 0;
+}
+
+U0 isatty()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    // Dbg;
+    @isatty;
+    POP_SYSV_REGS
+}
+
+I64 @fwrite(U8* ptr, I64 size, I64 nmemb, U64 stream)
+{
+    U8* tmp;
+    switch (stream) {
+    case stdout:
+    case stderr:
+        tmp = CAlloc((size * nmemb) + 1, adam_task->code_heap);
+        MemCpy(tmp, ptr, (size * nmemb));
+#ifdef QEMU_RUN_TESTS
+        QemuDebugMsg(tmp);
+#endif
+        DocPutS(adam_task->put_doc, tmp);
+        Free(tmp);
+        //        if (!MemCmp(tmp, "VERIFICATION FAILED", 19))
+        //            Break;
+        break;
+    default:
+        break;
+    }
+    return size * nmemb;
+}
+
+U0 fwrite()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @fwrite(p0, p1, p2, p3);
+    POP_SYSV_REGS
+}
+
+U64 @getentropy(U8* buffer, U64 length)
+{
+    I64 i;
+    for (i = 0; i < length; i++)
+        buffer[i] = RandU64;
+    return 0;
+}
+
+U0 getentropy()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @getentropy(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 htonl()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    EndianU32(p0);
+    POP_SYSV_REGS
+}
+
+U0 ntohl()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    EndianU32(p0);
+    POP_SYSV_REGS
+}
+
+U0 htons()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    EndianU16(p0);
+    POP_SYSV_REGS
+}
+
+U0 ntohs()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    EndianU16(p0);
+    POP_SYSV_REGS
+}
+
+U64 @malloc(I64 size)
+{
+    return MAlloc(size, adam_task->code_heap);
+}
+
+U0 malloc()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @malloc(p0);
+    POP_SYSV_REGS
+}
+
+U0 memcmp()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    MemCmp(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 memcpy()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    MemCpy(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U8* @memmove(U8* dest, U8* src, I64 n)
+{
+    I64 i;
+    U8* from = src;
+    U8* to = dest;
+    if (from == to || n == 0)
+        return dest;
+    if (to > from && to - from < n) {
+        /* to overlaps with from */
+        /*  <from......>         */
+        /*         <to........>  */
+        /* copy in reverse, to avoid overwriting from */
+        for (i = n - 1; i >= 0; i--)
+            to[i] = from[i];
+        return dest;
+    }
+    if (from > to && from - to < n) {
+        /* to overlaps with from */
+        /*        <from......>   */
+        /*  <to........>         */
+        /* copy forwards, to avoid overwriting from */
+        for (i = 0; i < n; i++)
+            to[i] = from[i];
+        return dest;
+    }
+    MemCpy(dest, src, n);
+    return dest;
+}
+
+U0 memmove()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @memmove(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 memset()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    MemSet(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 putc()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PutChars(p0);
+    POP_SYSV_REGS
+}
+
+U0 rand()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    RandU64;
+    POP_SYSV_REGS
+}
+
+U8* @realloc(U8* ptr, I64 size)
+{
+    U8* new;
+    if (!ptr) {
+        new = MAlloc(size, adam_task->code_heap);
+    } else {
+        new = MAlloc(size, adam_task->code_heap);
+        MemCpy(new, ptr, size);
+        Free(ptr);
+    }
+    return new;
+}
+
+U0 realloc()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @realloc(p0, p1);
+    POP_SYSV_REGS
+}
+
+// FIXME: It is non-obvious how to take a [u8] and convert it to a
+//        formatted string in Jakt, so we have to do this hack for
+//        now. Hopefully, this will change soon.
+U0 sprintf()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    StrPrint(p0, p1, p2, p3, p4, p5);
+    POP_SYSV_REGS
+}
+
+I64 @strncmp(U8* s1, U8* s2, I32 n)
+{
+    U64 u1, u2;
+
+    while (n-- > 0) {
+        u1 = *s1++;
+        u2 = *s2++;
+        u1 = u1 & 0xff;
+        u2 = u2 & 0xff;
+        if (u1 != u2)
+            return u1 - u2;
+        if (u1 == '\0')
+            return 0;
+    }
+    return 0;
+}
+
+U0 strncmp()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @strncmp(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 strcmp()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    StrCmp(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 strlen()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    StrLen(p0);
+    POP_SYSV_REGS
+}
+
+I64 tos_nist_offset = 5020;
+#define NIST_TIME_OFFSET (tos_nist_offset - local_time_offset / CDATE_FREQ)
+
+public
+I64 CDate2Unix(CDate dt)
+{ // TempleOS datetime to Unix timestamp.
+    return ToI64((dt - Str2Date("1/1/1970")) / CDATE_FREQ + NIST_TIME_OFFSET);
+}
+
+I64 @time(I64* ptr)
+{
+    no_warn ptr;
+    return CDate2Unix(Now);
+}
+
+U0 time()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @time(p0);
+    POP_SYSV_REGS
+}
+
+U0 toupper()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    ToUpper(p0);
+    POP_SYSV_REGS
+}
+
+U0 __assert_fail()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    "%s:%d: %s: %s\n", p1, p2, p3, p0;
+    Break;
+    POP_SYSV_REGS
+}
diff --git a/System/FFI/New.HC b/System/FFI/New.HC
new file mode 100644
index 0000000..e8cdd42
--- /dev/null
+++ b/System/FFI/New.HC
@@ -0,0 +1,35 @@
+U0 _ZdlPv()
+{
+    // operator delete(void*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Free(p0);
+    POP_SYSV_REGS
+}
+
+U0 _ZdlPvm()
+{
+    // operator delete(void*, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Free(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Znwm()
+{
+    // operator new(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    MAlloc(p0, adam_task);
+    POP_SYSV_REGS
+}
+
+U0 _ZnwmRKSt9nothrow_t()
+{
+    // operator new(unsigned long, std::nothrow_t const&)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    MAlloc(p0, adam_task);
+    POP_SYSV_REGS
+}
diff --git a/System/Jakt/DC.HC b/System/Jakt/DC.HC
new file mode 100644
index 0000000..ea20b52
--- /dev/null
+++ b/System/Jakt/DC.HC
@@ -0,0 +1,288 @@
+U0 _Z8dc_aliasm()
+{
+    // dc_alias(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    DCAlias(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z7dc_blotmmmm()
+{
+    // dc_blot(unsigned long, unsigned long, unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GrBlot(p0, p1, p2, p3);
+    POP_SYSV_REGS
+}
+
+U8* @dc_buffer(CDC* dc) { return dc->body; }
+
+U0 _Z9dc_bufferm()
+{
+    // dc_buffer(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_buffer(p0);
+    POP_SYSV_REGS
+}
+
+I64 @dc_color(CDC* dc) { return dc->color; }
+
+U0 _Z8dc_colorm()
+{
+    // dc_color(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_color(p0);
+    POP_SYSV_REGS
+}
+
+U0 @dc_copy(CDC* dest, I64 x, I64 y, CDC* src)
+{
+
+    // If position is off-screen, return
+    if (x > dest->width - 1 || y > dest->height - 1)
+        return;
+
+    // If device context dimensions match, MemCpy and return
+    if (dest->width_internal == src->width_internal && dest->height == src->height) {
+        MemCpy(dest->body, src->body, dest->width_internal * dest->height);
+        return;
+    }
+
+    CDC* dc1 = DCAlias(dest);
+    CDC* dc2 = DCAlias(src);
+
+    I64 src_line = 0;
+    I64 src_row = 0;
+    I64 clip_y = 0;
+
+    // Handle horizontal clipping left
+    while (x < 0) {
+        dc2->x0++;
+        x++;
+    }
+
+    // Handle vertical clipping top
+    while (y < 0) {
+        dc2->body += src->width_internal;
+        dc2->y0++;
+        y++;
+    }
+
+    // default, clip line to copy as width-left off screen
+    src_line = src->width - dc2->x0;
+
+    if (-dc2->x0 + x + src->width >= dest->width) {
+        src_line -= ((-dc2->x0 + x + src->width) - dest->width);
+    }
+
+    dc2->body += dc2->x0;
+    clip_y = y;
+
+    while (src_row < (src->height - dc2->y0) && clip_y < dest->height) {
+        MemCpy(dc1->body + (y * dest->width) + x, dc2->body, src_line);
+        dc2->body += src->width_internal;
+        dc1->body += dest->width_internal;
+        clip_y++;
+        src_row++;
+    }
+
+    Free(dc2);
+    Free(dc1);
+}
+
+U0 _Z7dc_copymmmm()
+{
+    // dc_copy(unsigned long, unsigned long, unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_copy(p0, p1, p2, p3);
+    POP_SYSV_REGS
+}
+
+U0 _Z10dc_destroym()
+{
+    // dc_destroy(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    DCDel(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z14dc_draw_circlemlll()
+{
+    // dc_draw_circle(unsigned long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GrCircle3(p0, p1, p2, 0, p3);
+    POP_SYSV_REGS
+}
+
+U0 _Z19dc_draw_filled_rectmllll()
+{
+    // dc_draw_filled_rect(unsigned long, long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GrRect(p0, p1, p2, p3, p4);
+    POP_SYSV_REGS
+}
+
+U0 _Z12dc_draw_linemllll()
+{
+    // dc_draw_line(unsigned long, long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GrLine3(p0, p1, p2, 0, p3, p4, 0);
+    POP_SYSV_REGS
+}
+
+U0 _Z13dc_draw_pixelmll()
+{
+    // dc_draw_pixel(unsigned long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GrPlot(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 _Z7dc_fillmm()
+{
+    // dc_fill(unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    DCFill(p0, p1);
+    POP_SYSV_REGS
+}
+
+CDC* @dc_gr_dc() { return gr.dc; }
+
+U0 _Z8dc_gr_dcv()
+{
+    // dc_gr_dc()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_gr_dc();
+    POP_SYSV_REGS
+}
+
+I64 @dc_height(CDC* dc) { return dc->height; }
+
+U0 _Z9dc_heightm()
+{
+    // dc_height(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_height(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z17dc_load_from_filePKc()
+{
+    // dc_load_from_file(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GRRead(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z6dc_newmm()
+{
+    // dc_new(unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    DCNew(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 _Z11dc_pixel_atmll()
+{
+    // dc_pixel_at(unsigned long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GrPeek(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 _Z16dc_replace_colormmm()
+{
+    // dc_replace_color(unsigned long, unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    DCColorChg(p0, p1, p2);
+    POP_SYSV_REGS
+}
+
+U0 _Z13dc_screenshotv()
+{
+    // dc_screenshot()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    DCScrnCapture(1);
+    POP_SYSV_REGS
+}
+
+U0 _Z15dc_save_to_filePKcm()
+{
+    // dc_save_to_file(char const*, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GRWrite(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 @dc_set_color(CDC* dc, I64 color) { dc->color = color; }
+
+U0 _Z12dc_set_colorml()
+{
+    // dc_set_color(unsigned long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_set_color(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 @dc_set_thickness(CDC* dc, I64 thickness) { dc->thick = thickness; }
+
+U0 _Z16dc_set_thicknessml()
+{
+    // dc_set_thickness(unsigned long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_set_thickness(p0, p1);
+    POP_SYSV_REGS
+}
+
+I64 @dc_thickness(CDC* dc) { return dc->thick; }
+
+U0 _Z12dc_thicknessm()
+{
+    // dc_thickness(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_thickness(p0);
+    POP_SYSV_REGS
+}
+
+I64 @dc_width(CDC* dc) { return dc->width; }
+
+U0 _Z8dc_widthm()
+{
+    // dc_width(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_width(p0);
+    POP_SYSV_REGS
+}
+
+I64 @dc_width_internal(CDC* dc) { return dc->width_internal; }
+
+U0 _Z17dc_width_internalm()
+{
+    // dc_width_internal(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @dc_width_internal(p0);
+    POP_SYSV_REGS
+}
\ No newline at end of file
diff --git a/System/Jakt/IOPort.HC b/System/Jakt/IOPort.HC
new file mode 100644
index 0000000..e192c14
--- /dev/null
+++ b/System/Jakt/IOPort.HC
@@ -0,0 +1,53 @@
+U0 _Z14ioport_read_u8t()
+{
+    // ioport_read_u8(unsigned short)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    InU8(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z15ioport_read_u16t()
+{
+    // ioport_read_u16(unsigned short)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    InU16(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z15ioport_read_u32t()
+{
+    // ioport_read_u32(unsigned short)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    InU32(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z15ioport_write_u8th()
+{
+    // ioport_write_u8(unsigned short, unsigned char)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    OutU8(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 _Z16ioport_write_u16tt()
+{
+    // ioport_write_u16(unsigned short, unsigned short)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    OutU16(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 _Z16ioport_write_u32tj()
+{
+    // ioport_write_u32(unsigned short, unsigned int)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    OutU32(p0, p1);
+    POP_SYSV_REGS
+}
diff --git a/System/Jakt/Input.HC b/System/Jakt/Input.HC
new file mode 100644
index 0000000..b427a40
--- /dev/null
+++ b/System/Jakt/Input.HC
@@ -0,0 +1,72 @@
+U0 _Z16input_get_stringPKc()
+{
+    // input_get_string(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    GetStr(p0);
+    POP_SYSV_REGS
+}
+
+Bool @input_key_down(U8 scancode) { return Bt(kbd.down_bitmap, scancode); }
+
+U0 _Z14input_key_downh()
+{
+    // input_key_down(unsigned char)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @input_key_down(p0);
+    POP_SYSV_REGS
+}
+
+Bool @input_mouse_left() { return ms.lb; }
+
+U0 _Z16input_mouse_leftv()
+{
+    // input_mouse_left()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @input_mouse_left();
+    POP_SYSV_REGS
+}
+
+Bool @input_mouse_right() { return ms.rb; }
+
+U0 _Z17input_mouse_rightv()
+{
+    // input_mouse_right()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @input_mouse_right();
+    POP_SYSV_REGS
+}
+
+I64 @input_mouse_x() { return ms.pos.x; }
+
+U0 _Z13input_mouse_xv()
+{
+    // input_mouse_x()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @input_mouse_x();
+    POP_SYSV_REGS
+}
+
+I64 @input_mouse_y() { return ms.pos.y; }
+
+U0 _Z13input_mouse_yv()
+{
+    // input_mouse_y()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @input_mouse_y();
+    POP_SYSV_REGS
+}
+
+U0 _Z17input_press_a_keyv()
+{
+    // input_press_a_key()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PressAKey;
+    POP_SYSV_REGS
+}
\ No newline at end of file
diff --git a/System/Jakt/OS.HC b/System/Jakt/OS.HC
new file mode 100644
index 0000000..d0b7680
--- /dev/null
+++ b/System/Jakt/OS.HC
@@ -0,0 +1,282 @@
+U0 _Z8os_blinkPKc()
+{
+    // os_blink(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    F64 frequency = Str2F64(p0);
+    Print("called os_blink(%.1f)\n", frequency);
+    Blink(frequency);
+    POP_SYSV_REGS
+}
+
+U64 @os_call(U64 addr, U64 val)
+{
+    if (!addr || !val)
+        return NULL;
+    U64 (*func)(U64 arg);
+    func = addr;
+    return func(val);
+}
+
+U0 _Z7os_callmm()
+{
+    // os_call(unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @os_call(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 _Z16os_device_callocj()
+{
+    // os_device_calloc(unsigned int)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    CAllocAligned(p0, 4096, adam_task->code_heap);
+    POP_SYSV_REGS
+}
+
+U0 _Z7os_exitv()
+{
+    // os_exit()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    UserTaskCont;
+    POP_SYSV_REGS
+}
+
+U8* @os_file_picker(U8* path, U8* glob)
+{
+    U8* full_path = CAlloc(StrLen(path) + StrLen(glob) + 4, adam_task);
+    CatPrint(full_path, "%s/%s", path, glob);
+
+    CDirEntry* de = FilesFind(full_path);
+    Free(full_path);
+
+    CDirEntry* tmpde;
+    U8* file_list = NULL;
+    U8* selected_file = NULL;
+    I64 list_pos = 0;
+    I64 list_size = 0;
+
+    tmpde = de;
+    while (tmpde) {
+        list_size += StrLen(tmpde->name) + 2;
+        tmpde = tmpde->next;
+    }
+
+    file_list = CAlloc(list_size, adam_task);
+
+    tmpde = de;
+    while (tmpde) {
+        StrCpy(file_list + list_pos, tmpde->name);
+        list_pos += StrLen(tmpde->name) + 1;
+        tmpde = tmpde->next;
+    }
+
+    I64 list_index = Adam("PopUpPickLst(0x%08x);\n", file_list);
+    Free(file_list);
+    list_pos = 0;
+
+    if (list_index < 0) {
+        DirTreeDel(de);
+        return StrNew("", adam_task);
+    }
+
+    tmpde = de;
+    while (tmpde) {
+        if (list_index == list_pos) {
+            selected_file = CAlloc(StrLen(path) + StrLen(tmpde->name) + 4, adam_task);
+            CatPrint(selected_file, "%s/%s", path, tmpde->name);
+            break;
+        }
+        StrCpy(file_list + list_pos, tmpde->name);
+        list_pos++;
+        tmpde = tmpde->next;
+    }
+
+    DirTreeDel(de);
+    return selected_file;
+}
+
+U0 _Z14os_file_pickerPKcS0_()
+{
+    // os_file_picker(char const*, char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @os_file_picker(p0, p1);
+    POP_SYSV_REGS
+}
+
+U8* @os_files_list(U8* path)
+{
+    U8* full_path = CAlloc(StrLen(path) + 4, adam_task);
+    CatPrint(full_path, "%s", path);
+
+    CDirEntry* de = FilesFind(full_path);
+    Free(full_path);
+
+    CDateStruct ds;
+    CDirEntry* tmpde;
+    U8* file_list = NULL;
+    I64 list_size = 0;
+
+    tmpde = de;
+    while (tmpde) {
+        list_size += StrLen(tmpde->name) + 48; // Should be enough for filename, date,
+                                               // filesize + semicolon separators
+        tmpde = tmpde->next;
+    }
+
+    if (!list_size)
+        return NULL;
+
+    file_list = CAlloc(list_size, adam_task);
+
+    tmpde = de;
+    I64 counter = 0;
+
+    while (tmpde) {
+        if (counter > 0) {
+            StrCpy(file_list + StrLen(file_list), "|");
+        }
+        StrCpy(file_list + StrLen(file_list), tmpde->name);
+        if (tmpde->attr & RS_ATTR_DIR)
+            StrCpy(file_list + StrLen(file_list), "/");
+        StrCpy(file_list + StrLen(file_list), ";");
+        Date2Struct(&ds, tmpde->datetime);
+        StrPrint(file_list + StrLen(file_list), "%04d-%02d-%02d %02d:%02d", ds.year,
+            ds.mon, ds.day_of_mon, ds.hour, ds.min);
+        StrCpy(file_list + StrLen(file_list), ";");
+        StrPrint(file_list + StrLen(file_list), "%d", tmpde->size);
+        tmpde = tmpde->next;
+        counter++;
+    }
+
+    DirTreeDel(de);
+    return file_list;
+}
+
+U0 _Z14os_path_existsPKc()
+{
+    // os_path_exists(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    FileFind(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z13os_files_listPKc()
+{
+    // os_files_list(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @os_files_list(p0);
+    POP_SYSV_REGS
+}
+
+Bool @os_is_vm()
+{
+    CRAXRBCRCXRDX res;
+    CPUId(0x40000000, &res);
+    if (res.rbx == 0x4B4D564B)
+        return TRUE;
+    return FALSE;
+}
+
+U0 _Z8os_is_vmv()
+{
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @os_is_vm;
+    POP_SYSV_REGS
+}
+
+U0 @os_pc_speaker(F64 frequency)
+{
+    I64 period;
+    if (!frequency)
+        OutU8(0x61, InU8(0x61) & ~3);
+    else {
+        period = ClampI64(SYS_TIMER_FREQ / frequency, 1, U16_MAX);
+        OutU8(0x43, 0xB6);
+        OutU8(0x42, period);
+        OutU8(0x42, period.u8[1]);
+        OutU8(0x61, 3 | InU8(0x61));
+    }
+}
+
+U0 _Z13os_pc_speakerPKc()
+{
+    // os_pc_speaker(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    F64 frequency = Str2F64(p0);
+    @os_pc_speaker(frequency);
+    POP_SYSV_REGS
+}
+
+U0 _Z9os_randomv()
+{
+    // os_random()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    RandU64;
+    POP_SYSV_REGS
+}
+
+U0 _Z19os_read_entire_filePKcPl()
+{
+    // os_read_entire_file(char const*, long*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    FileRead(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 @os_screenshot()
+{
+    CDC* dc = DCScrnCapture(, adam_task);
+    // Image.Write("B:/screenshot.png", dc);
+    DCDel(dc);
+}
+
+U0 _Z13os_screenshotv()
+{
+    // os_screenshot()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @os_screenshot;
+    POP_SYSV_REGS
+}
+
+U8* @os_to_uppercase(U8* instr)
+{
+    if (!instr)
+        return NULL;
+    if (!StrLen(instr))
+        return NULL;
+    U8* outstr = CAlloc(StrLen(instr) + 1, adam_task);
+    I64 i;
+    for (i = 0; i < StrLen(instr); i++)
+        outstr[i] = ToUpper(instr[i]);
+    return outstr;
+}
+
+U0 _Z15os_to_uppercasePKc()
+{
+    // os_to_uppercase(char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @os_to_uppercase(p0);
+    POP_SYSV_REGS
+}
+
+U0 _Z20os_write_entire_filePKcPhl()
+{
+    // os_write_entire_file(char const*, unsigned char*, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    FileWrite(p0, p1, p2);
+    POP_SYSV_REGS
+}
diff --git a/System/Jakt/PCI.HC b/System/Jakt/PCI.HC
new file mode 100644
index 0000000..1a37903
--- /dev/null
+++ b/System/Jakt/PCI.HC
@@ -0,0 +1,62 @@
+U0 _Z8pci_findl()
+{
+    // pci_find(long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIClassFind(p0, 0);
+    POP_SYSV_REGS
+}
+
+U0 _Z11pci_read_u8llll()
+{
+    // pci_read_u8(long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIReadU8(p0, p1, p2, p3);
+    POP_SYSV_REGS
+}
+
+U0 _Z12pci_read_u16llll()
+{
+    // pci_read_u16(long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIReadU16(p0, p1, p2, p3);
+    POP_SYSV_REGS
+}
+
+U0 _Z12pci_read_u32llll()
+{
+    // pci_read_u32(long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIReadU32(p0, p1, p2, p3);
+    POP_SYSV_REGS
+}
+
+U0 _Z12pci_write_u8llllh()
+{
+    // pci_write_u8(long, long, long, long, unsigned char)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIWriteU8(p0, p1, p2, p3, p4);
+    POP_SYSV_REGS
+}
+
+U0 _Z13pci_write_u16llllt()
+{
+    // pci_write_u16(long, long, long, long, unsigned short)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIWriteU16(p0, p1, p2, p3, p4);
+    POP_SYSV_REGS
+}
+
+U0 _Z13pci_write_u32llllj()
+{
+    // pci_write_u32(long, long, long, long, unsigned int)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    PCIWriteU32(p0, p1, p2, p3, p4);
+    POP_SYSV_REGS
+}
diff --git a/System/Jakt/Time.HC b/System/Jakt/Time.HC
new file mode 100644
index 0000000..f6f3f31
--- /dev/null
+++ b/System/Jakt/Time.HC
@@ -0,0 +1,37 @@
+U0 _Z9time_busyl()
+{
+    // time_busy(long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Busy(p0);
+    POP_SYSV_REGS
+}
+
+I64 @time_jiffies() { return cnts.jiffies; }
+
+U0 _Z12time_jiffiesv()
+{
+    // time_jiffies()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @time_jiffies;
+    POP_SYSV_REGS
+}
+
+U0 _Z8time_nowv()
+{
+    // time_now()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Now;
+    POP_SYSV_REGS
+}
+
+U0 _Z10time_sleepl()
+{
+    // time_sleep(long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Sleep(p0);
+    POP_SYSV_REGS
+}
\ No newline at end of file
diff --git a/System/Jakt/Window.HC b/System/Jakt/Window.HC
new file mode 100644
index 0000000..7e91097
--- /dev/null
+++ b/System/Jakt/Window.HC
@@ -0,0 +1,94 @@
+U0 @window_draw_it(CTask* task, CDC* dc)
+{
+    if (task->user_data)
+        @dc_copy(dc, task->pix_left, task->pix_top, task->user_data);
+}
+
+CTask* @window_user()
+{
+    CTask* task = Spawn(&UserCmdLine, , , 0);
+    TaskWait(task);
+    XTalk(task,
+        "while (1) { StrCpy(Fs->task_title, Fs->task_name); Sleep(1); };\n");
+    return task;
+}
+
+CTask* @window_create()
+{
+    CTask* task = @window_user;
+    task->draw_it = &@window_draw_it;
+    return task;
+}
+
+U0 _Z13window_createv()
+{
+    // window_create()
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @window_create();
+    POP_SYSV_REGS
+}
+
+U0 _Z14window_destroym()
+{
+    // window_destroy(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    Kill(p0);
+    POP_SYSV_REGS
+}
+
+Bool @window_is_focused(CTask* task) { return task == sys_focus_task; }
+
+U0 _Z17window_is_focusedm()
+{
+    // window_is_focused(unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @window_is_focused(p0);
+    POP_SYSV_REGS
+}
+
+U0 @window_set_coordinates(CTask* task, I64 top, I64 left, I64 bottom,
+    I64 right)
+{
+    task->win_top = top;
+    task->win_left = left;
+    task->win_bottom = bottom;
+    task->win_right = right;
+}
+
+U0 _Z22window_set_coordinatesmllll()
+{
+    // window_set_coordinates(unsigned long, long, long, long, long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @window_set_coordinates(p0, p1, p2, p3, p4);
+    POP_SYSV_REGS
+}
+
+U0 @window_set_context(CTask* task, CDC* dc) { task->user_data = dc; }
+
+U0 _Z18window_set_contextmm()
+{
+    // window_set_context(unsigned long, unsigned long)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @window_set_context(p0, p1);
+    POP_SYSV_REGS
+}
+
+U0 @window_set_title(CTask* task, U8* title)
+{
+    StrCpy(task->task_name, title);
+    StrCpy(task->task_title, title);
+}
+
+U0 _Z16window_set_titlemPKc()
+{
+    // window_set_title(unsigned long, char const*)
+    PUSH_SYSV_REGS
+    GET_SYSV_ARGS
+    @window_set_title(p0, p1);
+    POP_SYSV_REGS
+}
diff --git a/System/Libraries/Json.HC b/System/Libraries/Json.HC
new file mode 100644
index 0000000..373e87f
--- /dev/null
+++ b/System/Libraries/Json.HC
@@ -0,0 +1,1573 @@
+#define JSON_SAME -1
+#define JSON_UNDEFINED 0
+#define JSON_OBJECT 1
+#define JSON_ARRAY 2
+#define JSON_STRING 3
+#define JSON_NUMBER 4
+#define JSON_BOOLEAN 5
+#define JSON_NULL 6
+#define JSON_HTML 7
+
+#define JSON_ELEMENT_IS_KEY 1
+#define JSON_ELEMENT_IS_ITEM 2
+
+#define JSON_STATE_OBJECT_OR_ARRAY 0
+
+#define JSON_STATE_OBJECT 100
+#define JSON_STATE_OBJECT_KEY 101
+#define JSON_STATE_OBJECT_SEPARATOR 102
+#define JSON_STATE_OBJECT_TYPE 103
+#define JSON_STATE_OBJECT_NEXT 104
+
+#define JSON_STATE_OBJECT_OBJECT 105
+#define JSON_STATE_OBJECT_ARRAY 106
+#define JSON_STATE_OBJECT_STRING 107
+#define JSON_STATE_OBJECT_NUMBER 108
+#define JSON_STATE_OBJECT_BOOLEAN 109
+#define JSON_STATE_OBJECT_NULL 110
+
+#define JSON_STATE_ARRAY 200
+#define JSON_STATE_ARRAY_TYPE 201
+#define JSON_STATE_ARRAY_NEXT 202
+
+#define JSON_STATE_ARRAY_OBJECT 203
+#define JSON_STATE_ARRAY_ARRAY 204
+#define JSON_STATE_ARRAY_STRING 205
+#define JSON_STATE_ARRAY_NUMBER 206
+#define JSON_STATE_ARRAY_BOOLEAN 207
+#define JSON_STATE_ARRAY_NULL 208
+
+#define JSON_PARSER_FIFO_SIZE 32768
+
+#define JSON_WRAPPER_MAGIC_NUMBER 0xDEADC0DEDEADC0DE
+
+#define JSON_SIG 0xFABACEAE
+
+class @json_stringify_string
+{
+    I64 length;
+    U8* value;
+    I64 capacity;
+    CTask* mem_task;
+}
+
+class @json_element
+{
+    U32 sig;
+    @json_element* prev;
+    @json_element* next;
+    I64 type;
+};
+
+class @json_key : @json_element
+{
+    U8* name;
+    U64 value;
+};
+
+class @json_item : @json_element { U64 value; };
+
+class @json_object : @json_element
+{
+    I64 length;
+    @json_key* keys;
+    CTask* mem_task;
+};
+
+class @json_array : @json_element
+{
+    I64 length;
+    @json_item* items;
+    @json_item* last_item;
+    CTask* mem_task;
+};
+
+extern class @json_callable_object;
+
+class @json_callable_array : @json_array
+{
+    U64 (*@)(I64 index, Bool return_item = FALSE);
+    @json_callable_array* (*a)(I64 index, Bool return_item = FALSE);
+    @json_callable_object* (*o)(I64 index, Bool return_item = FALSE);
+    U0 (*append)(U64 value, I64 type = NULL);
+    Bool (*contains)(U64 value, I64 type = NULL, Bool match_case = FALSE);
+    U0 (*insert)(I64 index, U64 value, I64 type = NULL);
+    U0 (*prepend)(U64 value, I64 type = NULL);
+    U0 (*remove)(I64 index);
+};
+
+class @json_callable_object : @json_object
+{
+    U64 (*@)(U8* key, Bool return_key = FALSE);
+    @json_callable_array* (*a)(U8* key, Bool return_key = FALSE);
+    @json_callable_object* (*o)(U8* key, Bool return_key = FALSE);
+    U0 (*set)(U8* key, U64 value, I64 type = JSON_SAME);
+    U0 (*unset)(U8* key);
+};
+
+class @json_parser
+{
+    U8* stream;
+    U8 token;
+    CFifoU8* consumed;
+    I64 pos;
+    I64 state;
+    Bool debug;
+    CTask* mem_task;
+};
+
+#define JsonArray @json_callable_array
+#define JsonElement @json_element
+#define JsonItem @json_item
+#define JsonKey @json_key
+#define JsonObject @json_callable_object
+
+U0 @json_debug_parser_state(@json_parser* parser)
+{
+    switch (parser->state) {
+    case JSON_STATE_OBJECT:
+        "JSON_STATE_OBJECT\n";
+        break;
+    case JSON_STATE_OBJECT_KEY:
+        "JSON_STATE_OBJECT_KEY\n";
+        break;
+    case JSON_STATE_OBJECT_SEPARATOR:
+        "JSON_STATE_OBJECT_SEPARATOR\n";
+        break;
+    case JSON_STATE_OBJECT_TYPE:
+        "JSON_STATE_OBJECT_TYPE\n";
+        break;
+    case JSON_STATE_OBJECT_NEXT:
+        "JSON_STATE_OBJECT_NEXT\n";
+        break;
+    case JSON_STATE_OBJECT_STRING:
+        "JSON_STATE_OBJECT_STRING\n";
+        break;
+    case JSON_STATE_OBJECT_NUMBER:
+        "JSON_STATE_OBJECT_NUMBER\n";
+        break;
+    case JSON_STATE_ARRAY:
+        "JSON_STATE_ARRAY\n";
+        break;
+    case JSON_STATE_ARRAY_TYPE:
+        "JSON_STATE_ARRAY_TYPE\n";
+        break;
+    case JSON_STATE_ARRAY_NEXT:
+        "JSON_STATE_ARRAY_NEXT\n";
+        break;
+    case JSON_STATE_ARRAY_STRING:
+        "JSON_STATE_ARRAY_STRING\n";
+        break;
+    case JSON_STATE_ARRAY_NUMBER:
+        "JSON_STATE_ARRAY_NUMBER\n";
+        break;
+    }
+}
+
+@json_item* @json_create_item(U64 value, I64 type = NULL, CTask* mem_task)
+{
+    @json_item* item = CAlloc(sizeof(@json_item), mem_task);
+    item->sig = JSON_SIG;
+    item->type = type;
+    if (!item->type) {
+        if (value(@json_element*)->sig == JSON_SIG) {
+            item->type = value(@json_element*)->type;
+        } else if (value > 0x1000) {
+            item->type = JSON_STRING;
+        } else {
+            item->type = JSON_BOOLEAN;
+            value = value != 0;
+        }
+    }
+    if (item->type == JSON_STRING)
+        item->value = StrNew(value, mem_task);
+    else
+        item->value = value;
+    return item;
+}
+
+U0 @json_append_item(@json_array* arr, U64 value, I64 type = NULL)
+{
+    if (!arr)
+        return;
+    if (arr->type != JSON_ARRAY)
+        return;
+    @json_item* append_item = @json_create_item(value, type, arr->mem_task);
+    @json_item* item = arr->last_item;
+    if (!item) {
+        append_item->prev = NULL;
+        append_item->next = NULL;
+        arr->items = append_item;
+        arr->last_item = append_item;
+        arr->length++;
+        return;
+    }
+    item->next = append_item;
+    append_item->prev = item;
+    arr->last_item = append_item;
+    arr->length++;
+}
+
+U8* @json_string_from_fifo(CFifoU8* f, CTask* mem_task)
+{
+    U8 ch;
+    I64 i = 0;
+    U8* str = CAlloc(FifoU8Cnt(f) + 1, mem_task);
+    while (FifoU8Cnt(f)) {
+        FifoU8Rem(f, &ch);
+        str[i] = ch;
+        i++;
+    }
+    FifoU8Flush(f);
+    return str;
+}
+
+U0 @json_insert_key(@json_object* obj, @json_key* key)
+{
+    if (!obj)
+        return;
+    if (!obj->keys) {
+        obj->keys = key;
+        obj->length++;
+        return;
+    }
+    @json_key* k = obj->keys;
+    while (k->next)
+        k = k->next;
+    k->next = key;
+    key->prev = k;
+    obj->length++;
+}
+
+U0 @json_rstrip(U8* str)
+{
+    if (!str || !StrLen(str))
+        return;
+    I64 r_pos = StrLen(str) - 1;
+    while (str[r_pos] == ' ' || str[r_pos] == '\r' || str[r_pos] == '\n' || str[r_pos] == '\t') {
+        str[r_pos] = NULL;
+        --r_pos;
+    }
+}
+
+extern @json_element* @json_parse_object_or_array(@json_parser* parser);
+
+I64 @json_unescape_char(I64 escape_ch)
+{
+    I64 return_ch = escape_ch;
+
+    // FIXME: unicode
+    switch (escape_ch) {
+    case 'b':
+        return_ch = 0x08;
+        break;
+    case 'f':
+        return_ch = 0x0c;
+        break;
+    case 'n':
+        return_ch = 0x0a;
+        break;
+    case 'r':
+        return_ch = 0x0d;
+        break;
+    case 't':
+        return_ch = 0x09;
+        break;
+    default:
+        break;
+    }
+
+    return return_ch;
+}
+
+U0 @json_parse_object(@json_parser* parser, @json_object* obj)
+{
+    @json_key* key = NULL;
+    while (1) {
+        switch (parser->stream[parser->pos]) {
+        case '\\':
+            // NOTE: We keep escaped unicode in its original form, and let the program ingesting the JSON handle the UTF-8 conversion.
+            if (parser->state == JSON_STATE_OBJECT_STRING) {
+                if (parser->stream[parser->pos + 1] == 'u') {
+                    FifoU8Ins(parser->consumed, '\\');
+                } else {
+                    FifoU8Ins(parser->consumed, @json_unescape_char(parser->stream[++parser->pos]));
+                }
+            }
+            break;
+        case '}':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_NUMBER:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(key->value);
+                key->value = Str2F64(key->value);
+                @json_insert_key(obj, key);
+                return;
+                break;
+            case JSON_STATE_OBJECT_BOOLEAN:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(key->value);
+                if (StrCmp("true", key->value) && StrCmp("false", key->value)) {
+                    PrintErr("@json_parse_object: Illegal boolean value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                if (!StrCmp("true", key->value))
+                    key->value = TRUE;
+                else
+                    key->value = FALSE;
+                @json_insert_key(obj, key);
+                return;
+                break;
+            case JSON_STATE_OBJECT_NULL:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(key->value);
+                if (StrCmp("null", key->value)) {
+                    PrintErr("@json_parse_object: Illegal null value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                key->value = NULL;
+                @json_insert_key(obj, key);
+                return;
+                break;
+            case JSON_STATE_OBJECT:
+            case JSON_STATE_OBJECT_NEXT:
+                return;
+                break;
+            }
+            break;
+        case ',':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_NUMBER:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                key->value = Str2F64(key->value);
+                @json_insert_key(obj, key);
+                parser->state = JSON_STATE_OBJECT;
+                break;
+            case JSON_STATE_OBJECT_BOOLEAN:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(key->value);
+                if (StrCmp("true", key->value) && StrCmp("false", key->value)) {
+                    PrintErr("@json_parse_object: Illegal boolean value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                if (!StrCmp("true", key->value))
+                    key->value = TRUE;
+                else
+                    key->value = FALSE;
+                @json_insert_key(obj, key);
+                parser->state = JSON_STATE_OBJECT;
+                break;
+            case JSON_STATE_OBJECT_NULL:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(key->value);
+                if (StrCmp("null", key->value)) {
+                    PrintErr("@json_parse_object: Illegal null value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                key->value = NULL;
+                @json_insert_key(obj, key);
+                parser->state = JSON_STATE_OBJECT;
+                break;
+            case JSON_STATE_OBJECT_NEXT:
+                parser->state = JSON_STATE_OBJECT;
+                break;
+            }
+            break;
+        case ':':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_SEPARATOR:
+                parser->state = JSON_STATE_OBJECT_TYPE;
+                break;
+            }
+            break;
+        case '[':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                key->type = JSON_ARRAY;
+                key->value = @json_parse_object_or_array(parser);
+                @json_insert_key(obj, key);
+                parser->state = JSON_STATE_OBJECT_NEXT;
+                break;
+            }
+            break;
+        case '{':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                key->type = JSON_OBJECT;
+                key->value = @json_parse_object_or_array(parser);
+                @json_insert_key(obj, key);
+                parser->state = JSON_STATE_OBJECT_NEXT;
+                break;
+            }
+            break;
+        case '"':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_STRING:
+                key->value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_insert_key(obj, key);
+                parser->state = JSON_STATE_OBJECT_NEXT;
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                key->type = JSON_STRING;
+                parser->state = JSON_STATE_OBJECT_STRING;
+                break;
+            case JSON_STATE_OBJECT_KEY:
+                key->name = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                parser->state = JSON_STATE_OBJECT_SEPARATOR;
+                break;
+            case JSON_STATE_OBJECT:
+                key = CAlloc(sizeof(@json_key), parser->mem_task);
+                key->sig = JSON_SIG;
+                parser->state = JSON_STATE_OBJECT_KEY;
+                break;
+            }
+            break;
+        case '-':
+        case '0' ... '9':
+        case '.':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+            case JSON_STATE_OBJECT_NUMBER:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                key->type = JSON_NUMBER;
+                parser->state = JSON_STATE_OBJECT_NUMBER;
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        case 't':
+        case 'f':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                key->type = JSON_BOOLEAN;
+                parser->state = JSON_STATE_OBJECT_BOOLEAN;
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        case 'n':
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                key->type = JSON_NULL;
+                parser->state = JSON_STATE_OBJECT_NULL;
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        default:
+            switch (parser->state) {
+            case JSON_STATE_OBJECT_KEY:
+            case JSON_STATE_OBJECT_STRING:
+            case JSON_STATE_OBJECT_BOOLEAN:
+            case JSON_STATE_OBJECT_NULL:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        }
+        if (parser->debug) {
+            @json_debug_parser_state(parser);
+            "Object: %08X, Pos: %d, Token: %c\n", obj, parser->pos,
+                parser->stream[parser->pos];
+            Sleep(50);
+        }
+        parser->pos++;
+    }
+}
+
+U0 @json_parse_array(@json_parser* parser, @json_array* arr)
+{
+    U64 value = NULL;
+    while (1) {
+        if (parser->state == JSON_STATE_ARRAY) {
+            switch (parser->stream[parser->pos]) {
+            case 0:
+                PrintErr("@json_parse_array: Malformed array");
+                while (1)
+                    Sleep(1);
+                break;
+            case ']':
+                return;
+                break;
+            }
+            parser->state = JSON_STATE_ARRAY_TYPE;
+        }
+        switch (parser->stream[parser->pos]) {
+        case '\\':
+            // NOTE: We keep escaped unicode in its original form, and let the program ingesting the JSON handle the UTF-8 conversion.
+            if (parser->state == JSON_STATE_ARRAY_STRING) {
+                if (parser->stream[parser->pos + 1] == 'u') {
+                    FifoU8Ins(parser->consumed, '\\');
+                } else {
+                    FifoU8Ins(parser->consumed, @json_unescape_char(parser->stream[++parser->pos]));
+                }
+            }
+            break;
+        case ']':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_ARRAY_NUMBER:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(value);
+                value = Str2F64(value);
+                @json_append_item(arr, value, JSON_NUMBER);
+                return;
+                break;
+            case JSON_STATE_ARRAY_BOOLEAN:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(value);
+                if (StrCmp("true", value) && StrCmp("false", value)) {
+                    PrintErr("@json_parse_array: Illegal boolean value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                if (!StrCmp("true", value))
+                    value = TRUE;
+                else
+                    value = FALSE;
+                @json_append_item(arr, value, JSON_BOOLEAN);
+                break;
+            case JSON_STATE_ARRAY_NULL:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(value);
+                if (StrCmp("null", value)) {
+                    PrintErr("@json_parse_array: Illegal null value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                value = NULL;
+                @json_append_item(arr, value, JSON_NULL);
+                break;
+            case JSON_STATE_ARRAY:
+            case JSON_STATE_ARRAY_NEXT:
+                return;
+                break;
+            }
+            break;
+        case ',':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_ARRAY_NUMBER:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                value = Str2F64(value);
+                @json_append_item(arr, value, JSON_NUMBER);
+                parser->state = JSON_STATE_ARRAY;
+                break;
+            case JSON_STATE_ARRAY_BOOLEAN:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(value);
+                if (StrCmp("true", value) && StrCmp("false", value)) {
+                    PrintErr("@json_parse_array: Illegal boolean value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                if (!StrCmp("true", value))
+                    value = TRUE;
+                else
+                    value = FALSE;
+                @json_append_item(arr, value, JSON_BOOLEAN);
+                parser->state = JSON_STATE_ARRAY;
+                break;
+            case JSON_STATE_ARRAY_NULL:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_rstrip(value);
+                if (StrCmp("null", value)) {
+                    PrintErr("@json_parse_array: Illegal null value at position %d",
+                        parser->pos);
+                    while (1)
+                        Sleep(1);
+                }
+                value = NULL;
+                @json_append_item(arr, value, JSON_NULL);
+                parser->state = JSON_STATE_ARRAY;
+                break;
+            case JSON_STATE_ARRAY_NEXT:
+                parser->state = JSON_STATE_ARRAY;
+                break;
+            }
+            break;
+        case '[':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_ARRAY_TYPE:
+                value = @json_parse_object_or_array(parser);
+                @json_append_item(arr, value, JSON_ARRAY);
+                parser->state = JSON_STATE_ARRAY_NEXT;
+                break;
+            }
+            break;
+        case '{':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_ARRAY_TYPE:
+                value = @json_parse_object_or_array(parser);
+                @json_append_item(arr, value, JSON_OBJECT);
+                parser->state = JSON_STATE_ARRAY_NEXT;
+                break;
+            }
+            break;
+        case '"':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                value = @json_string_from_fifo(parser->consumed, parser->mem_task);
+                @json_append_item(arr, value, JSON_STRING);
+                parser->state = JSON_STATE_ARRAY_NEXT;
+                break;
+            case JSON_STATE_ARRAY_TYPE:
+                parser->state = JSON_STATE_ARRAY_STRING;
+                break;
+            }
+            break;
+        case '-':
+        case '0' ... '9':
+        case '.':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+            case JSON_STATE_ARRAY_NUMBER:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_ARRAY_TYPE:
+                parser->state = JSON_STATE_ARRAY_NUMBER;
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        case 't':
+        case 'f':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_ARRAY_TYPE:
+                parser->state = JSON_STATE_ARRAY_BOOLEAN;
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        case 'n':
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            case JSON_STATE_OBJECT_TYPE:
+                parser->state = JSON_STATE_ARRAY_NULL;
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        default:
+            switch (parser->state) {
+            case JSON_STATE_ARRAY_STRING:
+            case JSON_STATE_ARRAY_BOOLEAN:
+            case JSON_STATE_ARRAY_NULL:
+                FifoU8Ins(parser->consumed, parser->stream[parser->pos]);
+                break;
+            }
+            break;
+        }
+        if (parser->debug) {
+            @json_debug_parser_state(parser);
+            "Array: %08X, Pos: %d, Token: %c\n", arr, parser->pos,
+                parser->stream[parser->pos];
+            Sleep(50);
+        }
+        parser->pos++;
+    }
+}
+
+extern @json_callable_array* @json_create_callable_array(@json_array* arr);
+extern @json_callable_object* @json_create_callable_object(@json_object* obj);
+
+@json_element* @json_parse_object_or_array(@json_parser* parser)
+{
+    @json_element* el = CAlloc(sizeof(@json_array), parser->mem_task);
+    el->sig = JSON_SIG;
+    while (1) {
+        switch (parser->stream[parser->pos]) {
+        case 0:
+            return el;
+            break;
+        case ' ':
+        case '\r':
+        case '\n':
+        case '\t':
+            break;
+        case '{':
+            el->type = JSON_OBJECT;
+            el(@json_object*)->mem_task = parser->mem_task;
+            parser->pos++;
+            parser->state = JSON_STATE_OBJECT;
+            @json_parse_object(parser, el);
+            return @json_create_callable_object(el);
+            break;
+        case '[':
+            el->type = JSON_ARRAY;
+            el(@json_array*)->mem_task = parser->mem_task;
+            parser->pos++;
+            parser->state = JSON_STATE_ARRAY;
+            @json_parse_array(parser, el);
+            return @json_create_callable_array(el);
+            break;
+        default:
+            PrintErr("@json_parse_object_or_array: Invalid token: '%c'", parser->stream[parser->pos]);
+            while (1) {
+                Sleep(1);
+            };
+            break;
+        }
+        parser->pos++;
+        Sleep(1);
+    }
+}
+
+@json_element* @json_parse(U8* str, CTask* mem_task)
+{
+    if (!str || !mem_task) {
+        return NULL;
+    }
+    @json_parser* parser = CAlloc(sizeof(@json_parser), mem_task);
+    parser->mem_task = mem_task;
+    parser->consumed = FifoU8New(JSON_PARSER_FIFO_SIZE, parser->mem_task);
+    parser->stream = str;
+    return @json_parse_object_or_array(parser);
+}
+
+U0 @json_prepend_item(@json_array* arr, U64 value, I64 type)
+{
+    if (!arr)
+        return;
+    if (arr->type != JSON_ARRAY)
+        return;
+    @json_item* prepend_item = @json_create_item(value, type, arr->mem_task);
+    @json_item* items = arr->items;
+    arr->items = prepend_item;
+    arr->items->next = items;
+    arr->length++;
+}
+
+U0 @json_insert_item(@json_array* arr, I64 index, U64 value, I64 type)
+{
+    if (!arr)
+        return;
+    if (arr->type != JSON_ARRAY)
+        return;
+    if (index <= 0) {
+        @json_prepend_item(arr, value, type);
+        return;
+    }
+    if (index >= arr->length) {
+        @json_append_item(arr, value, type);
+        return;
+    }
+    @json_item* insert_item = @json_create_item(value, type, arr->mem_task);
+    @json_item* insert_at_item = arr->items;
+    @json_item* insert_after_item = NULL;
+    I64 i;
+    for (i = 0; i < index; i++)
+        insert_at_item = insert_at_item->next;
+    insert_after_item = insert_at_item->prev;
+    insert_after_item->next = insert_item;
+    insert_item->prev = insert_after_item;
+    insert_item->next = insert_at_item;
+    insert_at_item->prev = insert_item;
+    arr->length++;
+}
+
+U0 @json_stringify_check_capacity(@json_stringify_string* str)
+{
+    if (str->length >= str->capacity) {
+        str->capacity *= 2;
+        U8* new_value = CAlloc(str->capacity * 2, str->mem_task);
+        MemCpy(new_value, str->value, str->length);
+        str->value = new_value;
+    }
+}
+
+U0 @json_stringify_append_char(@json_stringify_string* str, U8 char)
+{
+    // FIXME: unicode
+    switch (char) {
+    case '\\':
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = '\\';
+        break;
+    case 0x08:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'b';
+        break;
+    case 0x0c:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'f';
+        break;
+    case 0x0a:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'n';
+        break;
+    case 0x0d:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'r';
+        break;
+    case 0x09:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 't';
+        break;
+    default:
+        str->value[str->length++] = char;
+        break;
+    }
+    str->value[str->length] = NULL;
+    @json_stringify_check_capacity(str);
+}
+
+U0 @json_stringify_append_char_escape_quotes(@json_stringify_string* str, U8 char)
+{
+    // FIXME: unicode
+    switch (char) {
+    case '"':
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = '"';
+        break;
+    case '\\':
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = '\\';
+        break;
+    case 0x08:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'b';
+        break;
+    case 0x0c:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'f';
+        break;
+    case 0x0a:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'n';
+        break;
+    case 0x0d:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 'r';
+        break;
+    case 0x09:
+        str->value[str->length++] = '\\';
+        str->value[str->length++] = 't';
+        break;
+    default:
+        str->value[str->length++] = char;
+        break;
+    }
+    str->value[str->length] = NULL;
+    @json_stringify_check_capacity(str);
+}
+
+U0 @json_stringify_append_str(@json_stringify_string* str, U8* str2)
+{
+    while (*str2) {
+        // NOTE: We keep escaped unicode in its original form, and let the program ingesting the JSON handle the UTF-8 conversion.
+        if (*str2 == '\\' && *(str2 + 1) == 'u') {
+            str->value[str->length++] = '\\';
+            str->value[str->length++] = 'u';
+            str->value[str->length] = NULL;
+            str2++;
+        } else {
+            @json_stringify_append_char_escape_quotes(str, *str2);
+        }
+        str2++;
+        @json_stringify_check_capacity(str);
+    }
+}
+
+U0 @json_stringify_append_number(@json_stringify_string* str, F64 num)
+{
+    U8 buf[256];
+    MemSet(buf, 0, 256);
+    StrPrint(buf, "%.6f", num);
+    I64 i = StrLen(buf) - 1;
+    while (buf[i] == '0') {
+        buf[i] = NULL;
+        i--;
+    }
+    i = StrLen(buf) - 1;
+    if (buf[i] == '.')
+        buf[i] = NULL;
+    @json_stringify_append_str(str, buf);
+}
+
+extern U0 @json_stringify_object_or_array(@json_stringify_string* str, @json_element* el);
+
+U0 @json_stringify_object(@json_stringify_string* str, @json_object* obj)
+{
+    @json_stringify_append_char(str, '{');
+    @json_key* key = obj->keys;
+    while (key) {
+        @json_stringify_append_char(str, '"');
+        @json_stringify_append_str(str, key->name);
+        @json_stringify_append_char(str, '"');
+        @json_stringify_append_char(str, ':');
+        switch (key->type) {
+        case JSON_OBJECT:
+        case JSON_ARRAY:
+            @json_stringify_object_or_array(str, key->value);
+            break;
+        case JSON_STRING:
+            @json_stringify_append_char(str, '"');
+            @json_stringify_append_str(str, key->value);
+            @json_stringify_append_char(str, '"');
+            break;
+        case JSON_NUMBER:
+            @json_stringify_append_number(str, key->value);
+            break;
+        case JSON_BOOLEAN:
+            @json_stringify_append_str(str, @t(key->value, "true", "false"));
+            break;
+        case JSON_NULL:
+            @json_stringify_append_str(str, "null");
+            break;
+        default:
+            PrintErr("@json_stringify_object: Invalid element type: %d", key->type);
+            while (1) {
+                Sleep(1);
+            };
+        }
+        if (key->next)
+            @json_stringify_append_char(str, ',');
+        key = key->next;
+    }
+    @json_stringify_append_char(str, '}');
+}
+
+U0 @json_stringify_array(@json_stringify_string* str, @json_array* arr)
+{
+    @json_stringify_append_char(str, '[');
+    @json_item* item = arr->items;
+    while (item) {
+        switch (item->type) {
+        case JSON_OBJECT:
+        case JSON_ARRAY:
+            @json_stringify_object_or_array(str, item->value);
+            break;
+        case JSON_STRING:
+            @json_stringify_append_char(str, '"');
+            @json_stringify_append_str(str, item->value);
+            @json_stringify_append_char(str, '"');
+            break;
+        case JSON_NUMBER:
+            @json_stringify_append_number(str, item->value);
+            break;
+        case JSON_BOOLEAN:
+            @json_stringify_append_str(str, @t(item->value, "true", "false"));
+            break;
+        case JSON_NULL:
+            @json_stringify_append_str(str, "null");
+            break;
+        default:
+            PrintErr("@json_stringify_array: Invalid element type: %d", item->type);
+            while (1) {
+                Sleep(1);
+            };
+        }
+        if (item->next)
+            @json_stringify_append_char(str, ',');
+        item = item->next;
+    }
+    @json_stringify_append_char(str, ']');
+}
+
+U0 @json_stringify_object_or_array(@json_stringify_string* str, @json_element* el)
+{
+    while (el) {
+        switch (el->type) {
+        case JSON_OBJECT:
+            @json_stringify_object(str, el);
+            break;
+        case JSON_ARRAY:
+            @json_stringify_array(str, el);
+            break;
+        default:
+            PrintErr("@json_stringify_object_or_array: Invalid element type: %d", el->type);
+            while (1) {
+                Sleep(1);
+            };
+            break;
+        }
+        el = el->next;
+    }
+}
+
+U8* @json_stringify(@json_element* el, CTask* mem_task)
+{
+    if (!el || !mem_task) {
+        return NULL;
+    }
+    @json_stringify_string* str = CAlloc(sizeof(@json_stringify_string), mem_task);
+    str->mem_task = mem_task;
+    str->capacity = 256;
+    str->value = CAlloc(str->capacity * 2, str->mem_task);
+    @json_stringify_object_or_array(str, el);
+    return str->value;
+}
+
+U64 @json_get(@json_object* obj, U8* key, Bool return_key = FALSE)
+{
+    if (!obj || !key)
+        return NULL;
+    if (!obj->keys || obj->type != JSON_OBJECT)
+        return NULL;
+    @json_key* iter_key = obj->keys;
+    while (iter_key) {
+        if (!StrICmp(iter_key->name, key))
+            if (return_key)
+                return iter_key;
+            else
+                return iter_key->value;
+        iter_key = iter_key->next;
+    }
+    return NULL;
+}
+
+U0 @json_set(@json_object* obj, U8* key, U64 value, I64 type = JSON_SAME)
+{
+    if (!obj || !key || !type)
+        return;
+    if (obj->type != JSON_OBJECT)
+        return;
+    @json_key* iter_key = obj->keys;
+    while (iter_key) {
+        if (!StrCmp(iter_key->name, key)) {
+            if (type != JSON_SAME)
+                iter_key->type = type;
+            iter_key->value = value;
+            return;
+        }
+        iter_key = iter_key->next;
+    }
+    @json_key* new_key = CAlloc(sizeof(@json_key), obj->mem_task);
+    new_key->sig = JSON_SIG;
+    new_key->name = StrNew(key, obj->mem_task);
+    new_key->type = type;
+    if (new_key->type == JSON_STRING)
+        new_key->value = StrNew(value, obj->mem_task);
+    else
+        new_key->value = value;
+    @json_insert_key(obj, new_key);
+}
+
+U0 @json_unset(@json_object* obj, U8* key)
+{
+    if (!obj || !key)
+        return;
+    if (obj->type != JSON_OBJECT)
+        return;
+    @json_key* iter_key = obj->keys;
+    @json_key* prev_key = NULL;
+    @json_key* next_key = NULL;
+    while (iter_key) {
+        if (!StrCmp(iter_key->name, key)) {
+            prev_key = iter_key->prev;
+            next_key = iter_key->next;
+            if (prev_key)
+                prev_key->next = next_key;
+            if (next_key)
+                next_key->prev = prev_key;
+            return;
+        }
+        iter_key = iter_key->next;
+    }
+}
+
+U64 @json_callable_object_get_wrapper_function(U8* key, Bool return_key = FALSE)
+{
+    return @json_get(JSON_WRAPPER_MAGIC_NUMBER, key, return_key);
+}
+
+U0 @json_callable_object_set_wrapper_function(U8* key, U64 value, I64 type = JSON_SAME)
+{
+    @json_set(JSON_WRAPPER_MAGIC_NUMBER, key, value, type);
+}
+
+U0 @json_callable_object_unset_wrapper_function(U8* key)
+{
+    @json_unset(JSON_WRAPPER_MAGIC_NUMBER, key);
+}
+
+@json_callable_object* @json_create_callable_object(@json_object* obj)
+{
+    @json_callable_object* cobj = CAlloc(sizeof(@json_callable_object), obj->mem_task);
+    MemCpy(cobj, obj, sizeof(@json_object));
+    Free(obj);
+
+    U64 a;
+    I64 buffer_size = (MSize(&@json_callable_object_get_wrapper_function) + MSize(&@json_callable_object_set_wrapper_function) + MSize(&@json_callable_object_unset_wrapper_function));
+    buffer_size += buffer_size % 16;
+
+    U64 code_ptr = CAlloc(buffer_size, cobj->mem_task->code_heap);
+    I64 code_size = 0;
+
+    // Create a copy of function and patch Get
+    code_size = MSize(&@json_callable_object_get_wrapper_function);
+    cobj->@ = code_ptr;
+    MemCpy(cobj->@, &@json_callable_object_get_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = cobj->@;
+    a += 0x13;
+    MemSetI64(a, cobj, 1);
+
+    a = cobj->@;
+    a += 0x1c;
+    @patch_call_rel32(a, &@json_get);
+
+    // Create a copy of function and patch Set
+    code_size = MSize(&@json_callable_object_set_wrapper_function);
+    cobj->set = code_ptr;
+    MemCpy(cobj->set, &@json_callable_object_set_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = cobj->set;
+    a += 0x1a;
+    MemSetI64(a, cobj, 1);
+
+    a = cobj->set;
+    a += 0x23;
+    @patch_call_rel32(a, &@json_set);
+
+    // Create a copy of function and patch Unset
+    code_size = MSize(&@json_callable_object_unset_wrapper_function);
+    cobj->unset = code_ptr;
+    MemCpy(cobj->unset, &@json_callable_object_unset_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = cobj->unset;
+    a += 0x0c;
+    MemSetI64(a, cobj, 1);
+
+    a = cobj->unset;
+    a += 0x15;
+    @patch_call_rel32(a, &@json_unset);
+
+    cobj->a = cobj->@;
+    cobj->o = cobj->@;
+
+    return cobj;
+}
+
+@json_callable_object* @json_create_object(CTask* mem_task)
+{
+    if (!mem_task) {
+        return NULL;
+    }
+    @json_object* obj = CAlloc(sizeof(@json_object), mem_task);
+    obj->mem_task = mem_task;
+    obj->sig = JSON_SIG;
+    obj->type = JSON_OBJECT;
+    return @json_create_callable_object(obj);
+}
+
+U64 @json_array_index(@json_array* arr, I64 index, Bool return_item = FALSE)
+{
+    if (!arr)
+        return NULL;
+    if (arr->type != JSON_ARRAY)
+        return NULL;
+    if (index < 0)
+        return NULL;
+    if (!arr->length)
+        return NULL;
+    if (index > 0 && index > arr->length - 1)
+        return NULL;
+    @json_item* item = arr->items;
+    if (!item)
+        return NULL;
+    I64 i;
+    for (i = 0; i < index; i++)
+        item = item->next;
+    if (return_item)
+        return item;
+    else
+        return item->value;
+}
+
+Bool @json_array_contains(@json_array* arr, U64 value, I64 type = NULL, Bool match_case = FALSE)
+{
+    if (!arr)
+        return FALSE;
+    if (arr->type != JSON_ARRAY)
+        return FALSE;
+    if (!arr->length)
+        return FALSE;
+
+    if (!type) {
+        if (value > 0x1000) {
+            type = JSON_STRING;
+        } else {
+            type = JSON_BOOLEAN;
+        }
+    }
+
+    I64 i;
+    @json_item* item = NULL;
+    for (i = 0; i < arr->length; i++) {
+        item = @json_array_index(arr, i, TRUE);
+        if (item->type == type) {
+            switch (type) {
+            case JSON_STRING:
+                if (match_case) {
+                    if (!StrCmp(value, item->value)) {
+                        return TRUE;
+                    }
+                } else {
+                    if (!StrICmp(value, item->value)) {
+                        return TRUE;
+                    }
+                }
+                break;
+            case JSON_BOOLEAN:
+            case JSON_NULL:
+            case JSON_NUMBER:
+                if (item->value == value) {
+                    return TRUE;
+                }
+                break;
+            default:
+                break;
+            }
+        }
+    }
+    return FALSE;
+}
+
+U0 @json_remove_item(@json_array* arr, I64 index)
+{
+    if (!arr)
+        return;
+    if (arr->type != JSON_ARRAY)
+        return;
+    if (index < 0)
+        return;
+    if (!arr->length)
+        return;
+    if (index > 0 && index > arr->length - 1)
+        return;
+    @json_item* item = arr->items;
+    if (!item)
+        return;
+    @json_item* prev_item = NULL;
+    @json_item* next_item = NULL;
+    I64 i;
+    for (i = 0; i < index; i++)
+        item = item->next;
+    if (!index) {
+        arr->items = item->next;
+        goto @json_remove_item_final;
+    }
+    prev_item = item->prev;
+    next_item = item->next;
+    if (arr->last_item == item)
+        arr->last_item = item->prev;
+    prev_item->next = next_item;
+    if (next_item)
+        next_item->prev = prev_item;
+    @json_remove_item_final : --arr->length;
+    if (!arr->length)
+        arr->last_item = NULL;
+}
+
+U64 @json_callable_array_index_wrapper_function(I64 index, Bool return_item = FALSE)
+{
+    return @json_array_index(JSON_WRAPPER_MAGIC_NUMBER, index, return_item);
+}
+
+U0 @json_callable_array_append_wrapper_function(U64 value, I64 type = NULL)
+{
+    @json_append_item(JSON_WRAPPER_MAGIC_NUMBER, value, type);
+}
+
+Bool @json_callable_array_contains_wrapper_function(U64 value, I64 type = NULL, Bool match_case = FALSE)
+{
+    return @json_array_contains(JSON_WRAPPER_MAGIC_NUMBER, value, type, match_case);
+}
+
+U0 @json_callable_array_insert_wrapper_function(I64 index, U64 value, I64 type = NULL)
+{
+    @json_insert_item(JSON_WRAPPER_MAGIC_NUMBER, index, value, type);
+}
+
+U0 @json_callable_array_prepend_wrapper_function(U64 value, I64 type = NULL)
+{
+    @json_prepend_item(JSON_WRAPPER_MAGIC_NUMBER, value, type);
+}
+
+U0 @json_callable_array_remove_wrapper_function(I64 index)
+{
+    @json_remove_item(JSON_WRAPPER_MAGIC_NUMBER, index);
+}
+
+@json_callable_array* @json_create_callable_array(@json_array* arr)
+{
+    // Alloc callable object and copy instance
+    @json_callable_array* carr = CAlloc(sizeof(@json_callable_array), arr->mem_task);
+    MemCpy(carr, arr, sizeof(@json_array));
+    Free(arr);
+
+    U64 a;
+    I64 buffer_size = (MSize(&@json_callable_array_index_wrapper_function) + MSize(&@json_callable_array_append_wrapper_function) + MSize(&@json_callable_array_contains_wrapper_function) + MSize(&@json_callable_array_prepend_wrapper_function) + MSize(&@json_callable_array_insert_wrapper_function) + MSize(&@json_callable_array_remove_wrapper_function));
+    buffer_size += buffer_size % 16;
+
+    U64 code_ptr = CAlloc(buffer_size, carr->mem_task->code_heap);
+    I64 code_size = 0;
+
+    // Create a copy of function and patch Index
+    code_size = MSize(&@json_callable_array_index_wrapper_function);
+    carr->@ = code_ptr;
+    MemCpy(carr->@, &@json_callable_array_index_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = carr->@;
+    a += 0x13;
+    MemSetI64(a, carr, 1);
+
+    a = carr->@;
+    a += 0x1c;
+    @patch_call_rel32(a, &@json_array_index);
+
+    carr->a = carr->@;
+    carr->o = carr->@;
+
+    // Create a copy of function and patch Append
+    code_size = MSize(&@json_callable_array_append_wrapper_function);
+    carr->append = code_ptr;
+    MemCpy(carr->append, &@json_callable_array_append_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = carr->append;
+    a += 0x12;
+    MemSetI64(a, carr, 1);
+
+    a = carr->append;
+    a += 0x1b;
+    @patch_call_rel32(a, &@json_append_item);
+
+    // Create a copy of function and patch Contains
+    code_size = MSize(&@json_callable_array_contains_wrapper_function);
+    carr->contains = code_ptr;
+    MemCpy(carr->contains, &@json_callable_array_contains_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = carr->contains;
+    a += 0x1b;
+    MemSetI64(a, carr, 1);
+
+    a = carr->contains;
+    a += 0x24;
+    @patch_call_rel32(a, &@json_array_contains);
+
+    // Create a copy of function and patch Prepend
+    code_size = MSize(&@json_callable_array_prepend_wrapper_function);
+    carr->prepend = code_ptr;
+    MemCpy(carr->prepend, &@json_callable_array_prepend_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = carr->prepend;
+    a += 0x12;
+    MemSetI64(a, carr, 1);
+
+    a = carr->prepend;
+    a += 0x1b;
+    @patch_call_rel32(a, &@json_prepend_item);
+
+    // Create a copy of function and patch Insert
+    code_size = MSize(&@json_callable_array_insert_wrapper_function);
+    carr->insert = code_ptr;
+    MemCpy(carr->insert, &@json_callable_array_insert_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = carr->insert;
+    a += 0x1a;
+    MemSetI64(a, carr, 1);
+
+    a = carr->insert;
+    a += 0x23;
+    @patch_call_rel32(a, &@json_insert_item);
+
+    // Create a copy of function and patch Remove
+    code_size = MSize(&@json_callable_array_remove_wrapper_function);
+    carr->remove = code_ptr;
+    MemCpy(carr->remove, &@json_callable_array_remove_wrapper_function, code_size);
+    code_ptr += code_size;
+
+    a = carr->remove;
+    a += 0x0c;
+    MemSetI64(a, carr, 1);
+
+    a = carr->remove;
+    a += 0x15;
+    @patch_call_rel32(a, &@json_remove_item);
+
+    return carr;
+}
+
+@json_array* @json_create_array(CTask* mem_task)
+{
+    if (!mem_task) {
+        return NULL;
+    }
+    @json_array* arr = CAlloc(sizeof(@json_array), mem_task);
+    arr->mem_task = mem_task;
+    arr->sig = JSON_SIG;
+    arr->type = JSON_ARRAY;
+    return @json_create_callable_array(arr);
+}
+
+U64 @json_parse_file(U8* path, CTask* mem_task)
+{
+    if (!path || !mem_task || !FileFind(path)) {
+        return NULL;
+    }
+    U64 res = NULL;
+    U8* json_string = FileRead(path);
+    if (json_string) {
+        res = @json_parse(json_string, mem_task);
+    }
+    return res;
+}
+
+U0 @json_dump_to_file(U8* path, @json_element* el, CTask* mem_task)
+{
+    if (!path || !el || !mem_task)
+        return;
+    U8* json_string = @json_stringify(el, mem_task);
+    FileWrite(path, json_string, StrLen(json_string));
+}
+
+@json_array* @json_element_value_as_array(@json_element* el, CTask* mem_task, I64 key_or_item)
+{
+    if (!el || !mem_task || !key_or_item) {
+        return NULL;
+    }
+    switch (el->type) {
+    case JSON_ARRAY:
+        switch (key_or_item) {
+        case JSON_ELEMENT_IS_ITEM:
+            return el(@json_item*)->value;
+        case JSON_ELEMENT_IS_KEY:
+            return el(@json_key*)->value;
+        default:
+            return NULL;
+        }
+        break;
+    default:
+        break;
+    }
+    @json_array* arr = CAlloc(sizeof(@json_array), mem_task);
+    arr->mem_task = mem_task;
+    arr->sig = JSON_SIG;
+    arr->type = JSON_ARRAY;
+    switch (key_or_item) {
+    case JSON_ELEMENT_IS_ITEM:
+        @json_append_item(arr, el(@json_item*)->value, el->type);
+        break;
+    case JSON_ELEMENT_IS_KEY:
+        @json_append_item(arr, el(@json_key*)->value, el->type);
+        break;
+    default:
+        break;
+    }
+    return @json_create_callable_array(arr);
+}
+
+@json_array* @json_item_value_as_array(@json_item* item, CTask* mem_task)
+{
+    return @json_element_value_as_array(item, mem_task, JSON_ELEMENT_IS_ITEM);
+}
+
+@json_array* @json_key_value_as_array(@json_key* key, CTask* mem_task)
+{
+    return @json_element_value_as_array(key, mem_task, JSON_ELEMENT_IS_KEY);
+}
+
+@json_element* @json_clone(@json_element* el, CTask* mem_task)
+{
+    if (!el || !mem_task) {
+        return NULL;
+    }
+    U8* tmp = @json_stringify(el, mem_task);
+    if (!tmp) {
+        return NULL;
+    }
+    return @json_parse(tmp, mem_task);
+}
+
+class @json
+{
+    @json_element* (*Clone)(@json_element* el, CTask* mem_task);
+    @json_array* (*CreateArray)(CTask* mem_task);
+    @json_object* (*CreateObject)(CTask* mem_task);
+    U0 (*DumpToFile)(U8* path, @json_element* el, CTask* mem_task);
+    @json_array* (*KeyValueAsArray)(@json_key* key, CTask* mem_task);
+    @json_array* (*ItemValueAsArray)(@json_item* item, CTask* mem_task);
+    @json_element* (*Parse)(U8* str, CTask* mem_task);
+    U64 (*ParseFile)(U8* path, CTask* mem_task);
+    U8* (*Stringify)(@json_element* el, CTask* mem_task);
+};
+
+@json Json;
+Json.Clone = &@json_clone;
+Json.CreateArray = &@json_create_array;
+Json.CreateObject = &@json_create_object;
+Json.DumpToFile = &@json_dump_to_file;
+Json.ItemValueAsArray = &@json_item_value_as_array;
+Json.KeyValueAsArray = &@json_key_value_as_array;
+Json.Parse = &@json_parse;
+Json.ParseFile = &@json_parse_file;
+Json.Stringify = &@json_stringify;
+
+"json ";
diff --git a/System/Libraries/String.HC b/System/Libraries/String.HC
new file mode 100644
index 0000000..bb568f9
--- /dev/null
+++ b/System/Libraries/String.HC
@@ -0,0 +1,196 @@
+#define TRIM_BOTH 0
+#define TRIM_LEFT 1
+#define TRIM_RIGHT 2
+
+I64 StrLenI32(I32* str)
+{
+    I64 res = 0;
+    while (*str) {
+        ++res;
+        ++str;
+    }
+    return res;
+}
+
+U0 @string_append(U8* dst, U8* fmt, ...)
+{
+    U8* buf;
+    if (argc) {
+        buf = StrPrintJoin(NULL, fmt, argc, argv);
+    } else {
+        buf = StrNew(fmt, adam_task);
+    }
+    U8* src = buf;
+    StrCpy(dst + StrLen(dst), src);
+    Free(buf);
+}
+
+Bool @string_is_number(U8* s)
+{
+    while (*s) {
+        switch (*s) {
+        case '-':
+        case '.':
+        case '0' ... '9':
+            break;
+        default:
+            return FALSE;
+            break;
+        }
+        s++;
+    }
+    return TRUE;
+}
+
+Bool @string_begins_with(U8* fragment, U8* str)
+{
+    if (!fragment || !str)
+        return FALSE;
+    if (StrLen(fragment) > StrLen(str))
+        return FALSE;
+    return !MemCmp(fragment, str, StrLen(fragment));
+}
+
+Bool @string_ends_with(U8* fragment, U8* str)
+{
+    if (!fragment || !str)
+        return FALSE;
+    if (StrLen(fragment) > StrLen(str))
+        return FALSE;
+    return !MemCmp(fragment, str + StrLen(str) - StrLen(fragment), StrLen(fragment));
+}
+
+U0 @string_lower(U8* s)
+{
+    while (*s) {
+        switch (*s) {
+        case 'A' ... 'Z':
+            *s += 32;
+            break;
+        default:
+            break;
+        }
+        ++s;
+    }
+}
+
+U8* @string_replace(U8* s, U8* oldW, U8* newW)
+{
+    if (!StrFind(oldW, s)) {
+        return StrNew(s, adam_task);
+    }
+    U8* result;
+    I64 i, cnt = 0;
+    I64 newWlen = StrLen(newW);
+    I64 oldWlen = StrLen(oldW);
+    for (i = 0; s[i] != NULL; i++) {
+        if (StrFind(oldW, &s[i]) == &s[i]) {
+            cnt++;
+
+            i += oldWlen - 1;
+        }
+    }
+    result = MAlloc(i + cnt * (newWlen - oldWlen) + 1, adam_task);
+    i = 0;
+    while (*s) {
+        if (StrFind(oldW, s) == s) {
+            StrCpy(&result[i], newW);
+            i += newWlen;
+            s += oldWlen;
+        } else
+            result[i++] = *s++;
+    }
+    result[i] = NULL;
+    return result;
+}
+
+U8** @string_split(U8* s, U8 ch = '\n', I64* cnt)
+{
+    U8 check_buf[4];
+    StrPrint(check_buf, "%c", ch);
+    if (!StrFind(check_buf, s)) {
+        U8** same_arr = CAlloc(sizeof(U8*) * 1, adam_task);
+        same_arr[0] = s;
+        *cnt = 1;
+        return same_arr;
+    }
+    U8* p = s;
+    cnt[0] = 0;
+    while (*p) {
+        if (*p == ch)
+            cnt[0]++;
+        p++;
+    }
+    if (!(cnt[0]))
+        return NULL;
+    cnt[0]++;
+    I64 i = -1;
+    U8** arr = CAlloc(sizeof(U8*) * cnt[0], adam_task);
+    p = s;
+    while (*p) {
+        if (*p == ch || i < 0) {
+            i++;
+            arr[i] = p;
+            if (*p == ch) {
+                arr[i]++;
+                *p = NULL;
+            }
+        }
+        p++;
+    }
+    return arr;
+}
+
+Bool @string_trim_ch(U8 s_ch, U8 trim_ch)
+{
+    if (!s_ch) {
+        return FALSE;
+    }
+    if (!trim_ch) {
+        return (s_ch == ' ' || s_ch == '\r' || s_ch == '\n' || s_ch == '\t');
+    } else {
+        return (s_ch == trim_ch);
+    }
+}
+
+U0 @string_trim(U8* s, U8 ch = NULL, I64 mode = TRIM_BOTH)
+{
+    Bool trim_ch = @string_trim_ch(*s, ch);
+    if (mode == TRIM_BOTH || mode == TRIM_LEFT) {
+        while (trim_ch) {
+            StrCpy(s, s + 1);
+            trim_ch = @string_trim_ch(*s, ch);
+        }
+    }
+    trim_ch = @string_trim_ch(s[StrLen(s) - 1], ch);
+    if (mode == TRIM_BOTH || mode == TRIM_RIGHT) {
+        while (trim_ch) {
+            s[StrLen(s) - 1] = NULL;
+            trim_ch = @string_trim_ch(s[StrLen(s) - 1], ch);
+        }
+    }
+}
+
+class @string
+{
+    U0 (*Append)(U8* dst, U8* fmt, ...);
+    Bool (*BeginsWith)(U8* fragment, U8* str);
+    Bool (*EndsWith)(U8* fragment, U8* str);
+    Bool (*IsNumber)(U8* s);
+    U0 (*Lower)(U8* s);
+    U8* (*Replace)(U8* s, U8* oldW, U8* newW);
+    U8** (*Split)(U8* s, U8 ch = '\n', I64 * cnt);
+    U0 (*Trim)(U8* s, U8 ch = NULL, I64 mode = TRIM_BOTH);
+};
+
+@string String;
+String.Append = &@string_append;
+String.BeginsWith = &@string_begins_with;
+String.EndsWith = &@string_ends_with;
+String.IsNumber = &@string_is_number;
+String.Lower = &@string_lower;
+String.Replace = &@string_replace;
+String.Split = &@string_split;
+String.Trim = &@string_trim;
+
+"string ";
diff --git a/System/MakeSystem.HC b/System/MakeSystem.HC
new file mode 100644
index 0000000..2caf4fe
--- /dev/null
+++ b/System/MakeSystem.HC
@@ -0,0 +1,56 @@
+/* clang-format off */
+
+#include "Setup/Util";
+//#include "Setup/MAlloc";
+#include "Setup/Environment";
+
+// FFI support files
+#include "FFI/Base";
+#include "FFI/LibC";
+#include "FFI/New";
+#include "FFI/ELF64";
+
+// stb_image library
+#include "Utilities/Image";
+load_elf("M:/build/bin/image");
+
+// Jakt support files
+#include "Jakt/OS";
+#include "Jakt/IOPort";
+#include "Jakt/PCI";
+#include "Jakt/Time";
+
+// Networking APIs
+#include "Api/Dns.HC";
+#include "Api/Icmp.HC";
+#include "Api/Ipv4.HC";
+#include "Api/MD5.HC";
+#include "Api/NetInfo.HC";
+#include "Api/Tcp.HC";
+//#include "Api/Tls.HC";
+
+#include "Libraries/Json";
+#include "Libraries/String";
+
+load_elf("M:/build/bin/net");
+
+// Networking Utilities
+#include "Utilities/Dns";
+#include "Utilities/NetRep";
+#include "Utilities/Ping";
+//#include "Utilities/Time";
+
+Spawn(_start, , "Net Task", NET_TASK_CPU);
+
+#include "9P/Debug";
+#include "9P/Defs";
+#include "9P/Draw";
+#include "9P/Requests";
+
+#include "NeinTerm";
+
+//#include "NeinTask";
+
+//TimeSync;
+
+/* clang-format on */
diff --git a/System/Setup/Environment.HC b/System/Setup/Environment.HC
new file mode 100644
index 0000000..9f9735e
--- /dev/null
+++ b/System/Setup/Environment.HC
@@ -0,0 +1,23 @@
+#define NET_TASK_CPU 3
+
+// Before continuing, we:
+
+// 1. Mark memory in code heap below 0x1000000 as used.
+sys_code_bp->mem_free_lst->next->pags = 0;
+
+// 2. Free up 64MB at bottom of code heap for non-HolyC programs
+sys_code_bp->mem_free_lst = ShrinkMemBlkByPags(sys_code_bp->mem_free_lst, 131072);
+
+U0 NoBeep(I8, Bool) {};
+@patch_jmp_rel32(&Beep, &NoBeep); // Don't delay on beep when entering debugger
+//@patch_jmp_rel32(&Fault2, &Reboot); // Reboot instead of crashing to the debugger
+
+// 4. Disable exclusive access to BlkDev and Drv
+
+Bool FakeLock(U64)
+{
+    return TRUE;
+}
+
+@patch_jmp_rel32(&BlkDevLock, &FakeLock);
+@patch_jmp_rel32(&DrvLock, &FakeLock);
diff --git a/System/Setup/Util.HC b/System/Setup/Util.HC
new file mode 100644
index 0000000..0f79800
--- /dev/null
+++ b/System/Setup/Util.HC
@@ -0,0 +1,180 @@
+AutoComplete(0);
+
+U0 @sse_enable()
+{
+    /* clang-format off */
+    asm
+    {
+        MOV_EAX_CR0
+        AND AX, 0xFFFB // clear coprocessor emulation CR0.EM
+        OR AX, 0x2 // set coprocessor monitoring  CR0.MP
+        MOV_CR0_EAX
+        MOV_EAX_CR4
+        OR AX, 3 << 9 // set CR4.OSFXSR and CR4.OSXMMEXCPT at the same time
+        MOV_CR4_EAX
+    }
+    /* clang-format on */
+}
+
+U0 @sse_enable_on_all_cores()
+{
+    I64 i;
+    for (i = 1; i < mp_cnt; i++)
+        Spawn(&@sse_enable, , , i);
+}
+
+// Enable SSE
+@sse_enable;
+@sse_enable_on_all_cores;
+
+U0 @patch_call_rel32(U32 from, U32 to)
+{
+    *(from(U8*)) = 0xE8;
+    *((from + 1)(I32*)) = to - from - 5;
+}
+
+U0 @patch_jmp_rel32(U32 from, U32 to)
+{
+    *(from(U8*)) = 0xE9;
+    *((from + 1)(I32*)) = to - from - 5;
+}
+
+I64 tos_nist_offset = 5603; // UTC -4
+#define NIST_TIME_OFFSET (tos_nist_offset - local_time_offset / CDATE_FREQ)
+
+public
+I64 CDate2Unix(CDate dt)
+{ // TempleOS datetime to Unix timestamp.
+    return ToI64((dt - Str2Date("1/1/1970")) / CDATE_FREQ + NIST_TIME_OFFSET);
+}
+
+public
+CDate Unix2CDate(I64 timestamp)
+{ // Unix timestamp to TempleOS datetime.
+    return (timestamp - NIST_TIME_OFFSET) * CDATE_FREQ + Str2Date("1/1/1970");
+}
+
+// FIXME: Put these in a "Builtin" library?
+U0 FifoU8Cpy(CFifoU8* f, U8* s)
+{
+    if (!f || !s)
+        return;
+    while (*s)
+        FifoU8Ins(f, *s++);
+}
+Bool KeyDown(I64 sc) return Bt(kbd.down_bitmap, sc);
+I64 T(Bool _condition, I64 _true, I64 _false)
+{
+    if (_condition)
+        return _true;
+    return _false;
+}
+
+asm
+    {
+_MEMCPY_U16::
+        PUSH    RBP
+        MOV     RBP,RSP
+        PUSH    RSI
+        PUSH    RDI
+        CLD
+        MOV     RDI,U64 SF_ARG1[RBP]
+        MOV     RSI,U64 SF_ARG2[RBP]
+        MOV     RCX,U64 SF_ARG3[RBP]
+        REP_MOVSW
+        MOV     RAX,RDI
+        POP     RDI
+        POP     RSI
+        POP     RBP
+        RET1    24    
+_MEMCPY_U32::
+        PUSH    RBP
+        MOV     RBP,RSP
+        PUSH    RSI
+        PUSH    RDI
+        CLD
+        MOV     RDI,U64 SF_ARG1[RBP]
+        MOV     RSI,U64 SF_ARG2[RBP]
+        MOV     RCX,U64 SF_ARG3[RBP]
+        REP_MOVSD
+        MOV     RAX,RDI
+        POP     RDI
+        POP     RSI
+        POP     RBP
+        RET1    24    
+_MEMCPY_U64::
+        PUSH    RBP
+        MOV     RBP,RSP
+        PUSH    RSI
+        PUSH    RDI
+        CLD
+        MOV     RDI,U64 SF_ARG1[RBP]
+        MOV     RSI,U64 SF_ARG2[RBP]
+        MOV     RCX,U64 SF_ARG3[RBP]
+        REP_MOVSQ
+        MOV     RAX,RDI
+        POP     RDI
+        POP     RSI
+        POP     RBP
+        RET1    24
+    }
+
+public _extern _MEMCPY_U16 U16* MemCpyU16(U16* dst, U16* src, I64 cnt);
+public
+_extern _MEMCPY_U32 U32* MemCpyU32(U32* dst, U32* src, I64 cnt);
+public
+_extern _MEMCPY_U64 U64* MemCpyU64(U64* dst, U64* src, I64 cnt);
+
+I64 @lerp(U32 val, U32 mx1, U32 mx2)
+{
+    F64 r = (val & mx1) / ToF64(mx1);
+    return ToI64(r * mx2);
+}
+
+CMemBlk* ShrinkMemBlkByPags(CMemBlk* from, I64 count)
+{
+    from->pags -= count;
+    U64 to = from;
+    to += count * MEM_PAG_SIZE;
+    MemCpy(to, from, MEM_PAG_SIZE);
+    return to;
+}
+
+I64 @t(Bool _condition, I64 _true, I64 _false)
+{
+    if (_condition)
+        return _true;
+    return _false;
+}
+
+U0 dd() { DocDump(adam_task->put_doc); }
+
+Bool FifoU8Last(CFifoU8* f, U8* _b)
+{ // Peek at back of fifo and don't remove.
+    PUSHFD
+    CLI if (f->in_ptr == f->out_ptr)
+    {
+        POPFD
+        return FALSE;
+    }
+    else
+    {
+        I64 last_ptr = f->in_ptr - 1;
+        if (last_ptr < 0)
+            last_ptr = FifoU8Cnt(f) - 1;
+        *_b = f->buf[last_ptr];
+        POPFD
+        return TRUE;
+    }
+}
+
+U8 reverse_bits(U8 n)
+{
+    // Swap nibbles
+    n = ((n >> 4) & 0x0F) | ((n & 0x0F) << 4);
+    // Swap adjacent pairs
+    n = ((n >> 2) & 0x33) | ((n & 0x33) << 2);
+    // Swap adjacent bits
+    n = ((n >> 1) & 0x55) | ((n & 0x55) << 1);
+    return n;
+}
diff --git a/System/Utilities/Dns.HC b/System/Utilities/Dns.HC
new file mode 100644
index 0000000..42c97a7
--- /dev/null
+++ b/System/Utilities/Dns.HC
@@ -0,0 +1,10 @@
+U0 DnsQuery(U8* host)
+{
+    U32 result = @dns_query(host);
+    if (result == U32_MAX) {
+        "Error looking up host %s\n", host;
+        return;
+    }
+    "Query for %s: %d.%d.%d.%d\n", host, result.u8[3], result.u8[2], result.u8[1],
+        result.u8[0];
+}
diff --git a/System/Utilities/Image.HC b/System/Utilities/Image.HC
new file mode 100644
index 0000000..b0d89ee
--- /dev/null
+++ b/System/Utilities/Image.HC
@@ -0,0 +1,414 @@
+Silent(1); // This is needed to suppress "Function should return val" warnings for wrappers to non-HolyC functions
+
+// class @image
+// {
+//     CDC* (*FromBuffer)(U8* buffer, I64 len);
+//     CDC* (*Load)(U8* filename);
+//     CDC* (*Write)(U8* filename, CDC* dc);
+// };
+//
+// @image Image;
+
+class @image_frame
+{
+    CDC* dc;
+    CSprite* sprite;
+    I64 delay;
+};
+
+class @image_collection
+{
+    @image_frame** frames;
+    I64 count;
+    I64 current;
+    I64 jiffies;
+    I64 index;
+    @image_collection* next;
+};
+
+I64 @image_cbgr24_to_4_bit(CBGR24* ptr, Bool dither_probability)
+{
+    I64 res, k;
+    if (dither_probability) {
+        k = RandU32;
+        if (SqrI64(ptr->r) + SqrI64(ptr->g) + SqrI64(ptr->b) >= 3 * SqrI64(k.u8[0]))
+            res = 8;
+        else
+            res = 0;
+        if (ptr->r >= k.u8[1])
+            res |= RED;
+        if (ptr->g >= k.u8[2])
+            res |= GREEN;
+        if (ptr->b >= k.u8[3])
+            res |= BLUE;
+    } else {
+        if (SqrI64(ptr->r) + SqrI64(ptr->g) + SqrI64(ptr->b) >= SqrI64(0x80)) {
+            res = 8;
+            if (ptr->r >= 0x80)
+                res |= RED;
+            if (ptr->g >= 0x80)
+                res |= GREEN;
+            if (ptr->b >= 0x80)
+                res |= BLUE;
+        } else {
+            res = 0;
+            if (ptr->r >= 0x40)
+                res |= RED;
+            if (ptr->g >= 0x40)
+                res |= GREEN;
+            if (ptr->b >= 0x40)
+                res |= BLUE;
+        }
+    }
+    return res;
+}
+
+#define IMAGE_DITHER_NONE 0
+#define IMAGE_DITHER_NATIVE 1
+#define IMAGE_DITHER_FLOYDSTEINBERG 2
+
+U0 @image_render_4bit_floydstein(U8* buffer, I32 width, I32 height)
+{
+    U64 reg RDI rdi = buffer;
+    U64 reg RSI rsi = width;
+    U64 reg RDX rdx = height;
+    no_warn rdi, rsi, rdx;
+    asm {
+        MOV RAX, RENDER_4BIT_FLOYDSTEIN
+        CALL RAX
+    }
+}
+
+CDC* @image_render_16color_native(U8* pixels, I32 x, I32 y, Bool dither)
+{
+    I64 i;
+    I64 j;
+    I64 cnt = 0;
+    CBGR24 cbgr24;
+    CDC* dc = DCNew(x, y);
+    for (i = 0; i < y; i++)
+        for (j = 0; j < x; j++) {
+            cbgr24.r = pixels[cnt];
+            cbgr24.g = pixels[cnt + 1];
+            cbgr24.b = pixels[cnt + 2];
+            if (!pixels[cnt + 3])
+                dc->color = TRANSPARENT;
+            else
+                dc->color = @image_cbgr24_to_4_bit(&cbgr24, dither);
+            GrPlot(dc, j, y - i - 1);
+            cnt += 4;
+        }
+    return dc;
+}
+
+CBGR24 @image_palette_std[COLORS_NUM] = {
+    0x000000, 0x0000AA, 0x00AA00, 0x00AAAA,
+    0xAA0000, 0xAA00AA, 0xAA5500, 0xAAAAAA,
+    0x555555, 0x5555FF, 0x55FF55, 0x55FFFF,
+    0xFF5555, 0xFF55FF, 0xFFFF55, 0xFFFFFF
+};
+
+CBGR24 @image_dif_rgb(CBGR24 from, CBGR24 to)
+{
+    CBGR24 dif;
+    dif.r = to.r - from.r;
+    dif.g = to.g - from.g;
+    dif.b = to.b - from.b;
+    return dif;
+}
+
+F64 @image_dist_rgb(CBGR24 from, CBGR24 to)
+{
+    CBGR24 dif = @image_dif_rgb(from, to);
+    F64 dist = dif.r * dif.r + dif.g * dif.g + dif.b * dif.b;
+    return dist;
+}
+
+I64 @image_get_4bit_color(CBGR24* cbgr24)
+{
+    F64 dist = -1, tempDist;
+    I64 i;
+    I64 color = TRANSPARENT;
+    for (i = 0; i < COLORS_NUM; i++) {
+        tempDist = @image_dist_rgb(*cbgr24, @image_palette_std[i]);
+        if (tempDist < dist || dist < 0) {
+            dist = tempDist;
+            color = i;
+        }
+    }
+    return color;
+}
+
+CDC* @image_render_16color_floydsteinberg(U8* pixels, I32 width, I32 height)
+{
+    @image_render_4bit_floydstein(pixels, width, height);
+    I64 i;
+    I64 j;
+    I64 cnt = 0;
+    CBGR24 cbgr24;
+    CDC* dc = DCNew(width, height);
+    for (i = 0; i < height; i++)
+        for (j = 0; j < width; j++) {
+            cbgr24.r = pixels[cnt];
+            cbgr24.g = pixels[cnt + 1];
+            cbgr24.b = pixels[cnt + 2];
+            if (!pixels[cnt + 3])
+                dc->color = TRANSPARENT;
+            else
+                dc->color = @image_get_4bit_color(&cbgr24);
+            GrPlot(dc, j, height - i - 1);
+            cnt += 4;
+        }
+    return dc;
+}
+
+CDC* @image_generate_dc_from_pixels(U8* pixels, I32 width, I32 height, Bool dither = IMAGE_DITHER_FLOYDSTEINBERG)
+{
+    switch (dither) {
+    case IMAGE_DITHER_NONE:
+    case IMAGE_DITHER_NATIVE:
+        return @image_render_16color_native(pixels, width, height, dither);
+        break;
+    case IMAGE_DITHER_FLOYDSTEINBERG:
+        return @image_render_16color_floydsteinberg(pixels, width, height);
+        break;
+    default:
+        break;
+    }
+    return NULL;
+}
+
+U8* @image_load_gif_from_memory(U8* buffer, I64 len, I64** delays, I64* x, I64* y,
+    I64* z)
+{
+    U64 reg RDI rdi = buffer;
+    U64 reg RSI rsi = len;
+    U64 reg RDX rdx = delays;
+    U64 reg RCX rcx = x;
+    U64 reg R8 r8 = y;
+    U64 reg R9 r9 = z;
+    no_warn rdi, rsi, rdx, rcx, r8, r9;
+    asm {
+        MOV RAX, IMAGE_LOAD_GIF_FROM_MEMORY
+        CALL RAX
+    }
+}
+
+U8* @stbi_failure_reason()
+{
+    asm {
+        MOV RAX, STBI_FAILURE_REASON
+        CALL RAX
+    }
+}
+
+I32 @stbi_info_from_memory(U8* buffer, I64 len, I64* x, I64* y, I64* comp)
+{
+    U64 reg RDI rdi = buffer;
+    U64 reg RSI rsi = len;
+    U64 reg RDX rdx = x;
+    U64 reg RCX rcx = y;
+    U64 reg R8 r8 = comp;
+    no_warn rdi, rsi, rdx, rcx, r8;
+    asm {
+        MOV RAX, STBI_INFO_FROM_MEMORY
+        CALL RAX
+    }
+}
+
+U8* @stbi_load_from_memory(U8* buffer, I64 len, I64* x, I64* y,
+    I64* channels_in_file, I64 desired_channels)
+{
+    U64 reg RDI rdi = buffer;
+    U64 reg RSI rsi = len;
+    U64 reg RDX rdx = x;
+    U64 reg RCX rcx = y;
+    U64 reg R8 r8 = channels_in_file;
+    U64 reg R9 r9 = desired_channels;
+    no_warn rdi, rsi, rdx, rcx, r8, r9;
+    asm {
+        MOV RAX, STBI_LOAD_FROM_MEMORY
+        CALL RAX
+    }
+}
+
+U32* @stbi_write_png_to_mem(U32* pixels, I32 stride_bytes, I32 x, I32 y, I32 n, I32* out_len)
+{
+    U64 reg RDI rdi = pixels;
+    U64 reg RSI rsi = stride_bytes;
+    U64 reg RDX rdx = x;
+    U64 reg RCX rcx = y;
+    U64 reg R8 r8 = n;
+    U64 reg R9 r9 = out_len;
+    no_warn rdi, rsi, rdx, rcx, r8, r9;
+    asm {
+        MOV RAX, STBI_WRITE_PNG_TO_MEM
+        CALL RAX
+    }
+}
+
+CDC* @image_load(U8* filename)
+{
+    if (!filename || !FileFind(filename)) {
+        // PrintErr("Image file not found.\n");
+        return NULL;
+    }
+    I64 len;
+    I32 x;
+    I32 y;
+    I32 comp;
+    U8* buffer = FileRead(filename, &len);
+    I32 code = @stbi_info_from_memory(buffer, len, &x, &y, &comp);
+    if (code != 1) {
+        Free(buffer);
+        return NULL;
+    }
+    U8* pixels = @stbi_load_from_memory(buffer, len, &x, &y, &comp, 4);
+    Free(buffer);
+    CDC* dc = @image_generate_dc_from_pixels(pixels, x, y);
+    Free(pixels);
+    return dc;
+}
+
+U32 @image_rgba_color_table[16] = {
+    0xff000000, 0xffaa0000, 0xff00aa00, 0xffaaaa00,
+    0xff0000aa, 0xffaa00aa, 0xff0055aa, 0xffaaaaaa,
+    0xff555555, 0xffff5555, 0xff55ff55, 0xffffff55,
+    0xff5555ff, 0xffff55ff, 0xff55ffff, 0xffffffff
+};
+
+U32 @image_get_rgba_color(I64 color)
+{
+    if (color > 15)
+        return 0;
+    return @image_rgba_color_table[color];
+}
+
+U32* @image_get_rgba_buffer_from_dc_body(CDC* dc)
+{
+    if (!dc)
+        return NULL;
+    U32* pixels = CAlloc((dc->width * dc->height) * 4, adam_task);
+    I64 x;
+    I64 y;
+    I64 p = 0;
+    for (y = 0; y < dc->height; y++)
+        for (x = 0; x < dc->width; x++)
+            pixels[p++] = @image_get_rgba_color(GrPeek(dc, x, y));
+    return pixels;
+}
+
+U0 @image_write(U8* filename, CDC* dc)
+{
+    if (!dc) {
+        PrintErr("Device context is NULL.\n");
+        return;
+    }
+    I32 out_len;
+    U32* rgba_buffer = @image_get_rgba_buffer_from_dc_body(dc);
+    if (!rgba_buffer) {
+        PrintErr("RGBA buffer is NULL.\n");
+        return;
+    }
+    U8* png_buffer = @stbi_write_png_to_mem(rgba_buffer, dc->width * 4, dc->width, dc->height, 4, &out_len);
+    if (!png_buffer) {
+        PrintErr("PNG buffer is NULL.\n");
+        Free(rgba_buffer);
+        return;
+    }
+    FileWrite(filename, png_buffer, out_len);
+    Free(rgba_buffer);
+    Free(png_buffer);
+}
+
+U32 @image_pixel_flip_rgb_bgr(U32 src)
+{
+    U32 dst;
+    dst.u8[0] = src.u8[2];
+    dst.u8[1] = src.u8[1];
+    dst.u8[2] = src.u8[0];
+    dst.u8[3] = src.u8[3];
+    return dst;
+}
+
+CDC* @image_from_buffer(U8* buffer, I64 len)
+{
+    I32 x = 0;
+    I32 y = 0;
+    U8* pixels = NULL;
+    CDC* dc = NULL;
+
+    I32 comp;
+    I32 code = @stbi_info_from_memory(buffer, len, &x, &y, &comp);
+    if (code != 1) {
+        return NULL;
+    }
+    pixels = @stbi_load_from_memory(buffer, len, &x, &y, &comp, 4);
+    if (!pixels)
+        PopUpOk(@stbi_failure_reason);
+    dc = @image_generate_dc_from_pixels(pixels, x, y);
+    Free(pixels);
+    return dc;
+}
+
+@image_collection* @image_collection_from_buffer(U8* buffer, I64 len)
+{
+    I64 i;
+    I32* delays;
+    I32 x;
+    I32 y;
+    I32 z;
+    I32 comp;
+    I32 code = @stbi_info_from_memory(buffer, len, &x, &y, &comp);
+    if (code != 1) {
+        return NULL;
+    }
+    U64 pixels = @image_load_gif_from_memory(buffer, len, &delays, &x, &y, &z);
+    if (!pixels)
+        PopUpOk(@stbi_failure_reason);
+    if (!z)
+        return NULL; // no frames?
+    @image_collection* collection = CAlloc(sizeof(@image_collection), adam_task);
+    @image_frame* frame;
+    collection->frames = CAlloc(sizeof(@image_frame*) * z, adam_task);
+    collection->count = z;
+    for (i = 0; i < z; i++) {
+        frame = CAlloc(sizeof(@image_frame), adam_task);
+        frame->dc = @image_generate_dc_from_pixels(pixels, x, y);
+        frame->sprite = DC2Sprite(frame->dc);
+        frame->delay = delays[i];
+        collection->frames[i] = frame;
+        pixels += (x * y) * 4;
+    }
+    return collection;
+}
+
+// Image.FromBuffer = &@image_from_buffer;
+// Image.Load = &@image_load;
+// Image.Write = &@image_write;
+
+Silent(0);
+
+U0 Screenshot(U8* custom_filename = NULL, Bool output_filename_to_focus_task = FALSE)
+{
+    CDC* dc = DCScrnCapture;
+    U8 filename[256];
+    CDateStruct ds;
+    if (custom_filename)
+        StrCpy(filename, custom_filename);
+    else {
+        Date2Struct(&ds, Now);
+        StrPrint(filename, "C:/Tmp/ScrnShots/%04d-%02d-%02d-%02d-%02d-%02d.png", ds.year, ds.mon, ds.day_of_mon, ds.hour, ds.min, ds.sec);
+    }
+    @image_write(filename, dc);
+    DCDel(dc);
+    if (output_filename_to_focus_task)
+        XTalk(sys_focus_task, filename);
+};
+
+U0 @screenshot_hotkey(I64)
+{
+    Screenshot("C:/Home/Screenshot.png", TRUE);
+}
+
+CtrlAltCBSet('S', &@screenshot_hotkey, "", , FALSE);
diff --git a/System/Utilities/NetRep.HC b/System/Utilities/NetRep.HC
new file mode 100644
index 0000000..f56bacd
--- /dev/null
+++ b/System/Utilities/NetRep.HC
@@ -0,0 +1,22 @@
+U0 NetRep()
+{
+    NetInfoRequest* req = @net_info_request;
+    "MAC address       : %02x:%02x:%02x:%02x:%02x:%02x\n", req->mac_address.u8[5], req->mac_address.u8[4],
+        req->mac_address.u8[3], req->mac_address.u8[2],
+        req->mac_address.u8[1], req->mac_address.u8[0];
+    "IPv4 address      : %d.%d.%d.%d\n", req->ipv4_address.u8[3], req->ipv4_address.u8[2],
+        req->ipv4_address.u8[1], req->ipv4_address.u8[0];
+    "IPv4 netmask      : %d.%d.%d.%d\n", req->ipv4_netmask.u8[3], req->ipv4_netmask.u8[2],
+        req->ipv4_netmask.u8[1], req->ipv4_netmask.u8[0];
+    "IPv4 network      : %d.%d.%d.%d\n", req->ipv4_network.u8[3], req->ipv4_network.u8[2],
+        req->ipv4_network.u8[1], req->ipv4_network.u8[0];
+    "IPv4 gateway      : %d.%d.%d.%d\n", req->ipv4_gateway.u8[3], req->ipv4_gateway.u8[2],
+        req->ipv4_gateway.u8[1], req->ipv4_gateway.u8[0];
+    "DNS server (port) : %d.%d.%d.%d (%d)\n", req->dns_server_address.u8[3], req->dns_server_address.u8[2],
+        req->dns_server_address.u8[1], req->dns_server_address.u8[0], req->dns_server_port;
+    "RX bytes          : %d\n", req->rx_bytes;
+    "RX frames         : %d\n", req->rx_frames;
+    "TX bytes          : %d\n", req->tx_bytes;
+    "TX frames         : %d\n", req->tx_frames;
+    Free(req);
+}
diff --git a/System/Utilities/Ping.HC b/System/Utilities/Ping.HC
new file mode 100644
index 0000000..386c441
--- /dev/null
+++ b/System/Utilities/Ping.HC
@@ -0,0 +1,72 @@
+#define PING_ERR_INVALID_HOST 1
+#define PING_ERR_HOST_NOT_FOUND 2
+
+#define PING_PAYLOAD_SIZE 56
+
+I64 @ping_err(I64 code)
+{
+    switch (code) {
+    case PING_ERR_INVALID_HOST:
+        "Invalid host specified\n";
+        return 1;
+        break;
+    case PING_ERR_HOST_NOT_FOUND:
+        "Host not found\n";
+        return 2;
+        break;
+    default:
+        "Unspecified error\n";
+        return -1;
+    }
+}
+
+I64 Ping(U8* host, I64 count = 4)
+{
+    if (!host)
+        return @ping_err(PING_ERR_INVALID_HOST);
+    if (!StrLen(host))
+        return @ping_err(PING_ERR_INVALID_HOST);
+
+    U32 addr = @dns_query(host);
+    if (addr == U32_MAX)
+        return @ping_err(PING_ERR_HOST_NOT_FOUND);
+
+    U16 iden = (RandU16 * SysTimerRead) & 0xFFFF;
+    I64 start_jiffies;
+    U32 reply = NULL;
+    I64 res = 0;
+    U16 seq = 0;
+    I64 loss = 0;
+
+    IcmpRequest* request = CAlloc(sizeof(IcmpRequest), Fs->code_heap);
+
+    "PING %s (%d.%d.%d.%d): %d data bytes\n",
+        host, addr.u8[3], addr.u8[2], addr.u8[1], addr.u8[0], PING_PAYLOAD_SIZE;
+
+    I64 i;
+    for (i = 0; i < count; i++) {
+        start_jiffies = cnts.jiffies;
+        reply = @icmp_echo_request(addr, iden, seq, request, i);
+        if (!reply) {
+            "Request timeout for icmp_seq %d\n", seq;
+            ++loss;
+            res = 1;
+        } else {
+            "%d bytes from %d.%d.%d.%d: icmp_seq=%d ttl=%d time=%d ms\n",
+                reply.u16[1], addr.u8[3], addr.u8[2], addr.u8[1], addr.u8[0], seq, reply.u16[0], cnts.jiffies - start_jiffies;
+        }
+        while (cnts.jiffies < start_jiffies + 1000 && i < (count - 1))
+            Sleep(1);
+        ++seq;
+    }
+
+    Free(request);
+
+    "--- %d.%d.%d.%d ping statistics ---\n", addr.u8[3], addr.u8[2], addr.u8[1], addr.u8[0];
+    "%d packets transmitted, %d packets received, %0f",
+        seq, seq - loss, (loss * 1.0 / seq * 1.0) * 100;
+    PutChars(37);
+    " packet loss\n";
+
+    return res;
+}
\ No newline at end of file
diff --git a/System/Utilities/Time.HC b/System/Utilities/Time.HC
new file mode 100644
index 0000000..8025c76
--- /dev/null
+++ b/System/Utilities/Time.HC
@@ -0,0 +1,92 @@
+U0 @time_cmos_update_byte(I64 time_reg, I64 val)
+{
+    OutU8(0x70, time_reg);
+    OutU8(0x71, val);
+}
+
+I64 @time_dec_to_bcd(I64 val)
+{
+    return (((val / 10) << 4) | (val % 10));
+}
+
+U0 @time_update(U8* date_str, I64 mS_delta, I64 hour_offset)
+{
+    no_warn mS_delta;
+    Bool is_bcd;
+    OutU8(0x70, 0x0B);
+    if (InU8(0x71) & 4)
+        is_bcd = FALSE;
+    else
+        is_bcd = TRUE;
+
+    I64 date_argc;
+    U8** date_argv = String.Split(date_str, ' ', &date_argc);
+
+    I64 month = DefineMatch(date_argv[2], "ST_MONTHS") + 1;
+    I64 day = Str2I64(date_argv[1]);
+    I64 year = Str2I64(date_argv[3] + 2);
+    I64 century = 20;
+
+    date_argv[4][2] = NULL;
+    date_argv[4][5] = NULL;
+
+    I64 hour = Str2I64(date_argv[4]);
+    I64 minute = Str2I64(date_argv[4] + 3);
+    I64 second = Str2I64(date_argv[4] + 6);
+
+    // FIXME: Handle month boundaries, and 12 hour time
+    hour += hour_offset;
+    if (hour < 0) {
+        hour += 24;
+        --day;
+    } else if (hour > 23) {
+        hour -= 24;
+        ++day;
+    }
+
+    if (is_bcd) {
+        century = @time_dec_to_bcd(century);
+        year = @time_dec_to_bcd(year);
+        month = @time_dec_to_bcd(month);
+        day = @time_dec_to_bcd(day);
+        hour = @time_dec_to_bcd(hour);
+        minute = @time_dec_to_bcd(minute);
+        second = @time_dec_to_bcd(second);
+    }
+
+    @time_cmos_update_byte(0x32, century);
+    @time_cmos_update_byte(0x09, year);
+    @time_cmos_update_byte(0x08, month);
+    @time_cmos_update_byte(0x07, day);
+    @time_cmos_update_byte(0x04, hour);
+    @time_cmos_update_byte(0x02, minute);
+    @time_cmos_update_byte(0x00, second);
+}
+
+I64 @time_tz_offset()
+{
+    return -4;
+}
+
+U0 @time_query(Bool set = FALSE)
+{
+    U8 buf[1024];
+    @http_url* url = @http_parse_url("http://time.google.com");
+    @http_response* resp = Http.Head(url, &buf);
+    while (resp->state != HTTP_STATE_DONE)
+        Sleep(1);
+    I64 mS_delta = cnts.jiffies;
+    "Set current date and time to %s ", resp->headers->@("Date");
+    if (!set)
+        set = YorN;
+    else
+        "\n";
+    if (set)
+        @time_update(resp->headers->@("Date"), mS_delta, @time_tz_offset);
+}
+
+U0 TimeSync()
+{
+    Sleep(500);
+    @time_query(1);
+}
diff --git a/System/Utilities/TrueType.HC b/System/Utilities/TrueType.HC
new file mode 100644
index 0000000..9d9ab7f
--- /dev/null
+++ b/System/Utilities/TrueType.HC
@@ -0,0 +1,80 @@
+Silent(1); // This is needed to suppress "Function should return val" warnings for wrappers to non-HolyC functions
+
+class stbtt__buf {
+    U8* data;
+    I32 cursor;
+    I32 size;
+};
+
+class stbtt_fontinfo {
+    U8* userdata;
+    U8* data;                                          // pointer to .ttf file
+    I32 fontstart;                                     // offset of start of font
+    I32 numGlyphs;                                     // number of glyphs, needed for range checking
+    I32 loca, head, glyf, hhea, hmtx, kern, gpos, svg; // table locations as offset from start of .ttf
+    I32 index_map;                                     // a cmap mapping for our chosen character encoding
+    I32 indexToLocFormat;                              // format needed to map from glyph index to glyph
+
+    stbtt__buf cff;         // cff font data
+    stbtt__buf charstrings; // the charstring index
+    stbtt__buf gsubrs;      // global charstring subroutines index
+    stbtt__buf subrs;       // private charstring subroutines index
+    stbtt__buf fontdicts;   // array of font dicts
+    stbtt__buf fdselect;    // map from glyph to fontdict
+};
+
+I32 @stbtt_InitFont(stbtt_fontinfo* info, U8* data, I32 offset)
+{
+    U64 reg RDI rdi = info;
+    U64 reg RSI rsi = data;
+    U64 reg RDX rdx = offset;
+    no_warn rdi, rsi, rdx;
+    asm {
+        MOV RAX, STBTT_INITFONT
+        CALL RAX
+    }
+}
+
+U8* @stbtt_RenderText(stbtt_fontinfo* info, I32 b_w, I32 b_h, I32 l_h, I32* word, I32* advance = NULL)
+{
+    U64 reg RDI rdi = info;
+    U64 reg RSI rsi = b_w;
+    U64 reg RDX rdx = b_h;
+    U64 reg RCX rcx = l_h;
+    U64 reg R8 r8 = word;
+    U64 reg R9 r9 = advance;
+    no_warn rdi, rsi, rdx, rcx, r8, r9;
+    stbtt_pos = 0;
+    asm {
+        MOV RAX, STBTT_RENDERTEXT
+        CALL RAX
+    }
+}
+
+I32 @stbtt_GetTextWidth(stbtt_fontinfo* info, I32 l_h, I32* word, I32* advance = NULL)
+{
+    U64 reg RDI rdi = info;
+    U64 reg RSI rsi = l_h;
+    U64 reg RDX rdx = word;
+    U64 reg RCX rcx = advance;
+    no_warn rdi, rsi, rdx, rcx;
+    asm {
+        MOV RAX, STBTT_GETTEXTWIDTH
+        CALL RAX
+    }
+}
+
+U8* @stbtt_GetFontNameDefault(stbtt_fontinfo* font, I32* length)
+{
+    U64 reg RDI rdi = font;
+    U64 reg RSI rsi = length;
+    no_warn rdi, rsi;
+    asm {
+        MOV RAX, STBTT_GETFONTNAMEDEFAULT
+        CALL RAX
+    }
+}
+
+Silent(0);
+
+"truetype ";
diff --git a/script.rc b/script.rc
new file mode 100644
index 0000000..0f7eb64
--- /dev/null
+++ b/script.rc
@@ -0,0 +1,9 @@
+mount -nc /fd/0 /mnt/term || exit
+bind -q /mnt/term/dev/cons /dev/cons
+if(test -r /mnt/term/dev/kbd){
+	</dev/cons >/dev/cons >[2=1] aux/kbdfs -dq -m /mnt/term/dev
+	bind -q /mnt/term/dev/cons /dev/cons
+}
+</dev/cons >/dev/cons >[2=1] service=neinterm rc -li
+echo -n $status >/mnt/term/env/rstatus >[2]/dev/null"
+echo -n hangup >/proc/$pid/notepg
diff --git a/scripts/build-all b/scripts/build-all
new file mode 100755
index 0000000..922de74
--- /dev/null
+++ b/scripts/build-all
@@ -0,0 +1,207 @@
+#!/usr/bin/python3 -u
+from pathlib import Path
+import glob
+import json
+import os
+import subprocess
+import sys
+import time
+
+if len(sys.argv) < 2:
+    raise ValueError('wrong number of arguments')
+
+home_path = str(Path.home()) + '/'
+
+project_path = sys.argv[1] + '/'
+project_name = project_path.rsplit('/')[-2]
+
+build_options_file = project_path + 'build_options.json'
+
+try:
+    build_options = json.loads(open(build_options_file, "rb").read())
+except:
+    build_options = {}
+
+# Default settings
+
+default_settings = {
+    'isoc_file': project_path + 'build/isoc/NeinTerm.ISO.C',
+    'redsea_path': project_path + 'build/redsea',
+    'jakt_compiler_path': home_path + 'cloned/jakt/build/bin/jakt',
+    'jakt_runtime_path': home_path + 'cloned/jakt/runtime',
+    'jakt_lib_path': home_path + 'cloned/jakt/build/lib/x86_64-unknown-linux-unknown/',
+    'qemu_bin_path': 'qemu-system-x86_64',
+    'qemu_slipstream_iso_file': project_path + 'build/isoc/bootable.iso',
+    'templeos_iso_file': home_path + 'iso/TempleOS.ISO'
+}
+
+isoc_file = build_options['isoc_file'] if 'isoc_file' in build_options else default_settings['isoc_file']
+redsea_path = build_options['redsea_path'] if 'redsea_path' in build_options else default_settings['redsea_path']
+
+jakt_compiler_path = build_options['jakt_compiler_path'] if 'jakt_compiler_path' in build_options else default_settings['jakt_compiler_path']
+jakt_runtime_path = build_options['jakt_runtime_path'] if 'jakt_runtime_path' in build_options else default_settings['jakt_runtime_path']
+jakt_lib_path = build_options['jakt_lib_path'] if 'jakt_lib_path' in build_options else default_settings['jakt_lib_path']
+
+qemu_bin_path = build_options['qemu_bin_path'] if 'qemu_bin_path' in build_options else default_settings['qemu_bin_path']
+qemu_slipstream_iso_file = build_options['qemu_slipstream_iso_file'] if 'qemu_slipstream_iso_file' in build_options else default_settings['qemu_slipstream_iso_file']
+
+templeos_iso_file = build_options['templeos_iso_file'] if 'templeos_iso_file' in build_options else default_settings['templeos_iso_file']
+
+qemu_args = build_options['qemu_args'] if 'qemu_args' in build_options else [
+    '-display sdl,grab-mod=rctrl',
+    '-enable-kvm',
+    '-machine vmport=on',
+    '-smp cores=4',
+    '-m 4096',
+    '-netdev tap,id=mynet0,ifname=tap0,script=no,downscript=no',
+    '-device ac97',
+    '-device virtio-net,netdev=mynet0',
+    '-device vmware-svga',
+    '-cdrom ' + qemu_slipstream_iso_file,
+    '-debugcon stdio',
+    '-boot d'
+]
+
+qemu_run_cmd = ' '.join([qemu_bin_path] + qemu_args)
+
+def build_options_bool(key):
+    if key not in build_options:
+        return False
+    return build_options[key] == True
+
+def clang_format_src_files():
+    print("build-all: clang-format-src-files")
+    exclude_paths = ["stb_", "openlibm", "tlse", ".iso.c"]
+    format_file_extensions = [".c", ".cpp", ".h", ".hc"]
+    for src_file in glob.glob(project_path + "**", recursive=True):
+        exclude_file = False
+        for exclude_path in exclude_paths:
+            if src_file.lower().find(exclude_path) > 0:
+                exclude_file = True
+        if exclude_file:
+            continue
+        for format_file_extension in format_file_extensions:
+            if src_file.lower().endswith(format_file_extension):
+                print(src_file)
+                res = os.system('clang-format -i --style=file:' + project_path + '.clang-format ' + src_file)
+                if res:
+                    raise ValueError("build-all: step 'clang-format-src-files' failed, error code " + str(res))
+
+def refresh_build_path():
+    print("build-all: refresh-build-path")
+    res = os.system('rm -rf ' + project_path + 'build && mkdir -p ' + project_path + 'build/bin && mkdir -p ' + project_path + 'build/isoc && mkdir -p ' + project_path + 'build/lib && mkdir -p ' + project_path + 'build/redsea')
+    if res:
+        raise ValueError("build-all: step 'refresh-build-path' failed, error code " + str(res))
+
+def build_image():
+    print("build-all: build-image")
+    build_specific_options = '-Wl,--section-start=.text=0x1004000 -Wl,--section-start=.plt=0x1002020 -no-pie'
+    res = os.system('cd ' + project_path + '&& cd src/image && gcc -o ../../build/bin/image ' + build_specific_options + ' -O0 -mno-mmx -mno-red-zone image.c')
+    if res:
+        raise ValueError("build-all: step 'build-image' failed, error code " + str(res))
+
+def build_libtemple():
+    print("build-all: build-libtemple")
+    res = os.system('cd ' + project_path + 'src/libtemple && g++ -c -o ../../build/libtemple.o libtemple.cpp && gcc -shared -o ../../build/lib/libtemple.so ../../build/libtemple.o && rm ' + project_path + 'build/libtemple.o')
+    if res:
+        raise ValueError("build-all: step 'build-libtemple' failed, error code " + str(res))
+
+def transpile_net_to_sepples():
+    print("build-all: transpile-net-to-sepples")
+    res = os.system('cd ' + project_path + 'src/net && ' + jakt_compiler_path + ' -S -R ' + jakt_runtime_path + ' -B ' + project_path + 'build/net -O net.jakt')
+    if res:
+        raise ValueError("build-all: step 'transpile-net-to-sepples' failed, error code " + str(res))
+
+def build_net():
+    print("build-all: build-net")
+    build_specific_options = '-Wno-invalid-offsetof -Wl,--section-start=.text=0x1404000 -Wl,--section-start=.plt=0x1402020 -no-pie'
+    res = os.system('cd ' + project_path + 'build/net && clang++-20 ' + build_specific_options + ' -O3 -I ' + jakt_runtime_path + ' -I ' + project_path + '/src/libtemple -fcolor-diagnostics -std=c++20 -fno-exceptions -Wno-user-defined-literals -Wno-deprecated-declarations -Wno-parentheses-equality -Wno-unqualified-std-cast-call -Wno-unknown-warning-option -Wno-int-to-pointer-cast -mno-red-zone -o ../bin/net *.cpp ../lib/libtemple.so ' + jakt_lib_path + 'libjakt_runtime_x86_64-unknown-linux-unknown.a ' + jakt_lib_path + 'libjakt_main_x86_64-unknown-linux-unknown.a && cd .. && rm -rf net')
+    if res:
+        raise ValueError("build-all: step 'build-net' failed, error code " + str(res))
+
+def address_string_for_symbol(file, symbol):
+    p = subprocess.Popen('readelf -s --wide "' + file + '" | grep \'' + symbol + '$\' | awk \'{sub("000000000", "0x", $2); print $2}\'', shell=True, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
+    return str(p.communicate()[0][:-1].decode(encoding='utf-8'))
+
+def image_hc_fixup(macro, symbol, image_bin_path, image_hc_path):
+    os.system('echo -e "#define ' + macro + ' ' + address_string_for_symbol(image_bin_path, symbol) + '\n" | cat - ' + image_hc_path + ' | sponge ' + image_hc_path)
+    return
+
+def generate_iso_c_file():
+    print("build-all: generate-iso-c-file")
+    step_error_message = "build-all: step 'generate-iso-c-file' failed, error code "
+
+    try:
+        os.remove(isoc_file)
+    except:
+        pass
+    res = os.system('isoc-mount --rw ' + isoc_file + ' ' + redsea_path)
+    if res:
+        raise ValueError(step_error_message + str(res))
+    time.sleep(0.25)
+
+    copy_files_cmd_line = 'rsync -av --inplace --progress ' + project_path + ' ' + redsea_path
+    copy_files_cmd_line += ' --exclude .clang-format'
+    copy_files_cmd_line += ' --exclude .git'
+    copy_files_cmd_line += ' --exclude .gitignore'
+    copy_files_cmd_line += ' --exclude .vscode'
+    copy_files_cmd_line += ' --exclude build/isoc'
+    copy_files_cmd_line += ' --exclude build/lib'
+    copy_files_cmd_line += ' --exclude build/redsea'
+    copy_files_cmd_line += ' --exclude scripts'
+    copy_files_cmd_line += ' --exclude src'
+    res = os.system(copy_files_cmd_line)
+    if res:
+        raise ValueError(step_error_message + str(res))
+
+    if 'custom_files_path' in build_options:
+        copy_custom_files_cmd_line = 'rsync -av --inplace --progress ' + build_options['custom_files_path'] + ' ' + redsea_path
+        res = os.system(copy_custom_files_cmd_line)
+        if res:
+            raise ValueError(step_error_message + str(res))
+
+    # Fixup addresses for Image.HC
+    image_bin_path = redsea_path + '/build/bin/image'
+    image_hc_path = redsea_path + '/System/Utilities/Image.HC'
+
+    image_hc_fixup('IMAGE_LOAD_GIF_FROM_MEMORY', 'image_load_gif_from_memory', image_bin_path, image_hc_path)
+    image_hc_fixup('STBI_WRITE_PNG_TO_MEM', 'stbi_write_png_to_mem', image_bin_path, image_hc_path)
+    image_hc_fixup('STBI_LOAD_FROM_MEMORY', 'stbi_load_from_memory', image_bin_path, image_hc_path)
+    image_hc_fixup('STBI_INFO_FROM_MEMORY', 'stbi_info_from_memory', image_bin_path, image_hc_path)
+    image_hc_fixup('STBI_FAILURE_REASON', 'stbi_failure_reason', image_bin_path, image_hc_path)
+    image_hc_fixup('RENDER_4BIT_FLOYDSTEIN', 'render_4bit_floydstein', image_bin_path, image_hc_path)
+
+    time.sleep(0.25)
+
+    res = os.system('sync && fusermount -u ' + redsea_path)
+    if res:
+        raise ValueError(step_error_message + str(res))
+    time.sleep(0.25)
+
+def generate_slipstream_iso_file():
+    print("build-all: generate-slipstream-iso-file")
+    res = os.system('templeos-slipstream ' + templeos_iso_file + ' ' + isoc_file + ' ' + qemu_slipstream_iso_file)
+    if res:
+        raise ValueError("build-all: step 'generate-slipstream-iso-file' failed, error code " + str(res))
+
+def run():
+    print("build-all: run")
+    res = os.system(qemu_run_cmd)
+    if res:
+        raise ValueError("build-all: step 'run' failed, error code " + str(res))
+
+def build_all():
+    if not build_options_bool('skip_clang_format'):
+        clang_format_src_files()
+    #if not build_options_bool('skip_rebuild'):
+    if 2==3:
+        refresh_build_path()
+        build_image()
+        build_libtemple()
+        transpile_net_to_sepples()
+        build_net()
+    generate_iso_c_file()
+    generate_slipstream_iso_file()
+    run()
+
+build_all()
diff --git a/src/image/image.c b/src/image/image.c
new file mode 100644
index 0000000..42f031c
--- /dev/null
+++ b/src/image/image.c
@@ -0,0 +1,209 @@
+#define STBI_WRITE_NO_STDIO
+#define STB_IMAGE_WRITE_STATIC
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+#define STB_IMAGE_IMPLEMENTATION
+#define STBI_NO_LINEAR
+#define STBI_NO_STDIO
+#define STBI_NO_SIMD
+#define STBI_NO_HDR
+
+#include "stb_image.h"
+#include "stb_image_write.h"
+
+int main() { return 0; }
+
+STBIDEF stbi_uc* image_load_gif_from_memory(stbi_uc const* buffer, int len,
+    int** delays, int* x, int* y,
+    int* z)
+{
+    int comp;
+    return stbi_load_gif_from_memory(buffer, len, delays, x, y, z, &comp, 4);
+}
+
+/* dither.c: MIT License
+
+Copyright (c) 2016 jonmortiboy
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+*/
+
+typedef struct RGB {
+    int r;
+    int g;
+    int b;
+} RGB;
+
+int imgw, imgh;
+
+// Define the 4bit colour palette
+int numCols = 16;
+RGB cols4bit[] = {
+    { 0, 0, 0 }, { 0, 0, 170 }, { 0, 170, 0 }, { 0, 170, 170 },
+    { 170, 0, 0 }, { 170, 0, 170 }, { 170, 85, 0 }, { 170, 170, 170 },
+    { 85, 85, 85 }, { 85, 85, 255 }, { 85, 255, 85 }, { 85, 255, 255 },
+    { 255, 85, 85 }, { 255, 85, 255 }, { 255, 255, 85 }, { 255, 255, 255 }
+};
+RGB* cols = cols4bit;
+
+RGB getRGB(uint32_t* pixels, int x, int y);
+void setRGB(uint32_t* pixels, int x, int y, RGB rgb);
+RGB difRGB(RGB from, RGB to);
+RGB addRGB(RGB a, RGB b);
+RGB divRGB(RGB rgb, double d);
+RGB mulRGB(RGB rgb, double d);
+RGB nearestRGB(RGB rgb, RGB* rgbs, int numRGBs);
+double distRGB(RGB from, RGB to);
+
+void render_4bit_floydstein(uint32_t* pixels, int width, int height);
+
+RGB getRGB(uint32_t* pixels, int x, int y)
+{
+    RGB rgb;
+    rgb.r = 0;
+    rgb.g = 0;
+    rgb.b = 0;
+
+    if (x < 0 || x >= imgw || y < 0 || y >= imgh)
+        return rgb;
+
+    rgb.r = (pixels[y * imgw + x] & 0xff);
+    rgb.g = (pixels[y * imgw + x] & 0xff00) >> 8;
+    rgb.b = (pixels[y * imgw + x] & 0xff0000) >> 16;
+
+    return rgb;
+}
+
+void setRGB(uint32_t* pixels, int x, int y, RGB rgb)
+{
+    if (x < 0 || x >= imgw || y < 0 || y >= imgh)
+        return;
+
+    uint32_t alpha = pixels[y * imgw + x] & 0xff000000;
+    pixels[y * imgw + x] = alpha + (rgb.r) + (rgb.g << 8) + (rgb.b << 16);
+}
+
+RGB difRGB(RGB from, RGB to)
+{
+    RGB dif;
+    dif.r = to.r - from.r;
+    dif.g = to.g - from.g;
+    dif.b = to.b - from.b;
+
+    return dif;
+}
+
+RGB addRGB(RGB a, RGB b)
+{
+    RGB sum;
+    sum.r = a.r + b.r;
+    sum.g = a.g + b.g;
+    sum.b = a.b + b.b;
+
+    if (sum.r > 255)
+        sum.r = 255;
+    if (sum.r < 0)
+        sum.r = 0;
+    if (sum.g > 255)
+        sum.g = 255;
+    if (sum.g < 0)
+        sum.g = 0;
+    if (sum.b > 255)
+        sum.b = 255;
+    if (sum.b < 0)
+        sum.b = 0;
+
+    return sum;
+}
+
+RGB divRGB(RGB rgb, double d)
+{
+    RGB div;
+    div.r = (int)((double)rgb.r / d);
+    div.g = (int)((double)rgb.g / d);
+    div.b = (int)((double)rgb.b / d);
+
+    return div;
+}
+
+RGB mulRGB(RGB rgb, double d)
+{
+    RGB mul;
+    mul.r = (int)((double)rgb.r * d);
+    mul.g = (int)((double)rgb.g * d);
+    mul.b = (int)((double)rgb.b * d);
+
+    return mul;
+}
+
+double distRGB(RGB from, RGB to)
+{
+    RGB dif = difRGB(from, to);
+    double dist = dif.r * dif.r + dif.g * dif.g + dif.b * dif.b;
+
+    return dist;
+}
+
+RGB nearestRGB(RGB rgb, RGB rgbs[], int numRGBs)
+{
+    double dist = -1, tempDist;
+    RGB nearest;
+
+    int i;
+    for (i = 0; i < numRGBs; i++) {
+        tempDist = distRGB(rgb, rgbs[i]);
+
+        if (tempDist < dist || dist < 0) {
+            dist = tempDist;
+            nearest = rgbs[i];
+        }
+    }
+
+    return nearest;
+}
+
+void render_4bit_floydstein(uint32_t* pixels, int width, int height)
+{
+
+    int i, x, y;
+    imgw = width;
+    imgh = height;
+    RGB rgb, nearest, rgberror;
+    for (i = 0; i < imgw * imgh; i++) {
+        rgb = getRGB(pixels, i % imgw, i / imgw);
+        nearest = nearestRGB(rgb, cols, numCols);
+
+        rgberror = difRGB(nearest, rgb);
+        rgberror = divRGB(rgberror, 16);
+
+        x = i % imgw;
+        y = i / imgw;
+
+        setRGB(pixels, x + 1, y,
+            addRGB(getRGB(pixels, x + 1, y), mulRGB(rgberror, 7)));
+        setRGB(pixels, x - 1, y + 1,
+            addRGB(getRGB(pixels, x - 1, y + 1), mulRGB(rgberror, 3)));
+        setRGB(pixels, x, y + 1,
+            addRGB(getRGB(pixels, x, y + 1), mulRGB(rgberror, 5)));
+        setRGB(pixels, x + 1, y + 1,
+            addRGB(getRGB(pixels, x + 1, y + 1), rgberror));
+
+        setRGB(pixels, i % imgw, i / imgw, nearest);
+    }
+}
\ No newline at end of file
diff --git a/src/image/stb_image.h b/src/image/stb_image.h
new file mode 100644
index 0000000..d152aa9
--- /dev/null
+++ b/src/image/stb_image.h
@@ -0,0 +1,8634 @@
+/* stb_image - v2.30 - public domain image loader - http://nothings.org/stb
+                                  no warranty implied; use at your own risk
+
+   Do this:
+      #define STB_IMAGE_IMPLEMENTATION
+   before you include this file in *one* C or C++ file to create the implementation.
+
+   // i.e. it should look like this:
+   #include ...
+   #include ...
+   #include ...
+   #define STB_IMAGE_IMPLEMENTATION
+   #include "stb_image.h"
+
+   You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+   And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+   QUICK NOTES:
+      Primarily of interest to game developers and other people who can
+          avoid problematic images and only need the trivial interface
+
+      JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+      PNG 1/2/4/8/16-bit-per-channel
+
+      TGA (not sure what subset, if a subset)
+      BMP non-1bpp, non-RLE
+      PSD (composited view only, no extra channels, 8/16 bit-per-channel)
+
+      GIF (*comp always reports as 4-channel)
+      HDR (radiance rgbE format)
+      PIC (Softimage PIC)
+      PNM (PPM and PGM binary only)
+
+      Animated GIF still needs a proper API, but here's one way to do it:
+          http://gist.github.com/urraka/685d9a6340b26b830d49
+
+      - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+      - decode from arbitrary I/O callbacks
+      - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+   Full documentation under "DOCUMENTATION" below.
+
+
+LICENSE
+
+  See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+      2.30  (2024-05-31) avoid erroneous gcc warning
+      2.29  (2023-05-xx) optimizations
+      2.28  (2023-01-29) many error fixes, security errors, just tons of stuff
+      2.27  (2021-07-11) document stbi_info better, 16-bit PNM support, bug fixes
+      2.26  (2020-07-13) many minor fixes
+      2.25  (2020-02-02) fix warnings
+      2.24  (2020-02-02) fix warnings; thread-local failure_reason and flip_vertically
+      2.23  (2019-08-11) fix clang static analysis warning
+      2.22  (2019-03-04) gif fixes, fix warnings
+      2.21  (2019-02-25) fix typo in comment
+      2.20  (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs
+      2.19  (2018-02-11) fix warning
+      2.18  (2018-01-30) fix warnings
+      2.17  (2018-01-29) bugfix, 1-bit BMP, 16-bitness query, fix warnings
+      2.16  (2017-07-23) all functions have 16-bit variants; optimizations; bugfixes
+      2.15  (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+      2.14  (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+      2.13  (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
+      2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+      2.11  (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
+                         RGB-format JPEG; remove white matting in PSD;
+                         allocate large structures on the stack;
+                         correct channel count for PNG & BMP
+      2.10  (2016-01-22) avoid warning introduced in 2.09
+      2.09  (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
+
+   See end of file for full revision history.
+
+
+ ============================    Contributors    =========================
+
+ Image formats                          Extensions, features
+    Sean Barrett (jpeg, png, bmp)          Jetro Lauha (stbi_info)
+    Nicolas Schulz (hdr, psd)              Martin "SpartanJ" Golini (stbi_info)
+    Jonathan Dummer (tga)                  James "moose2000" Brown (iPhone PNG)
+    Jean-Marc Lienher (gif)                Ben "Disch" Wenger (io callbacks)
+    Tom Seddon (pic)                       Omar Cornut (1/2/4-bit PNG)
+    Thatcher Ulrich (psd)                  Nicolas Guillemot (vertical flip)
+    Ken Miller (pgm, ppm)                  Richard Mitton (16-bit PSD)
+    github:urraka (animated gif)           Junggon Kim (PNM comments)
+    Christopher Forseth (animated gif)     Daniel Gibson (16-bit TGA)
+                                           socks-the-fox (16-bit PNG)
+                                           Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes                  Mikhail Morozov (1-bit BMP)
+    Fabian "ryg" Giesen                    Anael Seghezzi (is-16-bit query)
+    Arseny Kapoulkine                      Simon Breuss (16-bit PNM)
+    John-Mark Allen
+    Carmelo J Fdez-Aguera
+
+ Bug & warning fixes
+    Marc LeBlanc            David Woo          Guillaume George     Martins Mozeiko
+    Christpher Lloyd        Jerry Jansson      Joseph Thomson       Blazej Dariusz Roszkowski
+    Phil Jordan                                Dave Moore           Roy Eltham
+    Hayaki Saito            Nathan Reed        Won Chun
+    Luke Graham             Johan Duparc       Nick Verigakis       the Horde3D community
+    Thomas Ruf              Ronny Chevalier                         github:rlyeh
+    Janez Zemva             John Bartholomew   Michal Cichon        github:romigrou
+    Jonathan Blow           Ken Hamada         Tero Hanninen        github:svdijk
+    Eugene Golushkov        Laurent Gomila     Cort Stratton        github:snagar
+    Aruelien Pocheville     Sergio Gonzalez    Thibault Reuille     github:Zelex
+    Cass Everitt            Ryamond Barbiero                        github:grim210
+    Paul Du Bois            Engin Manap        Aldo Culquicondor    github:sammyhw
+    Philipp Wiesemann       Dale Weiler        Oriol Ferrer Mesia   github:phprus
+    Josh Tobin              Neil Bickford      Matthew Gregan       github:poppolopoppo
+    Julian Raschke          Gregory Mullen     Christian Floisand   github:darealshinji
+    Baldur Karlsson         Kevin Schmidt      JR Smith             github:Michaelangel007
+                            Brad Weinberger    Matvey Cherevko      github:mosra
+    Luca Sas                Alexander Veselov  Zack Middleton       [reserved]
+    Ryan C. Gordon          [reserved]                              [reserved]
+                     DO NOT ADD YOUR NAME HERE
+
+                     Jacko Dirks
+
+  To add your name to the credits, pick a random blank space in the middle and fill it.
+  80% of merge conflicts on stb PRs are due to people adding their name at the end
+  of the credits.
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#    define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+//    - no 12-bit-per-channel JPEG
+//    - no JPEGs with arithmetic coding
+//    - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+//    int x,y,n;
+//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+//    // ... process data if not NULL ...
+//    // ... x = width, y = height, n = # 8-bit components per pixel ...
+//    // ... replace '0' with '1'..'4' to force that many components per pixel
+//    // ... but 'n' will always be the number that it would have been if you said 0
+//    stbi_image_free(data);
+//
+// Standard parameters:
+//    int *x                 -- outputs image width in pixels
+//    int *y                 -- outputs image height in pixels
+//    int *channels_in_file  -- outputs # of image components in image file
+//    int desired_channels   -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data, or NULL on an allocation failure or if the image is
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'desired_channels' if desired_channels is non-zero, or
+// *channels_in_file otherwise. If desired_channels is non-zero,
+// *channels_in_file has the number of components that _would_ have been
+// output otherwise. E.g. if you set desired_channels to 4, you will always
+// get RGBA output, but you can check *channels_in_file to see if it's trivially
+// opaque because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+//     N=#comp     components
+//       1           grey
+//       2           grey, alpha
+//       3           red, green, blue
+//       4           red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *channels_in_file will be unchanged. The function
+// stbi_failure_reason() can be queried for an extremely brief, end-user
+// unfriendly explanation of why the load failed. Define STBI_NO_FAILURE_STRINGS
+// to avoid compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// To query the width, height and component count of an image without having to
+// decode the full file, you can use the stbi_info family of functions:
+//
+//   int x,y,n,ok;
+//   ok = stbi_info(filename, &x, &y, &n);
+//   // returns ok=1 and sets x, y, n if image is a supported format,
+//   // 0 otherwise.
+//
+// Note that stb_image pervasively uses ints in its public API for sizes,
+// including sizes of memory buffers. This is now part of the API and thus
+// hard to change without causing breakage. As a result, the various image
+// loaders all have certain limits on image size; these differ somewhat
+// by format but generally boil down to either just under 2GB or just under
+// 1GB. When the decoded image would be larger than this, stb_image decoding
+// will fail.
+//
+// Additionally, stb_image will reject image files that have any of their
+// dimensions set to a larger value than the configurable STBI_MAX_DIMENSIONS,
+// which defaults to 2**24 = 16777216 pixels. Due to the above memory limit,
+// the only way to have an image with such dimensions load correctly
+// is for it to have a rather extreme aspect ratio. Either way, the
+// assumption here is that such larger images are likely to be malformed
+// or malicious. If you do need to load an image with individual dimensions
+// larger than that, and it still fits in the overall size limit, you can
+// #define STBI_MAX_DIMENSIONS on your own to be something larger.
+//
+// ===========================================================================
+//
+// UNICODE:
+//
+//   If compiling for Windows and you wish to use Unicode filenames, compile
+//   with
+//       #define STBI_WINDOWS_UTF8
+//   and pass utf8-encoded filenames. Call stbi_convert_wchar_to_utf8 to convert
+//   Windows wchar_t filenames to utf8.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+//    1. easy to use
+//    2. easy to maintain
+//    3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy-to-use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// provide more explicit reasons why performance can't be emphasized.
+//
+//    - Portable ("ease of use")
+//    - Small source code footprint ("easy to maintain")
+//    - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support   (disable by defining STBI_NO_HDR)
+//
+// stb_image supports loading HDR images in general, and currently the Radiance
+// .HDR file format specifically. You can still load any file through the existing
+// interface; if you attempt to load an HDR file, it will be automatically remapped
+// to LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+//     stbi_hdr_to_ldr_gamma(2.2f);
+//     stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+//     stbi_ldr_to_hdr_scale(1.0f);
+//     stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+//     stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// We optionally support converting iPhone-formatted PNGs (which store
+// premultiplied BGRA) back to RGB, even though they're internally encoded
+// differently. To enable this conversion, call
+// stbi_convert_iphone_png_to_rgb(1).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+//  - You can suppress implementation of any of the decoders to reduce
+//    your code footprint by #defining one or more of the following
+//    symbols before creating the implementation.
+//
+//        STBI_NO_JPEG
+//        STBI_NO_PNG
+//        STBI_NO_BMP
+//        STBI_NO_PSD
+//        STBI_NO_TGA
+//        STBI_NO_GIF
+//        STBI_NO_HDR
+//        STBI_NO_PIC
+//        STBI_NO_PNM   (.ppm and .pgm)
+//
+//  - You can request *only* certain decoders and suppress all other ones
+//    (this will be more forward-compatible, as addition of new decoders
+//    doesn't require you to disable them explicitly):
+//
+//        STBI_ONLY_JPEG
+//        STBI_ONLY_PNG
+//        STBI_ONLY_BMP
+//        STBI_ONLY_PSD
+//        STBI_ONLY_TGA
+//        STBI_ONLY_GIF
+//        STBI_ONLY_HDR
+//        STBI_ONLY_PIC
+//        STBI_ONLY_PNM   (.ppm and .pgm)
+//
+//   - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+//     want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
+//  - If you define STBI_MAX_DIMENSIONS, stb_image will reject images greater
+//    than that size (in either width or height) without further processing.
+//    This is to let programs in the wild set an upper bound to prevent
+//    denial-of-service attacks on untrusted data, as one could generate a
+//    valid image of gigantic dimensions and force stb_image to allocate a
+//    huge block of memory and spend disproportionate time decoding it. By
+//    default this is set to (1 << 24), which is 16777216, but that's still
+//    very big.
+
+#    ifndef STBI_NO_STDIO
+#        include <stdio.h>
+#    endif // STBI_NO_STDIO
+
+#    define STBI_VERSION 1
+
+enum {
+    STBI_default = 0, // only used for desired_channels
+
+    STBI_grey = 1,
+    STBI_grey_alpha = 2,
+    STBI_rgb = 3,
+    STBI_rgb_alpha = 4
+};
+
+#    include <stdlib.h>
+typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
+
+#    ifdef __cplusplus
+extern "C" {
+#    endif
+
+#    ifndef STBIDEF
+#        ifdef STB_IMAGE_STATIC
+#            define STBIDEF static
+#        else
+#            define STBIDEF extern
+#        endif
+#    endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+    int (*read)(void* user, char* data, int size); // fill 'data' with 'size' bytes.  return number of bytes actually read
+    void (*skip)(void* user, int n);               // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+    int (*eof)(void* user);                        // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc* stbi_load_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_uc* stbi_load_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels);
+
+#    ifndef STBI_NO_STDIO
+STBIDEF stbi_uc* stbi_load(char const* filename, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_uc* stbi_load_from_file(FILE* f, int* x, int* y, int* channels_in_file, int desired_channels);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#    endif
+
+#    ifndef STBI_NO_GIF
+STBIDEF stbi_uc* stbi_load_gif_from_memory(stbi_uc const* buffer, int len, int** delays, int* x, int* y, int* z, int* comp, int req_comp);
+#    endif
+
+#    ifdef STBI_WINDOWS_UTF8
+STBIDEF int stbi_convert_wchar_to_utf8(char* buffer, size_t bufferlen, wchar_t const* input);
+#    endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us* stbi_load_16_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_us* stbi_load_16_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels);
+
+#    ifndef STBI_NO_STDIO
+STBIDEF stbi_us* stbi_load_16(char const* filename, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_us* stbi_load_from_file_16(FILE* f, int* x, int* y, int* channels_in_file, int desired_channels);
+#    endif
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
+#    ifndef STBI_NO_LINEAR
+STBIDEF float* stbi_loadf_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF float* stbi_loadf_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels);
+
+#        ifndef STBI_NO_STDIO
+STBIDEF float* stbi_loadf(char const* filename, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF float* stbi_loadf_from_file(FILE* f, int* x, int* y, int* channels_in_file, int desired_channels);
+#        endif
+#    endif
+
+#    ifndef STBI_NO_HDR
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
+STBIDEF void stbi_hdr_to_ldr_scale(float scale);
+#    endif // STBI_NO_HDR
+
+#    ifndef STBI_NO_LINEAR
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
+STBIDEF void stbi_ldr_to_hdr_scale(float scale);
+#    endif // STBI_NO_LINEAR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const* clbk, void* user);
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const* buffer, int len);
+#    ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr(char const* filename);
+STBIDEF int stbi_is_hdr_from_file(FILE* f);
+#    endif // STBI_NO_STDIO
+
+// get a VERY brief reason for failure
+// on most compilers (and ALL modern mainstream compilers) this is threadsafe
+STBIDEF const char* stbi_failure_reason(void);
+
+// free the loaded image -- this is just free()
+STBIDEF void stbi_image_free(void* retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int stbi_info_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp);
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* comp);
+STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const* buffer, int len);
+STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const* clbk, void* user);
+
+#    ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const* filename, int* x, int* y, int* comp);
+STBIDEF int stbi_info_from_file(FILE* f, int* x, int* y, int* comp);
+STBIDEF int stbi_is_16_bit(char const* filename);
+STBIDEF int stbi_is_16_bit_from_file(FILE* f);
+#    endif
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// as above, but only applies to images loaded on the thread that calls the function
+// this function is only available if your compiler supports thread-local variables;
+// calling it will fail to link if your compiler doesn't
+STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply);
+STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert);
+STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char* stbi_zlib_decode_malloc_guesssize(char const* buffer, int len, int initial_size, int* outlen);
+STBIDEF char* stbi_zlib_decode_malloc_guesssize_headerflag(char const* buffer, int len, int initial_size, int* outlen, int parse_header);
+STBIDEF char* stbi_zlib_decode_malloc(char const* buffer, int len, int* outlen);
+STBIDEF int stbi_zlib_decode_buffer(char* obuffer, int olen, char const* ibuffer, int ilen);
+
+STBIDEF char* stbi_zlib_decode_noheader_malloc(char const* buffer, int len, int* outlen);
+STBIDEF int stbi_zlib_decode_noheader_buffer(char* obuffer, int olen, char const* ibuffer, int ilen);
+
+#    ifdef __cplusplus
+}
+#    endif
+
+//
+//
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#    if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP)   \
+        || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
+        || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
+        || defined(STBI_ONLY_ZLIB)
+#        ifndef STBI_ONLY_JPEG
+#            define STBI_NO_JPEG
+#        endif
+#        ifndef STBI_ONLY_PNG
+#            define STBI_NO_PNG
+#        endif
+#        ifndef STBI_ONLY_BMP
+#            define STBI_NO_BMP
+#        endif
+#        ifndef STBI_ONLY_PSD
+#            define STBI_NO_PSD
+#        endif
+#        ifndef STBI_ONLY_TGA
+#            define STBI_NO_TGA
+#        endif
+#        ifndef STBI_ONLY_GIF
+#            define STBI_NO_GIF
+#        endif
+#        ifndef STBI_ONLY_HDR
+#            define STBI_NO_HDR
+#        endif
+#        ifndef STBI_ONLY_PIC
+#            define STBI_NO_PIC
+#        endif
+#        ifndef STBI_ONLY_PNM
+#            define STBI_NO_PNM
+#        endif
+#    endif
+
+#    if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
+#        define STBI_NO_ZLIB
+#    endif
+
+#    include <limits.h>
+#    include <stdarg.h>
+#    include <stddef.h> // ptrdiff_t on osx
+#    include <stdlib.h>
+#    include <string.h>
+
+#    if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
+#        include <math.h> // ldexp, pow
+#    endif
+
+#    ifndef STBI_NO_STDIO
+#        include <stdio.h>
+#    endif
+
+#    ifndef STBI_ASSERT
+#        include <assert.h>
+#        define STBI_ASSERT(x) assert(x)
+#    endif
+
+#    ifdef __cplusplus
+#        define STBI_EXTERN extern "C"
+#    else
+#        define STBI_EXTERN extern
+#    endif
+
+#    ifndef _MSC_VER
+#        ifdef __cplusplus
+#            define stbi_inline inline
+#        else
+#            define stbi_inline
+#        endif
+#    else
+#        define stbi_inline __forceinline
+#    endif
+
+#    ifndef STBI_NO_THREAD_LOCALS
+#        if defined(__cplusplus) && __cplusplus >= 201103L
+#            define STBI_THREAD_LOCAL thread_local
+#        elif defined(__GNUC__) && __GNUC__ < 5
+#            define STBI_THREAD_LOCAL __thread
+#        elif defined(_MSC_VER)
+#            define STBI_THREAD_LOCAL __declspec(thread)
+#        elif defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__)
+#            define STBI_THREAD_LOCAL _Thread_local
+#        endif
+
+#        ifndef STBI_THREAD_LOCAL
+#            if defined(__GNUC__)
+#                define STBI_THREAD_LOCAL __thread
+#            endif
+#        endif
+#    endif
+
+#    if defined(_MSC_VER) || defined(__SYMBIAN32__)
+typedef unsigned short stbi__uint16;
+typedef signed short stbi__int16;
+typedef unsigned int stbi__uint32;
+typedef signed int stbi__int32;
+#    else
+#        include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t stbi__int32;
+#    endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32) == 4 ? 1 : -1];
+
+#    ifdef _MSC_VER
+#        define STBI_NOTUSED(v) (void)(v)
+#    else
+#        define STBI_NOTUSED(v) (void)sizeof(v)
+#    endif
+
+#    ifdef _MSC_VER
+#        define STBI_HAS_LROTL
+#    endif
+
+#    ifdef STBI_HAS_LROTL
+#        define stbi_lrot(x, y) _lrotl(x, y)
+#    else
+#        define stbi_lrot(x, y) (((x) << (y)) | ((x) >> (-(y) & 31)))
+#    endif
+
+#    if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
+// ok
+#    elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
+// ok
+#    else
+#        error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
+#    endif
+
+#    ifndef STBI_MALLOC
+#        define STBI_MALLOC(sz) malloc(sz)
+#        define STBI_REALLOC(p, newsz) realloc(p, newsz)
+#        define STBI_FREE(p) free(p)
+#    endif
+
+#    ifndef STBI_REALLOC_SIZED
+#        define STBI_REALLOC_SIZED(p, oldsz, newsz) STBI_REALLOC(p, newsz)
+#    endif
+
+// x86/x64 detection
+#    if defined(__x86_64__) || defined(_M_X64)
+#        define STBI__X64_TARGET
+#    elif defined(__i386) || defined(_M_IX86)
+#        define STBI__X86_TARGET
+#    endif
+
+#    if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
+// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
+#        define STBI_NO_SIMD
+#    endif
+
+#    if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
+// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
+//
+// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
+// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
+// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
+// simultaneously enabling "-mstackrealign".
+//
+// See https://github.com/nothings/stb/issues/81 for more information.
+//
+// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
+// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
+#        define STBI_NO_SIMD
+#    endif
+
+#    if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
+#        define STBI_SSE2
+#        include <emmintrin.h>
+
+#        ifdef _MSC_VER
+
+#            if _MSC_VER >= 1400    // not VC6
+#                include <intrin.h> // __cpuid
+static int stbi__cpuid3(void)
+{
+    int info[4];
+    __cpuid(info, 1);
+    return info[3];
+}
+#            else
+static int stbi__cpuid3(void)
+{
+    int res;
+    __asm {
+      mov  eax,1
+      cpuid
+      mov  res,edx
+    }
+    return res;
+}
+#            endif
+
+#            define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+
+#            if !defined(STBI_NO_JPEG) && defined(STBI_SSE2)
+static int stbi__sse2_available(void)
+{
+    int info3 = stbi__cpuid3();
+    return ((info3 >> 26) & 1) != 0;
+}
+#            endif
+
+#        else // assume GCC-style if not VC++
+#            define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+#            if !defined(STBI_NO_JPEG) && defined(STBI_SSE2)
+static int stbi__sse2_available(void)
+{
+    // If we're even attempting to compile this on GCC/Clang, that means
+    // -msse2 is on, which means the compiler is allowed to use SSE2
+    // instructions at will, and so are we.
+    return 1;
+}
+#            endif
+
+#        endif
+#    endif
+
+// ARM NEON
+#    if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#        undef STBI_NEON
+#    endif
+
+#    ifdef STBI_NEON
+#        include <arm_neon.h>
+#        ifdef _MSC_VER
+#            define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+#        else
+#            define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#        endif
+#    endif
+
+#    ifndef STBI_SIMD_ALIGN
+#        define STBI_SIMD_ALIGN(type, name) type name
+#    endif
+
+#    ifndef STBI_MAX_DIMENSIONS
+#        define STBI_MAX_DIMENSIONS (1 << 24)
+#    endif
+
+///////////////////////////////////////////////
+//
+//  stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+    stbi__uint32 img_x, img_y;
+    int img_n, img_out_n;
+
+    stbi_io_callbacks io;
+    void* io_user_data;
+
+    int read_from_callbacks;
+    int buflen;
+    stbi_uc buffer_start[128];
+    int callback_already_read;
+
+    stbi_uc *img_buffer, *img_buffer_end;
+    stbi_uc *img_buffer_original, *img_buffer_original_end;
+} stbi__context;
+
+static void stbi__refill_buffer(stbi__context* s);
+
+// initialize a memory-decode context
+static void stbi__start_mem(stbi__context* s, stbi_uc const* buffer, int len)
+{
+    s->io.read = NULL;
+    s->read_from_callbacks = 0;
+    s->callback_already_read = 0;
+    s->img_buffer = s->img_buffer_original = (stbi_uc*)buffer;
+    s->img_buffer_end = s->img_buffer_original_end = (stbi_uc*)buffer + len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context* s, stbi_io_callbacks* c, void* user)
+{
+    s->io = *c;
+    s->io_user_data = user;
+    s->buflen = sizeof(s->buffer_start);
+    s->read_from_callbacks = 1;
+    s->callback_already_read = 0;
+    s->img_buffer = s->img_buffer_original = s->buffer_start;
+    stbi__refill_buffer(s);
+    s->img_buffer_original_end = s->img_buffer_end;
+}
+
+#    ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void* user, char* data, int size)
+{
+    return (int)fread(data, 1, size, (FILE*)user);
+}
+
+static void stbi__stdio_skip(void* user, int n)
+{
+    int ch;
+    fseek((FILE*)user, n, SEEK_CUR);
+    ch = fgetc((FILE*)user); /* have to read a byte to reset feof()'s flag */
+    if (ch != EOF) {
+        ungetc(ch, (FILE*)user); /* push byte back onto stream if valid. */
+    }
+}
+
+static int stbi__stdio_eof(void* user)
+{
+    return feof((FILE*)user) || ferror((FILE*)user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks = {
+    stbi__stdio_read,
+    stbi__stdio_skip,
+    stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context* s, FILE* f)
+{
+    stbi__start_callbacks(s, &stbi__stdio_callbacks, (void*)f);
+}
+
+// static void stop_file(stbi__context *s) { }
+
+#    endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context* s)
+{
+    // conceptually rewind SHOULD rewind to the beginning of the stream,
+    // but we just rewind to the beginning of the initial buffer, because
+    // we only use it after doing 'test', which only ever looks at at most 92 bytes
+    s->img_buffer = s->img_buffer_original;
+    s->img_buffer_end = s->img_buffer_original_end;
+}
+
+enum {
+    STBI_ORDER_RGB,
+    STBI_ORDER_BGR
+};
+
+typedef struct
+{
+    int bits_per_channel;
+    int num_channels;
+    int channel_order;
+} stbi__result_info;
+
+#    ifndef STBI_NO_JPEG
+static int stbi__jpeg_test(stbi__context* s);
+static void* stbi__jpeg_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__jpeg_info(stbi__context* s, int* x, int* y, int* comp);
+#    endif
+
+#    ifndef STBI_NO_PNG
+static int stbi__png_test(stbi__context* s);
+static void* stbi__png_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__png_info(stbi__context* s, int* x, int* y, int* comp);
+static int stbi__png_is16(stbi__context* s);
+#    endif
+
+#    ifndef STBI_NO_BMP
+static int stbi__bmp_test(stbi__context* s);
+static void* stbi__bmp_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__bmp_info(stbi__context* s, int* x, int* y, int* comp);
+#    endif
+
+#    ifndef STBI_NO_TGA
+static int stbi__tga_test(stbi__context* s);
+static void* stbi__tga_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__tga_info(stbi__context* s, int* x, int* y, int* comp);
+#    endif
+
+#    ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context* s);
+static void* stbi__psd_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri, int bpc);
+static int stbi__psd_info(stbi__context* s, int* x, int* y, int* comp);
+static int stbi__psd_is16(stbi__context* s);
+#    endif
+
+#    ifndef STBI_NO_HDR
+static int stbi__hdr_test(stbi__context* s);
+static float* stbi__hdr_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__hdr_info(stbi__context* s, int* x, int* y, int* comp);
+#    endif
+
+#    ifndef STBI_NO_PIC
+static int stbi__pic_test(stbi__context* s);
+static void* stbi__pic_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__pic_info(stbi__context* s, int* x, int* y, int* comp);
+#    endif
+
+#    ifndef STBI_NO_GIF
+static int stbi__gif_test(stbi__context* s);
+static void* stbi__gif_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static void* stbi__load_gif_main(stbi__context* s, int** delays, int* x, int* y, int* z, int* comp, int req_comp);
+static int stbi__gif_info(stbi__context* s, int* x, int* y, int* comp);
+#    endif
+
+#    ifndef STBI_NO_PNM
+static int stbi__pnm_test(stbi__context* s);
+static void* stbi__pnm_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__pnm_info(stbi__context* s, int* x, int* y, int* comp);
+static int stbi__pnm_is16(stbi__context* s);
+#    endif
+
+static
+#    ifdef STBI_THREAD_LOCAL
+    STBI_THREAD_LOCAL
+#    endif
+    char const* stbi__g_failure_reason;
+
+STBIDEF const char* stbi_failure_reason(void)
+{
+    return stbi__g_failure_reason;
+}
+
+#    ifndef STBI_NO_FAILURE_STRINGS
+static int stbi__err(char const* str)
+{
+    stbi__g_failure_reason = str;
+    return 0;
+}
+#    endif
+
+static void* stbi__malloc(size_t size)
+{
+    return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+    if (b < 0)
+        return 0;
+    // now 0 <= b <= INT_MAX, hence also
+    // 0 <= INT_MAX - b <= INTMAX.
+    // And "a + b <= INT_MAX" (which might overflow) is the
+    // same as a <= INT_MAX - b (no overflow)
+    return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+    if (a < 0 || b < 0)
+        return 0;
+    if (b == 0)
+        return 1; // mul-by-0 is always safe
+    // portable way to check for no overflows in a*b
+    return a <= INT_MAX / b;
+}
+
+#    if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR)
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+    return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a * b, add);
+}
+#    endif
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+    return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a * b, c) && stbi__addsizes_valid(a * b * c, add);
+}
+
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+#    if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM)
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
+{
+    return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a * b, c) && stbi__mul2sizes_valid(a * b * c, d) && stbi__addsizes_valid(a * b * c * d, add);
+}
+#    endif
+
+#    if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR)
+// mallocs with size overflow checking
+static void* stbi__malloc_mad2(int a, int b, int add)
+{
+    if (!stbi__mad2sizes_valid(a, b, add))
+        return NULL;
+    return stbi__malloc(a * b + add);
+}
+#    endif
+
+static void* stbi__malloc_mad3(int a, int b, int c, int add)
+{
+    if (!stbi__mad3sizes_valid(a, b, c, add))
+        return NULL;
+    return stbi__malloc(a * b * c + add);
+}
+
+#    if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM)
+static void* stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+    if (!stbi__mad4sizes_valid(a, b, c, d, add))
+        return NULL;
+    return stbi__malloc(a * b * c * d + add);
+}
+#    endif
+
+// returns 1 if the sum of two signed ints is valid (between -2^31 and 2^31-1 inclusive), 0 on overflow.
+static int stbi__addints_valid(int a, int b)
+{
+    if ((a >= 0) != (b >= 0))
+        return 1; // a and b have different signs, so no overflow
+    if (a < 0 && b < 0)
+        return a >= INT_MIN - b; // same as a + b >= INT_MIN; INT_MIN - b cannot overflow since b < 0.
+    return a <= INT_MAX - b;
+}
+
+// returns 1 if the product of two ints fits in a signed short, 0 on overflow.
+static int stbi__mul2shorts_valid(int a, int b)
+{
+    if (b == 0 || b == -1)
+        return 1; // multiplication by 0 is always 0; check for -1 so SHRT_MIN/b doesn't overflow
+    if ((a >= 0) == (b >= 0))
+        return a <= SHRT_MAX / b; // product is positive, so similar to mul2sizes_valid
+    if (b < 0)
+        return a <= SHRT_MIN / b; // same as a * b >= SHRT_MIN
+    return a >= SHRT_MIN / b;
+}
+
+// stbi__err - error
+// stbi__errpf - error returning pointer to float
+// stbi__errpuc - error returning pointer to unsigned char
+
+#    ifdef STBI_NO_FAILURE_STRINGS
+#        define stbi__err(x, y) 0
+#    elif defined(STBI_FAILURE_USERMSG)
+#        define stbi__err(x, y) stbi__err(y)
+#    else
+#        define stbi__err(x, y) stbi__err(x)
+#    endif
+
+#    define stbi__errpf(x, y) ((float*)(size_t)(stbi__err(x, y) ? NULL : NULL))
+#    define stbi__errpuc(x, y) ((unsigned char*)(size_t)(stbi__err(x, y) ? NULL : NULL))
+
+STBIDEF void stbi_image_free(void* retval_from_stbi_load)
+{
+    STBI_FREE(retval_from_stbi_load);
+}
+
+#    ifndef STBI_NO_LINEAR
+static float* stbi__ldr_to_hdr(stbi_uc* data, int x, int y, int comp);
+#    endif
+
+#    ifndef STBI_NO_HDR
+static stbi_uc* stbi__hdr_to_ldr(float* data, int x, int y, int comp);
+#    endif
+
+static int stbi__vertically_flip_on_load_global = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+    stbi__vertically_flip_on_load_global = flag_true_if_should_flip;
+}
+
+#    ifndef STBI_THREAD_LOCAL
+#        define stbi__vertically_flip_on_load stbi__vertically_flip_on_load_global
+#    else
+static STBI_THREAD_LOCAL int stbi__vertically_flip_on_load_local, stbi__vertically_flip_on_load_set;
+
+STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip)
+{
+    stbi__vertically_flip_on_load_local = flag_true_if_should_flip;
+    stbi__vertically_flip_on_load_set = 1;
+}
+
+#        define stbi__vertically_flip_on_load (stbi__vertically_flip_on_load_set \
+                ? stbi__vertically_flip_on_load_local                            \
+                : stbi__vertically_flip_on_load_global)
+#    endif // STBI_THREAD_LOCAL
+
+static void* stbi__load_main(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri, int bpc)
+{
+    memset(ri, 0, sizeof(*ri));         // make sure it's initialized if we add new fields
+    ri->bits_per_channel = 8;           // default is 8 so most paths don't have to be changed
+    ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+    ri->num_channels = 0;
+
+// test the formats with a very explicit header first (at least a FOURCC
+// or distinctive magic number first)
+#    ifndef STBI_NO_PNG
+    if (stbi__png_test(s))
+        return stbi__png_load(s, x, y, comp, req_comp, ri);
+#    endif
+#    ifndef STBI_NO_BMP
+    if (stbi__bmp_test(s))
+        return stbi__bmp_load(s, x, y, comp, req_comp, ri);
+#    endif
+#    ifndef STBI_NO_GIF
+    if (stbi__gif_test(s))
+        return stbi__gif_load(s, x, y, comp, req_comp, ri);
+#    endif
+#    ifndef STBI_NO_PSD
+    if (stbi__psd_test(s))
+        return stbi__psd_load(s, x, y, comp, req_comp, ri, bpc);
+#    else
+    STBI_NOTUSED(bpc);
+#    endif
+#    ifndef STBI_NO_PIC
+    if (stbi__pic_test(s))
+        return stbi__pic_load(s, x, y, comp, req_comp, ri);
+#    endif
+
+// then the formats that can end up attempting to load with just 1 or 2
+// bytes matching expectations; these are prone to false positives, so
+// try them later
+#    ifndef STBI_NO_JPEG
+    if (stbi__jpeg_test(s))
+        return stbi__jpeg_load(s, x, y, comp, req_comp, ri);
+#    endif
+#    ifndef STBI_NO_PNM
+    if (stbi__pnm_test(s))
+        return stbi__pnm_load(s, x, y, comp, req_comp, ri);
+#    endif
+
+#    ifndef STBI_NO_HDR
+    if (stbi__hdr_test(s)) {
+        float* hdr = stbi__hdr_load(s, x, y, comp, req_comp, ri);
+        return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+    }
+#    endif
+
+#    ifndef STBI_NO_TGA
+    // test tga last because it's a crappy test!
+    if (stbi__tga_test(s))
+        return stbi__tga_load(s, x, y, comp, req_comp, ri);
+#    endif
+
+    return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static stbi_uc* stbi__convert_16_to_8(stbi__uint16* orig, int w, int h, int channels)
+{
+    int i;
+    int img_len = w * h * channels;
+    stbi_uc* reduced;
+
+    reduced = (stbi_uc*)stbi__malloc(img_len);
+    if (reduced == NULL)
+        return stbi__errpuc("outofmem", "Out of memory");
+
+    for (i = 0; i < img_len; ++i)
+        reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+    STBI_FREE(orig);
+    return reduced;
+}
+
+static stbi__uint16* stbi__convert_8_to_16(stbi_uc* orig, int w, int h, int channels)
+{
+    int i;
+    int img_len = w * h * channels;
+    stbi__uint16* enlarged;
+
+    enlarged = (stbi__uint16*)stbi__malloc(img_len * 2);
+    if (enlarged == NULL)
+        return (stbi__uint16*)stbi__errpuc("outofmem", "Out of memory");
+
+    for (i = 0; i < img_len; ++i)
+        enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+    STBI_FREE(orig);
+    return enlarged;
+}
+
+static void stbi__vertical_flip(void* image, int w, int h, int bytes_per_pixel)
+{
+    return;
+    int row;
+    size_t bytes_per_row = (size_t)w * bytes_per_pixel;
+    stbi_uc temp[2048];
+    stbi_uc* bytes = (stbi_uc*)image;
+
+    for (row = 0; row < (h >> 1); row++) {
+        stbi_uc* row0 = bytes + row * bytes_per_row;
+        stbi_uc* row1 = bytes + (h - row - 1) * bytes_per_row;
+        // swap row0 with row1
+        size_t bytes_left = bytes_per_row;
+        while (bytes_left) {
+            size_t bytes_copy = (bytes_left < sizeof(temp)) ? bytes_left : sizeof(temp);
+            memcpy(temp, row0, bytes_copy);
+            memcpy(row0, row1, bytes_copy);
+            memcpy(row1, temp, bytes_copy);
+            row0 += bytes_copy;
+            row1 += bytes_copy;
+            bytes_left -= bytes_copy;
+        }
+    }
+}
+
+#    ifndef STBI_NO_GIF
+static void stbi__vertical_flip_slices(void* image, int w, int h, int z, int bytes_per_pixel)
+{
+    return;
+    int slice;
+    int slice_size = w * h * bytes_per_pixel;
+
+    stbi_uc* bytes = (stbi_uc*)image;
+    for (slice = 0; slice < z; ++slice) {
+        stbi__vertical_flip(bytes, w, h, bytes_per_pixel);
+        bytes += slice_size;
+    }
+}
+#    endif
+
+static unsigned char* stbi__load_and_postprocess_8bit(stbi__context* s, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__result_info ri;
+    void* result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+    if (result == NULL)
+        return NULL;
+
+    // it is the responsibility of the loaders to make sure we get either 8 or 16 bit.
+    STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16);
+
+    if (ri.bits_per_channel != 8) {
+        result = stbi__convert_16_to_8((stbi__uint16*)result, *x, *y, req_comp == 0 ? *comp : req_comp);
+        ri.bits_per_channel = 8;
+    }
+
+    // @TODO: move stbi__convert_format to here
+
+    if (stbi__vertically_flip_on_load) {
+        int channels = req_comp ? req_comp : *comp;
+        stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi_uc));
+    }
+
+    return (unsigned char*)result;
+}
+
+static stbi__uint16* stbi__load_and_postprocess_16bit(stbi__context* s, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__result_info ri;
+    void* result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+    if (result == NULL)
+        return NULL;
+
+    // it is the responsibility of the loaders to make sure we get either 8 or 16 bit.
+    STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16);
+
+    if (ri.bits_per_channel != 16) {
+        result = stbi__convert_8_to_16((stbi_uc*)result, *x, *y, req_comp == 0 ? *comp : req_comp);
+        ri.bits_per_channel = 16;
+    }
+
+    // @TODO: move stbi__convert_format16 to here
+    // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+    if (stbi__vertically_flip_on_load) {
+        int channels = req_comp ? req_comp : *comp;
+        stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi__uint16));
+    }
+
+    return (stbi__uint16*)result;
+}
+
+#    if !defined(STBI_NO_HDR) && !defined(STBI_NO_LINEAR)
+static void stbi__float_postprocess(float* result, int* x, int* y, int* comp, int req_comp)
+{
+    if (stbi__vertically_flip_on_load && result != NULL) {
+        int channels = req_comp ? req_comp : *comp;
+        stbi__vertical_flip(result, *x, *y, channels * sizeof(float));
+    }
+}
+#    endif
+
+#    ifndef STBI_NO_STDIO
+
+#        if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+STBI_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, char const* str, int cbmb, wchar_t* widestr, int cchwide);
+STBI_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, wchar_t const* widestr, int cchwide, char* str, int cbmb, char const* defchar, int* used_default);
+#        endif
+
+#        if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+STBIDEF int stbi_convert_wchar_to_utf8(char* buffer, size_t bufferlen, wchar_t const* input)
+{
+    return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int)bufferlen, NULL, NULL);
+}
+#        endif
+
+static FILE* stbi__fopen(char const* filename, char const* mode)
+{
+    FILE* f;
+#        if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+    wchar_t wMode[64];
+    wchar_t wFilename[1024];
+    if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename) / sizeof(*wFilename)))
+        return 0;
+
+    if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode) / sizeof(*wMode)))
+        return 0;
+
+#            if defined(_MSC_VER) && _MSC_VER >= 1400
+    if (0 != _wfopen_s(&f, wFilename, wMode))
+        f = 0;
+#            else
+    f = _wfopen(wFilename, wMode);
+#            endif
+
+#        elif defined(_MSC_VER) && _MSC_VER >= 1400
+    if (0 != fopen_s(&f, filename, mode))
+        f = 0;
+#        else
+    f = fopen(filename, mode);
+#        endif
+    return f;
+}
+
+STBIDEF stbi_uc* stbi_load(char const* filename, int* x, int* y, int* comp, int req_comp)
+{
+    FILE* f = stbi__fopen(filename, "rb");
+    unsigned char* result;
+    if (!f)
+        return stbi__errpuc("can't fopen", "Unable to open file");
+    result = stbi_load_from_file(f, x, y, comp, req_comp);
+    fclose(f);
+    return result;
+}
+
+STBIDEF stbi_uc* stbi_load_from_file(FILE* f, int* x, int* y, int* comp, int req_comp)
+{
+    unsigned char* result;
+    stbi__context s;
+    stbi__start_file(&s, f);
+    result = stbi__load_and_postprocess_8bit(&s, x, y, comp, req_comp);
+    if (result) {
+        // need to 'unget' all the characters in the IO buffer
+        fseek(f, -(int)(s.img_buffer_end - s.img_buffer), SEEK_CUR);
+    }
+    return result;
+}
+
+STBIDEF stbi__uint16* stbi_load_from_file_16(FILE* f, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__uint16* result;
+    stbi__context s;
+    stbi__start_file(&s, f);
+    result = stbi__load_and_postprocess_16bit(&s, x, y, comp, req_comp);
+    if (result) {
+        // need to 'unget' all the characters in the IO buffer
+        fseek(f, -(int)(s.img_buffer_end - s.img_buffer), SEEK_CUR);
+    }
+    return result;
+}
+
+STBIDEF stbi_us* stbi_load_16(char const* filename, int* x, int* y, int* comp, int req_comp)
+{
+    FILE* f = stbi__fopen(filename, "rb");
+    stbi__uint16* result;
+    if (!f)
+        return (stbi_us*)stbi__errpuc("can't fopen", "Unable to open file");
+    result = stbi_load_from_file_16(f, x, y, comp, req_comp);
+    fclose(f);
+    return result;
+}
+
+#    endif //! STBI_NO_STDIO
+
+STBIDEF stbi_us* stbi_load_16_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels)
+{
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+    return stbi__load_and_postprocess_16bit(&s, x, y, channels_in_file, desired_channels);
+}
+
+STBIDEF stbi_us* stbi_load_16_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels)
+{
+    stbi__context s;
+    stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+    return stbi__load_and_postprocess_16bit(&s, x, y, channels_in_file, desired_channels);
+}
+
+STBIDEF stbi_uc* stbi_load_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+    return stbi__load_and_postprocess_8bit(&s, x, y, comp, req_comp);
+}
+
+STBIDEF stbi_uc* stbi_load_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__context s;
+    stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+    return stbi__load_and_postprocess_8bit(&s, x, y, comp, req_comp);
+}
+
+#    ifndef STBI_NO_GIF
+STBIDEF stbi_uc* stbi_load_gif_from_memory(stbi_uc const* buffer, int len, int** delays, int* x, int* y, int* z, int* comp, int req_comp)
+{
+    unsigned char* result;
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+
+    result = (unsigned char*)stbi__load_gif_main(&s, delays, x, y, z, comp, req_comp);
+    if (stbi__vertically_flip_on_load) {
+        stbi__vertical_flip_slices(result, *x, *y, *z, *comp);
+    }
+
+    return result;
+}
+#    endif
+
+#    ifndef STBI_NO_LINEAR
+static float* stbi__loadf_main(stbi__context* s, int* x, int* y, int* comp, int req_comp)
+{
+    unsigned char* data;
+#        ifndef STBI_NO_HDR
+    if (stbi__hdr_test(s)) {
+        stbi__result_info ri;
+        float* hdr_data = stbi__hdr_load(s, x, y, comp, req_comp, &ri);
+        if (hdr_data)
+            stbi__float_postprocess(hdr_data, x, y, comp, req_comp);
+        return hdr_data;
+    }
+#        endif
+    data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
+    if (data)
+        return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
+    return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
+}
+
+STBIDEF float* stbi_loadf_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+    return stbi__loadf_main(&s, x, y, comp, req_comp);
+}
+
+STBIDEF float* stbi_loadf_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__context s;
+    stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+    return stbi__loadf_main(&s, x, y, comp, req_comp);
+}
+
+#        ifndef STBI_NO_STDIO
+STBIDEF float* stbi_loadf(char const* filename, int* x, int* y, int* comp, int req_comp)
+{
+    float* result;
+    FILE* f = stbi__fopen(filename, "rb");
+    if (!f)
+        return stbi__errpf("can't fopen", "Unable to open file");
+    result = stbi_loadf_from_file(f, x, y, comp, req_comp);
+    fclose(f);
+    return result;
+}
+
+STBIDEF float* stbi_loadf_from_file(FILE* f, int* x, int* y, int* comp, int req_comp)
+{
+    stbi__context s;
+    stbi__start_file(&s, f);
+    return stbi__loadf_main(&s, x, y, comp, req_comp);
+}
+#        endif // !STBI_NO_STDIO
+
+#    endif // !STBI_NO_LINEAR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
+// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
+// reports false!
+
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const* buffer, int len)
+{
+#    ifndef STBI_NO_HDR
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+    return stbi__hdr_test(&s);
+#    else
+    STBI_NOTUSED(buffer);
+    STBI_NOTUSED(len);
+    return 0;
+#    endif
+}
+
+#    ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr(char const* filename)
+{
+    FILE* f = stbi__fopen(filename, "rb");
+    int result = 0;
+    if (f) {
+        result = stbi_is_hdr_from_file(f);
+        fclose(f);
+    }
+    return result;
+}
+
+STBIDEF int stbi_is_hdr_from_file(FILE* f)
+{
+#        ifndef STBI_NO_HDR
+    long pos = ftell(f);
+    int res;
+    stbi__context s;
+    stbi__start_file(&s, f);
+    res = stbi__hdr_test(&s);
+    fseek(f, pos, SEEK_SET);
+    return res;
+#        else
+    STBI_NOTUSED(f);
+    return 0;
+#        endif
+}
+#    endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const* clbk, void* user)
+{
+#    ifndef STBI_NO_HDR
+    stbi__context s;
+    stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+    return stbi__hdr_test(&s);
+#    else
+    STBI_NOTUSED(clbk);
+    STBI_NOTUSED(user);
+    return 0;
+#    endif
+}
+
+#    ifndef STBI_NO_LINEAR
+static float stbi__l2h_gamma = 2.2f, stbi__l2h_scale = 1.0f;
+
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#    endif
+
+static float stbi__h2l_gamma_i = 1.0f / 2.2f, stbi__h2l_scale_i = 1.0f;
+
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1 / gamma; }
+STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1 / scale; }
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum {
+    STBI__SCAN_load = 0,
+    STBI__SCAN_type,
+    STBI__SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context* s)
+{
+    int n = (s->io.read)(s->io_user_data, (char*)s->buffer_start, s->buflen);
+    s->callback_already_read += (int)(s->img_buffer - s->img_buffer_original);
+    if (n == 0) {
+        // at end of file, treat same as if from memory, but need to handle case
+        // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
+        s->read_from_callbacks = 0;
+        s->img_buffer = s->buffer_start;
+        s->img_buffer_end = s->buffer_start + 1;
+        *s->img_buffer = 0;
+    } else {
+        s->img_buffer = s->buffer_start;
+        s->img_buffer_end = s->buffer_start + n;
+    }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context* s)
+{
+    if (s->img_buffer < s->img_buffer_end)
+        return *s->img_buffer++;
+    if (s->read_from_callbacks) {
+        stbi__refill_buffer(s);
+        return *s->img_buffer++;
+    }
+    return 0;
+}
+
+#    if defined(STBI_NO_JPEG) && defined(STBI_NO_HDR) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#    else
+stbi_inline static int stbi__at_eof(stbi__context* s)
+{
+    if (s->io.read) {
+        if (!(s->io.eof)(s->io_user_data))
+            return 0;
+        // if feof() is true, check if buffer = end
+        // special case: we've only got the special 0 character at the end
+        if (s->read_from_callbacks == 0)
+            return 1;
+    }
+
+    return s->img_buffer >= s->img_buffer_end;
+}
+#    endif
+
+#    if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC)
+// nothing
+#    else
+static void stbi__skip(stbi__context* s, int n)
+{
+    if (n == 0)
+        return; // already there!
+    if (n < 0) {
+        s->img_buffer = s->img_buffer_end;
+        return;
+    }
+    if (s->io.read) {
+        int blen = (int)(s->img_buffer_end - s->img_buffer);
+        if (blen < n) {
+            s->img_buffer = s->img_buffer_end;
+            (s->io.skip)(s->io_user_data, n - blen);
+            return;
+        }
+    }
+    s->img_buffer += n;
+}
+#    endif
+
+#    if defined(STBI_NO_PNG) && defined(STBI_NO_TGA) && defined(STBI_NO_HDR) && defined(STBI_NO_PNM)
+// nothing
+#    else
+static int stbi__getn(stbi__context* s, stbi_uc* buffer, int n)
+{
+    if (s->io.read) {
+        int blen = (int)(s->img_buffer_end - s->img_buffer);
+        if (blen < n) {
+            int res, count;
+
+            memcpy(buffer, s->img_buffer, blen);
+
+            count = (s->io.read)(s->io_user_data, (char*)buffer + blen, n - blen);
+            res = (count == (n - blen));
+            s->img_buffer = s->img_buffer_end;
+            return res;
+        }
+    }
+
+    if (s->img_buffer + n <= s->img_buffer_end) {
+        memcpy(buffer, s->img_buffer, n);
+        s->img_buffer += n;
+        return 1;
+    } else
+        return 0;
+}
+#    endif
+
+#    if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC)
+// nothing
+#    else
+static int stbi__get16be(stbi__context* s)
+{
+    int z = stbi__get8(s);
+    return (z << 8) + stbi__get8(s);
+}
+#    endif
+
+#    if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC)
+// nothing
+#    else
+static stbi__uint32 stbi__get32be(stbi__context* s)
+{
+    stbi__uint32 z = stbi__get16be(s);
+    return (z << 16) + stbi__get16be(s);
+}
+#    endif
+
+#    if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
+// nothing
+#    else
+static int stbi__get16le(stbi__context* s)
+{
+    int z = stbi__get8(s);
+    return z + (stbi__get8(s) << 8);
+}
+#    endif
+
+#    ifndef STBI_NO_BMP
+static stbi__uint32 stbi__get32le(stbi__context* s)
+{
+    stbi__uint32 z = stbi__get16le(s);
+    z += (stbi__uint32)stbi__get16le(s) << 16;
+    return z;
+}
+#    endif
+
+#    define STBI__BYTECAST(x) ((stbi_uc)((x) & 255)) // truncate int to byte without warnings
+
+#    if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#    else
+//////////////////////////////////////////////////////////////////////////////
+//
+//  generic converter from built-in img_n to req_comp
+//    individual types do this automatically as much as possible (e.g. jpeg
+//    does all cases internally since it needs to colorspace convert anyway,
+//    and it never has alpha, so very few cases ). png can automatically
+//    interleave an alpha=255 channel, but falls back to this for other cases
+//
+//  assume data buffer is malloced, so malloc a new one and free that one
+//  only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+    return (stbi_uc)(((r * 77) + (g * 150) + (29 * b)) >> 8);
+}
+#    endif
+
+#    if defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#    else
+static unsigned char* stbi__convert_format(unsigned char* data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+    int i, j;
+    unsigned char* good;
+
+    if (req_comp == img_n)
+        return data;
+    STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+    good = (unsigned char*)stbi__malloc_mad3(req_comp, x, y, 0);
+    if (good == NULL) {
+        STBI_FREE(data);
+        return stbi__errpuc("outofmem", "Out of memory");
+    }
+
+    for (j = 0; j < (int)y; ++j) {
+        unsigned char* src = data + j * x * img_n;
+        unsigned char* dest = good + j * x * req_comp;
+
+#        define STBI__COMBO(a, b) ((a) * 8 + (b))
+#        define STBI__CASE(a, b) \
+        case STBI__COMBO(a, b):  \
+            for (i = x - 1; i >= 0; --i, src += a, dest += b)
+        // convert source image with img_n components to one with req_comp components;
+        // avoid switch per pixel, so use switch per scanline and massive macros
+        switch (STBI__COMBO(img_n, req_comp)) {
+            STBI__CASE(1, 2)
+            {
+                dest[0] = src[0];
+                dest[1] = 255;
+            }
+            break;
+            STBI__CASE(1, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+            break;
+            STBI__CASE(1, 4)
+            {
+                dest[0] = dest[1] = dest[2] = src[0];
+                dest[3] = 255;
+            }
+            break;
+            STBI__CASE(2, 1) { dest[0] = src[0]; }
+            break;
+            STBI__CASE(2, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+            break;
+            STBI__CASE(2, 4)
+            {
+                dest[0] = dest[1] = dest[2] = src[0];
+                dest[3] = src[1];
+            }
+            break;
+            STBI__CASE(3, 4)
+            {
+                dest[0] = src[0];
+                dest[1] = src[1];
+                dest[2] = src[2];
+                dest[3] = 255;
+            }
+            break;
+            STBI__CASE(3, 1) { dest[0] = stbi__compute_y(src[0], src[1], src[2]); }
+            break;
+            STBI__CASE(3, 2)
+            {
+                dest[0] = stbi__compute_y(src[0], src[1], src[2]);
+                dest[1] = 255;
+            }
+            break;
+            STBI__CASE(4, 1) { dest[0] = stbi__compute_y(src[0], src[1], src[2]); }
+            break;
+            STBI__CASE(4, 2)
+            {
+                dest[0] = stbi__compute_y(src[0], src[1], src[2]);
+                dest[1] = src[3];
+            }
+            break;
+            STBI__CASE(4, 3)
+            {
+                dest[0] = src[0];
+                dest[1] = src[1];
+                dest[2] = src[2];
+            }
+            break;
+        default:
+            STBI_ASSERT(0);
+            STBI_FREE(data);
+            STBI_FREE(good);
+            return stbi__errpuc("unsupported", "Unsupported format conversion");
+        }
+#        undef STBI__CASE
+    }
+
+    STBI_FREE(data);
+    return good;
+}
+#    endif
+
+#    if defined(STBI_NO_PNG) && defined(STBI_NO_PSD)
+// nothing
+#    else
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+    return (stbi__uint16)(((r * 77) + (g * 150) + (29 * b)) >> 8);
+}
+#    endif
+
+#    if defined(STBI_NO_PNG) && defined(STBI_NO_PSD)
+// nothing
+#    else
+static stbi__uint16* stbi__convert_format16(stbi__uint16* data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+    int i, j;
+    stbi__uint16* good;
+
+    if (req_comp == img_n)
+        return data;
+    STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+    good = (stbi__uint16*)stbi__malloc(req_comp * x * y * 2);
+    if (good == NULL) {
+        STBI_FREE(data);
+        return (stbi__uint16*)stbi__errpuc("outofmem", "Out of memory");
+    }
+
+    for (j = 0; j < (int)y; ++j) {
+        stbi__uint16* src = data + j * x * img_n;
+        stbi__uint16* dest = good + j * x * req_comp;
+
+#        define STBI__COMBO(a, b) ((a) * 8 + (b))
+#        define STBI__CASE(a, b) \
+        case STBI__COMBO(a, b):  \
+            for (i = x - 1; i >= 0; --i, src += a, dest += b)
+        // convert source image with img_n components to one with req_comp components;
+        // avoid switch per pixel, so use switch per scanline and massive macros
+        switch (STBI__COMBO(img_n, req_comp)) {
+            STBI__CASE(1, 2)
+            {
+                dest[0] = src[0];
+                dest[1] = 0xffff;
+            }
+            break;
+            STBI__CASE(1, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+            break;
+            STBI__CASE(1, 4)
+            {
+                dest[0] = dest[1] = dest[2] = src[0];
+                dest[3] = 0xffff;
+            }
+            break;
+            STBI__CASE(2, 1) { dest[0] = src[0]; }
+            break;
+            STBI__CASE(2, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+            break;
+            STBI__CASE(2, 4)
+            {
+                dest[0] = dest[1] = dest[2] = src[0];
+                dest[3] = src[1];
+            }
+            break;
+            STBI__CASE(3, 4)
+            {
+                dest[0] = src[0];
+                dest[1] = src[1];
+                dest[2] = src[2];
+                dest[3] = 0xffff;
+            }
+            break;
+            STBI__CASE(3, 1) { dest[0] = stbi__compute_y_16(src[0], src[1], src[2]); }
+            break;
+            STBI__CASE(3, 2)
+            {
+                dest[0] = stbi__compute_y_16(src[0], src[1], src[2]);
+                dest[1] = 0xffff;
+            }
+            break;
+            STBI__CASE(4, 1) { dest[0] = stbi__compute_y_16(src[0], src[1], src[2]); }
+            break;
+            STBI__CASE(4, 2)
+            {
+                dest[0] = stbi__compute_y_16(src[0], src[1], src[2]);
+                dest[1] = src[3];
+            }
+            break;
+            STBI__CASE(4, 3)
+            {
+                dest[0] = src[0];
+                dest[1] = src[1];
+                dest[2] = src[2];
+            }
+            break;
+        default:
+            STBI_ASSERT(0);
+            STBI_FREE(data);
+            STBI_FREE(good);
+            return (stbi__uint16*)stbi__errpuc("unsupported", "Unsupported format conversion");
+        }
+#        undef STBI__CASE
+    }
+
+    STBI_FREE(data);
+    return good;
+}
+#    endif
+
+#    ifndef STBI_NO_LINEAR
+static float* stbi__ldr_to_hdr(stbi_uc* data, int x, int y, int comp)
+{
+    int i, k, n;
+    float* output;
+    if (!data)
+        return NULL;
+    output = (float*)stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+    if (output == NULL) {
+        STBI_FREE(data);
+        return stbi__errpf("outofmem", "Out of memory");
+    }
+    // compute number of non-alpha components
+    if (comp & 1)
+        n = comp;
+    else
+        n = comp - 1;
+    for (i = 0; i < x * y; ++i) {
+        for (k = 0; k < n; ++k) {
+            output[i * comp + k] = (float)(pow(data[i * comp + k] / 255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+        }
+    }
+    if (n < comp) {
+        for (i = 0; i < x * y; ++i) {
+            output[i * comp + n] = data[i * comp + n] / 255.0f;
+        }
+    }
+    STBI_FREE(data);
+    return output;
+}
+#    endif
+
+#    ifndef STBI_NO_HDR
+#        define stbi__float2int(x) ((int)(x))
+static stbi_uc* stbi__hdr_to_ldr(float* data, int x, int y, int comp)
+{
+    int i, k, n;
+    stbi_uc* output;
+    if (!data)
+        return NULL;
+    output = (stbi_uc*)stbi__malloc_mad3(x, y, comp, 0);
+    if (output == NULL) {
+        STBI_FREE(data);
+        return stbi__errpuc("outofmem", "Out of memory");
+    }
+    // compute number of non-alpha components
+    if (comp & 1)
+        n = comp;
+    else
+        n = comp - 1;
+    for (i = 0; i < x * y; ++i) {
+        for (k = 0; k < n; ++k) {
+            float z = (float)pow(data[i * comp + k] * stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
+            if (z < 0)
+                z = 0;
+            if (z > 255)
+                z = 255;
+            output[i * comp + k] = (stbi_uc)stbi__float2int(z);
+        }
+        if (k < comp) {
+            float z = data[i * comp + k] * 255 + 0.5f;
+            if (z < 0)
+                z = 0;
+            if (z > 255)
+                z = 255;
+            output[i * comp + k] = (stbi_uc)stbi__float2int(z);
+        }
+    }
+    STBI_FREE(data);
+    return output;
+}
+#    endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  "baseline" JPEG/JFIF decoder
+//
+//    simple implementation
+//      - doesn't support delayed output of y-dimension
+//      - simple interface (only one output format: 8-bit interleaved RGB)
+//      - doesn't try to recover corrupt jpegs
+//      - doesn't allow partial loading, loading multiple at once
+//      - still fast on x86 (copying globals into locals doesn't help x86)
+//      - allocates lots of intermediate memory (full size of all components)
+//        - non-interleaved case requires this anyway
+//        - allows good upsampling (see next)
+//    high-quality
+//      - upsampled channels are bilinearly interpolated, even across blocks
+//      - quality integer IDCT derived from IJG's 'slow'
+//    performance
+//      - fast huffman; reasonable integer IDCT
+//      - some SIMD kernels for common paths on targets with SSE2/NEON
+//      - uses a lot of intermediate memory, could cache poorly
+
+#    ifndef STBI_NO_JPEG
+
+// huffman decoding acceleration
+#        define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
+
+typedef struct
+{
+    stbi_uc fast[1 << FAST_BITS];
+    // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
+    stbi__uint16 code[256];
+    stbi_uc values[256];
+    stbi_uc size[257];
+    unsigned int maxcode[18];
+    int delta[17]; // old 'firstsymbol' - old 'firstcode'
+} stbi__huffman;
+
+typedef struct
+{
+    stbi__context* s;
+    stbi__huffman huff_dc[4];
+    stbi__huffman huff_ac[4];
+    stbi__uint16 dequant[4][64];
+    stbi__int16 fast_ac[4][1 << FAST_BITS];
+
+    // sizes for components, interleaved MCUs
+    int img_h_max, img_v_max;
+    int img_mcu_x, img_mcu_y;
+    int img_mcu_w, img_mcu_h;
+
+    // definition of jpeg image component
+    struct
+    {
+        int id;
+        int h, v;
+        int tq;
+        int hd, ha;
+        int dc_pred;
+
+        int x, y, w2, h2;
+        stbi_uc* data;
+        void *raw_data, *raw_coeff;
+        stbi_uc* linebuf;
+        short* coeff;         // progressive only
+        int coeff_w, coeff_h; // number of 8x8 coefficient blocks
+    } img_comp[4];
+
+    stbi__uint32 code_buffer; // jpeg entropy-coded buffer
+    int code_bits;            // number of valid bits
+    unsigned char marker;     // marker seen while filling entropy buffer
+    int nomore;               // flag if we saw a marker so must stop
+
+    int progressive;
+    int spec_start;
+    int spec_end;
+    int succ_high;
+    int succ_low;
+    int eob_run;
+    int jfif;
+    int app14_color_transform; // Adobe APP14 tag
+    int rgb;
+
+    int scan_n, order[4];
+    int restart_interval, todo;
+
+    // kernels
+    void (*idct_block_kernel)(stbi_uc* out, int out_stride, short data[64]);
+    void (*YCbCr_to_RGB_kernel)(stbi_uc* out, stbi_uc const* y, stbi_uc const* pcb, stbi_uc const* pcr, int count, int step);
+    stbi_uc* (*resample_row_hv_2_kernel)(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs);
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman* h, int* count)
+{
+    int i, j, k = 0;
+    unsigned int code;
+    // build size list for each symbol (from JPEG spec)
+    for (i = 0; i < 16; ++i) {
+        for (j = 0; j < count[i]; ++j) {
+            h->size[k++] = (stbi_uc)(i + 1);
+            if (k >= 257)
+                return stbi__err("bad size list", "Corrupt JPEG");
+        }
+    }
+    h->size[k] = 0;
+
+    // compute actual symbols (from jpeg spec)
+    code = 0;
+    k = 0;
+    for (j = 1; j <= 16; ++j) {
+        // compute delta to add to code to compute symbol id
+        h->delta[j] = k - code;
+        if (h->size[k] == j) {
+            while (h->size[k] == j)
+                h->code[k++] = (stbi__uint16)(code++);
+            if (code - 1 >= (1u << j))
+                return stbi__err("bad code lengths", "Corrupt JPEG");
+        }
+        // compute largest code + 1 for this size, preshifted as needed later
+        h->maxcode[j] = code << (16 - j);
+        code <<= 1;
+    }
+    h->maxcode[j] = 0xffffffff;
+
+    // build non-spec acceleration table; 255 is flag for not-accelerated
+    memset(h->fast, 255, 1 << FAST_BITS);
+    for (i = 0; i < k; ++i) {
+        int s = h->size[i];
+        if (s <= FAST_BITS) {
+            int c = h->code[i] << (FAST_BITS - s);
+            int m = 1 << (FAST_BITS - s);
+            for (j = 0; j < m; ++j) {
+                h->fast[c + j] = (stbi_uc)i;
+            }
+        }
+    }
+    return 1;
+}
+
+// build a table that decodes both magnitude and value of small ACs in
+// one go.
+static void stbi__build_fast_ac(stbi__int16* fast_ac, stbi__huffman* h)
+{
+    int i;
+    for (i = 0; i < (1 << FAST_BITS); ++i) {
+        stbi_uc fast = h->fast[i];
+        fast_ac[i] = 0;
+        if (fast < 255) {
+            int rs = h->values[fast];
+            int run = (rs >> 4) & 15;
+            int magbits = rs & 15;
+            int len = h->size[fast];
+
+            if (magbits && len + magbits <= FAST_BITS) {
+                // magnitude code followed by receive_extend code
+                int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
+                int m = 1 << (magbits - 1);
+                if (k < m)
+                    k += (~0U << magbits) + 1;
+                // if the result is small enough, we can fit it in fast_ac table
+                if (k >= -128 && k <= 127)
+                    fast_ac[i] = (stbi__int16)((k * 256) + (run * 16) + (len + magbits));
+            }
+        }
+    }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg* j)
+{
+    do {
+        unsigned int b = j->nomore ? 0 : stbi__get8(j->s);
+        if (b == 0xff) {
+            int c = stbi__get8(j->s);
+            while (c == 0xff)
+                c = stbi__get8(j->s); // consume fill bytes
+            if (c != 0) {
+                j->marker = (unsigned char)c;
+                j->nomore = 1;
+                return;
+            }
+        }
+        j->code_buffer |= b << (24 - j->code_bits);
+        j->code_bits += 8;
+    } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 const stbi__bmask[17] = { 0, 1, 3, 7, 15, 31, 63, 127, 255, 511, 1023, 2047, 4095, 8191, 16383, 32767, 65535 };
+
+// decode a jpeg huffman value from the bitstream
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg* j, stbi__huffman* h)
+{
+    unsigned int temp;
+    int c, k;
+
+    if (j->code_bits < 16)
+        stbi__grow_buffer_unsafe(j);
+
+    // look at the top FAST_BITS and determine what symbol ID it is,
+    // if the code is <= FAST_BITS
+    c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS) - 1);
+    k = h->fast[c];
+    if (k < 255) {
+        int s = h->size[k];
+        if (s > j->code_bits)
+            return -1;
+        j->code_buffer <<= s;
+        j->code_bits -= s;
+        return h->values[k];
+    }
+
+    // naive test is to shift the code_buffer down so k bits are
+    // valid, then test against maxcode. To speed this up, we've
+    // preshifted maxcode left so that it has (16-k) 0s at the
+    // end; in other words, regardless of the number of bits, it
+    // wants to be compared against something shifted to have 16;
+    // that way we don't need to shift inside the loop.
+    temp = j->code_buffer >> 16;
+    for (k = FAST_BITS + 1;; ++k)
+        if (temp < h->maxcode[k])
+            break;
+    if (k == 17) {
+        // error! code not found
+        j->code_bits -= 16;
+        return -1;
+    }
+
+    if (k > j->code_bits)
+        return -1;
+
+    // convert the huffman code to the symbol id
+    c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+    if (c < 0 || c >= 256) // symbol id out of bounds!
+        return -1;
+    STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
+
+    // convert the id to a symbol
+    j->code_bits -= k;
+    j->code_buffer <<= k;
+    return h->values[c];
+}
+
+// bias[n] = (-1<<n) + 1
+static int const stbi__jbias[16] = { 0, -1, -3, -7, -15, -31, -63, -127, -255, -511, -1023, -2047, -4095, -8191, -16383, -32767 };
+
+// combined JPEG 'receive' and JPEG 'extend', since baseline
+// always extends everything it receives.
+stbi_inline static int stbi__extend_receive(stbi__jpeg* j, int n)
+{
+    unsigned int k;
+    int sgn;
+    if (j->code_bits < n)
+        stbi__grow_buffer_unsafe(j);
+    if (j->code_bits < n)
+        return 0; // ran out of bits from stream, return 0s intead of continuing
+
+    sgn = j->code_buffer >> 31; // sign bit always in MSB; 0 if MSB clear (positive), 1 if MSB set (negative)
+    k = stbi_lrot(j->code_buffer, n);
+    j->code_buffer = k & ~stbi__bmask[n];
+    k &= stbi__bmask[n];
+    j->code_bits -= n;
+    return k + (stbi__jbias[n] & (sgn - 1));
+}
+
+// get some unsigned bits
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg* j, int n)
+{
+    unsigned int k;
+    if (j->code_bits < n)
+        stbi__grow_buffer_unsafe(j);
+    if (j->code_bits < n)
+        return 0; // ran out of bits from stream, return 0s intead of continuing
+    k = stbi_lrot(j->code_buffer, n);
+    j->code_buffer = k & ~stbi__bmask[n];
+    k &= stbi__bmask[n];
+    j->code_bits -= n;
+    return k;
+}
+
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg* j)
+{
+    unsigned int k;
+    if (j->code_bits < 1)
+        stbi__grow_buffer_unsafe(j);
+    if (j->code_bits < 1)
+        return 0; // ran out of bits from stream, return 0s intead of continuing
+    k = j->code_buffer;
+    j->code_buffer <<= 1;
+    --j->code_bits;
+    return k & 0x80000000;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc const stbi__jpeg_dezigzag[64 + 15] = {
+    0, 1, 8, 16, 9, 2, 3, 10,
+    17, 24, 32, 25, 18, 11, 4, 5,
+    12, 19, 26, 33, 40, 48, 41, 34,
+    27, 20, 13, 6, 7, 14, 21, 28,
+    35, 42, 49, 56, 57, 50, 43, 36,
+    29, 22, 15, 23, 30, 37, 44, 51,
+    58, 59, 52, 45, 38, 31, 39, 46,
+    53, 60, 61, 54, 47, 55, 62, 63,
+    // let corrupt input sample past end
+    63, 63, 63, 63, 63, 63, 63, 63,
+    63, 63, 63, 63, 63, 63, 63
+};
+
+// decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg* j, short data[64], stbi__huffman* hdc, stbi__huffman* hac, stbi__int16* fac, int b, stbi__uint16* dequant)
+{
+    int diff, dc, k;
+    int t;
+
+    if (j->code_bits < 16)
+        stbi__grow_buffer_unsafe(j);
+    t = stbi__jpeg_huff_decode(j, hdc);
+    if (t < 0 || t > 15)
+        return stbi__err("bad huffman code", "Corrupt JPEG");
+
+    // 0 all the ac values now so we can do it 32-bits at a time
+    memset(data, 0, 64 * sizeof(data[0]));
+
+    diff = t ? stbi__extend_receive(j, t) : 0;
+    if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff))
+        return stbi__err("bad delta", "Corrupt JPEG");
+    dc = j->img_comp[b].dc_pred + diff;
+    j->img_comp[b].dc_pred = dc;
+    if (!stbi__mul2shorts_valid(dc, dequant[0]))
+        return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+    data[0] = (short)(dc * dequant[0]);
+
+    // decode AC components, see JPEG spec
+    k = 1;
+    do {
+        unsigned int zig;
+        int c, r, s;
+        if (j->code_bits < 16)
+            stbi__grow_buffer_unsafe(j);
+        c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS) - 1);
+        r = fac[c];
+        if (r) {                // fast-AC path
+            k += (r >> 4) & 15; // run
+            s = r & 15;         // combined length
+            if (s > j->code_bits)
+                return stbi__err("bad huffman code", "Combined length longer than code bits available");
+            j->code_buffer <<= s;
+            j->code_bits -= s;
+            // decode into unzigzag'd location
+            zig = stbi__jpeg_dezigzag[k++];
+            data[zig] = (short)((r >> 8) * dequant[zig]);
+        } else {
+            int rs = stbi__jpeg_huff_decode(j, hac);
+            if (rs < 0)
+                return stbi__err("bad huffman code", "Corrupt JPEG");
+            s = rs & 15;
+            r = rs >> 4;
+            if (s == 0) {
+                if (rs != 0xf0)
+                    break; // end block
+                k += 16;
+            } else {
+                k += r;
+                // decode into unzigzag'd location
+                zig = stbi__jpeg_dezigzag[k++];
+                data[zig] = (short)(stbi__extend_receive(j, s) * dequant[zig]);
+            }
+        }
+    } while (k < 64);
+    return 1;
+}
+
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg* j, short data[64], stbi__huffman* hdc, int b)
+{
+    int diff, dc;
+    int t;
+    if (j->spec_end != 0)
+        return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+    if (j->code_bits < 16)
+        stbi__grow_buffer_unsafe(j);
+
+    if (j->succ_high == 0) {
+        // first scan for DC coefficient, must be first
+        memset(data, 0, 64 * sizeof(data[0])); // 0 all the ac values now
+        t = stbi__jpeg_huff_decode(j, hdc);
+        if (t < 0 || t > 15)
+            return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+        diff = t ? stbi__extend_receive(j, t) : 0;
+
+        if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff))
+            return stbi__err("bad delta", "Corrupt JPEG");
+        dc = j->img_comp[b].dc_pred + diff;
+        j->img_comp[b].dc_pred = dc;
+        if (!stbi__mul2shorts_valid(dc, 1 << j->succ_low))
+            return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+        data[0] = (short)(dc * (1 << j->succ_low));
+    } else {
+        // refinement scan for DC coefficient
+        if (stbi__jpeg_get_bit(j))
+            data[0] += (short)(1 << j->succ_low);
+    }
+    return 1;
+}
+
+// @OPTIMIZE: store non-zigzagged during the decode passes,
+// and only de-zigzag when dequantizing
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg* j, short data[64], stbi__huffman* hac, stbi__int16* fac)
+{
+    int k;
+    if (j->spec_start == 0)
+        return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+    if (j->succ_high == 0) {
+        int shift = j->succ_low;
+
+        if (j->eob_run) {
+            --j->eob_run;
+            return 1;
+        }
+
+        k = j->spec_start;
+        do {
+            unsigned int zig;
+            int c, r, s;
+            if (j->code_bits < 16)
+                stbi__grow_buffer_unsafe(j);
+            c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS) - 1);
+            r = fac[c];
+            if (r) {                // fast-AC path
+                k += (r >> 4) & 15; // run
+                s = r & 15;         // combined length
+                if (s > j->code_bits)
+                    return stbi__err("bad huffman code", "Combined length longer than code bits available");
+                j->code_buffer <<= s;
+                j->code_bits -= s;
+                zig = stbi__jpeg_dezigzag[k++];
+                data[zig] = (short)((r >> 8) * (1 << shift));
+            } else {
+                int rs = stbi__jpeg_huff_decode(j, hac);
+                if (rs < 0)
+                    return stbi__err("bad huffman code", "Corrupt JPEG");
+                s = rs & 15;
+                r = rs >> 4;
+                if (s == 0) {
+                    if (r < 15) {
+                        j->eob_run = (1 << r);
+                        if (r)
+                            j->eob_run += stbi__jpeg_get_bits(j, r);
+                        --j->eob_run;
+                        break;
+                    }
+                    k += 16;
+                } else {
+                    k += r;
+                    zig = stbi__jpeg_dezigzag[k++];
+                    data[zig] = (short)(stbi__extend_receive(j, s) * (1 << shift));
+                }
+            }
+        } while (k <= j->spec_end);
+    } else {
+        // refinement scan for these AC coefficients
+
+        short bit = (short)(1 << j->succ_low);
+
+        if (j->eob_run) {
+            --j->eob_run;
+            for (k = j->spec_start; k <= j->spec_end; ++k) {
+                short* p = &data[stbi__jpeg_dezigzag[k]];
+                if (*p != 0)
+                    if (stbi__jpeg_get_bit(j))
+                        if ((*p & bit) == 0) {
+                            if (*p > 0)
+                                *p += bit;
+                            else
+                                *p -= bit;
+                        }
+            }
+        } else {
+            k = j->spec_start;
+            do {
+                int r, s;
+                int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
+                if (rs < 0)
+                    return stbi__err("bad huffman code", "Corrupt JPEG");
+                s = rs & 15;
+                r = rs >> 4;
+                if (s == 0) {
+                    if (r < 15) {
+                        j->eob_run = (1 << r) - 1;
+                        if (r)
+                            j->eob_run += stbi__jpeg_get_bits(j, r);
+                        r = 64; // force end of block
+                    } else {
+                        // r=15 s=0 should write 16 0s, so we just do
+                        // a run of 15 0s and then write s (which is 0),
+                        // so we don't have to do anything special here
+                    }
+                } else {
+                    if (s != 1)
+                        return stbi__err("bad huffman code", "Corrupt JPEG");
+                    // sign bit
+                    if (stbi__jpeg_get_bit(j))
+                        s = bit;
+                    else
+                        s = -bit;
+                }
+
+                // advance by r
+                while (k <= j->spec_end) {
+                    short* p = &data[stbi__jpeg_dezigzag[k++]];
+                    if (*p != 0) {
+                        if (stbi__jpeg_get_bit(j))
+                            if ((*p & bit) == 0) {
+                                if (*p > 0)
+                                    *p += bit;
+                                else
+                                    *p -= bit;
+                            }
+                    } else {
+                        if (r == 0) {
+                            *p = (short)s;
+                            break;
+                        }
+                        --r;
+                    }
+                }
+            } while (k <= j->spec_end);
+        }
+    }
+    return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+    // trick to use a single test to catch both cases
+    if ((unsigned int)x > 255) {
+        if (x < 0)
+            return 0;
+        if (x > 255)
+            return 255;
+    }
+    return (stbi_uc)x;
+}
+
+#        define stbi__f2f(x) ((int)(((x) * 4096 + 0.5)))
+#        define stbi__fsh(x) ((x) * 4096)
+
+// derived from jidctint -- DCT_ISLOW
+#        define STBI__IDCT_1D(s0, s1, s2, s3, s4, s5, s6, s7)       \
+            int t0, t1, t2, t3, p1, p2, p3, p4, p5, x0, x1, x2, x3; \
+            p2 = s2;                                                \
+            p3 = s6;                                                \
+            p1 = (p2 + p3) * stbi__f2f(0.5411961f);                 \
+            t2 = p1 + p3 * stbi__f2f(-1.847759065f);                \
+            t3 = p1 + p2 * stbi__f2f(0.765366865f);                 \
+            p2 = s0;                                                \
+            p3 = s4;                                                \
+            t0 = stbi__fsh(p2 + p3);                                \
+            t1 = stbi__fsh(p2 - p3);                                \
+            x0 = t0 + t3;                                           \
+            x3 = t0 - t3;                                           \
+            x1 = t1 + t2;                                           \
+            x2 = t1 - t2;                                           \
+            t0 = s7;                                                \
+            t1 = s5;                                                \
+            t2 = s3;                                                \
+            t3 = s1;                                                \
+            p3 = t0 + t2;                                           \
+            p4 = t1 + t3;                                           \
+            p1 = t0 + t3;                                           \
+            p2 = t1 + t2;                                           \
+            p5 = (p3 + p4) * stbi__f2f(1.175875602f);               \
+            t0 = t0 * stbi__f2f(0.298631336f);                      \
+            t1 = t1 * stbi__f2f(2.053119869f);                      \
+            t2 = t2 * stbi__f2f(3.072711026f);                      \
+            t3 = t3 * stbi__f2f(1.501321110f);                      \
+            p1 = p5 + p1 * stbi__f2f(-0.899976223f);                \
+            p2 = p5 + p2 * stbi__f2f(-2.562915447f);                \
+            p3 = p3 * stbi__f2f(-1.961570560f);                     \
+            p4 = p4 * stbi__f2f(-0.390180644f);                     \
+            t3 += p1 + p4;                                          \
+            t2 += p2 + p3;                                          \
+            t1 += p2 + p4;                                          \
+            t0 += p1 + p3;
+
+static void stbi__idct_block(stbi_uc* out, int out_stride, short data[64])
+{
+    int i, val[64], *v = val;
+    stbi_uc* o;
+    short* d = data;
+
+    // columns
+    for (i = 0; i < 8; ++i, ++d, ++v) {
+        // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
+        if (d[8] == 0 && d[16] == 0 && d[24] == 0 && d[32] == 0
+            && d[40] == 0 && d[48] == 0 && d[56] == 0) {
+            //    no shortcut                 0     seconds
+            //    (1|2|3|4|5|6|7)==0          0     seconds
+            //    all separate               -0.047 seconds
+            //    1 && 2|3 && 4|5 && 6|7:    -0.047 seconds
+            int dcterm = d[0] * 4;
+            v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+        } else {
+            STBI__IDCT_1D(d[0], d[8], d[16], d[24], d[32], d[40], d[48], d[56])
+            // constants scaled things up by 1<<12; let's bring them back
+            // down, but keep 2 extra bits of precision
+            x0 += 512;
+            x1 += 512;
+            x2 += 512;
+            x3 += 512;
+            v[0] = (x0 + t3) >> 10;
+            v[56] = (x0 - t3) >> 10;
+            v[8] = (x1 + t2) >> 10;
+            v[48] = (x1 - t2) >> 10;
+            v[16] = (x2 + t1) >> 10;
+            v[40] = (x2 - t1) >> 10;
+            v[24] = (x3 + t0) >> 10;
+            v[32] = (x3 - t0) >> 10;
+        }
+    }
+
+    for (i = 0, v = val, o = out; i < 8; ++i, v += 8, o += out_stride) {
+        // no fast case since the first 1D IDCT spread components out
+        STBI__IDCT_1D(v[0], v[1], v[2], v[3], v[4], v[5], v[6], v[7])
+        // constants scaled things up by 1<<12, plus we had 1<<2 from first
+        // loop, plus horizontal and vertical each scale by sqrt(8) so together
+        // we've got an extra 1<<3, so 1<<17 total we need to remove.
+        // so we want to round that, which means adding 0.5 * 1<<17,
+        // aka 65536. Also, we'll end up with -128 to 127 that we want
+        // to encode as 0..255 by adding 128, so we'll add that before the shift
+        x0 += 65536 + (128 << 17);
+        x1 += 65536 + (128 << 17);
+        x2 += 65536 + (128 << 17);
+        x3 += 65536 + (128 << 17);
+        // tried computing the shifts into temps, or'ing the temps to see
+        // if any were out of range, but that was slower
+        o[0] = stbi__clamp((x0 + t3) >> 17);
+        o[7] = stbi__clamp((x0 - t3) >> 17);
+        o[1] = stbi__clamp((x1 + t2) >> 17);
+        o[6] = stbi__clamp((x1 - t2) >> 17);
+        o[2] = stbi__clamp((x2 + t1) >> 17);
+        o[5] = stbi__clamp((x2 - t1) >> 17);
+        o[3] = stbi__clamp((x3 + t0) >> 17);
+        o[4] = stbi__clamp((x3 - t0) >> 17);
+    }
+}
+
+#        ifdef STBI_SSE2
+// sse2 integer IDCT. not the fastest possible implementation but it
+// produces bit-identical results to the generic C version so it's
+// fully "transparent".
+static void stbi__idct_simd(stbi_uc* out, int out_stride, short data[64])
+{
+    // This is constructed to match our regular (generic) integer IDCT exactly.
+    __m128i row0, row1, row2, row3, row4, row5, row6, row7;
+    __m128i tmp;
+
+// dot product constant: even elems=x, odd elems=y
+#            define dct_const(x, y) _mm_setr_epi16((x), (y), (x), (y), (x), (y), (x), (y))
+
+// out(0) = c0[even]*x + c0[odd]*y   (c0, x, y 16-bit, out 32-bit)
+// out(1) = c1[even]*x + c1[odd]*y
+#            define dct_rot(out0, out1, x, y, c0, c1)          \
+                __m128i c0##lo = _mm_unpacklo_epi16((x), (y)); \
+                __m128i c0##hi = _mm_unpackhi_epi16((x), (y)); \
+                __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+                __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+                __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+                __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+// out = in << 12  (in 16-bit, out 32-bit)
+#            define dct_widen(out, in)                                                              \
+                __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+                __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+// wide add
+#            define dct_wadd(out, a, b)                        \
+                __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+                __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+// wide sub
+#            define dct_wsub(out, a, b)                        \
+                __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+                __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+// butterfly a/b, add bias, then shift by "s" and pack
+#            define dct_bfly32o(out0, out1, a, b, bias, s)                                      \
+                {                                                                               \
+                    __m128i abiased_l = _mm_add_epi32(a##_l, bias);                             \
+                    __m128i abiased_h = _mm_add_epi32(a##_h, bias);                             \
+                    dct_wadd(sum, abiased, b);                                                  \
+                    dct_wsub(dif, abiased, b);                                                  \
+                    out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+                    out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+                }
+
+// 8-bit interleave step (for transposes)
+#            define dct_interleave8(a, b)    \
+                tmp = a;                     \
+                a = _mm_unpacklo_epi8(a, b); \
+                b = _mm_unpackhi_epi8(tmp, b)
+
+// 16-bit interleave step (for transposes)
+#            define dct_interleave16(a, b)    \
+                tmp = a;                      \
+                a = _mm_unpacklo_epi16(a, b); \
+                b = _mm_unpackhi_epi16(tmp, b)
+
+#            define dct_pass(bias, shift)                            \
+                {                                                    \
+                    /* even part */                                  \
+                    dct_rot(t2e, t3e, row2, row6, rot0_0, rot0_1);   \
+                    __m128i sum04 = _mm_add_epi16(row0, row4);       \
+                    __m128i dif04 = _mm_sub_epi16(row0, row4);       \
+                    dct_widen(t0e, sum04);                           \
+                    dct_widen(t1e, dif04);                           \
+                    dct_wadd(x0, t0e, t3e);                          \
+                    dct_wsub(x3, t0e, t3e);                          \
+                    dct_wadd(x1, t1e, t2e);                          \
+                    dct_wsub(x2, t1e, t2e);                          \
+                    /* odd part */                                   \
+                    dct_rot(y0o, y2o, row7, row3, rot2_0, rot2_1);   \
+                    dct_rot(y1o, y3o, row5, row1, rot3_0, rot3_1);   \
+                    __m128i sum17 = _mm_add_epi16(row1, row7);       \
+                    __m128i sum35 = _mm_add_epi16(row3, row5);       \
+                    dct_rot(y4o, y5o, sum17, sum35, rot1_0, rot1_1); \
+                    dct_wadd(x4, y0o, y4o);                          \
+                    dct_wadd(x5, y1o, y5o);                          \
+                    dct_wadd(x6, y2o, y5o);                          \
+                    dct_wadd(x7, y3o, y4o);                          \
+                    dct_bfly32o(row0, row7, x0, x7, bias, shift);    \
+                    dct_bfly32o(row1, row6, x1, x6, bias, shift);    \
+                    dct_bfly32o(row2, row5, x2, x5, bias, shift);    \
+                    dct_bfly32o(row3, row4, x3, x4, bias, shift);    \
+                }
+
+    __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
+    __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f(0.765366865f), stbi__f2f(0.5411961f));
+    __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
+    __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
+    __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f(0.298631336f), stbi__f2f(-1.961570560f));
+    __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f(3.072711026f));
+    __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f(2.053119869f), stbi__f2f(-0.390180644f));
+    __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f(1.501321110f));
+
+    // rounding biases in column/row passes, see stbi__idct_block for explanation.
+    __m128i bias_0 = _mm_set1_epi32(512);
+    __m128i bias_1 = _mm_set1_epi32(65536 + (128 << 17));
+
+    // load
+    row0 = _mm_load_si128((__m128i const*)(data + 0 * 8));
+    row1 = _mm_load_si128((__m128i const*)(data + 1 * 8));
+    row2 = _mm_load_si128((__m128i const*)(data + 2 * 8));
+    row3 = _mm_load_si128((__m128i const*)(data + 3 * 8));
+    row4 = _mm_load_si128((__m128i const*)(data + 4 * 8));
+    row5 = _mm_load_si128((__m128i const*)(data + 5 * 8));
+    row6 = _mm_load_si128((__m128i const*)(data + 6 * 8));
+    row7 = _mm_load_si128((__m128i const*)(data + 7 * 8));
+
+    // column pass
+    dct_pass(bias_0, 10);
+
+    {
+        // 16bit 8x8 transpose pass 1
+        dct_interleave16(row0, row4);
+        dct_interleave16(row1, row5);
+        dct_interleave16(row2, row6);
+        dct_interleave16(row3, row7);
+
+        // transpose pass 2
+        dct_interleave16(row0, row2);
+        dct_interleave16(row1, row3);
+        dct_interleave16(row4, row6);
+        dct_interleave16(row5, row7);
+
+        // transpose pass 3
+        dct_interleave16(row0, row1);
+        dct_interleave16(row2, row3);
+        dct_interleave16(row4, row5);
+        dct_interleave16(row6, row7);
+    }
+
+    // row pass
+    dct_pass(bias_1, 17);
+
+    {
+        // pack
+        __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
+        __m128i p1 = _mm_packus_epi16(row2, row3);
+        __m128i p2 = _mm_packus_epi16(row4, row5);
+        __m128i p3 = _mm_packus_epi16(row6, row7);
+
+        // 8bit 8x8 transpose pass 1
+        dct_interleave8(p0, p2); // a0e0a1e1...
+        dct_interleave8(p1, p3); // c0g0c1g1...
+
+        // transpose pass 2
+        dct_interleave8(p0, p1); // a0c0e0g0...
+        dct_interleave8(p2, p3); // b0d0f0h0...
+
+        // transpose pass 3
+        dct_interleave8(p0, p2); // a0b0c0d0...
+        dct_interleave8(p1, p3); // a4b4c4d4...
+
+        // store
+        _mm_storel_epi64((__m128i*)out, p0);
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p0, 0x4e));
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, p2);
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p2, 0x4e));
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, p1);
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p1, 0x4e));
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, p3);
+        out += out_stride;
+        _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p3, 0x4e));
+    }
+
+#            undef dct_const
+#            undef dct_rot
+#            undef dct_widen
+#            undef dct_wadd
+#            undef dct_wsub
+#            undef dct_bfly32o
+#            undef dct_interleave8
+#            undef dct_interleave16
+#            undef dct_pass
+}
+
+#        endif // STBI_SSE2
+
+#        ifdef STBI_NEON
+
+// NEON integer IDCT. should produce bit-identical
+// results to the generic C version.
+static void stbi__idct_simd(stbi_uc* out, int out_stride, short data[64])
+{
+    int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
+
+    int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
+    int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
+    int16x4_t rot0_2 = vdup_n_s16(stbi__f2f(0.765366865f));
+    int16x4_t rot1_0 = vdup_n_s16(stbi__f2f(1.175875602f));
+    int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
+    int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
+    int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
+    int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
+    int16x4_t rot3_0 = vdup_n_s16(stbi__f2f(0.298631336f));
+    int16x4_t rot3_1 = vdup_n_s16(stbi__f2f(2.053119869f));
+    int16x4_t rot3_2 = vdup_n_s16(stbi__f2f(3.072711026f));
+    int16x4_t rot3_3 = vdup_n_s16(stbi__f2f(1.501321110f));
+
+#            define dct_long_mul(out, inq, coeff)                        \
+                int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
+                int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
+
+#            define dct_long_mac(out, acc, inq, coeff)                            \
+                int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
+                int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
+
+#            define dct_widen(out, inq)                                 \
+                int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
+                int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
+
+// wide add
+#            define dct_wadd(out, a, b)                      \
+                int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
+                int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
+
+// wide sub
+#            define dct_wsub(out, a, b)                      \
+                int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
+                int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
+
+// butterfly a/b, then shift using "shiftop" by "s" and pack
+#            define dct_bfly32o(out0, out1, a, b, shiftop, s)                  \
+                {                                                              \
+                    dct_wadd(sum, a, b);                                       \
+                    dct_wsub(dif, a, b);                                       \
+                    out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
+                    out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
+                }
+
+#            define dct_pass(shiftop, shift)                         \
+                {                                                    \
+                    /* even part */                                  \
+                    int16x8_t sum26 = vaddq_s16(row2, row6);         \
+                    dct_long_mul(p1e, sum26, rot0_0);                \
+                    dct_long_mac(t2e, p1e, row6, rot0_1);            \
+                    dct_long_mac(t3e, p1e, row2, rot0_2);            \
+                    int16x8_t sum04 = vaddq_s16(row0, row4);         \
+                    int16x8_t dif04 = vsubq_s16(row0, row4);         \
+                    dct_widen(t0e, sum04);                           \
+                    dct_widen(t1e, dif04);                           \
+                    dct_wadd(x0, t0e, t3e);                          \
+                    dct_wsub(x3, t0e, t3e);                          \
+                    dct_wadd(x1, t1e, t2e);                          \
+                    dct_wsub(x2, t1e, t2e);                          \
+                    /* odd part */                                   \
+                    int16x8_t sum15 = vaddq_s16(row1, row5);         \
+                    int16x8_t sum17 = vaddq_s16(row1, row7);         \
+                    int16x8_t sum35 = vaddq_s16(row3, row5);         \
+                    int16x8_t sum37 = vaddq_s16(row3, row7);         \
+                    int16x8_t sumodd = vaddq_s16(sum17, sum35);      \
+                    dct_long_mul(p5o, sumodd, rot1_0);               \
+                    dct_long_mac(p1o, p5o, sum17, rot1_1);           \
+                    dct_long_mac(p2o, p5o, sum35, rot1_2);           \
+                    dct_long_mul(p3o, sum37, rot2_0);                \
+                    dct_long_mul(p4o, sum15, rot2_1);                \
+                    dct_wadd(sump13o, p1o, p3o);                     \
+                    dct_wadd(sump24o, p2o, p4o);                     \
+                    dct_wadd(sump23o, p2o, p3o);                     \
+                    dct_wadd(sump14o, p1o, p4o);                     \
+                    dct_long_mac(x4, sump13o, row7, rot3_0);         \
+                    dct_long_mac(x5, sump24o, row5, rot3_1);         \
+                    dct_long_mac(x6, sump23o, row3, rot3_2);         \
+                    dct_long_mac(x7, sump14o, row1, rot3_3);         \
+                    dct_bfly32o(row0, row7, x0, x7, shiftop, shift); \
+                    dct_bfly32o(row1, row6, x1, x6, shiftop, shift); \
+                    dct_bfly32o(row2, row5, x2, x5, shiftop, shift); \
+                    dct_bfly32o(row3, row4, x3, x4, shiftop, shift); \
+                }
+
+    // load
+    row0 = vld1q_s16(data + 0 * 8);
+    row1 = vld1q_s16(data + 1 * 8);
+    row2 = vld1q_s16(data + 2 * 8);
+    row3 = vld1q_s16(data + 3 * 8);
+    row4 = vld1q_s16(data + 4 * 8);
+    row5 = vld1q_s16(data + 5 * 8);
+    row6 = vld1q_s16(data + 6 * 8);
+    row7 = vld1q_s16(data + 7 * 8);
+
+    // add DC bias
+    row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
+
+    // column pass
+    dct_pass(vrshrn_n_s32, 10);
+
+    // 16bit 8x8 transpose
+    {
+// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
+// whether compilers actually get this is another story, sadly.
+#            define dct_trn16(x, y)                  \
+                {                                    \
+                    int16x8x2_t t = vtrnq_s16(x, y); \
+                    x = t.val[0];                    \
+                    y = t.val[1];                    \
+                }
+#            define dct_trn32(x, y)                                                                \
+                {                                                                                  \
+                    int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); \
+                    x = vreinterpretq_s16_s32(t.val[0]);                                           \
+                    y = vreinterpretq_s16_s32(t.val[1]);                                           \
+                }
+#            define dct_trn64(x, y)                                         \
+                {                                                           \
+                    int16x8_t x0 = x;                                       \
+                    int16x8_t y0 = y;                                       \
+                    x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0));   \
+                    y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); \
+                }
+
+        // pass 1
+        dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
+        dct_trn16(row2, row3);
+        dct_trn16(row4, row5);
+        dct_trn16(row6, row7);
+
+        // pass 2
+        dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
+        dct_trn32(row1, row3);
+        dct_trn32(row4, row6);
+        dct_trn32(row5, row7);
+
+        // pass 3
+        dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
+        dct_trn64(row1, row5);
+        dct_trn64(row2, row6);
+        dct_trn64(row3, row7);
+
+#            undef dct_trn16
+#            undef dct_trn32
+#            undef dct_trn64
+    }
+
+    // row pass
+    // vrshrn_n_s32 only supports shifts up to 16, we need
+    // 17. so do a non-rounding shift of 16 first then follow
+    // up with a rounding shift by 1.
+    dct_pass(vshrn_n_s32, 16);
+
+    {
+        // pack and round
+        uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
+        uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
+        uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
+        uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
+        uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
+        uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
+        uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
+        uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
+
+        // again, these can translate into one instruction, but often don't.
+#            define dct_trn8_8(x, y)               \
+                {                                  \
+                    uint8x8x2_t t = vtrn_u8(x, y); \
+                    x = t.val[0];                  \
+                    y = t.val[1];                  \
+                }
+#            define dct_trn8_16(x, y)                                                          \
+                {                                                                              \
+                    uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); \
+                    x = vreinterpret_u8_u16(t.val[0]);                                         \
+                    y = vreinterpret_u8_u16(t.val[1]);                                         \
+                }
+#            define dct_trn8_32(x, y)                                                          \
+                {                                                                              \
+                    uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); \
+                    x = vreinterpret_u8_u32(t.val[0]);                                         \
+                    y = vreinterpret_u8_u32(t.val[1]);                                         \
+                }
+
+        // sadly can't use interleaved stores here since we only write
+        // 8 bytes to each scan line!
+
+        // 8x8 8-bit transpose pass 1
+        dct_trn8_8(p0, p1);
+        dct_trn8_8(p2, p3);
+        dct_trn8_8(p4, p5);
+        dct_trn8_8(p6, p7);
+
+        // pass 2
+        dct_trn8_16(p0, p2);
+        dct_trn8_16(p1, p3);
+        dct_trn8_16(p4, p6);
+        dct_trn8_16(p5, p7);
+
+        // pass 3
+        dct_trn8_32(p0, p4);
+        dct_trn8_32(p1, p5);
+        dct_trn8_32(p2, p6);
+        dct_trn8_32(p3, p7);
+
+        // store
+        vst1_u8(out, p0);
+        out += out_stride;
+        vst1_u8(out, p1);
+        out += out_stride;
+        vst1_u8(out, p2);
+        out += out_stride;
+        vst1_u8(out, p3);
+        out += out_stride;
+        vst1_u8(out, p4);
+        out += out_stride;
+        vst1_u8(out, p5);
+        out += out_stride;
+        vst1_u8(out, p6);
+        out += out_stride;
+        vst1_u8(out, p7);
+
+#            undef dct_trn8_8
+#            undef dct_trn8_16
+#            undef dct_trn8_32
+    }
+
+#            undef dct_long_mul
+#            undef dct_long_mac
+#            undef dct_widen
+#            undef dct_wadd
+#            undef dct_wsub
+#            undef dct_bfly32o
+#            undef dct_pass
+}
+
+#        endif // STBI_NEON
+
+#        define STBI__MARKER_none 0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg* j)
+{
+    stbi_uc x;
+    if (j->marker != STBI__MARKER_none) {
+        x = j->marker;
+        j->marker = STBI__MARKER_none;
+        return x;
+    }
+    x = stbi__get8(j->s);
+    if (x != 0xff)
+        return STBI__MARKER_none;
+    while (x == 0xff)
+        x = stbi__get8(j->s); // consume repeated 0xff fill bytes
+    return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#        define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg* j)
+{
+    j->code_bits = 0;
+    j->code_buffer = 0;
+    j->nomore = 0;
+    j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
+    j->marker = STBI__MARKER_none;
+    j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+    j->eob_run = 0;
+    // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
+    // since we don't even allow 1<<30 pixels
+}
+
+static int stbi__parse_entropy_coded_data(stbi__jpeg* z)
+{
+    stbi__jpeg_reset(z);
+    if (!z->progressive) {
+        if (z->scan_n == 1) {
+            int i, j;
+            STBI_SIMD_ALIGN(short, data[64]);
+            int n = z->order[0];
+            // non-interleaved data, we just need to process one block at a time,
+            // in trivial scanline order
+            // number of blocks to do just depends on how many actual "pixels" this
+            // component has, independent of interleaved MCU blocking and such
+            int w = (z->img_comp[n].x + 7) >> 3;
+            int h = (z->img_comp[n].y + 7) >> 3;
+            for (j = 0; j < h; ++j) {
+                for (i = 0; i < w; ++i) {
+                    int ha = z->img_comp[n].ha;
+                    if (!stbi__jpeg_decode_block(z, data, z->huff_dc + z->img_comp[n].hd, z->huff_ac + ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq]))
+                        return 0;
+                    z->idct_block_kernel(z->img_comp[n].data + z->img_comp[n].w2 * j * 8 + i * 8, z->img_comp[n].w2, data);
+                    // every data block is an MCU, so countdown the restart interval
+                    if (--z->todo <= 0) {
+                        if (z->code_bits < 24)
+                            stbi__grow_buffer_unsafe(z);
+                        // if it's NOT a restart, then just bail, so we get corrupt data
+                        // rather than no data
+                        if (!STBI__RESTART(z->marker))
+                            return 1;
+                        stbi__jpeg_reset(z);
+                    }
+                }
+            }
+            return 1;
+        } else { // interleaved
+            int i, j, k, x, y;
+            STBI_SIMD_ALIGN(short, data[64]);
+            for (j = 0; j < z->img_mcu_y; ++j) {
+                for (i = 0; i < z->img_mcu_x; ++i) {
+                    // scan an interleaved mcu... process scan_n components in order
+                    for (k = 0; k < z->scan_n; ++k) {
+                        int n = z->order[k];
+                        // scan out an mcu's worth of this component; that's just determined
+                        // by the basic H and V specified for the component
+                        for (y = 0; y < z->img_comp[n].v; ++y) {
+                            for (x = 0; x < z->img_comp[n].h; ++x) {
+                                int x2 = (i * z->img_comp[n].h + x) * 8;
+                                int y2 = (j * z->img_comp[n].v + y) * 8;
+                                int ha = z->img_comp[n].ha;
+                                if (!stbi__jpeg_decode_block(z, data, z->huff_dc + z->img_comp[n].hd, z->huff_ac + ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq]))
+                                    return 0;
+                                z->idct_block_kernel(z->img_comp[n].data + z->img_comp[n].w2 * y2 + x2, z->img_comp[n].w2, data);
+                            }
+                        }
+                    }
+                    // after all interleaved components, that's an interleaved MCU,
+                    // so now count down the restart interval
+                    if (--z->todo <= 0) {
+                        if (z->code_bits < 24)
+                            stbi__grow_buffer_unsafe(z);
+                        if (!STBI__RESTART(z->marker))
+                            return 1;
+                        stbi__jpeg_reset(z);
+                    }
+                }
+            }
+            return 1;
+        }
+    } else {
+        if (z->scan_n == 1) {
+            int i, j;
+            int n = z->order[0];
+            // non-interleaved data, we just need to process one block at a time,
+            // in trivial scanline order
+            // number of blocks to do just depends on how many actual "pixels" this
+            // component has, independent of interleaved MCU blocking and such
+            int w = (z->img_comp[n].x + 7) >> 3;
+            int h = (z->img_comp[n].y + 7) >> 3;
+            for (j = 0; j < h; ++j) {
+                for (i = 0; i < w; ++i) {
+                    short* data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+                    if (z->spec_start == 0) {
+                        if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+                            return 0;
+                    } else {
+                        int ha = z->img_comp[n].ha;
+                        if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
+                            return 0;
+                    }
+                    // every data block is an MCU, so countdown the restart interval
+                    if (--z->todo <= 0) {
+                        if (z->code_bits < 24)
+                            stbi__grow_buffer_unsafe(z);
+                        if (!STBI__RESTART(z->marker))
+                            return 1;
+                        stbi__jpeg_reset(z);
+                    }
+                }
+            }
+            return 1;
+        } else { // interleaved
+            int i, j, k, x, y;
+            for (j = 0; j < z->img_mcu_y; ++j) {
+                for (i = 0; i < z->img_mcu_x; ++i) {
+                    // scan an interleaved mcu... process scan_n components in order
+                    for (k = 0; k < z->scan_n; ++k) {
+                        int n = z->order[k];
+                        // scan out an mcu's worth of this component; that's just determined
+                        // by the basic H and V specified for the component
+                        for (y = 0; y < z->img_comp[n].v; ++y) {
+                            for (x = 0; x < z->img_comp[n].h; ++x) {
+                                int x2 = (i * z->img_comp[n].h + x);
+                                int y2 = (j * z->img_comp[n].v + y);
+                                short* data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+                                if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+                                    return 0;
+                            }
+                        }
+                    }
+                    // after all interleaved components, that's an interleaved MCU,
+                    // so now count down the restart interval
+                    if (--z->todo <= 0) {
+                        if (z->code_bits < 24)
+                            stbi__grow_buffer_unsafe(z);
+                        if (!STBI__RESTART(z->marker))
+                            return 1;
+                        stbi__jpeg_reset(z);
+                    }
+                }
+            }
+            return 1;
+        }
+    }
+}
+
+static void stbi__jpeg_dequantize(short* data, stbi__uint16* dequant)
+{
+    int i;
+    for (i = 0; i < 64; ++i)
+        data[i] *= dequant[i];
+}
+
+static void stbi__jpeg_finish(stbi__jpeg* z)
+{
+    if (z->progressive) {
+        // dequantize and idct the data
+        int i, j, n;
+        for (n = 0; n < z->s->img_n; ++n) {
+            int w = (z->img_comp[n].x + 7) >> 3;
+            int h = (z->img_comp[n].y + 7) >> 3;
+            for (j = 0; j < h; ++j) {
+                for (i = 0; i < w; ++i) {
+                    short* data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+                    stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
+                    z->idct_block_kernel(z->img_comp[n].data + z->img_comp[n].w2 * j * 8 + i * 8, z->img_comp[n].w2, data);
+                }
+            }
+        }
+    }
+}
+
+static int stbi__process_marker(stbi__jpeg* z, int m)
+{
+    int L;
+    switch (m) {
+    case STBI__MARKER_none: // no marker found
+        return stbi__err("expected marker", "Corrupt JPEG");
+
+    case 0xDD: // DRI - specify restart interval
+        if (stbi__get16be(z->s) != 4)
+            return stbi__err("bad DRI len", "Corrupt JPEG");
+        z->restart_interval = stbi__get16be(z->s);
+        return 1;
+
+    case 0xDB: // DQT - define quantization table
+        L = stbi__get16be(z->s) - 2;
+        while (L > 0) {
+            int q = stbi__get8(z->s);
+            int p = q >> 4, sixteen = (p != 0);
+            int t = q & 15, i;
+            if (p != 0 && p != 1)
+                return stbi__err("bad DQT type", "Corrupt JPEG");
+            if (t > 3)
+                return stbi__err("bad DQT table", "Corrupt JPEG");
+
+            for (i = 0; i < 64; ++i)
+                z->dequant[t][stbi__jpeg_dezigzag[i]] = (stbi__uint16)(sixteen ? stbi__get16be(z->s) : stbi__get8(z->s));
+            L -= (sixteen ? 129 : 65);
+        }
+        return L == 0;
+
+    case 0xC4: // DHT - define huffman table
+        L = stbi__get16be(z->s) - 2;
+        while (L > 0) {
+            stbi_uc* v;
+            int sizes[16], i, n = 0;
+            int q = stbi__get8(z->s);
+            int tc = q >> 4;
+            int th = q & 15;
+            if (tc > 1 || th > 3)
+                return stbi__err("bad DHT header", "Corrupt JPEG");
+            for (i = 0; i < 16; ++i) {
+                sizes[i] = stbi__get8(z->s);
+                n += sizes[i];
+            }
+            if (n > 256)
+                return stbi__err("bad DHT header", "Corrupt JPEG"); // Loop over i < n would write past end of values!
+            L -= 17;
+            if (tc == 0) {
+                if (!stbi__build_huffman(z->huff_dc + th, sizes))
+                    return 0;
+                v = z->huff_dc[th].values;
+            } else {
+                if (!stbi__build_huffman(z->huff_ac + th, sizes))
+                    return 0;
+                v = z->huff_ac[th].values;
+            }
+            for (i = 0; i < n; ++i)
+                v[i] = stbi__get8(z->s);
+            if (tc != 0)
+                stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
+            L -= n;
+        }
+        return L == 0;
+    }
+
+    // check for comment block or APP blocks
+    if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+        L = stbi__get16be(z->s);
+        if (L < 2) {
+            if (m == 0xFE)
+                return stbi__err("bad COM len", "Corrupt JPEG");
+            else
+                return stbi__err("bad APP len", "Corrupt JPEG");
+        }
+        L -= 2;
+
+        if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
+            static unsigned char const tag[5] = { 'J', 'F', 'I', 'F', '\0' };
+            int ok = 1;
+            int i;
+            for (i = 0; i < 5; ++i)
+                if (stbi__get8(z->s) != tag[i])
+                    ok = 0;
+            L -= 5;
+            if (ok)
+                z->jfif = 1;
+        } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
+            static unsigned char const tag[6] = { 'A', 'd', 'o', 'b', 'e', '\0' };
+            int ok = 1;
+            int i;
+            for (i = 0; i < 6; ++i)
+                if (stbi__get8(z->s) != tag[i])
+                    ok = 0;
+            L -= 6;
+            if (ok) {
+                stbi__get8(z->s);                            // version
+                stbi__get16be(z->s);                         // flags0
+                stbi__get16be(z->s);                         // flags1
+                z->app14_color_transform = stbi__get8(z->s); // color transform
+                L -= 6;
+            }
+        }
+
+        stbi__skip(z->s, L);
+        return 1;
+    }
+
+    return stbi__err("unknown marker", "Corrupt JPEG");
+}
+
+// after we see SOS
+static int stbi__process_scan_header(stbi__jpeg* z)
+{
+    int i;
+    int Ls = stbi__get16be(z->s);
+    z->scan_n = stbi__get8(z->s);
+    if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int)z->s->img_n)
+        return stbi__err("bad SOS component count", "Corrupt JPEG");
+    if (Ls != 6 + 2 * z->scan_n)
+        return stbi__err("bad SOS len", "Corrupt JPEG");
+    for (i = 0; i < z->scan_n; ++i) {
+        int id = stbi__get8(z->s), which;
+        int q = stbi__get8(z->s);
+        for (which = 0; which < z->s->img_n; ++which)
+            if (z->img_comp[which].id == id)
+                break;
+        if (which == z->s->img_n)
+            return 0; // no match
+        z->img_comp[which].hd = q >> 4;
+        if (z->img_comp[which].hd > 3)
+            return stbi__err("bad DC huff", "Corrupt JPEG");
+        z->img_comp[which].ha = q & 15;
+        if (z->img_comp[which].ha > 3)
+            return stbi__err("bad AC huff", "Corrupt JPEG");
+        z->order[i] = which;
+    }
+
+    {
+        int aa;
+        z->spec_start = stbi__get8(z->s);
+        z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
+        aa = stbi__get8(z->s);
+        z->succ_high = (aa >> 4);
+        z->succ_low = (aa & 15);
+        if (z->progressive) {
+            if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+                return stbi__err("bad SOS", "Corrupt JPEG");
+        } else {
+            if (z->spec_start != 0)
+                return stbi__err("bad SOS", "Corrupt JPEG");
+            if (z->succ_high != 0 || z->succ_low != 0)
+                return stbi__err("bad SOS", "Corrupt JPEG");
+            z->spec_end = 63;
+        }
+    }
+
+    return 1;
+}
+
+static int stbi__free_jpeg_components(stbi__jpeg* z, int ncomp, int why)
+{
+    int i;
+    for (i = 0; i < ncomp; ++i) {
+        if (z->img_comp[i].raw_data) {
+            STBI_FREE(z->img_comp[i].raw_data);
+            z->img_comp[i].raw_data = NULL;
+            z->img_comp[i].data = NULL;
+        }
+        if (z->img_comp[i].raw_coeff) {
+            STBI_FREE(z->img_comp[i].raw_coeff);
+            z->img_comp[i].raw_coeff = 0;
+            z->img_comp[i].coeff = 0;
+        }
+        if (z->img_comp[i].linebuf) {
+            STBI_FREE(z->img_comp[i].linebuf);
+            z->img_comp[i].linebuf = NULL;
+        }
+    }
+    return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg* z, int scan)
+{
+    stbi__context* s = z->s;
+    int Lf, p, i, q, h_max = 1, v_max = 1, c;
+    Lf = stbi__get16be(s);
+    if (Lf < 11)
+        return stbi__err("bad SOF len", "Corrupt JPEG"); // JPEG
+    p = stbi__get8(s);
+    if (p != 8)
+        return stbi__err("only 8-bit", "JPEG format not supported: 8-bit only"); // JPEG baseline
+    s->img_y = stbi__get16be(s);
+    if (s->img_y == 0)
+        return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+    s->img_x = stbi__get16be(s);
+    if (s->img_x == 0)
+        return stbi__err("0 width", "Corrupt JPEG"); // JPEG requires
+    if (s->img_y > STBI_MAX_DIMENSIONS)
+        return stbi__err("too large", "Very large image (corrupt?)");
+    if (s->img_x > STBI_MAX_DIMENSIONS)
+        return stbi__err("too large", "Very large image (corrupt?)");
+    c = stbi__get8(s);
+    if (c != 3 && c != 1 && c != 4)
+        return stbi__err("bad component count", "Corrupt JPEG");
+    s->img_n = c;
+    for (i = 0; i < c; ++i) {
+        z->img_comp[i].data = NULL;
+        z->img_comp[i].linebuf = NULL;
+    }
+
+    if (Lf != 8 + 3 * s->img_n)
+        return stbi__err("bad SOF len", "Corrupt JPEG");
+
+    z->rgb = 0;
+    for (i = 0; i < s->img_n; ++i) {
+        static unsigned char const rgb[3] = { 'R', 'G', 'B' };
+        z->img_comp[i].id = stbi__get8(s);
+        if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+            ++z->rgb;
+        q = stbi__get8(s);
+        z->img_comp[i].h = (q >> 4);
+        if (!z->img_comp[i].h || z->img_comp[i].h > 4)
+            return stbi__err("bad H", "Corrupt JPEG");
+        z->img_comp[i].v = q & 15;
+        if (!z->img_comp[i].v || z->img_comp[i].v > 4)
+            return stbi__err("bad V", "Corrupt JPEG");
+        z->img_comp[i].tq = stbi__get8(s);
+        if (z->img_comp[i].tq > 3)
+            return stbi__err("bad TQ", "Corrupt JPEG");
+    }
+
+    if (scan != STBI__SCAN_load)
+        return 1;
+
+    if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0))
+        return stbi__err("too large", "Image too large to decode");
+
+    for (i = 0; i < s->img_n; ++i) {
+        if (z->img_comp[i].h > h_max)
+            h_max = z->img_comp[i].h;
+        if (z->img_comp[i].v > v_max)
+            v_max = z->img_comp[i].v;
+    }
+
+    // check that plane subsampling factors are integer ratios; our resamplers can't deal with fractional ratios
+    // and I've never seen a non-corrupted JPEG file actually use them
+    for (i = 0; i < s->img_n; ++i) {
+        if (h_max % z->img_comp[i].h != 0)
+            return stbi__err("bad H", "Corrupt JPEG");
+        if (v_max % z->img_comp[i].v != 0)
+            return stbi__err("bad V", "Corrupt JPEG");
+    }
+
+    // compute interleaved mcu info
+    z->img_h_max = h_max;
+    z->img_v_max = v_max;
+    z->img_mcu_w = h_max * 8;
+    z->img_mcu_h = v_max * 8;
+    // these sizes can't be more than 17 bits
+    z->img_mcu_x = (s->img_x + z->img_mcu_w - 1) / z->img_mcu_w;
+    z->img_mcu_y = (s->img_y + z->img_mcu_h - 1) / z->img_mcu_h;
+
+    for (i = 0; i < s->img_n; ++i) {
+        // number of effective pixels (e.g. for non-interleaved MCU)
+        z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max - 1) / h_max;
+        z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max - 1) / v_max;
+        // to simplify generation, we'll allocate enough memory to decode
+        // the bogus oversized data from using interleaved MCUs and their
+        // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
+        // discard the extra data until colorspace conversion
+        //
+        // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+        // so these muls can't overflow with 32-bit ints (which we require)
+        z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+        z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+        z->img_comp[i].coeff = 0;
+        z->img_comp[i].raw_coeff = 0;
+        z->img_comp[i].linebuf = NULL;
+        z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+        if (z->img_comp[i].raw_data == NULL)
+            return stbi__free_jpeg_components(z, i + 1, stbi__err("outofmem", "Out of memory"));
+        // align blocks for idct using mmx/sse
+        z->img_comp[i].data = (stbi_uc*)(((size_t)z->img_comp[i].raw_data + 15) & ~15);
+        if (z->progressive) {
+            // w2, h2 are multiples of 8 (see above)
+            z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+            z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+            z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+            if (z->img_comp[i].raw_coeff == NULL)
+                return stbi__free_jpeg_components(z, i + 1, stbi__err("outofmem", "Out of memory"));
+            z->img_comp[i].coeff = (short*)(((size_t)z->img_comp[i].raw_coeff + 15) & ~15);
+        }
+    }
+
+    return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (e.g. SOF)
+#        define stbi__DNL(x) ((x) == 0xdc)
+#        define stbi__SOI(x) ((x) == 0xd8)
+#        define stbi__EOI(x) ((x) == 0xd9)
+#        define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#        define stbi__SOS(x) ((x) == 0xda)
+
+#        define stbi__SOF_progressive(x) ((x) == 0xc2)
+
+static int stbi__decode_jpeg_header(stbi__jpeg* z, int scan)
+{
+    int m;
+    z->jfif = 0;
+    z->app14_color_transform = -1; // valid values are 0,1,2
+    z->marker = STBI__MARKER_none; // initialize cached marker to empty
+    m = stbi__get_marker(z);
+    if (!stbi__SOI(m))
+        return stbi__err("no SOI", "Corrupt JPEG");
+    if (scan == STBI__SCAN_type)
+        return 1;
+    m = stbi__get_marker(z);
+    while (!stbi__SOF(m)) {
+        if (!stbi__process_marker(z, m))
+            return 0;
+        m = stbi__get_marker(z);
+        while (m == STBI__MARKER_none) {
+            // some files have extra padding after their blocks, so ok, we'll scan
+            if (stbi__at_eof(z->s))
+                return stbi__err("no SOF", "Corrupt JPEG");
+            m = stbi__get_marker(z);
+        }
+    }
+    z->progressive = stbi__SOF_progressive(m);
+    if (!stbi__process_frame_header(z, scan))
+        return 0;
+    return 1;
+}
+
+static stbi_uc stbi__skip_jpeg_junk_at_end(stbi__jpeg* j)
+{
+    // some JPEGs have junk at end, skip over it but if we find what looks
+    // like a valid marker, resume there
+    while (!stbi__at_eof(j->s)) {
+        stbi_uc x = stbi__get8(j->s);
+        while (x == 0xff) { // might be a marker
+            if (stbi__at_eof(j->s))
+                return STBI__MARKER_none;
+            x = stbi__get8(j->s);
+            if (x != 0x00 && x != 0xff) {
+                // not a stuffed zero or lead-in to another marker, looks
+                // like an actual marker, return it
+                return x;
+            }
+            // stuffed zero has x=0 now which ends the loop, meaning we go
+            // back to regular scan loop.
+            // repeated 0xff keeps trying to read the next byte of the marker.
+        }
+    }
+    return STBI__MARKER_none;
+}
+
+// decode image to YCbCr format
+static int stbi__decode_jpeg_image(stbi__jpeg* j)
+{
+    int m;
+    for (m = 0; m < 4; m++) {
+        j->img_comp[m].raw_data = NULL;
+        j->img_comp[m].raw_coeff = NULL;
+    }
+    j->restart_interval = 0;
+    if (!stbi__decode_jpeg_header(j, STBI__SCAN_load))
+        return 0;
+    m = stbi__get_marker(j);
+    while (!stbi__EOI(m)) {
+        if (stbi__SOS(m)) {
+            if (!stbi__process_scan_header(j))
+                return 0;
+            if (!stbi__parse_entropy_coded_data(j))
+                return 0;
+            if (j->marker == STBI__MARKER_none) {
+                j->marker = stbi__skip_jpeg_junk_at_end(j);
+                // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+            }
+            m = stbi__get_marker(j);
+            if (STBI__RESTART(m))
+                m = stbi__get_marker(j);
+        } else if (stbi__DNL(m)) {
+            int Ld = stbi__get16be(j->s);
+            stbi__uint32 NL = stbi__get16be(j->s);
+            if (Ld != 4)
+                return stbi__err("bad DNL len", "Corrupt JPEG");
+            if (NL != j->s->img_y)
+                return stbi__err("bad DNL height", "Corrupt JPEG");
+            m = stbi__get_marker(j);
+        } else {
+            if (!stbi__process_marker(j, m))
+                return 1;
+            m = stbi__get_marker(j);
+        }
+    }
+    if (j->progressive)
+        stbi__jpeg_finish(j);
+    return 1;
+}
+
+// static jfif-centered resampling (across block boundaries)
+
+typedef stbi_uc* (*resample_row_func)(stbi_uc* out, stbi_uc* in0, stbi_uc* in1,
+    int w, int hs);
+
+#        define stbi__div4(x) ((stbi_uc)((x) >> 2))
+
+static stbi_uc* resample_row_1(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
+{
+    STBI_NOTUSED(out);
+    STBI_NOTUSED(in_far);
+    STBI_NOTUSED(w);
+    STBI_NOTUSED(hs);
+    return in_near;
+}
+
+static stbi_uc* stbi__resample_row_v_2(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
+{
+    // need to generate two samples vertically for every one in input
+    int i;
+    STBI_NOTUSED(hs);
+    for (i = 0; i < w; ++i)
+        out[i] = stbi__div4(3 * in_near[i] + in_far[i] + 2);
+    return out;
+}
+
+static stbi_uc* stbi__resample_row_h_2(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
+{
+    // need to generate two samples horizontally for every one in input
+    int i;
+    stbi_uc* input = in_near;
+
+    if (w == 1) {
+        // if only one sample, can't do any interpolation
+        out[0] = out[1] = input[0];
+        return out;
+    }
+
+    out[0] = input[0];
+    out[1] = stbi__div4(input[0] * 3 + input[1] + 2);
+    for (i = 1; i < w - 1; ++i) {
+        int n = 3 * input[i] + 2;
+        out[i * 2 + 0] = stbi__div4(n + input[i - 1]);
+        out[i * 2 + 1] = stbi__div4(n + input[i + 1]);
+    }
+    out[i * 2 + 0] = stbi__div4(input[w - 2] * 3 + input[w - 1] + 2);
+    out[i * 2 + 1] = input[w - 1];
+
+    STBI_NOTUSED(in_far);
+    STBI_NOTUSED(hs);
+
+    return out;
+}
+
+#        define stbi__div16(x) ((stbi_uc)((x) >> 4))
+
+static stbi_uc* stbi__resample_row_hv_2(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
+{
+    // need to generate 2x2 samples for every one in input
+    int i, t0, t1;
+    if (w == 1) {
+        out[0] = out[1] = stbi__div4(3 * in_near[0] + in_far[0] + 2);
+        return out;
+    }
+
+    t1 = 3 * in_near[0] + in_far[0];
+    out[0] = stbi__div4(t1 + 2);
+    for (i = 1; i < w; ++i) {
+        t0 = t1;
+        t1 = 3 * in_near[i] + in_far[i];
+        out[i * 2 - 1] = stbi__div16(3 * t0 + t1 + 8);
+        out[i * 2] = stbi__div16(3 * t1 + t0 + 8);
+    }
+    out[w * 2 - 1] = stbi__div4(t1 + 2);
+
+    STBI_NOTUSED(hs);
+
+    return out;
+}
+
+#        if defined(STBI_SSE2) || defined(STBI_NEON)
+static stbi_uc* stbi__resample_row_hv_2_simd(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
+{
+    // need to generate 2x2 samples for every one in input
+    int i = 0, t0, t1;
+
+    if (w == 1) {
+        out[0] = out[1] = stbi__div4(3 * in_near[0] + in_far[0] + 2);
+        return out;
+    }
+
+    t1 = 3 * in_near[0] + in_far[0];
+    // process groups of 8 pixels for as long as we can.
+    // note we can't handle the last pixel in a row in this loop
+    // because we need to handle the filter boundary conditions.
+    for (; i < ((w - 1) & ~7); i += 8) {
+#            if defined(STBI_SSE2)
+        // load and perform the vertical filtering pass
+        // this uses 3*x + y = 4*x + (y - x)
+        __m128i zero = _mm_setzero_si128();
+        __m128i farb = _mm_loadl_epi64((__m128i*)(in_far + i));
+        __m128i nearb = _mm_loadl_epi64((__m128i*)(in_near + i));
+        __m128i farw = _mm_unpacklo_epi8(farb, zero);
+        __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
+        __m128i diff = _mm_sub_epi16(farw, nearw);
+        __m128i nears = _mm_slli_epi16(nearw, 2);
+        __m128i curr = _mm_add_epi16(nears, diff); // current row
+
+        // horizontal filter works the same based on shifted vers of current
+        // row. "prev" is current row shifted right by 1 pixel; we need to
+        // insert the previous pixel value (from t1).
+        // "next" is current row shifted left by 1 pixel, with first pixel
+        // of next block of 8 pixels added in.
+        __m128i prv0 = _mm_slli_si128(curr, 2);
+        __m128i nxt0 = _mm_srli_si128(curr, 2);
+        __m128i prev = _mm_insert_epi16(prv0, t1, 0);
+        __m128i next = _mm_insert_epi16(nxt0, 3 * in_near[i + 8] + in_far[i + 8], 7);
+
+        // horizontal filter, polyphase implementation since it's convenient:
+        // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+        // odd  pixels = 3*cur + next = cur*4 + (next - cur)
+        // note the shared term.
+        __m128i bias = _mm_set1_epi16(8);
+        __m128i curs = _mm_slli_epi16(curr, 2);
+        __m128i prvd = _mm_sub_epi16(prev, curr);
+        __m128i nxtd = _mm_sub_epi16(next, curr);
+        __m128i curb = _mm_add_epi16(curs, bias);
+        __m128i even = _mm_add_epi16(prvd, curb);
+        __m128i odd = _mm_add_epi16(nxtd, curb);
+
+        // interleave even and odd pixels, then undo scaling.
+        __m128i int0 = _mm_unpacklo_epi16(even, odd);
+        __m128i int1 = _mm_unpackhi_epi16(even, odd);
+        __m128i de0 = _mm_srli_epi16(int0, 4);
+        __m128i de1 = _mm_srli_epi16(int1, 4);
+
+        // pack and write output
+        __m128i outv = _mm_packus_epi16(de0, de1);
+        _mm_storeu_si128((__m128i*)(out + i * 2), outv);
+#            elif defined(STBI_NEON)
+        // load and perform the vertical filtering pass
+        // this uses 3*x + y = 4*x + (y - x)
+        uint8x8_t farb = vld1_u8(in_far + i);
+        uint8x8_t nearb = vld1_u8(in_near + i);
+        int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+        int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
+        int16x8_t curr = vaddq_s16(nears, diff); // current row
+
+        // horizontal filter works the same based on shifted vers of current
+        // row. "prev" is current row shifted right by 1 pixel; we need to
+        // insert the previous pixel value (from t1).
+        // "next" is current row shifted left by 1 pixel, with first pixel
+        // of next block of 8 pixels added in.
+        int16x8_t prv0 = vextq_s16(curr, curr, 7);
+        int16x8_t nxt0 = vextq_s16(curr, curr, 1);
+        int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
+        int16x8_t next = vsetq_lane_s16(3 * in_near[i + 8] + in_far[i + 8], nxt0, 7);
+
+        // horizontal filter, polyphase implementation since it's convenient:
+        // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+        // odd  pixels = 3*cur + next = cur*4 + (next - cur)
+        // note the shared term.
+        int16x8_t curs = vshlq_n_s16(curr, 2);
+        int16x8_t prvd = vsubq_s16(prev, curr);
+        int16x8_t nxtd = vsubq_s16(next, curr);
+        int16x8_t even = vaddq_s16(curs, prvd);
+        int16x8_t odd = vaddq_s16(curs, nxtd);
+
+        // undo scaling and round, then store with even/odd phases interleaved
+        uint8x8x2_t o;
+        o.val[0] = vqrshrun_n_s16(even, 4);
+        o.val[1] = vqrshrun_n_s16(odd, 4);
+        vst2_u8(out + i * 2, o);
+#            endif
+
+        // "previous" value for next iter
+        t1 = 3 * in_near[i + 7] + in_far[i + 7];
+    }
+
+    t0 = t1;
+    t1 = 3 * in_near[i] + in_far[i];
+    out[i * 2] = stbi__div16(3 * t1 + t0 + 8);
+
+    for (++i; i < w; ++i) {
+        t0 = t1;
+        t1 = 3 * in_near[i] + in_far[i];
+        out[i * 2 - 1] = stbi__div16(3 * t0 + t1 + 8);
+        out[i * 2] = stbi__div16(3 * t1 + t0 + 8);
+    }
+    out[w * 2 - 1] = stbi__div4(t1 + 2);
+
+    STBI_NOTUSED(hs);
+
+    return out;
+}
+#        endif
+
+static stbi_uc* stbi__resample_row_generic(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
+{
+    // resample with nearest-neighbor
+    int i, j;
+    STBI_NOTUSED(in_far);
+    for (i = 0; i < w; ++i)
+        for (j = 0; j < hs; ++j)
+            out[i * hs + j] = in_near[i];
+    return out;
+}
+
+// this is a reduced-precision calculation of YCbCr-to-RGB introduced
+// to make sure the code produces the same results in both SIMD and scalar
+#        define stbi__float2fixed(x) (((int)((x) * 4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc* out, stbi_uc const* y, stbi_uc const* pcb, stbi_uc const* pcr, int count, int step)
+{
+    int i;
+    for (i = 0; i < count; ++i) {
+        int y_fixed = (y[i] << 20) + (1 << 19); // rounding
+        int r, g, b;
+        int cr = pcr[i] - 128;
+        int cb = pcb[i] - 128;
+        r = y_fixed + cr * stbi__float2fixed(1.40200f);
+        g = y_fixed + (cr * -stbi__float2fixed(0.71414f)) + ((cb * -stbi__float2fixed(0.34414f)) & 0xffff0000);
+        b = y_fixed + cb * stbi__float2fixed(1.77200f);
+        r >>= 20;
+        g >>= 20;
+        b >>= 20;
+        if ((unsigned)r > 255) {
+            if (r < 0)
+                r = 0;
+            else
+                r = 255;
+        }
+        if ((unsigned)g > 255) {
+            if (g < 0)
+                g = 0;
+            else
+                g = 255;
+        }
+        if ((unsigned)b > 255) {
+            if (b < 0)
+                b = 0;
+            else
+                b = 255;
+        }
+        out[0] = (stbi_uc)r;
+        out[1] = (stbi_uc)g;
+        out[2] = (stbi_uc)b;
+        out[3] = 255;
+        out += step;
+    }
+}
+
+#        if defined(STBI_SSE2) || defined(STBI_NEON)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc* out, stbi_uc const* y, stbi_uc const* pcb, stbi_uc const* pcr, int count, int step)
+{
+    int i = 0;
+
+#            ifdef STBI_SSE2
+    // step == 3 is pretty ugly on the final interleave, and i'm not convinced
+    // it's useful in practice (you wouldn't use it for textures, for example).
+    // so just accelerate step == 4 case.
+    if (step == 4) {
+        // this is a fairly straightforward implementation and not super-optimized.
+        __m128i signflip = _mm_set1_epi8(-0x80);
+        __m128i cr_const0 = _mm_set1_epi16((short)(1.40200f * 4096.0f + 0.5f));
+        __m128i cr_const1 = _mm_set1_epi16(-(short)(0.71414f * 4096.0f + 0.5f));
+        __m128i cb_const0 = _mm_set1_epi16(-(short)(0.34414f * 4096.0f + 0.5f));
+        __m128i cb_const1 = _mm_set1_epi16((short)(1.77200f * 4096.0f + 0.5f));
+        __m128i y_bias = _mm_set1_epi8((char)(unsigned char)128);
+        __m128i xw = _mm_set1_epi16(255); // alpha channel
+
+        for (; i + 7 < count; i += 8) {
+            // load
+            __m128i y_bytes = _mm_loadl_epi64((__m128i*)(y + i));
+            __m128i cr_bytes = _mm_loadl_epi64((__m128i*)(pcr + i));
+            __m128i cb_bytes = _mm_loadl_epi64((__m128i*)(pcb + i));
+            __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
+            __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
+
+            // unpack to short (and left-shift cr, cb by 8)
+            __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
+            __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
+            __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
+
+            // color transform
+            __m128i yws = _mm_srli_epi16(yw, 4);
+            __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
+            __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
+            __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
+            __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
+            __m128i rws = _mm_add_epi16(cr0, yws);
+            __m128i gwt = _mm_add_epi16(cb0, yws);
+            __m128i bws = _mm_add_epi16(yws, cb1);
+            __m128i gws = _mm_add_epi16(gwt, cr1);
+
+            // descale
+            __m128i rw = _mm_srai_epi16(rws, 4);
+            __m128i bw = _mm_srai_epi16(bws, 4);
+            __m128i gw = _mm_srai_epi16(gws, 4);
+
+            // back to byte, set up for transpose
+            __m128i brb = _mm_packus_epi16(rw, bw);
+            __m128i gxb = _mm_packus_epi16(gw, xw);
+
+            // transpose to interleave channels
+            __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
+            __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
+            __m128i o0 = _mm_unpacklo_epi16(t0, t1);
+            __m128i o1 = _mm_unpackhi_epi16(t0, t1);
+
+            // store
+            _mm_storeu_si128((__m128i*)(out + 0), o0);
+            _mm_storeu_si128((__m128i*)(out + 16), o1);
+            out += 32;
+        }
+    }
+#            endif
+
+#            ifdef STBI_NEON
+    // in this version, step=3 support would be easy to add. but is there demand?
+    if (step == 4) {
+        // this is a fairly straightforward implementation and not super-optimized.
+        uint8x8_t signflip = vdup_n_u8(0x80);
+        int16x8_t cr_const0 = vdupq_n_s16((short)(1.40200f * 4096.0f + 0.5f));
+        int16x8_t cr_const1 = vdupq_n_s16(-(short)(0.71414f * 4096.0f + 0.5f));
+        int16x8_t cb_const0 = vdupq_n_s16(-(short)(0.34414f * 4096.0f + 0.5f));
+        int16x8_t cb_const1 = vdupq_n_s16((short)(1.77200f * 4096.0f + 0.5f));
+
+        for (; i + 7 < count; i += 8) {
+            // load
+            uint8x8_t y_bytes = vld1_u8(y + i);
+            uint8x8_t cr_bytes = vld1_u8(pcr + i);
+            uint8x8_t cb_bytes = vld1_u8(pcb + i);
+            int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
+            int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
+
+            // expand to s16
+            int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
+            int16x8_t crw = vshll_n_s8(cr_biased, 7);
+            int16x8_t cbw = vshll_n_s8(cb_biased, 7);
+
+            // color transform
+            int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
+            int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
+            int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
+            int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
+            int16x8_t rws = vaddq_s16(yws, cr0);
+            int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
+            int16x8_t bws = vaddq_s16(yws, cb1);
+
+            // undo scaling, round, convert to byte
+            uint8x8x4_t o;
+            o.val[0] = vqrshrun_n_s16(rws, 4);
+            o.val[1] = vqrshrun_n_s16(gws, 4);
+            o.val[2] = vqrshrun_n_s16(bws, 4);
+            o.val[3] = vdup_n_u8(255);
+
+            // store, interleaving r/g/b/a
+            vst4_u8(out, o);
+            out += 8 * 4;
+        }
+    }
+#            endif
+
+    for (; i < count; ++i) {
+        int y_fixed = (y[i] << 20) + (1 << 19); // rounding
+        int r, g, b;
+        int cr = pcr[i] - 128;
+        int cb = pcb[i] - 128;
+        r = y_fixed + cr * stbi__float2fixed(1.40200f);
+        g = y_fixed + cr * -stbi__float2fixed(0.71414f) + ((cb * -stbi__float2fixed(0.34414f)) & 0xffff0000);
+        b = y_fixed + cb * stbi__float2fixed(1.77200f);
+        r >>= 20;
+        g >>= 20;
+        b >>= 20;
+        if ((unsigned)r > 255) {
+            if (r < 0)
+                r = 0;
+            else
+                r = 255;
+        }
+        if ((unsigned)g > 255) {
+            if (g < 0)
+                g = 0;
+            else
+                g = 255;
+        }
+        if ((unsigned)b > 255) {
+            if (b < 0)
+                b = 0;
+            else
+                b = 255;
+        }
+        out[0] = (stbi_uc)r;
+        out[1] = (stbi_uc)g;
+        out[2] = (stbi_uc)b;
+        out[3] = 255;
+        out += step;
+    }
+}
+#        endif
+
+// set up the kernels
+static void stbi__setup_jpeg(stbi__jpeg* j)
+{
+    j->idct_block_kernel = stbi__idct_block;
+    j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
+    j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
+
+#        ifdef STBI_SSE2
+    if (stbi__sse2_available()) {
+        j->idct_block_kernel = stbi__idct_simd;
+        j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+        j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+    }
+#        endif
+
+#        ifdef STBI_NEON
+    j->idct_block_kernel = stbi__idct_simd;
+    j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+    j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+#        endif
+}
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg* j)
+{
+    stbi__free_jpeg_components(j, j->s->img_n, 0);
+}
+
+typedef struct
+{
+    resample_row_func resample;
+    stbi_uc *line0, *line1;
+    int hs, vs;  // expansion factor in each axis
+    int w_lores; // horizontal pixels pre-expansion
+    int ystep;   // how far through vertical expansion we are
+    int ypos;    // which pre-expansion row we're on
+} stbi__resample;
+
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+    unsigned int t = x * y + 128;
+    return (stbi_uc)((t + (t >> 8)) >> 8);
+}
+
+static stbi_uc* load_jpeg_image(stbi__jpeg* z, int* out_x, int* out_y, int* comp, int req_comp)
+{
+    int n, decode_n, is_rgb;
+    z->s->img_n = 0; // make stbi__cleanup_jpeg safe
+
+    // validate req_comp
+    if (req_comp < 0 || req_comp > 4)
+        return stbi__errpuc("bad req_comp", "Internal error");
+
+    // load a jpeg image from whichever source, but leave in YCbCr format
+    if (!stbi__decode_jpeg_image(z)) {
+        stbi__cleanup_jpeg(z);
+        return NULL;
+    }
+
+    // determine actual number of components to generate
+    n = req_comp ? req_comp : z->s->img_n >= 3 ? 3
+                                               : 1;
+
+    is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
+
+    if (z->s->img_n == 3 && n < 3 && !is_rgb)
+        decode_n = 1;
+    else
+        decode_n = z->s->img_n;
+
+    // nothing to do if no components requested; check this now to avoid
+    // accessing uninitialized coutput[0] later
+    if (decode_n <= 0) {
+        stbi__cleanup_jpeg(z);
+        return NULL;
+    }
+
+    // resample and color-convert
+    {
+        int k;
+        unsigned int i, j;
+        stbi_uc* output;
+        stbi_uc* coutput[4] = { NULL, NULL, NULL, NULL };
+
+        stbi__resample res_comp[4];
+
+        for (k = 0; k < decode_n; ++k) {
+            stbi__resample* r = &res_comp[k];
+
+            // allocate line buffer big enough for upsampling off the edges
+            // with upsample factor of 4
+            z->img_comp[k].linebuf = (stbi_uc*)stbi__malloc(z->s->img_x + 3);
+            if (!z->img_comp[k].linebuf) {
+                stbi__cleanup_jpeg(z);
+                return stbi__errpuc("outofmem", "Out of memory");
+            }
+
+            r->hs = z->img_h_max / z->img_comp[k].h;
+            r->vs = z->img_v_max / z->img_comp[k].v;
+            r->ystep = r->vs >> 1;
+            r->w_lores = (z->s->img_x + r->hs - 1) / r->hs;
+            r->ypos = 0;
+            r->line0 = r->line1 = z->img_comp[k].data;
+
+            if (r->hs == 1 && r->vs == 1)
+                r->resample = resample_row_1;
+            else if (r->hs == 1 && r->vs == 2)
+                r->resample = stbi__resample_row_v_2;
+            else if (r->hs == 2 && r->vs == 1)
+                r->resample = stbi__resample_row_h_2;
+            else if (r->hs == 2 && r->vs == 2)
+                r->resample = z->resample_row_hv_2_kernel;
+            else
+                r->resample = stbi__resample_row_generic;
+        }
+
+        // can't error after this so, this is safe
+        output = (stbi_uc*)stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+        if (!output) {
+            stbi__cleanup_jpeg(z);
+            return stbi__errpuc("outofmem", "Out of memory");
+        }
+
+        // now go ahead and resample
+        for (j = 0; j < z->s->img_y; ++j) {
+            stbi_uc* out = output + n * z->s->img_x * j;
+            for (k = 0; k < decode_n; ++k) {
+                stbi__resample* r = &res_comp[k];
+                int y_bot = r->ystep >= (r->vs >> 1);
+                coutput[k] = r->resample(z->img_comp[k].linebuf,
+                    y_bot ? r->line1 : r->line0,
+                    y_bot ? r->line0 : r->line1,
+                    r->w_lores, r->hs);
+                if (++r->ystep >= r->vs) {
+                    r->ystep = 0;
+                    r->line0 = r->line1;
+                    if (++r->ypos < z->img_comp[k].y)
+                        r->line1 += z->img_comp[k].w2;
+                }
+            }
+            if (n >= 3) {
+                stbi_uc* y = coutput[0];
+                if (z->s->img_n == 3) {
+                    if (is_rgb) {
+                        for (i = 0; i < z->s->img_x; ++i) {
+                            out[0] = y[i];
+                            out[1] = coutput[1][i];
+                            out[2] = coutput[2][i];
+                            out[3] = 255;
+                            out += n;
+                        }
+                    } else {
+                        z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+                    }
+                } else if (z->s->img_n == 4) {
+                    if (z->app14_color_transform == 0) { // CMYK
+                        for (i = 0; i < z->s->img_x; ++i) {
+                            stbi_uc m = coutput[3][i];
+                            out[0] = stbi__blinn_8x8(coutput[0][i], m);
+                            out[1] = stbi__blinn_8x8(coutput[1][i], m);
+                            out[2] = stbi__blinn_8x8(coutput[2][i], m);
+                            out[3] = 255;
+                            out += n;
+                        }
+                    } else if (z->app14_color_transform == 2) { // YCCK
+                        z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+                        for (i = 0; i < z->s->img_x; ++i) {
+                            stbi_uc m = coutput[3][i];
+                            out[0] = stbi__blinn_8x8(255 - out[0], m);
+                            out[1] = stbi__blinn_8x8(255 - out[1], m);
+                            out[2] = stbi__blinn_8x8(255 - out[2], m);
+                            out += n;
+                        }
+                    } else { // YCbCr + alpha?  Ignore the fourth channel for now
+                        z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+                    }
+                } else
+                    for (i = 0; i < z->s->img_x; ++i) {
+                        out[0] = out[1] = out[2] = y[i];
+                        out[3] = 255; // not used if n==3
+                        out += n;
+                    }
+            } else {
+                if (is_rgb) {
+                    if (n == 1)
+                        for (i = 0; i < z->s->img_x; ++i)
+                            *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+                    else {
+                        for (i = 0; i < z->s->img_x; ++i, out += 2) {
+                            out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+                            out[1] = 255;
+                        }
+                    }
+                } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
+                    for (i = 0; i < z->s->img_x; ++i) {
+                        stbi_uc m = coutput[3][i];
+                        stbi_uc r = stbi__blinn_8x8(coutput[0][i], m);
+                        stbi_uc g = stbi__blinn_8x8(coutput[1][i], m);
+                        stbi_uc b = stbi__blinn_8x8(coutput[2][i], m);
+                        out[0] = stbi__compute_y(r, g, b);
+                        out[1] = 255;
+                        out += n;
+                    }
+                } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
+                    for (i = 0; i < z->s->img_x; ++i) {
+                        out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+                        out[1] = 255;
+                        out += n;
+                    }
+                } else {
+                    stbi_uc* y = coutput[0];
+                    if (n == 1)
+                        for (i = 0; i < z->s->img_x; ++i)
+                            out[i] = y[i];
+                    else
+                        for (i = 0; i < z->s->img_x; ++i) {
+                            *out++ = y[i];
+                            *out++ = 255;
+                        }
+                }
+            }
+        }
+        stbi__cleanup_jpeg(z);
+        *out_x = z->s->img_x;
+        *out_y = z->s->img_y;
+        if (comp)
+            *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+        return output;
+    }
+}
+
+static void* stbi__jpeg_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    unsigned char* result;
+    stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+    if (!j)
+        return stbi__errpuc("outofmem", "Out of memory");
+    memset(j, 0, sizeof(stbi__jpeg));
+    STBI_NOTUSED(ri);
+    j->s = s;
+    stbi__setup_jpeg(j);
+    result = load_jpeg_image(j, x, y, comp, req_comp);
+    STBI_FREE(j);
+    return result;
+}
+
+static int stbi__jpeg_test(stbi__context* s)
+{
+    int r;
+    stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+    if (!j)
+        return stbi__err("outofmem", "Out of memory");
+    memset(j, 0, sizeof(stbi__jpeg));
+    j->s = s;
+    stbi__setup_jpeg(j);
+    r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+    stbi__rewind(s);
+    STBI_FREE(j);
+    return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg* j, int* x, int* y, int* comp)
+{
+    if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
+        stbi__rewind(j->s);
+        return 0;
+    }
+    if (x)
+        *x = j->s->img_x;
+    if (y)
+        *y = j->s->img_y;
+    if (comp)
+        *comp = j->s->img_n >= 3 ? 3 : 1;
+    return 1;
+}
+
+static int stbi__jpeg_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    int result;
+    stbi__jpeg* j = (stbi__jpeg*)(stbi__malloc(sizeof(stbi__jpeg)));
+    if (!j)
+        return stbi__err("outofmem", "Out of memory");
+    memset(j, 0, sizeof(stbi__jpeg));
+    j->s = s;
+    result = stbi__jpeg_info_raw(j, x, y, comp);
+    STBI_FREE(j);
+    return result;
+}
+#    endif
+
+// public domain zlib decode    v0.2  Sean Barrett 2006-11-18
+//    simple implementation
+//      - all input must be provided in an upfront buffer
+//      - all output is written to a single output buffer (can malloc/realloc)
+//    performance
+//      - fast huffman
+
+#    ifndef STBI_NO_ZLIB
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#        define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
+#        define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
+#        define STBI__ZNSYMS 288 // number of symbols in literal/length alphabet
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+    stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+    stbi__uint16 firstcode[16];
+    int maxcode[17];
+    stbi__uint16 firstsymbol[16];
+    stbi_uc size[STBI__ZNSYMS];
+    stbi__uint16 value[STBI__ZNSYMS];
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+    n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1);
+    n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2);
+    n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4);
+    n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8);
+    return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+    STBI_ASSERT(bits <= 16);
+    // to bit reverse n bits, reverse 16 and shift
+    // e.g. 11 bits, bit reverse and shift away 5
+    return stbi__bitreverse16(v) >> (16 - bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman* z, stbi_uc const* sizelist, int num)
+{
+    int i, k = 0;
+    int code, next_code[16], sizes[17];
+
+    // DEFLATE spec for generating codes
+    memset(sizes, 0, sizeof(sizes));
+    memset(z->fast, 0, sizeof(z->fast));
+    for (i = 0; i < num; ++i)
+        ++sizes[sizelist[i]];
+    sizes[0] = 0;
+    for (i = 1; i < 16; ++i)
+        if (sizes[i] > (1 << i))
+            return stbi__err("bad sizes", "Corrupt PNG");
+    code = 0;
+    for (i = 1; i < 16; ++i) {
+        next_code[i] = code;
+        z->firstcode[i] = (stbi__uint16)code;
+        z->firstsymbol[i] = (stbi__uint16)k;
+        code = (code + sizes[i]);
+        if (sizes[i])
+            if (code - 1 >= (1 << i))
+                return stbi__err("bad codelengths", "Corrupt PNG");
+        z->maxcode[i] = code << (16 - i); // preshift for inner loop
+        code <<= 1;
+        k += sizes[i];
+    }
+    z->maxcode[16] = 0x10000; // sentinel
+    for (i = 0; i < num; ++i) {
+        int s = sizelist[i];
+        if (s) {
+            int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
+            stbi__uint16 fastv = (stbi__uint16)((s << 9) | i);
+            z->size[c] = (stbi_uc)s;
+            z->value[c] = (stbi__uint16)i;
+            if (s <= STBI__ZFAST_BITS) {
+                int j = stbi__bit_reverse(next_code[s], s);
+                while (j < (1 << STBI__ZFAST_BITS)) {
+                    z->fast[j] = fastv;
+                    j += (1 << s);
+                }
+            }
+            ++next_code[s];
+        }
+    }
+    return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+//    because PNG allows splitting the zlib stream arbitrarily,
+//    and it's annoying structurally to have PNG call ZLIB call PNG,
+//    we require PNG read all the IDATs and combine them into a single
+//    memory buffer
+
+typedef struct
+{
+    stbi_uc *zbuffer, *zbuffer_end;
+    int num_bits;
+    int hit_zeof_once;
+    stbi__uint32 code_buffer;
+
+    char* zout;
+    char* zout_start;
+    char* zout_end;
+    int z_expandable;
+
+    stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static int stbi__zeof(stbi__zbuf* z)
+{
+    return (z->zbuffer >= z->zbuffer_end);
+}
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf* z)
+{
+    return stbi__zeof(z) ? 0 : *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf* z)
+{
+    do {
+        if (z->code_buffer >= (1U << z->num_bits)) {
+            z->zbuffer = z->zbuffer_end; /* treat this as EOF so we fail. */
+            return;
+        }
+        z->code_buffer |= (unsigned int)stbi__zget8(z) << z->num_bits;
+        z->num_bits += 8;
+    } while (z->num_bits <= 24);
+}
+
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf* z, int n)
+{
+    unsigned int k;
+    if (z->num_bits < n)
+        stbi__fill_bits(z);
+    k = z->code_buffer & ((1 << n) - 1);
+    z->code_buffer >>= n;
+    z->num_bits -= n;
+    return k;
+}
+
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf* a, stbi__zhuffman* z)
+{
+    int b, s, k;
+    // not resolved by fast table, so compute it the slow way
+    // use jpeg approach, which requires MSbits at top
+    k = stbi__bit_reverse(a->code_buffer, 16);
+    for (s = STBI__ZFAST_BITS + 1;; ++s)
+        if (k < z->maxcode[s])
+            break;
+    if (s >= 16)
+        return -1; // invalid code!
+    // code size is s, so:
+    b = (k >> (16 - s)) - z->firstcode[s] + z->firstsymbol[s];
+    if (b >= STBI__ZNSYMS)
+        return -1; // some data was corrupt somewhere!
+    if (z->size[b] != s)
+        return -1; // was originally an assert, but report failure instead.
+    a->code_buffer >>= s;
+    a->num_bits -= s;
+    return z->value[b];
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf* a, stbi__zhuffman* z)
+{
+    int b, s;
+    if (a->num_bits < 16) {
+        if (stbi__zeof(a)) {
+            if (!a->hit_zeof_once) {
+                // This is the first time we hit eof, insert 16 extra padding btis
+                // to allow us to keep going; if we actually consume any of them
+                // though, that is invalid data. This is caught later.
+                a->hit_zeof_once = 1;
+                a->num_bits += 16; // add 16 implicit zero bits
+            } else {
+                // We already inserted our extra 16 padding bits and are again
+                // out, this stream is actually prematurely terminated.
+                return -1;
+            }
+        } else {
+            stbi__fill_bits(a);
+        }
+    }
+    b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+    if (b) {
+        s = b >> 9;
+        a->code_buffer >>= s;
+        a->num_bits -= s;
+        return b & 511;
+    }
+    return stbi__zhuffman_decode_slowpath(a, z);
+}
+
+static int stbi__zexpand(stbi__zbuf* z, char* zout, int n) // need to make room for n bytes
+{
+    char* q;
+    unsigned int cur, limit, old_limit;
+    z->zout = zout;
+    if (!z->z_expandable)
+        return stbi__err("output buffer limit", "Corrupt PNG");
+    cur = (unsigned int)(z->zout - z->zout_start);
+    limit = old_limit = (unsigned)(z->zout_end - z->zout_start);
+    if (UINT_MAX - cur < (unsigned)n)
+        return stbi__err("outofmem", "Out of memory");
+    while (cur + n > limit) {
+        if (limit > UINT_MAX / 2)
+            return stbi__err("outofmem", "Out of memory");
+        limit *= 2;
+    }
+    q = (char*)STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+    STBI_NOTUSED(old_limit);
+    if (q == NULL)
+        return stbi__err("outofmem", "Out of memory");
+    z->zout_start = q;
+    z->zout = q + cur;
+    z->zout_end = q + limit;
+    return 1;
+}
+
+static int const stbi__zlength_base[31] = {
+    3, 4, 5, 6, 7, 8, 9, 10, 11, 13,
+    15, 17, 19, 23, 27, 31, 35, 43, 51, 59,
+    67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0
+};
+
+static int const stbi__zlength_extra[31] = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0, 0, 0 };
+
+static int const stbi__zdist_base[32] = { 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
+    257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577, 0, 0 };
+
+static int const stbi__zdist_extra[32] = { 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13 };
+
+static int stbi__parse_huffman_block(stbi__zbuf* a)
+{
+    char* zout = a->zout;
+    for (;;) {
+        int z = stbi__zhuffman_decode(a, &a->z_length);
+        if (z < 256) {
+            if (z < 0)
+                return stbi__err("bad huffman code", "Corrupt PNG"); // error in huffman codes
+            if (zout >= a->zout_end) {
+                if (!stbi__zexpand(a, zout, 1))
+                    return 0;
+                zout = a->zout;
+            }
+            *zout++ = (char)z;
+        } else {
+            stbi_uc* p;
+            int len, dist;
+            if (z == 256) {
+                a->zout = zout;
+                if (a->hit_zeof_once && a->num_bits < 16) {
+                    // The first time we hit zeof, we inserted 16 extra zero bits into our bit
+                    // buffer so the decoder can just do its speculative decoding. But if we
+                    // actually consumed any of those bits (which is the case when num_bits < 16),
+                    // the stream actually read past the end so it is malformed.
+                    return stbi__err("unexpected end", "Corrupt PNG");
+                }
+                return 1;
+            }
+            if (z >= 286)
+                return stbi__err("bad huffman code", "Corrupt PNG"); // per DEFLATE, length codes 286 and 287 must not appear in compressed data
+            z -= 257;
+            len = stbi__zlength_base[z];
+            if (stbi__zlength_extra[z])
+                len += stbi__zreceive(a, stbi__zlength_extra[z]);
+            z = stbi__zhuffman_decode(a, &a->z_distance);
+            if (z < 0 || z >= 30)
+                return stbi__err("bad huffman code", "Corrupt PNG"); // per DEFLATE, distance codes 30 and 31 must not appear in compressed data
+            dist = stbi__zdist_base[z];
+            if (stbi__zdist_extra[z])
+                dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+            if (zout - a->zout_start < dist)
+                return stbi__err("bad dist", "Corrupt PNG");
+            if (len > a->zout_end - zout) {
+                if (!stbi__zexpand(a, zout, len))
+                    return 0;
+                zout = a->zout;
+            }
+            p = (stbi_uc*)(zout - dist);
+            if (dist == 1) { // run of one byte; common in images.
+                stbi_uc v = *p;
+                if (len) {
+                    do
+                        *zout++ = v;
+                    while (--len);
+                }
+            } else {
+                if (len) {
+                    do
+                        *zout++ = *p++;
+                    while (--len);
+                }
+            }
+        }
+    }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf* a)
+{
+    static stbi_uc const length_dezigzag[19] = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
+    stbi__zhuffman z_codelength;
+    stbi_uc lencodes[286 + 32 + 137]; // padding for maximum single op
+    stbi_uc codelength_sizes[19];
+    int i, n;
+
+    int hlit = stbi__zreceive(a, 5) + 257;
+    int hdist = stbi__zreceive(a, 5) + 1;
+    int hclen = stbi__zreceive(a, 4) + 4;
+    int ntot = hlit + hdist;
+
+    memset(codelength_sizes, 0, sizeof(codelength_sizes));
+    for (i = 0; i < hclen; ++i) {
+        int s = stbi__zreceive(a, 3);
+        codelength_sizes[length_dezigzag[i]] = (stbi_uc)s;
+    }
+    if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19))
+        return 0;
+
+    n = 0;
+    while (n < ntot) {
+        int c = stbi__zhuffman_decode(a, &z_codelength);
+        if (c < 0 || c >= 19)
+            return stbi__err("bad codelengths", "Corrupt PNG");
+        if (c < 16)
+            lencodes[n++] = (stbi_uc)c;
+        else {
+            stbi_uc fill = 0;
+            if (c == 16) {
+                c = stbi__zreceive(a, 2) + 3;
+                if (n == 0)
+                    return stbi__err("bad codelengths", "Corrupt PNG");
+                fill = lencodes[n - 1];
+            } else if (c == 17) {
+                c = stbi__zreceive(a, 3) + 3;
+            } else if (c == 18) {
+                c = stbi__zreceive(a, 7) + 11;
+            } else {
+                return stbi__err("bad codelengths", "Corrupt PNG");
+            }
+            if (ntot - n < c)
+                return stbi__err("bad codelengths", "Corrupt PNG");
+            memset(lencodes + n, fill, c);
+            n += c;
+        }
+    }
+    if (n != ntot)
+        return stbi__err("bad codelengths", "Corrupt PNG");
+    if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit))
+        return 0;
+    if (!stbi__zbuild_huffman(&a->z_distance, lencodes + hlit, hdist))
+        return 0;
+    return 1;
+}
+
+static int stbi__parse_uncompressed_block(stbi__zbuf* a)
+{
+    stbi_uc header[4];
+    int len, nlen, k;
+    if (a->num_bits & 7)
+        stbi__zreceive(a, a->num_bits & 7); // discard
+    // drain the bit-packed data into header
+    k = 0;
+    while (a->num_bits > 0) {
+        header[k++] = (stbi_uc)(a->code_buffer & 255); // suppress MSVC run-time check
+        a->code_buffer >>= 8;
+        a->num_bits -= 8;
+    }
+    if (a->num_bits < 0)
+        return stbi__err("zlib corrupt", "Corrupt PNG");
+    // now fill header the normal way
+    while (k < 4)
+        header[k++] = stbi__zget8(a);
+    len = header[1] * 256 + header[0];
+    nlen = header[3] * 256 + header[2];
+    if (nlen != (len ^ 0xffff))
+        return stbi__err("zlib corrupt", "Corrupt PNG");
+    if (a->zbuffer + len > a->zbuffer_end)
+        return stbi__err("read past buffer", "Corrupt PNG");
+    if (a->zout + len > a->zout_end)
+        if (!stbi__zexpand(a, a->zout, len))
+            return 0;
+    memcpy(a->zout, a->zbuffer, len);
+    a->zbuffer += len;
+    a->zout += len;
+    return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf* a)
+{
+    int cmf = stbi__zget8(a);
+    int cm = cmf & 15;
+    /* int cinfo = cmf >> 4; */
+    int flg = stbi__zget8(a);
+    if (stbi__zeof(a))
+        return stbi__err("bad zlib header", "Corrupt PNG"); // zlib spec
+    if ((cmf * 256 + flg) % 31 != 0)
+        return stbi__err("bad zlib header", "Corrupt PNG"); // zlib spec
+    if (flg & 32)
+        return stbi__err("no preset dict", "Corrupt PNG"); // preset dictionary not allowed in png
+    if (cm != 8)
+        return stbi__err("bad compression", "Corrupt PNG"); // DEFLATE required for png
+    // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+    return 1;
+}
+
+static stbi_uc const stbi__zdefault_length[STBI__ZNSYMS] = {
+    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+    9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+    9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+    9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+    7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8
+};
+static stbi_uc const stbi__zdefault_distance[32] = {
+    5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
+};
+/*
+Init algorithm:
+{
+   int i;   // use <= to match clearly with spec
+   for (i=0; i <= 143; ++i)     stbi__zdefault_length[i]   = 8;
+   for (   ; i <= 255; ++i)     stbi__zdefault_length[i]   = 9;
+   for (   ; i <= 279; ++i)     stbi__zdefault_length[i]   = 7;
+   for (   ; i <= 287; ++i)     stbi__zdefault_length[i]   = 8;
+
+   for (i=0; i <=  31; ++i)     stbi__zdefault_distance[i] = 5;
+}
+*/
+
+static int stbi__parse_zlib(stbi__zbuf* a, int parse_header)
+{
+    int final, type;
+    if (parse_header)
+        if (!stbi__parse_zlib_header(a))
+            return 0;
+    a->num_bits = 0;
+    a->code_buffer = 0;
+    a->hit_zeof_once = 0;
+    do {
+        final = stbi__zreceive(a, 1);
+        type = stbi__zreceive(a, 2);
+        if (type == 0) {
+            if (!stbi__parse_uncompressed_block(a))
+                return 0;
+        } else if (type == 3) {
+            return 0;
+        } else {
+            if (type == 1) {
+                // use fixed code lengths
+                if (!stbi__zbuild_huffman(&a->z_length, stbi__zdefault_length, STBI__ZNSYMS))
+                    return 0;
+                if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32))
+                    return 0;
+            } else {
+                if (!stbi__compute_huffman_codes(a))
+                    return 0;
+            }
+            if (!stbi__parse_huffman_block(a))
+                return 0;
+        }
+    } while (!final);
+    return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf* a, char* obuf, int olen, int exp, int parse_header)
+{
+    a->zout_start = obuf;
+    a->zout = obuf;
+    a->zout_end = obuf + olen;
+    a->z_expandable = exp;
+
+    return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char* stbi_zlib_decode_malloc_guesssize(char const* buffer, int len, int initial_size, int* outlen)
+{
+    stbi__zbuf a;
+    char* p = (char*)stbi__malloc(initial_size);
+    if (p == NULL)
+        return NULL;
+    a.zbuffer = (stbi_uc*)buffer;
+    a.zbuffer_end = (stbi_uc*)buffer + len;
+    if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
+        if (outlen)
+            *outlen = (int)(a.zout - a.zout_start);
+        return a.zout_start;
+    } else {
+        STBI_FREE(a.zout_start);
+        return NULL;
+    }
+}
+
+STBIDEF char* stbi_zlib_decode_malloc(char const* buffer, int len, int* outlen)
+{
+    return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
+}
+
+STBIDEF char* stbi_zlib_decode_malloc_guesssize_headerflag(char const* buffer, int len, int initial_size, int* outlen, int parse_header)
+{
+    stbi__zbuf a;
+    char* p = (char*)stbi__malloc(initial_size);
+    if (p == NULL)
+        return NULL;
+    a.zbuffer = (stbi_uc*)buffer;
+    a.zbuffer_end = (stbi_uc*)buffer + len;
+    if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
+        if (outlen)
+            *outlen = (int)(a.zout - a.zout_start);
+        return a.zout_start;
+    } else {
+        STBI_FREE(a.zout_start);
+        return NULL;
+    }
+}
+
+STBIDEF int stbi_zlib_decode_buffer(char* obuffer, int olen, char const* ibuffer, int ilen)
+{
+    stbi__zbuf a;
+    a.zbuffer = (stbi_uc*)ibuffer;
+    a.zbuffer_end = (stbi_uc*)ibuffer + ilen;
+    if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
+        return (int)(a.zout - a.zout_start);
+    else
+        return -1;
+}
+
+STBIDEF char* stbi_zlib_decode_noheader_malloc(char const* buffer, int len, int* outlen)
+{
+    stbi__zbuf a;
+    char* p = (char*)stbi__malloc(16384);
+    if (p == NULL)
+        return NULL;
+    a.zbuffer = (stbi_uc*)buffer;
+    a.zbuffer_end = (stbi_uc*)buffer + len;
+    if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
+        if (outlen)
+            *outlen = (int)(a.zout - a.zout_start);
+        return a.zout_start;
+    } else {
+        STBI_FREE(a.zout_start);
+        return NULL;
+    }
+}
+
+STBIDEF int stbi_zlib_decode_noheader_buffer(char* obuffer, int olen, char const* ibuffer, int ilen)
+{
+    stbi__zbuf a;
+    a.zbuffer = (stbi_uc*)ibuffer;
+    a.zbuffer_end = (stbi_uc*)ibuffer + ilen;
+    if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
+        return (int)(a.zout - a.zout_start);
+    else
+        return -1;
+}
+#    endif
+
+// public domain "baseline" PNG decoder   v0.10  Sean Barrett 2006-11-18
+//    simple implementation
+//      - only 8-bit samples
+//      - no CRC checking
+//      - allocates lots of intermediate memory
+//        - avoids problem of streaming data between subsystems
+//        - avoids explicit window management
+//    performance
+//      - uses stb_zlib, a PD zlib implementation with fast huffman decoding
+
+#    ifndef STBI_NO_PNG
+typedef struct
+{
+    stbi__uint32 length;
+    stbi__uint32 type;
+} stbi__pngchunk;
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context* s)
+{
+    stbi__pngchunk c;
+    c.length = stbi__get32be(s);
+    c.type = stbi__get32be(s);
+    return c;
+}
+
+static int stbi__check_png_header(stbi__context* s)
+{
+    static stbi_uc const png_sig[8] = { 137, 80, 78, 71, 13, 10, 26, 10 };
+    int i;
+    for (i = 0; i < 8; ++i)
+        if (stbi__get8(s) != png_sig[i])
+            return stbi__err("bad png sig", "Not a PNG");
+    return 1;
+}
+
+typedef struct
+{
+    stbi__context* s;
+    stbi_uc *idata, *expanded, *out;
+    int depth;
+} stbi__png;
+
+enum {
+    STBI__F_none = 0,
+    STBI__F_sub = 1,
+    STBI__F_up = 2,
+    STBI__F_avg = 3,
+    STBI__F_paeth = 4,
+    // synthetic filter used for first scanline to avoid needing a dummy row of 0s
+    STBI__F_avg_first
+};
+
+static stbi_uc first_row_filter[5] = {
+    STBI__F_none,
+    STBI__F_sub,
+    STBI__F_none,
+    STBI__F_avg_first,
+    STBI__F_sub // Paeth with b=c=0 turns out to be equivalent to sub
+};
+
+static int stbi__paeth(int a, int b, int c)
+{
+    // This formulation looks very different from the reference in the PNG spec, but is
+    // actually equivalent and has favorable data dependencies and admits straightforward
+    // generation of branch-free code, which helps performance significantly.
+    int thresh = c * 3 - (a + b);
+    int lo = a < b ? a : b;
+    int hi = a < b ? b : a;
+    int t0 = (hi <= thresh) ? lo : c;
+    int t1 = (thresh <= lo) ? hi : t0;
+    return t1;
+}
+
+static stbi_uc const stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0, 0, 0, 0x01 };
+
+// adds an extra all-255 alpha channel
+// dest == src is legal
+// img_n must be 1 or 3
+static void stbi__create_png_alpha_expand8(stbi_uc* dest, stbi_uc* src, stbi__uint32 x, int img_n)
+{
+    int i;
+    // must process data backwards since we allow dest==src
+    if (img_n == 1) {
+        for (i = x - 1; i >= 0; --i) {
+            dest[i * 2 + 1] = 255;
+            dest[i * 2 + 0] = src[i];
+        }
+    } else {
+        STBI_ASSERT(img_n == 3);
+        for (i = x - 1; i >= 0; --i) {
+            dest[i * 4 + 3] = 255;
+            dest[i * 4 + 2] = src[i * 3 + 2];
+            dest[i * 4 + 1] = src[i * 3 + 1];
+            dest[i * 4 + 0] = src[i * 3 + 0];
+        }
+    }
+}
+
+// create the png data from post-deflated data
+static int stbi__create_png_image_raw(stbi__png* a, stbi_uc* raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+{
+    int bytes = (depth == 16 ? 2 : 1);
+    stbi__context* s = a->s;
+    stbi__uint32 i, j, stride = x * out_n * bytes;
+    stbi__uint32 img_len, img_width_bytes;
+    stbi_uc* filter_buf;
+    int all_ok = 1;
+    int k;
+    int img_n = s->img_n; // copy it into a local for later
+
+    int output_bytes = out_n * bytes;
+    int filter_bytes = img_n * bytes;
+    int width = x;
+
+    STBI_ASSERT(out_n == s->img_n || out_n == s->img_n + 1);
+    a->out = (stbi_uc*)stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+    if (!a->out)
+        return stbi__err("outofmem", "Out of memory");
+
+    // note: error exits here don't need to clean up a->out individually,
+    // stbi__do_png always does on error.
+    if (!stbi__mad3sizes_valid(img_n, x, depth, 7))
+        return stbi__err("too large", "Corrupt PNG");
+    img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+    if (!stbi__mad2sizes_valid(img_width_bytes, y, img_width_bytes))
+        return stbi__err("too large", "Corrupt PNG");
+    img_len = (img_width_bytes + 1) * y;
+
+    // we used to check for exact match between raw_len and img_len on non-interlaced PNGs,
+    // but issue #276 reported a PNG in the wild that had extra data at the end (all zeros),
+    // so just check for raw_len < img_len always.
+    if (raw_len < img_len)
+        return stbi__err("not enough pixels", "Corrupt PNG");
+
+    // Allocate two scan lines worth of filter workspace buffer.
+    filter_buf = (stbi_uc*)stbi__malloc_mad2(img_width_bytes, 2, 0);
+    if (!filter_buf)
+        return stbi__err("outofmem", "Out of memory");
+
+    // Filtering for low-bit-depth images
+    if (depth < 8) {
+        filter_bytes = 1;
+        width = img_width_bytes;
+    }
+
+    for (j = 0; j < y; ++j) {
+        // cur/prior filter buffers alternate
+        stbi_uc* cur = filter_buf + (j & 1) * img_width_bytes;
+        stbi_uc* prior = filter_buf + (~j & 1) * img_width_bytes;
+        stbi_uc* dest = a->out + stride * j;
+        int nk = width * filter_bytes;
+        int filter = *raw++;
+
+        // check filter type
+        if (filter > 4) {
+            all_ok = stbi__err("invalid filter", "Corrupt PNG");
+            break;
+        }
+
+        // if first row, use special filter that doesn't sample previous row
+        if (j == 0)
+            filter = first_row_filter[filter];
+
+        // perform actual filtering
+        switch (filter) {
+        case STBI__F_none:
+            memcpy(cur, raw, nk);
+            break;
+        case STBI__F_sub:
+            memcpy(cur, raw, filter_bytes);
+            for (k = filter_bytes; k < nk; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + cur[k - filter_bytes]);
+            break;
+        case STBI__F_up:
+            for (k = 0; k < nk; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + prior[k]);
+            break;
+        case STBI__F_avg:
+            for (k = 0; k < filter_bytes; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + (prior[k] >> 1));
+            for (k = filter_bytes; k < nk; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k - filter_bytes]) >> 1));
+            break;
+        case STBI__F_paeth:
+            for (k = 0; k < filter_bytes; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + prior[k]); // prior[k] == stbi__paeth(0,prior[k],0)
+            for (k = filter_bytes; k < nk; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k - filter_bytes], prior[k], prior[k - filter_bytes]));
+            break;
+        case STBI__F_avg_first:
+            memcpy(cur, raw, filter_bytes);
+            for (k = filter_bytes; k < nk; ++k)
+                cur[k] = STBI__BYTECAST(raw[k] + (cur[k - filter_bytes] >> 1));
+            break;
+        }
+
+        raw += nk;
+
+        // expand decoded bits in cur to dest, also adding an extra alpha channel if desired
+        if (depth < 8) {
+            stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+            stbi_uc* in = cur;
+            stbi_uc* out = dest;
+            stbi_uc inb = 0;
+            stbi__uint32 nsmp = x * img_n;
+
+            // expand bits to bytes first
+            if (depth == 4) {
+                for (i = 0; i < nsmp; ++i) {
+                    if ((i & 1) == 0)
+                        inb = *in++;
+                    *out++ = scale * (inb >> 4);
+                    inb <<= 4;
+                }
+            } else if (depth == 2) {
+                for (i = 0; i < nsmp; ++i) {
+                    if ((i & 3) == 0)
+                        inb = *in++;
+                    *out++ = scale * (inb >> 6);
+                    inb <<= 2;
+                }
+            } else {
+                STBI_ASSERT(depth == 1);
+                for (i = 0; i < nsmp; ++i) {
+                    if ((i & 7) == 0)
+                        inb = *in++;
+                    *out++ = scale * (inb >> 7);
+                    inb <<= 1;
+                }
+            }
+
+            // insert alpha=255 values if desired
+            if (img_n != out_n)
+                stbi__create_png_alpha_expand8(dest, dest, x, img_n);
+        } else if (depth == 8) {
+            if (img_n == out_n)
+                memcpy(dest, cur, x * img_n);
+            else
+                stbi__create_png_alpha_expand8(dest, cur, x, img_n);
+        } else if (depth == 16) {
+            // convert the image data from big-endian to platform-native
+            stbi__uint16* dest16 = (stbi__uint16*)dest;
+            stbi__uint32 nsmp = x * img_n;
+
+            if (img_n == out_n) {
+                for (i = 0; i < nsmp; ++i, ++dest16, cur += 2)
+                    *dest16 = (cur[0] << 8) | cur[1];
+            } else {
+                STBI_ASSERT(img_n + 1 == out_n);
+                if (img_n == 1) {
+                    for (i = 0; i < x; ++i, dest16 += 2, cur += 2) {
+                        dest16[0] = (cur[0] << 8) | cur[1];
+                        dest16[1] = 0xffff;
+                    }
+                } else {
+                    STBI_ASSERT(img_n == 3);
+                    for (i = 0; i < x; ++i, dest16 += 4, cur += 6) {
+                        dest16[0] = (cur[0] << 8) | cur[1];
+                        dest16[1] = (cur[2] << 8) | cur[3];
+                        dest16[2] = (cur[4] << 8) | cur[5];
+                        dest16[3] = 0xffff;
+                    }
+                }
+            }
+        }
+    }
+
+    STBI_FREE(filter_buf);
+    if (!all_ok)
+        return 0;
+
+    return 1;
+}
+
+static int stbi__create_png_image(stbi__png* a, stbi_uc* image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+{
+    int bytes = (depth == 16 ? 2 : 1);
+    int out_bytes = out_n * bytes;
+    stbi_uc* final;
+    int p;
+    if (!interlaced)
+        return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
+
+    // de-interlacing
+    final = (stbi_uc*)stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+    if (!final)
+        return stbi__err("outofmem", "Out of memory");
+    for (p = 0; p < 7; ++p) {
+        int xorig[] = { 0, 4, 0, 2, 0, 1, 0 };
+        int yorig[] = { 0, 0, 4, 0, 2, 0, 1 };
+        int xspc[] = { 8, 8, 4, 4, 2, 2, 1 };
+        int yspc[] = { 8, 8, 8, 4, 4, 2, 2 };
+        int i, j, x, y;
+        // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+        x = (a->s->img_x - xorig[p] + xspc[p] - 1) / xspc[p];
+        y = (a->s->img_y - yorig[p] + yspc[p] - 1) / yspc[p];
+        if (x && y) {
+            stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
+            if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+                STBI_FREE(final);
+                return 0;
+            }
+            for (j = 0; j < y; ++j) {
+                for (i = 0; i < x; ++i) {
+                    int out_y = j * yspc[p] + yorig[p];
+                    int out_x = i * xspc[p] + xorig[p];
+                    memcpy(final + out_y * a->s->img_x * out_bytes + out_x * out_bytes,
+                        a->out + (j * x + i) * out_bytes, out_bytes);
+                }
+            }
+            STBI_FREE(a->out);
+            image_data += img_len;
+            image_data_len -= img_len;
+        }
+    }
+    a->out = final;
+
+    return 1;
+}
+
+static int stbi__compute_transparency(stbi__png* z, stbi_uc tc[3], int out_n)
+{
+    stbi__context* s = z->s;
+    stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+    stbi_uc* p = z->out;
+
+    // compute color-based transparency, assuming we've
+    // already got 255 as the alpha value in the output
+    STBI_ASSERT(out_n == 2 || out_n == 4);
+
+    if (out_n == 2) {
+        for (i = 0; i < pixel_count; ++i) {
+            p[1] = (p[0] == tc[0] ? 0 : 255);
+            p += 2;
+        }
+    } else {
+        for (i = 0; i < pixel_count; ++i) {
+            if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+                p[3] = 0;
+            p += 4;
+        }
+    }
+    return 1;
+}
+
+static int stbi__compute_transparency16(stbi__png* z, stbi__uint16 tc[3], int out_n)
+{
+    stbi__context* s = z->s;
+    stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+    stbi__uint16* p = (stbi__uint16*)z->out;
+
+    // compute color-based transparency, assuming we've
+    // already got 65535 as the alpha value in the output
+    STBI_ASSERT(out_n == 2 || out_n == 4);
+
+    if (out_n == 2) {
+        for (i = 0; i < pixel_count; ++i) {
+            p[1] = (p[0] == tc[0] ? 0 : 65535);
+            p += 2;
+        }
+    } else {
+        for (i = 0; i < pixel_count; ++i) {
+            if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+                p[3] = 0;
+            p += 4;
+        }
+    }
+    return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png* a, stbi_uc* palette, int len, int pal_img_n)
+{
+    stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
+    stbi_uc *p, *temp_out, *orig = a->out;
+
+    p = (stbi_uc*)stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+    if (p == NULL)
+        return stbi__err("outofmem", "Out of memory");
+
+    // between here and free(out) below, exitting would leak
+    temp_out = p;
+
+    if (pal_img_n == 3) {
+        for (i = 0; i < pixel_count; ++i) {
+            int n = orig[i] * 4;
+            p[0] = palette[n];
+            p[1] = palette[n + 1];
+            p[2] = palette[n + 2];
+            p += 3;
+        }
+    } else {
+        for (i = 0; i < pixel_count; ++i) {
+            int n = orig[i] * 4;
+            p[0] = palette[n];
+            p[1] = palette[n + 1];
+            p[2] = palette[n + 2];
+            p[3] = palette[n + 3];
+            p += 4;
+        }
+    }
+    STBI_FREE(a->out);
+    a->out = temp_out;
+
+    STBI_NOTUSED(len);
+
+    return 1;
+}
+
+static int stbi__unpremultiply_on_load_global = 0;
+static int stbi__de_iphone_flag_global = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+    stbi__unpremultiply_on_load_global = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+    stbi__de_iphone_flag_global = flag_true_if_should_convert;
+}
+
+#        ifndef STBI_THREAD_LOCAL
+#            define stbi__unpremultiply_on_load stbi__unpremultiply_on_load_global
+#            define stbi__de_iphone_flag stbi__de_iphone_flag_global
+#        else
+static STBI_THREAD_LOCAL int stbi__unpremultiply_on_load_local, stbi__unpremultiply_on_load_set;
+static STBI_THREAD_LOCAL int stbi__de_iphone_flag_local, stbi__de_iphone_flag_set;
+
+STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply)
+{
+    stbi__unpremultiply_on_load_local = flag_true_if_should_unpremultiply;
+    stbi__unpremultiply_on_load_set = 1;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert)
+{
+    stbi__de_iphone_flag_local = flag_true_if_should_convert;
+    stbi__de_iphone_flag_set = 1;
+}
+
+#            define stbi__unpremultiply_on_load (stbi__unpremultiply_on_load_set \
+                    ? stbi__unpremultiply_on_load_local                          \
+                    : stbi__unpremultiply_on_load_global)
+#            define stbi__de_iphone_flag (stbi__de_iphone_flag_set \
+                    ? stbi__de_iphone_flag_local                   \
+                    : stbi__de_iphone_flag_global)
+#        endif // STBI_THREAD_LOCAL
+
+static void stbi__de_iphone(stbi__png* z)
+{
+    stbi__context* s = z->s;
+    stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+    stbi_uc* p = z->out;
+
+    if (s->img_out_n == 3) { // convert bgr to rgb
+        for (i = 0; i < pixel_count; ++i) {
+            stbi_uc t = p[0];
+            p[0] = p[2];
+            p[2] = t;
+            p += 3;
+        }
+    } else {
+        STBI_ASSERT(s->img_out_n == 4);
+        if (stbi__unpremultiply_on_load) {
+            // convert bgr to rgb and unpremultiply
+            for (i = 0; i < pixel_count; ++i) {
+                stbi_uc a = p[3];
+                stbi_uc t = p[0];
+                if (a) {
+                    stbi_uc half = a / 2;
+                    p[0] = (p[2] * 255 + half) / a;
+                    p[1] = (p[1] * 255 + half) / a;
+                    p[2] = (t * 255 + half) / a;
+                } else {
+                    p[0] = p[2];
+                    p[2] = t;
+                }
+                p += 4;
+            }
+        } else {
+            // convert bgr to rgb
+            for (i = 0; i < pixel_count; ++i) {
+                stbi_uc t = p[0];
+                p[0] = p[2];
+                p[2] = t;
+                p += 4;
+            }
+        }
+    }
+}
+
+#        define STBI__PNG_TYPE(a, b, c, d) (((unsigned)(a) << 24) + ((unsigned)(b) << 16) + ((unsigned)(c) << 8) + (unsigned)(d))
+
+static int stbi__parse_png_file(stbi__png* z, int scan, int req_comp)
+{
+    stbi_uc palette[1024], pal_img_n = 0;
+    stbi_uc has_trans = 0, tc[3] = { 0 };
+    stbi__uint16 tc16[3];
+    stbi__uint32 ioff = 0, idata_limit = 0, i, pal_len = 0;
+    int first = 1, k, interlace = 0, color = 0, is_iphone = 0;
+    stbi__context* s = z->s;
+
+    z->expanded = NULL;
+    z->idata = NULL;
+    z->out = NULL;
+
+    if (!stbi__check_png_header(s))
+        return 0;
+
+    if (scan == STBI__SCAN_type)
+        return 1;
+
+    for (;;) {
+        stbi__pngchunk c = stbi__get_chunk_header(s);
+        switch (c.type) {
+        case STBI__PNG_TYPE('C', 'g', 'B', 'I'):
+            is_iphone = 1;
+            stbi__skip(s, c.length);
+            break;
+        case STBI__PNG_TYPE('I', 'H', 'D', 'R'): {
+            int comp, filter;
+            if (!first)
+                return stbi__err("multiple IHDR", "Corrupt PNG");
+            first = 0;
+            if (c.length != 13)
+                return stbi__err("bad IHDR len", "Corrupt PNG");
+            s->img_x = stbi__get32be(s);
+            s->img_y = stbi__get32be(s);
+            if (s->img_y > STBI_MAX_DIMENSIONS)
+                return stbi__err("too large", "Very large image (corrupt?)");
+            if (s->img_x > STBI_MAX_DIMENSIONS)
+                return stbi__err("too large", "Very large image (corrupt?)");
+            z->depth = stbi__get8(s);
+            if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16)
+                return stbi__err("1/2/4/8/16-bit only", "PNG not supported: 1/2/4/8/16-bit only");
+            color = stbi__get8(s);
+            if (color > 6)
+                return stbi__err("bad ctype", "Corrupt PNG");
+            if (color == 3 && z->depth == 16)
+                return stbi__err("bad ctype", "Corrupt PNG");
+            if (color == 3)
+                pal_img_n = 3;
+            else if (color & 1)
+                return stbi__err("bad ctype", "Corrupt PNG");
+            comp = stbi__get8(s);
+            if (comp)
+                return stbi__err("bad comp method", "Corrupt PNG");
+            filter = stbi__get8(s);
+            if (filter)
+                return stbi__err("bad filter method", "Corrupt PNG");
+            interlace = stbi__get8(s);
+            if (interlace > 1)
+                return stbi__err("bad interlace method", "Corrupt PNG");
+            if (!s->img_x || !s->img_y)
+                return stbi__err("0-pixel image", "Corrupt PNG");
+            if (!pal_img_n) {
+                s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
+                if ((1 << 30) / s->img_x / s->img_n < s->img_y)
+                    return stbi__err("too large", "Image too large to decode");
+            } else {
+                // if paletted, then pal_n is our final components, and
+                // img_n is # components to decompress/filter.
+                s->img_n = 1;
+                if ((1 << 30) / s->img_x / 4 < s->img_y)
+                    return stbi__err("too large", "Corrupt PNG");
+            }
+            // even with SCAN_header, have to scan to see if we have a tRNS
+            break;
+        }
+
+        case STBI__PNG_TYPE('P', 'L', 'T', 'E'): {
+            if (first)
+                return stbi__err("first not IHDR", "Corrupt PNG");
+            if (c.length > 256 * 3)
+                return stbi__err("invalid PLTE", "Corrupt PNG");
+            pal_len = c.length / 3;
+            if (pal_len * 3 != c.length)
+                return stbi__err("invalid PLTE", "Corrupt PNG");
+            for (i = 0; i < pal_len; ++i) {
+                palette[i * 4 + 0] = stbi__get8(s);
+                palette[i * 4 + 1] = stbi__get8(s);
+                palette[i * 4 + 2] = stbi__get8(s);
+                palette[i * 4 + 3] = 255;
+            }
+            break;
+        }
+
+        case STBI__PNG_TYPE('t', 'R', 'N', 'S'): {
+            if (first)
+                return stbi__err("first not IHDR", "Corrupt PNG");
+            if (z->idata)
+                return stbi__err("tRNS after IDAT", "Corrupt PNG");
+            if (pal_img_n) {
+                if (scan == STBI__SCAN_header) {
+                    s->img_n = 4;
+                    return 1;
+                }
+                if (pal_len == 0)
+                    return stbi__err("tRNS before PLTE", "Corrupt PNG");
+                if (c.length > pal_len)
+                    return stbi__err("bad tRNS len", "Corrupt PNG");
+                pal_img_n = 4;
+                for (i = 0; i < c.length; ++i)
+                    palette[i * 4 + 3] = stbi__get8(s);
+            } else {
+                if (!(s->img_n & 1))
+                    return stbi__err("tRNS with alpha", "Corrupt PNG");
+                if (c.length != (stbi__uint32)s->img_n * 2)
+                    return stbi__err("bad tRNS len", "Corrupt PNG");
+                has_trans = 1;
+                // non-paletted with tRNS = constant alpha. if header-scanning, we can stop now.
+                if (scan == STBI__SCAN_header) {
+                    ++s->img_n;
+                    return 1;
+                }
+                if (z->depth == 16) {
+                    for (k = 0; k < s->img_n && k < 3; ++k)       // extra loop test to suppress false GCC warning
+                        tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+                } else {
+                    for (k = 0; k < s->img_n && k < 3; ++k)
+                        tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+                }
+            }
+            break;
+        }
+
+        case STBI__PNG_TYPE('I', 'D', 'A', 'T'): {
+            if (first)
+                return stbi__err("first not IHDR", "Corrupt PNG");
+            if (pal_img_n && !pal_len)
+                return stbi__err("no PLTE", "Corrupt PNG");
+            if (scan == STBI__SCAN_header) {
+                // header scan definitely stops at first IDAT
+                if (pal_img_n)
+                    s->img_n = pal_img_n;
+                return 1;
+            }
+            if (c.length > (1u << 30))
+                return stbi__err("IDAT size limit", "IDAT section larger than 2^30 bytes");
+            if ((int)(ioff + c.length) < (int)ioff)
+                return 0;
+            if (ioff + c.length > idata_limit) {
+                stbi__uint32 idata_limit_old = idata_limit;
+                stbi_uc* p;
+                if (idata_limit == 0)
+                    idata_limit = c.length > 4096 ? c.length : 4096;
+                while (ioff + c.length > idata_limit)
+                    idata_limit *= 2;
+                STBI_NOTUSED(idata_limit_old);
+                p = (stbi_uc*)STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit);
+                if (p == NULL)
+                    return stbi__err("outofmem", "Out of memory");
+                z->idata = p;
+            }
+            if (!stbi__getn(s, z->idata + ioff, c.length))
+                return stbi__err("outofdata", "Corrupt PNG");
+            ioff += c.length;
+            break;
+        }
+
+        case STBI__PNG_TYPE('I', 'E', 'N', 'D'): {
+            stbi__uint32 raw_len, bpl;
+            if (first)
+                return stbi__err("first not IHDR", "Corrupt PNG");
+            if (scan != STBI__SCAN_load)
+                return 1;
+            if (z->idata == NULL)
+                return stbi__err("no IDAT", "Corrupt PNG");
+            // initial guess for decoded data size to avoid unnecessary reallocs
+            bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
+            raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
+            z->expanded = (stbi_uc*)stbi_zlib_decode_malloc_guesssize_headerflag((char*)z->idata, ioff, raw_len, (int*)&raw_len, !is_iphone);
+            if (z->expanded == NULL)
+                return 0; // zlib should set error
+            STBI_FREE(z->idata);
+            z->idata = NULL;
+            if ((req_comp == s->img_n + 1 && req_comp != 3 && !pal_img_n) || has_trans)
+                s->img_out_n = s->img_n + 1;
+            else
+                s->img_out_n = s->img_n;
+            if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace))
+                return 0;
+            if (has_trans) {
+                if (z->depth == 16) {
+                    if (!stbi__compute_transparency16(z, tc16, s->img_out_n))
+                        return 0;
+                } else {
+                    if (!stbi__compute_transparency(z, tc, s->img_out_n))
+                        return 0;
+                }
+            }
+            if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
+                stbi__de_iphone(z);
+            if (pal_img_n) {
+                // pal_img_n == 3 or 4
+                s->img_n = pal_img_n; // record the actual colors we had
+                s->img_out_n = pal_img_n;
+                if (req_comp >= 3)
+                    s->img_out_n = req_comp;
+                if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
+                    return 0;
+            } else if (has_trans) {
+                // non-paletted image with tRNS -> source image has (constant) alpha
+                ++s->img_n;
+            }
+            STBI_FREE(z->expanded);
+            z->expanded = NULL;
+            // end of PNG chunk, read and skip CRC
+            stbi__get32be(s);
+            return 1;
+        }
+
+        default:
+            // if critical, fail
+            if (first)
+                return stbi__err("first not IHDR", "Corrupt PNG");
+            if ((c.type & (1 << 29)) == 0) {
+#        ifndef STBI_NO_FAILURE_STRINGS
+                // not threadsafe
+                static char invalid_chunk[] = "XXXX PNG chunk not known";
+                invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+                invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+                invalid_chunk[2] = STBI__BYTECAST(c.type >> 8);
+                invalid_chunk[3] = STBI__BYTECAST(c.type >> 0);
+#        endif
+                return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+            }
+            stbi__skip(s, c.length);
+            break;
+        }
+        // end of PNG chunk, read and skip CRC
+        stbi__get32be(s);
+    }
+}
+
+static void* stbi__do_png(stbi__png* p, int* x, int* y, int* n, int req_comp, stbi__result_info* ri)
+{
+    void* result = NULL;
+    if (req_comp < 0 || req_comp > 4)
+        return stbi__errpuc("bad req_comp", "Internal error");
+    if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+        if (p->depth <= 8)
+            ri->bits_per_channel = 8;
+        else if (p->depth == 16)
+            ri->bits_per_channel = 16;
+        else
+            return stbi__errpuc("bad bits_per_channel", "PNG not supported: unsupported color depth");
+        result = p->out;
+        p->out = NULL;
+        if (req_comp && req_comp != p->s->img_out_n) {
+            if (ri->bits_per_channel == 8)
+                result = stbi__convert_format((unsigned char*)result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+            else
+                result = stbi__convert_format16((stbi__uint16*)result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+            p->s->img_out_n = req_comp;
+            if (result == NULL)
+                return result;
+        }
+        *x = p->s->img_x;
+        *y = p->s->img_y;
+        if (n)
+            *n = p->s->img_n;
+    }
+    STBI_FREE(p->out);
+    p->out = NULL;
+    STBI_FREE(p->expanded);
+    p->expanded = NULL;
+    STBI_FREE(p->idata);
+    p->idata = NULL;
+
+    return result;
+}
+
+static void* stbi__png_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    stbi__png p;
+    p.s = s;
+    return stbi__do_png(&p, x, y, comp, req_comp, ri);
+}
+
+static int stbi__png_test(stbi__context* s)
+{
+    int r;
+    r = stbi__check_png_header(s);
+    stbi__rewind(s);
+    return r;
+}
+
+static int stbi__png_info_raw(stbi__png* p, int* x, int* y, int* comp)
+{
+    if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
+        stbi__rewind(p->s);
+        return 0;
+    }
+    if (x)
+        *x = p->s->img_x;
+    if (y)
+        *y = p->s->img_y;
+    if (comp)
+        *comp = p->s->img_n;
+    return 1;
+}
+
+static int stbi__png_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    stbi__png p;
+    p.s = s;
+    return stbi__png_info_raw(&p, x, y, comp);
+}
+
+static int stbi__png_is16(stbi__context* s)
+{
+    stbi__png p;
+    p.s = s;
+    if (!stbi__png_info_raw(&p, NULL, NULL, NULL))
+        return 0;
+    if (p.depth != 16) {
+        stbi__rewind(p.s);
+        return 0;
+    }
+    return 1;
+}
+#    endif
+
+// Microsoft/Windows BMP image
+
+#    ifndef STBI_NO_BMP
+static int stbi__bmp_test_raw(stbi__context* s)
+{
+    int r;
+    int sz;
+    if (stbi__get8(s) != 'B')
+        return 0;
+    if (stbi__get8(s) != 'M')
+        return 0;
+    stbi__get32le(s); // discard filesize
+    stbi__get16le(s); // discard reserved
+    stbi__get16le(s); // discard reserved
+    stbi__get32le(s); // discard data offset
+    sz = stbi__get32le(s);
+    r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+    return r;
+}
+
+static int stbi__bmp_test(stbi__context* s)
+{
+    int r = stbi__bmp_test_raw(s);
+    stbi__rewind(s);
+    return r;
+}
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+    int n = 0;
+    if (z == 0)
+        return -1;
+    if (z >= 0x10000) {
+        n += 16;
+        z >>= 16;
+    }
+    if (z >= 0x00100) {
+        n += 8;
+        z >>= 8;
+    }
+    if (z >= 0x00010) {
+        n += 4;
+        z >>= 4;
+    }
+    if (z >= 0x00004) {
+        n += 2;
+        z >>= 2;
+    }
+    if (z >= 0x00002) {
+        n += 1; /* >>=  1;*/
+    }
+    return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+    a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2
+    a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4
+    a = (a + (a >> 4)) & 0x0f0f0f0f;                // max 8 per 4, now 8 bits
+    a = (a + (a >> 8));                             // max 16 per 8 bits
+    a = (a + (a >> 16));                            // max 32 per 8 bits
+    return a & 0xff;
+}
+
+// extract an arbitrarily-aligned N-bit value (N=bits)
+// from v, and then make it 8-bits long and fractionally
+// extend it to full full range.
+static int stbi__shiftsigned(unsigned int v, int shift, int bits)
+{
+    static unsigned int mul_table[9] = {
+        0,
+        0xff /*0b11111111*/,
+        0x55 /*0b01010101*/,
+        0x49 /*0b01001001*/,
+        0x11 /*0b00010001*/,
+        0x21 /*0b00100001*/,
+        0x41 /*0b01000001*/,
+        0x81 /*0b10000001*/,
+        0x01 /*0b00000001*/,
+    };
+    static unsigned int shift_table[9] = {
+        0,
+        0,
+        0,
+        1,
+        0,
+        2,
+        4,
+        6,
+        0,
+    };
+    if (shift < 0)
+        v <<= -shift;
+    else
+        v >>= shift;
+    STBI_ASSERT(v < 256);
+    v >>= (8 - bits);
+    STBI_ASSERT(bits >= 0 && bits <= 8);
+    return (int)((unsigned)v * mul_table[bits]) >> shift_table[bits];
+}
+
+typedef struct
+{
+    int bpp, offset, hsz;
+    unsigned int mr, mg, mb, ma, all_a;
+    int extra_read;
+} stbi__bmp_data;
+
+static int stbi__bmp_set_mask_defaults(stbi__bmp_data* info, int compress)
+{
+    // BI_BITFIELDS specifies masks explicitly, don't override
+    if (compress == 3)
+        return 1;
+
+    if (compress == 0) {
+        if (info->bpp == 16) {
+            info->mr = 31u << 10;
+            info->mg = 31u << 5;
+            info->mb = 31u << 0;
+        } else if (info->bpp == 32) {
+            info->mr = 0xffu << 16;
+            info->mg = 0xffu << 8;
+            info->mb = 0xffu << 0;
+            info->ma = 0xffu << 24;
+            info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+        } else {
+            // otherwise, use defaults, which is all-0
+            info->mr = info->mg = info->mb = info->ma = 0;
+        }
+        return 1;
+    }
+    return 0; // error
+}
+
+static void* stbi__bmp_parse_header(stbi__context* s, stbi__bmp_data* info)
+{
+    int hsz;
+    if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M')
+        return stbi__errpuc("not BMP", "Corrupt BMP");
+    stbi__get32le(s); // discard filesize
+    stbi__get16le(s); // discard reserved
+    stbi__get16le(s); // discard reserved
+    info->offset = stbi__get32le(s);
+    info->hsz = hsz = stbi__get32le(s);
+    info->mr = info->mg = info->mb = info->ma = 0;
+    info->extra_read = 14;
+
+    if (info->offset < 0)
+        return stbi__errpuc("bad BMP", "bad BMP");
+
+    if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124)
+        return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+    if (hsz == 12) {
+        s->img_x = stbi__get16le(s);
+        s->img_y = stbi__get16le(s);
+    } else {
+        s->img_x = stbi__get32le(s);
+        s->img_y = stbi__get32le(s);
+    }
+    if (stbi__get16le(s) != 1)
+        return stbi__errpuc("bad BMP", "bad BMP");
+    info->bpp = stbi__get16le(s);
+    if (hsz != 12) {
+        int compress = stbi__get32le(s);
+        if (compress == 1 || compress == 2)
+            return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+        if (compress >= 4)
+            return stbi__errpuc("BMP JPEG/PNG", "BMP type not supported: unsupported compression"); // this includes PNG/JPEG modes
+        if (compress == 3 && info->bpp != 16 && info->bpp != 32)
+            return stbi__errpuc("bad BMP", "bad BMP"); // bitfields requires 16 or 32 bits/pixel
+        stbi__get32le(s);                              // discard sizeof
+        stbi__get32le(s);                              // discard hres
+        stbi__get32le(s);                              // discard vres
+        stbi__get32le(s);                              // discard colorsused
+        stbi__get32le(s);                              // discard max important
+        if (hsz == 40 || hsz == 56) {
+            if (hsz == 56) {
+                stbi__get32le(s);
+                stbi__get32le(s);
+                stbi__get32le(s);
+                stbi__get32le(s);
+            }
+            if (info->bpp == 16 || info->bpp == 32) {
+                if (compress == 0) {
+                    stbi__bmp_set_mask_defaults(info, compress);
+                } else if (compress == 3) {
+                    info->mr = stbi__get32le(s);
+                    info->mg = stbi__get32le(s);
+                    info->mb = stbi__get32le(s);
+                    info->extra_read += 12;
+                    // not documented, but generated by photoshop and handled by mspaint
+                    if (info->mr == info->mg && info->mg == info->mb) {
+                        // ?!?!?
+                        return stbi__errpuc("bad BMP", "bad BMP");
+                    }
+                } else
+                    return stbi__errpuc("bad BMP", "bad BMP");
+            }
+        } else {
+            // V4/V5 header
+            int i;
+            if (hsz != 108 && hsz != 124)
+                return stbi__errpuc("bad BMP", "bad BMP");
+            info->mr = stbi__get32le(s);
+            info->mg = stbi__get32le(s);
+            info->mb = stbi__get32le(s);
+            info->ma = stbi__get32le(s);
+            if (compress != 3) // override mr/mg/mb unless in BI_BITFIELDS mode, as per docs
+                stbi__bmp_set_mask_defaults(info, compress);
+            stbi__get32le(s); // discard color space
+            for (i = 0; i < 12; ++i)
+                stbi__get32le(s); // discard color space parameters
+            if (hsz == 124) {
+                stbi__get32le(s); // discard rendering intent
+                stbi__get32le(s); // discard offset of profile data
+                stbi__get32le(s); // discard size of profile data
+                stbi__get32le(s); // discard reserved
+            }
+        }
+    }
+    return (void*)1;
+}
+
+static void* stbi__bmp_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    stbi_uc* out;
+    unsigned int mr = 0, mg = 0, mb = 0, ma = 0, all_a;
+    stbi_uc pal[256][4];
+    int psize = 0, i, j, width;
+    int flip_vertically, pad, target;
+    stbi__bmp_data info;
+    STBI_NOTUSED(ri);
+
+    info.all_a = 255;
+    if (stbi__bmp_parse_header(s, &info) == NULL)
+        return NULL; // error code already set
+
+    flip_vertically = ((int)s->img_y) > 0;
+    s->img_y = abs((int)s->img_y);
+
+    if (s->img_y > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+    if (s->img_x > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+
+    mr = info.mr;
+    mg = info.mg;
+    mb = info.mb;
+    ma = info.ma;
+    all_a = info.all_a;
+
+    if (info.hsz == 12) {
+        if (info.bpp < 24)
+            psize = (info.offset - info.extra_read - 24) / 3;
+    } else {
+        if (info.bpp < 16)
+            psize = (info.offset - info.extra_read - info.hsz) >> 2;
+    }
+    if (psize == 0) {
+        // accept some number of extra bytes after the header, but if the offset points either to before
+        // the header ends or implies a large amount of extra data, reject the file as malformed
+        int bytes_read_so_far = s->callback_already_read + (int)(s->img_buffer - s->img_buffer_original);
+        int header_limit = 1024;        // max we actually read is below 256 bytes currently.
+        int extra_data_limit = 256 * 4; // what ordinarily goes here is a palette; 256 entries*4 bytes is its max size.
+        if (bytes_read_so_far <= 0 || bytes_read_so_far > header_limit) {
+            return stbi__errpuc("bad header", "Corrupt BMP");
+        }
+        // we established that bytes_read_so_far is positive and sensible.
+        // the first half of this test rejects offsets that are either too small positives, or
+        // negative, and guarantees that info.offset >= bytes_read_so_far > 0. this in turn
+        // ensures the number computed in the second half of the test can't overflow.
+        if (info.offset < bytes_read_so_far || info.offset - bytes_read_so_far > extra_data_limit) {
+            return stbi__errpuc("bad offset", "Corrupt BMP");
+        } else {
+            stbi__skip(s, info.offset - bytes_read_so_far);
+        }
+    }
+
+    if (info.bpp == 24 && ma == 0xff000000)
+        s->img_n = 3;
+    else
+        s->img_n = ma ? 4 : 3;
+    if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
+        target = req_comp;
+    else
+        target = s->img_n; // if they want monochrome, we'll post-convert
+
+    // sanity-check size
+    if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+        return stbi__errpuc("too large", "Corrupt BMP");
+
+    out = (stbi_uc*)stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+    if (!out)
+        return stbi__errpuc("outofmem", "Out of memory");
+    if (info.bpp < 16) {
+        int z = 0;
+        if (psize == 0 || psize > 256) {
+            STBI_FREE(out);
+            return stbi__errpuc("invalid", "Corrupt BMP");
+        }
+        for (i = 0; i < psize; ++i) {
+            pal[i][2] = stbi__get8(s);
+            pal[i][1] = stbi__get8(s);
+            pal[i][0] = stbi__get8(s);
+            if (info.hsz != 12)
+                stbi__get8(s);
+            pal[i][3] = 255;
+        }
+        stbi__skip(s, info.offset - info.extra_read - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+        if (info.bpp == 1)
+            width = (s->img_x + 7) >> 3;
+        else if (info.bpp == 4)
+            width = (s->img_x + 1) >> 1;
+        else if (info.bpp == 8)
+            width = s->img_x;
+        else {
+            STBI_FREE(out);
+            return stbi__errpuc("bad bpp", "Corrupt BMP");
+        }
+        pad = (-width) & 3;
+        if (info.bpp == 1) {
+            for (j = 0; j < (int)s->img_y; ++j) {
+                int bit_offset = 7, v = stbi__get8(s);
+                for (i = 0; i < (int)s->img_x; ++i) {
+                    int color = (v >> bit_offset) & 0x1;
+                    out[z++] = pal[color][0];
+                    out[z++] = pal[color][1];
+                    out[z++] = pal[color][2];
+                    if (target == 4)
+                        out[z++] = 255;
+                    if (i + 1 == (int)s->img_x)
+                        break;
+                    if ((--bit_offset) < 0) {
+                        bit_offset = 7;
+                        v = stbi__get8(s);
+                    }
+                }
+                stbi__skip(s, pad);
+            }
+        } else {
+            for (j = 0; j < (int)s->img_y; ++j) {
+                for (i = 0; i < (int)s->img_x; i += 2) {
+                    int v = stbi__get8(s), v2 = 0;
+                    if (info.bpp == 4) {
+                        v2 = v & 15;
+                        v >>= 4;
+                    }
+                    out[z++] = pal[v][0];
+                    out[z++] = pal[v][1];
+                    out[z++] = pal[v][2];
+                    if (target == 4)
+                        out[z++] = 255;
+                    if (i + 1 == (int)s->img_x)
+                        break;
+                    v = (info.bpp == 8) ? stbi__get8(s) : v2;
+                    out[z++] = pal[v][0];
+                    out[z++] = pal[v][1];
+                    out[z++] = pal[v][2];
+                    if (target == 4)
+                        out[z++] = 255;
+                }
+                stbi__skip(s, pad);
+            }
+        }
+    } else {
+        int rshift = 0, gshift = 0, bshift = 0, ashift = 0, rcount = 0, gcount = 0, bcount = 0, acount = 0;
+        int z = 0;
+        int easy = 0;
+        stbi__skip(s, info.offset - info.extra_read - info.hsz);
+        if (info.bpp == 24)
+            width = 3 * s->img_x;
+        else if (info.bpp == 16)
+            width = 2 * s->img_x;
+        else /* bpp = 32 and pad = 0 */
+            width = 0;
+        pad = (-width) & 3;
+        if (info.bpp == 24) {
+            easy = 1;
+        } else if (info.bpp == 32) {
+            if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+                easy = 2;
+        }
+        if (!easy) {
+            if (!mr || !mg || !mb) {
+                STBI_FREE(out);
+                return stbi__errpuc("bad masks", "Corrupt BMP");
+            }
+            // right shift amt to put high bit in position #7
+            rshift = stbi__high_bit(mr) - 7;
+            rcount = stbi__bitcount(mr);
+            gshift = stbi__high_bit(mg) - 7;
+            gcount = stbi__bitcount(mg);
+            bshift = stbi__high_bit(mb) - 7;
+            bcount = stbi__bitcount(mb);
+            ashift = stbi__high_bit(ma) - 7;
+            acount = stbi__bitcount(ma);
+            if (rcount > 8 || gcount > 8 || bcount > 8 || acount > 8) {
+                STBI_FREE(out);
+                return stbi__errpuc("bad masks", "Corrupt BMP");
+            }
+        }
+        for (j = 0; j < (int)s->img_y; ++j) {
+            if (easy) {
+                for (i = 0; i < (int)s->img_x; ++i) {
+                    unsigned char a;
+                    out[z + 2] = stbi__get8(s);
+                    out[z + 1] = stbi__get8(s);
+                    out[z + 0] = stbi__get8(s);
+                    z += 3;
+                    a = (easy == 2 ? stbi__get8(s) : 255);
+                    all_a |= a;
+                    if (target == 4)
+                        out[z++] = a;
+                }
+            } else {
+                int bpp = info.bpp;
+                for (i = 0; i < (int)s->img_x; ++i) {
+                    stbi__uint32 v = (bpp == 16 ? (stbi__uint32)stbi__get16le(s) : stbi__get32le(s));
+                    unsigned int a;
+                    out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+                    out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+                    out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+                    a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+                    all_a |= a;
+                    if (target == 4)
+                        out[z++] = STBI__BYTECAST(a);
+                }
+            }
+            stbi__skip(s, pad);
+        }
+    }
+
+    // if alpha channel is all 0s, replace with all 255s
+    if (target == 4 && all_a == 0)
+        for (i = 4 * s->img_x * s->img_y - 1; i >= 0; i -= 4)
+            out[i] = 255;
+
+    if (flip_vertically) {
+        stbi_uc t;
+        for (j = 0; j < (int)s->img_y >> 1; ++j) {
+            stbi_uc* p1 = out + j * s->img_x * target;
+            stbi_uc* p2 = out + (s->img_y - 1 - j) * s->img_x * target;
+            for (i = 0; i < (int)s->img_x * target; ++i) {
+                t = p1[i];
+                p1[i] = p2[i];
+                p2[i] = t;
+            }
+        }
+    }
+
+    if (req_comp && req_comp != target) {
+        out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
+        if (out == NULL)
+            return out; // stbi__convert_format frees input on failure
+    }
+
+    *x = s->img_x;
+    *y = s->img_y;
+    if (comp)
+        *comp = s->img_n;
+    return out;
+}
+#    endif
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+#    ifndef STBI_NO_TGA
+// returns STBI_rgb or whatever, 0 on error
+static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
+{
+    // only RGB or RGBA (incl. 16bit) or grey allowed
+    if (is_rgb16)
+        *is_rgb16 = 0;
+    switch (bits_per_pixel) {
+    case 8:
+        return STBI_grey;
+    case 16:
+        if (is_grey)
+            return STBI_grey_alpha;
+        // fallthrough
+    case 15:
+        if (is_rgb16)
+            *is_rgb16 = 1;
+        return STBI_rgb;
+    case 24: // fallthrough
+    case 32:
+        return bits_per_pixel / 8;
+    default:
+        return 0;
+    }
+}
+
+static int stbi__tga_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+    int sz, tga_colormap_type;
+    stbi__get8(s);                     // discard Offset
+    tga_colormap_type = stbi__get8(s); // colormap type
+    if (tga_colormap_type > 1) {
+        stbi__rewind(s);
+        return 0; // only RGB or indexed allowed
+    }
+    tga_image_type = stbi__get8(s); // image type
+    if (tga_colormap_type == 1) {   // colormapped (paletted) image
+        if (tga_image_type != 1 && tga_image_type != 9) {
+            stbi__rewind(s);
+            return 0;
+        }
+        stbi__skip(s, 4);   // skip index of first colormap entry and number of entries
+        sz = stbi__get8(s); //   check bits per palette color entry
+        if ((sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32)) {
+            stbi__rewind(s);
+            return 0;
+        }
+        stbi__skip(s, 4); // skip image x and y origin
+        tga_colormap_bpp = sz;
+    } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+        if ((tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11)) {
+            stbi__rewind(s);
+            return 0; // only RGB or grey allowed, +/- RLE
+        }
+        stbi__skip(s, 9); // skip colormap specification and image x/y origin
+        tga_colormap_bpp = 0;
+    }
+    tga_w = stbi__get16le(s);
+    if (tga_w < 1) {
+        stbi__rewind(s);
+        return 0; // test width
+    }
+    tga_h = stbi__get16le(s);
+    if (tga_h < 1) {
+        stbi__rewind(s);
+        return 0; // test height
+    }
+    tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+    stbi__get8(s);                      // ignore alpha bits
+    if (tga_colormap_bpp != 0) {
+        if ((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
+            // when using a colormap, tga_bits_per_pixel is the size of the indexes
+            // I don't think anything but 8 or 16bit indexes makes sense
+            stbi__rewind(s);
+            return 0;
+        }
+        tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+    } else {
+        tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+    }
+    if (!tga_comp) {
+        stbi__rewind(s);
+        return 0;
+    }
+    if (x)
+        *x = tga_w;
+    if (y)
+        *y = tga_h;
+    if (comp)
+        *comp = tga_comp;
+    return 1; // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context* s)
+{
+    int res = 0;
+    int sz, tga_color_type;
+    stbi__get8(s);                  //   discard Offset
+    tga_color_type = stbi__get8(s); //   color type
+    if (tga_color_type > 1)
+        goto errorEnd;         //   only RGB or indexed allowed
+    sz = stbi__get8(s);        //   image type
+    if (tga_color_type == 1) { // colormapped (paletted) image
+        if (sz != 1 && sz != 9)
+            goto errorEnd;  // colortype 1 demands image type 1 or 9
+        stbi__skip(s, 4);   // skip index of first colormap entry and number of entries
+        sz = stbi__get8(s); //   check bits per palette color entry
+        if ((sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32))
+            goto errorEnd;
+        stbi__skip(s, 4); // skip image x and y origin
+    } else {              // "normal" image w/o colormap
+        if ((sz != 2) && (sz != 3) && (sz != 10) && (sz != 11))
+            goto errorEnd; // only RGB or grey allowed, +/- RLE
+        stbi__skip(s, 9);  // skip colormap specification and image x/y origin
+    }
+    if (stbi__get16le(s) < 1)
+        goto errorEnd; //   test width
+    if (stbi__get16le(s) < 1)
+        goto errorEnd;  //   test height
+    sz = stbi__get8(s); //   bits per pixel
+    if ((tga_color_type == 1) && (sz != 8) && (sz != 16))
+        goto errorEnd; // for colormapped images, bpp is size of an index
+    if ((sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32))
+        goto errorEnd;
+
+    res = 1; // if we got this far, everything's good and we can return 1 instead of 0
+
+errorEnd:
+    stbi__rewind(s);
+    return res;
+}
+
+// read 16bit value and convert to 24bit RGB
+static void stbi__tga_read_rgb16(stbi__context* s, stbi_uc* out)
+{
+    stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
+    stbi__uint16 fiveBitMask = 31;
+    // we have 3 channels with 5bits each
+    int r = (px >> 10) & fiveBitMask;
+    int g = (px >> 5) & fiveBitMask;
+    int b = px & fiveBitMask;
+    // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
+    out[0] = (stbi_uc)((r * 255) / 31);
+    out[1] = (stbi_uc)((g * 255) / 31);
+    out[2] = (stbi_uc)((b * 255) / 31);
+
+    // some people claim that the most significant bit might be used for alpha
+    // (possibly if an alpha-bit is set in the "image descriptor byte")
+    // but that only made 16bit test images completely translucent..
+    // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
+}
+
+static void* stbi__tga_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    //   read in the TGA header stuff
+    int tga_offset = stbi__get8(s);
+    int tga_indexed = stbi__get8(s);
+    int tga_image_type = stbi__get8(s);
+    int tga_is_RLE = 0;
+    int tga_palette_start = stbi__get16le(s);
+    int tga_palette_len = stbi__get16le(s);
+    int tga_palette_bits = stbi__get8(s);
+    int tga_x_origin = stbi__get16le(s);
+    int tga_y_origin = stbi__get16le(s);
+    int tga_width = stbi__get16le(s);
+    int tga_height = stbi__get16le(s);
+    int tga_bits_per_pixel = stbi__get8(s);
+    int tga_comp, tga_rgb16 = 0;
+    int tga_inverted = stbi__get8(s);
+    // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
+    //   image data
+    unsigned char* tga_data;
+    unsigned char* tga_palette = NULL;
+    int i, j;
+    unsigned char raw_data[4] = { 0 };
+    int RLE_count = 0;
+    int RLE_repeating = 0;
+    int read_next_pixel = 1;
+    STBI_NOTUSED(ri);
+    STBI_NOTUSED(tga_x_origin); // @TODO
+    STBI_NOTUSED(tga_y_origin); // @TODO
+
+    if (tga_height > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+    if (tga_width > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+
+    //   do a tiny bit of precessing
+    if (tga_image_type >= 8) {
+        tga_image_type -= 8;
+        tga_is_RLE = 1;
+    }
+    tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+    //   If I'm paletted, then I'll use the number of bits from the palette
+    if (tga_indexed)
+        tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+    else
+        tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+
+    if (!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+        return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
+
+    //   tga info
+    *x = tga_width;
+    *y = tga_height;
+    if (comp)
+        *comp = tga_comp;
+
+    if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+        return stbi__errpuc("too large", "Corrupt TGA");
+
+    tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+    if (!tga_data)
+        return stbi__errpuc("outofmem", "Out of memory");
+
+    // skip to the data's starting position (offset usually = 0)
+    stbi__skip(s, tga_offset);
+
+    if (!tga_indexed && !tga_is_RLE && !tga_rgb16) {
+        for (i = 0; i < tga_height; ++i) {
+            int row = tga_inverted ? tga_height - i - 1 : i;
+            stbi_uc* tga_row = tga_data + row * tga_width * tga_comp;
+            stbi__getn(s, tga_row, tga_width * tga_comp);
+        }
+    } else {
+        //   do I need to load a palette?
+        if (tga_indexed) {
+            if (tga_palette_len == 0) { /* you have to have at least one entry! */
+                STBI_FREE(tga_data);
+                return stbi__errpuc("bad palette", "Corrupt TGA");
+            }
+
+            //   any data to skip? (offset usually = 0)
+            stbi__skip(s, tga_palette_start);
+            //   load the palette
+            tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+            if (!tga_palette) {
+                STBI_FREE(tga_data);
+                return stbi__errpuc("outofmem", "Out of memory");
+            }
+            if (tga_rgb16) {
+                stbi_uc* pal_entry = tga_palette;
+                STBI_ASSERT(tga_comp == STBI_rgb);
+                for (i = 0; i < tga_palette_len; ++i) {
+                    stbi__tga_read_rgb16(s, pal_entry);
+                    pal_entry += tga_comp;
+                }
+            } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
+                STBI_FREE(tga_data);
+                STBI_FREE(tga_palette);
+                return stbi__errpuc("bad palette", "Corrupt TGA");
+            }
+        }
+        //   load the data
+        for (i = 0; i < tga_width * tga_height; ++i) {
+            //   if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
+            if (tga_is_RLE) {
+                if (RLE_count == 0) {
+                    //   yep, get the next byte as a RLE command
+                    int RLE_cmd = stbi__get8(s);
+                    RLE_count = 1 + (RLE_cmd & 127);
+                    RLE_repeating = RLE_cmd >> 7;
+                    read_next_pixel = 1;
+                } else if (!RLE_repeating) {
+                    read_next_pixel = 1;
+                }
+            } else {
+                read_next_pixel = 1;
+            }
+            //   OK, if I need to read a pixel, do it now
+            if (read_next_pixel) {
+                //   load however much data we did have
+                if (tga_indexed) {
+                    // read in index, then perform the lookup
+                    int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
+                    if (pal_idx >= tga_palette_len) {
+                        // invalid index
+                        pal_idx = 0;
+                    }
+                    pal_idx *= tga_comp;
+                    for (j = 0; j < tga_comp; ++j) {
+                        raw_data[j] = tga_palette[pal_idx + j];
+                    }
+                } else if (tga_rgb16) {
+                    STBI_ASSERT(tga_comp == STBI_rgb);
+                    stbi__tga_read_rgb16(s, raw_data);
+                } else {
+                    //   read in the data raw
+                    for (j = 0; j < tga_comp; ++j) {
+                        raw_data[j] = stbi__get8(s);
+                    }
+                }
+                //   clear the reading flag for the next pixel
+                read_next_pixel = 0;
+            } // end of reading a pixel
+
+            // copy data
+            for (j = 0; j < tga_comp; ++j)
+                tga_data[i * tga_comp + j] = raw_data[j];
+
+            //   in case we're in RLE mode, keep counting down
+            --RLE_count;
+        }
+        //   do I need to invert the image?
+        if (tga_inverted) {
+            for (j = 0; j * 2 < tga_height; ++j) {
+                int index1 = j * tga_width * tga_comp;
+                int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
+                for (i = tga_width * tga_comp; i > 0; --i) {
+                    unsigned char temp = tga_data[index1];
+                    tga_data[index1] = tga_data[index2];
+                    tga_data[index2] = temp;
+                    ++index1;
+                    ++index2;
+                }
+            }
+        }
+        //   clear my palette, if I had one
+        if (tga_palette != NULL) {
+            STBI_FREE(tga_palette);
+        }
+    }
+
+    // swap RGB - if the source data was RGB16, it already is in the right order
+    if (tga_comp >= 3 && !tga_rgb16) {
+        unsigned char* tga_pixel = tga_data;
+        for (i = 0; i < tga_width * tga_height; ++i) {
+            unsigned char temp = tga_pixel[0];
+            tga_pixel[0] = tga_pixel[2];
+            tga_pixel[2] = temp;
+            tga_pixel += tga_comp;
+        }
+    }
+
+    // convert to target component count
+    if (req_comp && req_comp != tga_comp)
+        tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+    //   the things I do to get rid of an error message, and yet keep
+    //   Microsoft's C compilers happy... [8^(
+    tga_palette_start = tga_palette_len = tga_palette_bits = tga_x_origin = tga_y_origin = 0;
+    STBI_NOTUSED(tga_palette_start);
+    //   OK, done
+    return tga_data;
+}
+#    endif
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+#    ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context* s)
+{
+    int r = (stbi__get32be(s) == 0x38425053);
+    stbi__rewind(s);
+    return r;
+}
+
+static int stbi__psd_decode_rle(stbi__context* s, stbi_uc* p, int pixelCount)
+{
+    int count, nleft, len;
+
+    count = 0;
+    while ((nleft = pixelCount - count) > 0) {
+        len = stbi__get8(s);
+        if (len == 128) {
+            // No-op.
+        } else if (len < 128) {
+            // Copy next len+1 bytes literally.
+            len++;
+            if (len > nleft)
+                return 0; // corrupt data
+            count += len;
+            while (len) {
+                *p = stbi__get8(s);
+                p += 4;
+                len--;
+            }
+        } else if (len > 128) {
+            stbi_uc val;
+            // Next -len+1 bytes in the dest are replicated from next source byte.
+            // (Interpret len as a negative 8-bit int.)
+            len = 257 - len;
+            if (len > nleft)
+                return 0; // corrupt data
+            val = stbi__get8(s);
+            count += len;
+            while (len) {
+                *p = val;
+                p += 4;
+                len--;
+            }
+        }
+    }
+
+    return 1;
+}
+
+static void* stbi__psd_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri, int bpc)
+{
+    int pixelCount;
+    int channelCount, compression;
+    int channel, i;
+    int bitdepth;
+    int w, h;
+    stbi_uc* out;
+    STBI_NOTUSED(ri);
+
+    // Check identifier
+    if (stbi__get32be(s) != 0x38425053) // "8BPS"
+        return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+    // Check file type version.
+    if (stbi__get16be(s) != 1)
+        return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+    // Skip 6 reserved bytes.
+    stbi__skip(s, 6);
+
+    // Read the number of channels (R, G, B, A, etc).
+    channelCount = stbi__get16be(s);
+    if (channelCount < 0 || channelCount > 16)
+        return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+    // Read the rows and columns of the image.
+    h = stbi__get32be(s);
+    w = stbi__get32be(s);
+
+    if (h > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+    if (w > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+
+    // Make sure the depth is 8 bits.
+    bitdepth = stbi__get16be(s);
+    if (bitdepth != 8 && bitdepth != 16)
+        return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
+
+    // Make sure the color mode is RGB.
+    // Valid options are:
+    //   0: Bitmap
+    //   1: Grayscale
+    //   2: Indexed color
+    //   3: RGB color
+    //   4: CMYK color
+    //   7: Multichannel
+    //   8: Duotone
+    //   9: Lab color
+    if (stbi__get16be(s) != 3)
+        return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+    // Skip the Mode Data.  (It's the palette for indexed color; other info for other modes.)
+    stbi__skip(s, stbi__get32be(s));
+
+    // Skip the image resources.  (resolution, pen tool paths, etc)
+    stbi__skip(s, stbi__get32be(s));
+
+    // Skip the reserved data.
+    stbi__skip(s, stbi__get32be(s));
+
+    // Find out if the data is compressed.
+    // Known values:
+    //   0: no compression
+    //   1: RLE compressed
+    compression = stbi__get16be(s);
+    if (compression > 1)
+        return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+    // Check size
+    if (!stbi__mad3sizes_valid(4, w, h, 0))
+        return stbi__errpuc("too large", "Corrupt PSD");
+
+    // Create the destination image.
+
+    if (!compression && bitdepth == 16 && bpc == 16) {
+        out = (stbi_uc*)stbi__malloc_mad3(8, w, h, 0);
+        ri->bits_per_channel = 16;
+    } else
+        out = (stbi_uc*)stbi__malloc(4 * w * h);
+
+    if (!out)
+        return stbi__errpuc("outofmem", "Out of memory");
+    pixelCount = w * h;
+
+    // Initialize the data to zero.
+    // memset( out, 0, pixelCount * 4 );
+
+    // Finally, the image data.
+    if (compression) {
+        // RLE as used by .PSD and .TIFF
+        // Loop until you get the number of unpacked bytes you are expecting:
+        //     Read the next source byte into n.
+        //     If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+        //     Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+        //     Else if n is 128, noop.
+        // Endloop
+
+        // The RLE-compressed data is preceded by a 2-byte data count for each row in the data,
+        // which we're going to just skip.
+        stbi__skip(s, h * channelCount * 2);
+
+        // Read the RLE data by channel.
+        for (channel = 0; channel < 4; channel++) {
+            stbi_uc* p;
+
+            p = out + channel;
+            if (channel >= channelCount) {
+                // Fill this channel with default data.
+                for (i = 0; i < pixelCount; i++, p += 4)
+                    *p = (channel == 3 ? 255 : 0);
+            } else {
+                // Read the RLE data.
+                if (!stbi__psd_decode_rle(s, p, pixelCount)) {
+                    STBI_FREE(out);
+                    return stbi__errpuc("corrupt", "bad RLE data");
+                }
+            }
+        }
+
+    } else {
+        // We're at the raw image data.  It's each channel in order (Red, Green, Blue, Alpha, ...)
+        // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
+
+        // Read the data by channel.
+        for (channel = 0; channel < 4; channel++) {
+            if (channel >= channelCount) {
+                // Fill this channel with default data.
+                if (bitdepth == 16 && bpc == 16) {
+                    stbi__uint16* q = ((stbi__uint16*)out) + channel;
+                    stbi__uint16 val = channel == 3 ? 65535 : 0;
+                    for (i = 0; i < pixelCount; i++, q += 4)
+                        *q = val;
+                } else {
+                    stbi_uc* p = out + channel;
+                    stbi_uc val = channel == 3 ? 255 : 0;
+                    for (i = 0; i < pixelCount; i++, p += 4)
+                        *p = val;
+                }
+            } else {
+                if (ri->bits_per_channel == 16) { // output bpc
+                    stbi__uint16* q = ((stbi__uint16*)out) + channel;
+                    for (i = 0; i < pixelCount; i++, q += 4)
+                        *q = (stbi__uint16)stbi__get16be(s);
+                } else {
+                    stbi_uc* p = out + channel;
+                    if (bitdepth == 16) { // input bpc
+                        for (i = 0; i < pixelCount; i++, p += 4)
+                            *p = (stbi_uc)(stbi__get16be(s) >> 8);
+                    } else {
+                        for (i = 0; i < pixelCount; i++, p += 4)
+                            *p = stbi__get8(s);
+                    }
+                }
+            }
+        }
+    }
+
+    // remove weird white matte from PSD
+    if (channelCount >= 4) {
+        if (ri->bits_per_channel == 16) {
+            for (i = 0; i < w * h; ++i) {
+                stbi__uint16* pixel = (stbi__uint16*)out + 4 * i;
+                if (pixel[3] != 0 && pixel[3] != 65535) {
+                    float a = pixel[3] / 65535.0f;
+                    float ra = 1.0f / a;
+                    float inv_a = 65535.0f * (1 - ra);
+                    pixel[0] = (stbi__uint16)(pixel[0] * ra + inv_a);
+                    pixel[1] = (stbi__uint16)(pixel[1] * ra + inv_a);
+                    pixel[2] = (stbi__uint16)(pixel[2] * ra + inv_a);
+                }
+            }
+        } else {
+            for (i = 0; i < w * h; ++i) {
+                unsigned char* pixel = out + 4 * i;
+                if (pixel[3] != 0 && pixel[3] != 255) {
+                    float a = pixel[3] / 255.0f;
+                    float ra = 1.0f / a;
+                    float inv_a = 255.0f * (1 - ra);
+                    pixel[0] = (unsigned char)(pixel[0] * ra + inv_a);
+                    pixel[1] = (unsigned char)(pixel[1] * ra + inv_a);
+                    pixel[2] = (unsigned char)(pixel[2] * ra + inv_a);
+                }
+            }
+        }
+    }
+
+    // convert to desired output format
+    if (req_comp && req_comp != 4) {
+        if (ri->bits_per_channel == 16)
+            out = (stbi_uc*)stbi__convert_format16((stbi__uint16*)out, 4, req_comp, w, h);
+        else
+            out = stbi__convert_format(out, 4, req_comp, w, h);
+        if (out == NULL)
+            return out; // stbi__convert_format frees input on failure
+    }
+
+    if (comp)
+        *comp = 4;
+    *y = h;
+    *x = w;
+
+    return out;
+}
+#    endif
+
+// *************************************************************************************************
+// Softimage PIC loader
+// by Tom Seddon
+//
+// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
+// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
+
+#    ifndef STBI_NO_PIC
+static int stbi__pic_is4(stbi__context* s, char const* str)
+{
+    int i;
+    for (i = 0; i < 4; ++i)
+        if (stbi__get8(s) != (stbi_uc)str[i])
+            return 0;
+
+    return 1;
+}
+
+static int stbi__pic_test_core(stbi__context* s)
+{
+    int i;
+
+    if (!stbi__pic_is4(s, "\x53\x80\xF6\x34"))
+        return 0;
+
+    for (i = 0; i < 84; ++i)
+        stbi__get8(s);
+
+    if (!stbi__pic_is4(s, "PICT"))
+        return 0;
+
+    return 1;
+}
+
+typedef struct
+{
+    stbi_uc size, type, channel;
+} stbi__pic_packet;
+
+static stbi_uc* stbi__readval(stbi__context* s, int channel, stbi_uc* dest)
+{
+    int mask = 0x80, i;
+
+    for (i = 0; i < 4; ++i, mask >>= 1) {
+        if (channel & mask) {
+            if (stbi__at_eof(s))
+                return stbi__errpuc("bad file", "PIC file too short");
+            dest[i] = stbi__get8(s);
+        }
+    }
+
+    return dest;
+}
+
+static void stbi__copyval(int channel, stbi_uc* dest, stbi_uc const* src)
+{
+    int mask = 0x80, i;
+
+    for (i = 0; i < 4; ++i, mask >>= 1)
+        if (channel & mask)
+            dest[i] = src[i];
+}
+
+static stbi_uc* stbi__pic_load_core(stbi__context* s, int width, int height, int* comp, stbi_uc* result)
+{
+    int act_comp = 0, num_packets = 0, y, chained;
+    stbi__pic_packet packets[10];
+
+    // this will (should...) cater for even some bizarre stuff like having data
+    // for the same channel in multiple packets.
+    do {
+        stbi__pic_packet* packet;
+
+        if (num_packets == sizeof(packets) / sizeof(packets[0]))
+            return stbi__errpuc("bad format", "too many packets");
+
+        packet = &packets[num_packets++];
+
+        chained = stbi__get8(s);
+        packet->size = stbi__get8(s);
+        packet->type = stbi__get8(s);
+        packet->channel = stbi__get8(s);
+
+        act_comp |= packet->channel;
+
+        if (stbi__at_eof(s))
+            return stbi__errpuc("bad file", "file too short (reading packets)");
+        if (packet->size != 8)
+            return stbi__errpuc("bad format", "packet isn't 8bpp");
+    } while (chained);
+
+    *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
+
+    for (y = 0; y < height; ++y) {
+        int packet_idx;
+
+        for (packet_idx = 0; packet_idx < num_packets; ++packet_idx) {
+            stbi__pic_packet* packet = &packets[packet_idx];
+            stbi_uc* dest = result + y * width * 4;
+
+            switch (packet->type) {
+            default:
+                return stbi__errpuc("bad format", "packet has bad compression type");
+
+            case 0: { // uncompressed
+                int x;
+
+                for (x = 0; x < width; ++x, dest += 4)
+                    if (!stbi__readval(s, packet->channel, dest))
+                        return 0;
+                break;
+            }
+
+            case 1: // Pure RLE
+            {
+                int left = width, i;
+
+                while (left > 0) {
+                    stbi_uc count, value[4];
+
+                    count = stbi__get8(s);
+                    if (stbi__at_eof(s))
+                        return stbi__errpuc("bad file", "file too short (pure read count)");
+
+                    if (count > left)
+                        count = (stbi_uc)left;
+
+                    if (!stbi__readval(s, packet->channel, value))
+                        return 0;
+
+                    for (i = 0; i < count; ++i, dest += 4)
+                        stbi__copyval(packet->channel, dest, value);
+                    left -= count;
+                }
+            } break;
+
+            case 2: { // Mixed RLE
+                int left = width;
+                while (left > 0) {
+                    int count = stbi__get8(s), i;
+                    if (stbi__at_eof(s))
+                        return stbi__errpuc("bad file", "file too short (mixed read count)");
+
+                    if (count >= 128) { // Repeated
+                        stbi_uc value[4];
+
+                        if (count == 128)
+                            count = stbi__get16be(s);
+                        else
+                            count -= 127;
+                        if (count > left)
+                            return stbi__errpuc("bad file", "scanline overrun");
+
+                        if (!stbi__readval(s, packet->channel, value))
+                            return 0;
+
+                        for (i = 0; i < count; ++i, dest += 4)
+                            stbi__copyval(packet->channel, dest, value);
+                    } else { // Raw
+                        ++count;
+                        if (count > left)
+                            return stbi__errpuc("bad file", "scanline overrun");
+
+                        for (i = 0; i < count; ++i, dest += 4)
+                            if (!stbi__readval(s, packet->channel, dest))
+                                return 0;
+                    }
+                    left -= count;
+                }
+                break;
+            }
+            }
+        }
+    }
+
+    return result;
+}
+
+static void* stbi__pic_load(stbi__context* s, int* px, int* py, int* comp, int req_comp, stbi__result_info* ri)
+{
+    stbi_uc* result;
+    int i, x, y, internal_comp;
+    STBI_NOTUSED(ri);
+
+    if (!comp)
+        comp = &internal_comp;
+
+    for (i = 0; i < 92; ++i)
+        stbi__get8(s);
+
+    x = stbi__get16be(s);
+    y = stbi__get16be(s);
+
+    if (y > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+    if (x > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+
+    if (stbi__at_eof(s))
+        return stbi__errpuc("bad file", "file too short (pic header)");
+    if (!stbi__mad3sizes_valid(x, y, 4, 0))
+        return stbi__errpuc("too large", "PIC image too large to decode");
+
+    stbi__get32be(s); // skip `ratio'
+    stbi__get16be(s); // skip `fields'
+    stbi__get16be(s); // skip `pad'
+
+    // intermediate buffer is RGBA
+    result = (stbi_uc*)stbi__malloc_mad3(x, y, 4, 0);
+    if (!result)
+        return stbi__errpuc("outofmem", "Out of memory");
+    memset(result, 0xff, x * y * 4);
+
+    if (!stbi__pic_load_core(s, x, y, comp, result)) {
+        STBI_FREE(result);
+        result = 0;
+    }
+    *px = x;
+    *py = y;
+    if (req_comp == 0)
+        req_comp = *comp;
+    result = stbi__convert_format(result, 4, req_comp, x, y);
+
+    return result;
+}
+
+static int stbi__pic_test(stbi__context* s)
+{
+    int r = stbi__pic_test_core(s);
+    stbi__rewind(s);
+    return r;
+}
+#    endif
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+
+#    ifndef STBI_NO_GIF
+typedef struct
+{
+    stbi__int16 prefix;
+    stbi_uc first;
+    stbi_uc suffix;
+} stbi__gif_lzw;
+
+typedef struct
+{
+    int w, h;
+    stbi_uc* out;        // output buffer (always 4 components)
+    stbi_uc* background; // The current "background" as far as a gif is concerned
+    stbi_uc* history;
+    int flags, bgindex, ratio, transparent, eflags;
+    stbi_uc pal[256][4];
+    stbi_uc lpal[256][4];
+    stbi__gif_lzw codes[8192];
+    stbi_uc* color_table;
+    int parse, step;
+    int lflags;
+    int start_x, start_y;
+    int max_x, max_y;
+    int cur_x, cur_y;
+    int line_size;
+    int delay;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context* s)
+{
+    int sz;
+    if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+        return 0;
+    sz = stbi__get8(s);
+    if (sz != '9' && sz != '7')
+        return 0;
+    if (stbi__get8(s) != 'a')
+        return 0;
+    return 1;
+}
+
+static int stbi__gif_test(stbi__context* s)
+{
+    int r = stbi__gif_test_raw(s);
+    stbi__rewind(s);
+    return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context* s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+    int i;
+    for (i = 0; i < num_entries; ++i) {
+        pal[i][2] = stbi__get8(s);
+        pal[i][1] = stbi__get8(s);
+        pal[i][0] = stbi__get8(s);
+        pal[i][3] = transp == i ? 0 : 255;
+    }
+}
+
+static int stbi__gif_header(stbi__context* s, stbi__gif* g, int* comp, int is_info)
+{
+    stbi_uc version;
+    if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+        return stbi__err("not GIF", "Corrupt GIF");
+
+    version = stbi__get8(s);
+    if (version != '7' && version != '9')
+        return stbi__err("not GIF", "Corrupt GIF");
+    if (stbi__get8(s) != 'a')
+        return stbi__err("not GIF", "Corrupt GIF");
+
+    stbi__g_failure_reason = "";
+    g->w = stbi__get16le(s);
+    g->h = stbi__get16le(s);
+    g->flags = stbi__get8(s);
+    g->bgindex = stbi__get8(s);
+    g->ratio = stbi__get8(s);
+    g->transparent = -1;
+
+    if (g->w > STBI_MAX_DIMENSIONS)
+        return stbi__err("too large", "Very large image (corrupt?)");
+    if (g->h > STBI_MAX_DIMENSIONS)
+        return stbi__err("too large", "Very large image (corrupt?)");
+
+    if (comp != 0)
+        *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
+
+    if (is_info)
+        return 1;
+
+    if (g->flags & 0x80)
+        stbi__gif_parse_colortable(s, g->pal, 2 << (g->flags & 7), -1);
+
+    return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context* s, int* x, int* y, int* comp)
+{
+    stbi__gif* g = (stbi__gif*)stbi__malloc(sizeof(stbi__gif));
+    if (!g)
+        return stbi__err("outofmem", "Out of memory");
+    if (!stbi__gif_header(s, g, comp, 1)) {
+        STBI_FREE(g);
+        stbi__rewind(s);
+        return 0;
+    }
+    if (x)
+        *x = g->w;
+    if (y)
+        *y = g->h;
+    STBI_FREE(g);
+    return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif* g, stbi__uint16 code)
+{
+    stbi_uc *p, *c;
+    int idx;
+
+    // recurse to decode the prefixes, since the linked-list is backwards,
+    // and working backwards through an interleaved image would be nasty
+    if (g->codes[code].prefix >= 0)
+        stbi__out_gif_code(g, g->codes[code].prefix);
+
+    if (g->cur_y >= g->max_y)
+        return;
+
+    idx = g->cur_x + g->cur_y;
+    p = &g->out[idx];
+    g->history[idx / 4] = 1;
+
+    c = &g->color_table[g->codes[code].suffix * 4];
+    if (c[3] > 128) { // don't render transparent pixels;
+        p[0] = c[2];
+        p[1] = c[1];
+        p[2] = c[0];
+        p[3] = c[3];
+    }
+    g->cur_x += 4;
+
+    if (g->cur_x >= g->max_x) {
+        g->cur_x = g->start_x;
+        g->cur_y += g->step;
+
+        while (g->cur_y >= g->max_y && g->parse > 0) {
+            g->step = (1 << g->parse) * g->line_size;
+            g->cur_y = g->start_y + (g->step >> 1);
+            --g->parse;
+        }
+    }
+}
+
+static stbi_uc* stbi__process_gif_raster(stbi__context* s, stbi__gif* g)
+{
+    stbi_uc lzw_cs;
+    stbi__int32 len, init_code;
+    stbi__uint32 first;
+    stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+    stbi__gif_lzw* p;
+
+    lzw_cs = stbi__get8(s);
+    if (lzw_cs > 12)
+        return NULL;
+    clear = 1 << lzw_cs;
+    first = 1;
+    codesize = lzw_cs + 1;
+    codemask = (1 << codesize) - 1;
+    bits = 0;
+    valid_bits = 0;
+    for (init_code = 0; init_code < clear; init_code++) {
+        g->codes[init_code].prefix = -1;
+        g->codes[init_code].first = (stbi_uc)init_code;
+        g->codes[init_code].suffix = (stbi_uc)init_code;
+    }
+
+    // support no starting clear code
+    avail = clear + 2;
+    oldcode = -1;
+
+    len = 0;
+    for (;;) {
+        if (valid_bits < codesize) {
+            if (len == 0) {
+                len = stbi__get8(s); // start new block
+                if (len == 0)
+                    return g->out;
+            }
+            --len;
+            bits |= (stbi__int32)stbi__get8(s) << valid_bits;
+            valid_bits += 8;
+        } else {
+            stbi__int32 code = bits & codemask;
+            bits >>= codesize;
+            valid_bits -= codesize;
+            // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+            if (code == clear) { // clear code
+                codesize = lzw_cs + 1;
+                codemask = (1 << codesize) - 1;
+                avail = clear + 2;
+                oldcode = -1;
+                first = 0;
+            } else if (code == clear + 1) { // end of stream code
+                stbi__skip(s, len);
+                while ((len = stbi__get8(s)) > 0)
+                    stbi__skip(s, len);
+                return g->out;
+            } else if (code <= avail) {
+                if (first) {
+                    return stbi__errpuc("no clear code", "Corrupt GIF");
+                }
+
+                if (oldcode >= 0) {
+                    p = &g->codes[avail++];
+                    if (avail > 8192) {
+                        return stbi__errpuc("too many codes", "Corrupt GIF");
+                    }
+
+                    p->prefix = (stbi__int16)oldcode;
+                    p->first = g->codes[oldcode].first;
+                    p->suffix = (code == avail) ? p->first : g->codes[code].first;
+                } else if (code == avail)
+                    return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+                stbi__out_gif_code(g, (stbi__uint16)code);
+
+                if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+                    codesize++;
+                    codemask = (1 << codesize) - 1;
+                }
+
+                oldcode = code;
+            } else {
+                return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+            }
+        }
+    }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+// two back is the image from two frames ago, used for a very specific disposal format
+static stbi_uc* stbi__gif_load_next(stbi__context* s, stbi__gif* g, int* comp, int req_comp, stbi_uc* two_back)
+{
+    int dispose;
+    int first_frame;
+    int pi;
+    int pcount;
+    STBI_NOTUSED(req_comp);
+
+    // on first frame, any non-written pixels get the background colour (non-transparent)
+    first_frame = 0;
+    if (g->out == 0) {
+        if (!stbi__gif_header(s, g, comp, 0))
+            return 0; // stbi__g_failure_reason set by stbi__gif_header
+        if (!stbi__mad3sizes_valid(4, g->w, g->h, 0))
+            return stbi__errpuc("too large", "GIF image is too large");
+        pcount = g->w * g->h;
+        g->out = (stbi_uc*)stbi__malloc(4 * pcount);
+        g->background = (stbi_uc*)stbi__malloc(4 * pcount);
+        g->history = (stbi_uc*)stbi__malloc(pcount);
+        if (!g->out || !g->background || !g->history)
+            return stbi__errpuc("outofmem", "Out of memory");
+
+        // image is treated as "transparent" at the start - ie, nothing overwrites the current background;
+        // background colour is only used for pixels that are not rendered first frame, after that "background"
+        // color refers to the color that was there the previous frame.
+        memset(g->out, 0x00, 4 * pcount);
+        memset(g->background, 0x00, 4 * pcount); // state of the background (starts transparent)
+        memset(g->history, 0x00, pcount);        // pixels that were affected previous frame
+        first_frame = 1;
+    } else {
+        // second frame - how do we dispose of the previous one?
+        dispose = (g->eflags & 0x1C) >> 2;
+        pcount = g->w * g->h;
+
+        if ((dispose == 3) && (two_back == 0)) {
+            dispose = 2; // if I don't have an image to revert back to, default to the old background
+        }
+
+        if (dispose == 3) { // use previous graphic
+            for (pi = 0; pi < pcount; ++pi) {
+                if (g->history[pi]) {
+                    memcpy(&g->out[pi * 4], &two_back[pi * 4], 4);
+                }
+            }
+        } else if (dispose == 2) {
+            // restore what was changed last frame to background before that frame;
+            for (pi = 0; pi < pcount; ++pi) {
+                if (g->history[pi]) {
+                    memcpy(&g->out[pi * 4], &g->background[pi * 4], 4);
+                }
+            }
+        } else {
+            // This is a non-disposal case eithe way, so just
+            // leave the pixels as is, and they will become the new background
+            // 1: do not dispose
+            // 0:  not specified.
+        }
+
+        // background is what out is after the undoing of the previou frame;
+        memcpy(g->background, g->out, 4 * g->w * g->h);
+    }
+
+    // clear my history;
+    memset(g->history, 0x00, g->w * g->h); // pixels that were affected previous frame
+
+    for (;;) {
+        int tag = stbi__get8(s);
+        switch (tag) {
+        case 0x2C: /* Image Descriptor */
+        {
+            stbi__int32 x, y, w, h;
+            stbi_uc* o;
+
+            x = stbi__get16le(s);
+            y = stbi__get16le(s);
+            w = stbi__get16le(s);
+            h = stbi__get16le(s);
+            if (((x + w) > (g->w)) || ((y + h) > (g->h)))
+                return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
+
+            g->line_size = g->w * 4;
+            g->start_x = x * 4;
+            g->start_y = y * g->line_size;
+            g->max_x = g->start_x + w * 4;
+            g->max_y = g->start_y + h * g->line_size;
+            g->cur_x = g->start_x;
+            g->cur_y = g->start_y;
+
+            // if the width of the specified rectangle is 0, that means
+            // we may not see *any* pixels or the image is malformed;
+            // to make sure this is caught, move the current y down to
+            // max_y (which is what out_gif_code checks).
+            if (w == 0)
+                g->cur_y = g->max_y;
+
+            g->lflags = stbi__get8(s);
+
+            if (g->lflags & 0x40) {
+                g->step = 8 * g->line_size; // first interlaced spacing
+                g->parse = 3;
+            } else {
+                g->step = g->line_size;
+                g->parse = 0;
+            }
+
+            if (g->lflags & 0x80) {
+                stbi__gif_parse_colortable(s, g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+                g->color_table = (stbi_uc*)g->lpal;
+            } else if (g->flags & 0x80) {
+                g->color_table = (stbi_uc*)g->pal;
+            } else
+                return stbi__errpuc("missing color table", "Corrupt GIF");
+
+            o = stbi__process_gif_raster(s, g);
+            if (!o)
+                return NULL;
+
+            // if this was the first frame,
+            pcount = g->w * g->h;
+            if (first_frame && (g->bgindex > 0)) {
+                // if first frame, any pixel not drawn to gets the background color
+                for (pi = 0; pi < pcount; ++pi) {
+                    if (g->history[pi] == 0) {
+                        g->pal[g->bgindex][3] = 255; // just in case it was made transparent, undo that; It will be reset next frame if need be;
+                        memcpy(&g->out[pi * 4], &g->pal[g->bgindex], 4);
+                    }
+                }
+            }
+
+            return o;
+        }
+
+        case 0x21: // Comment Extension.
+        {
+            int len;
+            int ext = stbi__get8(s);
+            if (ext == 0xF9) { // Graphic Control Extension.
+                len = stbi__get8(s);
+                if (len == 4) {
+                    g->eflags = stbi__get8(s);
+                    g->delay = 10 * stbi__get16le(s); // delay - 1/100th of a second, saving as 1/1000ths.
+
+                    // unset old transparent
+                    if (g->transparent >= 0) {
+                        g->pal[g->transparent][3] = 255;
+                    }
+                    if (g->eflags & 0x01) {
+                        g->transparent = stbi__get8(s);
+                        if (g->transparent >= 0) {
+                            g->pal[g->transparent][3] = 0;
+                        }
+                    } else {
+                        // don't need transparent
+                        stbi__skip(s, 1);
+                        g->transparent = -1;
+                    }
+                } else {
+                    stbi__skip(s, len);
+                    break;
+                }
+            }
+            while ((len = stbi__get8(s)) != 0) {
+                stbi__skip(s, len);
+            }
+            break;
+        }
+
+        case 0x3B:              // gif stream termination code
+            return (stbi_uc*)s; // using '1' causes warning on some compilers
+
+        default:
+            return stbi__errpuc("unknown code", "Corrupt GIF");
+        }
+    }
+}
+
+static void* stbi__load_gif_main_outofmem(stbi__gif* g, stbi_uc* out, int** delays)
+{
+    STBI_FREE(g->out);
+    STBI_FREE(g->history);
+    STBI_FREE(g->background);
+
+    if (out)
+        STBI_FREE(out);
+    if (delays && *delays)
+        STBI_FREE(*delays);
+    return stbi__errpuc("outofmem", "Out of memory");
+}
+
+static void* stbi__load_gif_main(stbi__context* s, int** delays, int* x, int* y, int* z, int* comp, int req_comp)
+{
+    if (stbi__gif_test(s)) {
+        int layers = 0;
+        stbi_uc* u = 0;
+        stbi_uc* out = 0;
+        stbi_uc* two_back = 0;
+        stbi__gif g;
+        int stride;
+        int out_size = 0;
+        int delays_size = 0;
+
+        STBI_NOTUSED(out_size);
+        STBI_NOTUSED(delays_size);
+
+        memset(&g, 0, sizeof(g));
+        if (delays) {
+            *delays = 0;
+        }
+
+        do {
+            u = stbi__gif_load_next(s, &g, comp, req_comp, two_back);
+            if (u == (stbi_uc*)s)
+                u = 0; // end of animated gif marker
+
+            if (u) {
+                *x = g.w;
+                *y = g.h;
+                ++layers;
+                stride = g.w * g.h * 4;
+
+                if (out) {
+                    void* tmp = (stbi_uc*)STBI_REALLOC_SIZED(out, out_size, layers * stride);
+                    if (!tmp)
+                        return stbi__load_gif_main_outofmem(&g, out, delays);
+                    else {
+                        out = (stbi_uc*)tmp;
+                        out_size = layers * stride;
+                    }
+
+                    if (delays) {
+                        int* new_delays = (int*)STBI_REALLOC_SIZED(*delays, delays_size, sizeof(int) * layers);
+                        if (!new_delays)
+                            return stbi__load_gif_main_outofmem(&g, out, delays);
+                        *delays = new_delays;
+                        delays_size = layers * sizeof(int);
+                    }
+                } else {
+                    out = (stbi_uc*)stbi__malloc(layers * stride);
+                    if (!out)
+                        return stbi__load_gif_main_outofmem(&g, out, delays);
+                    out_size = layers * stride;
+                    if (delays) {
+                        *delays = (int*)stbi__malloc(layers * sizeof(int));
+                        if (!*delays)
+                            return stbi__load_gif_main_outofmem(&g, out, delays);
+                        delays_size = layers * sizeof(int);
+                    }
+                }
+                memcpy(out + ((layers - 1) * stride), u, stride);
+                if (layers >= 2) {
+                    two_back = out - 2 * stride;
+                }
+
+                if (delays) {
+                    (*delays)[layers - 1U] = g.delay;
+                }
+            }
+        } while (u != 0);
+
+        // free temp buffer;
+        STBI_FREE(g.out);
+        STBI_FREE(g.history);
+        STBI_FREE(g.background);
+
+        // do the final conversion after loading everything;
+        if (req_comp && req_comp != 4)
+            out = stbi__convert_format(out, 4, req_comp, layers * g.w, g.h);
+
+        *z = layers;
+        return out;
+    } else {
+        return stbi__errpuc("not GIF", "Image was not as a gif type.");
+    }
+}
+
+static void* stbi__gif_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    stbi_uc* u = 0;
+    stbi__gif g;
+    memset(&g, 0, sizeof(g));
+    STBI_NOTUSED(ri);
+
+    u = stbi__gif_load_next(s, &g, comp, req_comp, 0);
+    if (u == (stbi_uc*)s)
+        u = 0; // end of animated gif marker
+    if (u) {
+        *x = g.w;
+        *y = g.h;
+
+        // moved conversion to after successful load so that the same
+        // can be done for multiple frames.
+        if (req_comp && req_comp != 4)
+            u = stbi__convert_format(u, 4, req_comp, g.w, g.h);
+    } else if (g.out) {
+        // if there was an error and we allocated an image buffer, free it!
+        STBI_FREE(g.out);
+    }
+
+    // free buffers needed for multiple frame loading;
+    STBI_FREE(g.history);
+    STBI_FREE(g.background);
+
+    return u;
+}
+
+static int stbi__gif_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    return stbi__gif_info_raw(s, x, y, comp);
+}
+#    endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#    ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context* s, char const* signature)
+{
+    int i;
+    for (i = 0; signature[i]; ++i)
+        if (stbi__get8(s) != signature[i])
+            return 0;
+    stbi__rewind(s);
+    return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+    int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
+    stbi__rewind(s);
+    if (!r) {
+        r = stbi__hdr_test_core(s, "#?RGBE\n");
+        stbi__rewind(s);
+    }
+    return r;
+}
+
+#        define STBI__HDR_BUFLEN 1024
+static char* stbi__hdr_gettoken(stbi__context* z, char* buffer)
+{
+    int len = 0;
+    char c = '\0';
+
+    c = (char)stbi__get8(z);
+
+    while (!stbi__at_eof(z) && c != '\n') {
+        buffer[len++] = c;
+        if (len == STBI__HDR_BUFLEN - 1) {
+            // flush to end of line
+            while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+                ;
+            break;
+        }
+        c = (char)stbi__get8(z);
+    }
+
+    buffer[len] = 0;
+    return buffer;
+}
+
+static void stbi__hdr_convert(float* output, stbi_uc* input, int req_comp)
+{
+    if (input[3] != 0) {
+        float f1;
+        // Exponent
+        f1 = (float)ldexp(1.0f, input[3] - (int)(128 + 8));
+        if (req_comp <= 2)
+            output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+        else {
+            output[0] = input[0] * f1;
+            output[1] = input[1] * f1;
+            output[2] = input[2] * f1;
+        }
+        if (req_comp == 2)
+            output[1] = 1;
+        if (req_comp == 4)
+            output[3] = 1;
+    } else {
+        switch (req_comp) {
+        case 4:
+            output[3] = 1; /* fallthrough */
+        case 3:
+            output[0] = output[1] = output[2] = 0;
+            break;
+        case 2:
+            output[1] = 1; /* fallthrough */
+        case 1:
+            output[0] = 0;
+            break;
+        }
+    }
+}
+
+static float* stbi__hdr_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    char buffer[STBI__HDR_BUFLEN];
+    char* token;
+    int valid = 0;
+    int width, height;
+    stbi_uc* scanline;
+    float* hdr_data;
+    int len;
+    unsigned char count, value;
+    int i, j, k, c1, c2, z;
+    char const* headerToken;
+    STBI_NOTUSED(ri);
+
+    // Check identifier
+    headerToken = stbi__hdr_gettoken(s, buffer);
+    if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
+        return stbi__errpf("not HDR", "Corrupt HDR image");
+
+    // Parse header
+    for (;;) {
+        token = stbi__hdr_gettoken(s, buffer);
+        if (token[0] == 0)
+            break;
+        if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0)
+            valid = 1;
+    }
+
+    if (!valid)
+        return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+    // Parse width and height
+    // can't use sscanf() if we're not using stdio!
+    token = stbi__hdr_gettoken(s, buffer);
+    if (strncmp(token, "-Y ", 3))
+        return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+    token += 3;
+    height = (int)strtol(token, &token, 10);
+    while (*token == ' ')
+        ++token;
+    if (strncmp(token, "+X ", 3))
+        return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+    token += 3;
+    width = (int)strtol(token, NULL, 10);
+
+    if (height > STBI_MAX_DIMENSIONS)
+        return stbi__errpf("too large", "Very large image (corrupt?)");
+    if (width > STBI_MAX_DIMENSIONS)
+        return stbi__errpf("too large", "Very large image (corrupt?)");
+
+    *x = width;
+    *y = height;
+
+    if (comp)
+        *comp = 3;
+    if (req_comp == 0)
+        req_comp = 3;
+
+    if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+        return stbi__errpf("too large", "HDR image is too large");
+
+    // Read data
+    hdr_data = (float*)stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+    if (!hdr_data)
+        return stbi__errpf("outofmem", "Out of memory");
+
+    // Load image data
+    // image data is stored as some number of sca
+    if (width < 8 || width >= 32768) {
+        // Read flat data
+        for (j = 0; j < height; ++j) {
+            for (i = 0; i < width; ++i) {
+                stbi_uc rgbe[4];
+            main_decode_loop:
+                stbi__getn(s, rgbe, 4);
+                stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+            }
+        }
+    } else {
+        // Read RLE-encoded data
+        scanline = NULL;
+
+        for (j = 0; j < height; ++j) {
+            c1 = stbi__get8(s);
+            c2 = stbi__get8(s);
+            len = stbi__get8(s);
+            if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+                // not run-length encoded, so we have to actually use THIS data as a decoded
+                // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+                stbi_uc rgbe[4];
+                rgbe[0] = (stbi_uc)c1;
+                rgbe[1] = (stbi_uc)c2;
+                rgbe[2] = (stbi_uc)len;
+                rgbe[3] = (stbi_uc)stbi__get8(s);
+                stbi__hdr_convert(hdr_data, rgbe, req_comp);
+                i = 1;
+                j = 0;
+                STBI_FREE(scanline);
+                goto main_decode_loop; // yes, this makes no sense
+            }
+            len <<= 8;
+            len |= stbi__get8(s);
+            if (len != width) {
+                STBI_FREE(hdr_data);
+                STBI_FREE(scanline);
+                return stbi__errpf("invalid decoded scanline length", "corrupt HDR");
+            }
+            if (scanline == NULL) {
+                scanline = (stbi_uc*)stbi__malloc_mad2(width, 4, 0);
+                if (!scanline) {
+                    STBI_FREE(hdr_data);
+                    return stbi__errpf("outofmem", "Out of memory");
+                }
+            }
+
+            for (k = 0; k < 4; ++k) {
+                int nleft;
+                i = 0;
+                while ((nleft = width - i) > 0) {
+                    count = stbi__get8(s);
+                    if (count > 128) {
+                        // Run
+                        value = stbi__get8(s);
+                        count -= 128;
+                        if ((count == 0) || (count > nleft)) {
+                            STBI_FREE(hdr_data);
+                            STBI_FREE(scanline);
+                            return stbi__errpf("corrupt", "bad RLE data in HDR");
+                        }
+                        for (z = 0; z < count; ++z)
+                            scanline[i++ * 4 + k] = value;
+                    } else {
+                        // Dump
+                        if ((count == 0) || (count > nleft)) {
+                            STBI_FREE(hdr_data);
+                            STBI_FREE(scanline);
+                            return stbi__errpf("corrupt", "bad RLE data in HDR");
+                        }
+                        for (z = 0; z < count; ++z)
+                            scanline[i++ * 4 + k] = stbi__get8(s);
+                    }
+                }
+            }
+            for (i = 0; i < width; ++i)
+                stbi__hdr_convert(hdr_data + (j * width + i) * req_comp, scanline + i * 4, req_comp);
+        }
+        if (scanline)
+            STBI_FREE(scanline);
+    }
+
+    return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    char buffer[STBI__HDR_BUFLEN];
+    char* token;
+    int valid = 0;
+    int dummy;
+
+    if (!x)
+        x = &dummy;
+    if (!y)
+        y = &dummy;
+    if (!comp)
+        comp = &dummy;
+
+    if (stbi__hdr_test(s) == 0) {
+        stbi__rewind(s);
+        return 0;
+    }
+
+    for (;;) {
+        token = stbi__hdr_gettoken(s, buffer);
+        if (token[0] == 0)
+            break;
+        if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0)
+            valid = 1;
+    }
+
+    if (!valid) {
+        stbi__rewind(s);
+        return 0;
+    }
+    token = stbi__hdr_gettoken(s, buffer);
+    if (strncmp(token, "-Y ", 3)) {
+        stbi__rewind(s);
+        return 0;
+    }
+    token += 3;
+    *y = (int)strtol(token, &token, 10);
+    while (*token == ' ')
+        ++token;
+    if (strncmp(token, "+X ", 3)) {
+        stbi__rewind(s);
+        return 0;
+    }
+    token += 3;
+    *x = (int)strtol(token, NULL, 10);
+    *comp = 3;
+    return 1;
+}
+#    endif // STBI_NO_HDR
+
+#    ifndef STBI_NO_BMP
+static int stbi__bmp_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    void* p;
+    stbi__bmp_data info;
+
+    info.all_a = 255;
+    p = stbi__bmp_parse_header(s, &info);
+    if (p == NULL) {
+        stbi__rewind(s);
+        return 0;
+    }
+    if (x)
+        *x = s->img_x;
+    if (y)
+        *y = s->img_y;
+    if (comp) {
+        if (info.bpp == 24 && info.ma == 0xff000000)
+            *comp = 3;
+        else
+            *comp = info.ma ? 4 : 3;
+    }
+    return 1;
+}
+#    endif
+
+#    ifndef STBI_NO_PSD
+static int stbi__psd_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    int channelCount, dummy, depth;
+    if (!x)
+        x = &dummy;
+    if (!y)
+        y = &dummy;
+    if (!comp)
+        comp = &dummy;
+    if (stbi__get32be(s) != 0x38425053) {
+        stbi__rewind(s);
+        return 0;
+    }
+    if (stbi__get16be(s) != 1) {
+        stbi__rewind(s);
+        return 0;
+    }
+    stbi__skip(s, 6);
+    channelCount = stbi__get16be(s);
+    if (channelCount < 0 || channelCount > 16) {
+        stbi__rewind(s);
+        return 0;
+    }
+    *y = stbi__get32be(s);
+    *x = stbi__get32be(s);
+    depth = stbi__get16be(s);
+    if (depth != 8 && depth != 16) {
+        stbi__rewind(s);
+        return 0;
+    }
+    if (stbi__get16be(s) != 3) {
+        stbi__rewind(s);
+        return 0;
+    }
+    *comp = 4;
+    return 1;
+}
+
+static int stbi__psd_is16(stbi__context* s)
+{
+    int channelCount, depth;
+    if (stbi__get32be(s) != 0x38425053) {
+        stbi__rewind(s);
+        return 0;
+    }
+    if (stbi__get16be(s) != 1) {
+        stbi__rewind(s);
+        return 0;
+    }
+    stbi__skip(s, 6);
+    channelCount = stbi__get16be(s);
+    if (channelCount < 0 || channelCount > 16) {
+        stbi__rewind(s);
+        return 0;
+    }
+    STBI_NOTUSED(stbi__get32be(s));
+    STBI_NOTUSED(stbi__get32be(s));
+    depth = stbi__get16be(s);
+    if (depth != 16) {
+        stbi__rewind(s);
+        return 0;
+    }
+    return 1;
+}
+#    endif
+
+#    ifndef STBI_NO_PIC
+static int stbi__pic_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    int act_comp = 0, num_packets = 0, chained, dummy;
+    stbi__pic_packet packets[10];
+
+    if (!x)
+        x = &dummy;
+    if (!y)
+        y = &dummy;
+    if (!comp)
+        comp = &dummy;
+
+    if (!stbi__pic_is4(s, "\x53\x80\xF6\x34")) {
+        stbi__rewind(s);
+        return 0;
+    }
+
+    stbi__skip(s, 88);
+
+    *x = stbi__get16be(s);
+    *y = stbi__get16be(s);
+    if (stbi__at_eof(s)) {
+        stbi__rewind(s);
+        return 0;
+    }
+    if ((*x) != 0 && (1 << 28) / (*x) < (*y)) {
+        stbi__rewind(s);
+        return 0;
+    }
+
+    stbi__skip(s, 8);
+
+    do {
+        stbi__pic_packet* packet;
+
+        if (num_packets == sizeof(packets) / sizeof(packets[0]))
+            return 0;
+
+        packet = &packets[num_packets++];
+        chained = stbi__get8(s);
+        packet->size = stbi__get8(s);
+        packet->type = stbi__get8(s);
+        packet->channel = stbi__get8(s);
+        act_comp |= packet->channel;
+
+        if (stbi__at_eof(s)) {
+            stbi__rewind(s);
+            return 0;
+        }
+        if (packet->size != 8) {
+            stbi__rewind(s);
+            return 0;
+        }
+    } while (chained);
+
+    *comp = (act_comp & 0x10 ? 4 : 3);
+
+    return 1;
+}
+#    endif
+
+// *************************************************************************************************
+// Portable Gray Map and Portable Pixel Map loader
+// by Ken Miller
+//
+// PGM: http://netpbm.sourceforge.net/doc/pgm.html
+// PPM: http://netpbm.sourceforge.net/doc/ppm.html
+//
+// Known limitations:
+//    Does not support comments in the header section
+//    Does not support ASCII image data (formats P2 and P3)
+
+#    ifndef STBI_NO_PNM
+
+static int stbi__pnm_test(stbi__context* s)
+{
+    char p, t;
+    p = (char)stbi__get8(s);
+    t = (char)stbi__get8(s);
+    if (p != 'P' || (t != '5' && t != '6')) {
+        stbi__rewind(s);
+        return 0;
+    }
+    return 1;
+}
+
+static void* stbi__pnm_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+    stbi_uc* out;
+    STBI_NOTUSED(ri);
+
+    ri->bits_per_channel = stbi__pnm_info(s, (int*)&s->img_x, (int*)&s->img_y, (int*)&s->img_n);
+    if (ri->bits_per_channel == 0)
+        return 0;
+
+    if (s->img_y > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+    if (s->img_x > STBI_MAX_DIMENSIONS)
+        return stbi__errpuc("too large", "Very large image (corrupt?)");
+
+    *x = s->img_x;
+    *y = s->img_y;
+    if (comp)
+        *comp = s->img_n;
+
+    if (!stbi__mad4sizes_valid(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0))
+        return stbi__errpuc("too large", "PNM too large");
+
+    out = (stbi_uc*)stbi__malloc_mad4(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0);
+    if (!out)
+        return stbi__errpuc("outofmem", "Out of memory");
+    if (!stbi__getn(s, out, s->img_n * s->img_x * s->img_y * (ri->bits_per_channel / 8))) {
+        STBI_FREE(out);
+        return stbi__errpuc("bad PNM", "PNM file truncated");
+    }
+
+    if (req_comp && req_comp != s->img_n) {
+        if (ri->bits_per_channel == 16) {
+            out = (stbi_uc*)stbi__convert_format16((stbi__uint16*)out, s->img_n, req_comp, s->img_x, s->img_y);
+        } else {
+            out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+        }
+        if (out == NULL)
+            return out; // stbi__convert_format frees input on failure
+    }
+    return out;
+}
+
+static int stbi__pnm_isspace(char c)
+{
+    return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
+}
+
+static void stbi__pnm_skip_whitespace(stbi__context* s, char* c)
+{
+    for (;;) {
+        while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
+            *c = (char)stbi__get8(s);
+
+        if (stbi__at_eof(s) || *c != '#')
+            break;
+
+        while (!stbi__at_eof(s) && *c != '\n' && *c != '\r')
+            *c = (char)stbi__get8(s);
+    }
+}
+
+static int stbi__pnm_isdigit(char c)
+{
+    return c >= '0' && c <= '9';
+}
+
+static int stbi__pnm_getinteger(stbi__context* s, char* c)
+{
+    int value = 0;
+
+    while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
+        value = value * 10 + (*c - '0');
+        *c = (char)stbi__get8(s);
+        if ((value > 214748364) || (value == 214748364 && *c > '7'))
+            return stbi__err("integer parse overflow", "Parsing an integer in the PPM header overflowed a 32-bit int");
+    }
+
+    return value;
+}
+
+static int stbi__pnm_info(stbi__context* s, int* x, int* y, int* comp)
+{
+    int maxv, dummy;
+    char c, p, t;
+
+    if (!x)
+        x = &dummy;
+    if (!y)
+        y = &dummy;
+    if (!comp)
+        comp = &dummy;
+
+    stbi__rewind(s);
+
+    // Get identifier
+    p = (char)stbi__get8(s);
+    t = (char)stbi__get8(s);
+    if (p != 'P' || (t != '5' && t != '6')) {
+        stbi__rewind(s);
+        return 0;
+    }
+
+    *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
+
+    c = (char)stbi__get8(s);
+    stbi__pnm_skip_whitespace(s, &c);
+
+    *x = stbi__pnm_getinteger(s, &c); // read width
+    if (*x == 0)
+        return stbi__err("invalid width", "PPM image header had zero or overflowing width");
+    stbi__pnm_skip_whitespace(s, &c);
+
+    *y = stbi__pnm_getinteger(s, &c); // read height
+    if (*y == 0)
+        return stbi__err("invalid width", "PPM image header had zero or overflowing width");
+    stbi__pnm_skip_whitespace(s, &c);
+
+    maxv = stbi__pnm_getinteger(s, &c); // read max value
+    if (maxv > 65535)
+        return stbi__err("max value > 65535", "PPM image supports only 8-bit and 16-bit images");
+    else if (maxv > 255)
+        return 16;
+    else
+        return 8;
+}
+
+static int stbi__pnm_is16(stbi__context* s)
+{
+    if (stbi__pnm_info(s, NULL, NULL, NULL) == 16)
+        return 1;
+    return 0;
+}
+#    endif
+
+static int stbi__info_main(stbi__context* s, int* x, int* y, int* comp)
+{
+#    ifndef STBI_NO_JPEG
+    if (stbi__jpeg_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_PNG
+    if (stbi__png_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_GIF
+    if (stbi__gif_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_BMP
+    if (stbi__bmp_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_PSD
+    if (stbi__psd_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_PIC
+    if (stbi__pic_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_PNM
+    if (stbi__pnm_info(s, x, y, comp))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_HDR
+    if (stbi__hdr_info(s, x, y, comp))
+        return 1;
+#    endif
+
+// test tga last because it's a crappy test!
+#    ifndef STBI_NO_TGA
+    if (stbi__tga_info(s, x, y, comp))
+        return 1;
+#    endif
+    return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static int stbi__is_16_main(stbi__context* s)
+{
+#    ifndef STBI_NO_PNG
+    if (stbi__png_is16(s))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_PSD
+    if (stbi__psd_is16(s))
+        return 1;
+#    endif
+
+#    ifndef STBI_NO_PNM
+    if (stbi__pnm_is16(s))
+        return 1;
+#    endif
+    return 0;
+}
+
+#    ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const* filename, int* x, int* y, int* comp)
+{
+    FILE* f = stbi__fopen(filename, "rb");
+    int result;
+    if (!f)
+        return stbi__err("can't fopen", "Unable to open file");
+    result = stbi_info_from_file(f, x, y, comp);
+    fclose(f);
+    return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE* f, int* x, int* y, int* comp)
+{
+    int r;
+    stbi__context s;
+    long pos = ftell(f);
+    stbi__start_file(&s, f);
+    r = stbi__info_main(&s, x, y, comp);
+    fseek(f, pos, SEEK_SET);
+    return r;
+}
+
+STBIDEF int stbi_is_16_bit(char const* filename)
+{
+    FILE* f = stbi__fopen(filename, "rb");
+    int result;
+    if (!f)
+        return stbi__err("can't fopen", "Unable to open file");
+    result = stbi_is_16_bit_from_file(f);
+    fclose(f);
+    return result;
+}
+
+STBIDEF int stbi_is_16_bit_from_file(FILE* f)
+{
+    int r;
+    stbi__context s;
+    long pos = ftell(f);
+    stbi__start_file(&s, f);
+    r = stbi__is_16_main(&s);
+    fseek(f, pos, SEEK_SET);
+    return r;
+}
+#    endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp)
+{
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+    return stbi__info_main(&s, x, y, comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const* c, void* user, int* x, int* y, int* comp)
+{
+    stbi__context s;
+    stbi__start_callbacks(&s, (stbi_io_callbacks*)c, user);
+    return stbi__info_main(&s, x, y, comp);
+}
+
+STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const* buffer, int len)
+{
+    stbi__context s;
+    stbi__start_mem(&s, buffer, len);
+    return stbi__is_16_main(&s);
+}
+
+STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const* c, void* user)
+{
+    stbi__context s;
+    stbi__start_callbacks(&s, (stbi_io_callbacks*)c, user);
+    return stbi__is_16_main(&s);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+   revision history:
+      2.20  (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs
+      2.19  (2018-02-11) fix warning
+      2.18  (2018-01-30) fix warnings
+      2.17  (2018-01-29) change sbti__shiftsigned to avoid clang -O2 bug
+                         1-bit BMP
+                         *_is_16_bit api
+                         avoid warnings
+      2.16  (2017-07-23) all functions have 16-bit variants;
+                         STBI_NO_STDIO works again;
+                         compilation fixes;
+                         fix rounding in unpremultiply;
+                         optimize vertical flip;
+                         disable raw_len validation;
+                         documentation fixes
+      2.15  (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+                         warning fixes; disable run-time SSE detection on gcc;
+                         uniform handling of optional "return" values;
+                         thread-safe initialization of zlib tables
+      2.14  (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+      2.13  (2016-11-29) add 16-bit API, only supported for PNG right now
+      2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+      2.11  (2016-04-02) allocate large structures on the stack
+                         remove white matting for transparent PSD
+                         fix reported channel count for PNG & BMP
+                         re-enable SSE2 in non-gcc 64-bit
+                         support RGB-formatted JPEG
+                         read 16-bit PNGs (only as 8-bit)
+      2.10  (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
+      2.09  (2016-01-16) allow comments in PNM files
+                         16-bit-per-pixel TGA (not bit-per-component)
+                         info() for TGA could break due to .hdr handling
+                         info() for BMP to shares code instead of sloppy parse
+                         can use STBI_REALLOC_SIZED if allocator doesn't support realloc
+                         code cleanup
+      2.08  (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+      2.07  (2015-09-13) fix compiler warnings
+                         partial animated GIF support
+                         limited 16-bpc PSD support
+                         #ifdef unused functions
+                         bug with < 92 byte PIC,PNM,HDR,TGA
+      2.06  (2015-04-19) fix bug where PSD returns wrong '*comp' value
+      2.05  (2015-04-19) fix bug in progressive JPEG handling, fix warning
+      2.04  (2015-04-15) try to re-enable SIMD on MinGW 64-bit
+      2.03  (2015-04-12) extra corruption checking (mmozeiko)
+                         stbi_set_flip_vertically_on_load (nguillemot)
+                         fix NEON support; fix mingw support
+      2.02  (2015-01-19) fix incorrect assert, fix warning
+      2.01  (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
+      2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+      2.00  (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
+                         progressive JPEG (stb)
+                         PGM/PPM support (Ken Miller)
+                         STBI_MALLOC,STBI_REALLOC,STBI_FREE
+                         GIF bugfix -- seemingly never worked
+                         STBI_NO_*, STBI_ONLY_*
+      1.48  (2014-12-14) fix incorrectly-named assert()
+      1.47  (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
+                         optimize PNG (ryg)
+                         fix bug in interlaced PNG with user-specified channel count (stb)
+      1.46  (2014-08-26)
+              fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
+      1.45  (2014-08-16)
+              fix MSVC-ARM internal compiler error by wrapping malloc
+      1.44  (2014-08-07)
+              various warning fixes from Ronny Chevalier
+      1.43  (2014-07-15)
+              fix MSVC-only compiler problem in code changed in 1.42
+      1.42  (2014-07-09)
+              don't define _CRT_SECURE_NO_WARNINGS (affects user code)
+              fixes to stbi__cleanup_jpeg path
+              added STBI_ASSERT to avoid requiring assert.h
+      1.41  (2014-06-25)
+              fix search&replace from 1.36 that messed up comments/error messages
+      1.40  (2014-06-22)
+              fix gcc struct-initialization warning
+      1.39  (2014-06-15)
+              fix to TGA optimization when req_comp != number of components in TGA;
+              fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
+              add support for BMP version 5 (more ignored fields)
+      1.38  (2014-06-06)
+              suppress MSVC warnings on integer casts truncating values
+              fix accidental rename of 'skip' field of I/O
+      1.37  (2014-06-04)
+              remove duplicate typedef
+      1.36  (2014-06-03)
+              convert to header file single-file library
+              if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
+      1.35  (2014-05-27)
+              various warnings
+              fix broken STBI_SIMD path
+              fix bug where stbi_load_from_file no longer left file pointer in correct place
+              fix broken non-easy path for 32-bit BMP (possibly never used)
+              TGA optimization by Arseny Kapoulkine
+      1.34  (unknown)
+              use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
+      1.33  (2011-07-14)
+              make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
+      1.32  (2011-07-13)
+              support for "info" function for all supported filetypes (SpartanJ)
+      1.31  (2011-06-20)
+              a few more leak fixes, bug in PNG handling (SpartanJ)
+      1.30  (2011-06-11)
+              added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+              removed deprecated format-specific test/load functions
+              removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
+              error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
+              fix inefficiency in decoding 32-bit BMP (David Woo)
+      1.29  (2010-08-16)
+              various warning fixes from Aurelien Pocheville
+      1.28  (2010-08-01)
+              fix bug in GIF palette transparency (SpartanJ)
+      1.27  (2010-08-01)
+              cast-to-stbi_uc to fix warnings
+      1.26  (2010-07-24)
+              fix bug in file buffering for PNG reported by SpartanJ
+      1.25  (2010-07-17)
+              refix trans_data warning (Won Chun)
+      1.24  (2010-07-12)
+              perf improvements reading from files on platforms with lock-heavy fgetc()
+              minor perf improvements for jpeg
+              deprecated type-specific functions so we'll get feedback if they're needed
+              attempt to fix trans_data warning (Won Chun)
+      1.23    fixed bug in iPhone support
+      1.22  (2010-07-10)
+              removed image *writing* support
+              stbi_info support from Jetro Lauha
+              GIF support from Jean-Marc Lienher
+              iPhone PNG-extensions from James Brown
+              warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva)
+      1.21    fix use of 'stbi_uc' in header (reported by jon blow)
+      1.20    added support for Softimage PIC, by Tom Seddon
+      1.19    bug in interlaced PNG corruption check (found by ryg)
+      1.18  (2008-08-02)
+              fix a threading bug (local mutable static)
+      1.17    support interlaced PNG
+      1.16    major bugfix - stbi__convert_format converted one too many pixels
+      1.15    initialize some fields for thread safety
+      1.14    fix threadsafe conversion bug
+              header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
+      1.13    threadsafe
+      1.12    const qualifiers in the API
+      1.11    Support installable IDCT, colorspace conversion routines
+      1.10    Fixes for 64-bit (don't use "unsigned long")
+              optimized upsampling by Fabian "ryg" Giesen
+      1.09    Fix format-conversion for PSD code (bad global variables!)
+      1.08    Thatcher Ulrich's PSD code integrated by Nicolas Schulz
+      1.07    attempt to fix C++ warning/errors again
+      1.06    attempt to fix C++ warning/errors again
+      1.05    fix TGA loading to return correct *comp and use good luminance calc
+      1.04    default float alpha is 1, not 255; use 'void *' for stbi_image_free
+      1.03    bugfixes to STBI_NO_STDIO, STBI_NO_HDR
+      1.02    support for (subset of) HDR files, float interface for preferred access to them
+      1.01    fix bug: possible bug in handling right-side up bmps... not sure
+              fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
+      1.00    interface to zlib that skips zlib header
+      0.99    correct handling of alpha in palette
+      0.98    TGA loader by lonesock; dynamically add loaders (untested)
+      0.97    jpeg errors on too large a file; also catch another malloc failure
+      0.96    fix detection of invalid v value - particleman@mollyrocket forum
+      0.95    during header scan, seek to markers in case of padding
+      0.94    STBI_NO_STDIO to disable stdio usage; rename all #defines the same
+      0.93    handle jpegtran output; verbose errors
+      0.92    read 4,8,16,24,32-bit BMP files of several formats
+      0.91    output 24-bit Windows 3.0 BMP files
+      0.90    fix a few more warnings; bump version number to approach 1.0
+      0.61    bugfixes due to Marc LeBlanc, Christopher Lloyd
+      0.60    fix compiling as c++
+      0.59    fix warnings: merge Dave Moore's -Wall fixes
+      0.58    fix bug: zlib uncompressed mode len/nlen was wrong endian
+      0.57    fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
+      0.56    fix bug: zlib uncompressed mode len vs. nlen
+      0.55    fix bug: restart_interval not initialized to 0
+      0.54    allow NULL for 'int *comp'
+      0.53    fix bug in png 3->4; speedup png decoding
+      0.52    png handles req_comp=3,4 directly; minor cleanup; jpeg comments
+      0.51    obey req_comp requests, 1-component jpegs return as 1-component,
+              on 'test' only check type, not whether we support this variant
+      0.50  (2006-11-19)
+              first released version
+*/
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
diff --git a/src/image/stb_image_write.h b/src/image/stb_image_write.h
new file mode 100644
index 0000000..6ed3dd8
--- /dev/null
+++ b/src/image/stb_image_write.h
@@ -0,0 +1,1807 @@
+/* stb_image_write - v1.16 - public domain - http://nothings.org/stb
+   writes out PNG/BMP/TGA/JPEG/HDR images to C stdio - Sean Barrett 2010-2015
+                                     no warranty implied; use at your own risk
+
+   Before #including,
+
+       #define STB_IMAGE_WRITE_IMPLEMENTATION
+
+   in the file that you want to have the implementation.
+
+   Will probably not work correctly with strict-aliasing optimizations.
+
+ABOUT:
+
+   This header file is a library for writing images to C stdio or a callback.
+
+   The PNG output is not optimal; it is 20-50% larger than the file
+   written by a decent optimizing implementation; though providing a custom
+   zlib compress function (see STBIW_ZLIB_COMPRESS) can mitigate that.
+   This library is designed for source code compactness and simplicity,
+   not optimal image file size or run-time performance.
+
+BUILDING:
+
+   You can #define STBIW_ASSERT(x) before the #include to avoid using assert.h.
+   You can #define STBIW_MALLOC(), STBIW_REALLOC(), and STBIW_FREE() to replace
+   malloc,realloc,free.
+   You can #define STBIW_MEMMOVE() to replace memmove()
+   You can #define STBIW_ZLIB_COMPRESS to use a custom zlib-style compress function
+   for PNG compression (instead of the builtin one), it must have the following signature:
+   unsigned char * my_compress(unsigned char *data, int data_len, int *out_len, int quality);
+   The returned data will be freed with STBIW_FREE() (free() by default),
+   so it must be heap allocated with STBIW_MALLOC() (malloc() by default),
+
+UNICODE:
+
+   If compiling for Windows and you wish to use Unicode filenames, compile
+   with
+       #define STBIW_WINDOWS_UTF8
+   and pass utf8-encoded filenames. Call stbiw_convert_wchar_to_utf8 to convert
+   Windows wchar_t filenames to utf8.
+
+USAGE:
+
+   There are five functions, one for each image file format:
+
+     int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);
+     int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
+     int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
+     int stbi_write_jpg(char const *filename, int w, int h, int comp, const void *data, int quality);
+     int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data);
+
+     void stbi_flip_vertically_on_write(int flag); // flag is non-zero to flip data vertically
+
+   There are also five equivalent functions that use an arbitrary write function. You are
+   expected to open/close your file-equivalent before and after calling these:
+
+     int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data, int stride_in_bytes);
+     int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data);
+     int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data);
+     int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data);
+     int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality);
+
+   where the callback is:
+      void stbi_write_func(void *context, void *data, int size);
+
+   You can configure it with these global variables:
+      int stbi_write_tga_with_rle;             // defaults to true; set to 0 to disable RLE
+      int stbi_write_png_compression_level;    // defaults to 8; set to higher for more compression
+      int stbi_write_force_png_filter;         // defaults to -1; set to 0..5 to force a filter mode
+
+
+   You can define STBI_WRITE_NO_STDIO to disable the file variant of these
+   functions, so the library will not use stdio.h at all. However, this will
+   also disable HDR writing, because it requires stdio for formatted output.
+
+   Each function returns 0 on failure and non-0 on success.
+
+   The functions create an image file defined by the parameters. The image
+   is a rectangle of pixels stored from left-to-right, top-to-bottom.
+   Each pixel contains 'comp' channels of data stored interleaved with 8-bits
+   per channel, in the following order: 1=Y, 2=YA, 3=RGB, 4=RGBA. (Y is
+   monochrome color.) The rectangle is 'w' pixels wide and 'h' pixels tall.
+   The *data pointer points to the first byte of the top-left-most pixel.
+   For PNG, "stride_in_bytes" is the distance in bytes from the first byte of
+   a row of pixels to the first byte of the next row of pixels.
+
+   PNG creates output files with the same number of components as the input.
+   The BMP format expands Y to RGB in the file format and does not
+   output alpha.
+
+   PNG supports writing rectangles of data even when the bytes storing rows of
+   data are not consecutive in memory (e.g. sub-rectangles of a larger image),
+   by supplying the stride between the beginning of adjacent rows. The other
+   formats do not. (Thus you cannot write a native-format BMP through the BMP
+   writer, both because it is in BGR order and because it may have padding
+   at the end of the line.)
+
+   PNG allows you to set the deflate compression level by setting the global
+   variable 'stbi_write_png_compression_level' (it defaults to 8).
+
+   HDR expects linear float data. Since the format is always 32-bit rgb(e)
+   data, alpha (if provided) is discarded, and for monochrome data it is
+   replicated across all three channels.
+
+   TGA supports RLE or non-RLE compressed data. To use non-RLE-compressed
+   data, set the global variable 'stbi_write_tga_with_rle' to 0.
+
+   JPEG does ignore alpha channels in input data; quality is between 1 and 100.
+   Higher quality looks better but results in a bigger image.
+   JPEG baseline (no JPEG progressive).
+
+CREDITS:
+
+
+   Sean Barrett           -    PNG/BMP/TGA
+   Baldur Karlsson        -    HDR
+   Jean-Sebastien Guay    -    TGA monochrome
+   Tim Kelsey             -    misc enhancements
+   Alan Hickman           -    TGA RLE
+   Emmanuel Julien        -    initial file IO callback implementation
+   Jon Olick              -    original jo_jpeg.cpp code
+   Daniel Gibson          -    integrate JPEG, allow external zlib
+   Aarni Koskela          -    allow choosing PNG filter
+
+   bugfixes:
+      github:Chribba
+      Guillaume Chereau
+      github:jry2
+      github:romigrou
+      Sergio Gonzalez
+      Jonas Karlsson
+      Filip Wasil
+      Thatcher Ulrich
+      github:poppolopoppo
+      Patrick Boettcher
+      github:xeekworx
+      Cap Petschulat
+      Simon Rodriguez
+      Ivan Tikhonov
+      github:ignotion
+      Adam Schackart
+      Andrew Kensler
+
+LICENSE
+
+  See end of file for license information.
+
+*/
+
+#ifndef INCLUDE_STB_IMAGE_WRITE_H
+#    define INCLUDE_STB_IMAGE_WRITE_H
+
+#    include <stdlib.h>
+
+// if STB_IMAGE_WRITE_STATIC causes problems, try defining STBIWDEF to 'inline' or 'static inline'
+#    ifndef STBIWDEF
+#        ifdef STB_IMAGE_WRITE_STATIC
+#            define STBIWDEF static
+#        else
+#            ifdef __cplusplus
+#                define STBIWDEF extern "C"
+#            else
+#                define STBIWDEF extern
+#            endif
+#        endif
+#    endif
+
+#    ifndef STB_IMAGE_WRITE_STATIC // C++ forbids static forward declarations
+STBIWDEF int stbi_write_tga_with_rle;
+STBIWDEF int stbi_write_png_compression_level;
+STBIWDEF int stbi_write_force_png_filter;
+#    endif
+
+#    ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_png(char const* filename, int w, int h, int comp, void const* data, int stride_in_bytes);
+STBIWDEF int stbi_write_bmp(char const* filename, int w, int h, int comp, void const* data);
+STBIWDEF int stbi_write_tga(char const* filename, int w, int h, int comp, void const* data);
+STBIWDEF int stbi_write_hdr(char const* filename, int w, int h, int comp, float const* data);
+STBIWDEF int stbi_write_jpg(char const* filename, int x, int y, int comp, void const* data, int quality);
+
+#        ifdef STBIW_WINDOWS_UTF8
+STBIWDEF int stbiw_convert_wchar_to_utf8(char* buffer, size_t bufferlen, wchar_t const* input);
+#        endif
+#    endif
+
+typedef void stbi_write_func(void* context, void* data, int size);
+
+STBIWDEF int stbi_write_png_to_func(stbi_write_func* func, void* context, int w, int h, int comp, void const* data, int stride_in_bytes);
+STBIWDEF int stbi_write_bmp_to_func(stbi_write_func* func, void* context, int w, int h, int comp, void const* data);
+STBIWDEF int stbi_write_tga_to_func(stbi_write_func* func, void* context, int w, int h, int comp, void const* data);
+STBIWDEF int stbi_write_hdr_to_func(stbi_write_func* func, void* context, int w, int h, int comp, float const* data);
+STBIWDEF int stbi_write_jpg_to_func(stbi_write_func* func, void* context, int x, int y, int comp, void const* data, int quality);
+
+STBIWDEF void stbi_flip_vertically_on_write(int flip_boolean);
+
+#endif // INCLUDE_STB_IMAGE_WRITE_H
+
+#ifdef STB_IMAGE_WRITE_IMPLEMENTATION
+
+#    ifdef _WIN32
+#        ifndef _CRT_SECURE_NO_WARNINGS
+#            define _CRT_SECURE_NO_WARNINGS
+#        endif
+#        ifndef _CRT_NONSTDC_NO_DEPRECATE
+#            define _CRT_NONSTDC_NO_DEPRECATE
+#        endif
+#    endif
+
+#    ifndef STBI_WRITE_NO_STDIO
+#        include <stdio.h>
+#    endif // STBI_WRITE_NO_STDIO
+
+#    include <math.h>
+#    include <stdarg.h>
+#    include <stdlib.h>
+#    include <string.h>
+
+#    if defined(STBIW_MALLOC) && defined(STBIW_FREE) && (defined(STBIW_REALLOC) || defined(STBIW_REALLOC_SIZED))
+// ok
+#    elif !defined(STBIW_MALLOC) && !defined(STBIW_FREE) && !defined(STBIW_REALLOC) && !defined(STBIW_REALLOC_SIZED)
+// ok
+#    else
+#        error "Must define all or none of STBIW_MALLOC, STBIW_FREE, and STBIW_REALLOC (or STBIW_REALLOC_SIZED)."
+#    endif
+
+#    ifndef STBIW_MALLOC
+#        define STBIW_MALLOC(sz) malloc(sz)
+#        define STBIW_REALLOC(p, newsz) realloc(p, newsz)
+#        define STBIW_FREE(p) free(p)
+#    endif
+
+#    ifndef STBIW_REALLOC_SIZED
+#        define STBIW_REALLOC_SIZED(p, oldsz, newsz) STBIW_REALLOC(p, newsz)
+#    endif
+
+#    ifndef STBIW_MEMMOVE
+#        define STBIW_MEMMOVE(a, b, sz) memmove(a, b, sz)
+#    endif
+
+#    ifndef STBIW_ASSERT
+#        include <assert.h>
+#        define STBIW_ASSERT(x) assert(x)
+#    endif
+
+#    define STBIW_UCHAR(x) (unsigned char)((x) & 0xff)
+
+#    ifdef STB_IMAGE_WRITE_STATIC
+static int stbi_write_png_compression_level = 8;
+static int stbi_write_tga_with_rle = 1;
+static int stbi_write_force_png_filter = -1;
+#    else
+int stbi_write_png_compression_level = 8;
+int stbi_write_tga_with_rle = 1;
+int stbi_write_force_png_filter = -1;
+#    endif
+
+static int stbi__flip_vertically_on_write = 0;
+
+STBIWDEF void stbi_flip_vertically_on_write(int flag)
+{
+    stbi__flip_vertically_on_write = flag;
+}
+
+typedef struct
+{
+    stbi_write_func* func;
+    void* context;
+    unsigned char buffer[64];
+    int buf_used;
+} stbi__write_context;
+
+// initialize a callback-based context
+static void stbi__start_write_callbacks(stbi__write_context* s, stbi_write_func* c, void* context)
+{
+    s->func = c;
+    s->context = context;
+}
+
+#    ifndef STBI_WRITE_NO_STDIO
+
+static void stbi__stdio_write(void* context, void* data, int size)
+{
+    fwrite(data, 1, size, (FILE*)context);
+}
+
+#        if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8)
+#            ifdef __cplusplus
+#                define STBIW_EXTERN extern "C"
+#            else
+#                define STBIW_EXTERN extern
+#            endif
+STBIW_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, char const* str, int cbmb, wchar_t* widestr, int cchwide);
+STBIW_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, wchar_t const* widestr, int cchwide, char* str, int cbmb, char const* defchar, int* used_default);
+
+STBIWDEF int stbiw_convert_wchar_to_utf8(char* buffer, size_t bufferlen, wchar_t const* input)
+{
+    return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int)bufferlen, NULL, NULL);
+}
+#        endif
+
+static FILE* stbiw__fopen(char const* filename, char const* mode)
+{
+    FILE* f;
+#        if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8)
+    wchar_t wMode[64];
+    wchar_t wFilename[1024];
+    if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename) / sizeof(*wFilename)))
+        return 0;
+
+    if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode) / sizeof(*wMode)))
+        return 0;
+
+#            if defined(_MSC_VER) && _MSC_VER >= 1400
+    if (0 != _wfopen_s(&f, wFilename, wMode))
+        f = 0;
+#            else
+    f = _wfopen(wFilename, wMode);
+#            endif
+
+#        elif defined(_MSC_VER) && _MSC_VER >= 1400
+    if (0 != fopen_s(&f, filename, mode))
+        f = 0;
+#        else
+    f = fopen(filename, mode);
+#        endif
+    return f;
+}
+
+static int stbi__start_write_file(stbi__write_context* s, char const* filename)
+{
+    FILE* f = stbiw__fopen(filename, "wb");
+    stbi__start_write_callbacks(s, stbi__stdio_write, (void*)f);
+    return f != NULL;
+}
+
+static void stbi__end_write_file(stbi__write_context* s)
+{
+    fclose((FILE*)s->context);
+}
+
+#    endif // !STBI_WRITE_NO_STDIO
+
+typedef unsigned int stbiw_uint32;
+typedef int stb_image_write_test[sizeof(stbiw_uint32) == 4 ? 1 : -1];
+
+static void stbiw__writefv(stbi__write_context* s, char const* fmt, va_list v)
+{
+    while (*fmt) {
+        switch (*fmt++) {
+        case ' ':
+            break;
+        case '1': {
+            unsigned char x = STBIW_UCHAR(va_arg(v, int));
+            s->func(s->context, &x, 1);
+            break;
+        }
+        case '2': {
+            int x = va_arg(v, int);
+            unsigned char b[2];
+            b[0] = STBIW_UCHAR(x);
+            b[1] = STBIW_UCHAR(x >> 8);
+            s->func(s->context, b, 2);
+            break;
+        }
+        case '4': {
+            stbiw_uint32 x = va_arg(v, int);
+            unsigned char b[4];
+            b[0] = STBIW_UCHAR(x);
+            b[1] = STBIW_UCHAR(x >> 8);
+            b[2] = STBIW_UCHAR(x >> 16);
+            b[3] = STBIW_UCHAR(x >> 24);
+            s->func(s->context, b, 4);
+            break;
+        }
+        default:
+            STBIW_ASSERT(0);
+            return;
+        }
+    }
+}
+
+static void stbiw__writef(stbi__write_context* s, char const* fmt, ...)
+{
+    va_list v;
+    va_start(v, fmt);
+    stbiw__writefv(s, fmt, v);
+    va_end(v);
+}
+
+static void stbiw__write_flush(stbi__write_context* s)
+{
+    if (s->buf_used) {
+        s->func(s->context, &s->buffer, s->buf_used);
+        s->buf_used = 0;
+    }
+}
+
+static void stbiw__putc(stbi__write_context* s, unsigned char c)
+{
+    s->func(s->context, &c, 1);
+}
+
+static void stbiw__write1(stbi__write_context* s, unsigned char a)
+{
+    if ((size_t)s->buf_used + 1 > sizeof(s->buffer))
+        stbiw__write_flush(s);
+    s->buffer[s->buf_used++] = a;
+}
+
+static void stbiw__write3(stbi__write_context* s, unsigned char a, unsigned char b, unsigned char c)
+{
+    int n;
+    if ((size_t)s->buf_used + 3 > sizeof(s->buffer))
+        stbiw__write_flush(s);
+    n = s->buf_used;
+    s->buf_used = n + 3;
+    s->buffer[n + 0] = a;
+    s->buffer[n + 1] = b;
+    s->buffer[n + 2] = c;
+}
+
+static void stbiw__write_pixel(stbi__write_context* s, int rgb_dir, int comp, int write_alpha, int expand_mono, unsigned char* d)
+{
+    unsigned char bg[3] = { 255, 0, 255 }, px[3];
+    int k;
+
+    if (write_alpha < 0)
+        stbiw__write1(s, d[comp - 1]);
+
+    switch (comp) {
+    case 2: // 2 pixels = mono + alpha, alpha is written separately, so same as 1-channel case
+    case 1:
+        if (expand_mono)
+            stbiw__write3(s, d[0], d[0], d[0]); // monochrome bmp
+        else
+            stbiw__write1(s, d[0]); // monochrome TGA
+        break;
+    case 4:
+        if (!write_alpha) {
+            // composite against pink background
+            for (k = 0; k < 3; ++k)
+                px[k] = bg[k] + ((d[k] - bg[k]) * d[3]) / 255;
+            stbiw__write3(s, px[1 - rgb_dir], px[1], px[1 + rgb_dir]);
+            break;
+        }
+        /* FALLTHROUGH */
+    case 3:
+        stbiw__write3(s, d[1 - rgb_dir], d[1], d[1 + rgb_dir]);
+        break;
+    }
+    if (write_alpha > 0)
+        stbiw__write1(s, d[comp - 1]);
+}
+
+static void stbiw__write_pixels(stbi__write_context* s, int rgb_dir, int vdir, int x, int y, int comp, void* data, int write_alpha, int scanline_pad, int expand_mono)
+{
+    stbiw_uint32 zero = 0;
+    int i, j, j_end;
+
+    if (y <= 0)
+        return;
+
+    if (stbi__flip_vertically_on_write)
+        vdir *= -1;
+
+    if (vdir < 0) {
+        j_end = -1;
+        j = y - 1;
+    } else {
+        j_end = y;
+        j = 0;
+    }
+
+    for (; j != j_end; j += vdir) {
+        for (i = 0; i < x; ++i) {
+            unsigned char* d = (unsigned char*)data + (j * x + i) * comp;
+            stbiw__write_pixel(s, rgb_dir, comp, write_alpha, expand_mono, d);
+        }
+        stbiw__write_flush(s);
+        s->func(s->context, &zero, scanline_pad);
+    }
+}
+
+static int stbiw__outfile(stbi__write_context* s, int rgb_dir, int vdir, int x, int y, int comp, int expand_mono, void* data, int alpha, int pad, char const* fmt, ...)
+{
+    if (y < 0 || x < 0) {
+        return 0;
+    } else {
+        va_list v;
+        va_start(v, fmt);
+        stbiw__writefv(s, fmt, v);
+        va_end(v);
+        stbiw__write_pixels(s, rgb_dir, vdir, x, y, comp, data, alpha, pad, expand_mono);
+        return 1;
+    }
+}
+
+static int stbi_write_bmp_core(stbi__write_context* s, int x, int y, int comp, void const* data)
+{
+    if (comp != 4) {
+        // write RGB bitmap
+        int pad = (-x * 3) & 3;
+        return stbiw__outfile(s, -1, -1, x, y, comp, 1, (void*)data, 0, pad,
+            "11 4 22 4"
+            "4 44 22 444444",
+            'B', 'M', 14 + 40 + (x * 3 + pad) * y, 0, 0, 14 + 40, // file header
+            40, x, y, 1, 24, 0, 0, 0, 0, 0, 0);                   // bitmap header
+    } else {
+        // RGBA bitmaps need a v4 header
+        // use BI_BITFIELDS mode with 32bpp and alpha mask
+        // (straight BI_RGB with alpha mask doesn't work in most readers)
+        return stbiw__outfile(s, -1, -1, x, y, comp, 1, (void*)data, 1, 0,
+            "11 4 22 4"
+            "4 44 22 444444 4444 4 444 444 444 444",
+            'B', 'M', 14 + 108 + x * y * 4, 0, 0, 14 + 108,                                                                  // file header
+            108, x, y, 1, 32, 3, 0, 0, 0, 0, 0, 0xff0000, 0xff00, 0xff, 0xff000000u, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0); // bitmap V4 header
+    }
+}
+
+STBIWDEF int stbi_write_bmp_to_func(stbi_write_func* func, void* context, int x, int y, int comp, void const* data)
+{
+    stbi__write_context s = { 0 };
+    stbi__start_write_callbacks(&s, func, context);
+    return stbi_write_bmp_core(&s, x, y, comp, data);
+}
+
+#    ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_bmp(char const* filename, int x, int y, int comp, void const* data)
+{
+    stbi__write_context s = { 0 };
+    if (stbi__start_write_file(&s, filename)) {
+        int r = stbi_write_bmp_core(&s, x, y, comp, data);
+        stbi__end_write_file(&s);
+        return r;
+    } else
+        return 0;
+}
+#    endif //! STBI_WRITE_NO_STDIO
+
+static int stbi_write_tga_core(stbi__write_context* s, int x, int y, int comp, void* data)
+{
+    int has_alpha = (comp == 2 || comp == 4);
+    int colorbytes = has_alpha ? comp - 1 : comp;
+    int format = colorbytes < 2 ? 3 : 2; // 3 color channels (RGB/RGBA) = 2, 1 color channel (Y/YA) = 3
+
+    if (y < 0 || x < 0)
+        return 0;
+
+    if (!stbi_write_tga_with_rle) {
+        return stbiw__outfile(s, -1, -1, x, y, comp, 0, (void*)data, has_alpha, 0,
+            "111 221 2222 11", 0, 0, format, 0, 0, 0, 0, 0, x, y, (colorbytes + has_alpha) * 8, has_alpha * 8);
+    } else {
+        int i, j, k;
+        int jend, jdir;
+
+        stbiw__writef(s, "111 221 2222 11", 0, 0, format + 8, 0, 0, 0, 0, 0, x, y, (colorbytes + has_alpha) * 8, has_alpha * 8);
+
+        if (stbi__flip_vertically_on_write) {
+            j = 0;
+            jend = y;
+            jdir = 1;
+        } else {
+            j = y - 1;
+            jend = -1;
+            jdir = -1;
+        }
+        for (; j != jend; j += jdir) {
+            unsigned char* row = (unsigned char*)data + j * x * comp;
+            int len;
+
+            for (i = 0; i < x; i += len) {
+                unsigned char* begin = row + i * comp;
+                int diff = 1;
+                len = 1;
+
+                if (i < x - 1) {
+                    ++len;
+                    diff = memcmp(begin, row + (i + 1) * comp, comp);
+                    if (diff) {
+                        unsigned char const* prev = begin;
+                        for (k = i + 2; k < x && len < 128; ++k) {
+                            if (memcmp(prev, row + k * comp, comp)) {
+                                prev += comp;
+                                ++len;
+                            } else {
+                                --len;
+                                break;
+                            }
+                        }
+                    } else {
+                        for (k = i + 2; k < x && len < 128; ++k) {
+                            if (!memcmp(begin, row + k * comp, comp)) {
+                                ++len;
+                            } else {
+                                break;
+                            }
+                        }
+                    }
+                }
+
+                if (diff) {
+                    unsigned char header = STBIW_UCHAR(len - 1);
+                    stbiw__write1(s, header);
+                    for (k = 0; k < len; ++k) {
+                        stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin + k * comp);
+                    }
+                } else {
+                    unsigned char header = STBIW_UCHAR(len - 129);
+                    stbiw__write1(s, header);
+                    stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin);
+                }
+            }
+        }
+        stbiw__write_flush(s);
+    }
+    return 1;
+}
+
+STBIWDEF int stbi_write_tga_to_func(stbi_write_func* func, void* context, int x, int y, int comp, void const* data)
+{
+    stbi__write_context s = { 0 };
+    stbi__start_write_callbacks(&s, func, context);
+    return stbi_write_tga_core(&s, x, y, comp, (void*)data);
+}
+
+#    ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_tga(char const* filename, int x, int y, int comp, void const* data)
+{
+    stbi__write_context s = { 0 };
+    if (stbi__start_write_file(&s, filename)) {
+        int r = stbi_write_tga_core(&s, x, y, comp, (void*)data);
+        stbi__end_write_file(&s);
+        return r;
+    } else
+        return 0;
+}
+#    endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR writer
+// by Baldur Karlsson
+
+#    define stbiw__max(a, b) ((a) > (b) ? (a) : (b))
+
+#    ifndef STBI_WRITE_NO_STDIO
+
+static void stbiw__linear_to_rgbe(unsigned char* rgbe, float* linear)
+{
+    int exponent;
+    float maxcomp = stbiw__max(linear[0], stbiw__max(linear[1], linear[2]));
+
+    if (maxcomp < 1e-32f) {
+        rgbe[0] = rgbe[1] = rgbe[2] = rgbe[3] = 0;
+    } else {
+        float normalize = (float)frexp(maxcomp, &exponent) * 256.0f / maxcomp;
+
+        rgbe[0] = (unsigned char)(linear[0] * normalize);
+        rgbe[1] = (unsigned char)(linear[1] * normalize);
+        rgbe[2] = (unsigned char)(linear[2] * normalize);
+        rgbe[3] = (unsigned char)(exponent + 128);
+    }
+}
+
+static void stbiw__write_run_data(stbi__write_context* s, int length, unsigned char databyte)
+{
+    unsigned char lengthbyte = STBIW_UCHAR(length + 128);
+    STBIW_ASSERT(length + 128 <= 255);
+    s->func(s->context, &lengthbyte, 1);
+    s->func(s->context, &databyte, 1);
+}
+
+static void stbiw__write_dump_data(stbi__write_context* s, int length, unsigned char* data)
+{
+    unsigned char lengthbyte = STBIW_UCHAR(length);
+    STBIW_ASSERT(length <= 128); // inconsistent with spec but consistent with official code
+    s->func(s->context, &lengthbyte, 1);
+    s->func(s->context, data, length);
+}
+
+static void stbiw__write_hdr_scanline(stbi__write_context* s, int width, int ncomp, unsigned char* scratch, float* scanline)
+{
+    unsigned char scanlineheader[4] = { 2, 2, 0, 0 };
+    unsigned char rgbe[4];
+    float linear[3];
+    int x;
+
+    scanlineheader[2] = (width & 0xff00) >> 8;
+    scanlineheader[3] = (width & 0x00ff);
+
+    /* skip RLE for images too small or large */
+    if (width < 8 || width >= 32768) {
+        for (x = 0; x < width; x++) {
+            switch (ncomp) {
+            case 4: /* fallthrough */
+            case 3:
+                linear[2] = scanline[x * ncomp + 2];
+                linear[1] = scanline[x * ncomp + 1];
+                linear[0] = scanline[x * ncomp + 0];
+                break;
+            default:
+                linear[0] = linear[1] = linear[2] = scanline[x * ncomp + 0];
+                break;
+            }
+            stbiw__linear_to_rgbe(rgbe, linear);
+            s->func(s->context, rgbe, 4);
+        }
+    } else {
+        int c, r;
+        /* encode into scratch buffer */
+        for (x = 0; x < width; x++) {
+            switch (ncomp) {
+            case 4: /* fallthrough */
+            case 3:
+                linear[2] = scanline[x * ncomp + 2];
+                linear[1] = scanline[x * ncomp + 1];
+                linear[0] = scanline[x * ncomp + 0];
+                break;
+            default:
+                linear[0] = linear[1] = linear[2] = scanline[x * ncomp + 0];
+                break;
+            }
+            stbiw__linear_to_rgbe(rgbe, linear);
+            scratch[x + width * 0] = rgbe[0];
+            scratch[x + width * 1] = rgbe[1];
+            scratch[x + width * 2] = rgbe[2];
+            scratch[x + width * 3] = rgbe[3];
+        }
+
+        s->func(s->context, scanlineheader, 4);
+
+        /* RLE each component separately */
+        for (c = 0; c < 4; c++) {
+            unsigned char* comp = &scratch[width * c];
+
+            x = 0;
+            while (x < width) {
+                // find first run
+                r = x;
+                while (r + 2 < width) {
+                    if (comp[r] == comp[r + 1] && comp[r] == comp[r + 2])
+                        break;
+                    ++r;
+                }
+                if (r + 2 >= width)
+                    r = width;
+                // dump up to first run
+                while (x < r) {
+                    int len = r - x;
+                    if (len > 128)
+                        len = 128;
+                    stbiw__write_dump_data(s, len, &comp[x]);
+                    x += len;
+                }
+                // if there's a run, output it
+                if (r + 2 < width) { // same test as what we break out of in search loop, so only true if we break'd
+                    // find next byte after run
+                    while (r < width && comp[r] == comp[x])
+                        ++r;
+                    // output run up to r
+                    while (x < r) {
+                        int len = r - x;
+                        if (len > 127)
+                            len = 127;
+                        stbiw__write_run_data(s, len, comp[x]);
+                        x += len;
+                    }
+                }
+            }
+        }
+    }
+}
+
+static int stbi_write_hdr_core(stbi__write_context* s, int x, int y, int comp, float* data)
+{
+    if (y <= 0 || x <= 0 || data == NULL)
+        return 0;
+    else {
+        // Each component is stored separately. Allocate scratch space for full output scanline.
+        unsigned char* scratch = (unsigned char*)STBIW_MALLOC(x * 4);
+        int i, len;
+        char buffer[128];
+        char header[] = "#?RADIANCE\n# Written by stb_image_write.h\nFORMAT=32-bit_rle_rgbe\n";
+        s->func(s->context, header, sizeof(header) - 1);
+
+#        ifdef __STDC_LIB_EXT1__
+        len = sprintf_s(buffer, sizeof(buffer), "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n", y, x);
+#        else
+        len = sprintf(buffer, "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n", y, x);
+#        endif
+        s->func(s->context, buffer, len);
+
+        for (i = 0; i < y; i++)
+            stbiw__write_hdr_scanline(s, x, comp, scratch, data + comp * x * (stbi__flip_vertically_on_write ? y - 1 - i : i));
+        STBIW_FREE(scratch);
+        return 1;
+    }
+}
+
+STBIWDEF int stbi_write_hdr_to_func(stbi_write_func* func, void* context, int x, int y, int comp, float const* data)
+{
+    stbi__write_context s = { 0 };
+    stbi__start_write_callbacks(&s, func, context);
+    return stbi_write_hdr_core(&s, x, y, comp, (float*)data);
+}
+
+STBIWDEF int stbi_write_hdr(char const* filename, int x, int y, int comp, float const* data)
+{
+    stbi__write_context s = { 0 };
+    if (stbi__start_write_file(&s, filename)) {
+        int r = stbi_write_hdr_core(&s, x, y, comp, (float*)data);
+        stbi__end_write_file(&s);
+        return r;
+    } else
+        return 0;
+}
+#    endif // STBI_WRITE_NO_STDIO
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PNG writer
+//
+
+#    ifndef STBIW_ZLIB_COMPRESS
+// stretchy buffer; stbiw__sbpush() == vector<>::push_back() -- stbiw__sbcount() == vector<>::size()
+#        define stbiw__sbraw(a) ((int*)(void*)(a) - 2)
+#        define stbiw__sbm(a) stbiw__sbraw(a)[0]
+#        define stbiw__sbn(a) stbiw__sbraw(a)[1]
+
+#        define stbiw__sbneedgrow(a, n) ((a) == 0 || stbiw__sbn(a) + n >= stbiw__sbm(a))
+#        define stbiw__sbmaybegrow(a, n) (stbiw__sbneedgrow(a, (n)) ? stbiw__sbgrow(a, n) : 0)
+#        define stbiw__sbgrow(a, n) stbiw__sbgrowf((void**)&(a), (n), sizeof(*(a)))
+
+#        define stbiw__sbpush(a, v) (stbiw__sbmaybegrow(a, 1), (a)[stbiw__sbn(a)++] = (v))
+#        define stbiw__sbcount(a) ((a) ? stbiw__sbn(a) : 0)
+#        define stbiw__sbfree(a) ((a) ? STBIW_FREE(stbiw__sbraw(a)), 0 : 0)
+
+static void* stbiw__sbgrowf(void** arr, int increment, int itemsize)
+{
+    int m = *arr ? 2 * stbiw__sbm(*arr) + increment : increment + 1;
+    void* p = STBIW_REALLOC_SIZED(*arr ? stbiw__sbraw(*arr) : 0, *arr ? (stbiw__sbm(*arr) * itemsize + sizeof(int) * 2) : 0, itemsize * m + sizeof(int) * 2);
+    STBIW_ASSERT(p);
+    if (p) {
+        if (!*arr)
+            ((int*)p)[1] = 0;
+        *arr = (void*)((int*)p + 2);
+        stbiw__sbm(*arr) = m;
+    }
+    return *arr;
+}
+
+static unsigned char* stbiw__zlib_flushf(unsigned char* data, unsigned int* bitbuffer, int* bitcount)
+{
+    while (*bitcount >= 8) {
+        stbiw__sbpush(data, STBIW_UCHAR(*bitbuffer));
+        *bitbuffer >>= 8;
+        *bitcount -= 8;
+    }
+    return data;
+}
+
+static int stbiw__zlib_bitrev(int code, int codebits)
+{
+    int res = 0;
+    while (codebits--) {
+        res = (res << 1) | (code & 1);
+        code >>= 1;
+    }
+    return res;
+}
+
+static unsigned int stbiw__zlib_countm(unsigned char* a, unsigned char* b, int limit)
+{
+    int i;
+    for (i = 0; i < limit && i < 258; ++i)
+        if (a[i] != b[i])
+            break;
+    return i;
+}
+
+static unsigned int stbiw__zhash(unsigned char* data)
+{
+    stbiw_uint32 hash = data[0] + (data[1] << 8) + (data[2] << 16);
+    hash ^= hash << 3;
+    hash += hash >> 5;
+    hash ^= hash << 4;
+    hash += hash >> 17;
+    hash ^= hash << 25;
+    hash += hash >> 6;
+    return hash;
+}
+
+#        define stbiw__zlib_flush() (out = stbiw__zlib_flushf(out, &bitbuf, &bitcount))
+#        define stbiw__zlib_add(code, codebits) \
+            (bitbuf |= (code) << bitcount, bitcount += (codebits), stbiw__zlib_flush())
+#        define stbiw__zlib_huffa(b, c) stbiw__zlib_add(stbiw__zlib_bitrev(b, c), c)
+// default huffman tables
+#        define stbiw__zlib_huff1(n) stbiw__zlib_huffa(0x30 + (n), 8)
+#        define stbiw__zlib_huff2(n) stbiw__zlib_huffa(0x190 + (n) - 144, 9)
+#        define stbiw__zlib_huff3(n) stbiw__zlib_huffa(0 + (n) - 256, 7)
+#        define stbiw__zlib_huff4(n) stbiw__zlib_huffa(0xc0 + (n) - 280, 8)
+#        define stbiw__zlib_huff(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : (n) <= 255 ? stbiw__zlib_huff2(n) \
+                : (n) <= 279                                                        ? stbiw__zlib_huff3(n) \
+                                                                                    : stbiw__zlib_huff4(n))
+#        define stbiw__zlib_huffb(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : stbiw__zlib_huff2(n))
+
+#        define stbiw__ZHASH 16384
+
+#    endif // STBIW_ZLIB_COMPRESS
+
+STBIWDEF unsigned char* stbi_zlib_compress(unsigned char* data, int data_len, int* out_len, int quality)
+{
+#    ifdef STBIW_ZLIB_COMPRESS
+    // user provided a zlib compress implementation, use that
+    return STBIW_ZLIB_COMPRESS(data, data_len, out_len, quality);
+#    else  // use builtin
+    static unsigned short lengthc[] = { 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 259 };
+    static unsigned char lengtheb[] = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0 };
+    static unsigned short distc[] = { 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577, 32768 };
+    static unsigned char disteb[] = { 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13 };
+    unsigned int bitbuf = 0;
+    int i, j, bitcount = 0;
+    unsigned char* out = NULL;
+    unsigned char*** hash_table = (unsigned char***)STBIW_MALLOC(stbiw__ZHASH * sizeof(unsigned char**));
+    if (hash_table == NULL)
+        return NULL;
+    if (quality < 5)
+        quality = 5;
+
+    stbiw__sbpush(out, 0x78); // DEFLATE 32K window
+    stbiw__sbpush(out, 0x5e); // FLEVEL = 1
+    stbiw__zlib_add(1, 1);    // BFINAL = 1
+    stbiw__zlib_add(1, 2);    // BTYPE = 1 -- fixed huffman
+
+    for (i = 0; i < stbiw__ZHASH; ++i)
+        hash_table[i] = NULL;
+
+    i = 0;
+    while (i < data_len - 3) {
+        // hash next 3 bytes of data to be compressed
+        int h = stbiw__zhash(data + i) & (stbiw__ZHASH - 1), best = 3;
+        unsigned char* bestloc = 0;
+        unsigned char** hlist = hash_table[h];
+        int n = stbiw__sbcount(hlist);
+        for (j = 0; j < n; ++j) {
+            if (hlist[j] - data > i - 32768) { // if entry lies within window
+                int d = stbiw__zlib_countm(hlist[j], data + i, data_len - i);
+                if (d >= best) {
+                    best = d;
+                    bestloc = hlist[j];
+                }
+            }
+        }
+        // when hash table entry is too long, delete half the entries
+        if (hash_table[h] && stbiw__sbn(hash_table[h]) == 2 * quality) {
+            STBIW_MEMMOVE(hash_table[h], hash_table[h] + quality, sizeof(hash_table[h][0]) * quality);
+            stbiw__sbn(hash_table[h]) = quality;
+        }
+        stbiw__sbpush(hash_table[h], data + i);
+
+        if (bestloc) {
+            // "lazy matching" - check match at *next* byte, and if it's better, do cur byte as literal
+            h = stbiw__zhash(data + i + 1) & (stbiw__ZHASH - 1);
+            hlist = hash_table[h];
+            n = stbiw__sbcount(hlist);
+            for (j = 0; j < n; ++j) {
+                if (hlist[j] - data > i - 32767) {
+                    int e = stbiw__zlib_countm(hlist[j], data + i + 1, data_len - i - 1);
+                    if (e > best) { // if next match is better, bail on current match
+                        bestloc = NULL;
+                        break;
+                    }
+                }
+            }
+        }
+
+        if (bestloc) {
+            int d = (int)(data + i - bestloc); // distance back
+            STBIW_ASSERT(d <= 32767 && best <= 258);
+            for (j = 0; best > lengthc[j + 1] - 1; ++j)
+                ;
+            stbiw__zlib_huff(j + 257);
+            if (lengtheb[j])
+                stbiw__zlib_add(best - lengthc[j], lengtheb[j]);
+            for (j = 0; d > distc[j + 1] - 1; ++j)
+                ;
+            stbiw__zlib_add(stbiw__zlib_bitrev(j, 5), 5);
+            if (disteb[j])
+                stbiw__zlib_add(d - distc[j], disteb[j]);
+            i += best;
+        } else {
+            stbiw__zlib_huffb(data[i]);
+            ++i;
+        }
+    }
+    // write out final bytes
+    for (; i < data_len; ++i)
+        stbiw__zlib_huffb(data[i]);
+    stbiw__zlib_huff(256); // end of block
+    // pad with 0 bits to byte boundary
+    while (bitcount)
+        stbiw__zlib_add(0, 1);
+
+    for (i = 0; i < stbiw__ZHASH; ++i)
+        (void)stbiw__sbfree(hash_table[i]);
+    STBIW_FREE(hash_table);
+
+    // store uncompressed instead if compression was worse
+    if (stbiw__sbn(out) > data_len + 2 + ((data_len + 32766) / 32767) * 5) {
+        stbiw__sbn(out) = 2; // truncate to DEFLATE 32K window and FLEVEL = 1
+        for (j = 0; j < data_len;) {
+            int blocklen = data_len - j;
+            if (blocklen > 32767)
+                blocklen = 32767;
+            stbiw__sbpush(out, data_len - j == blocklen); // BFINAL = ?, BTYPE = 0 -- no compression
+            stbiw__sbpush(out, STBIW_UCHAR(blocklen));    // LEN
+            stbiw__sbpush(out, STBIW_UCHAR(blocklen >> 8));
+            stbiw__sbpush(out, STBIW_UCHAR(~blocklen)); // NLEN
+            stbiw__sbpush(out, STBIW_UCHAR(~blocklen >> 8));
+            memcpy(out + stbiw__sbn(out), data + j, blocklen);
+            stbiw__sbn(out) += blocklen;
+            j += blocklen;
+        }
+    }
+
+    {
+        // compute adler32 on input
+        unsigned int s1 = 1, s2 = 0;
+        int blocklen = (int)(data_len % 5552);
+        j = 0;
+        while (j < data_len) {
+            for (i = 0; i < blocklen; ++i) {
+                s1 += data[j + i];
+                s2 += s1;
+            }
+            s1 %= 65521;
+            s2 %= 65521;
+            j += blocklen;
+            blocklen = 5552;
+        }
+        stbiw__sbpush(out, STBIW_UCHAR(s2 >> 8));
+        stbiw__sbpush(out, STBIW_UCHAR(s2));
+        stbiw__sbpush(out, STBIW_UCHAR(s1 >> 8));
+        stbiw__sbpush(out, STBIW_UCHAR(s1));
+    }
+    *out_len = stbiw__sbn(out);
+    // make returned pointer freeable
+    STBIW_MEMMOVE(stbiw__sbraw(out), out, *out_len);
+    return (unsigned char*)stbiw__sbraw(out);
+#    endif // STBIW_ZLIB_COMPRESS
+}
+
+static unsigned int stbiw__crc32(unsigned char* buffer, int len)
+{
+#    ifdef STBIW_CRC32
+    return STBIW_CRC32(buffer, len);
+#    else
+    static unsigned int crc_table[256] = {
+        0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3,
+        0x0eDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91,
+        0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7,
+        0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5,
+        0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B,
+        0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59,
+        0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F,
+        0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D,
+        0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433,
+        0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01,
+        0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457,
+        0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65,
+        0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB,
+        0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9,
+        0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F,
+        0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD,
+        0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683,
+        0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1,
+        0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7,
+        0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5,
+        0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B,
+        0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79,
+        0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F,
+        0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D,
+        0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713,
+        0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21,
+        0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777,
+        0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45,
+        0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB,
+        0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9,
+        0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF,
+        0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D
+    };
+
+    unsigned int crc = ~0u;
+    int i;
+    for (i = 0; i < len; ++i)
+        crc = (crc >> 8) ^ crc_table[buffer[i] ^ (crc & 0xff)];
+    return ~crc;
+#    endif
+}
+
+#    define stbiw__wpng4(o, a, b, c, d) ((o)[0] = STBIW_UCHAR(a), (o)[1] = STBIW_UCHAR(b), (o)[2] = STBIW_UCHAR(c), (o)[3] = STBIW_UCHAR(d), (o) += 4)
+#    define stbiw__wp32(data, v) stbiw__wpng4(data, (v) >> 24, (v) >> 16, (v) >> 8, (v));
+#    define stbiw__wptag(data, s) stbiw__wpng4(data, s[0], s[1], s[2], s[3])
+
+static void stbiw__wpcrc(unsigned char** data, int len)
+{
+    unsigned int crc = stbiw__crc32(*data - len - 4, len + 4);
+    stbiw__wp32(*data, crc);
+}
+
+static unsigned char stbiw__paeth(int a, int b, int c)
+{
+    int p = a + b - c, pa = abs(p - a), pb = abs(p - b), pc = abs(p - c);
+    if (pa <= pb && pa <= pc)
+        return STBIW_UCHAR(a);
+    if (pb <= pc)
+        return STBIW_UCHAR(b);
+    return STBIW_UCHAR(c);
+}
+
+// @OPTIMIZE: provide an option that always forces left-predict or paeth predict
+static void stbiw__encode_png_line(unsigned char* pixels, int stride_bytes, int width, int height, int y, int n, int filter_type, signed char* line_buffer)
+{
+    static int mapping[] = { 0, 1, 2, 3, 4 };
+    static int firstmap[] = { 0, 1, 0, 5, 6 };
+    int* mymap = (y != 0) ? mapping : firstmap;
+    int i;
+    int type = mymap[filter_type];
+    unsigned char* z = pixels + stride_bytes * (stbi__flip_vertically_on_write ? height - 1 - y : y);
+    int signed_stride = stbi__flip_vertically_on_write ? -stride_bytes : stride_bytes;
+
+    if (type == 0) {
+        memcpy(line_buffer, z, width * n);
+        return;
+    }
+
+    // first loop isn't optimized since it's just one pixel
+    for (i = 0; i < n; ++i) {
+        switch (type) {
+        case 1:
+            line_buffer[i] = z[i];
+            break;
+        case 2:
+            line_buffer[i] = z[i] - z[i - signed_stride];
+            break;
+        case 3:
+            line_buffer[i] = z[i] - (z[i - signed_stride] >> 1);
+            break;
+        case 4:
+            line_buffer[i] = (signed char)(z[i] - stbiw__paeth(0, z[i - signed_stride], 0));
+            break;
+        case 5:
+            line_buffer[i] = z[i];
+            break;
+        case 6:
+            line_buffer[i] = z[i];
+            break;
+        }
+    }
+    switch (type) {
+    case 1:
+        for (i = n; i < width * n; ++i)
+            line_buffer[i] = z[i] - z[i - n];
+        break;
+    case 2:
+        for (i = n; i < width * n; ++i)
+            line_buffer[i] = z[i] - z[i - signed_stride];
+        break;
+    case 3:
+        for (i = n; i < width * n; ++i)
+            line_buffer[i] = z[i] - ((z[i - n] + z[i - signed_stride]) >> 1);
+        break;
+    case 4:
+        for (i = n; i < width * n; ++i)
+            line_buffer[i] = z[i] - stbiw__paeth(z[i - n], z[i - signed_stride], z[i - signed_stride - n]);
+        break;
+    case 5:
+        for (i = n; i < width * n; ++i)
+            line_buffer[i] = z[i] - (z[i - n] >> 1);
+        break;
+    case 6:
+        for (i = n; i < width * n; ++i)
+            line_buffer[i] = z[i] - stbiw__paeth(z[i - n], 0, 0);
+        break;
+    }
+}
+
+STBIWDEF unsigned char* stbi_write_png_to_mem(unsigned char const* pixels, int stride_bytes, int x, int y, int n, int* out_len)
+{
+    int force_filter = stbi_write_force_png_filter;
+    int ctype[5] = { -1, 0, 4, 2, 6 };
+    unsigned char sig[8] = { 137, 80, 78, 71, 13, 10, 26, 10 };
+    unsigned char *out, *o, *filt, *zlib;
+    signed char* line_buffer;
+    int j, zlen;
+
+    if (stride_bytes == 0)
+        stride_bytes = x * n;
+
+    if (force_filter >= 5) {
+        force_filter = -1;
+    }
+
+    filt = (unsigned char*)STBIW_MALLOC((x * n + 1) * y);
+    if (!filt)
+        return 0;
+    line_buffer = (signed char*)STBIW_MALLOC(x * n);
+    if (!line_buffer) {
+        STBIW_FREE(filt);
+        return 0;
+    }
+    for (j = 0; j < y; ++j) {
+        int filter_type;
+        if (force_filter > -1) {
+            filter_type = force_filter;
+            stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, force_filter, line_buffer);
+        } else { // Estimate the best filter by running through all of them:
+            int best_filter = 0, best_filter_val = 0x7fffffff, est, i;
+            for (filter_type = 0; filter_type < 5; filter_type++) {
+                stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, filter_type, line_buffer);
+
+                // Estimate the entropy of the line using this filter; the less, the better.
+                est = 0;
+                for (i = 0; i < x * n; ++i) {
+                    est += abs((signed char)line_buffer[i]);
+                }
+                if (est < best_filter_val) {
+                    best_filter_val = est;
+                    best_filter = filter_type;
+                }
+            }
+            if (filter_type != best_filter) { // If the last iteration already got us the best filter, don't redo it
+                stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, best_filter, line_buffer);
+                filter_type = best_filter;
+            }
+        }
+        // when we get here, filter_type contains the filter type, and line_buffer contains the data
+        filt[j * (x * n + 1)] = (unsigned char)filter_type;
+        STBIW_MEMMOVE(filt + j * (x * n + 1) + 1, line_buffer, x * n);
+    }
+    STBIW_FREE(line_buffer);
+    zlib = stbi_zlib_compress(filt, y * (x * n + 1), &zlen, stbi_write_png_compression_level);
+    STBIW_FREE(filt);
+    if (!zlib)
+        return 0;
+
+    // each tag requires 12 bytes of overhead
+    out = (unsigned char*)STBIW_MALLOC(8 + 12 + 13 + 12 + zlen + 12);
+    if (!out)
+        return 0;
+    *out_len = 8 + 12 + 13 + 12 + zlen + 12;
+
+    o = out;
+    STBIW_MEMMOVE(o, sig, 8);
+    o += 8;
+    stbiw__wp32(o, 13); // header length
+    stbiw__wptag(o, "IHDR");
+    stbiw__wp32(o, x);
+    stbiw__wp32(o, y);
+    *o++ = 8;
+    *o++ = STBIW_UCHAR(ctype[n]);
+    *o++ = 0;
+    *o++ = 0;
+    *o++ = 0;
+    stbiw__wpcrc(&o, 13);
+
+    stbiw__wp32(o, zlen);
+    stbiw__wptag(o, "IDAT");
+    STBIW_MEMMOVE(o, zlib, zlen);
+    o += zlen;
+    STBIW_FREE(zlib);
+    stbiw__wpcrc(&o, zlen);
+
+    stbiw__wp32(o, 0);
+    stbiw__wptag(o, "IEND");
+    stbiw__wpcrc(&o, 0);
+
+    STBIW_ASSERT(o == out + *out_len);
+
+    return out;
+}
+
+#    ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_png(char const* filename, int x, int y, int comp, void const* data, int stride_bytes)
+{
+    FILE* f;
+    int len;
+    unsigned char* png = stbi_write_png_to_mem((unsigned char const*)data, stride_bytes, x, y, comp, &len);
+    if (png == NULL)
+        return 0;
+
+    f = stbiw__fopen(filename, "wb");
+    if (!f) {
+        STBIW_FREE(png);
+        return 0;
+    }
+    fwrite(png, 1, len, f);
+    fclose(f);
+    STBIW_FREE(png);
+    return 1;
+}
+#    endif
+
+STBIWDEF int stbi_write_png_to_func(stbi_write_func* func, void* context, int x, int y, int comp, void const* data, int stride_bytes)
+{
+    int len;
+    unsigned char* png = stbi_write_png_to_mem((unsigned char const*)data, stride_bytes, x, y, comp, &len);
+    if (png == NULL)
+        return 0;
+    func(context, png, len);
+    STBIW_FREE(png);
+    return 1;
+}
+
+/* ***************************************************************************
+ *
+ * JPEG writer
+ *
+ * This is based on Jon Olick's jo_jpeg.cpp:
+ * public domain Simple, Minimalistic JPEG writer - http://www.jonolick.com/code.html
+ */
+
+static unsigned char const stbiw__jpg_ZigZag[] = { 0, 1, 5, 6, 14, 15, 27, 28, 2, 4, 7, 13, 16, 26, 29, 42, 3, 8, 12, 17, 25, 30, 41, 43, 9, 11, 18,
+    24, 31, 40, 44, 53, 10, 19, 23, 32, 39, 45, 52, 54, 20, 22, 33, 38, 46, 51, 55, 60, 21, 34, 37, 47, 50, 56, 59, 61, 35, 36, 48, 49, 57, 58, 62, 63 };
+
+static void stbiw__jpg_writeBits(stbi__write_context* s, int* bitBufP, int* bitCntP, unsigned short const* bs)
+{
+    int bitBuf = *bitBufP, bitCnt = *bitCntP;
+    bitCnt += bs[1];
+    bitBuf |= bs[0] << (24 - bitCnt);
+    while (bitCnt >= 8) {
+        unsigned char c = (bitBuf >> 16) & 255;
+        stbiw__putc(s, c);
+        if (c == 255) {
+            stbiw__putc(s, 0);
+        }
+        bitBuf <<= 8;
+        bitCnt -= 8;
+    }
+    *bitBufP = bitBuf;
+    *bitCntP = bitCnt;
+}
+
+static void stbiw__jpg_DCT(float* d0p, float* d1p, float* d2p, float* d3p, float* d4p, float* d5p, float* d6p, float* d7p)
+{
+    float d0 = *d0p, d1 = *d1p, d2 = *d2p, d3 = *d3p, d4 = *d4p, d5 = *d5p, d6 = *d6p, d7 = *d7p;
+    float z1, z2, z3, z4, z5, z11, z13;
+
+    float tmp0 = d0 + d7;
+    float tmp7 = d0 - d7;
+    float tmp1 = d1 + d6;
+    float tmp6 = d1 - d6;
+    float tmp2 = d2 + d5;
+    float tmp5 = d2 - d5;
+    float tmp3 = d3 + d4;
+    float tmp4 = d3 - d4;
+
+    // Even part
+    float tmp10 = tmp0 + tmp3; // phase 2
+    float tmp13 = tmp0 - tmp3;
+    float tmp11 = tmp1 + tmp2;
+    float tmp12 = tmp1 - tmp2;
+
+    d0 = tmp10 + tmp11; // phase 3
+    d4 = tmp10 - tmp11;
+
+    z1 = (tmp12 + tmp13) * 0.707106781f; // c4
+    d2 = tmp13 + z1;                     // phase 5
+    d6 = tmp13 - z1;
+
+    // Odd part
+    tmp10 = tmp4 + tmp5; // phase 2
+    tmp11 = tmp5 + tmp6;
+    tmp12 = tmp6 + tmp7;
+
+    // The rotator is modified from fig 4-8 to avoid extra negations.
+    z5 = (tmp10 - tmp12) * 0.382683433f; // c6
+    z2 = tmp10 * 0.541196100f + z5;      // c2-c6
+    z4 = tmp12 * 1.306562965f + z5;      // c2+c6
+    z3 = tmp11 * 0.707106781f;           // c4
+
+    z11 = tmp7 + z3; // phase 5
+    z13 = tmp7 - z3;
+
+    *d5p = z13 + z2; // phase 6
+    *d3p = z13 - z2;
+    *d1p = z11 + z4;
+    *d7p = z11 - z4;
+
+    *d0p = d0;
+    *d2p = d2;
+    *d4p = d4;
+    *d6p = d6;
+}
+
+static void stbiw__jpg_calcBits(int val, unsigned short bits[2])
+{
+    int tmp1 = val < 0 ? -val : val;
+    val = val < 0 ? val - 1 : val;
+    bits[1] = 1;
+    while (tmp1 >>= 1) {
+        ++bits[1];
+    }
+    bits[0] = val & ((1 << bits[1]) - 1);
+}
+
+static int stbiw__jpg_processDU(stbi__write_context* s, int* bitBuf, int* bitCnt, float* CDU, int du_stride, float* fdtbl, int DC, unsigned short const HTDC[256][2], unsigned short const HTAC[256][2])
+{
+    unsigned short const EOB[2] = { HTAC[0x00][0], HTAC[0x00][1] };
+    unsigned short const M16zeroes[2] = { HTAC[0xF0][0], HTAC[0xF0][1] };
+    int dataOff, i, j, n, diff, end0pos, x, y;
+    int DU[64];
+
+    // DCT rows
+    for (dataOff = 0, n = du_stride * 8; dataOff < n; dataOff += du_stride) {
+        stbiw__jpg_DCT(&CDU[dataOff], &CDU[dataOff + 1], &CDU[dataOff + 2], &CDU[dataOff + 3], &CDU[dataOff + 4], &CDU[dataOff + 5], &CDU[dataOff + 6], &CDU[dataOff + 7]);
+    }
+    // DCT columns
+    for (dataOff = 0; dataOff < 8; ++dataOff) {
+        stbiw__jpg_DCT(&CDU[dataOff], &CDU[dataOff + du_stride], &CDU[dataOff + du_stride * 2], &CDU[dataOff + du_stride * 3], &CDU[dataOff + du_stride * 4],
+            &CDU[dataOff + du_stride * 5], &CDU[dataOff + du_stride * 6], &CDU[dataOff + du_stride * 7]);
+    }
+    // Quantize/descale/zigzag the coefficients
+    for (y = 0, j = 0; y < 8; ++y) {
+        for (x = 0; x < 8; ++x, ++j) {
+            float v;
+            i = y * du_stride + x;
+            v = CDU[i] * fdtbl[j];
+            // DU[stbiw__jpg_ZigZag[j]] = (int)(v < 0 ? ceilf(v - 0.5f) : floorf(v + 0.5f));
+            // ceilf() and floorf() are C99, not C89, but I /think/ they're not needed here anyway?
+            DU[stbiw__jpg_ZigZag[j]] = (int)(v < 0 ? v - 0.5f : v + 0.5f);
+        }
+    }
+
+    // Encode DC
+    diff = DU[0] - DC;
+    if (diff == 0) {
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTDC[0]);
+    } else {
+        unsigned short bits[2];
+        stbiw__jpg_calcBits(diff, bits);
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTDC[bits[1]]);
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits);
+    }
+    // Encode ACs
+    end0pos = 63;
+    for (; (end0pos > 0) && (DU[end0pos] == 0); --end0pos) {
+    }
+    // end0pos = first element in reverse order !=0
+    if (end0pos == 0) {
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB);
+        return DU[0];
+    }
+    for (i = 1; i <= end0pos; ++i) {
+        int startpos = i;
+        int nrzeroes;
+        unsigned short bits[2];
+        for (; DU[i] == 0 && i <= end0pos; ++i) {
+        }
+        nrzeroes = i - startpos;
+        if (nrzeroes >= 16) {
+            int lng = nrzeroes >> 4;
+            int nrmarker;
+            for (nrmarker = 1; nrmarker <= lng; ++nrmarker)
+                stbiw__jpg_writeBits(s, bitBuf, bitCnt, M16zeroes);
+            nrzeroes &= 15;
+        }
+        stbiw__jpg_calcBits(DU[i], bits);
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTAC[(nrzeroes << 4) + bits[1]]);
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits);
+    }
+    if (end0pos != 63) {
+        stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB);
+    }
+    return DU[0];
+}
+
+static int stbi_write_jpg_core(stbi__write_context* s, int width, int height, int comp, void const* data, int quality)
+{
+    // Constants that don't pollute global namespace
+    static unsigned char const std_dc_luminance_nrcodes[] = { 0, 0, 1, 5, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0 };
+    static unsigned char const std_dc_luminance_values[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 };
+    static unsigned char const std_ac_luminance_nrcodes[] = { 0, 0, 2, 1, 3, 3, 2, 4, 3, 5, 5, 4, 4, 0, 0, 1, 0x7d };
+    static unsigned char const std_ac_luminance_values[] = {
+        0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12, 0x21, 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07, 0x22, 0x71, 0x14, 0x32, 0x81, 0x91, 0xa1, 0x08,
+        0x23, 0x42, 0xb1, 0xc1, 0x15, 0x52, 0xd1, 0xf0, 0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0a, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x25, 0x26, 0x27, 0x28,
+        0x29, 0x2a, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59,
+        0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89,
+        0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6,
+        0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xe1, 0xe2,
+        0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa
+    };
+    static unsigned char const std_dc_chrominance_nrcodes[] = { 0, 0, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
+    static unsigned char const std_dc_chrominance_values[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 };
+    static unsigned char const std_ac_chrominance_nrcodes[] = { 0, 0, 2, 1, 2, 4, 4, 3, 4, 7, 5, 4, 4, 0, 1, 2, 0x77 };
+    static unsigned char const std_ac_chrominance_values[] = {
+        0x00, 0x01, 0x02, 0x03, 0x11, 0x04, 0x05, 0x21, 0x31, 0x06, 0x12, 0x41, 0x51, 0x07, 0x61, 0x71, 0x13, 0x22, 0x32, 0x81, 0x08, 0x14, 0x42, 0x91,
+        0xa1, 0xb1, 0xc1, 0x09, 0x23, 0x33, 0x52, 0xf0, 0x15, 0x62, 0x72, 0xd1, 0x0a, 0x16, 0x24, 0x34, 0xe1, 0x25, 0xf1, 0x17, 0x18, 0x19, 0x1a, 0x26,
+        0x27, 0x28, 0x29, 0x2a, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58,
+        0x59, 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
+        0x88, 0x89, 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4,
+        0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda,
+        0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa
+    };
+    // Huffman tables
+    static unsigned short const YDC_HT[256][2] = { { 0, 2 }, { 2, 3 }, { 3, 3 }, { 4, 3 }, { 5, 3 }, { 6, 3 }, { 14, 4 }, { 30, 5 }, { 62, 6 }, { 126, 7 }, { 254, 8 }, { 510, 9 } };
+    static unsigned short const UVDC_HT[256][2] = { { 0, 2 }, { 1, 2 }, { 2, 2 }, { 6, 3 }, { 14, 4 }, { 30, 5 }, { 62, 6 }, { 126, 7 }, { 254, 8 }, { 510, 9 }, { 1022, 10 }, { 2046, 11 } };
+    static unsigned short const YAC_HT[256][2] = {
+        { 10, 4 }, { 0, 2 }, { 1, 2 }, { 4, 3 }, { 11, 4 }, { 26, 5 }, { 120, 7 }, { 248, 8 }, { 1014, 10 }, { 65410, 16 }, { 65411, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 12, 4 }, { 27, 5 }, { 121, 7 }, { 502, 9 }, { 2038, 11 }, { 65412, 16 }, { 65413, 16 }, { 65414, 16 }, { 65415, 16 }, { 65416, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 28, 5 }, { 249, 8 }, { 1015, 10 }, { 4084, 12 }, { 65417, 16 }, { 65418, 16 }, { 65419, 16 }, { 65420, 16 }, { 65421, 16 }, { 65422, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 58, 6 }, { 503, 9 }, { 4085, 12 }, { 65423, 16 }, { 65424, 16 }, { 65425, 16 }, { 65426, 16 }, { 65427, 16 }, { 65428, 16 }, { 65429, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 59, 6 }, { 1016, 10 }, { 65430, 16 }, { 65431, 16 }, { 65432, 16 }, { 65433, 16 }, { 65434, 16 }, { 65435, 16 }, { 65436, 16 }, { 65437, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 122, 7 }, { 2039, 11 }, { 65438, 16 }, { 65439, 16 }, { 65440, 16 }, { 65441, 16 }, { 65442, 16 }, { 65443, 16 }, { 65444, 16 }, { 65445, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 123, 7 }, { 4086, 12 }, { 65446, 16 }, { 65447, 16 }, { 65448, 16 }, { 65449, 16 }, { 65450, 16 }, { 65451, 16 }, { 65452, 16 }, { 65453, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 250, 8 }, { 4087, 12 }, { 65454, 16 }, { 65455, 16 }, { 65456, 16 }, { 65457, 16 }, { 65458, 16 }, { 65459, 16 }, { 65460, 16 }, { 65461, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 504, 9 }, { 32704, 15 }, { 65462, 16 }, { 65463, 16 }, { 65464, 16 }, { 65465, 16 }, { 65466, 16 }, { 65467, 16 }, { 65468, 16 }, { 65469, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 505, 9 }, { 65470, 16 }, { 65471, 16 }, { 65472, 16 }, { 65473, 16 }, { 65474, 16 }, { 65475, 16 }, { 65476, 16 }, { 65477, 16 }, { 65478, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 506, 9 }, { 65479, 16 }, { 65480, 16 }, { 65481, 16 }, { 65482, 16 }, { 65483, 16 }, { 65484, 16 }, { 65485, 16 }, { 65486, 16 }, { 65487, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 1017, 10 }, { 65488, 16 }, { 65489, 16 }, { 65490, 16 }, { 65491, 16 }, { 65492, 16 }, { 65493, 16 }, { 65494, 16 }, { 65495, 16 }, { 65496, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 1018, 10 }, { 65497, 16 }, { 65498, 16 }, { 65499, 16 }, { 65500, 16 }, { 65501, 16 }, { 65502, 16 }, { 65503, 16 }, { 65504, 16 }, { 65505, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 2040, 11 }, { 65506, 16 }, { 65507, 16 }, { 65508, 16 }, { 65509, 16 }, { 65510, 16 }, { 65511, 16 }, { 65512, 16 }, { 65513, 16 }, { 65514, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 65515, 16 }, { 65516, 16 }, { 65517, 16 }, { 65518, 16 }, { 65519, 16 }, { 65520, 16 }, { 65521, 16 }, { 65522, 16 }, { 65523, 16 }, { 65524, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 2041, 11 }, { 65525, 16 }, { 65526, 16 }, { 65527, 16 }, { 65528, 16 }, { 65529, 16 }, { 65530, 16 }, { 65531, 16 }, { 65532, 16 }, { 65533, 16 }, { 65534, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }
+    };
+    static unsigned short const UVAC_HT[256][2] = {
+        { 0, 2 }, { 1, 2 }, { 4, 3 }, { 10, 4 }, { 24, 5 }, { 25, 5 }, { 56, 6 }, { 120, 7 }, { 500, 9 }, { 1014, 10 }, { 4084, 12 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 11, 4 }, { 57, 6 }, { 246, 8 }, { 501, 9 }, { 2038, 11 }, { 4085, 12 }, { 65416, 16 }, { 65417, 16 }, { 65418, 16 }, { 65419, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 26, 5 }, { 247, 8 }, { 1015, 10 }, { 4086, 12 }, { 32706, 15 }, { 65420, 16 }, { 65421, 16 }, { 65422, 16 }, { 65423, 16 }, { 65424, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 27, 5 }, { 248, 8 }, { 1016, 10 }, { 4087, 12 }, { 65425, 16 }, { 65426, 16 }, { 65427, 16 }, { 65428, 16 }, { 65429, 16 }, { 65430, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 58, 6 }, { 502, 9 }, { 65431, 16 }, { 65432, 16 }, { 65433, 16 }, { 65434, 16 }, { 65435, 16 }, { 65436, 16 }, { 65437, 16 }, { 65438, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 59, 6 }, { 1017, 10 }, { 65439, 16 }, { 65440, 16 }, { 65441, 16 }, { 65442, 16 }, { 65443, 16 }, { 65444, 16 }, { 65445, 16 }, { 65446, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 121, 7 }, { 2039, 11 }, { 65447, 16 }, { 65448, 16 }, { 65449, 16 }, { 65450, 16 }, { 65451, 16 }, { 65452, 16 }, { 65453, 16 }, { 65454, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 122, 7 }, { 2040, 11 }, { 65455, 16 }, { 65456, 16 }, { 65457, 16 }, { 65458, 16 }, { 65459, 16 }, { 65460, 16 }, { 65461, 16 }, { 65462, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 249, 8 }, { 65463, 16 }, { 65464, 16 }, { 65465, 16 }, { 65466, 16 }, { 65467, 16 }, { 65468, 16 }, { 65469, 16 }, { 65470, 16 }, { 65471, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 503, 9 }, { 65472, 16 }, { 65473, 16 }, { 65474, 16 }, { 65475, 16 }, { 65476, 16 }, { 65477, 16 }, { 65478, 16 }, { 65479, 16 }, { 65480, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 504, 9 }, { 65481, 16 }, { 65482, 16 }, { 65483, 16 }, { 65484, 16 }, { 65485, 16 }, { 65486, 16 }, { 65487, 16 }, { 65488, 16 }, { 65489, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 505, 9 }, { 65490, 16 }, { 65491, 16 }, { 65492, 16 }, { 65493, 16 }, { 65494, 16 }, { 65495, 16 }, { 65496, 16 }, { 65497, 16 }, { 65498, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 506, 9 }, { 65499, 16 }, { 65500, 16 }, { 65501, 16 }, { 65502, 16 }, { 65503, 16 }, { 65504, 16 }, { 65505, 16 }, { 65506, 16 }, { 65507, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 2041, 11 }, { 65508, 16 }, { 65509, 16 }, { 65510, 16 }, { 65511, 16 }, { 65512, 16 }, { 65513, 16 }, { 65514, 16 }, { 65515, 16 }, { 65516, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 16352, 14 }, { 65517, 16 }, { 65518, 16 }, { 65519, 16 }, { 65520, 16 }, { 65521, 16 }, { 65522, 16 }, { 65523, 16 }, { 65524, 16 }, { 65525, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 },
+        { 1018, 10 }, { 32707, 15 }, { 65526, 16 }, { 65527, 16 }, { 65528, 16 }, { 65529, 16 }, { 65530, 16 }, { 65531, 16 }, { 65532, 16 }, { 65533, 16 }, { 65534, 16 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }
+    };
+    static int const YQT[] = { 16, 11, 10, 16, 24, 40, 51, 61, 12, 12, 14, 19, 26, 58, 60, 55, 14, 13, 16, 24, 40, 57, 69, 56, 14, 17, 22, 29, 51, 87, 80, 62, 18, 22,
+        37, 56, 68, 109, 103, 77, 24, 35, 55, 64, 81, 104, 113, 92, 49, 64, 78, 87, 103, 121, 120, 101, 72, 92, 95, 98, 112, 100, 103, 99 };
+    static int const UVQT[] = { 17, 18, 24, 47, 99, 99, 99, 99, 18, 21, 26, 66, 99, 99, 99, 99, 24, 26, 56, 99, 99, 99, 99, 99, 47, 66, 99, 99, 99, 99, 99, 99,
+        99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99, 99 };
+    static float const aasf[] = { 1.0f * 2.828427125f, 1.387039845f * 2.828427125f, 1.306562965f * 2.828427125f, 1.175875602f * 2.828427125f,
+        1.0f * 2.828427125f, 0.785694958f * 2.828427125f, 0.541196100f * 2.828427125f, 0.275899379f * 2.828427125f };
+
+    int row, col, i, k, subsample;
+    float fdtbl_Y[64], fdtbl_UV[64];
+    unsigned char YTable[64], UVTable[64];
+
+    if (!data || !width || !height || comp > 4 || comp < 1) {
+        return 0;
+    }
+
+    quality = quality ? quality : 90;
+    subsample = quality <= 90 ? 1 : 0;
+    quality = quality < 1 ? 1 : quality > 100 ? 100
+                                              : quality;
+    quality = quality < 50 ? 5000 / quality : 200 - quality * 2;
+
+    for (i = 0; i < 64; ++i) {
+        int uvti, yti = (YQT[i] * quality + 50) / 100;
+        YTable[stbiw__jpg_ZigZag[i]] = (unsigned char)(yti < 1 ? 1 : yti > 255 ? 255
+                                                                               : yti);
+        uvti = (UVQT[i] * quality + 50) / 100;
+        UVTable[stbiw__jpg_ZigZag[i]] = (unsigned char)(uvti < 1 ? 1 : uvti > 255 ? 255
+                                                                                  : uvti);
+    }
+
+    for (row = 0, k = 0; row < 8; ++row) {
+        for (col = 0; col < 8; ++col, ++k) {
+            fdtbl_Y[k] = 1 / (YTable[stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]);
+            fdtbl_UV[k] = 1 / (UVTable[stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]);
+        }
+    }
+
+    // Write Headers
+    {
+        static unsigned char const head0[] = { 0xFF, 0xD8, 0xFF, 0xE0, 0, 0x10, 'J', 'F', 'I', 'F', 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 0xFF, 0xDB, 0, 0x84, 0 };
+        static unsigned char const head2[] = { 0xFF, 0xDA, 0, 0xC, 3, 1, 0, 2, 0x11, 3, 0x11, 0, 0x3F, 0 };
+        unsigned char const head1[] = { 0xFF, 0xC0, 0, 0x11, 8, (unsigned char)(height >> 8), STBIW_UCHAR(height), (unsigned char)(width >> 8), STBIW_UCHAR(width),
+            3, 1, (unsigned char)(subsample ? 0x22 : 0x11), 0, 2, 0x11, 1, 3, 0x11, 1, 0xFF, 0xC4, 0x01, 0xA2, 0 };
+        s->func(s->context, (void*)head0, sizeof(head0));
+        s->func(s->context, (void*)YTable, sizeof(YTable));
+        stbiw__putc(s, 1);
+        s->func(s->context, UVTable, sizeof(UVTable));
+        s->func(s->context, (void*)head1, sizeof(head1));
+        s->func(s->context, (void*)(std_dc_luminance_nrcodes + 1), sizeof(std_dc_luminance_nrcodes) - 1);
+        s->func(s->context, (void*)std_dc_luminance_values, sizeof(std_dc_luminance_values));
+        stbiw__putc(s, 0x10); // HTYACinfo
+        s->func(s->context, (void*)(std_ac_luminance_nrcodes + 1), sizeof(std_ac_luminance_nrcodes) - 1);
+        s->func(s->context, (void*)std_ac_luminance_values, sizeof(std_ac_luminance_values));
+        stbiw__putc(s, 1); // HTUDCinfo
+        s->func(s->context, (void*)(std_dc_chrominance_nrcodes + 1), sizeof(std_dc_chrominance_nrcodes) - 1);
+        s->func(s->context, (void*)std_dc_chrominance_values, sizeof(std_dc_chrominance_values));
+        stbiw__putc(s, 0x11); // HTUACinfo
+        s->func(s->context, (void*)(std_ac_chrominance_nrcodes + 1), sizeof(std_ac_chrominance_nrcodes) - 1);
+        s->func(s->context, (void*)std_ac_chrominance_values, sizeof(std_ac_chrominance_values));
+        s->func(s->context, (void*)head2, sizeof(head2));
+    }
+
+    // Encode 8x8 macroblocks
+    {
+        static unsigned short const fillBits[] = { 0x7F, 7 };
+        int DCY = 0, DCU = 0, DCV = 0;
+        int bitBuf = 0, bitCnt = 0;
+        // comp == 2 is grey+alpha (alpha is ignored)
+        int ofsG = comp > 2 ? 1 : 0, ofsB = comp > 2 ? 2 : 0;
+        unsigned char const* dataR = (unsigned char const*)data;
+        unsigned char const* dataG = dataR + ofsG;
+        unsigned char const* dataB = dataR + ofsB;
+        int x, y, pos;
+        if (subsample) {
+            for (y = 0; y < height; y += 16) {
+                for (x = 0; x < width; x += 16) {
+                    float Y[256], U[256], V[256];
+                    for (row = y, pos = 0; row < y + 16; ++row) {
+                        // row >= height => use last input row
+                        int clamped_row = (row < height) ? row : height - 1;
+                        int base_p = (stbi__flip_vertically_on_write ? (height - 1 - clamped_row) : clamped_row) * width * comp;
+                        for (col = x; col < x + 16; ++col, ++pos) {
+                            // if col >= width => use pixel from last input column
+                            int p = base_p + ((col < width) ? col : (width - 1)) * comp;
+                            float r = dataR[p], g = dataG[p], b = dataB[p];
+                            Y[pos] = +0.29900f * r + 0.58700f * g + 0.11400f * b - 128;
+                            U[pos] = -0.16874f * r - 0.33126f * g + 0.50000f * b;
+                            V[pos] = +0.50000f * r - 0.41869f * g - 0.08131f * b;
+                        }
+                    }
+                    DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y + 0, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+                    DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y + 8, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+                    DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y + 128, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+                    DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y + 136, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+
+                    // subsample U,V
+                    {
+                        float subU[64], subV[64];
+                        int yy, xx;
+                        for (yy = 0, pos = 0; yy < 8; ++yy) {
+                            for (xx = 0; xx < 8; ++xx, ++pos) {
+                                int j = yy * 32 + xx * 2;
+                                subU[pos] = (U[j + 0] + U[j + 1] + U[j + 16] + U[j + 17]) * 0.25f;
+                                subV[pos] = (V[j + 0] + V[j + 1] + V[j + 16] + V[j + 17]) * 0.25f;
+                            }
+                        }
+                        DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subU, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT);
+                        DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subV, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT);
+                    }
+                }
+            }
+        } else {
+            for (y = 0; y < height; y += 8) {
+                for (x = 0; x < width; x += 8) {
+                    float Y[64], U[64], V[64];
+                    for (row = y, pos = 0; row < y + 8; ++row) {
+                        // row >= height => use last input row
+                        int clamped_row = (row < height) ? row : height - 1;
+                        int base_p = (stbi__flip_vertically_on_write ? (height - 1 - clamped_row) : clamped_row) * width * comp;
+                        for (col = x; col < x + 8; ++col, ++pos) {
+                            // if col >= width => use pixel from last input column
+                            int p = base_p + ((col < width) ? col : (width - 1)) * comp;
+                            float r = dataR[p], g = dataG[p], b = dataB[p];
+                            Y[pos] = +0.29900f * r + 0.58700f * g + 0.11400f * b - 128;
+                            U[pos] = -0.16874f * r - 0.33126f * g + 0.50000f * b;
+                            V[pos] = +0.50000f * r - 0.41869f * g - 0.08131f * b;
+                        }
+                    }
+
+                    DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y, 8, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+                    DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, U, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT);
+                    DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, V, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT);
+                }
+            }
+        }
+
+        // Do the bit alignment of the EOI marker
+        stbiw__jpg_writeBits(s, &bitBuf, &bitCnt, fillBits);
+    }
+
+    // EOI
+    stbiw__putc(s, 0xFF);
+    stbiw__putc(s, 0xD9);
+
+    return 1;
+}
+
+STBIWDEF int stbi_write_jpg_to_func(stbi_write_func* func, void* context, int x, int y, int comp, void const* data, int quality)
+{
+    stbi__write_context s = { 0 };
+    stbi__start_write_callbacks(&s, func, context);
+    return stbi_write_jpg_core(&s, x, y, comp, (void*)data, quality);
+}
+
+#    ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_jpg(char const* filename, int x, int y, int comp, void const* data, int quality)
+{
+    stbi__write_context s = { 0 };
+    if (stbi__start_write_file(&s, filename)) {
+        int r = stbi_write_jpg_core(&s, x, y, comp, data, quality);
+        stbi__end_write_file(&s);
+        return r;
+    } else
+        return 0;
+}
+#    endif
+
+#endif // STB_IMAGE_WRITE_IMPLEMENTATION
+
+/* Revision history
+      1.16  (2021-07-11)
+             make Deflate code emit uncompressed blocks when it would otherwise expand
+             support writing BMPs with alpha channel
+      1.15  (2020-07-13) unknown
+      1.14  (2020-02-02) updated JPEG writer to downsample chroma channels
+      1.13
+      1.12
+      1.11  (2019-08-11)
+
+      1.10  (2019-02-07)
+             support utf8 filenames in Windows; fix warnings and platform ifdefs
+      1.09  (2018-02-11)
+             fix typo in zlib quality API, improve STB_I_W_STATIC in C++
+      1.08  (2018-01-29)
+             add stbi__flip_vertically_on_write, external zlib, zlib quality, choose PNG filter
+      1.07  (2017-07-24)
+             doc fix
+      1.06 (2017-07-23)
+             writing JPEG (using Jon Olick's code)
+      1.05   ???
+      1.04 (2017-03-03)
+             monochrome BMP expansion
+      1.03   ???
+      1.02 (2016-04-02)
+             avoid allocating large structures on the stack
+      1.01 (2016-01-16)
+             STBIW_REALLOC_SIZED: support allocators with no realloc support
+             avoid race-condition in crc initialization
+             minor compile issues
+      1.00 (2015-09-14)
+             installable file IO function
+      0.99 (2015-09-13)
+             warning fixes; TGA rle support
+      0.98 (2015-04-08)
+             added STBIW_MALLOC, STBIW_ASSERT etc
+      0.97 (2015-01-18)
+             fixed HDR asserts, rewrote HDR rle logic
+      0.96 (2015-01-17)
+             add HDR output
+             fix monochrome BMP
+      0.95 (2014-08-17)
+             add monochrome TGA output
+      0.94 (2014-05-31)
+             rename private functions to avoid conflicts with stb_image.h
+      0.93 (2014-05-27)
+             warning fixes
+      0.92 (2010-08-01)
+             casts to unsigned char to fix warnings
+      0.91 (2010-07-17)
+             first public release
+      0.90   first internal release
+*/
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
diff --git a/src/libtemple/ioport.h b/src/libtemple/ioport.h
new file mode 100644
index 0000000..69e69e3
--- /dev/null
+++ b/src/libtemple/ioport.h
@@ -0,0 +1,6 @@
+u8 ioport_read_u8(u16 address);
+u16 ioport_read_u16(u16 address);
+u32 ioport_read_u32(u16 address);
+void ioport_write_u8(u16 address, u8 value);
+void ioport_write_u16(u16 address, u16 value);
+void ioport_write_u32(u16 address, u32 value);
diff --git a/src/libtemple/libtemple.cpp b/src/libtemple/libtemple.cpp
new file mode 100644
index 0000000..2311264
--- /dev/null
+++ b/src/libtemple/libtemple.cpp
@@ -0,0 +1,81 @@
+unsigned long ioport_read_u8(unsigned short address) { return 0; }
+
+unsigned long ioport_read_u16(unsigned short address) { return 0; }
+
+unsigned long ioport_read_u32(unsigned short address) { return 0; }
+
+void ioport_write_u8(unsigned short address, unsigned char value) { }
+
+void ioport_write_u16(unsigned short address, unsigned short value) { }
+
+void ioport_write_u32(unsigned short address, unsigned int value) { }
+
+bool os_blink(char const* frequency_as_string) { return 0; }
+
+unsigned long os_call(unsigned long function_name, unsigned long arg) { return 0; }
+
+unsigned int os_device_calloc(unsigned int size) { return 0; }
+
+void os_exit() { }
+
+char const* os_file_picker(char const* path, char const* glob) { return 0; }
+
+char const* os_files_list(char const* path) { return 0; }
+
+bool os_is_vm() { return 0; }
+
+bool os_path_exists(char const* path) { return 0; }
+
+void os_pc_speaker(char const* frequency_as_string) { }
+
+unsigned long os_random() { return 0; }
+
+unsigned long os_read_entire_file(char const* filename, long* size)
+{
+    return 0;
+}
+
+void os_screenshot() { }
+
+char const* os_to_uppercase(char const* input_string) { return 0; }
+
+void os_write_entire_file(char const* filename, unsigned char* buffer,
+    long size) { }
+
+long pci_find(long class_code) { return 0; }
+
+unsigned long pci_read_u8(long bus, long device, long fun, long offset)
+{
+    return 0;
+}
+
+unsigned long pci_read_u16(long bus, long device, long fun, long offset)
+{
+    return 0;
+}
+
+unsigned long pci_read_u32(long bus, long device, long fun, long offset)
+{
+    return 0;
+}
+
+void pci_write_u8(long bus, long device, long fun, long offset,
+    unsigned char value) { }
+
+void pci_write_u16(long bus, long device, long fun, long offset,
+    unsigned short value) { }
+
+void pci_write_u32(long bus, long device, long fun, long offset,
+    unsigned int value) { }
+
+void time_busy(long duration) { }
+
+long time_jiffies() { return 0; }
+
+long time_now() { return 0; }
+
+void time_sleep(long duration) { }
+
+void* truetype_malloc(int size) { return 0; }
+
+void truetype_free(void* ptr) { }
\ No newline at end of file
diff --git a/src/libtemple/os.h b/src/libtemple/os.h
new file mode 100644
index 0000000..24e489e
--- /dev/null
+++ b/src/libtemple/os.h
@@ -0,0 +1,15 @@
+bool os_blink(char const* frequency_as_string);
+unsigned long os_call(unsigned long function_name, unsigned long arg);
+unsigned int os_device_calloc(unsigned int size);
+void os_exit();
+char const* os_file_picker(char const* path, char const* glob);
+char const* os_files_list(char const* path);
+bool os_is_vm();
+bool os_path_exists(char const* path);
+void os_pc_speaker(char const* frequency_as_string);
+unsigned long os_random();
+u8* os_read_entire_file(char const* filename, i64* size);
+void os_screenshot();
+char const* os_to_uppercase(char const* input_string);
+void os_write_entire_file(char const* filename, unsigned char* buffer,
+    i64 size);
\ No newline at end of file
diff --git a/src/libtemple/pci.h b/src/libtemple/pci.h
new file mode 100644
index 0000000..33d7fea
--- /dev/null
+++ b/src/libtemple/pci.h
@@ -0,0 +1,10 @@
+long pci_find(long class_code);
+unsigned long pci_read_u8(long bus, long device, long fun, long offset);
+unsigned long pci_read_u16(long bus, long device, long fun, long offset);
+unsigned long pci_read_u32(long bus, long device, long fun, long offset);
+void pci_write_u8(long bus, long device, long fun, long offset,
+    unsigned char value);
+void pci_write_u16(long bus, long device, long fun, long offset,
+    unsigned short value);
+void pci_write_u32(long bus, long device, long fun, long offset,
+    unsigned int value);
\ No newline at end of file
diff --git a/src/libtemple/time.h b/src/libtemple/time.h
new file mode 100644
index 0000000..e2fe7c4
--- /dev/null
+++ b/src/libtemple/time.h
@@ -0,0 +1,4 @@
+void time_busy(i64 duration);
+i64 time_jiffies();
+i64 time_now();
+void time_sleep(i64 duration);
\ No newline at end of file
diff --git a/src/net/devices/virtio.h b/src/net/devices/virtio.h
new file mode 100644
index 0000000..989e1d0
--- /dev/null
+++ b/src/net/devices/virtio.h
@@ -0,0 +1,28 @@
+struct virtio_queue_buf {
+    u64 address;
+    u32 length;
+    u16 flags;
+    u16 next;
+};
+struct virtio_avail {
+    u16 flags;
+    u16 index;
+    u16 ring[256];
+    u16 int_index;
+};
+struct virtio_used_item {
+    u32 index;
+    u32 length;
+};
+struct virtio_used {
+    u16 flags;
+    u16 index;
+    virtio_used_item ring[256];
+    u16 int_index;
+};
+struct virtio_queue {
+    virtio_queue_buf buffers[256];
+    virtio_avail available;
+    u8 padding[3578];
+    virtio_used used;
+};
\ No newline at end of file
diff --git a/src/net/devices/virtio.jakt b/src/net/devices/virtio.jakt
new file mode 100644
index 0000000..647af28
--- /dev/null
+++ b/src/net/devices/virtio.jakt
@@ -0,0 +1,143 @@
+import relative parent::os::os { OS }
+import relative parent::os::pci { PCI, PCIDevice }
+
+enum VirtIOConfig: u8 {
+    acknowledge = 1
+    driver = 2
+    driver_ok = 4
+}
+
+enum VirtIOReg: u16 {
+    host_features = 0
+    guest_features = 4
+    queue_page_frame_number = 8
+    queue_size = 12
+    queue_select = 14
+    queue_notify = 16
+    status = 18
+    isr = 19
+    config = 20
+}
+
+class VirtIO {
+    public pci_device: PCIDevice
+    public rq_index: i64
+    public rq_size: u16
+    public rq: u32
+    public tq_size: u16
+    public tq: u32
+    public fn rx_frame(mut this) throws -> [u8] {
+        mut frame: [u8] = []
+        mut queue_notify: bool = false
+        unsafe {
+          cpp {
+"
+#include <../../src/net/devices/virtio.h>
+virtio_queue *rq = (virtio_queue*)this->rq;
+i64 i = this->rq_index;
+i64 used_index = rq->used.index;
+if (used_index < i)
+  used_index += 0x10000;
+if (used_index && i != used_index) {
+  virtio_used_item* item = rq->used.ring;
+  u8* buffer = (u8*)rq->buffers[item[i % 256].index + 1].address;
+  i64 length = item[i % 256].length - 10;
+  for (i64 j = 0; j < length; j++)
+    frame.push(buffer[j]);
+  this->rq_index = used_index % 0x10000;
+  rq->available.index++;
+  queue_notify = true;
+}
+"            
+          }
+        }
+        if queue_notify {
+          .pci_device.io_write_u16(offset: VirtIOReg::queue_notify as! u16, value: 0)
+        }
+        return frame
+    }
+    public fn tx_frame(mut this, anon mut data: [u8]) throws {
+        mut size = data.size()
+        unsafe {
+            cpp {
+"
+#include <../../src/net/devices/virtio.h>
+virtio_queue *tq = (virtio_queue*)this->tq;
+int tq_idx = tq->available.index % 256;
+int tq_idx2 = tq_idx % 128;
+memset((u8*)tq->buffers[tq_idx2 * 2].address, 0, 10);
+u8 *buffer = (u8*)tq->buffers[(tq_idx2 * 2) + 1].address;
+    for (int i = 0; i < size; i++)
+        buffer[i] = data[i];
+tq->buffers[tq_idx2 * 2].length = 10;
+tq->buffers[tq_idx2 * 2].flags = 1;
+tq->buffers[tq_idx2 * 2].next = (tq_idx2 * 2) + 1;
+tq->buffers[(tq_idx2 * 2) + 1].length = size;
+tq->buffers[(tq_idx2 * 2) + 1].flags = 0;
+tq->buffers[(tq_idx2 * 2) + 1].next = 0;
+tq->available.ring[tq_idx] = tq_idx2 * 2;
+tq->available.index++;
+"
+            }
+        }
+        .pci_device.io_write_u16(offset: VirtIOReg::queue_notify as! u16, value: 1)
+    }
+    fn reset_device(this) {
+        .pci_device.io_write_u8(offset: VirtIOReg::status as! u16, value: 0)
+    }
+    fn found_driver(this) throws {
+        .pci_device.io_write_u8(offset: VirtIOReg::status as! u16,
+            value: .pci_device.io_read_u8(VirtIOReg::status as! u16) | VirtIOConfig::acknowledge as! u8 | VirtIOConfig::driver as! u8)
+    }
+    fn setup_rx_queue(mut this) throws {
+        .pci_device.io_write_u16(offset: VirtIOReg::queue_select as! u16, value: 0)
+        .rq_size = .pci_device.io_read_u16(VirtIOReg::queue_size as! u16)
+        .rq = OS::device_calloc(16384)
+        .pci_device.io_write_u32(offset: VirtIOReg::queue_page_frame_number as! u16, value: .rq / 4096)
+    }
+    fn setup_tx_queue(mut this) throws {
+        .pci_device.io_write_u16(offset: VirtIOReg::queue_select as! u16, value: 1)
+        .tq_size = .pci_device.io_read_u16(VirtIOReg::queue_size as! u16)
+        .tq = OS::device_calloc(16384)
+        .pci_device.io_write_u32(offset: VirtIOReg::queue_page_frame_number as! u16, value: .tq / 4096)
+    }
+    fn init_queue_buffers(this) {
+        unsafe {
+            cpp {
+"
+#include <../../src/net/devices/virtio.h>
+virtio_queue *rq = (virtio_queue*)this->rq;
+virtio_queue *tq = (virtio_queue*)this->tq;
+for (int i = 0; i < 128; i++) {
+    rq->buffers[i * 2].address = (u64)calloc(1, 16);
+    rq->buffers[i * 2].length = 10;
+    rq->buffers[i * 2].flags = 3;
+    rq->buffers[i * 2].next = (i * 2) + 1;
+    rq->buffers[(i * 2) + 1].address = (u64)calloc(1, 2048);
+    rq->buffers[(i * 2) + 1].length = 2048;
+    rq->buffers[(i * 2) + 1].flags = 2;
+    rq->buffers[(i * 2) + 1].next = 0;
+    rq->available.ring[i] = i * 2;
+    rq->available.ring[i + 128] = i * 2;
+    tq->buffers[i * 2].address = (u64)calloc(1, 16);
+    tq->buffers[(i * 2) + 1].address = (u64)calloc(1, 2048);
+}
+rq->available.index = 1;
+"
+            }
+        }
+    }
+    fn init_ok(this) throws {
+        .pci_device.io_write_u8(offset: VirtIOReg::status as! u16, 
+          value: .pci_device.io_read_u8(VirtIOReg::status as! u16) | VirtIOConfig::driver_ok as! u8)
+        .pci_device.io_write_u16(offset: VirtIOReg::queue_notify as! u16, value: 0)
+    }
+    public fn init(mut this) throws {
+        .reset_device()
+        .found_driver()
+        .setup_rx_queue()
+        .setup_tx_queue()
+        .init_queue_buffers()
+        .init_ok()
+    }
+}
\ No newline at end of file
diff --git a/src/net/lib/json.jakt b/src/net/lib/json.jakt
new file mode 100644
index 0000000..5cdd177
--- /dev/null
+++ b/src/net/lib/json.jakt
@@ -0,0 +1,350 @@
+/// Expect:
+/// - output: "JsonValue::JsonArray([JsonValue::Object([\"id\": JsonValue::Number(0.5), \"displayName\": JsonValue::JsonString(\"Air\"), \"name\": JsonValue::JsonString(\"air\"), \"hardness\": JsonValue::Number(3.9), \"resistance\": JsonValue::Number(0), \"minStateId\": JsonValue::Number(0), \"maxStateId\": JsonValue::Number(0), \"states\": JsonValue::JsonArray([])])])\n"
+
+enum JsonValue {
+    Null
+    Bool(bool)
+    Number(f64)
+    // FIXME: This variant should be called String
+    JsonString(String)
+    // FIXME: This variant should be called Array
+    JsonArray([JsonValue])
+    Object([String:JsonValue])
+}
+
+fn is_whitespace(anon c: u8) -> bool {
+    return match c {
+        b'\t' | b'\n' | b'\r' | b' ' => true
+        else => false
+    }
+}
+
+class JsonParser {
+    input: String
+    index: usize
+
+    public fn construct(input: String) throws -> JsonParser {
+        return JsonParser(input, index: 0)
+    }
+
+    fn eof(this) -> bool {
+        return .index >= .input.length()
+    }
+
+    public fn parse(mut this) throws -> JsonValue {
+        // FIXME: Jakt::JsonParser ignores trailing whitespace for some reason.
+        let value = .parse_helper()
+        if not .eof() {
+            // FIXME: "Didn't consume all input"
+            throw Error::from_errno(9000)
+        }
+        return value
+    }
+
+    fn skip_whitespace(mut this) {
+        while not .eof() {
+            if not is_whitespace(.input.byte_at(.index)) {
+                break
+            }
+            .index++
+        }
+    }
+
+    fn consume_and_unescape_string(mut this) throws -> String {
+        if not .consume_specific(b'"') {
+            // FIXME: "Expected '"'
+            throw Error::from_errno(9007)
+        }
+
+        mut builder = StringBuilder::create()
+
+        loop {
+            mut ch = 0u8
+            mut peek_index = .index
+            while peek_index < .input.length() {
+                ch = .input.byte_at(peek_index)
+                if ch == b'"' or ch == b'\\' {
+                    break
+                }
+                // FIXME: This is is_ascii_c0_control()
+                if ch < 0x20 {
+                    // FIXME: "Error while parsing string"
+                    throw Error::from_errno(9008)
+                }
+                peek_index++
+            }
+
+            while peek_index != .index {
+                builder.append(.input.byte_at(.index))
+                .index++
+            }
+
+            if .eof() {
+                break
+            }
+
+            if ch == b'"' {
+                break
+            }
+
+            if ch != b'\\' {
+                builder.append(.consume())
+                continue
+            }
+
+            .ignore()
+
+            match .peek() {
+                b'"' | b'/' | b'\\' | b'n' | b'r' | b't' | b'b' | b'f' => {
+                    let ch = .consume()
+                    builder.append(match ch {
+                        b'n' => b'\n'
+                        b'r' => b'\r'
+                        b't' => b'\t'
+                        b'b' => b'\b'
+                        b'f' => b'\f'
+                        else => ch
+                    })
+                }
+                b'u' => {
+                    eprintln("FIXME: Implement unicode literals")
+                    abort()
+                }
+                else => {
+                    // FIXME: "Error while parsing string"
+                    throw Error::from_errno(9009)
+                }
+            }
+        }
+
+        if not .consume_specific(b'"') {
+            // FIXME: "Expected '"'"
+            throw Error::from_errno(9010)
+        }
+
+        return builder.to_string()
+    }
+
+    fn ignore(mut this) {
+        .index++
+    }
+
+    fn peek(this) -> u8 {
+        if .eof() {
+            return 0
+        }
+        return .input.byte_at(.index)
+    }
+
+    fn consume(mut this) -> u8 {
+        let ch = .peek()
+        .index++
+        return ch
+    }
+
+    fn consume_specific(mut this, anon expected: u8) -> bool {
+        if .peek() != expected {
+            return false
+        }
+        .index++
+        return true
+    }
+
+    fn parse_helper(mut this) throws -> JsonValue {
+        .skip_whitespace()
+        return match .peek() {
+            b'{' => .parse_object()
+            b'[' => .parse_array()
+            b'"' => .parse_string()
+            b'-' => .parse_number()
+            b'0' | b'1' | b'2' | b'3' | b'4' | b'5' | b'6' | b'7' | b'8' | b'9' => .parse_number()
+            b'f' => .parse_false()
+            b't' => .parse_true()
+            b'n' => .parse_null()
+            else => .parse_failure(error_message: "Unexpected character")
+        }
+    }
+
+    fn parse_failure(this, error_message: String) throws -> JsonValue {
+        throw Error::from_errno(9001)
+    }
+
+    fn parse_array(mut this) throws -> JsonValue {
+        mut array: [JsonValue] = []
+        if (not .consume_specific(b'[')) {
+            // Expected '['
+            throw Error::from_errno(9014)
+        }
+        loop {
+            .skip_whitespace()
+            if .peek() == b']' {
+                break
+            }
+            array.push(.parse_helper())
+            .skip_whitespace()
+            if .peek() == b']' {
+                break
+            }
+            if not .consume_specific(b',') {
+                // Expected ','
+                throw Error::from_errno(9014)
+            }
+            .skip_whitespace()
+            if .peek() == b']' {
+                // Unexpected ']'
+                throw Error::from_errno(9014)
+            }
+        }
+        if not .consume_specific(b']') {
+            // Expected ']'
+            throw Error::from_errno(9015)
+        }
+        return JsonValue::JsonArray(array)
+    }
+
+    fn parse_object(mut this) throws -> JsonValue {
+        if not .consume_specific(b'{') {
+            // FIXME: "Expected '{'"
+            throw Error::from_errno(9002)
+        }
+
+        mut values: [String:JsonValue] = [:]
+
+        loop {
+            .skip_whitespace()
+            if .peek() == b'}' {
+                break
+            }
+            .skip_whitespace()
+            let key = .consume_and_unescape_string()
+            .skip_whitespace()
+            if not .consume_specific(b':') {
+                // FIXME: "Expected ':'"
+                throw Error::from_errno(9003)
+            }
+            .skip_whitespace()
+            let value = .parse_helper()
+            // FIXME: This should say `values[key] = value`, but the compiler doesn't wrap it in TRY()
+            values.set(key, value)
+            .skip_whitespace()
+            if .peek() == b'}' {
+                break
+            }
+            if not .consume_specific(b',') {
+                // FIXME: "Expected ','"
+                throw Error::from_errno(9004)
+            }
+            .skip_whitespace()
+            if .peek() == b'}' {
+                // FIXME: "Unexpected '}'"
+                throw Error::from_errno(9005)
+            }
+        }
+        if not .consume_specific(b'}') {
+            // FIXME: "Expected '}'"
+            throw Error::from_errno(9006)
+        }
+        return JsonValue::Object(values)
+    }
+
+    fn char_to_f64(anon num: u8) throws -> f64 {
+        // FIXME 1: Shouldn't need this function at all
+        // FIXME 2: Shouldn't need return in else branch
+        return match num {
+            0u8 => 0.0
+            1u8 => 1.0
+            2u8 => 2.0
+            3u8 => 3.0
+            4u8 => 4.0
+            5u8 => 5.0
+            6u8 => 6.0
+            7u8 => 7.0
+            8u8 => 8.0
+            9u8 => 9.0
+            else => {
+                // FIXME: "Unexpected number"
+                throw Error::from_errno(9017)
+            }
+        }
+    }
+
+    fn parse_number(mut this) throws -> JsonValue {
+        // FIXME: This implementation doesn't match JsonParser.cpp
+        let is_negative = .consume_specific(b'-')
+        mut decimal_start_index: usize? = None
+
+        mut value = 0.0
+
+        while not .eof() {
+            let ch = .peek()
+            if ch == b'.' {
+                if decimal_start_index.has_value() {
+                    // FIXME: "Unexpected '.'"
+                    throw Error::from_errno(9016)
+                }
+                decimal_start_index = .index++
+                continue
+            } else if not (ch >= b'0' and ch <= b'9') {
+                break
+            }
+
+            if not decimal_start_index.has_value() {
+                value *= 10.0
+                value += char_to_f64(ch - b'0')
+            } else {
+                mut num = char_to_f64(ch - b'0')
+                // FIXME: This should really be: `value += pow(10, -decimal_place)*num`, but: there's no pow function and you can't multiply float by usize
+                let decimal_place = .index - decimal_start_index.value()
+                for i in 0..decimal_place {
+                    num /= 10.0
+                }
+                value += num
+            }
+            .index++
+        }
+
+        if is_negative {
+            value *= -1.0
+        }
+
+        return JsonValue::Number(value)
+    }
+
+    fn parse_string(mut this) throws -> JsonValue {
+        return JsonValue::JsonString(.consume_and_unescape_string())
+    }
+
+    fn parse_false(mut this) throws -> JsonValue {
+        if (.consume() != b'f' or .consume() != b'a' or .consume() != b'l' or .consume() != b's' or .consume() != b'e') {
+            // FIXME: "Expected 'false'"
+            throw Error::from_errno(9011)
+        }
+        return JsonValue::Bool(false)
+    }
+
+    fn parse_true(mut this) throws -> JsonValue {
+        if (.consume() != b't' or .consume() != b'r' or .consume() != b'u' or .consume() != b'e') {
+            // FIXME: "Expected 'true'"
+            throw Error::from_errno(9012)
+        }
+        return JsonValue::Bool(true)
+    }
+
+    fn parse_null(mut this) throws -> JsonValue {
+        if (.consume() != b'n' or .consume() != b'u' or .consume() != b'l' or .consume() != b'l') {
+            // FIXME: "Expected 'null'"
+            throw Error::from_errno(9013)
+        }
+        return JsonValue::Null
+    }
+}
+
+// fn parse_json(input: String) throws -> JsonValue {
+//     mut parser = JsonParser::construct(input)
+//     return parser.parse()
+// }
+// 
+// fn main() {
+//     let value = parse_json(input: "[{\"id\":0.5,\"displayName\":\"Air\",\"name\":\"air\",\"hardness\":3.9,\"resistance\":0,\"minStateId\":0,\"maxStateId\":0,\"states\":[]}]")
+//     println("{}", value)
+// }
\ No newline at end of file
diff --git a/src/net/lib/util.jakt b/src/net/lib/util.jakt
new file mode 100644
index 0000000..434158d
--- /dev/null
+++ b/src/net/lib/util.jakt
@@ -0,0 +1,147 @@
+import relative parent::os::os { OS }
+
+struct Util {
+    fn get_address_u32_from_ipv4_u8_array(anon array: [u8]) -> u32 {
+        if array.size() != 4 {
+            return 0
+        }
+        mut address: u32 = (array[3] as! u32 & 0xff) as! u32
+        address += ((array[2] as! u32 & 0xff) << 8) as! u32
+        address += ((array[1] as! u32 & 0xff) << 16) as! u32
+        address += ((array[0] as! u32 & 0xff) << 24) as! u32
+        return address
+    }
+    fn get_hexadecimal_string_from_ipv4_u8_array(anon array: [u8]) throws -> String {
+        mut s = StringBuilder::create()
+        unsafe {
+            cpp {
+                "char *chars = (char*)calloc(32, 1);
+                sprintf(chars, \"%02x%02x%02x%02x\", array[0], array[1], array[2], array[3]);
+                s.append_c_string(chars);
+                delete(chars);"
+            }
+        }
+        return s.to_string()
+    }
+    fn get_md5_string_from_string(anon s: String) throws -> String {
+        mut sb = StringBuilder::create()
+        unsafe {
+            cpp {
+                "
+                char* md5 = (char*)os_call((u64)\"@saubari_get_md5_string_from_string\", (u64)s.characters());
+                sb.append_c_string(md5);
+                delete(md5);
+                "
+            }
+        }
+        return sb.to_string()
+    }
+    fn get_ipv4_u8_array_from_address_string(anon s: String) throws -> [u8] {
+        mut address: [u8] = []
+        let octet_strings = s.split(c'.')
+        for octet_string in octet_strings {
+            unsafe {
+                cpp {
+                    "auto value = octet_string.to_number<u32>();
+                    if (value.has_value()) {
+                        auto result = value.release_value();
+                        address.push(result & 0xff);
+                    }"
+                }
+            }
+        }
+        return address
+    }
+    fn get_ipv4_u8_array_from_address_u32(anon addr: u32) throws -> [u8] {
+        mut address: [u8] = []
+        // let source_address: [u8] = [ipv4_packet[12], ipv4_packet[13], ipv4_packet[14], ipv4_packet[15]]
+        address.push(((addr >> 24) & 0xff) as! u8)
+        address.push(((addr >> 16) & 0xff) as! u8)
+        address.push(((addr >> 8) & 0xff) as! u8)
+        address.push((addr & 0xff) as! u8)
+        return address
+    }
+    fn get_string_from_u8_array(anon array: [u8]) throws -> String {
+        mut s = StringBuilder::create()
+        unsafe {
+            cpp {
+                "for (int i = 0; i < array.size(); i++) {
+                    s.append(array[i]);
+                }"
+            }
+        }
+        return s.to_string()
+    }
+    fn get_u16_from_u8_array(anon array: [u8], anon offset: i64) -> u16{
+        return (array[offset] as! u16 << 8) + array[offset + 1] as! u16
+    }
+    fn get_u16_from_u8_arrayslice(anon array: ArraySlice<u8>, anon offset: i64) -> u16{
+        return (array[offset] as! u16 << 8) + array[offset + 1] as! u16
+    }
+    fn push_string_to_u8_array(anon mut array: [u8], anon s: String) throws {
+        for i in 0..s.length() {
+            unsafe {
+                cpp {
+                    "array.push(s.characters()[i]);"
+                }
+            }
+        }
+    }
+    fn push_u16_to_u8_array(anon mut array: [u8], anon value: u16) throws {
+        array.push((value >> 8) as! u8)
+        array.push((value & 0xff) as! u8)
+    }
+    fn push_u32_to_u8_array(anon mut array: [u8], anon value: u32) throws {
+        mut val_u32_to_u8: u32 = 0
+        val_u32_to_u8 = (value >> 24) & 0xff
+        array.push(val_u32_to_u8 as! u8)
+        val_u32_to_u8 = (value >> 16) & 0xff
+        array.push(val_u32_to_u8 as! u8)
+        val_u32_to_u8 = (value >> 8) & 0xff
+        array.push(val_u32_to_u8 as! u8)
+        array.push((value & 0xff) as! u8)
+    }
+    fn get_dictionary_from_json_file(anon json_file: String) throws -> [String:String] {
+        mut dictionary: [String:String] = Dictionary()
+        let json_bytes = OS::read_entire_file(json_file)
+        let json_string = get_string_from_u8_array(json_bytes)
+        unsafe {
+            cpp {
+                "auto json = JsonValue::from_string(json_string).value();
+                auto const& object = json.as_object();
+                object.for_each_member([&]([[maybe_unused]] auto& property_name, [[maybe_unused]] const JsonValue& property_value) {
+                    dictionary.set(property_name, property_value.deprecated_to_byte_string());
+                });"
+            }
+        }
+        return dictionary
+    }
+    fn get_dictionary_from_string(anon s: String) throws -> [String:String] {
+        mut dictionary: [String:String] = Dictionary()
+        unsafe {
+            cpp {
+                "auto json = JsonValue::from_string(s).value();
+                auto const& object = json.as_object();
+                object.for_each_member([&]([[maybe_unused]] auto& property_name, [[maybe_unused]] const JsonValue& property_value) {
+                    dictionary.set(property_name, property_value.deprecated_to_byte_string());
+                });"
+            }
+        }
+        return dictionary
+    }
+    fn string_from_file(anon filepath: String) throws -> String {
+        if filepath.is_empty() or not OS::path_exists(filepath) {
+            return ""
+        }
+        let array = OS::read_entire_file(filepath)
+        mut s = StringBuilder::create()
+        unsafe {
+            cpp {
+                "for (int i = 0; i < array.size(); i++) {
+                    s.append(array[i]);
+                }"
+            }
+        }
+        return s.to_string()
+    }
+}
\ No newline at end of file
diff --git a/src/net/net.jakt b/src/net/net.jakt
new file mode 100644
index 0000000..b7bdec9
--- /dev/null
+++ b/src/net/net.jakt
@@ -0,0 +1,173 @@
+import devices::virtio { VirtIO, VirtIOReg }
+
+import lib::util { Util }
+
+import os::os { OS }
+import os::pci { PCI, PCIDevice }
+import os::time { Time }
+
+import tcpip { TCPIP }
+
+class NetDevices {
+    public virtio: VirtIO
+    public fn create(pci_device: PCIDevice) throws -> NetDevices {
+        return NetDevices(
+            virtio: VirtIO(pci_device, rq_index: 0, rq_size: 0, rq: 0, tq_size: 0, tq: 0)
+        )
+    }
+}
+
+class Net {
+    public device: NetDevices
+    public mac_address: [u8]
+    public tcpip: TCPIP
+    public pci_device: PCIDevice
+    public fn init(config: [String:String]) throws -> Net {
+        let pci_device = PCI::find_device_by_class_code(0x020000)
+        mut net = Net(
+            device: NetDevices::create(pci_device)
+            mac_address: []
+            tcpip: TCPIP(
+                ipv4_address: Util::get_ipv4_u8_array_from_address_string(config["tcpip.ipv4_address"])
+                ipv4_netmask: Util::get_ipv4_u8_array_from_address_string(config["tcpip.ipv4_netmask"])
+                ipv4_network: Util::get_ipv4_u8_array_from_address_string(config["tcpip.ipv4_network"])
+                ipv4_gateway: Util::get_ipv4_u8_array_from_address_string(config["tcpip.ipv4_gateway"])
+                dns_server_address: Util::get_ipv4_u8_array_from_address_string(config["tcpip.ipv4_dns_server_address"])
+                dns_server_port: config["tcpip.ipv4_dns_server_port"].to_number<u32>().value() as! u16
+                mss_size: config["tcpip.mss_size"].to_number<u32>().value() as! u16
+                tx_queue: []
+                ttl: 64
+                arp_cache: Dictionary()
+                bound_sockets: Dictionary()
+                dns_cache: Dictionary()
+                tcp_sessions: []
+                pending_dns_lookups: Dictionary()
+                pending_dns_cached_entries: Dictionary()
+                pending_icmp_requests: Dictionary()
+                timestamp_last_arp_request: 0
+                rx_bytes: 0
+                rx_frames: 0
+                tx_bytes: 0
+                tx_frames: 0
+                )
+            pci_device
+        )
+        if net.pci_device.vendor_id() == 0x1af4 and net.pci_device.device_id() == 0x1000 {
+            println("[net] Found device: virtio-net, QEMU")
+            for i in 0u16..6u16 {
+                net.mac_address.push(net.pci_device.io_read_u8(VirtIOReg::config as! u16 + i))
+            }
+            net.device.virtio.init()
+            return net
+        }
+        println("[net] No supported vendor ids found")
+        OS::exit()
+        return net
+    }
+    fn process_ethernet_frame(mut this, anon frame: [u8]) throws {
+        let ethertype: u16 = (frame[12] as! u16 * 256) + frame[13] as! u16
+        match ethertype {
+            0x0806 => {
+                //println("ARP")
+                .tcpip.process_arp_packet(.mac_address, frame)
+            }
+            0x0800 => {
+                //println("IPv4")
+                .tcpip.process_ipv4_packet(.mac_address, frame)
+            }
+            0x86dd => {
+                //.tcpip.process_ipv6_packet(frame)
+            }
+            0x8035 => {
+                //.tcpip.process_rarp_packet(frame)
+            }
+            else => {
+                // unsupported
+            }
+        }
+    }
+    public fn process_events(mut this) throws {
+        mut received_frame = .rx_frame()
+        if received_frame.size() > 0 {
+            .tcpip.rx_bytes += received_frame.size() as! u64
+            .tcpip.rx_frames++
+            .process_ethernet_frame(received_frame)
+        }
+        .tcpip.tcp_transmit_pending_data_for_existing_sessions()
+        for frame in .tcpip.tx_queue {
+            .tx_frame(frame)
+        }
+        if .tcpip.tx_queue.size() > 0 {
+            .tcpip.tx_queue.shrink(0)
+        }
+        .tcpip.tcp_process_bind_request()
+        .tcpip.tcp_process_client_socket_request(.mac_address)
+        .tcpip.tcp_process_client_received_data()
+        .tcpip.tcp_process_client_send_requests(.mac_address)
+        .tcpip.dns_process_client_request(.mac_address)
+        .tcpip.icmp_process_client_request(.mac_address)
+        .tcpip.netinfo_process_client_request(.mac_address)
+    }
+    fn rx_frame(mut this) throws -> [u8] {
+        mut frame: [u8] = []
+        if .pci_device.vendor_id() == 0x1af4 and .pci_device.device_id() == 0x1000 {
+            frame = .device.virtio.rx_frame()
+        }
+        return frame
+    }
+    fn tx_frame(mut this, anon mut data: [u8]) throws {
+        if data.size() < 1 {
+            return
+        }
+        while data.size() < 60 {
+            data.push(0u8)
+        }
+        .tcpip.tx_bytes += data.size() as! u64
+        .tcpip.tx_frames++
+        if .pci_device.vendor_id() == 0x1af4 and .pci_device.device_id() == 0x1000 {
+            .device.virtio.tx_frame(data)
+        }
+    }
+}
+
+fn main() {
+    println("$WW,1$")
+    mut config = Util::get_dictionary_from_json_file("M:/System/Config/Net.json")
+    mut net = Net::init(config)
+
+    println("[net] PCI device is {}", net.pci_device)
+    print("[net] MAC address is ")
+    for i in 0u16..5u16 {
+        print("{:0>2x}:", net.mac_address[i])
+    }
+    println("{:0>2x}", net.mac_address[5])
+    print("[net] IPv4 address is ")
+    for i in 0u16..3u16 {
+        print("{:d}.", net.tcpip.ipv4_address[i])
+    }
+    println("{:d}", net.tcpip.ipv4_address[3])
+    println(" ")
+
+    // Update the ARP cache entry for IPv4 gateway address
+    net.tcpip.send_arp_request(net.mac_address, net.tcpip.ipv4_gateway)
+
+    mut prev_rx_frames = net.tcpip.rx_frames
+    mut prev_tx_frames = net.tcpip.tx_frames
+    mut prev_jiffies = Time::jiffies()
+
+    while true {
+        net.process_events()
+        if (prev_rx_frames != net.tcpip.rx_frames) or (prev_tx_frames != net.tcpip.tx_frames) {
+            prev_rx_frames = net.tcpip.rx_frames
+            prev_tx_frames = net.tcpip.tx_frames
+            prev_jiffies = Time::jiffies()
+        }
+        if Time::jiffies() < prev_jiffies + 250 {
+            Time::sleep(0)
+        } else {
+            Time::sleep(1)
+        }
+    }
+
+    OS::exit()
+}
\ No newline at end of file
diff --git a/src/net/os/ioport.jakt b/src/net/os/ioport.jakt
new file mode 100644
index 0000000..a01eaa2
--- /dev/null
+++ b/src/net/os/ioport.jakt
@@ -0,0 +1,29 @@
+import extern c "ioport.h" {
+    extern fn ioport_read_u8(address: u16) -> u8
+    extern fn ioport_read_u16(address: u16) -> u16
+    extern fn ioport_read_u32(address: u16) -> u32
+    extern fn ioport_write_u8(address: u16, value: u8)
+    extern fn ioport_write_u16(address: u16, value: u16)
+    extern fn ioport_write_u32(address: u16, value: u32)
+}
+
+struct IOPort {
+    fn read_u8(anon address: u16) throws -> u8 {
+        return ioport_read_u8(address)
+    }
+    fn read_u16(anon address: u16) throws -> u16 {
+        return ioport_read_u16(address)
+    }
+    fn read_u32(anon address: u16) throws -> u32 {
+        return ioport_read_u32(address)
+    }
+    fn write_u8(address: u16, value: u8) {
+        return ioport_write_u8(address, value)
+    }
+    fn write_u16(address: u16, value: u16) {
+        return ioport_write_u16(address, value)
+    }
+    fn write_u32(address: u16, value: u32) {
+        return ioport_write_u32(address, value)
+    }
+}
diff --git a/src/net/os/os.jakt b/src/net/os/os.jakt
new file mode 100644
index 0000000..358ac9f
--- /dev/null
+++ b/src/net/os/os.jakt
@@ -0,0 +1,154 @@
+import extern c "os.h" {
+    extern fn os_blink(frequency: raw c_char) -> bool
+    extern fn os_call(function_name: u64, arg: u64) -> u64
+    extern fn os_device_calloc(size: u32) -> u32
+    extern fn os_exit()
+    extern fn os_file_picker(path: raw c_char, glob: raw c_char)
+    extern fn os_files_list(path: raw c_char)
+    extern fn os_is_vm() -> bool
+    extern fn os_path_exists(anon path: raw c_char) -> bool
+    extern fn os_pc_speaker(frequency: raw c_char)
+    extern fn os_random() -> u64
+    extern fn os_screenshot()
+    extern fn os_to_uppercase(anon input_string: raw c_char) -> raw c_char
+}
+
+struct OS {
+    fn blink(frequency: f64 = 2.5) throws -> bool {
+        let frequency_as_string = format("{}", frequency)
+        return os_blink(frequency: frequency_as_string.c_string())
+    }
+    fn call(anon function_name: String, anon arg: String) throws -> u64 {
+        mut res: u64 = 0
+        unsafe {
+            cpp {
+                "
+                res = os_call((u64)function_name.characters(), (u64)arg.characters());
+                "
+            }
+        }
+        return res
+    }
+    fn device_calloc(anon size: u32) throws -> u32 {
+        return os_device_calloc(size)
+    }
+    fn device_copy_buffer(anon buffer: [u8]) -> u32 {
+        mut address: u32 = 0
+        mut size = buffer.size()
+        unsafe {
+            cpp {
+                "u8 *data = (u8*)os_device_calloc(size);
+                for (int i = 0; i < size; i++)
+                    data[i] = buffer[i];
+                address = (uintptr_t)data;"
+            }
+        }
+        return address
+    }
+    fn exit() {
+        os_exit()
+    }
+    fn file_picker(path: String, glob: String) throws -> String {
+        mut s = StringBuilder::create()
+        unsafe {
+            cpp {
+                "char const *chars = os_file_picker(path.characters(), glob.characters());
+                s.append_c_string(chars);
+                delete(chars);"
+            }
+        }
+        return s.to_string()
+    }
+    fn files_list(path: String) throws -> [String] {
+        mut s = StringBuilder::create()
+        unsafe {
+            cpp {
+                "char const *chars = os_files_list(path.characters());
+                if (chars) {
+                    s.append_c_string(chars);
+                    delete(chars);
+                }"
+            }
+        }
+        return s.to_string().split(c'|')
+    }
+    fn path_exists(anon path: String) -> bool {
+        return os_path_exists(path.c_string())
+    }
+    fn is_vm() -> bool {
+        return os_is_vm()
+    }
+    fn pc_speaker(frequency: f64) throws {
+        let frequency_as_string = format("{}", frequency)
+        os_pc_speaker(frequency: frequency_as_string.c_string())
+    }
+    fn put_char(ch: u8) {
+        unsafe {
+            cpp {
+                "putchar(ch);"
+            }
+        }
+    }
+    fn random() -> u64 {
+        return os_random()
+    }
+    fn read_entire_file(anon filename: String) throws -> [u8] {
+        mut size = 0
+        mut buffer: [u8] = []
+        unsafe {
+            cpp {
+                "u8 *data = os_read_entire_file(filename.characters(), &size);
+                for (int i = 0; i < size; i++)
+                  buffer.push(data[i]);
+                free(data);"
+            }
+        }
+        return buffer
+    }
+    fn read_device_memory(address: u32, size: i64) throws -> [u8] {
+        mut buffer: [u8] = [];
+        unsafe {
+            cpp {
+                "u8 *device_memory = (u8*)address;
+                for (int i = 0; i < size; i++)
+                    buffer.push(device_memory[i]);"
+            }
+        }
+        return buffer
+    }
+    fn read_u16_from_device_memory(anon address: u32) throws -> u16 {
+        mut value: u16 = 0
+        unsafe {
+            cpp {
+                "value = *(u16*)address;"
+            }
+        }
+        return value
+    }
+    fn screenshot() {
+        os_screenshot()
+    }
+    fn to_uppercase(anon input_string: String) throws -> String {
+        mut s = StringBuilder::create()
+        unsafe {
+            cpp {
+                "char const *chars = os_to_uppercase(input_string.characters());
+                s.append_c_string(chars);
+                delete(chars);"
+            }
+        }
+        return s.to_string()
+    }
+    fn write_entire_file(filename: String, buffer: [u8]) {
+        mut size = buffer.size()
+        unsafe {
+            cpp {
+                "unsigned char *data = (unsigned char *)malloc(size);
+                for (int i = 0; i < size; i++)
+                    data[i] = buffer[i];
+                os_write_entire_file(filename.characters(), data, size);
+                free(data);"
+            }
+        }
+    }
+}
diff --git a/src/net/os/pci.jakt b/src/net/os/pci.jakt
new file mode 100644
index 0000000..6d15132
--- /dev/null
+++ b/src/net/os/pci.jakt
@@ -0,0 +1,103 @@
+import ioport { IOPort }
+
+import extern c "pci.h" {
+    extern fn pci_find(anon class_code: i64) -> i64
+    extern fn pci_read_u8(anon bus: i64, anon device: i64, anon fun: i64, anon offset: i64) -> u8
+    extern fn pci_read_u16(anon bus: i64, anon device: i64, anon fun: i64, anon offset: i64) -> u16
+    extern fn pci_read_u32(anon bus: i64, anon device: i64, anon fun: i64, anon offset: i64) -> u32
+    extern fn pci_write_u8(anon bus: i64, anon device: i64, anon fun: i64, anon offset: i64, anon value: u8)
+    extern fn pci_write_u16(anon bus: i64, anon device: i64, anon fun: i64, anon offset: i64, anon value: u16)
+    extern fn pci_write_u32(anon bus: i64, anon device: i64, anon fun: i64, anon offset: i64, anon value: u32)
+}
+
+struct PCIDevice {
+    bus: i64
+    device: i64
+    fun: i64
+    pci_vendor_id: u16
+    pci_device_id: u16
+    bar: [u32]
+    public fn enable_bus_master(mut this) {
+        .set_command(.command() | 0x4)
+    }
+    public fn read_u8(this, anon offset: i64) -> u8 {
+        return pci_read_u8(.bus, .device, .fun, offset)
+    }
+    public fn read_u16(this, anon offset: i64) -> u16 {
+        return pci_read_u16(.bus, .device, .fun, offset)
+    }
+    public fn read_u32(this, anon offset: i64) -> u32 {
+        return pci_read_u32(.bus, .device, .fun, offset)
+    }
+    public fn write_u8(this, offset: i64, value: u8) {
+        pci_write_u8(.bus, .device, .fun, offset, value)
+    }
+    public fn write_u16(this, offset: i64, value: u16) {
+        pci_write_u16(.bus, .device, .fun, offset, value)
+    }
+    public fn write_u32(this, offset: i64, value: u32) {
+        pci_write_u32(.bus, .device, .fun, offset, value)
+    }
+    public fn io_read_u8(this, anon offset: u16) throws -> u8 {
+        return IOPort::read_u8(.bar[0] as! u16 + offset)
+    }
+    public fn io_read_u16(this, anon offset: u16) throws -> u16 {
+        return IOPort::read_u16(.bar[0] as! u16 + offset)
+    }
+    public fn io_read_u32(this, anon offset: u16) throws -> u32 {
+        return IOPort::read_u32(.bar[0] as! u16 + offset)
+    }
+    public fn io_write_u8(this, offset: u16, value: u8) {
+        IOPort::write_u8(address: .bar[0] as! u16 + offset, value)
+    }
+    public fn io_write_u16(this, offset: u16, value: u16) {
+        IOPort::write_u16(address: .bar[0] as! u16 + offset, value)
+    }
+    public fn io_write_u32(this, offset: u16, value: u32) {
+        IOPort::write_u32(address: .bar[0] as! u16 + offset, value)
+    }
+    public fn vendor_id(this) -> u16 {
+        return .pci_vendor_id
+    }
+    public fn device_id(this) -> u16 {
+        return .pci_device_id
+    }
+    public fn command(this) -> u16 {
+        return pci_read_u16(.bus, .device, .fun, 0x4)
+    }
+    public fn set_command(this, anon value: u16) {
+        pci_write_u16(.bus, .device, .fun, offset: 0x4, value)
+    }
+    public fn status(this) -> u16 {
+        return pci_read_u16(.bus, .device, .fun, 0x6)
+    }
+}
+
+fn lookup_bar(bus: i64, device: i64, fun: i64, anon index: i64) -> u32 {
+    if index < 0 or index > 5 {
+        return 0xFFFFFFFF
+    }
+    return pci_read_u32(bus, device, fun, 0x10 + (index * 4)) & 0xFFFFFFFC
+}
+
+struct PCI {
+    public fn find_device_by_class_code(anon class_code: i64) throws -> PCIDevice {
+        let result = pci_find(class_code)
+
+        if result < 0 {
+            eprintln("error: device not found")
+            throw Error::from_errno(1)
+        }
+
+        let bus = (result >> 16) & 0xff
+        let device = (result >> 8) & 0xff 
+        let fun = result & 0xff
+        let pci_vendor_id = pci_read_u16(bus, device, fun, 0x0)
+        let pci_device_id = pci_read_u16(bus, device, fun, 0x2)
+        mut bar: [u32] = []
+        for i in 0..5 {
+            bar.push(lookup_bar(bus, device, fun, i))
+        }
+        return PCIDevice(bus, device, fun, pci_vendor_id, pci_device_id, bar)
+    }
+}
diff --git a/src/net/os/time.jakt b/src/net/os/time.jakt
new file mode 100644
index 0000000..c84daaa
--- /dev/null
+++ b/src/net/os/time.jakt
@@ -0,0 +1,94 @@
+import extern c "time.h" {
+    extern fn time_busy(anon duration: i64)
+    extern fn time_jiffies() -> i64
+    extern fn time_now() -> i64
+    extern fn time_sleep(anon duration: i64)
+}
+
+struct Time {
+    fn busy(anon duration: i64) {
+        time_busy(duration)
+    }
+    fn jiffies() throws -> i64 {
+        return time_jiffies()
+    }
+    fn now() throws -> i64 {
+        return time_now()
+    }
+    fn cdate_to_unix(anon cdate: i64) -> i64 {
+        // (cdate - Str2Date("1/1/1970") / CDATE_FREQ + NIST_TIME_OFFSET
+        return (cdate - 3090344933588992) / 49710 + 8575
+    }
+    fn unix_to_cdate(anon unix: i64) -> i64 {
+        // (unix - NIST_TIME_OFFSET) * CDATE_FREQ + Str2Date("1/1/1970")
+        return (unix - 8575) * 49710 + 3090344933588992
+    }
+    fn sleep(anon duration: i64) {
+        time_sleep(duration)
+    }
+    fn timestamp_from_unix(anon timestamp: i64) -> String {
+
+        let SECS_PER_DAY   = 86400
+        let DAYS_PER_YEAR    = 365
+        let DAYS_PER_LYEAR   = 366
+        let DAYS_PER_LYEAR_PERIOD = 146097
+        let YEARS_PER_LYEAR_PERIOD = 400
+
+        mut days = timestamp / SECS_PER_DAY
+        mut remainder = timestamp - (days * SECS_PER_DAY)
+        if timestamp < 0 and remainder == 0 {
+            days++
+            remainder -= SECS_PER_DAY
+        }
+
+        mut cur_year = 0
+        mut months: [i64] = []
+        mut tmp_days = 0
+
+        let month_tab = [ 0,  31,  59,  90, 120, 151, 181, 212, 243, 273, 304, 334 ]
+        let month_tab_leap = [ -1,  30,  59,  90, 120, 151, 181, 212, 243, 273, 304, 334 ]
+
+    	tmp_days = days;
+    	if tmp_days >= DAYS_PER_LYEAR_PERIOD or tmp_days <= -DAYS_PER_LYEAR_PERIOD {
+    		cur_year += YEARS_PER_LYEAR_PERIOD * (tmp_days / DAYS_PER_LYEAR_PERIOD);
+    		tmp_days -= DAYS_PER_LYEAR_PERIOD * (tmp_days / DAYS_PER_LYEAR_PERIOD);
+    	}
+    	while tmp_days >= DAYS_PER_LYEAR {
+    		cur_year++;
+    		if cur_year % 4 == 0 {
+    			tmp_days -= DAYS_PER_LYEAR;
+    		} else {
+    			tmp_days -= DAYS_PER_YEAR;
+    		}
+    	}
+        if cur_year % 4 == 0 {
+            months = month_tab_leap
+        } else {
+            months = month_tab
+        }
+
+        mut i = 11
+        while i > 0 {
+            if tmp_days > months[i] {
+                break;
+            }
+            i--
+        }
+
+        let year = 1970 + cur_year
+        let month = i + 1
+        let day = tmp_days - months[i]
+
+        let hours = remainder / 3600
+        let minutes = (remainder - hours * 3600) / 60
+        let seconds = remainder % 60
+
+        mut sb = StringBuilder::create()
+        sb.clear()
+        sb.appendff("{:0>4d}-{:0>2d}-{:0>2d}T{:0>2d}:{:0>2d}:{:0>2d}.000Z", year, month, day, hours, minutes, seconds)
+        return sb.to_string()
+    }
+    fn timestamp_from_cdate(anon cdate: i64) -> String {
+        return timestamp_from_unix(cdate_to_unix(cdate))
+    }
+}
\ No newline at end of file
diff --git a/src/net/tcpip.jakt b/src/net/tcpip.jakt
new file mode 100644
index 0000000..fc2b6bb
--- /dev/null
+++ b/src/net/tcpip.jakt
@@ -0,0 +1,1755 @@
+import lib::util { Util }
+
+import os::os { OS }
+import os::time { Time }
+
+struct IcmpRequest {
+  host: u64
+  identifier: u64
+  sequence_number: u64
+  response_type_pointer: u64
+}
+
+enum TcpSessionState: i64 {
+    Idle = 0
+    Established = 1
+    Closed = 2
+    Connecting = 4
+}
+
+struct TcpSession {
+    local_mac: [u8]
+    remote_mac: [u8]
+    local_address: [u8]
+    remote_address: [u8]
+    local_port: u16
+    remote_port: u16
+    local_sequence_number: u32
+    acknowledgement_number: u32
+    timestamp_last_echo_reply: u32
+    timestamp_last_jiffies: i64
+    timestamp_origin: i64
+    last_window_size: u16
+    last_tx_sequence_number: u32
+    last_tx_was_acked: bool
+    tx_chunk_counter: i64
+    state: TcpSessionState
+    pending_data_to_transmit: [u8]
+    received_data: [u8]
+    received_frames: [[u8]]
+    is_bound_socket: bool
+    socket: u64
+}
+
+enum TcpBindError: u64 {
+    BadRequest = 1
+    SocketIsAlreadyBound = 2
+}
+
+class TCPIP {
+    public ipv4_address: [u8]
+    public ipv4_netmask: [u8]
+    public ipv4_network: [u8]
+    public ipv4_gateway: [u8]
+    public dns_server_address: [u8]
+    public dns_server_port: u16
+    public mss_size: u16
+    public tx_queue: [[u8]]
+    public ttl: u8
+    public arp_cache: [String:[u8]]
+    public bound_sockets: [u16:u64]
+    public dns_cache: [String:[u8]]
+    public tcp_sessions: [TcpSession]
+    public pending_dns_lookups: [u16:u64]
+    public pending_dns_cached_entries: [u16:String]
+    public pending_icmp_requests: [u16:u64]
+    public timestamp_last_arp_request: i64
+    public rx_bytes: u64
+    public rx_frames: u64
+    public tx_bytes: u64
+    public tx_frames: u64
+    fn calculate_header_checksum(this, anon mut packet: [u8], offset: i64, count: i64) -> u16 {
+        mut sum: i64 = 0
+        for i in 0..count {
+            if (i & 1) == 1 {
+                sum += packet[offset + i] as! i64
+            } else {
+                sum += (packet[offset + i] as! i64 * 256)
+            }
+        }
+        mut calculated_checksum: u16 = 0
+        let overflowable_checksum: i64 = (sum & 0xffff) + (sum >> 16)
+        if overflowable_checksum < 65536 {
+            calculated_checksum = ~((sum & 0xffff) as! u16 + (sum >> 16) as! u16)
+        } else {
+            calculated_checksum = ~(unchecked_add <u16>((sum & 0xffff) as! u16, (sum >> 16) as! u16)) - 1
+        }
+        return calculated_checksum
+    }
+    fn push_ethernet_header(this, mut packet: [u8], destination_mac: [u8], source_mac: [u8], ethertype: u16) throws {
+        for i in 0..6 {
+            packet.push(destination_mac[i])
+        }
+        for i in 0..6 {
+            packet.push(source_mac[i])
+        }
+        Util::push_u16_to_u8_array(packet, ethertype)
+    }
+    fn push_arp_packet(this, mut packet: [u8], opcode: u16, sender_mac: [u8], sender_ipv4: [u8], target_mac: [u8], target_ipv4: [u8]) throws {
+        let hardware_protocol_header_data: [u8] = [0x00, 0x01, 0x08, 0x00, 0x06, 0x04]
+        for i in 0..6 {
+            packet.push(hardware_protocol_header_data[i])
+        }
+        Util::push_u16_to_u8_array(packet, opcode)
+        for i in 0..6 {
+            packet.push(sender_mac[i])
+        }
+        for i in 0..4 {
+            packet.push(sender_ipv4[i])
+        }
+        for i in 0..6 {
+            packet.push(target_mac[i])
+        }
+        for i in 0..4 {
+            packet.push(target_ipv4[i])
+        }        
+    }
+    fn push_ipv4_header(this, mut packet: [u8], total_length: u16, flags: u8, protocol: u8, source_address: [u8], destination_address: [u8]) throws {
+        packet.push(0x45u8) // version & header length
+        packet.push(0x00u8) // differentiated services field
+        Util::push_u16_to_u8_array(packet, total_length) // total length of entire packet
+        let identification: u64 = OS::random() & 0xffff
+        Util::push_u16_to_u8_array(packet, identification as! u16)
+        packet.push(flags)
+        packet.push(0x00u8) // fragment size (unused)
+        packet.push(.ttl) // time-to-live
+        packet.push(protocol) // protocol number
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+        for i in 0..4 {
+            packet.push(source_address[i])
+        }
+        for i in 0..4 {
+            packet.push(destination_address[i])
+        }
+        let checksum = .calculate_header_checksum(packet, offset: 14, count: 20)
+        packet[24] = (checksum >> 8) as! u8
+        packet[25] = (checksum & 0xff) as! u8        
+    }
+    fn push_udp_header(this, mut packet: [u8], source_port: u16, destination_port: u16, length: i64) throws {
+        Util::push_u16_to_u8_array(packet, source_port)
+        Util::push_u16_to_u8_array(packet, destination_port)
+        Util::push_u16_to_u8_array(packet, length as! u16)
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+    }
+    fn send_arp_reply(mut this, anon mac_address: [u8], anon sender_mac_address: [u8], anon sender_ipv4_address: [u8]) throws {
+        mut packet: [u8] = []
+        .push_ethernet_header(packet, destination_mac: sender_mac_address, source_mac: mac_address, ethertype: 0x0806)
+        .push_arp_packet(
+            packet
+            opcode: 0x0002
+            sender_mac: mac_address
+            sender_ipv4: .ipv4_address
+            target_mac: sender_mac_address
+            target_ipv4: sender_ipv4_address
+        )
+        .tx_queue.push(packet)
+    }
+    public fn send_arp_request(mut this, anon mac_address: [u8], anon target_ipv4_address: [u8]) throws {
+        if .timestamp_last_arp_request > 0 and (.timestamp_last_arp_request + 5000) > Time::jiffies() {
+            return
+        }
+        mut packet: [u8] = []
+        .push_ethernet_header(packet, destination_mac: [0xff, 0xff, 0xff, 0xff, 0xff, 0xff], source_mac: mac_address, ethertype: 0x0806)
+        .push_arp_packet(
+            packet
+            opcode: 0x0001
+            sender_mac: mac_address
+            sender_ipv4: .ipv4_address
+            target_mac: [0x00, 0x00, 0x00, 0x00, 0x00, 0x00]
+            target_ipv4: target_ipv4_address
+        )
+        .tx_queue.push(packet)
+        .timestamp_last_arp_request = Time::jiffies()
+    }
+    fn insert_arp_cache_entry(mut this, ipv4_address: [u8], mac_address: [u8]) throws {
+        .arp_cache[Util::get_hexadecimal_string_from_ipv4_u8_array(ipv4_address)] = mac_address
+    }
+    public fn process_arp_packet(mut this, anon mac_address: [u8], anon frame: [u8]) throws {
+        let arp_packet = frame[14..]
+
+        let hardware_type = Util::get_u16_from_u8_arrayslice(arp_packet, 0)
+        let protocol_type = Util::get_u16_from_u8_arrayslice(arp_packet, 2)
+        let hardware_size = arp_packet[4]
+        let protocol_size = arp_packet[5]
+        let opcode = Util::get_u16_from_u8_arrayslice(arp_packet, 6)
+
+        let sender_mac_address: [u8] = [arp_packet[8], arp_packet[9], arp_packet[10], 
+            arp_packet[11], arp_packet[12], arp_packet[13]]
+        let sender_ipv4_address: [u8] = [arp_packet[14], arp_packet[15], arp_packet[16], arp_packet[17]] 
+
+        let target_mac_address: [u8] = [arp_packet[18], arp_packet[19], arp_packet[20], 
+            arp_packet[21], arp_packet[22], arp_packet[23]]
+        let target_ipv4_address: [u8] = [arp_packet[24], arp_packet[25], arp_packet[26], arp_packet[27]] 
+
+        // If any of these mismatch, the packet is invalid
+        if hardware_type != 0x0001 or protocol_type != 0x0800 or 
+            hardware_size != 6 or protocol_size != 4 {
+                return
+        }
+
+        // Add the sender to ARP cache
+        .insert_arp_cache_entry(ipv4_address: sender_ipv4_address, mac_address: sender_mac_address)
+        //print("Updated ARP cache entry for {}: ", sender_ipv4_address)
+        //for i in 0..6 {
+        //    print("{:0>2x}", sender_mac_address[i])
+        //    if (i < 5) {
+        //        print(":")
+        //    }
+        //}
+        //println(" ")
+
+        match opcode {
+            1 => {
+                // ARP request, check if our ipv4 address matches
+                for i in 0..4 {
+                    if .ipv4_address[i] != target_ipv4_address[i] {
+                        // doesn't match, break
+                        break
+                    }
+                }
+                // matches, send ARP reply
+                .send_arp_reply(mac_address, sender_mac_address, sender_ipv4_address)
+            }
+            else => {
+                // unsupported opcode, ignore
+            }
+        }
+    }
+    public fn send_dns_query(mut this, anon mac_address: [u8], anon query: String, anon pointer_to_u32: u64) throws {
+
+        mut cached_result: u32 = 0
+        if .dns_cache.contains(query) {
+            cached_result += (.dns_cache[query][0] as! u32 * (256 * 256 * 256) as! u32)
+            cached_result += (.dns_cache[query][1] as! u32 * (256 * 256) as! u32)
+            cached_result += (.dns_cache[query][2] as! u32 * 256 as! u32)
+            cached_result += .dns_cache[query][3] as! u32
+            unsafe {
+                cpp {
+                    "u32 *r = (u32*)pointer_to_u32;
+                    r[0] = cached_result;"
+                }
+            }
+            return
+        }
+
+        mut packet: [u8] = []
+        mut dns_packet: [u8] = []
+        
+        let transaction_id: u64 = OS::random() & 0xffff
+        Util::push_u16_to_u8_array(dns_packet, transaction_id as! u16)
+
+        .pending_dns_lookups.set(transaction_id as! u16, pointer_to_u32)
+        .pending_dns_cached_entries.set(transaction_id as! u16, query)
+
+        let dns_header_data: [u8] = [0x01, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]
+        dns_packet.push_values(&dns_header_data)
+
+        let segments = query.split(c'.')
+        for segment in segments {
+            dns_packet.push(segment.length() as! u8)
+            //println("segment_length: {} bytes", segment.length())
+            unsafe {
+                cpp {
+                    "const char *raw_string = segment.characters();
+                    for (int i = 0; i < segment.length(); i++)
+                        dns_packet.push(raw_string[i]);"
+                }
+            }
+        }
+        dns_packet.push(0x00u8)
+
+        let dns_footer_data: [u8] = [0x00, 0x01, 0x00, 0x01]
+        dns_packet.push_values(&dns_footer_data)
+
+        let source_port: u16 = (OS::random() & 0xffff) as! u16
+        let destination_port: u16 = .dns_server_port
+
+        let destination_mac = .arp_cache[Util::get_hexadecimal_string_from_ipv4_u8_array(.ipv4_gateway)]
+        .push_ethernet_header(packet, destination_mac, source_mac: mac_address, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: 28 + dns_packet.size() as! u16, flags: 0x00, protocol: 0x11, source_address: .ipv4_address, destination_address: .dns_server_address)
+        .push_udp_header(packet, source_port, destination_port, length: 8 + dns_packet.size() as! i64)
+
+        packet.push_values(&dns_packet)
+
+        // Calculate UDP checksum
+
+        mut checksum_packet: [u8] = []
+        // Source and destination IP
+        for i in 26..34 {
+            checksum_packet.push(packet[i])
+        }
+        // Protocol
+        checksum_packet.push(0x00u8)
+        checksum_packet.push(0x11u8)
+        // UDP Packet length
+        let udp_packet_length: u16 = 8 + dns_packet.size() as! u16
+        Util::push_u16_to_u8_array(checksum_packet, udp_packet_length)
+        // Source Port
+        Util::push_u16_to_u8_array(checksum_packet, source_port)
+        // Destination port
+        Util::push_u16_to_u8_array(checksum_packet, destination_port)
+        // UDP Packet length (again)
+        Util::push_u16_to_u8_array(checksum_packet, udp_packet_length)
+        // Data
+        checksum_packet.push_values(&dns_packet)
+
+        let checksum = .calculate_header_checksum(checksum_packet, offset: 0, count: checksum_packet.size() as! i64)
+        packet[40] = (checksum >> 8) as! u8
+        packet[41] = (checksum & 0xff) as! u8        
+
+        .tx_queue.push(packet)
+
+    }
+    public fn netinfo_process_client_request(mut this, anon mac_address: [u8]) throws {
+        unsafe {
+            cpp {
+                "u64 *request = (u64*)0x300030;
+                if (*request) {
+                    int i = 0;
+                    u64 *r = (u64*)*request;
+
+                    for (i = 0; i < 6; i++) {
+                        r[0] = (r[0] << 8) | mac_address[i];
+                        if (i < 4) {
+                            r[1] = (r[1] << 8) | this->ipv4_address[i];
+                            r[2] = (r[2] << 8) | this->ipv4_netmask[i];
+                            r[3] = (r[3] << 8) | this->ipv4_network[i];
+                            r[4] = (r[4] << 8) | this->ipv4_gateway[i];
+                            r[5] = (r[5] << 8) | this->dns_server_address[i];
+                        }
+                    }
+
+                    r[6] = this->dns_server_port;
+
+                    r[7] = this->rx_bytes;
+                    r[8] = this->rx_frames;
+                    r[9] = this->tx_bytes;
+                    r[10] = this->tx_frames;
+
+                    u32 *p = (u32*)r[11];
+                    p[0] = 1; // pointer_to_u32
+
+                    *request = 0;
+                }"
+            }
+        }
+    }
+
+    public fn icmp_process_client_request(mut this, anon mac_address: [u8]) throws {
+        mut did_receive_request = false
+        mut destination_mac: [u8] = []
+        mut addr: u32 = 0
+        mut iden: u64 = 0
+        mut seq: u64 = 0
+        mut pointer_to_u32: u64 = 0
+        unsafe {
+            cpp {
+                "u64 *ls_request = (u64*)0x300020;
+                if (*ls_request) {
+                    u64 *ls_r = (u64*)*ls_request;
+                    addr = ls_r[0];
+                }"
+            }
+        }
+        // FIXME: This will crash if we 1) Ping a host on the local subnet that does not exist, then 2) subsequently ping the gateway.
+        if (addr > 0) {
+            mut remote_address_hex_string = Util::get_hexadecimal_string_from_ipv4_u8_array(.ipv4_gateway)
+            if addr != Util::get_address_u32_from_ipv4_u8_array(.ipv4_gateway) {
+                if .is_local_ipv4_address(Util::get_ipv4_u8_array_from_address_u32(addr)) {
+                    remote_address_hex_string = Util::get_hexadecimal_string_from_ipv4_u8_array(Util::get_ipv4_u8_array_from_address_u32(addr))
+                    if not .arp_cache.contains(remote_address_hex_string) {
+                        // send arp request
+                        //println("send arp request for: {}", remote_address_hex_string)
+                        .send_arp_request(mac_address, Util::get_ipv4_u8_array_from_address_u32(addr))
+                        return
+                    }
+                }
+            }
+            destination_mac = .arp_cache[remote_address_hex_string]
+        } else {
+            return
+        }
+        unsafe {
+            cpp {
+                "u64 *request = (u64*)0x300020;
+                if (*request) {
+                    did_receive_request = true;
+                    u64 *r = (u64*)*request;
+                    addr = r[0];
+                    iden = r[1];
+                    seq = r[2];
+                    pointer_to_u32 = r[3];
+                    *request = 0;
+                }"
+            }
+        }
+        if (did_receive_request) {
+            .send_icmp_request(mac_address, destination_mac, addr, iden, seq, pointer_to_u32)
+            did_receive_request = false
+        }
+    }
+
+    public fn dns_process_client_request(mut this, anon mac_address: [u8]) throws {
+        mut did_receive_request = false
+        mut request_is_ipv4_address = true
+        mut s = StringBuilder::create()
+        mut pointer_to_u32: u64 = 0
+        unsafe {
+            cpp {
+                "u64 *request = (u64*)0x300010;
+                if (*request) {
+                    did_receive_request = true;
+                    u64 *r = (u64*)*request;
+
+                    char const *chars = (char const*)r[0];
+                    s.append_c_string(chars);
+                    for (int i = 0; i < s.to_string().length(); i++)
+                        if ((chars[i] < '0' || chars[i] > '9') && chars[i] != '.')
+                            request_is_ipv4_address = false;
+                    delete(chars);
+
+                    pointer_to_u32 = r[1];
+
+                    *request = 0;
+                }"
+            }
+        }
+        if (did_receive_request) {
+            if (request_is_ipv4_address) {
+                let octets = s.to_string().split(c'.')
+                mut u32_address: u32 = 0
+                u32_address += octets[3].to_number<u32>().value()
+                u32_address += octets[2].to_number<u32>().value() << 8
+                u32_address += octets[1].to_number<u32>().value() << 16
+                u32_address += octets[0].to_number<u32>().value() << 24
+                unsafe {
+                    cpp {
+                        "u32 *r = (u32*)pointer_to_u32;
+                        r[0] = u32_address;"
+                    }
+                }
+            } else {
+                .send_dns_query(mac_address, s.to_string(), pointer_to_u32)
+            }
+            did_receive_request = false
+        }
+    }
+    fn send_icmp_request(mut this, anon mac_address: [u8], anon destination_mac: [u8], anon addr: u32, anon iden: u64, anon seq: u64, anon pointer_to_u32: u64) throws {
+        mut packet: [u8] = []
+
+        mut destination_address: [u8] = []
+        Util::push_u32_to_u8_array(destination_address, addr)
+
+        .pending_icmp_requests.set(iden as! u16, pointer_to_u32)
+
+        let total_length: u16 = 84
+
+        .push_ethernet_header(packet, destination_mac, source_mac: mac_address, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length, flags: 0x40, protocol: 0x01, source_address: .ipv4_address, destination_address)
+
+        packet.push(0x08u8) // type
+        for i in 0..3 {
+            packet.push(0x00u8) // code, checksum placeholder
+        }
+
+        Util::push_u16_to_u8_array(packet, iden as! u16)
+        Util::push_u16_to_u8_array(packet, seq as! u16)
+
+        for i in 0..8 {
+            packet.push(0x00u8) // FIXME: timestamp
+        }
+
+        mut ch: u8 = 0x20
+        for i in 0..48 {
+            packet.push(ch) // data
+            ch++
+        }
+
+        let checksum = .calculate_header_checksum(packet, offset: 34, count: packet.size() as! i64 - 34)
+        packet[36] = (checksum >> 8) as! u8
+        packet[37] = (checksum & 0xff) as! u8
+
+        .tx_queue.push(packet)
+    }
+    fn process_icmp_reply(mut this, anon frame: [u8]) throws {
+        let iden = Util::get_u16_from_u8_array(frame, 38)
+        let ttl: u16 = frame[22] as! u16
+        let payload_size: u16 = Util::get_u16_from_u8_array(frame, 16) - 20
+        mut result: u32 = 0
+        result = (payload_size as! u32) << 16
+        result += ttl as! u32
+        mut pointer_to_u32: u64 = 0
+        if .pending_icmp_requests.contains(iden) {
+            pointer_to_u32 = .pending_icmp_requests[iden]
+        }
+        if pointer_to_u32 > 0 {
+            unsafe {
+                cpp {
+                    "u32 *r = (u32*)pointer_to_u32;
+                    r[0] = result;"
+                }
+            }
+        }
+    }
+    fn send_icmp_reply(mut this, anon mac_address: [u8], anon frame: [u8]) throws {
+        mut packet: [u8] = []
+        let ipv4_packet = frame[14..]
+        let icmp_packet = frame[34..]
+
+        mut destination_mac: [u8] = []
+        for octet in frame[6..12] {
+            destination_mac.push(octet)
+        }
+        mut destination_address: [u8] = []
+        for octet in ipv4_packet[12..16] {
+            destination_address.push(octet)
+        }
+
+        .push_ethernet_header(packet, destination_mac, source_mac: mac_address, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: frame.size() as! u16 - 14, flags: 0x40, protocol: 0x01, source_address: .ipv4_address, destination_address)
+
+        for i in 0..4 {
+            packet.push(0x00u8) // type, code, checksum placeholder
+        }
+        for i in 4..8 {
+            packet.push(icmp_packet[i]) // identifier & sequence number
+        }
+        for i in 8..icmp_packet.size() {
+            packet.push(icmp_packet[i]) // data
+        }
+        let checksum = .calculate_header_checksum(packet, offset: 34, count: packet.size() as! i64 - 34)
+        packet[36] = (checksum >> 8) as! u8
+        packet[37] = (checksum & 0xff) as! u8
+
+        .tx_queue.push(packet)
+    }
+    public fn process_icmp_packet(mut this, anon mac_address: [u8], anon frame: [u8]) throws {
+        let icmp_packet = frame[34..]
+
+        let type = icmp_packet[0]
+        let code = icmp_packet[1]
+        let checksum: u16 = Util::get_u16_from_u8_arrayslice(icmp_packet, 2)
+        let identifier: u16 = Util::get_u16_from_u8_arrayslice(icmp_packet, 4)
+        let sequence_number: u16 = Util::get_u16_from_u8_arrayslice(icmp_packet, 6)
+
+        match type {
+            0 => {
+                // ICMP reply
+                .process_icmp_reply(frame)
+            }
+            8 => {
+                // ICMP request, send ICMP reply
+                .send_icmp_reply(mac_address, frame)
+            }
+            else => {
+                // unsupported
+            }
+        }
+    }
+    fn tcp_is_fin(this, flags: u16) -> bool {
+        if (flags & 1) == 1 { 
+            return true
+        }
+        return false
+    }
+    fn tcp_is_syn(this, flags: u16) -> bool {
+        if (flags & 2) == 2 { 
+            return true
+        }
+        return false
+    }
+    fn tcp_is_reset(this, flags: u16) -> bool {
+        if (flags & 4) == 4 { 
+            return true
+        }
+        return false
+    }
+    fn tcp_is_push(this, flags: u16) -> bool {
+        if (flags & 8) == 8 { 
+            return true
+        }
+        return false
+    }
+    fn tcp_is_ack(this, flags: u16) -> bool {
+        if (flags & 16) == 16 { 
+            return true
+        }
+        return false
+    }
+    fn tcp_is_urgent(this, flags: u16) -> bool {
+        if (flags & 32) == 32 { 
+            return true
+        }
+        return false
+    }
+    fn tcp_syn_packet(mut this, anon mut session: TcpSession) throws {
+        mut packet: [u8] = []
+
+        let header_length: u16 = 40
+        let flags: u16 = 0xa002 // SYN
+
+        mut ipv4_total_length: u16 = 0
+        ipv4_total_length += 20; // IPv4
+        ipv4_total_length += header_length as! u16 // TCP
+
+        .push_ethernet_header(packet, destination_mac: session.remote_mac, source_mac: session.local_mac, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: ipv4_total_length, flags: 0x40, protocol: 0x06, source_address: session.local_address, destination_address: session.remote_address)
+
+        // FIXME: put this into .push_tcp_header
+
+        Util::push_u16_to_u8_array(packet, session.local_port)
+        Util::push_u16_to_u8_array(packet, session.remote_port)
+        Util::push_u32_to_u8_array(packet, session.local_sequence_number)
+        Util::push_u32_to_u8_array(packet, 0 as! u32) // Acknowledgement number
+        Util::push_u16_to_u8_array(packet, flags)
+        Util::push_u16_to_u8_array(packet, session.last_window_size)
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Urgent pointer
+
+        mut tcp_options: [u8] = [0x02, 0x04]
+        Util::push_u16_to_u8_array(tcp_options, .mss_size)
+        Util::push_u32_to_u8_array(tcp_options, 0x0101080a)
+        packet.push_values(&tcp_options)
+        let timestamp: u32 = (session.timestamp_origin + (Time::jiffies() - session.timestamp_origin)) as! u32
+        Util::push_u32_to_u8_array(packet, timestamp)
+        Util::push_u32_to_u8_array(packet, session.timestamp_last_echo_reply)
+
+        packet.push(0x01u8)
+        packet.push(0x03u8)
+        packet.push(0x03u8)
+        packet.push(0x07u8)
+
+        // Calculate TCP checksum
+
+        mut checksum_packet: [u8] = []
+        // Source and destination IP
+        for i in 26..34 {
+            checksum_packet.push(packet[i])
+        }
+        // Protocol
+        checksum_packet.push(0x00u8)
+        checksum_packet.push(0x06u8)
+        // TCP Packet length
+        let tcp_packet_length: u16 = header_length as! u16
+        Util::push_u16_to_u8_array(checksum_packet, tcp_packet_length)
+
+        for i in 34..packet.size() {
+            checksum_packet.push(packet[i])
+        }
+
+        mut checksum = .calculate_header_checksum(checksum_packet, offset: 0, count: checksum_packet.size() as! i64)
+        packet[50] = (checksum >> 8) as! u8
+        packet[51] = (checksum & 0xff) as! u8
+
+        .tx_queue.push(packet)
+
+    }
+    fn tcp_psh_packet(mut this, anon mut session: TcpSession) throws {
+        mut packet: [u8] = []
+
+        let header_length: u16 = 32
+        let flags: u16 = 0x8018 // PSH, ACK
+
+        let maximum_payload_size_to_transmit: i64 = 1024
+        mut payload_size: i64 = maximum_payload_size_to_transmit
+        let payload_offset: i64 = session.tx_chunk_counter * maximum_payload_size_to_transmit
+        if session.pending_data_to_transmit.size() as! i64 - payload_offset < maximum_payload_size_to_transmit {
+            payload_size = session.pending_data_to_transmit.size() as! i64 - payload_offset
+        }
+
+        mut ipv4_total_length: u16 = 0
+        ipv4_total_length += 20; // IPv4
+        ipv4_total_length += header_length + payload_size as! u16 // TCP
+
+        .push_ethernet_header(packet, destination_mac: session.remote_mac, source_mac: session.local_mac, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: ipv4_total_length, flags: 0x40, protocol: 0x06, source_address: session.local_address, destination_address: session.remote_address)
+
+
+        // FIXME: put this into .push_tcp_header
+
+        Util::push_u16_to_u8_array(packet, session.local_port)
+        Util::push_u16_to_u8_array(packet, session.remote_port)
+        Util::push_u32_to_u8_array(packet, session.local_sequence_number)
+        Util::push_u32_to_u8_array(packet, session.acknowledgement_number)
+        Util::push_u16_to_u8_array(packet, flags)
+        Util::push_u16_to_u8_array(packet, session.last_window_size)
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Urgent pointer
+
+        mut tcp_options: [u8] = [0x01, 0x01, 0x08, 0x0a]
+        packet.push_values(&tcp_options)
+
+        let timestamp: u32 = (session.timestamp_origin + (Time::jiffies() - session.timestamp_origin)) as! u32
+        Util::push_u32_to_u8_array(packet, timestamp)
+        Util::push_u32_to_u8_array(packet, session.timestamp_last_echo_reply)
+
+        for i in payload_offset..(payload_offset + payload_size) {
+            packet.push(session.pending_data_to_transmit[i])
+        }
+
+        // Calculate TCP checksum
+
+        mut checksum_packet: [u8] = []
+        // Source and destination IP
+        for i in 26..34 {
+            checksum_packet.push(packet[i])
+        }
+        // Protocol
+        checksum_packet.push(0x00u8)
+        checksum_packet.push(0x06u8)
+        // TCP Packet length
+        let tcp_packet_length: u16 = header_length + payload_size as! u16
+        Util::push_u16_to_u8_array(checksum_packet, tcp_packet_length)
+
+        for i in 34..packet.size() {
+            checksum_packet.push(packet[i])
+        }
+
+        mut checksum = .calculate_header_checksum(checksum_packet, offset: 0, count: checksum_packet.size() as! i64)
+        packet[50] = (checksum >> 8) as! u8
+        packet[51] = (checksum & 0xff) as! u8
+
+        .tx_queue.push(packet)
+
+    }
+    fn tcp_rst_packet(mut this, anon mut session: TcpSession) throws {
+        mut packet: [u8] = []
+
+        let header_length: u16 = 32
+        let flags: u16 = 0x8004 // RST
+
+        let maximum_payload_size_to_transmit: i64 = 1024
+        mut payload_size: i64 = maximum_payload_size_to_transmit
+        let payload_offset: i64 = session.tx_chunk_counter * maximum_payload_size_to_transmit
+        if session.pending_data_to_transmit.size() as! i64 - payload_offset < maximum_payload_size_to_transmit {
+            payload_size = session.pending_data_to_transmit.size() as! i64 - payload_offset
+        }
+
+        mut ipv4_total_length: u16 = 0
+        ipv4_total_length += 20; // IPv4
+        ipv4_total_length += header_length + payload_size as! u16 // TCP
+
+        .push_ethernet_header(packet, destination_mac: session.remote_mac, source_mac: session.local_mac, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: ipv4_total_length, flags: 0x40, protocol: 0x06, source_address: session.local_address, destination_address: session.remote_address)
+
+
+        // FIXME: put this into .push_tcp_header
+
+        Util::push_u16_to_u8_array(packet, session.local_port)
+        Util::push_u16_to_u8_array(packet, session.remote_port)
+        Util::push_u32_to_u8_array(packet, session.local_sequence_number)
+        Util::push_u32_to_u8_array(packet, session.acknowledgement_number)
+        Util::push_u16_to_u8_array(packet, flags)
+        Util::push_u16_to_u8_array(packet, session.last_window_size)
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Urgent pointer
+
+        mut tcp_options: [u8] = [0x01, 0x01, 0x08, 0x0a]
+        packet.push_values(&tcp_options)
+
+        let timestamp: u32 = (session.timestamp_origin + (Time::jiffies() - session.timestamp_origin)) as! u32
+        Util::push_u32_to_u8_array(packet, timestamp)
+        Util::push_u32_to_u8_array(packet, session.timestamp_last_echo_reply)
+
+        for i in payload_offset..(payload_offset + payload_size) {
+            packet.push(session.pending_data_to_transmit[i])
+        }
+
+        // Calculate TCP checksum
+
+        mut checksum_packet: [u8] = []
+        // Source and destination IP
+        for i in 26..34 {
+            checksum_packet.push(packet[i])
+        }
+        // Protocol
+        checksum_packet.push(0x00u8)
+        checksum_packet.push(0x06u8)
+        // TCP Packet length
+        let tcp_packet_length: u16 = header_length + payload_size as! u16
+        Util::push_u16_to_u8_array(checksum_packet, tcp_packet_length)
+
+        for i in 34..packet.size() {
+            checksum_packet.push(packet[i])
+        }
+
+        mut checksum = .calculate_header_checksum(checksum_packet, offset: 0, count: checksum_packet.size() as! i64)
+        packet[50] = (checksum >> 8) as! u8
+        packet[51] = (checksum & 0xff) as! u8
+
+        .tx_queue.push(packet)
+
+    }
+
+    public fn tcp_transmit_pending_data_for_existing_sessions(mut this) throws {
+        for i in 0..this.tcp_sessions.size() {
+            if .tcp_sessions[i].pending_data_to_transmit.size() > 0 {
+                if (.tcp_sessions[i].tx_chunk_counter == 0) or (.tcp_sessions[i].tx_chunk_counter > 0 and .tcp_sessions[i].last_tx_was_acked) {
+                    let maximum_payload_size_to_transmit: i64 = 1024
+                    let payload_offset: i64 = .tcp_sessions[i].tx_chunk_counter * maximum_payload_size_to_transmit
+                    mut truncated_packet: [u8] = []
+                    .tcp_psh_packet(.tcp_sessions[i])
+                    if (.tcp_sessions[i].pending_data_to_transmit.size() as! i64 - payload_offset >= maximum_payload_size_to_transmit) {
+                        .tcp_sessions[i].last_tx_sequence_number = .tcp_sessions[i].local_sequence_number + maximum_payload_size_to_transmit as! u32
+                        .tcp_sessions[i].last_tx_was_acked = false
+                        .tcp_sessions[i].tx_chunk_counter++
+                    } else {
+                        // We have transmitted all of the pending data.
+                        .tcp_sessions[i].pending_data_to_transmit.shrink(0)
+                        .tcp_sessions[i].last_tx_sequence_number = 0
+                        .tcp_sessions[i].last_tx_was_acked = false
+                        .tcp_sessions[i].tx_chunk_counter = 0
+                    }
+                }
+            }
+        }
+    }
+    fn tcp_ack_packet(mut this, anon mut session: TcpSession, anon frame: [u8], mut flags: u16) throws -> u32 {
+        mut packet: [u8] = []
+
+        let syn_ipv4_packet = frame[14..]
+        let syn_tcp_packet = frame[34..]
+
+        // Swap source/destination ports
+        mut destination_port: u16 = Util::get_u16_from_u8_arrayslice(syn_tcp_packet, 0)
+        mut source_port: u16 = Util::get_u16_from_u8_arrayslice(syn_tcp_packet, 2)
+
+        let header_length: u16 = ((syn_tcp_packet[12] as! u16) >> 4) * 4
+
+        flags += ((header_length / 4) << 12)
+        flags |= 16 // ACK
+
+        mut sequence_number: u32 = session.local_sequence_number
+        mut acknowledgement_number: u32 = 0
+        for i in 0..4 {
+            acknowledgement_number += (syn_tcp_packet[7 - i] as! u32) * match i {
+                1 => 256 as! u32
+                2 => (256 * 256) as! u32
+                3 => (256 * 256 * 256) as! u32
+                else => 1 as! u32
+            }
+        }
+
+        if (.tcp_is_syn(flags) or .tcp_is_fin(flags)) {
+            acknowledgement_number += 1
+        }
+        if (.tcp_is_push(flags)) {
+            // advance ACK by number of bytes
+            flags -= 8
+        }
+        acknowledgement_number += (syn_tcp_packet.size() as! u32 - header_length as! u32)
+
+        let window_size: u16 = 0xffff
+        let urgent_pointer: u16 = Util::get_u16_from_u8_arrayslice(syn_tcp_packet, 18)
+
+        mut destination_mac: [u8] = []
+        for octet in frame[6..12] {
+            destination_mac.push(octet)
+        }
+        mut destination_address: [u8] = []
+        for octet in syn_ipv4_packet[12..16] {
+            destination_address.push(octet)
+        }
+
+        mut ipv4_total_length: u16 = 0
+        ipv4_total_length += 20; // IPv4
+        ipv4_total_length += header_length // TCP
+
+        .push_ethernet_header(packet, destination_mac, source_mac: session.local_mac, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: ipv4_total_length, flags: 0x40, protocol: 0x06, source_address: .ipv4_address, destination_address)
+
+        // FIXME: put this into .push_tcp_header
+
+        Util::push_u16_to_u8_array(packet, source_port)
+        Util::push_u16_to_u8_array(packet, destination_port)
+        Util::push_u32_to_u8_array(packet, sequence_number)
+        Util::push_u32_to_u8_array(packet, acknowledgement_number)
+        Util::push_u16_to_u8_array(packet, flags)
+        Util::push_u16_to_u8_array(packet, window_size)
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+        Util::push_u16_to_u8_array(packet, urgent_pointer)
+
+        mut tcp_options_offset = 20
+        mut tcp_option_length = 0
+        while tcp_options_offset < header_length as! i64 {
+            match syn_tcp_packet[tcp_options_offset] {
+                0u8 => {
+                    packet.push(0u8)
+                    tcp_options_offset++
+                }
+                1u8 => {
+                    packet.push(1u8)
+                    tcp_options_offset++
+                }
+                8u8 => {
+                    packet.push(8u8)
+                    packet.push(10u8)
+                    let timestamp: u32 = (session.timestamp_origin + (Time::jiffies() - session.timestamp_origin)) as! u32
+                    Util::push_u32_to_u8_array(packet, timestamp)
+                    Util::push_u32_to_u8_array(packet, session.timestamp_last_echo_reply)
+                    tcp_options_offset += 10
+                }
+                else => {
+                    packet.push(syn_tcp_packet[tcp_options_offset])
+                    packet.push(syn_tcp_packet[tcp_options_offset + 1])
+                    tcp_option_length = syn_tcp_packet[tcp_options_offset + 1] as! i64
+                    for i in 2..tcp_option_length {
+                        packet.push(syn_tcp_packet[tcp_options_offset + i])
+                    }
+                    tcp_options_offset += tcp_option_length
+                }
+            }
+        }
+
+        // Calculate TCP checksum
+
+        mut checksum_packet: [u8] = []
+        // Source and destination IP
+        for i in 26..34 {
+            checksum_packet.push(packet[i])
+        }
+        // Protocol
+        checksum_packet.push(0x00u8)
+        checksum_packet.push(0x06u8)
+        // TCP Packet length
+        let tcp_packet_length: u16 = 40 as! u16
+        Util::push_u16_to_u8_array(checksum_packet, tcp_packet_length)
+
+        for i in 34..packet.size() {
+            checksum_packet.push(packet[i])
+        }
+
+        mut checksum = .calculate_header_checksum(checksum_packet, offset: 0, count: checksum_packet.size() as! i64)
+        if (not .tcp_is_syn(flags)) {
+            checksum += 8
+        }
+        packet[50] = (checksum >> 8) as! u8
+        packet[51] = (checksum & 0xff) as! u8        
+
+        .tx_queue.push(packet)
+        return acknowledgement_number
+
+    }
+    fn tcp_synack_ack_packet(mut this, anon mut session: TcpSession, anon frame: [u8], mut flags: u16) throws -> u32 {
+        mut packet: [u8] = []
+
+        let syn_ipv4_packet = frame[14..]
+        let syn_tcp_packet = frame[34..]
+
+        // Swap source/destination ports
+        mut destination_port: u16 = Util::get_u16_from_u8_arrayslice(syn_tcp_packet, 0)
+        mut source_port: u16 = Util::get_u16_from_u8_arrayslice(syn_tcp_packet, 2)
+
+        let header_length: u16 = (20 + 12) as! u16
+
+        flags += ((header_length / 4) << 12)
+        flags |= 16 // ACK
+
+        mut sequence_number: u32 = session.local_sequence_number
+        mut acknowledgement_number: u32 = 0
+        for i in 0..4 {
+            acknowledgement_number += (syn_tcp_packet[7 - i] as! u32) * match i {
+                1 => 256 as! u32
+                2 => (256 * 256) as! u32
+                3 => (256 * 256 * 256) as! u32
+                else => 1 as! u32
+            }
+        }
+
+        acknowledgement_number += 1
+
+        let window_size: u16 = 0xffff
+        let urgent_pointer: u16 = Util::get_u16_from_u8_arrayslice(syn_tcp_packet, 18)
+
+        mut destination_mac: [u8] = []
+        for octet in frame[6..12] {
+            destination_mac.push(octet)
+        }
+        mut destination_address: [u8] = []
+        for octet in syn_ipv4_packet[12..16] {
+            destination_address.push(octet)
+        }
+
+        mut ipv4_total_length: u16 = 0
+        ipv4_total_length += 20; // IPv4
+        ipv4_total_length += header_length // TCP
+
+        .push_ethernet_header(packet, destination_mac, source_mac: session.local_mac, ethertype: 0x0800)
+        .push_ipv4_header(packet, total_length: ipv4_total_length, flags: 0x40, protocol: 0x06, source_address: .ipv4_address, destination_address)
+
+        // FIXME: put this into .push_tcp_header
+
+        Util::push_u16_to_u8_array(packet, source_port)
+        Util::push_u16_to_u8_array(packet, destination_port)
+        Util::push_u32_to_u8_array(packet, sequence_number)
+        Util::push_u32_to_u8_array(packet, acknowledgement_number)
+        Util::push_u16_to_u8_array(packet, flags)
+        Util::push_u16_to_u8_array(packet, window_size)
+        Util::push_u16_to_u8_array(packet, 0 as! u16) // Checksum placeholder
+        Util::push_u16_to_u8_array(packet, urgent_pointer)
+
+        // Options: NOP, NOP, Timestamp
+        packet.push(0x01u8)
+        packet.push(0x01u8)
+        Util::push_u16_to_u8_array(packet, 0x080a)
+        let timestamp: u32 = (session.timestamp_origin + (Time::jiffies() - session.timestamp_origin)) as! u32
+        Util::push_u32_to_u8_array(packet, timestamp)
+        Util::push_u32_to_u8_array(packet, session.timestamp_last_echo_reply)
+
+        // Calculate TCP checksum
+
+        mut checksum_packet: [u8] = []
+        // Source and destination IP
+        for i in 26..34 {
+            checksum_packet.push(packet[i])
+        }
+        // Protocol
+        checksum_packet.push(0x00u8)
+        checksum_packet.push(0x06u8)
+        // TCP Packet length
+        let tcp_packet_length: u16 = 40 as! u16
+        Util::push_u16_to_u8_array(checksum_packet, tcp_packet_length)
+
+        for i in 34..packet.size() {
+            checksum_packet.push(packet[i])
+        }
+
+        mut checksum = .calculate_header_checksum(checksum_packet, offset: 0, count: checksum_packet.size() as! i64)
+        checksum += 8
+        packet[50] = (checksum >> 8) as! u8
+        packet[51] = (checksum & 0xff) as! u8        
+
+        .tx_queue.push(packet)
+        return acknowledgement_number
+
+    }    
+    fn tcp_session_matches_current_session(mut this, anon session: TcpSession, anon local_address: [u8], anon remote_address: [u8], anon local_port: u16, anon remote_port: u16) -> bool {
+        if session.state as! i64 == TcpSessionState::Closed as! i64 {
+            return false
+        }
+        if (session.local_port != local_port) or (session.remote_port != remote_port) {
+            return false
+        }
+        for i in 0..4 {
+            if (session.local_address[i] != local_address[i]) {
+                return false
+            }
+            if (session.remote_address[i] != remote_address[i]) {
+                return false
+            }
+        }
+        return true
+    }
+    fn tcp_send(mut this, anon mut session: TcpSession, anon data: [u8]) throws {
+        session.pending_data_to_transmit.push_values(&data)
+    }
+    public fn tcp_process_client_send_requests(mut this, anon mac_address: [u8]) throws {
+        for i in 0..this.tcp_sessions.size() {
+            let socket = .tcp_sessions[i].socket
+            if socket > 0 {
+                mut data: [u8] = []
+                unsafe {
+                    cpp {
+                        "u64 *s = (u64*)socket;
+                        if (s[10] == 1) {
+                            u8 *buffer = (u8*)s[7];
+                            for (int i=0; i < s[9]; i++) {
+                                data.push(buffer[i]);
+                            }
+                            s[10] = 0;
+                        }"
+                    }
+                }
+                if data.size() > 0 {
+                    .tcp_send(.tcp_sessions[i], data)
+                }
+            }
+        }
+    }
+    public fn tcp_process_client_received_data(mut this) throws {
+        for i in 0..this.tcp_sessions.size() {
+            mut session = .tcp_sessions[i]
+            let socket = session.socket
+            mut length = session.received_data.size()
+            mut max_length: usize = 0
+            mut close_connection: bool = false
+            // s[4] = 0; // receive_buffer_size
+            if (socket > 0) {
+                unsafe {
+                    cpp {
+                        "u64 *s = (u64*)socket;
+                        max_length = s[4];
+                        close_connection = s[11];
+                        if (close_connection) { s[11] = 0; };"
+                    }
+                }
+            }
+            if close_connection {
+                .tcp_rst_packet(session)
+                .tcp_sessions[i].state = TcpSessionState::Closed
+            }
+            // FIXME: Should the client be responsible for malloc()ing the receive buffer?
+            if length > 65536 {
+                length = 65536
+            }
+            if (length > max_length) {
+                length = max_length
+            }
+            if (socket > 0) and (length == 0) and (.tcp_sessions[i].state as! i64 == TcpSessionState::Closed as! i64) {
+                .tcp_update_socket_session_state(.tcp_sessions[i].state, .tcp_sessions[i].socket)
+                mut client_is_not_ready_to_receive_data = 0
+                unsafe {
+                    cpp {
+                        "u64 *s = (u64*)socket;
+                        client_is_not_ready_to_receive_data = s[6];"
+                    }
+                }
+                let is_bound_socket = .tcp_sessions[i].is_bound_socket
+                if (not is_bound_socket and client_is_not_ready_to_receive_data == 0) or (is_bound_socket) {
+                    // really close the connection
+                    unsafe {
+                        cpp {
+                            "u64 *s = (u64*)socket;
+                            s[5] = 0;
+                            s[6] = 1;
+                            if (s[3] > 0)
+                                free((u8*)s[3]);
+                            if (s[7] > 0)
+                                free((u8*)s[7]);
+                            "
+                        }
+                    }
+                    .tcp_sessions[i].socket = 0
+                }
+            }
+            if socket > 0 and length > 0 {
+                mut client_is_not_ready_to_receive_data = 0
+                unsafe {
+                    cpp {
+                        "u64 *s = (u64*)socket;
+                        client_is_not_ready_to_receive_data = s[6];"
+                    }
+                }
+                if (client_is_not_ready_to_receive_data == 0) {
+                    unsafe {
+                        cpp {
+                            "u64 *s = (u64*)socket;
+                            u8 *buffer = (u8*)s[3];
+                            for (int i=0; i < length; i++) {
+                                buffer[i] = session.received_data[i];
+                            }
+                            s[5] = length;
+                            s[6] = 1;"
+                        }
+                    }
+                    if (length == session.received_data.size() as! usize) {
+                        session.received_data.shrink(0)
+                    } else {
+                        mut remaining_received_data: [u8] = []
+                        for j in length..session.received_data.size() {
+                            remaining_received_data.push(session.received_data[j])
+                        }
+                        session.received_data.shrink(0)
+                        for j in 0..remaining_received_data.size() {
+                            session.received_data.push(remaining_received_data[j])
+                        }
+                    }
+                }
+            }
+        }
+    }
+    fn tcp_is_next_frame(mut this, index: i64, frame: [u8]) throws -> bool {
+        let tcp_packet = frame[34..]
+        mut sequence_number: u32 = 0
+        for o in 0..4 {
+            sequence_number += (tcp_packet[7 - o] as! u32) * match o {
+                1 => 256 as! u32
+                2 => (256 * 256) as! u32
+                3 => (256 * 256 * 256) as! u32
+                else => 1 as! u32
+            }
+        }
+        if (.tcp_sessions[index].acknowledgement_number == 0) {
+            // session not established yet, we're (hopefully) in a SYN/ACK
+            return true
+        }
+        if (sequence_number == .tcp_sessions[index].acknowledgement_number) {
+            return true
+        }
+        return false
+    }
+    fn tcp_handle_received_frame(mut this, anon i: i64, anon frame: [u8]) throws {
+        let ipv4_packet = frame[14..]
+        let tcp_packet = frame[34..]
+
+        let total_length = Util::get_u16_from_u8_arrayslice(ipv4_packet, 2) + 14
+
+        let source_address: [u8] = [ipv4_packet[12], ipv4_packet[13], ipv4_packet[14], ipv4_packet[15]]
+        let destination_address: [u8] = [ipv4_packet[16], ipv4_packet[17], ipv4_packet[18], ipv4_packet[19]]
+
+        mut source_port: u16 = Util::get_u16_from_u8_arrayslice(tcp_packet, 0)
+        mut destination_port: u16 = Util::get_u16_from_u8_arrayslice(tcp_packet, 2)
+
+        let header_length: u16 = ((tcp_packet[12] as! u16) >> 4) * 4
+        let flags: u16 = Util::get_u16_from_u8_arrayslice(tcp_packet, 12) & 0xfff
+
+        mut echo_reply: u32 = 0
+        mut tcp_options_offset = 54
+
+        if (.tcp_sessions[i].state as! i64 == TcpSessionState::Closed as! i64) {
+            return
+        }
+
+        while tcp_options_offset < frame.size() as! i64 {
+            match frame[tcp_options_offset] {
+                0u8 => { tcp_options_offset++ }
+                1u8 => { tcp_options_offset++ }
+                8u8 => {
+                    tcp_options_offset++
+                    echo_reply += (frame[tcp_options_offset + 1] as! u32 * (256 * 256 * 256) as! u32)
+                    echo_reply += (frame[tcp_options_offset + 2] as! u32 * (256 * 256) as! u32)
+                    echo_reply += (frame[tcp_options_offset + 3] as! u32 * 256 as! u32)
+                    echo_reply += frame[tcp_options_offset + 4] as! u32
+                    break
+                }
+                else => { tcp_options_offset += frame[tcp_options_offset + 1] as! i64 }
+            }
+        }
+
+        if (flags & 0xff) != 0x10 {
+            // Received a packet with flags other than just ACK
+            for i in 0..this.tcp_sessions.size() {
+                if (.tcp_session_matches_current_session(.tcp_sessions[i], destination_address, source_address, destination_port, source_port)) {
+                    mut received_data: [u8] = []
+                    for j in frame[(34u16 + header_length)..total_length] {
+                        received_data.push(j)
+                    }
+                    for j in 0..received_data.size() {
+                        .tcp_sessions[i].received_data.push(received_data[j])
+                    }
+                    if (.tcp_is_fin(flags)) {
+                        // Received request to close connection
+                        //println("Closing connection to IPv4 address: {}", source_address)
+                        if (.tcp_is_push(flags)) {
+                            .tcp_sessions[i].acknowledgement_number = .tcp_ack_packet(.tcp_sessions[i], frame, flags: 9u16)
+                        } else {
+                            .tcp_sessions[i].acknowledgement_number = .tcp_ack_packet(.tcp_sessions[i], frame, flags: 1u16)
+                        }
+                        .tcp_sessions[i].state = TcpSessionState::Closed
+                    } else if (.tcp_is_syn(flags)) and (.tcp_is_ack(flags)) {
+                        // Received SYN, ACK
+                        .tcp_sessions[i].timestamp_last_echo_reply = echo_reply
+                        .tcp_sessions[i].timestamp_last_jiffies = Time::jiffies()
+                        .tcp_sessions[i].local_sequence_number++
+                        .tcp_sessions[i].acknowledgement_number = .tcp_synack_ack_packet(.tcp_sessions[i], frame, flags: 0u16)
+                        .tcp_sessions[i].state = TcpSessionState::Established
+                        .tcp_update_socket_session_state(.tcp_sessions[i].state, .tcp_sessions[i].socket)
+                    } else {
+                        .tcp_sessions[i].timestamp_last_echo_reply = echo_reply
+                        .tcp_sessions[i].timestamp_last_jiffies = Time::jiffies()
+                        mut acknowledgement_number: u32 = 0
+                        for o in 0..4 {
+                            acknowledgement_number += (tcp_packet[11 - o] as! u32) * match o {
+                                1 => 256 as! u32
+                                2 => (256 * 256) as! u32
+                                3 => (256 * 256 * 256) as! u32
+                                else => 1 as! u32
+                            }
+                        }
+                        .tcp_sessions[i].local_sequence_number = acknowledgement_number
+                        if (.tcp_is_push(flags)) {
+                            .tcp_sessions[i].acknowledgement_number = .tcp_ack_packet(.tcp_sessions[i], frame, flags: 8u16)
+                        } else {
+                            .tcp_sessions[i].acknowledgement_number = .tcp_ack_packet(.tcp_sessions[i], frame, flags: 0u16)
+                        }
+                    }
+                }
+            }
+        } else if (flags & 0xff) == 0x10 {
+            // Received a packet with flags == just ACK
+            for i in 0..this.tcp_sessions.size() {
+                if (.tcp_session_matches_current_session(.tcp_sessions[i], destination_address, source_address, destination_port, source_port)) {
+                    mut received_data: [u8] = []
+                    for j in frame[(34u16 + header_length)..total_length] {
+                        received_data.push(j)
+                    }
+                    for j in 0..received_data.size() {
+                        .tcp_sessions[i].received_data.push(received_data[j])
+                    }
+                    mut sequence_number: u32 = 0
+                    for o in 0..4 {
+                        sequence_number += (tcp_packet[7 - o] as! u32) * match o {
+                            1 => 256 as! u32
+                            2 => (256 * 256) as! u32
+                            3 => (256 * 256 * 256) as! u32
+                            else => 1 as! u32
+                        }
+                    }
+                    mut acknowledgement_number: u32 = 0
+                    for o in 0..4 {
+                        acknowledgement_number += (tcp_packet[11 - o] as! u32) * match o {
+                            1 => 256 as! u32
+                            2 => (256 * 256) as! u32
+                            3 => (256 * 256 * 256) as! u32
+                            else => 1 as! u32
+                        }
+                    }
+                    let window_size: u16 = 0xffff
+                    if acknowledgement_number == .tcp_sessions[i].last_tx_sequence_number {
+                        .tcp_sessions[i].last_tx_was_acked = true
+                    }
+                    .tcp_sessions[i].local_sequence_number = acknowledgement_number
+                    if received_data.size() > 0 {
+                        .tcp_sessions[i].acknowledgement_number = .tcp_ack_packet(.tcp_sessions[i], frame, flags: 0u16)
+                    } else {
+                        .tcp_sessions[i].acknowledgement_number = sequence_number + (received_data.size() as! u32)
+                    }
+                    .tcp_sessions[i].last_window_size = window_size
+                    .tcp_sessions[i].timestamp_last_echo_reply = echo_reply
+                    .tcp_sessions[i].timestamp_last_jiffies = Time::jiffies()
+                }
+            }
+        }
+    }
+    fn tcp_handle_received_frames(mut this, anon index: i64) throws {
+        mut unhandled_frames: [[u8]] = []
+        for frame in .tcp_sessions[index].received_frames {
+            if .tcp_is_next_frame(index, frame) {
+                .tcp_handle_received_frame(index, frame)
+            } else {
+                unhandled_frames.push(frame)
+            }
+        }
+        .tcp_sessions[index].received_frames.shrink(0)
+        for frame in unhandled_frames {
+            .tcp_sessions[index].received_frames.push(frame)
+        }
+    }
+    fn process_udp_packet(mut this, anon mac_address: [u8], anon frame: [u8]) throws {
+        let ipv4_packet = frame[14..]
+        let udp_packet = frame[34..]
+
+        let source_address: [u8] = [ipv4_packet[12], ipv4_packet[13], ipv4_packet[14], ipv4_packet[15]]
+        let destination_address: [u8] = [ipv4_packet[16], ipv4_packet[17], ipv4_packet[18], ipv4_packet[19]]
+
+        mut source_port: u16 = Util::get_u16_from_u8_arrayslice(udp_packet, 0)
+        mut destination_port: u16 = Util::get_u16_from_u8_arrayslice(udp_packet, 2)
+
+        if (Util::get_hexadecimal_string_from_ipv4_u8_array(source_address) == Util::get_hexadecimal_string_from_ipv4_u8_array(.dns_server_address)) and (source_port == .dns_server_port) {
+            let transaction_id: u16 = Util::get_u16_from_u8_arrayslice(udp_packet, 8)
+            let flags: u16 = Util::get_u16_from_u8_arrayslice(udp_packet, 10)
+            // questions : 12-13
+            // answer_rrs: 14-15
+            // authority_rrs: 16-17
+            // additional_rrs: 18-19
+
+            if .pending_dns_lookups.contains(transaction_id) {
+                mut result: u32 = 0
+                mut result_type: u16 = 0
+                // skip over query data to get answer
+                mut response_pos = 20
+                while (udp_packet[response_pos] > 0) {
+                    response_pos += 1 + udp_packet[response_pos] as! i64
+                }
+                response_pos += 5 // skip over null byte, type, class
+                response_pos += 2 // skip over answer: name
+
+                result_type = 256 * udp_packet[response_pos] as! u16
+                result_type += udp_packet[response_pos + 1] as! u16
+
+                mut pointer_to_u32 = .pending_dns_lookups[transaction_id]
+                mut cached_result: [u8] = []
+
+                match result_type {
+                    1 => {
+                        // A record
+                        response_pos += 10
+                        result += (udp_packet[response_pos] as! u32 * (256 * 256 * 256) as! u32)
+                        result += (udp_packet[response_pos + 1] as! u32 * (256 * 256) as! u32)
+                        result += (udp_packet[response_pos + 2] as! u32 * 256 as! u32)
+                        result += udp_packet[response_pos + 3] as! u32
+                        cached_result.push(udp_packet[response_pos])
+                        cached_result.push(udp_packet[response_pos + 1])
+                        cached_result.push(udp_packet[response_pos + 2])
+                        cached_result.push(udp_packet[response_pos + 3])
+                        .dns_cache[.pending_dns_cached_entries[transaction_id]] = cached_result
+                    }
+                    5 => {
+                        // CNAME record
+                        response_pos += 8 // skip over answer: type, class, ttl
+                        let data_length = (256 * udp_packet[response_pos] as! i64) + udp_packet[response_pos + 1] as! i64
+                        response_pos += 2
+                        let data_pos = response_pos
+                        mut pointer_pos = 0
+
+                        mut s = StringBuilder::create()
+                        mut length = 0
+
+                        while (response_pos < data_pos + data_length) {
+                            match udp_packet[response_pos] {
+                                0xc0 => {
+                                    response_pos++
+                                    pointer_pos = 8 + udp_packet[response_pos++] as! i64
+                                    while (udp_packet[pointer_pos] > 0) {
+                                        length = udp_packet[pointer_pos] as! i64
+                                        pointer_pos++
+                                        for i in 0..length {
+                                            s.append(udp_packet[pointer_pos++])
+                                        }
+                                        if udp_packet[pointer_pos] > 0 {
+                                            s.append('.')
+                                        }
+                                    }
+                                }
+                                else => {
+                                    length = udp_packet[response_pos++] as! i64
+                                    for i in 0..length {
+                                        s.append(udp_packet[response_pos++])
+                                    }
+                                    if udp_packet[response_pos] > 0 {
+                                        s.append('.')
+                                    }
+                                }
+                            }
+                        }
+                        .send_dns_query(mac_address, s.to_string(), pointer_to_u32)
+                        pointer_to_u32 = 0
+                    }
+                    else => {
+                        // Unsupported record
+                        response_pos += 10 // skip over answer: type, class, ttl, data length
+                        result = 0xFFFFFFFF
+                    }
+                }
+                if pointer_to_u32 > 0 {
+                    unsafe {
+                        cpp {
+                            "u32 *r = (u32*)pointer_to_u32;
+                            r[0] = result;"
+                        }
+                    }
+                }
+                .pending_dns_lookups.remove(transaction_id)
+            }
+        }
+
+    }
+    fn is_listening_port(this, anon port: u16) -> bool {
+        return .bound_sockets.contains(port)
+    }
+    fn process_tcp_packet(mut this, anon mac_address: [u8], anon frame: [u8]) throws {
+        let ipv4_packet = frame[14..]
+        let tcp_packet = frame[34..]
+
+        let source_address: [u8] = [ipv4_packet[12], ipv4_packet[13], ipv4_packet[14], ipv4_packet[15]]
+        let destination_address: [u8] = [ipv4_packet[16], ipv4_packet[17], ipv4_packet[18], ipv4_packet[19]]
+
+        mut source_port: u16 = Util::get_u16_from_u8_arrayslice(tcp_packet, 0)
+        mut destination_port: u16 = Util::get_u16_from_u8_arrayslice(tcp_packet, 2)
+
+        let header_length: u16 = ((tcp_packet[12] as! u16) >> 4) * 4
+        let flags: u16 = Util::get_u16_from_u8_arrayslice(tcp_packet, 12) & 0xfff
+
+        mut echo_reply: u32 = 0
+        mut tcp_options_offset = 54
+
+        while tcp_options_offset < frame.size() as! i64 {
+            match frame[tcp_options_offset] {
+                0u8 => { tcp_options_offset++ }
+                1u8 => { tcp_options_offset++ }
+                8u8 => {
+                    tcp_options_offset++
+                    echo_reply += (frame[tcp_options_offset + 1] as! u32 * (256 * 256 * 256) as! u32)
+                    echo_reply += (frame[tcp_options_offset + 2] as! u32 * (256 * 256) as! u32)
+                    echo_reply += (frame[tcp_options_offset + 3] as! u32 * 256 as! u32)
+                    echo_reply += frame[tcp_options_offset + 4] as! u32
+                    break
+                }
+                else => { tcp_options_offset += frame[tcp_options_offset + 1] as! i64 }
+            }
+        }
+
+        if .tcp_is_syn(flags) and not (.tcp_is_ack(flags)) and (.is_listening_port(destination_port)) {
+            // Received SYN, begin a new connection
+            mut remote_mac: [u8] = []
+            for i in 6..12 {
+                remote_mac.push(frame[i])
+            }
+            if .is_local_ipv4_address(source_address) {
+                if not .arp_cache.contains(Util::get_hexadecimal_string_from_ipv4_u8_array(source_address)) {
+                    // send arp request
+                    //println("send arp request for: {}", Util::get_hexadecimal_string_from_ipv4_u8_array(source_address))
+                    .send_arp_request(mac_address, source_address)
+                }
+            }
+            mut socket: u64 = 0
+            let function: u64 = .bound_sockets[destination_port]
+            let source_address_u32 = Util::get_address_u32_from_ipv4_u8_array(source_address)
+            unsafe {
+                cpp {
+                    "
+                    u64* s = (u64*)calloc(256, 1);
+
+                    s[0] = source_address_u32;
+		            s[1] = source_port;
+                    s[2] = 1; // TCP_SOCKET_STATE_ESTABLISHED
+                    s[3] = (u64)calloc(65536, 1); // receive_buffer_ptr
+                    s[4] = 0; // receive_buffer_size
+                    s[5] = 0; // receive_buffer_filled
+                    s[6] = 0; // receive_buffer_kick
+                    s[7] = (u64)calloc(65536, 1); // send_buffer_ptr
+                    s[8] = 65536; // send_buffer_size
+                    s[9] = 0; // send_buffer_filled
+                    s[10] = 0; // send_buffer_kick
+
+                    socket = (u64)s;
+                    if (function && socket) {
+                        os_call(function, socket);
+                    }
+                    "
+                }
+            }
+            mut session = TcpSession(
+                local_mac: mac_address
+                remote_mac: remote_mac
+                local_address: destination_address
+                remote_address: source_address
+                local_port: destination_port
+                remote_port: source_port
+                local_sequence_number: (OS::random() & 0xffffffff) as! u32
+                acknowledgement_number: 0
+                timestamp_last_echo_reply: 0
+                timestamp_last_jiffies: 0
+                timestamp_origin: Time::jiffies()
+                last_window_size: 0
+                last_tx_sequence_number: 0
+                last_tx_was_acked: false
+                tx_chunk_counter: 0
+                state: TcpSessionState::Established
+                pending_data_to_transmit: []
+                received_data: []
+                received_frames: []
+                is_bound_socket: true
+                socket)
+            //println("Accepting connection from IPv4 address: {}", source_address)
+            session.timestamp_last_echo_reply = echo_reply
+            session.timestamp_last_jiffies = Time::jiffies()
+            session.acknowledgement_number = .tcp_ack_packet(session, frame, flags)
+            .tcp_sessions.push(session)
+        } else {
+            // Add packet to list to be processed
+            for i in 0..this.tcp_sessions.size() {
+                if (.tcp_session_matches_current_session(.tcp_sessions[i], destination_address, source_address, destination_port, source_port)) {
+                    .tcp_sessions[i].received_frames.push(frame)
+                    .tcp_handle_received_frames(i as! i64)
+                }
+            }
+        }
+    }
+    public fn process_ipv4_packet(mut this, anon mac_address: [u8], anon frame: [u8]) throws {
+        let ipv4_packet = frame[14..]
+
+        let version = ipv4_packet[0] >> 4
+        let header_length = (ipv4_packet[0] as! i64 & 0xf) * 4
+        let dsf = ipv4_packet[1]
+        let total_length = Util::get_u16_from_u8_arrayslice(ipv4_packet, 2)
+        let identification = Util::get_u16_from_u8_arrayslice(ipv4_packet, 4)
+        let flags = ipv4_packet[6]
+        let ttl = ipv4_packet[8]
+        let protocol = ipv4_packet[9]
+        let checksum = Util::get_u16_from_u8_arrayslice(ipv4_packet, 10)
+        let source_address: [u8] = [ipv4_packet[12], ipv4_packet[13], ipv4_packet[14], ipv4_packet[15]]
+        let destination_address: [u8] = [ipv4_packet[16], ipv4_packet[17], ipv4_packet[18], ipv4_packet[19]]
+
+        match protocol {
+            1 => {
+                .process_icmp_packet(mac_address, frame)
+            }
+            6 => {
+                .process_tcp_packet(mac_address, frame)
+            }
+            17 => {
+                .process_udp_packet(mac_address, frame)
+            }
+            else => {
+                // unsupported
+            }
+        }
+    }
+    public fn set_ipv4_address(mut this, anon ipv4_address: [i64]) {
+        for i in 0..4 {
+            .ipv4_address[i] = ipv4_address[i] as! u8
+        }
+    }
+    public fn set_ipv4_netmask(mut this, anon ipv4_netmask: [i64]) {
+        for i in 0..4 {
+            .ipv4_netmask[i] = ipv4_netmask[i] as! u8
+        }
+    }
+    public fn set_ipv4_network(mut this, anon ipv4_network: [i64]) {
+        for i in 0..4 {
+            .ipv4_network[i] = ipv4_network[i] as! u8
+        }
+    }
+    public fn set_ipv4_gateway(mut this, anon ipv4_gateway: [i64]) {
+        for i in 0..4 {
+            .ipv4_gateway[i] = ipv4_gateway[i] as! u8
+        }
+    }
+    fn is_local_ipv4_address(this, anon check_ipv4_address: [u8]) -> bool {
+        let check_ipv4_address_u32 = Util::get_address_u32_from_ipv4_u8_array(check_ipv4_address)
+        let local_network_address_u32 = Util::get_address_u32_from_ipv4_u8_array(.ipv4_network)
+        let local_netmask_address_u32 = Util::get_address_u32_from_ipv4_u8_array(.ipv4_netmask)
+        // NOTE: The line below was added and the subsequent line was modified after jakt PR #1596 to suppress a warning from codegen output
+        // tcpip.cpp:2599:90: warning: & has lower precedence than ==; == will be evaluated first [-Wparentheses]
+        // 2599 | return (local_network_address_u32 & local_netmask_address_u32) == check_ipv4_address_u32 & local_netmask_address_u32;
+            // |        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
+        // tcpip.cpp:2599:90: note: place parentheses around the '==' expression to silence this warning
+        let check_ipv4_address_u32_and_with_netmask = check_ipv4_address_u32 & local_netmask_address_u32
+        return (local_network_address_u32 & local_netmask_address_u32) == check_ipv4_address_u32_and_with_netmask
+    }
+    fn tcp_update_socket_session_state(mut this, anon state: TcpSessionState, anon socket: u64) throws {
+        unsafe {
+            cpp {
+                "u64 *s = (u64*)socket;
+                s[2] = (u64)state;
+                "
+            }
+        }
+    }
+    public fn tcp_process_bind_request(mut this) {
+        mut did_receive_request = false
+        mut port: u16 = 0
+        mut function: u64 = 0
+        mut response_code: u64 = 0
+        unsafe {
+            cpp {
+                "
+                u64 *request = (u64*)0x300040;
+                if (*request) {
+                    did_receive_request = true;
+                    u64 *b = (u64*)*request;
+                    port = b[0];
+                    function = b[1];
+                }
+                "
+            }
+        }
+        if did_receive_request {
+            if .bound_sockets.contains(port) {
+                response_code = TcpBindError::SocketIsAlreadyBound as! u64
+            } else if port < 1 or function < 1 {
+                response_code = TcpBindError::BadRequest as! u64
+            } else {
+                .bound_sockets[port] = function
+            }
+            unsafe {
+                cpp {
+                    "
+                    u64 *request = (u64*)0x300040;
+                    u64 *b = (u64*)*request;
+                    b[2] = response_code;
+                    *request = 0;
+                    "
+                }
+            }
+        }
+    }
+    public fn tcp_process_client_socket_request(mut this, anon mac_address: [u8]) throws {
+        mut did_receive_request = false
+        mut remote_address: u32 = 0
+        mut remote_port: u16 = 0
+        mut session_remote_mac: [u8] = []
+        mut socket: u64 = 0
+        unsafe {
+            cpp {
+                "u64 *ls_request = (u64*)0x300000;
+                if (*ls_request) {
+                    u64 *ls_s = (u64*)*ls_request;
+                    remote_address = ls_s[0];
+                }"
+            }
+        }
+        if (remote_address > 0) {
+            mut remote_address_hex_string = Util::get_hexadecimal_string_from_ipv4_u8_array(.ipv4_gateway)
+            if .is_local_ipv4_address(Util::get_ipv4_u8_array_from_address_u32(remote_address)) {
+                remote_address_hex_string = Util::get_hexadecimal_string_from_ipv4_u8_array(Util::get_ipv4_u8_array_from_address_u32(remote_address))
+                if not .arp_cache.contains(remote_address_hex_string) {
+                    // send arp request
+                    //println("send arp request for: {}", remote_address_hex_string)
+                    .send_arp_request(mac_address, Util::get_ipv4_u8_array_from_address_u32(remote_address))
+                    return
+                }
+            }
+            session_remote_mac = .arp_cache[remote_address_hex_string]
+        } else {
+            return
+        }
+        unsafe {
+            cpp {
+                "u64 *request = (u64*)0x300000;
+                if (*request) {
+                    did_receive_request = true;
+                    u64 *s = (u64*)*request;
+
+                    s[2] = 5; // TCP_SOCKET_STATE_IDLE
+
+                    s[3] = (u64)calloc(65536, 1); // receive_buffer_ptr
+                    s[4] = 0; // receive_buffer_size
+                    s[5] = 0; // receive_buffer_filled
+                    s[6] = 1; // receive_buffer_kick
+
+                    s[7] = (u64)calloc(65536, 1); // send_buffer_ptr
+                    s[8] = 65536; // send_buffer_size
+                    s[9] = 0; // send_buffer_filled
+                    s[10] = 0; // send_buffer_kick
+
+                    remote_address = s[0];
+                    remote_port = s[1];
+                    socket = (u64)s;
+                    *request = 0;
+                }" 
+            }
+        }
+        if (did_receive_request) {
+            let local_sequence_number: u32 = (OS::random() & 0xffffffff) as! u32
+
+            mut session = TcpSession(
+                local_mac: mac_address
+                remote_mac: session_remote_mac
+                local_address: .ipv4_address
+                remote_address: Util::get_ipv4_u8_array_from_address_u32(remote_address)
+                local_port: (OS::random() & 0xffff) as! u16 // FIXME
+                remote_port
+                local_sequence_number
+                acknowledgement_number: 0
+                timestamp_last_echo_reply: 0
+                timestamp_last_jiffies: Time::jiffies()
+                timestamp_origin: Time::jiffies()
+                last_window_size: 0xffff
+                last_tx_sequence_number: 0
+                last_tx_was_acked: false
+                tx_chunk_counter: 0
+                state: TcpSessionState::Connecting
+                pending_data_to_transmit: []
+                received_data: []
+                received_frames: []
+                is_bound_socket: false
+                socket)
+            .tcp_syn_packet(session)
+            .tcp_update_socket_session_state(session.state, session.socket)
+            .tcp_sessions.push(session)
+            did_receive_request = false
+        }
+    }
+}
\ No newline at end of file