Browse Source

Merge pull request #83281 from capnm/thorvg_0.11.1

ThorVG: update to v0.11.1
Rémi Verschelde 1 year ago
parent
commit
9ca1d78d53

+ 5 - 0
modules/svg/SCsub

@@ -24,6 +24,9 @@ thirdparty_sources = [
     "src/loaders/svg/tvgSvgUtil.cpp",
     "src/loaders/svg/tvgXmlParser.cpp",
     "src/loaders/raw/tvgRawLoader.cpp",
+    "src/loaders/external_png/tvgPngLoader.cpp",
+    "src/loaders/jpg/tvgJpgd.cpp",
+    "src/loaders/jpg/tvgJpgLoader.cpp",
     # renderer common
     "src/renderer/tvgAccessor.cpp",
     # "src/renderer/tvgAnimation.cpp",
@@ -68,6 +71,8 @@ env_thirdparty.Prepend(
         thirdparty_dir + "src/renderer",
         thirdparty_dir + "src/renderer/sw_engine",
         thirdparty_dir + "src/loaders/raw",
+        thirdparty_dir + "src/loaders/external_png",
+        thirdparty_dir + "src/loaders/jpg",
     ]
 )
 

+ 7 - 2
modules/text_server_adv/gdextension_build/SConstruct

@@ -55,6 +55,9 @@ if env["thorvg_enabled"] and env["freetype_enabled"]:
         "src/loaders/svg/tvgSvgUtil.cpp",
         "src/loaders/svg/tvgXmlParser.cpp",
         "src/loaders/raw/tvgRawLoader.cpp",
+        "src/loaders/external_png/tvgPngLoader.cpp",
+        "src/loaders/jpg/tvgJpgd.cpp",
+        "src/loaders/jpg/tvgJpgLoader.cpp",
         # renderer common
         "src/renderer/tvgAccessor.cpp",
         # "src/renderer/tvgAnimation.cpp",
@@ -88,10 +91,12 @@ if env["thorvg_enabled"] and env["freetype_enabled"]:
         CPPPATH=[
             "../../../thirdparty/thorvg/inc",
             "../../../thirdparty/thorvg/src/common",
-            "../../../thirdparty/thorvg/src/loaders/svg",
-            "../../../thirdparty/thorvg/src/loaders/raw",
             "../../../thirdparty/thorvg/src/renderer",
             "../../../thirdparty/thorvg/src/renderer/sw_engine",
+            "../../../thirdparty/thorvg/src/loaders/svg",
+            "../../../thirdparty/thorvg/src/loaders/raw",
+            "../../../thirdparty/thorvg/src/loaders/external_png",
+            "../../../thirdparty/thorvg/src/loaders/jpg",
         ]
     )
 

+ 7 - 2
modules/text_server_fb/gdextension_build/SConstruct

@@ -50,6 +50,9 @@ if env["thorvg_enabled"] and env["freetype_enabled"]:
         "src/loaders/svg/tvgSvgUtil.cpp",
         "src/loaders/svg/tvgXmlParser.cpp",
         "src/loaders/raw/tvgRawLoader.cpp",
+        "src/loaders/external_png/tvgPngLoader.cpp",
+        "src/loaders/jpg/tvgJpgd.cpp",
+        "src/loaders/jpg/tvgJpgLoader.cpp",
         # renderer common
         "src/renderer/tvgAccessor.cpp",
         # "src/renderer/tvgAnimation.cpp",
@@ -83,10 +86,12 @@ if env["thorvg_enabled"] and env["freetype_enabled"]:
         CPPPATH=[
             "../../../thirdparty/thorvg/inc",
             "../../../thirdparty/thorvg/src/common",
-            "../../../thirdparty/thorvg/src/loaders/svg",
-            "../../../thirdparty/thorvg/src/loaders/raw",
             "../../../thirdparty/thorvg/src/renderer",
             "../../../thirdparty/thorvg/src/renderer/sw_engine",
+            "../../../thirdparty/thorvg/src/loaders/svg",
+            "../../../thirdparty/thorvg/src/loaders/raw",
+            "../../../thirdparty/thorvg/src/loaders/external_png",
+            "../../../thirdparty/thorvg/src/loaders/jpg",
         ]
     )
 

+ 2 - 1
thirdparty/README.md

@@ -804,7 +804,7 @@ instead of `miniz.h` as an external dependency.
 ## thorvg
 
 - Upstream: https://github.com/thorvg/thorvg
-- Version: 0.11.0 (12260198d12719ea20939b68492accfc155d9ff5, 2023)
+- Version: 0.11.1 (ca00e52125446a1a5cca20f9d8621b382cff5cb9, 2023)
 - License: MIT
 
 Files extracted from upstream source:
@@ -812,6 +812,7 @@ Files extracted from upstream source:
 See `thorvg/update-thorvg.sh` for extraction instructions. Set the version
 number and run the script.
 
+Apply patches from the `patches` folder.
 
 ## vhacd
 

+ 3 - 1
thirdparty/thorvg/inc/config.h

@@ -3,9 +3,11 @@
 
 #define THORVG_SW_RASTER_SUPPORT
 #define THORVG_SVG_LOADER_SUPPORT
+#define THORVG_PNG_LOADER_SUPPORT
+#define THORVG_JPG_LOADER_SUPPORT
 
 // For internal debugging:
 //#define THORVG_LOG_ENABLED
 
-#define THORVG_VERSION_STRING "0.11.0"
+#define THORVG_VERSION_STRING "0.11.1"
 #endif

+ 13 - 0
thirdparty/thorvg/patches/loader_jpg-fix_regression_bug.patch

@@ -0,0 +1,13 @@
+diff --git a/thirdparty/thorvg/src/loaders/jpg/tvgJpgd.cpp b/thirdparty/thorvg/src/loaders/jpg/tvgJpgd.cpp
+index 88d359aaa3..61a5dc1c0f 100644
+--- a/thirdparty/thorvg/src/loaders/jpg/tvgJpgd.cpp
++++ b/thirdparty/thorvg/src/loaders/jpg/tvgJpgd.cpp
+@@ -431,7 +431,7 @@ struct Row<1>
+ {
+     static void idct(int* pTemp, const jpgd_block_t* pSrc)
+     {
+-        const int dcval = pSrc[0] * (pSrc[0] * (PASS1_BITS * 2));
++        const int dcval = pSrc[0] * PASS1_BITS * 2;
+ 
+         pTemp[0] = dcval;
+         pTemp[1] = dcval;

+ 120 - 0
thirdparty/thorvg/src/loaders/external_png/tvgPngLoader.cpp

@@ -0,0 +1,120 @@
+/*
+ * Copyright (c) 2020 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+#include "tvgLoader.h"
+#include "tvgPngLoader.h"
+
+/************************************************************************/
+/* Internal Class Implementation                                        */
+/************************************************************************/
+
+
+/************************************************************************/
+/* External Class Implementation                                        */
+/************************************************************************/
+
+PngLoader::PngLoader()
+{
+    image = static_cast<png_imagep>(calloc(1, sizeof(png_image)));
+    image->version = PNG_IMAGE_VERSION;
+    image->opaque = NULL;
+}
+
+PngLoader::~PngLoader()
+{
+    if (content) {
+        free((void*)content);
+        content = nullptr;
+    }
+    free(image);
+}
+
+bool PngLoader::open(const string& path)
+{
+    image->opaque = NULL;
+
+    if (!png_image_begin_read_from_file(image, path.c_str())) return false;
+
+    w = (float)image->width;
+    h = (float)image->height;
+    cs = ColorSpace::ARGB8888;
+
+    return true;
+}
+
+bool PngLoader::open(const char* data, uint32_t size, bool copy)
+{
+    image->opaque = NULL;
+
+    if (!png_image_begin_read_from_memory(image, data, size)) return false;
+
+    w = (float)image->width;
+    h = (float)image->height;
+    cs = ColorSpace::ARGB8888;
+
+    return true;
+}
+
+
+bool PngLoader::read()
+{
+    png_bytep buffer;
+    image->format = PNG_FORMAT_BGRA;
+    buffer = static_cast<png_bytep>(malloc(PNG_IMAGE_SIZE((*image))));
+    if (!buffer) {
+        //out of memory, only time when libpng doesnt free its data
+        png_image_free(image);
+        return false;
+    }
+    if (!png_image_finish_read(image, NULL, buffer, 0, NULL)) {
+        free(buffer);
+        return false;
+    }
+    content = reinterpret_cast<uint32_t*>(buffer);
+
+    return true;
+}
+
+bool PngLoader::close()
+{
+    png_image_free(image);
+    return true;
+}
+
+unique_ptr<Surface> PngLoader::bitmap()
+{
+    if (!content) return nullptr;
+
+    //TODO: It's better to keep this surface instance in the loader side
+    auto surface = new Surface;
+    surface->buf32 = content;
+    surface->stride = w;
+    surface->w = w;
+    surface->h = h;
+    surface->cs = cs;
+    surface->channelSize = sizeof(uint32_t);
+    surface->owner = true;
+    surface->premultiplied = false;
+
+    return unique_ptr<Surface>(surface);
+}
+

+ 47 - 0
thirdparty/thorvg/src/loaders/external_png/tvgPngLoader.h

@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2020 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+#ifndef _TVG_PNG_LOADER_H_
+#define _TVG_PNG_LOADER_H_
+
+#include <png.h>
+
+class PngLoader : public LoadModule
+{
+public:
+    PngLoader();
+    ~PngLoader();
+
+    using LoadModule::open;
+    bool open(const string& path) override;
+    bool open(const char* data, uint32_t size, bool copy) override;
+    bool read() override;
+    bool close() override;
+
+    unique_ptr<Surface> bitmap() override;
+
+private:
+    png_imagep image = nullptr;
+    uint32_t* content = nullptr;
+};
+
+#endif //_TVG_PNG_LOADER_H_

+ 143 - 0
thirdparty/thorvg/src/loaders/jpg/tvgJpgLoader.cpp

@@ -0,0 +1,143 @@
+/*
+ * Copyright (c) 2021 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+#include <memory.h>
+#include "tvgLoader.h"
+#include "tvgJpgLoader.h"
+
+/************************************************************************/
+/* Internal Class Implementation                                        */
+/************************************************************************/
+
+void JpgLoader::clear()
+{
+    jpgdDelete(decoder);
+    if (freeData) free(data);
+    decoder = nullptr;
+    data = nullptr;
+    freeData = false;
+}
+
+
+/************************************************************************/
+/* External Class Implementation                                        */
+/************************************************************************/
+
+
+JpgLoader::~JpgLoader()
+{
+    jpgdDelete(decoder);
+    if (freeData) free(data);
+    free(image);
+}
+
+
+bool JpgLoader::open(const string& path)
+{
+    clear();
+
+    int width, height;
+    decoder = jpgdHeader(path.c_str(), &width, &height);
+    if (!decoder) return false;
+
+    w = static_cast<float>(width);
+    h = static_cast<float>(height);
+    cs = ColorSpace::ARGB8888;
+
+    return true;
+}
+
+
+bool JpgLoader::open(const char* data, uint32_t size, bool copy)
+{
+    clear();
+
+    if (copy) {
+        this->data = (char *) malloc(size);
+        if (!this->data) return false;
+        memcpy((char *)this->data, data, size);
+        freeData = true;
+    } else {
+        this->data = (char *) data;
+        freeData = false;
+    }
+
+    int width, height;
+    decoder = jpgdHeader(this->data, size, &width, &height);
+    if (!decoder) return false;
+
+    w = static_cast<float>(width);
+    h = static_cast<float>(height);
+    cs = ColorSpace::ARGB8888;
+
+    return true;
+}
+
+
+
+bool JpgLoader::read()
+{
+    if (!decoder || w <= 0 || h <= 0) return false;
+
+    TaskScheduler::request(this);
+
+    return true;
+}
+
+
+bool JpgLoader::close()
+{
+    this->done();
+    clear();
+    return true;
+}
+
+
+unique_ptr<Surface> JpgLoader::bitmap()
+{
+    this->done();
+
+    if (!image) return nullptr;
+
+    //TODO: It's better to keep this surface instance in the loader side
+    auto surface = new Surface;
+    surface->buf8 = image;
+    surface->stride = static_cast<uint32_t>(w);
+    surface->w = static_cast<uint32_t>(w);
+    surface->h = static_cast<uint32_t>(h);
+    surface->cs = cs;
+    surface->channelSize = sizeof(uint32_t);
+    surface->premultiplied = true;
+    surface->owner = true;
+
+    return unique_ptr<Surface>(surface);
+}
+
+
+void JpgLoader::run(unsigned tid)
+{
+    if (image) {
+        free(image);
+        image = nullptr;
+    }
+    image = jpgdDecompress(decoder);
+}

+ 52 - 0
thirdparty/thorvg/src/loaders/jpg/tvgJpgLoader.h

@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2021 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+#ifndef _TVG_JPG_LOADER_H_
+#define _TVG_JPG_LOADER_H_
+
+#include "tvgTaskScheduler.h"
+#include "tvgJpgd.h"
+
+class JpgLoader : public LoadModule, public Task
+{
+private:
+    jpeg_decoder* decoder = nullptr;
+    char* data = nullptr;
+    unsigned char *image = nullptr;
+    bool freeData = false;
+
+    void clear();
+
+public:
+    ~JpgLoader();
+
+    using LoadModule::open;
+    bool open(const string& path) override;
+    bool open(const char* data, uint32_t size, bool copy) override;
+    bool read() override;
+    bool close() override;
+
+    unique_ptr<Surface> bitmap() override;
+    void run(unsigned tid) override;
+};
+
+#endif //_TVG_JPG_LOADER_H_

+ 3026 - 0
thirdparty/thorvg/src/loaders/jpg/tvgJpgd.cpp

@@ -0,0 +1,3026 @@
+/*
+ * Copyright (c) 2021 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+// jpgd.cpp - C++ class for JPEG decompression.
+// Public domain, Rich Geldreich <[email protected]>
+// Alex Evans: Linear memory allocator (taken from jpge.h).
+// v1.04, May. 19, 2012: Code tweaks to fix VS2008 static code analysis warnings (all looked harmless)
+//
+// Supports progressive and baseline sequential JPEG image files, and the most common chroma subsampling factors: Y, H1V1, H2V1, H1V2, and H2V2.
+//
+// Chroma upsampling quality: H2V2 is upsampled in the frequency domain, H2V1 and H1V2 are upsampled using point sampling.
+// Chroma upsampling reference: "Fast Scheme for Image Size Change in the Compressed Domain"
+// http://vision.ai.uiuc.edu/~dugad/research/dct/index.html
+
+#include <memory.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <setjmp.h>
+#include <stdint.h>
+#include "tvgJpgd.h"
+
+#ifdef _MSC_VER
+  #pragma warning (disable : 4611) // warning C4611: interaction between '_setjmp' and C++ object destruction is non-portable
+  #define JPGD_NORETURN __declspec(noreturn)
+#elif defined(__GNUC__)
+  #define JPGD_NORETURN __attribute__ ((noreturn))
+#else
+  #define JPGD_NORETURN
+#endif
+
+/************************************************************************/
+/* Internal Class Implementation                                        */
+/************************************************************************/
+
+
+// Set to 1 to enable freq. domain chroma upsampling on images using H2V2 subsampling (0=faster nearest neighbor sampling).
+// This is slower, but results in higher quality on images with highly saturated colors.
+#define JPGD_SUPPORT_FREQ_DOMAIN_UPSAMPLING 1
+
+#define JPGD_ASSERT(x)
+#define JPGD_MAX(a,b) (((a)>(b)) ? (a) : (b))
+#define JPGD_MIN(a,b) (((a)<(b)) ? (a) : (b))
+
+typedef int16_t jpgd_quant_t;
+typedef int16_t jpgd_block_t;
+
+// Success/failure error codes.
+enum jpgd_status
+{
+    JPGD_SUCCESS = 0, JPGD_FAILED = -1, JPGD_DONE = 1,
+    JPGD_BAD_DHT_COUNTS = -256, JPGD_BAD_DHT_INDEX, JPGD_BAD_DHT_MARKER, JPGD_BAD_DQT_MARKER, JPGD_BAD_DQT_TABLE,
+    JPGD_BAD_PRECISION, JPGD_BAD_HEIGHT, JPGD_BAD_WIDTH, JPGD_TOO_MANY_COMPONENTS,
+    JPGD_BAD_SOF_LENGTH, JPGD_BAD_VARIABLE_MARKER, JPGD_BAD_DRI_LENGTH, JPGD_BAD_SOS_LENGTH,
+    JPGD_BAD_SOS_COMP_ID, JPGD_W_EXTRA_BYTES_BEFORE_MARKER, JPGD_NO_ARITHMITIC_SUPPORT, JPGD_UNEXPECTED_MARKER,
+    JPGD_NOT_JPEG, JPGD_UNSUPPORTED_MARKER, JPGD_BAD_DQT_LENGTH, JPGD_TOO_MANY_BLOCKS,
+    JPGD_UNDEFINED_QUANT_TABLE, JPGD_UNDEFINED_HUFF_TABLE, JPGD_NOT_SINGLE_SCAN, JPGD_UNSUPPORTED_COLORSPACE,
+    JPGD_UNSUPPORTED_SAMP_FACTORS, JPGD_DECODE_ERROR, JPGD_BAD_RESTART_MARKER, JPGD_ASSERTION_ERROR,
+    JPGD_BAD_SOS_SPECTRAL, JPGD_BAD_SOS_SUCCESSIVE, JPGD_STREAM_READ, JPGD_NOTENOUGHMEM
+};
+
+enum
+{
+    JPGD_IN_BUF_SIZE = 8192, JPGD_MAX_BLOCKS_PER_MCU = 10, JPGD_MAX_HUFF_TABLES = 8, JPGD_MAX_QUANT_TABLES = 4,
+    JPGD_MAX_COMPONENTS = 4, JPGD_MAX_COMPS_IN_SCAN = 4, JPGD_MAX_BLOCKS_PER_ROW = 8192, JPGD_MAX_HEIGHT = 16384, JPGD_MAX_WIDTH = 16384
+};
+
+// Input stream interface.
+// Derive from this class to read input data from sources other than files or memory. Set m_eof_flag to true when no more data is available.
+// The decoder is rather greedy: it will keep on calling this method until its internal input buffer is full, or until the EOF flag is set.
+// It the input stream contains data after the JPEG stream's EOI (end of image) marker it will probably be pulled into the internal buffer.
+// Call the get_total_bytes_read() method to determine the actual size of the JPEG stream after successful decoding.
+struct jpeg_decoder_stream
+{
+    jpeg_decoder_stream() { }
+    virtual ~jpeg_decoder_stream() { }
+
+    // The read() method is called when the internal input buffer is empty.
+    // Parameters:
+    // pBuf - input buffer
+    // max_bytes_to_read - maximum bytes that can be written to pBuf
+    // pEOF_flag - set this to true if at end of stream (no more bytes remaining)
+    // Returns -1 on error, otherwise return the number of bytes actually written to the buffer (which may be 0).
+    // Notes: This method will be called in a loop until you set *pEOF_flag to true or the internal buffer is full.
+    virtual int read(uint8_t *pBuf, int max_bytes_to_read, bool *pEOF_flag) = 0;
+};
+
+
+// stdio FILE stream class.
+class jpeg_decoder_file_stream : public jpeg_decoder_stream
+{
+    jpeg_decoder_file_stream(const jpeg_decoder_file_stream &);
+    jpeg_decoder_file_stream &operator =(const jpeg_decoder_file_stream &);
+
+    FILE *m_pFile = nullptr;
+    bool m_eof_flag = false;
+    bool m_error_flag = false;
+
+public:
+    jpeg_decoder_file_stream() {}
+    virtual ~jpeg_decoder_file_stream();
+    bool open(const char *Pfilename);
+    void close();
+    virtual int read(uint8_t *pBuf, int max_bytes_to_read, bool *pEOF_flag);
+  };
+
+
+// Memory stream class.
+class jpeg_decoder_mem_stream : public jpeg_decoder_stream
+{
+    const uint8_t *m_pSrc_data;
+    uint32_t m_ofs, m_size;
+
+public:
+    jpeg_decoder_mem_stream() : m_pSrc_data(nullptr), m_ofs(0), m_size(0) {}
+    jpeg_decoder_mem_stream(const uint8_t *pSrc_data, uint32_t size) : m_pSrc_data(pSrc_data), m_ofs(0), m_size(size) {}
+    virtual ~jpeg_decoder_mem_stream() {}
+    bool open(const uint8_t *pSrc_data, uint32_t size);
+    void close() { m_pSrc_data = nullptr; m_ofs = 0; m_size = 0; }
+    virtual int read(uint8_t *pBuf, int max_bytes_to_read, bool *pEOF_flag);
+};
+
+
+class jpeg_decoder
+{
+public:
+    // Call get_error_code() after constructing to determine if the stream is valid or not. You may call the get_width(), get_height(), etc.
+    // methods after the constructor is called. You may then either destruct the object, or begin decoding the image by calling begin_decoding(), then decode() on each scanline.
+    jpeg_decoder(jpeg_decoder_stream *pStream);
+    ~jpeg_decoder();
+
+    // Call this method after constructing the object to begin decompression.
+    // If JPGD_SUCCESS is returned you may then call decode() on each scanline.
+    int begin_decoding();
+    // Returns the next scan line.
+    // For grayscale images, pScan_line will point to a buffer containing 8-bit pixels (get_bytes_per_pixel() will return 1).
+    // Otherwise, it will always point to a buffer containing 32-bit RGBA pixels (A will always be 255, and get_bytes_per_pixel() will return 4).
+    // Returns JPGD_SUCCESS if a scan line has been returned.
+    // Returns JPGD_DONE if all scan lines have been returned.
+    // Returns JPGD_FAILED if an error occurred. Call get_error_code() for a more info.
+    int decode(const void** pScan_line, uint32_t* pScan_line_len);
+    inline jpgd_status get_error_code() const { return m_error_code; }
+    inline int get_width() const { return m_image_x_size; }
+    inline int get_height() const { return m_image_y_size; }
+    inline int get_num_components() const { return m_comps_in_frame; }
+    inline int get_bytes_per_pixel() const { return m_dest_bytes_per_pixel; }
+    inline int get_bytes_per_scan_line() const { return m_image_x_size * get_bytes_per_pixel(); }
+    // Returns the total number of bytes actually consumed by the decoder (which should equal the actual size of the JPEG file).
+    inline int get_total_bytes_read() const { return m_total_bytes_read; }
+
+private:
+    jpeg_decoder(const jpeg_decoder &);
+    jpeg_decoder &operator =(const jpeg_decoder &);
+
+    typedef void (*pDecode_block_func)(jpeg_decoder *, int, int, int);
+
+    struct huff_tables
+    {
+      bool ac_table;
+      uint32_t  look_up[256];
+      uint32_t  look_up2[256];
+      uint8_t code_size[256];
+      uint32_t  tree[512];
+    };
+
+    struct coeff_buf
+    {
+      uint8_t *pData;
+      int block_num_x, block_num_y;
+      int block_len_x, block_len_y;
+      int block_size;
+    };
+
+    struct mem_block
+    {
+      mem_block *m_pNext;
+      size_t m_used_count;
+      size_t m_size;
+      char m_data[1];
+    };
+
+    jmp_buf m_jmp_state;
+    mem_block *m_pMem_blocks;
+    int m_image_x_size;
+    int m_image_y_size;
+    jpeg_decoder_stream *m_pStream;
+    int m_progressive_flag;
+    uint8_t m_huff_ac[JPGD_MAX_HUFF_TABLES];
+    uint8_t* m_huff_num[JPGD_MAX_HUFF_TABLES];      // pointer to number of Huffman codes per bit size
+    uint8_t* m_huff_val[JPGD_MAX_HUFF_TABLES];      // pointer to Huffman codes per bit size
+    jpgd_quant_t* m_quant[JPGD_MAX_QUANT_TABLES]; // pointer to quantization tables
+    int m_scan_type;                              // Gray, Yh1v1, Yh1v2, Yh2v1, Yh2v2 (CMYK111, CMYK4114 no longer supported)
+    int m_comps_in_frame;                         // # of components in frame
+    int m_comp_h_samp[JPGD_MAX_COMPONENTS];       // component's horizontal sampling factor
+    int m_comp_v_samp[JPGD_MAX_COMPONENTS];       // component's vertical sampling factor
+    int m_comp_quant[JPGD_MAX_COMPONENTS];        // component's quantization table selector
+    int m_comp_ident[JPGD_MAX_COMPONENTS];        // component's ID
+    int m_comp_h_blocks[JPGD_MAX_COMPONENTS];
+    int m_comp_v_blocks[JPGD_MAX_COMPONENTS];
+    int m_comps_in_scan;                          // # of components in scan
+    int m_comp_list[JPGD_MAX_COMPS_IN_SCAN];      // components in this scan
+    int m_comp_dc_tab[JPGD_MAX_COMPONENTS];       // component's DC Huffman coding table selector
+    int m_comp_ac_tab[JPGD_MAX_COMPONENTS];       // component's AC Huffman coding table selector
+    int m_spectral_start;                         // spectral selection start
+    int m_spectral_end;                           // spectral selection end
+    int m_successive_low;                         // successive approximation low
+    int m_successive_high;                        // successive approximation high
+    int m_max_mcu_x_size;                         // MCU's max. X size in pixels
+    int m_max_mcu_y_size;                         // MCU's max. Y size in pixels
+    int m_blocks_per_mcu;
+    int m_max_blocks_per_row;
+    int m_mcus_per_row, m_mcus_per_col;
+    int m_mcu_org[JPGD_MAX_BLOCKS_PER_MCU];
+    int m_total_lines_left;                       // total # lines left in image
+    int m_mcu_lines_left;                         // total # lines left in this MCU
+    int m_real_dest_bytes_per_scan_line;
+    int m_dest_bytes_per_scan_line;               // rounded up
+    int m_dest_bytes_per_pixel;                   // 4 (RGB) or 1 (Y)
+    huff_tables* m_pHuff_tabs[JPGD_MAX_HUFF_TABLES];
+    coeff_buf* m_dc_coeffs[JPGD_MAX_COMPONENTS];
+    coeff_buf* m_ac_coeffs[JPGD_MAX_COMPONENTS];
+    int m_eob_run;
+    int m_block_y_mcu[JPGD_MAX_COMPONENTS];
+    uint8_t* m_pIn_buf_ofs;
+    int m_in_buf_left;
+    int m_tem_flag;
+    bool m_eof_flag;
+    uint8_t m_in_buf_pad_start[128];
+    uint8_t m_in_buf[JPGD_IN_BUF_SIZE + 128];
+    uint8_t m_in_buf_pad_end[128];
+    int m_bits_left;
+    uint32_t m_bit_buf;
+    int m_restart_interval;
+    int m_restarts_left;
+    int m_next_restart_num;
+    int m_max_mcus_per_row;
+    int m_max_blocks_per_mcu;
+    int m_expanded_blocks_per_mcu;
+    int m_expanded_blocks_per_row;
+    int m_expanded_blocks_per_component;
+    bool  m_freq_domain_chroma_upsample;
+    int m_max_mcus_per_col;
+    uint32_t m_last_dc_val[JPGD_MAX_COMPONENTS];
+    jpgd_block_t* m_pMCU_coefficients;
+    int m_mcu_block_max_zag[JPGD_MAX_BLOCKS_PER_MCU];
+    uint8_t* m_pSample_buf;
+    int m_crr[256];
+    int m_cbb[256];
+    int m_crg[256];
+    int m_cbg[256];
+    uint8_t* m_pScan_line_0;
+    uint8_t* m_pScan_line_1;
+    jpgd_status m_error_code;
+    bool m_ready_flag;
+    int m_total_bytes_read;
+
+    void free_all_blocks();
+    JPGD_NORETURN void stop_decoding(jpgd_status status);
+    void *alloc(size_t n, bool zero = false);
+    void word_clear(void *p, uint16_t c, uint32_t n);
+    void prep_in_buffer();
+    void read_dht_marker();
+    void read_dqt_marker();
+    void read_sof_marker();
+    void skip_variable_marker();
+    void read_dri_marker();
+    void read_sos_marker();
+    int next_marker();
+    int process_markers();
+    void locate_soi_marker();
+    void locate_sof_marker();
+    int locate_sos_marker();
+    void init(jpeg_decoder_stream * pStream);
+    void create_look_ups();
+    void fix_in_buffer();
+    void transform_mcu(int mcu_row);
+    void transform_mcu_expand(int mcu_row);
+    coeff_buf* coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y);
+    inline jpgd_block_t *coeff_buf_getp(coeff_buf *cb, int block_x, int block_y);
+    void load_next_row();
+    void decode_next_row();
+    void make_huff_table(int index, huff_tables *pH);
+    void check_quant_tables();
+    void check_huff_tables();
+    void calc_mcu_block_order();
+    int init_scan();
+    void init_frame();
+    void process_restart();
+    void decode_scan(pDecode_block_func decode_block_func);
+    void init_progressive();
+    void init_sequential();
+    void decode_start();
+    void decode_init(jpeg_decoder_stream * pStream);
+    void H2V2Convert();
+    void H2V1Convert();
+    void H1V2Convert();
+    void H1V1Convert();
+    void gray_convert();
+    void expanded_convert();
+    void find_eoi();
+    inline uint32_t get_char();
+    inline uint32_t get_char(bool *pPadding_flag);
+    inline void stuff_char(uint8_t q);
+    inline uint8_t get_octet();
+    inline uint32_t get_bits(int num_bits);
+    inline uint32_t get_bits_no_markers(int numbits);
+    inline int huff_decode(huff_tables *pH);
+    inline int huff_decode(huff_tables *pH, int& extrabits);
+    static inline uint8_t clamp(int i);
+    static void decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+};
+
+
+// DCT coefficients are stored in this sequence.
+static int g_ZAG[64] = {  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 };
+
+enum JPEG_MARKER
+{
+  M_SOF0  = 0xC0, M_SOF1  = 0xC1, M_SOF2  = 0xC2, M_SOF3  = 0xC3, M_SOF5  = 0xC5, M_SOF6  = 0xC6, M_SOF7  = 0xC7, M_JPG   = 0xC8,
+  M_SOF9  = 0xC9, M_SOF10 = 0xCA, M_SOF11 = 0xCB, M_SOF13 = 0xCD, M_SOF14 = 0xCE, M_SOF15 = 0xCF, M_DHT   = 0xC4, M_DAC   = 0xCC,
+  M_RST0  = 0xD0, M_RST1  = 0xD1, M_RST2  = 0xD2, M_RST3  = 0xD3, M_RST4  = 0xD4, M_RST5  = 0xD5, M_RST6  = 0xD6, M_RST7  = 0xD7,
+  M_SOI   = 0xD8, M_EOI   = 0xD9, M_SOS   = 0xDA, M_DQT   = 0xDB, M_DNL   = 0xDC, M_DRI   = 0xDD, M_DHP   = 0xDE, M_EXP   = 0xDF,
+  M_APP0  = 0xE0, M_APP15 = 0xEF, M_JPG0  = 0xF0, M_JPG13 = 0xFD, M_COM   = 0xFE, M_TEM   = 0x01, M_ERROR = 0x100, RST0   = 0xD0
+};
+
+enum JPEG_SUBSAMPLING { JPGD_GRAYSCALE = 0, JPGD_YH1V1, JPGD_YH2V1, JPGD_YH1V2, JPGD_YH2V2 };
+
+#define CONST_BITS  13
+#define PASS1_BITS  2
+#define SCALEDONE ((int32_t)1)
+#define DESCALE(x,n)  (((x) + (SCALEDONE << ((n)-1))) >> (n))
+#define DESCALE_ZEROSHIFT(x,n)  (((x) + (128 << (n)) + (SCALEDONE << ((n)-1))) >> (n))
+#define MULTIPLY(var, cnst)  ((var) * (cnst))
+#define CLAMP(i) ((static_cast<uint32_t>(i) > 255) ? (((~i) >> 31) & 0xFF) : (i))
+
+#define FIX_0_298631336  ((int32_t)2446)        /* FIX(0.298631336) */
+#define FIX_0_390180644  ((int32_t)3196)        /* FIX(0.390180644) */
+#define FIX_0_541196100  ((int32_t)4433)        /* FIX(0.541196100) */
+#define FIX_0_765366865  ((int32_t)6270)        /* FIX(0.765366865) */
+#define FIX_0_899976223  ((int32_t)7373)        /* FIX(0.899976223) */
+#define FIX_1_175875602  ((int32_t)9633)        /* FIX(1.175875602) */
+#define FIX_1_501321110  ((int32_t)12299)       /* FIX(1.501321110) */
+#define FIX_1_847759065  ((int32_t)15137)       /* FIX(1.847759065) */
+#define FIX_1_961570560  ((int32_t)16069)       /* FIX(1.961570560) */
+#define FIX_2_053119869  ((int32_t)16819)       /* FIX(2.053119869) */
+#define FIX_2_562915447  ((int32_t)20995)       /* FIX(2.562915447) */
+#define FIX_3_072711026  ((int32_t)25172)       /* FIX(3.072711026) */
+
+
+// Compiler creates a fast path 1D IDCT for X non-zero columns
+template <int NONZERO_COLS>
+struct Row
+{
+    static void idct(int* pTemp, const jpgd_block_t* pSrc)
+    {
+        // ACCESS_COL() will be optimized at compile time to either an array access, or 0.
+        #define ACCESS_COL(x) (((x) < NONZERO_COLS) ? (int)pSrc[x] : 0)
+
+        const int z2 = ACCESS_COL(2), z3 = ACCESS_COL(6);
+        const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+        const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
+        const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
+
+        const int tmp0 = static_cast<unsigned int>(ACCESS_COL(0) + ACCESS_COL(4)) << CONST_BITS;
+        const int tmp1 = static_cast<unsigned int>(ACCESS_COL(0) - ACCESS_COL(4)) << CONST_BITS;
+
+        const int tmp10 = tmp0 + tmp3, tmp13 = tmp0 - tmp3, tmp11 = tmp1 + tmp2, tmp12 = tmp1 - tmp2;
+
+        const int atmp0 = ACCESS_COL(7), atmp1 = ACCESS_COL(5), atmp2 = ACCESS_COL(3), atmp3 = ACCESS_COL(1);
+
+        const int bz1 = atmp0 + atmp3, bz2 = atmp1 + atmp2, bz3 = atmp0 + atmp2, bz4 = atmp1 + atmp3;
+        const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
+
+        const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
+        const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
+        const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
+        const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
+
+        const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
+        const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
+        const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
+        const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
+
+        pTemp[0] = DESCALE(tmp10 + btmp3, CONST_BITS-PASS1_BITS);
+        pTemp[7] = DESCALE(tmp10 - btmp3, CONST_BITS-PASS1_BITS);
+        pTemp[1] = DESCALE(tmp11 + btmp2, CONST_BITS-PASS1_BITS);
+        pTemp[6] = DESCALE(tmp11 - btmp2, CONST_BITS-PASS1_BITS);
+        pTemp[2] = DESCALE(tmp12 + btmp1, CONST_BITS-PASS1_BITS);
+        pTemp[5] = DESCALE(tmp12 - btmp1, CONST_BITS-PASS1_BITS);
+        pTemp[3] = DESCALE(tmp13 + btmp0, CONST_BITS-PASS1_BITS);
+        pTemp[4] = DESCALE(tmp13 - btmp0, CONST_BITS-PASS1_BITS);
+    }
+};
+
+
+template <>
+struct Row<0>
+{
+    static void idct(int* pTemp, const jpgd_block_t* pSrc)
+    {
+#ifdef _MSC_VER
+      pTemp; pSrc;
+#endif
+    }
+};
+
+
+template <>
+struct Row<1>
+{
+    static void idct(int* pTemp, const jpgd_block_t* pSrc)
+    {
+        const int dcval = pSrc[0] * PASS1_BITS * 2;
+
+        pTemp[0] = dcval;
+        pTemp[1] = dcval;
+        pTemp[2] = dcval;
+        pTemp[3] = dcval;
+        pTemp[4] = dcval;
+        pTemp[5] = dcval;
+        pTemp[6] = dcval;
+        pTemp[7] = dcval;
+    }
+};
+
+
+// Compiler creates a fast path 1D IDCT for X non-zero rows
+template <int NONZERO_ROWS>
+struct Col
+{
+    static void idct(uint8_t* pDst_ptr, const int* pTemp)
+    {
+        // ACCESS_ROW() will be optimized at compile time to either an array access, or 0.
+        #define ACCESS_ROW(x) (((x) < NONZERO_ROWS) ? pTemp[x * 8] : 0)
+
+        const int z2 = ACCESS_ROW(2);
+        const int z3 = ACCESS_ROW(6);
+
+        const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+        const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
+        const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
+
+        const int tmp0 = static_cast<unsigned int>(ACCESS_ROW(0) + ACCESS_ROW(4)) << CONST_BITS;
+        const int tmp1 = static_cast<unsigned int>(ACCESS_ROW(0) - ACCESS_ROW(4)) << CONST_BITS;
+
+        const int tmp10 = tmp0 + tmp3, tmp13 = tmp0 - tmp3, tmp11 = tmp1 + tmp2, tmp12 = tmp1 - tmp2;
+
+        const int atmp0 = ACCESS_ROW(7), atmp1 = ACCESS_ROW(5), atmp2 = ACCESS_ROW(3), atmp3 = ACCESS_ROW(1);
+
+        const int bz1 = atmp0 + atmp3, bz2 = atmp1 + atmp2, bz3 = atmp0 + atmp2, bz4 = atmp1 + atmp3;
+        const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
+
+        const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
+        const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
+        const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
+        const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
+
+        const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
+        const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
+        const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
+        const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
+
+        int i = DESCALE_ZEROSHIFT(tmp10 + btmp3, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*0] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp10 - btmp3, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*7] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp11 + btmp2, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*1] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp11 - btmp2, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*6] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp12 + btmp1, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*2] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp12 - btmp1, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*5] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp13 + btmp0, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*3] = (uint8_t)CLAMP(i);
+
+        i = DESCALE_ZEROSHIFT(tmp13 - btmp0, CONST_BITS+PASS1_BITS+3);
+        pDst_ptr[8*4] = (uint8_t)CLAMP(i);
+    }
+};
+
+
+template <>
+struct Col<1>
+{
+    static void idct(uint8_t* pDst_ptr, const int* pTemp)
+    {
+        int dcval = DESCALE_ZEROSHIFT(pTemp[0], PASS1_BITS+3);
+        const uint8_t dcval_clamped = (uint8_t)CLAMP(dcval);
+        pDst_ptr[0*8] = dcval_clamped;
+        pDst_ptr[1*8] = dcval_clamped;
+        pDst_ptr[2*8] = dcval_clamped;
+        pDst_ptr[3*8] = dcval_clamped;
+        pDst_ptr[4*8] = dcval_clamped;
+        pDst_ptr[5*8] = dcval_clamped;
+        pDst_ptr[6*8] = dcval_clamped;
+        pDst_ptr[7*8] = dcval_clamped;
+    }
+};
+
+
+static const uint8_t s_idct_row_table[] = {
+    1,0,0,0,0,0,0,0, 2,0,0,0,0,0,0,0, 2,1,0,0,0,0,0,0, 2,1,1,0,0,0,0,0, 2,2,1,0,0,0,0,0, 3,2,1,0,0,0,0,0, 4,2,1,0,0,0,0,0, 4,3,1,0,0,0,0,0,
+    4,3,2,0,0,0,0,0, 4,3,2,1,0,0,0,0, 4,3,2,1,1,0,0,0, 4,3,2,2,1,0,0,0, 4,3,3,2,1,0,0,0, 4,4,3,2,1,0,0,0, 5,4,3,2,1,0,0,0, 6,4,3,2,1,0,0,0,
+    6,5,3,2,1,0,0,0, 6,5,4,2,1,0,0,0, 6,5,4,3,1,0,0,0, 6,5,4,3,2,0,0,0, 6,5,4,3,2,1,0,0, 6,5,4,3,2,1,1,0, 6,5,4,3,2,2,1,0, 6,5,4,3,3,2,1,0,
+    6,5,4,4,3,2,1,0, 6,5,5,4,3,2,1,0, 6,6,5,4,3,2,1,0, 7,6,5,4,3,2,1,0, 8,6,5,4,3,2,1,0, 8,7,5,4,3,2,1,0, 8,7,6,4,3,2,1,0, 8,7,6,5,3,2,1,0,
+    8,7,6,5,4,2,1,0, 8,7,6,5,4,3,1,0, 8,7,6,5,4,3,2,0, 8,7,6,5,4,3,2,1, 8,7,6,5,4,3,2,2, 8,7,6,5,4,3,3,2, 8,7,6,5,4,4,3,2, 8,7,6,5,5,4,3,2,
+    8,7,6,6,5,4,3,2, 8,7,7,6,5,4,3,2, 8,8,7,6,5,4,3,2, 8,8,8,6,5,4,3,2, 8,8,8,7,5,4,3,2, 8,8,8,7,6,4,3,2, 8,8,8,7,6,5,3,2, 8,8,8,7,6,5,4,2,
+    8,8,8,7,6,5,4,3, 8,8,8,7,6,5,4,4, 8,8,8,7,6,5,5,4, 8,8,8,7,6,6,5,4, 8,8,8,7,7,6,5,4, 8,8,8,8,7,6,5,4, 8,8,8,8,8,6,5,4, 8,8,8,8,8,7,5,4,
+    8,8,8,8,8,7,6,4, 8,8,8,8,8,7,6,5, 8,8,8,8,8,7,6,6, 8,8,8,8,8,7,7,6, 8,8,8,8,8,8,7,6, 8,8,8,8,8,8,8,6, 8,8,8,8,8,8,8,7, 8,8,8,8,8,8,8,8,
+};
+
+
+static const uint8_t s_idct_col_table[] = { 1, 1, 2, 3, 3, 3, 3, 3, 3, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 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 };
+
+
+void idct(const jpgd_block_t* pSrc_ptr, uint8_t* pDst_ptr, int block_max_zag)
+{
+    JPGD_ASSERT(block_max_zag >= 1);
+    JPGD_ASSERT(block_max_zag <= 64);
+
+    if (block_max_zag <= 1) {
+        int k = ((pSrc_ptr[0] + 4) >> 3) + 128;
+        k = CLAMP(k);
+        k = k | (k<<8);
+        k = k | (k<<16);
+        for (int i = 8; i > 0; i--) {
+            *(int*)&pDst_ptr[0] = k;
+            *(int*)&pDst_ptr[4] = k;
+            pDst_ptr += 8;
+        }
+      return;
+    }
+
+    int temp[64];
+    const jpgd_block_t* pSrc = pSrc_ptr;
+    int* pTemp = temp;
+    const uint8_t* pRow_tab = &s_idct_row_table[(block_max_zag - 1) * 8];
+    int i;
+    for (i = 8; i > 0; i--, pRow_tab++) {
+        switch (*pRow_tab) {
+            case 0: Row<0>::idct(pTemp, pSrc); break;
+            case 1: Row<1>::idct(pTemp, pSrc); break;
+            case 2: Row<2>::idct(pTemp, pSrc); break;
+            case 3: Row<3>::idct(pTemp, pSrc); break;
+            case 4: Row<4>::idct(pTemp, pSrc); break;
+            case 5: Row<5>::idct(pTemp, pSrc); break;
+            case 6: Row<6>::idct(pTemp, pSrc); break;
+            case 7: Row<7>::idct(pTemp, pSrc); break;
+            case 8: Row<8>::idct(pTemp, pSrc); break;
+        }
+        pSrc += 8;
+        pTemp += 8;
+    }
+
+    pTemp = temp;
+
+    const int nonzero_rows = s_idct_col_table[block_max_zag - 1];
+    for (i = 8; i > 0; i--) {
+        switch (nonzero_rows) {
+            case 1: Col<1>::idct(pDst_ptr, pTemp); break;
+            case 2: Col<2>::idct(pDst_ptr, pTemp); break;
+            case 3: Col<3>::idct(pDst_ptr, pTemp); break;
+            case 4: Col<4>::idct(pDst_ptr, pTemp); break;
+            case 5: Col<5>::idct(pDst_ptr, pTemp); break;
+            case 6: Col<6>::idct(pDst_ptr, pTemp); break;
+            case 7: Col<7>::idct(pDst_ptr, pTemp); break;
+            case 8: Col<8>::idct(pDst_ptr, pTemp); break;
+        }
+        pTemp++;
+        pDst_ptr++;
+    }
+}
+
+
+void idct_4x4(const jpgd_block_t* pSrc_ptr, uint8_t* pDst_ptr)
+{
+    int temp[64];
+    int* pTemp = temp;
+    const jpgd_block_t* pSrc = pSrc_ptr;
+
+    for (int i = 4; i > 0; i--) {
+        Row<4>::idct(pTemp, pSrc);
+        pSrc += 8;
+        pTemp += 8;
+    }
+
+    pTemp = temp;
+
+    for (int i = 8; i > 0; i--) {
+        Col<4>::idct(pDst_ptr, pTemp);
+        pTemp++;
+        pDst_ptr++;
+    }
+}
+
+
+// Retrieve one character from the input stream.
+inline uint32_t jpeg_decoder::get_char()
+{
+    // Any bytes remaining in buffer?
+    if (!m_in_buf_left) {
+        // Try to get more bytes.
+        prep_in_buffer();
+        // Still nothing to get?
+        if (!m_in_buf_left) {
+            // Pad the end of the stream with 0xFF 0xD9 (EOI marker)
+            int t = m_tem_flag;
+            m_tem_flag ^= 1;
+            if (t) return 0xD9;
+            else return 0xFF;
+        }
+    }
+    uint32_t c = *m_pIn_buf_ofs++;
+    m_in_buf_left--;
+    return c;
+}
+
+
+// Same as previous method, except can indicate if the character is a pad character or not.
+inline uint32_t jpeg_decoder::get_char(bool *pPadding_flag)
+{
+    if (!m_in_buf_left) {
+        prep_in_buffer();
+        if (!m_in_buf_left) {
+            *pPadding_flag = true;
+            int t = m_tem_flag;
+            m_tem_flag ^= 1;
+            if (t) return 0xD9;
+            else return 0xFF;
+        }
+    }
+    *pPadding_flag = false;
+    uint32_t c = *m_pIn_buf_ofs++;
+    m_in_buf_left--;
+
+    return c;
+}
+
+
+// Inserts a previously retrieved character back into the input buffer.
+inline void jpeg_decoder::stuff_char(uint8_t q)
+{
+    *(--m_pIn_buf_ofs) = q;
+    m_in_buf_left++;
+}
+
+
+// Retrieves one character from the input stream, but does not read past markers. Will continue to return 0xFF when a marker is encountered.
+inline uint8_t jpeg_decoder::get_octet()
+{
+    bool padding_flag;
+    int c = get_char(&padding_flag);
+
+    if (c == 0xFF) {
+        if (padding_flag) return 0xFF;
+
+        c = get_char(&padding_flag);
+        if (padding_flag) {
+            stuff_char(0xFF);
+            return 0xFF;
+        }
+        if (c == 0x00) return 0xFF;
+        else {
+            stuff_char(static_cast<uint8_t>(c));
+            stuff_char(0xFF);
+            return 0xFF;
+        }
+    }
+    return static_cast<uint8_t>(c);
+}
+
+
+// Retrieves a variable number of bits from the input stream. Does not recognize markers.
+inline uint32_t jpeg_decoder::get_bits(int num_bits)
+{
+    if (!num_bits) return 0;
+
+    uint32_t i = m_bit_buf >> (32 - num_bits);
+
+    if ((m_bits_left -= num_bits) <= 0) {
+        m_bit_buf <<= (num_bits += m_bits_left);
+        uint32_t c1 = get_char();
+        uint32_t c2 = get_char();
+        m_bit_buf = (m_bit_buf & 0xFFFF0000) | (c1 << 8) | c2;
+        m_bit_buf <<= -m_bits_left;
+        m_bits_left += 16;
+        JPGD_ASSERT(m_bits_left >= 0);
+    }
+    else m_bit_buf <<= num_bits;
+
+    return i;
+}
+
+
+// Retrieves a variable number of bits from the input stream. Markers will not be read into the input bit buffer. Instead, an infinite number of all 1's will be returned when a marker is encountered.
+inline uint32_t jpeg_decoder::get_bits_no_markers(int num_bits)
+{
+    if (!num_bits)return 0;
+
+    uint32_t i = m_bit_buf >> (32 - num_bits);
+
+    if ((m_bits_left -= num_bits) <= 0) {
+        m_bit_buf <<= (num_bits += m_bits_left);
+        if ((m_in_buf_left < 2) || (m_pIn_buf_ofs[0] == 0xFF) || (m_pIn_buf_ofs[1] == 0xFF)) {
+            uint32_t c1 = get_octet();
+            uint32_t c2 = get_octet();
+            m_bit_buf |= (c1 << 8) | c2;
+        } else {
+            m_bit_buf |= ((uint32_t)m_pIn_buf_ofs[0] << 8) | m_pIn_buf_ofs[1];
+            m_in_buf_left -= 2;
+            m_pIn_buf_ofs += 2;
+        }
+        m_bit_buf <<= -m_bits_left;
+        m_bits_left += 16;
+        JPGD_ASSERT(m_bits_left >= 0);
+    } else m_bit_buf <<= num_bits;
+
+    return i;
+}
+
+
+// Decodes a Huffman encoded symbol.
+inline int jpeg_decoder::huff_decode(huff_tables *pH)
+{
+    int symbol;
+
+    // Check first 8-bits: do we have a complete symbol?
+    if ((symbol = pH->look_up[m_bit_buf >> 24]) < 0) {
+        // Decode more bits, use a tree traversal to find symbol.
+        int ofs = 23;
+        do {
+            symbol = pH->tree[-(int)(symbol + ((m_bit_buf >> ofs) & 1))];
+            ofs--;
+        } while (symbol < 0);
+        get_bits_no_markers(8 + (23 - ofs));
+    } else get_bits_no_markers(pH->code_size[symbol]);
+
+  return symbol;
+}
+
+
+// Decodes a Huffman encoded symbol.
+inline int jpeg_decoder::huff_decode(huff_tables *pH, int& extra_bits)
+{
+    int symbol;
+
+    // Check first 8-bits: do we have a complete symbol?
+    if ((symbol = pH->look_up2[m_bit_buf >> 24]) < 0) {
+        // Use a tree traversal to find symbol.
+        int ofs = 23;
+        do {
+            symbol = pH->tree[-(int)(symbol + ((m_bit_buf >> ofs) & 1))];
+            ofs--;
+        } while (symbol < 0);
+
+        get_bits_no_markers(8 + (23 - ofs));
+        extra_bits = get_bits_no_markers(symbol & 0xF);
+    } else {
+        JPGD_ASSERT(((symbol >> 8) & 31) == pH->code_size[symbol & 255] + ((symbol & 0x8000) ? (symbol & 15) : 0));
+
+        if (symbol & 0x8000) {
+            get_bits_no_markers((symbol >> 8) & 31);
+            extra_bits = symbol >> 16;
+        } else  {
+            int code_size = (symbol >> 8) & 31;
+            int num_extra_bits = symbol & 0xF;
+            int bits = code_size + num_extra_bits;
+            if (bits <= (m_bits_left + 16)) extra_bits = get_bits_no_markers(bits) & ((1 << num_extra_bits) - 1);
+            else {
+                get_bits_no_markers(code_size);
+                extra_bits = get_bits_no_markers(num_extra_bits);
+            }
+        }
+        symbol &= 0xFF;
+    }
+    return symbol;
+}
+
+
+// Tables and macro used to fully decode the DPCM differences.
+static const int s_extend_test[16] = { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };
+static const unsigned int s_extend_offset[16] = { 0, ((~0u)<<1) + 1, ((~0u)<<2) + 1, ((~0u)<<3) + 1, ((~0u)<<4) + 1, ((~0u)<<5) + 1, ((~0u)<<6) + 1, ((~0u)<<7) + 1, ((~0u)<<8) + 1, ((~0u)<<9) + 1, ((~0u)<<10) + 1, ((~0u)<<11) + 1, ((~0u)<<12) + 1, ((~0u)<<13) + 1, ((~0u)<<14) + 1, ((~0u)<<15) + 1 };
+
+// The logical AND's in this macro are to shut up static code analysis (aren't really necessary - couldn't find another way to do this)
+#define JPGD_HUFF_EXTEND(x, s) (((x) < s_extend_test[s & 15]) ? ((x) + s_extend_offset[s & 15]) : (x))
+
+
+// Clamps a value between 0-255.
+inline uint8_t jpeg_decoder::clamp(int i)
+{
+    if (static_cast<uint32_t>(i) > 255) i = (((~i) >> 31) & 0xFF);
+    return static_cast<uint8_t>(i);
+}
+
+
+namespace DCT_Upsample
+{
+    struct Matrix44
+    {
+        typedef int Element_Type;
+        enum { NUM_ROWS = 4, NUM_COLS = 4 };
+
+        Element_Type v[NUM_ROWS][NUM_COLS];
+
+        inline int rows() const { return NUM_ROWS; }
+        inline int cols() const { return NUM_COLS; }
+        inline const Element_Type & at(int r, int c) const { return v[r][c]; }
+        inline       Element_Type & at(int r, int c)       { return v[r][c]; }
+
+        inline Matrix44() {}
+
+        inline Matrix44& operator += (const Matrix44& a)
+        {
+            for (int r = 0; r < NUM_ROWS; r++) {
+                at(r, 0) += a.at(r, 0);
+                at(r, 1) += a.at(r, 1);
+                at(r, 2) += a.at(r, 2);
+                at(r, 3) += a.at(r, 3);
+            }
+            return *this;
+        }
+
+        inline Matrix44& operator -= (const Matrix44& a)
+        {
+            for (int r = 0; r < NUM_ROWS; r++) {
+                at(r, 0) -= a.at(r, 0);
+                at(r, 1) -= a.at(r, 1);
+                at(r, 2) -= a.at(r, 2);
+                at(r, 3) -= a.at(r, 3);
+            }
+            return *this;
+        }
+
+        friend inline Matrix44 operator + (const Matrix44& a, const Matrix44& b)
+        {
+            Matrix44 ret;
+            for (int r = 0; r < NUM_ROWS; r++) {
+                ret.at(r, 0) = a.at(r, 0) + b.at(r, 0);
+                ret.at(r, 1) = a.at(r, 1) + b.at(r, 1);
+                ret.at(r, 2) = a.at(r, 2) + b.at(r, 2);
+                ret.at(r, 3) = a.at(r, 3) + b.at(r, 3);
+            }
+            return ret;
+        }
+
+        friend inline Matrix44 operator - (const Matrix44& a, const Matrix44& b)
+        {
+            Matrix44 ret;
+            for (int r = 0; r < NUM_ROWS; r++) {
+                ret.at(r, 0) = a.at(r, 0) - b.at(r, 0);
+                ret.at(r, 1) = a.at(r, 1) - b.at(r, 1);
+                ret.at(r, 2) = a.at(r, 2) - b.at(r, 2);
+                ret.at(r, 3) = a.at(r, 3) - b.at(r, 3);
+            }
+            return ret;
+        }
+
+        static inline void add_and_store(jpgd_block_t* pDst, const Matrix44& a, const Matrix44& b)
+        {
+            for (int r = 0; r < 4; r++) {
+                pDst[0*8 + r] = static_cast<jpgd_block_t>(a.at(r, 0) + b.at(r, 0));
+                pDst[1*8 + r] = static_cast<jpgd_block_t>(a.at(r, 1) + b.at(r, 1));
+                pDst[2*8 + r] = static_cast<jpgd_block_t>(a.at(r, 2) + b.at(r, 2));
+                pDst[3*8 + r] = static_cast<jpgd_block_t>(a.at(r, 3) + b.at(r, 3));
+            }
+        }
+
+        static inline void sub_and_store(jpgd_block_t* pDst, const Matrix44& a, const Matrix44& b)
+        {
+            for (int r = 0; r < 4; r++) {
+                pDst[0*8 + r] = static_cast<jpgd_block_t>(a.at(r, 0) - b.at(r, 0));
+                pDst[1*8 + r] = static_cast<jpgd_block_t>(a.at(r, 1) - b.at(r, 1));
+                pDst[2*8 + r] = static_cast<jpgd_block_t>(a.at(r, 2) - b.at(r, 2));
+                pDst[3*8 + r] = static_cast<jpgd_block_t>(a.at(r, 3) - b.at(r, 3));
+            }
+        }
+    };
+
+    const int FRACT_BITS = 10;
+    const int SCALE = 1 << FRACT_BITS;
+
+    typedef int Temp_Type;
+    #define D(i) (((i) + (SCALE >> 1)) >> FRACT_BITS)
+    #define F(i) ((int)((i) * SCALE + .5f))
+
+    // Any decent C++ compiler will optimize this at compile time to a 0, or an array access.
+    #define AT(c, r) ((((c)>=NUM_COLS)||((r)>=NUM_ROWS)) ? 0 : pSrc[(c)+(r)*8])
+
+    // NUM_ROWS/NUM_COLS = # of non-zero rows/cols in input matrix
+    template<int NUM_ROWS, int NUM_COLS>
+    struct P_Q
+    {
+        static void calc(Matrix44& P, Matrix44& Q, const jpgd_block_t* pSrc)
+        {
+            // 4x8 = 4x8 times 8x8, matrix 0 is constant
+            const Temp_Type X000 = AT(0, 0);
+            const Temp_Type X001 = AT(0, 1);
+            const Temp_Type X002 = AT(0, 2);
+            const Temp_Type X003 = AT(0, 3);
+            const Temp_Type X004 = AT(0, 4);
+            const Temp_Type X005 = AT(0, 5);
+            const Temp_Type X006 = AT(0, 6);
+            const Temp_Type X007 = AT(0, 7);
+            const Temp_Type X010 = D(F(0.415735f) * AT(1, 0) + F(0.791065f) * AT(3, 0) + F(-0.352443f) * AT(5, 0) + F(0.277785f) * AT(7, 0));
+            const Temp_Type X011 = D(F(0.415735f) * AT(1, 1) + F(0.791065f) * AT(3, 1) + F(-0.352443f) * AT(5, 1) + F(0.277785f) * AT(7, 1));
+            const Temp_Type X012 = D(F(0.415735f) * AT(1, 2) + F(0.791065f) * AT(3, 2) + F(-0.352443f) * AT(5, 2) + F(0.277785f) * AT(7, 2));
+            const Temp_Type X013 = D(F(0.415735f) * AT(1, 3) + F(0.791065f) * AT(3, 3) + F(-0.352443f) * AT(5, 3) + F(0.277785f) * AT(7, 3));
+            const Temp_Type X014 = D(F(0.415735f) * AT(1, 4) + F(0.791065f) * AT(3, 4) + F(-0.352443f) * AT(5, 4) + F(0.277785f) * AT(7, 4));
+            const Temp_Type X015 = D(F(0.415735f) * AT(1, 5) + F(0.791065f) * AT(3, 5) + F(-0.352443f) * AT(5, 5) + F(0.277785f) * AT(7, 5));
+            const Temp_Type X016 = D(F(0.415735f) * AT(1, 6) + F(0.791065f) * AT(3, 6) + F(-0.352443f) * AT(5, 6) + F(0.277785f) * AT(7, 6));
+            const Temp_Type X017 = D(F(0.415735f) * AT(1, 7) + F(0.791065f) * AT(3, 7) + F(-0.352443f) * AT(5, 7) + F(0.277785f) * AT(7, 7));
+            const Temp_Type X020 = AT(4, 0);
+            const Temp_Type X021 = AT(4, 1);
+            const Temp_Type X022 = AT(4, 2);
+            const Temp_Type X023 = AT(4, 3);
+            const Temp_Type X024 = AT(4, 4);
+            const Temp_Type X025 = AT(4, 5);
+            const Temp_Type X026 = AT(4, 6);
+            const Temp_Type X027 = AT(4, 7);
+            const Temp_Type X030 = D(F(0.022887f) * AT(1, 0) + F(-0.097545f) * AT(3, 0) + F(0.490393f) * AT(5, 0) + F(0.865723f) * AT(7, 0));
+            const Temp_Type X031 = D(F(0.022887f) * AT(1, 1) + F(-0.097545f) * AT(3, 1) + F(0.490393f) * AT(5, 1) + F(0.865723f) * AT(7, 1));
+            const Temp_Type X032 = D(F(0.022887f) * AT(1, 2) + F(-0.097545f) * AT(3, 2) + F(0.490393f) * AT(5, 2) + F(0.865723f) * AT(7, 2));
+            const Temp_Type X033 = D(F(0.022887f) * AT(1, 3) + F(-0.097545f) * AT(3, 3) + F(0.490393f) * AT(5, 3) + F(0.865723f) * AT(7, 3));
+            const Temp_Type X034 = D(F(0.022887f) * AT(1, 4) + F(-0.097545f) * AT(3, 4) + F(0.490393f) * AT(5, 4) + F(0.865723f) * AT(7, 4));
+            const Temp_Type X035 = D(F(0.022887f) * AT(1, 5) + F(-0.097545f) * AT(3, 5) + F(0.490393f) * AT(5, 5) + F(0.865723f) * AT(7, 5));
+            const Temp_Type X036 = D(F(0.022887f) * AT(1, 6) + F(-0.097545f) * AT(3, 6) + F(0.490393f) * AT(5, 6) + F(0.865723f) * AT(7, 6));
+            const Temp_Type X037 = D(F(0.022887f) * AT(1, 7) + F(-0.097545f) * AT(3, 7) + F(0.490393f) * AT(5, 7) + F(0.865723f) * AT(7, 7));
+
+            // 4x4 = 4x8 times 8x4, matrix 1 is constant
+            P.at(0, 0) = X000;
+            P.at(0, 1) = D(X001 * F(0.415735f) + X003 * F(0.791065f) + X005 * F(-0.352443f) + X007 * F(0.277785f));
+            P.at(0, 2) = X004;
+            P.at(0, 3) = D(X001 * F(0.022887f) + X003 * F(-0.097545f) + X005 * F(0.490393f) + X007 * F(0.865723f));
+            P.at(1, 0) = X010;
+            P.at(1, 1) = D(X011 * F(0.415735f) + X013 * F(0.791065f) + X015 * F(-0.352443f) + X017 * F(0.277785f));
+            P.at(1, 2) = X014;
+            P.at(1, 3) = D(X011 * F(0.022887f) + X013 * F(-0.097545f) + X015 * F(0.490393f) + X017 * F(0.865723f));
+            P.at(2, 0) = X020;
+            P.at(2, 1) = D(X021 * F(0.415735f) + X023 * F(0.791065f) + X025 * F(-0.352443f) + X027 * F(0.277785f));
+            P.at(2, 2) = X024;
+            P.at(2, 3) = D(X021 * F(0.022887f) + X023 * F(-0.097545f) + X025 * F(0.490393f) + X027 * F(0.865723f));
+            P.at(3, 0) = X030;
+            P.at(3, 1) = D(X031 * F(0.415735f) + X033 * F(0.791065f) + X035 * F(-0.352443f) + X037 * F(0.277785f));
+            P.at(3, 2) = X034;
+            P.at(3, 3) = D(X031 * F(0.022887f) + X033 * F(-0.097545f) + X035 * F(0.490393f) + X037 * F(0.865723f));
+            // 40 muls 24 adds
+
+            // 4x4 = 4x8 times 8x4, matrix 1 is constant
+            Q.at(0, 0) = D(X001 * F(0.906127f) + X003 * F(-0.318190f) + X005 * F(0.212608f) + X007 * F(-0.180240f));
+            Q.at(0, 1) = X002;
+            Q.at(0, 2) = D(X001 * F(-0.074658f) + X003 * F(0.513280f) + X005 * F(0.768178f) + X007 * F(-0.375330f));
+            Q.at(0, 3) = X006;
+            Q.at(1, 0) = D(X011 * F(0.906127f) + X013 * F(-0.318190f) + X015 * F(0.212608f) + X017 * F(-0.180240f));
+            Q.at(1, 1) = X012;
+            Q.at(1, 2) = D(X011 * F(-0.074658f) + X013 * F(0.513280f) + X015 * F(0.768178f) + X017 * F(-0.375330f));
+            Q.at(1, 3) = X016;
+            Q.at(2, 0) = D(X021 * F(0.906127f) + X023 * F(-0.318190f) + X025 * F(0.212608f) + X027 * F(-0.180240f));
+            Q.at(2, 1) = X022;
+            Q.at(2, 2) = D(X021 * F(-0.074658f) + X023 * F(0.513280f) + X025 * F(0.768178f) + X027 * F(-0.375330f));
+            Q.at(2, 3) = X026;
+            Q.at(3, 0) = D(X031 * F(0.906127f) + X033 * F(-0.318190f) + X035 * F(0.212608f) + X037 * F(-0.180240f));
+            Q.at(3, 1) = X032;
+            Q.at(3, 2) = D(X031 * F(-0.074658f) + X033 * F(0.513280f) + X035 * F(0.768178f) + X037 * F(-0.375330f));
+            Q.at(3, 3) = X036;
+            // 40 muls 24 adds
+        }
+    };
+
+
+    template<int NUM_ROWS, int NUM_COLS>
+    struct R_S
+    {
+        static void calc(Matrix44& R, Matrix44& S, const jpgd_block_t* pSrc)
+        {
+            // 4x8 = 4x8 times 8x8, matrix 0 is constant
+            const Temp_Type X100 = D(F(0.906127f) * AT(1, 0) + F(-0.318190f) * AT(3, 0) + F(0.212608f) * AT(5, 0) + F(-0.180240f) * AT(7, 0));
+            const Temp_Type X101 = D(F(0.906127f) * AT(1, 1) + F(-0.318190f) * AT(3, 1) + F(0.212608f) * AT(5, 1) + F(-0.180240f) * AT(7, 1));
+            const Temp_Type X102 = D(F(0.906127f) * AT(1, 2) + F(-0.318190f) * AT(3, 2) + F(0.212608f) * AT(5, 2) + F(-0.180240f) * AT(7, 2));
+            const Temp_Type X103 = D(F(0.906127f) * AT(1, 3) + F(-0.318190f) * AT(3, 3) + F(0.212608f) * AT(5, 3) + F(-0.180240f) * AT(7, 3));
+            const Temp_Type X104 = D(F(0.906127f) * AT(1, 4) + F(-0.318190f) * AT(3, 4) + F(0.212608f) * AT(5, 4) + F(-0.180240f) * AT(7, 4));
+            const Temp_Type X105 = D(F(0.906127f) * AT(1, 5) + F(-0.318190f) * AT(3, 5) + F(0.212608f) * AT(5, 5) + F(-0.180240f) * AT(7, 5));
+            const Temp_Type X106 = D(F(0.906127f) * AT(1, 6) + F(-0.318190f) * AT(3, 6) + F(0.212608f) * AT(5, 6) + F(-0.180240f) * AT(7, 6));
+            const Temp_Type X107 = D(F(0.906127f) * AT(1, 7) + F(-0.318190f) * AT(3, 7) + F(0.212608f) * AT(5, 7) + F(-0.180240f) * AT(7, 7));
+            const Temp_Type X110 = AT(2, 0);
+            const Temp_Type X111 = AT(2, 1);
+            const Temp_Type X112 = AT(2, 2);
+            const Temp_Type X113 = AT(2, 3);
+            const Temp_Type X114 = AT(2, 4);
+            const Temp_Type X115 = AT(2, 5);
+            const Temp_Type X116 = AT(2, 6);
+            const Temp_Type X117 = AT(2, 7);
+            const Temp_Type X120 = D(F(-0.074658f) * AT(1, 0) + F(0.513280f) * AT(3, 0) + F(0.768178f) * AT(5, 0) + F(-0.375330f) * AT(7, 0));
+            const Temp_Type X121 = D(F(-0.074658f) * AT(1, 1) + F(0.513280f) * AT(3, 1) + F(0.768178f) * AT(5, 1) + F(-0.375330f) * AT(7, 1));
+            const Temp_Type X122 = D(F(-0.074658f) * AT(1, 2) + F(0.513280f) * AT(3, 2) + F(0.768178f) * AT(5, 2) + F(-0.375330f) * AT(7, 2));
+            const Temp_Type X123 = D(F(-0.074658f) * AT(1, 3) + F(0.513280f) * AT(3, 3) + F(0.768178f) * AT(5, 3) + F(-0.375330f) * AT(7, 3));
+            const Temp_Type X124 = D(F(-0.074658f) * AT(1, 4) + F(0.513280f) * AT(3, 4) + F(0.768178f) * AT(5, 4) + F(-0.375330f) * AT(7, 4));
+            const Temp_Type X125 = D(F(-0.074658f) * AT(1, 5) + F(0.513280f) * AT(3, 5) + F(0.768178f) * AT(5, 5) + F(-0.375330f) * AT(7, 5));
+            const Temp_Type X126 = D(F(-0.074658f) * AT(1, 6) + F(0.513280f) * AT(3, 6) + F(0.768178f) * AT(5, 6) + F(-0.375330f) * AT(7, 6));
+            const Temp_Type X127 = D(F(-0.074658f) * AT(1, 7) + F(0.513280f) * AT(3, 7) + F(0.768178f) * AT(5, 7) + F(-0.375330f) * AT(7, 7));
+            const Temp_Type X130 = AT(6, 0);
+            const Temp_Type X131 = AT(6, 1);
+            const Temp_Type X132 = AT(6, 2);
+            const Temp_Type X133 = AT(6, 3);
+            const Temp_Type X134 = AT(6, 4);
+            const Temp_Type X135 = AT(6, 5);
+            const Temp_Type X136 = AT(6, 6);
+            const Temp_Type X137 = AT(6, 7);
+            // 80 muls 48 adds
+
+            // 4x4 = 4x8 times 8x4, matrix 1 is constant
+            R.at(0, 0) = X100;
+            R.at(0, 1) = D(X101 * F(0.415735f) + X103 * F(0.791065f) + X105 * F(-0.352443f) + X107 * F(0.277785f));
+            R.at(0, 2) = X104;
+            R.at(0, 3) = D(X101 * F(0.022887f) + X103 * F(-0.097545f) + X105 * F(0.490393f) + X107 * F(0.865723f));
+            R.at(1, 0) = X110;
+            R.at(1, 1) = D(X111 * F(0.415735f) + X113 * F(0.791065f) + X115 * F(-0.352443f) + X117 * F(0.277785f));
+            R.at(1, 2) = X114;
+            R.at(1, 3) = D(X111 * F(0.022887f) + X113 * F(-0.097545f) + X115 * F(0.490393f) + X117 * F(0.865723f));
+            R.at(2, 0) = X120;
+            R.at(2, 1) = D(X121 * F(0.415735f) + X123 * F(0.791065f) + X125 * F(-0.352443f) + X127 * F(0.277785f));
+            R.at(2, 2) = X124;
+            R.at(2, 3) = D(X121 * F(0.022887f) + X123 * F(-0.097545f) + X125 * F(0.490393f) + X127 * F(0.865723f));
+            R.at(3, 0) = X130;
+            R.at(3, 1) = D(X131 * F(0.415735f) + X133 * F(0.791065f) + X135 * F(-0.352443f) + X137 * F(0.277785f));
+            R.at(3, 2) = X134;
+            R.at(3, 3) = D(X131 * F(0.022887f) + X133 * F(-0.097545f) + X135 * F(0.490393f) + X137 * F(0.865723f));
+            // 40 muls 24 adds
+            // 4x4 = 4x8 times 8x4, matrix 1 is constant
+            S.at(0, 0) = D(X101 * F(0.906127f) + X103 * F(-0.318190f) + X105 * F(0.212608f) + X107 * F(-0.180240f));
+            S.at(0, 1) = X102;
+            S.at(0, 2) = D(X101 * F(-0.074658f) + X103 * F(0.513280f) + X105 * F(0.768178f) + X107 * F(-0.375330f));
+            S.at(0, 3) = X106;
+            S.at(1, 0) = D(X111 * F(0.906127f) + X113 * F(-0.318190f) + X115 * F(0.212608f) + X117 * F(-0.180240f));
+            S.at(1, 1) = X112;
+            S.at(1, 2) = D(X111 * F(-0.074658f) + X113 * F(0.513280f) + X115 * F(0.768178f) + X117 * F(-0.375330f));
+            S.at(1, 3) = X116;
+            S.at(2, 0) = D(X121 * F(0.906127f) + X123 * F(-0.318190f) + X125 * F(0.212608f) + X127 * F(-0.180240f));
+            S.at(2, 1) = X122;
+            S.at(2, 2) = D(X121 * F(-0.074658f) + X123 * F(0.513280f) + X125 * F(0.768178f) + X127 * F(-0.375330f));
+            S.at(2, 3) = X126;
+            S.at(3, 0) = D(X131 * F(0.906127f) + X133 * F(-0.318190f) + X135 * F(0.212608f) + X137 * F(-0.180240f));
+            S.at(3, 1) = X132;
+            S.at(3, 2) = D(X131 * F(-0.074658f) + X133 * F(0.513280f) + X135 * F(0.768178f) + X137 * F(-0.375330f));
+            S.at(3, 3) = X136;
+            // 40 muls 24 adds
+        }
+    };
+} // end namespace DCT_Upsample
+
+
+// Unconditionally frees all allocated m_blocks.
+void jpeg_decoder::free_all_blocks()
+{
+    delete(m_pStream);
+    m_pStream = nullptr;
+
+    for (mem_block *b = m_pMem_blocks; b; ) {
+        mem_block *n = b->m_pNext;
+        free(b);
+        b = n;
+    }
+    m_pMem_blocks = nullptr;
+}
+
+
+// This method handles all errors. It will never return.
+// It could easily be changed to use C++ exceptions.
+JPGD_NORETURN void jpeg_decoder::stop_decoding(jpgd_status status)
+{
+    m_error_code = status;
+    free_all_blocks();
+    longjmp(m_jmp_state, status);
+}
+
+
+void *jpeg_decoder::alloc(size_t nSize, bool zero)
+{
+    nSize = (JPGD_MAX(nSize, 1) + 3) & ~3;
+    char *rv = nullptr;
+    for (mem_block *b = m_pMem_blocks; b; b = b->m_pNext) {
+        if ((b->m_used_count + nSize) <= b->m_size) {
+            rv = b->m_data + b->m_used_count;
+            b->m_used_count += nSize;
+            break;
+        }
+    }
+    if (!rv) {
+        int capacity = JPGD_MAX(32768 - 256, (nSize + 2047) & ~2047);
+        mem_block *b = (mem_block*)malloc(sizeof(mem_block) + capacity);
+        if (!b) stop_decoding(JPGD_NOTENOUGHMEM);
+        b->m_pNext = m_pMem_blocks; m_pMem_blocks = b;
+        b->m_used_count = nSize;
+        b->m_size = capacity;
+        rv = b->m_data;
+    }
+    if (zero) memset(rv, 0, nSize);
+    return rv;
+}
+
+
+void jpeg_decoder::word_clear(void *p, uint16_t c, uint32_t n)
+{
+    uint8_t *pD = (uint8_t*)p;
+    const uint8_t l = c & 0xFF, h = (c >> 8) & 0xFF;
+    while (n) {
+        pD[0] = l; pD[1] = h; pD += 2;
+        n--;
+    }
+}
+
+
+// Refill the input buffer.
+// This method will sit in a loop until (A) the buffer is full or (B)
+// the stream's read() method reports and end of file condition.
+void jpeg_decoder::prep_in_buffer()
+{
+    m_in_buf_left = 0;
+    m_pIn_buf_ofs = m_in_buf;
+
+    if (m_eof_flag) return;
+
+    do {
+        int bytes_read = m_pStream->read(m_in_buf + m_in_buf_left, JPGD_IN_BUF_SIZE - m_in_buf_left, &m_eof_flag);
+        if (bytes_read == -1) stop_decoding(JPGD_STREAM_READ);
+        m_in_buf_left += bytes_read;
+    } while ((m_in_buf_left < JPGD_IN_BUF_SIZE) && (!m_eof_flag));
+
+    m_total_bytes_read += m_in_buf_left;
+
+    // Pad the end of the block with M_EOI (prevents the decompressor from going off the rails if the stream is invalid).
+    // (This dates way back to when this decompressor was written in C/asm, and the all-asm Huffman decoder did some fancy things to increase perf.)
+    word_clear(m_pIn_buf_ofs + m_in_buf_left, 0xD9FF, 64);
+}
+
+
+// Read a Huffman code table.
+void jpeg_decoder::read_dht_marker()
+{
+    int i, index, count;
+    uint8_t huff_num[17];
+    uint8_t huff_val[256];
+    uint32_t num_left = get_bits(16);
+
+    if (num_left < 2) stop_decoding(JPGD_BAD_DHT_MARKER);
+    num_left -= 2;
+
+    while (num_left) {
+        index = get_bits(8);
+        huff_num[0] = 0;
+        count = 0;
+
+        for (i = 1; i <= 16; i++) {
+            huff_num[i] = static_cast<uint8_t>(get_bits(8));
+            count += huff_num[i];
+        }
+
+        if (count > 255) stop_decoding(JPGD_BAD_DHT_COUNTS);
+
+        for (i = 0; i < count; i++)
+            huff_val[i] = static_cast<uint8_t>(get_bits(8));
+
+        i = 1 + 16 + count;
+
+        if (num_left < (uint32_t)i) stop_decoding(JPGD_BAD_DHT_MARKER);
+        num_left -= i;
+
+        if ((index & 0x10) > 0x10) stop_decoding(JPGD_BAD_DHT_INDEX);
+        index = (index & 0x0F) + ((index & 0x10) >> 4) * (JPGD_MAX_HUFF_TABLES >> 1);
+        if (index >= JPGD_MAX_HUFF_TABLES) stop_decoding(JPGD_BAD_DHT_INDEX);
+
+        if (!m_huff_num[index]) m_huff_num[index] = (uint8_t *)alloc(17);
+        if (!m_huff_val[index]) m_huff_val[index] = (uint8_t *)alloc(256);
+
+        m_huff_ac[index] = (index & 0x10) != 0;
+        memcpy(m_huff_num[index], huff_num, 17);
+        memcpy(m_huff_val[index], huff_val, 256);
+    }
+}
+
+
+// Read a quantization table.
+void jpeg_decoder::read_dqt_marker()
+{
+    int n, i, prec;
+    uint32_t temp;
+    uint32_t num_left = get_bits(16);
+    if (num_left < 2) stop_decoding(JPGD_BAD_DQT_MARKER);
+    num_left -= 2;
+
+    while (num_left) {
+        n = get_bits(8);
+        prec = n >> 4;
+        n &= 0x0F;
+
+        if (n >= JPGD_MAX_QUANT_TABLES) stop_decoding(JPGD_BAD_DQT_TABLE);
+
+        if (!m_quant[n]) m_quant[n] = (jpgd_quant_t *)alloc(64 * sizeof(jpgd_quant_t));
+
+        // read quantization entries, in zag order
+        for (i = 0; i < 64; i++) {
+            temp = get_bits(8);
+            if (prec) temp = (temp << 8) + get_bits(8);
+            m_quant[n][i] = static_cast<jpgd_quant_t>(temp);
+        }
+        i = 64 + 1;
+        if (prec) i += 64;
+        if (num_left < (uint32_t)i) stop_decoding(JPGD_BAD_DQT_LENGTH);
+        num_left -= i;
+    }
+}
+
+
+// Read the start of frame (SOF) marker.
+void jpeg_decoder::read_sof_marker()
+{
+    int i;
+    uint32_t num_left = get_bits(16);
+
+    if (get_bits(8) != 8) stop_decoding(JPGD_BAD_PRECISION);   /* precision: sorry, only 8-bit precision is supported right now */
+
+    m_image_y_size = get_bits(16);
+    if ((m_image_y_size < 1) || (m_image_y_size > JPGD_MAX_HEIGHT)) stop_decoding(JPGD_BAD_HEIGHT);
+
+    m_image_x_size = get_bits(16);
+    if ((m_image_x_size < 1) || (m_image_x_size > JPGD_MAX_WIDTH)) stop_decoding(JPGD_BAD_WIDTH);
+
+    m_comps_in_frame = get_bits(8);
+    if (m_comps_in_frame > JPGD_MAX_COMPONENTS) stop_decoding(JPGD_TOO_MANY_COMPONENTS);
+
+    if (num_left != (uint32_t)(m_comps_in_frame * 3 + 8)) stop_decoding(JPGD_BAD_SOF_LENGTH);
+
+    for (i = 0; i < m_comps_in_frame; i++) {
+        m_comp_ident[i]  = get_bits(8);
+        m_comp_h_samp[i] = get_bits(4);
+        m_comp_v_samp[i] = get_bits(4);
+        m_comp_quant[i]  = get_bits(8);
+    }
+}
+
+
+// Used to skip unrecognized markers.
+void jpeg_decoder::skip_variable_marker()
+{
+    uint32_t num_left = get_bits(16);
+    if (num_left < 2) stop_decoding(JPGD_BAD_VARIABLE_MARKER);
+    num_left -= 2;
+
+    while (num_left) {
+        get_bits(8);
+        num_left--;
+    }
+}
+
+
+// Read a define restart interval (DRI) marker.
+void jpeg_decoder::read_dri_marker()
+{
+    if (get_bits(16) != 4) stop_decoding(JPGD_BAD_DRI_LENGTH);
+    m_restart_interval = get_bits(16);
+}
+
+
+// Read a start of scan (SOS) marker.
+void jpeg_decoder::read_sos_marker()
+{
+    int i, ci, c, cc;
+    uint32_t num_left = get_bits(16);
+    int n = get_bits(8);
+
+    m_comps_in_scan = n;
+    num_left -= 3;
+
+    if ( (num_left != (uint32_t)(n * 2 + 3)) || (n < 1) || (n > JPGD_MAX_COMPS_IN_SCAN) ) stop_decoding(JPGD_BAD_SOS_LENGTH);
+
+    for (i = 0; i < n; i++) {
+        cc = get_bits(8);
+        c = get_bits(8);
+        num_left -= 2;
+
+        for (ci = 0; ci < m_comps_in_frame; ci++)
+          if (cc == m_comp_ident[ci]) break;
+
+        if (ci >= m_comps_in_frame) stop_decoding(JPGD_BAD_SOS_COMP_ID);
+
+        m_comp_list[i]    = ci;
+        m_comp_dc_tab[ci] = (c >> 4) & 15;
+        m_comp_ac_tab[ci] = (c & 15) + (JPGD_MAX_HUFF_TABLES >> 1);
+    }
+    m_spectral_start  = get_bits(8);
+    m_spectral_end    = get_bits(8);
+    m_successive_high = get_bits(4);
+    m_successive_low  = get_bits(4);
+
+    if (!m_progressive_flag) {
+        m_spectral_start = 0;
+        m_spectral_end = 63;
+    }
+    num_left -= 3;
+
+    while (num_left) {    /* read past whatever is num_left */
+        get_bits(8);
+        num_left--;
+    }
+}
+
+
+// Finds the next marker.
+int jpeg_decoder::next_marker()
+{
+    uint32_t c;
+
+    do {
+        do {
+            c = get_bits(8);
+        } while (c != 0xFF);
+
+        do {
+            c = get_bits(8);
+        } while (c == 0xFF);
+    } while (c == 0);
+
+    return c;
+}
+
+
+// Process markers. Returns when an SOFx, SOI, EOI, or SOS marker is
+// encountered.
+int jpeg_decoder::process_markers()
+{
+    int c;
+
+    for ( ; ; ) {
+        c = next_marker();
+        switch (c) {
+            case M_SOF0:
+            case M_SOF1:
+            case M_SOF2:
+            case M_SOF3:
+            case M_SOF5:
+            case M_SOF6:
+            case M_SOF7:
+      //      case M_JPG:
+            case M_SOF9:
+            case M_SOF10:
+            case M_SOF11:
+            case M_SOF13:
+            case M_SOF14:
+            case M_SOF15:
+            case M_SOI:
+            case M_EOI:
+            case M_SOS: return c;
+            case M_DHT: {
+                read_dht_marker();
+                break;
+            }
+            // No arithmitic support - dumb patents!
+            case M_DAC: {
+                stop_decoding(JPGD_NO_ARITHMITIC_SUPPORT);
+                break;
+            }
+            case M_DQT: {
+                read_dqt_marker();
+                break;
+            }
+            case M_DRI: {
+                read_dri_marker();
+                break;
+            }
+            //case M_APP0:  /* no need to read the JFIF marker */
+            case M_JPG:
+            case M_RST0:    /* no parameters */
+            case M_RST1:
+            case M_RST2:
+            case M_RST3:
+            case M_RST4:
+            case M_RST5:
+            case M_RST6:
+            case M_RST7:
+            case M_TEM: {
+                stop_decoding(JPGD_UNEXPECTED_MARKER);
+                break;
+            }
+            default: {   /* must be DNL, DHP, EXP, APPn, JPGn, COM, or RESn or APP0 */
+                skip_variable_marker();
+                break;
+            }
+        }
+    }
+}
+
+
+// Finds the start of image (SOI) marker.
+// This code is rather defensive: it only checks the first 512 bytes to avoid
+// false positives.
+void jpeg_decoder::locate_soi_marker()
+{
+    uint32_t lastchar = get_bits(8);
+    uint32_t thischar = get_bits(8);
+
+    /* ok if it's a normal JPEG file without a special header */
+    if ((lastchar == 0xFF) && (thischar == M_SOI)) return;
+
+    uint32_t bytesleft = 4096; //512;
+
+    while (true) {
+        if (--bytesleft == 0) stop_decoding(JPGD_NOT_JPEG);
+
+        lastchar = thischar;
+        thischar = get_bits(8);
+
+        if (lastchar == 0xFF) {
+          if (thischar == M_SOI) break;
+          else if (thischar == M_EOI) stop_decoding(JPGD_NOT_JPEG); // get_bits will keep returning M_EOI if we read past the end
+        }
+    }
+
+    // Check the next character after marker: if it's not 0xFF, it can't be the start of the next marker, so the file is bad.
+    thischar = (m_bit_buf >> 24) & 0xFF;
+    if (thischar != 0xFF) stop_decoding(JPGD_NOT_JPEG);
+}
+
+
+// Find a start of frame (SOF) marker.
+void jpeg_decoder::locate_sof_marker()
+{
+    locate_soi_marker();
+    int c = process_markers();
+
+    switch (c) {
+        case M_SOF2: m_progressive_flag = true;
+        case M_SOF0:  /* baseline DCT */
+        case M_SOF1: { /* extended sequential DCT */
+          read_sof_marker();
+          break;
+        }
+        case M_SOF9: {  /* Arithmitic coding */
+          stop_decoding(JPGD_NO_ARITHMITIC_SUPPORT);
+          break;
+        }
+        default: {
+          stop_decoding(JPGD_UNSUPPORTED_MARKER);
+          break;
+        }
+    }
+}
+
+
+// Find a start of scan (SOS) marker.
+int jpeg_decoder::locate_sos_marker()
+{
+    int c = process_markers();
+    if (c == M_EOI) return false;
+    else if (c != M_SOS) stop_decoding(JPGD_UNEXPECTED_MARKER);
+    read_sos_marker();
+    return true;
+}
+
+
+// Reset everything to default/uninitialized state.
+void jpeg_decoder::init(jpeg_decoder_stream *pStream)
+{
+    m_pMem_blocks = nullptr;
+    m_error_code = JPGD_SUCCESS;
+    m_ready_flag = false;
+    m_image_x_size = m_image_y_size = 0;
+    m_pStream = pStream;
+    m_progressive_flag = false;
+
+    memset(m_huff_ac, 0, sizeof(m_huff_ac));
+    memset(m_huff_num, 0, sizeof(m_huff_num));
+    memset(m_huff_val, 0, sizeof(m_huff_val));
+    memset(m_quant, 0, sizeof(m_quant));
+
+    m_scan_type = 0;
+    m_comps_in_frame = 0;
+
+    memset(m_comp_h_samp, 0, sizeof(m_comp_h_samp));
+    memset(m_comp_v_samp, 0, sizeof(m_comp_v_samp));
+    memset(m_comp_quant, 0, sizeof(m_comp_quant));
+    memset(m_comp_ident, 0, sizeof(m_comp_ident));
+    memset(m_comp_h_blocks, 0, sizeof(m_comp_h_blocks));
+    memset(m_comp_v_blocks, 0, sizeof(m_comp_v_blocks));
+
+    m_comps_in_scan = 0;
+    memset(m_comp_list, 0, sizeof(m_comp_list));
+    memset(m_comp_dc_tab, 0, sizeof(m_comp_dc_tab));
+    memset(m_comp_ac_tab, 0, sizeof(m_comp_ac_tab));
+
+    m_spectral_start = 0;
+    m_spectral_end = 0;
+    m_successive_low = 0;
+    m_successive_high = 0;
+    m_max_mcu_x_size = 0;
+    m_max_mcu_y_size = 0;
+    m_blocks_per_mcu = 0;
+    m_max_blocks_per_row = 0;
+    m_mcus_per_row = 0;
+    m_mcus_per_col = 0;
+    m_expanded_blocks_per_component = 0;
+    m_expanded_blocks_per_mcu = 0;
+    m_expanded_blocks_per_row = 0;
+    m_freq_domain_chroma_upsample = false;
+
+    memset(m_mcu_org, 0, sizeof(m_mcu_org));
+
+    m_total_lines_left = 0;
+    m_mcu_lines_left = 0;
+    m_real_dest_bytes_per_scan_line = 0;
+    m_dest_bytes_per_scan_line = 0;
+    m_dest_bytes_per_pixel = 0;
+
+    memset(m_pHuff_tabs, 0, sizeof(m_pHuff_tabs));
+
+    memset(m_dc_coeffs, 0, sizeof(m_dc_coeffs));
+    memset(m_ac_coeffs, 0, sizeof(m_ac_coeffs));
+    memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
+
+    m_eob_run = 0;
+
+    memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
+
+    m_pIn_buf_ofs = m_in_buf;
+    m_in_buf_left = 0;
+    m_eof_flag = false;
+    m_tem_flag = 0;
+
+    memset(m_in_buf_pad_start, 0, sizeof(m_in_buf_pad_start));
+    memset(m_in_buf, 0, sizeof(m_in_buf));
+    memset(m_in_buf_pad_end, 0, sizeof(m_in_buf_pad_end));
+
+    m_restart_interval = 0;
+    m_restarts_left    = 0;
+    m_next_restart_num = 0;
+
+    m_max_mcus_per_row = 0;
+    m_max_blocks_per_mcu = 0;
+    m_max_mcus_per_col = 0;
+
+    memset(m_last_dc_val, 0, sizeof(m_last_dc_val));
+    m_pMCU_coefficients = nullptr;
+    m_pSample_buf = nullptr;
+
+    m_total_bytes_read = 0;
+
+    m_pScan_line_0 = nullptr;
+    m_pScan_line_1 = nullptr;
+
+    // Ready the input buffer.
+    prep_in_buffer();
+
+    // Prime the bit buffer.
+    m_bits_left = 16;
+    m_bit_buf = 0;
+
+    get_bits(16);
+    get_bits(16);
+
+    for (int i = 0; i < JPGD_MAX_BLOCKS_PER_MCU; i++) {
+        m_mcu_block_max_zag[i] = 64;
+    }
+}
+
+#define SCALEBITS 16
+#define ONE_HALF  ((int) 1 << (SCALEBITS-1))
+#define FIX(x)    ((int) ((x) * (1L<<SCALEBITS) + 0.5f))
+
+
+// Create a few tables that allow us to quickly convert YCbCr to RGB.
+void jpeg_decoder::create_look_ups()
+{
+  for (int i = 0; i <= 255; i++) {
+      int k = i - 128;
+      m_crr[i] = ( FIX(1.40200f)  * k + ONE_HALF) >> SCALEBITS;
+      m_cbb[i] = ( FIX(1.77200f)  * k + ONE_HALF) >> SCALEBITS;
+      m_crg[i] = (-FIX(0.71414f)) * k;
+      m_cbg[i] = (-FIX(0.34414f)) * k + ONE_HALF;
+  }
+}
+
+
+// This method throws back into the stream any bytes that where read
+// into the bit buffer during initial marker scanning.
+void jpeg_decoder::fix_in_buffer()
+{
+    // In case any 0xFF's where pulled into the buffer during marker scanning.
+    JPGD_ASSERT((m_bits_left & 7) == 0);
+
+    if (m_bits_left == 16) stuff_char( (uint8_t)(m_bit_buf & 0xFF));
+    if (m_bits_left >= 8) stuff_char( (uint8_t)((m_bit_buf >> 8) & 0xFF));
+
+    stuff_char((uint8_t)((m_bit_buf >> 16) & 0xFF));
+    stuff_char((uint8_t)((m_bit_buf >> 24) & 0xFF));
+
+    m_bits_left = 16;
+    get_bits_no_markers(16);
+    get_bits_no_markers(16);
+}
+
+
+void jpeg_decoder::transform_mcu(int mcu_row)
+{
+    jpgd_block_t* pSrc_ptr = m_pMCU_coefficients;
+    uint8_t* pDst_ptr = m_pSample_buf + mcu_row * m_blocks_per_mcu * 64;
+
+    for (int mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++) {
+        idct(pSrc_ptr, pDst_ptr, m_mcu_block_max_zag[mcu_block]);
+        pSrc_ptr += 64;
+        pDst_ptr += 64;
+    }
+}
+
+
+static const uint8_t s_max_rc[64] =
+{
+    17, 18, 34, 50, 50, 51, 52, 52, 52, 68, 84, 84, 84, 84, 85, 86, 86, 86, 86, 86,
+    102, 118, 118, 118, 118, 118, 118, 119, 120, 120, 120, 120, 120, 120, 120, 136,
+    136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136,
+    136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136, 136
+};
+
+
+void jpeg_decoder::transform_mcu_expand(int mcu_row)
+{
+    jpgd_block_t* pSrc_ptr = m_pMCU_coefficients;
+    uint8_t* pDst_ptr = m_pSample_buf + mcu_row * m_expanded_blocks_per_mcu * 64;
+
+    // Y IDCT
+    int mcu_block;
+    for (mcu_block = 0; mcu_block < m_expanded_blocks_per_component; mcu_block++) {
+        idct(pSrc_ptr, pDst_ptr, m_mcu_block_max_zag[mcu_block]);
+        pSrc_ptr += 64;
+        pDst_ptr += 64;
+    }
+
+    // Chroma IDCT, with upsampling
+    jpgd_block_t temp_block[64];
+
+    for (int i = 0; i < 2; i++) {
+        DCT_Upsample::Matrix44 P, Q, R, S;
+        JPGD_ASSERT(m_mcu_block_max_zag[mcu_block] >= 1);
+        JPGD_ASSERT(m_mcu_block_max_zag[mcu_block] <= 64);
+
+        int max_zag = m_mcu_block_max_zag[mcu_block++] - 1;
+        if (max_zag <= 0) max_zag = 0; // should never happen, only here to shut up static analysis
+
+        switch (s_max_rc[max_zag]) {
+            case 1*16+1:
+                DCT_Upsample::P_Q<1, 1>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<1, 1>::calc(R, S, pSrc_ptr);
+                break;
+            case 1*16+2:
+                DCT_Upsample::P_Q<1, 2>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<1, 2>::calc(R, S, pSrc_ptr);
+                break;
+            case 2*16+2:
+                DCT_Upsample::P_Q<2, 2>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<2, 2>::calc(R, S, pSrc_ptr);
+                break;
+            case 3*16+2:
+                DCT_Upsample::P_Q<3, 2>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<3, 2>::calc(R, S, pSrc_ptr);
+                break;
+            case 3*16+3:
+                DCT_Upsample::P_Q<3, 3>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<3, 3>::calc(R, S, pSrc_ptr);
+                break;
+            case 3*16+4:
+                DCT_Upsample::P_Q<3, 4>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<3, 4>::calc(R, S, pSrc_ptr);
+                break;
+            case 4*16+4:
+                DCT_Upsample::P_Q<4, 4>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<4, 4>::calc(R, S, pSrc_ptr);
+                break;
+            case 5*16+4:
+                DCT_Upsample::P_Q<5, 4>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<5, 4>::calc(R, S, pSrc_ptr);
+                break;
+            case 5*16+5:
+                DCT_Upsample::P_Q<5, 5>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<5, 5>::calc(R, S, pSrc_ptr);
+                break;
+            case 5*16+6:
+                DCT_Upsample::P_Q<5, 6>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<5, 6>::calc(R, S, pSrc_ptr);
+                break;
+            case 6*16+6:
+                DCT_Upsample::P_Q<6, 6>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<6, 6>::calc(R, S, pSrc_ptr);
+                break;
+            case 7*16+6:
+                DCT_Upsample::P_Q<7, 6>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<7, 6>::calc(R, S, pSrc_ptr);
+                break;
+            case 7*16+7:
+                DCT_Upsample::P_Q<7, 7>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<7, 7>::calc(R, S, pSrc_ptr);
+                break;
+            case 7*16+8:
+                DCT_Upsample::P_Q<7, 8>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<7, 8>::calc(R, S, pSrc_ptr);
+                break;
+            case 8*16+8:
+                DCT_Upsample::P_Q<8, 8>::calc(P, Q, pSrc_ptr);
+                DCT_Upsample::R_S<8, 8>::calc(R, S, pSrc_ptr);
+                break;
+            default:
+                JPGD_ASSERT(false);
+        }
+        DCT_Upsample::Matrix44 a(P + Q); P -= Q;
+        DCT_Upsample::Matrix44& b = P;
+        DCT_Upsample::Matrix44 c(R + S); R -= S;
+        DCT_Upsample::Matrix44& d = R;
+
+        DCT_Upsample::Matrix44::add_and_store(temp_block, a, c);
+        idct_4x4(temp_block, pDst_ptr);
+        pDst_ptr += 64;
+
+        DCT_Upsample::Matrix44::sub_and_store(temp_block, a, c);
+        idct_4x4(temp_block, pDst_ptr);
+        pDst_ptr += 64;
+
+        DCT_Upsample::Matrix44::add_and_store(temp_block, b, d);
+        idct_4x4(temp_block, pDst_ptr);
+        pDst_ptr += 64;
+
+        DCT_Upsample::Matrix44::sub_and_store(temp_block, b, d);
+        idct_4x4(temp_block, pDst_ptr);
+        pDst_ptr += 64;
+        pSrc_ptr += 64;
+    }
+}
+
+
+// Loads and dequantizes the next row of (already decoded) coefficients.
+// Progressive images only.
+void jpeg_decoder::load_next_row()
+{
+    int i;
+    jpgd_block_t *p;
+    jpgd_quant_t *q;
+    int mcu_row, mcu_block;
+    int component_num, component_id;
+    int block_x_mcu[JPGD_MAX_COMPONENTS];
+
+    memset(block_x_mcu, 0, JPGD_MAX_COMPONENTS * sizeof(int));
+
+    for (mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++) {
+        int block_x_mcu_ofs = 0, block_y_mcu_ofs = 0;
+
+        for (mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++) {
+            component_id = m_mcu_org[mcu_block];
+            q = m_quant[m_comp_quant[component_id]];
+            p = m_pMCU_coefficients + 64 * mcu_block;
+
+            jpgd_block_t* pAC = coeff_buf_getp(m_ac_coeffs[component_id], block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
+            jpgd_block_t* pDC = coeff_buf_getp(m_dc_coeffs[component_id], block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
+            p[0] = pDC[0];
+            memcpy(&p[1], &pAC[1], 63 * sizeof(jpgd_block_t));
+
+            for (i = 63; i > 0; i--) {
+                if (p[g_ZAG[i]]) break;
+            }
+
+            m_mcu_block_max_zag[mcu_block] = i + 1;
+
+            for ( ; i >= 0; i--) {
+                if (p[g_ZAG[i]]) {
+                    p[g_ZAG[i]] = static_cast<jpgd_block_t>(p[g_ZAG[i]] * q[i]);
+                }
+            }
+
+            if (m_comps_in_scan == 1) block_x_mcu[component_id]++;
+            else {
+                if (++block_x_mcu_ofs == m_comp_h_samp[component_id]) block_x_mcu_ofs = 0;
+                if (++block_y_mcu_ofs == m_comp_v_samp[component_id]) {
+                    block_y_mcu_ofs = 0;
+                    block_x_mcu[component_id] += m_comp_h_samp[component_id];
+                }
+            }
+        }
+        if (m_freq_domain_chroma_upsample) transform_mcu_expand(mcu_row);
+        else transform_mcu(mcu_row);
+    }
+    if (m_comps_in_scan == 1) m_block_y_mcu[m_comp_list[0]]++;
+    else {
+        for (component_num = 0; component_num < m_comps_in_scan; component_num++) {
+            component_id = m_comp_list[component_num];
+            m_block_y_mcu[component_id] += m_comp_v_samp[component_id];
+        }
+    }
+}
+
+
+// Restart interval processing.
+void jpeg_decoder::process_restart()
+{
+    int i;
+    int c = 0;
+
+    // Align to a byte boundry
+    // FIXME: Is this really necessary? get_bits_no_markers() never reads in markers!
+    //get_bits_no_markers(m_bits_left & 7);
+
+    // Let's scan a little bit to find the marker, but not _too_ far.
+    // 1536 is a "fudge factor" that determines how much to scan.
+    for (i = 1536; i > 0; i--) {
+        if (get_char() == 0xFF) break;
+    }
+    if (i == 0) stop_decoding(JPGD_BAD_RESTART_MARKER);
+
+    for ( ; i > 0; i--) {
+        if ((c = get_char()) != 0xFF) break;
+    }
+    if (i == 0) stop_decoding(JPGD_BAD_RESTART_MARKER);
+
+    // Is it the expected marker? If not, something bad happened.
+    if (c != (m_next_restart_num + M_RST0)) stop_decoding(JPGD_BAD_RESTART_MARKER);
+
+    // Reset each component's DC prediction values.
+    memset(&m_last_dc_val, 0, m_comps_in_frame * sizeof(uint32_t));
+
+    m_eob_run = 0;
+    m_restarts_left = m_restart_interval;
+    m_next_restart_num = (m_next_restart_num + 1) & 7;
+
+    // Get the bit buffer going again...
+    m_bits_left = 16;
+    get_bits_no_markers(16);
+    get_bits_no_markers(16);
+}
+
+
+static inline int dequantize_ac(int c, int q)
+{
+    c *= q;
+    return c;
+}
+
+// Decodes and dequantizes the next row of coefficients.
+void jpeg_decoder::decode_next_row()
+{
+    for (int mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++) {
+        if ((m_restart_interval) && (m_restarts_left == 0)) process_restart();
+
+        jpgd_block_t* p = m_pMCU_coefficients;
+
+        for (int mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++, p += 64) {
+            int component_id = m_mcu_org[mcu_block];
+            jpgd_quant_t* q = m_quant[m_comp_quant[component_id]];
+
+            int r, s;
+            s = huff_decode(m_pHuff_tabs[m_comp_dc_tab[component_id]], r);
+            s = JPGD_HUFF_EXTEND(r, s);
+
+            m_last_dc_val[component_id] = (s += m_last_dc_val[component_id]);
+
+            p[0] = static_cast<jpgd_block_t>(s * q[0]);
+
+            int prev_num_set = m_mcu_block_max_zag[mcu_block];
+            huff_tables *pH = m_pHuff_tabs[m_comp_ac_tab[component_id]];
+            int k;
+            for (k = 1; k < 64; k++) {
+                int extra_bits;
+                s = huff_decode(pH, extra_bits);
+                r = s >> 4;
+                s &= 15;
+
+                if (s) {
+                    if (r) {
+                        if ((k + r) > 63) stop_decoding(JPGD_DECODE_ERROR);
+                        if (k < prev_num_set) {
+                            int n = JPGD_MIN(r, prev_num_set - k);
+                            int kt = k;
+                            while (n--) p[g_ZAG[kt++]] = 0;
+                        }
+                        k += r;
+                    }
+                    s = JPGD_HUFF_EXTEND(extra_bits, s);
+                    JPGD_ASSERT(k < 64);
+                    p[g_ZAG[k]] = static_cast<jpgd_block_t>(dequantize_ac(s, q[k])); //s * q[k];
+                } else {
+                    if (r == 15) {
+                        if ((k + 16) > 64) stop_decoding(JPGD_DECODE_ERROR);
+                        if (k < prev_num_set) {
+                            int n = JPGD_MIN(16, prev_num_set - k);
+                            int kt = k;
+                            while (n--) {
+                                JPGD_ASSERT(kt <= 63);
+                                p[g_ZAG[kt++]] = 0;
+                            }
+                        }
+                        k += 16 - 1; // - 1 because the loop counter is k
+                        JPGD_ASSERT(p[g_ZAG[k]] == 0);
+                    } else  break;
+                }
+            }
+
+            if (k < prev_num_set) {
+                int kt = k;
+                while (kt < prev_num_set) p[g_ZAG[kt++]] = 0;
+            }
+
+            m_mcu_block_max_zag[mcu_block] = k;
+        }
+        if (m_freq_domain_chroma_upsample) transform_mcu_expand(mcu_row);
+        else transform_mcu(mcu_row);
+        m_restarts_left--;
+    }
+}
+
+
+// YCbCr H1V1 (1x1:1:1, 3 m_blocks per MCU) to RGB
+void jpeg_decoder::H1V1Convert()
+{
+    int row = m_max_mcu_y_size - m_mcu_lines_left;
+    uint8_t *d = m_pScan_line_0;
+    uint8_t *s = m_pSample_buf + row * 8;
+
+    for (int i = m_max_mcus_per_row; i > 0; i--) {
+        for (int j = 0; j < 8; j++) {
+            int y = s[j];
+            int cb = s[64+j];
+            int cr = s[128+j];
+
+            d[0] = clamp(y + m_crr[cr]);
+            d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
+            d[2] = clamp(y + m_cbb[cb]);
+            d[3] = 255;
+            d += 4;
+        }
+        s += 64*3;
+    }
+}
+
+
+// YCbCr H2V1 (2x1:1:1, 4 m_blocks per MCU) to RGB
+void jpeg_decoder::H2V1Convert()
+{
+    int row = m_max_mcu_y_size - m_mcu_lines_left;
+    uint8_t *d0 = m_pScan_line_0;
+    uint8_t *y = m_pSample_buf + row * 8;
+    uint8_t *c = m_pSample_buf + 2*64 + row * 8;
+
+    for (int i = m_max_mcus_per_row; i > 0; i--) {
+        for (int l = 0; l < 2; l++) {
+            for (int j = 0; j < 4; j++) {
+                int cb = c[0];
+                int cr = c[64];
+
+                int rc = m_crr[cr];
+                int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
+                int bc = m_cbb[cb];
+
+                int yy = y[j<<1];
+                d0[0] = clamp(yy+rc);
+                d0[1] = clamp(yy+gc);
+                d0[2] = clamp(yy+bc);
+                d0[3] = 255;
+
+                yy = y[(j<<1)+1];
+                d0[4] = clamp(yy+rc);
+                d0[5] = clamp(yy+gc);
+                d0[6] = clamp(yy+bc);
+                d0[7] = 255;
+                d0 += 8;
+                c++;
+            }
+            y += 64;
+        }
+        y += 64*4 - 64*2;
+        c += 64*4 - 8;
+    }
+}
+
+
+// YCbCr H2V1 (1x2:1:1, 4 m_blocks per MCU) to RGB
+void jpeg_decoder::H1V2Convert()
+{
+    int row = m_max_mcu_y_size - m_mcu_lines_left;
+    uint8_t *d0 = m_pScan_line_0;
+    uint8_t *d1 = m_pScan_line_1;
+    uint8_t *y;
+    uint8_t *c;
+
+    if (row < 8) y = m_pSample_buf + row * 8;
+    else y = m_pSample_buf + 64*1 + (row & 7) * 8;
+
+    c = m_pSample_buf + 64*2 + (row >> 1) * 8;
+
+    for (int i = m_max_mcus_per_row; i > 0; i--) {
+        for (int j = 0; j < 8; j++) {
+            int cb = c[0+j];
+            int cr = c[64+j];
+
+            int rc = m_crr[cr];
+            int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
+            int bc = m_cbb[cb];
+
+            int yy = y[j];
+            d0[0] = clamp(yy+rc);
+            d0[1] = clamp(yy+gc);
+            d0[2] = clamp(yy+bc);
+            d0[3] = 255;
+
+            yy = y[8+j];
+            d1[0] = clamp(yy+rc);
+            d1[1] = clamp(yy+gc);
+            d1[2] = clamp(yy+bc);
+            d1[3] = 255;
+
+            d0 += 4;
+            d1 += 4;
+        }
+        y += 64*4;
+        c += 64*4;
+    }
+}
+
+
+// YCbCr H2V2 (2x2:1:1, 6 m_blocks per MCU) to RGB
+void jpeg_decoder::H2V2Convert()
+{
+    int row = m_max_mcu_y_size - m_mcu_lines_left;
+    uint8_t *d0 = m_pScan_line_0;
+    uint8_t *d1 = m_pScan_line_1;
+    uint8_t *y;
+    uint8_t *c;
+
+    if (row < 8) y = m_pSample_buf + row * 8;
+    else y = m_pSample_buf + 64*2 + (row & 7) * 8;
+
+    c = m_pSample_buf + 64*4 + (row >> 1) * 8;
+
+    for (int i = m_max_mcus_per_row; i > 0; i--) {
+        for (int l = 0; l < 2; l++) {
+            for (int j = 0; j < 8; j += 2) {
+                int cb = c[0];
+                int cr = c[64];
+
+                int rc = m_crr[cr];
+                int gc = ((m_crg[cr] + m_cbg[cb]) >> 16);
+                int bc = m_cbb[cb];
+
+                int yy = y[j];
+                d0[0] = clamp(yy+rc);
+                d0[1] = clamp(yy+gc);
+                d0[2] = clamp(yy+bc);
+                d0[3] = 255;
+
+                yy = y[j+1];
+                d0[4] = clamp(yy+rc);
+                d0[5] = clamp(yy+gc);
+                d0[6] = clamp(yy+bc);
+                d0[7] = 255;
+
+                yy = y[j+8];
+                d1[0] = clamp(yy+rc);
+                d1[1] = clamp(yy+gc);
+                d1[2] = clamp(yy+bc);
+                d1[3] = 255;
+
+                yy = y[j+8+1];
+                d1[4] = clamp(yy+rc);
+                d1[5] = clamp(yy+gc);
+                d1[6] = clamp(yy+bc);
+                d1[7] = 255;
+
+                d0 += 8;
+                d1 += 8;
+
+                c++;
+            }
+            y += 64;
+        }
+        y += 64*6 - 64*2;
+        c += 64*6 - 8;
+    }
+}
+
+
+// Y (1 block per MCU) to 8-bit grayscale
+void jpeg_decoder::gray_convert()
+{
+    int row = m_max_mcu_y_size - m_mcu_lines_left;
+    uint8_t *d = m_pScan_line_0;
+    uint8_t *s = m_pSample_buf + row * 8;
+
+    for (int i = m_max_mcus_per_row; i > 0; i--) {
+        *(uint32_t *)d = *(uint32_t *)s;
+        *(uint32_t *)(&d[4]) = *(uint32_t *)(&s[4]);
+        s += 64;
+        d += 8;
+    }
+}
+
+
+void jpeg_decoder::expanded_convert()
+{
+    int row = m_max_mcu_y_size - m_mcu_lines_left;
+    uint8_t* Py = m_pSample_buf + (row / 8) * 64 * m_comp_h_samp[0] + (row & 7) * 8;
+    uint8_t* d = m_pScan_line_0;
+
+    for (int i = m_max_mcus_per_row; i > 0; i--) {
+        for (int k = 0; k < m_max_mcu_x_size; k += 8) {
+            const int Y_ofs = k * 8;
+            const int Cb_ofs = Y_ofs + 64 * m_expanded_blocks_per_component;
+            const int Cr_ofs = Y_ofs + 64 * m_expanded_blocks_per_component * 2;
+            for (int j = 0; j < 8; j++) {
+                int y = Py[Y_ofs + j];
+                int cb = Py[Cb_ofs + j];
+                int cr = Py[Cr_ofs + j];
+
+                d[0] = clamp(y + m_crr[cr]);
+                d[1] = clamp(y + ((m_crg[cr] + m_cbg[cb]) >> 16));
+                d[2] = clamp(y + m_cbb[cb]);
+                d[3] = 255;
+
+                d += 4;
+            }
+        }
+        Py += 64 * m_expanded_blocks_per_mcu;
+    }
+}
+
+
+// Find end of image (EOI) marker, so we can return to the user the exact size of the input stream.
+void jpeg_decoder::find_eoi()
+{
+    if (!m_progressive_flag) {
+        // Attempt to read the EOI marker.
+        //get_bits_no_markers(m_bits_left & 7);
+
+        // Prime the bit buffer
+        m_bits_left = 16;
+        get_bits(16);
+        get_bits(16);
+
+        // The next marker _should_ be EOI
+        process_markers();
+    }
+    m_total_bytes_read -= m_in_buf_left;
+}
+
+
+int jpeg_decoder::decode(const void** pScan_line, uint32_t* pScan_line_len)
+{
+    if ((m_error_code) || (!m_ready_flag)) return JPGD_FAILED;
+    if (m_total_lines_left == 0) return JPGD_DONE;
+    if (m_mcu_lines_left == 0) {
+        if (setjmp(m_jmp_state)) return JPGD_FAILED;
+        if (m_progressive_flag) load_next_row();
+        else decode_next_row();
+        // Find the EOI marker if that was the last row.
+        if (m_total_lines_left <= m_max_mcu_y_size) find_eoi();
+        m_mcu_lines_left = m_max_mcu_y_size;
+    }
+
+    if (m_freq_domain_chroma_upsample) {
+        expanded_convert();
+        *pScan_line = m_pScan_line_0;
+    } else {
+        switch (m_scan_type) {
+            case JPGD_YH2V2: {
+                if ((m_mcu_lines_left & 1) == 0) {
+                    H2V2Convert();
+                    *pScan_line = m_pScan_line_0;
+                }
+              else *pScan_line = m_pScan_line_1;
+              break;
+            }
+            case JPGD_YH2V1: {
+                H2V1Convert();
+                *pScan_line = m_pScan_line_0;
+                break;
+            }
+            case JPGD_YH1V2: {
+                if ((m_mcu_lines_left & 1) == 0) {
+                    H1V2Convert();
+                    *pScan_line = m_pScan_line_0;
+                } else *pScan_line = m_pScan_line_1;
+                break;
+            }
+            case JPGD_YH1V1: {
+                H1V1Convert();
+                *pScan_line = m_pScan_line_0;
+                break;
+            }
+            case JPGD_GRAYSCALE: {
+                gray_convert();
+                *pScan_line = m_pScan_line_0;
+                break;
+            }
+        }
+    }
+
+    *pScan_line_len = m_real_dest_bytes_per_scan_line;
+    m_mcu_lines_left--;
+    m_total_lines_left--;
+
+    return JPGD_SUCCESS;
+}
+
+
+// Creates the tables needed for efficient Huffman decoding.
+void jpeg_decoder::make_huff_table(int index, huff_tables *pH)
+{
+    int p, i, l, si;
+    uint8_t huffsize[257];
+    uint32_t huffcode[257];
+    uint32_t code;
+    uint32_t subtree;
+    int code_size;
+    int lastp;
+    int nextfreeentry;
+    int currententry;
+
+    pH->ac_table = m_huff_ac[index] != 0;
+    p = 0;
+
+    for (l = 1; l <= 16; l++)  {
+        for (i = 1; i <= m_huff_num[index][l]; i++) {
+            huffsize[p++] = static_cast<uint8_t>(l);
+        }
+    }
+
+    huffsize[p] = 0;
+    lastp = p;
+    code = 0;
+    si = huffsize[0];
+    p = 0;
+
+    while (huffsize[p]) {
+        while (huffsize[p] == si) {
+            huffcode[p++] = code;
+            code++;
+        }
+        code <<= 1;
+        si++;
+    }
+
+    memset(pH->look_up, 0, sizeof(pH->look_up));
+    memset(pH->look_up2, 0, sizeof(pH->look_up2));
+    memset(pH->tree, 0, sizeof(pH->tree));
+    memset(pH->code_size, 0, sizeof(pH->code_size));
+
+    nextfreeentry = -1;
+    p = 0;
+
+    while (p < lastp) {
+        i = m_huff_val[index][p];
+        code = huffcode[p];
+        code_size = huffsize[p];
+        pH->code_size[i] = static_cast<uint8_t>(code_size);
+
+        if (code_size <= 8) {
+            code <<= (8 - code_size);
+            for (l = 1 << (8 - code_size); l > 0; l--) {
+                JPGD_ASSERT(i < 256);
+                pH->look_up[code] = i;
+                bool has_extrabits = false;
+                int extra_bits = 0;
+                int num_extra_bits = i & 15;
+                int bits_to_fetch = code_size;
+
+                if (num_extra_bits) {
+                    int total_codesize = code_size + num_extra_bits;
+                    if (total_codesize <= 8) {
+                        has_extrabits = true;
+                        extra_bits = ((1 << num_extra_bits) - 1) & (code >> (8 - total_codesize));
+                        JPGD_ASSERT(extra_bits <= 0x7FFF);
+                        bits_to_fetch += num_extra_bits;
+                    }
+                }
+                if (!has_extrabits) pH->look_up2[code] = i | (bits_to_fetch << 8);
+                else pH->look_up2[code] = i | 0x8000 | (extra_bits << 16) | (bits_to_fetch << 8);
+                code++;
+            }
+        } else {
+            subtree = (code >> (code_size - 8)) & 0xFF;
+            currententry = pH->look_up[subtree];
+
+            if (currententry == 0) {
+                pH->look_up[subtree] = currententry = nextfreeentry;
+                pH->look_up2[subtree] = currententry = nextfreeentry;
+                nextfreeentry -= 2;
+            }
+
+            code <<= (16 - (code_size - 8));
+
+            for (l = code_size; l > 9; l--) {
+                if ((code & 0x8000) == 0) currententry--;
+                if (pH->tree[-currententry - 1] == 0) {
+                    pH->tree[-currententry - 1] = nextfreeentry;
+                    currententry = nextfreeentry;
+                    nextfreeentry -= 2;
+                } else currententry = pH->tree[-currententry - 1];
+                code <<= 1;
+            }
+            if ((code & 0x8000) == 0) currententry--;
+            pH->tree[-currententry - 1] = i;
+        }
+        p++;
+    }
+}
+
+
+// Verifies the quantization tables needed for this scan are available.
+void jpeg_decoder::check_quant_tables()
+{
+    for (int i = 0; i < m_comps_in_scan; i++) {
+        if (m_quant[m_comp_quant[m_comp_list[i]]] == nullptr) stop_decoding(JPGD_UNDEFINED_QUANT_TABLE);
+    }
+}
+
+
+// Verifies that all the Huffman tables needed for this scan are available.
+void jpeg_decoder::check_huff_tables()
+{
+    for (int i = 0; i < m_comps_in_scan; i++) {
+      if ((m_spectral_start == 0) && (m_huff_num[m_comp_dc_tab[m_comp_list[i]]] == nullptr)) stop_decoding(JPGD_UNDEFINED_HUFF_TABLE);
+      if ((m_spectral_end > 0) && (m_huff_num[m_comp_ac_tab[m_comp_list[i]]] == nullptr)) stop_decoding(JPGD_UNDEFINED_HUFF_TABLE);
+    }
+
+    for (int i = 0; i < JPGD_MAX_HUFF_TABLES; i++) {
+        if (m_huff_num[i]) {
+            if (!m_pHuff_tabs[i]) m_pHuff_tabs[i] = (huff_tables *)alloc(sizeof(huff_tables));
+            make_huff_table(i, m_pHuff_tabs[i]);
+        }
+    }
+}
+
+
+// Determines the component order inside each MCU.
+// Also calcs how many MCU's are on each row, etc.
+void jpeg_decoder::calc_mcu_block_order()
+{
+    int component_num, component_id;
+    int max_h_samp = 0, max_v_samp = 0;
+
+    for (component_id = 0; component_id < m_comps_in_frame; component_id++) {
+        if (m_comp_h_samp[component_id] > max_h_samp) {
+          max_h_samp = m_comp_h_samp[component_id];
+        }
+        if (m_comp_v_samp[component_id] > max_v_samp) {
+          max_v_samp = m_comp_v_samp[component_id];
+        }
+    }
+
+    for (component_id = 0; component_id < m_comps_in_frame; component_id++) {
+        m_comp_h_blocks[component_id] = ((((m_image_x_size * m_comp_h_samp[component_id]) + (max_h_samp - 1)) / max_h_samp) + 7) / 8;
+        m_comp_v_blocks[component_id] = ((((m_image_y_size * m_comp_v_samp[component_id]) + (max_v_samp - 1)) / max_v_samp) + 7) / 8;
+    }
+
+    if (m_comps_in_scan == 1) {
+        m_mcus_per_row = m_comp_h_blocks[m_comp_list[0]];
+        m_mcus_per_col = m_comp_v_blocks[m_comp_list[0]];
+    } else {
+        m_mcus_per_row = (((m_image_x_size + 7) / 8) + (max_h_samp - 1)) / max_h_samp;
+        m_mcus_per_col = (((m_image_y_size + 7) / 8) + (max_v_samp - 1)) / max_v_samp;
+    }
+
+    if (m_comps_in_scan == 1) {
+        m_mcu_org[0] = m_comp_list[0];
+        m_blocks_per_mcu = 1;
+    } else {
+        m_blocks_per_mcu = 0;
+
+        for (component_num = 0; component_num < m_comps_in_scan; component_num++) {
+            int num_blocks;
+            component_id = m_comp_list[component_num];
+            num_blocks = m_comp_h_samp[component_id] * m_comp_v_samp[component_id];
+            while (num_blocks--) m_mcu_org[m_blocks_per_mcu++] = component_id;
+        }
+    }
+}
+
+
+// Starts a new scan.
+int jpeg_decoder::init_scan()
+{
+    if (!locate_sos_marker()) return false;
+
+    calc_mcu_block_order();
+    check_huff_tables();
+    check_quant_tables();
+
+    memset(m_last_dc_val, 0, m_comps_in_frame * sizeof(uint32_t));
+
+    m_eob_run = 0;
+
+    if (m_restart_interval) {
+        m_restarts_left = m_restart_interval;
+        m_next_restart_num = 0;
+    }
+    fix_in_buffer();
+    return true;
+}
+
+
+// Starts a frame. Determines if the number of components or sampling factors
+// are supported.
+void jpeg_decoder::init_frame()
+{
+    int i;
+
+    if (m_comps_in_frame == 1) {
+        if ((m_comp_h_samp[0] != 1) || (m_comp_v_samp[0] != 1)) stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
+        m_scan_type = JPGD_GRAYSCALE;
+        m_max_blocks_per_mcu = 1;
+        m_max_mcu_x_size = 8;
+        m_max_mcu_y_size = 8;
+    } else if (m_comps_in_frame == 3) {
+        if (((m_comp_h_samp[1] != 1) || (m_comp_v_samp[1] != 1)) || ((m_comp_h_samp[2] != 1) || (m_comp_v_samp[2] != 1)))
+            stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
+
+        if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 1)) {
+            m_scan_type = JPGD_YH1V1;
+            m_max_blocks_per_mcu = 3;
+            m_max_mcu_x_size = 8;
+            m_max_mcu_y_size = 8;
+        } else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 1)) {
+            m_scan_type = JPGD_YH2V1;
+            m_max_blocks_per_mcu = 4;
+            m_max_mcu_x_size = 16;
+            m_max_mcu_y_size = 8;
+        } else if ((m_comp_h_samp[0] == 1) && (m_comp_v_samp[0] == 2)) {
+            m_scan_type = JPGD_YH1V2;
+            m_max_blocks_per_mcu = 4;
+            m_max_mcu_x_size = 8;
+            m_max_mcu_y_size = 16;
+        } else if ((m_comp_h_samp[0] == 2) && (m_comp_v_samp[0] == 2)) {
+            m_scan_type = JPGD_YH2V2;
+            m_max_blocks_per_mcu = 6;
+            m_max_mcu_x_size = 16;
+            m_max_mcu_y_size = 16;
+        } else stop_decoding(JPGD_UNSUPPORTED_SAMP_FACTORS);
+    } else stop_decoding(JPGD_UNSUPPORTED_COLORSPACE);
+
+    m_max_mcus_per_row = (m_image_x_size + (m_max_mcu_x_size - 1)) / m_max_mcu_x_size;
+    m_max_mcus_per_col = (m_image_y_size + (m_max_mcu_y_size - 1)) / m_max_mcu_y_size;
+
+    // These values are for the *destination* pixels: after conversion.
+    if (m_scan_type == JPGD_GRAYSCALE) m_dest_bytes_per_pixel = 1;
+    else m_dest_bytes_per_pixel = 4;
+
+    m_dest_bytes_per_scan_line = ((m_image_x_size + 15) & 0xFFF0) * m_dest_bytes_per_pixel;
+    m_real_dest_bytes_per_scan_line = (m_image_x_size * m_dest_bytes_per_pixel);
+
+    // Initialize two scan line buffers.
+    m_pScan_line_0 = (uint8_t *)alloc(m_dest_bytes_per_scan_line, true);
+    if ((m_scan_type == JPGD_YH1V2) || (m_scan_type == JPGD_YH2V2)) {
+        m_pScan_line_1 = (uint8_t *)alloc(m_dest_bytes_per_scan_line, true);
+    }
+
+    m_max_blocks_per_row = m_max_mcus_per_row * m_max_blocks_per_mcu;
+
+    // Should never happen
+    if (m_max_blocks_per_row > JPGD_MAX_BLOCKS_PER_ROW) stop_decoding(JPGD_ASSERTION_ERROR);
+
+    // Allocate the coefficient buffer, enough for one MCU
+    m_pMCU_coefficients = (jpgd_block_t*)alloc(m_max_blocks_per_mcu * 64 * sizeof(jpgd_block_t));
+
+    for (i = 0; i < m_max_blocks_per_mcu; i++) {
+        m_mcu_block_max_zag[i] = 64;
+    }
+
+    m_expanded_blocks_per_component = m_comp_h_samp[0] * m_comp_v_samp[0];
+    m_expanded_blocks_per_mcu = m_expanded_blocks_per_component * m_comps_in_frame;
+    m_expanded_blocks_per_row = m_max_mcus_per_row * m_expanded_blocks_per_mcu;
+    // Freq. domain chroma upsampling is only supported for H2V2 subsampling factor (the most common one I've seen).
+    m_freq_domain_chroma_upsample = false;
+#if JPGD_SUPPORT_FREQ_DOMAIN_UPSAMPLING
+    m_freq_domain_chroma_upsample = (m_expanded_blocks_per_mcu == 4*3);
+#endif
+
+    if (m_freq_domain_chroma_upsample)
+        m_pSample_buf = (uint8_t *)alloc(m_expanded_blocks_per_row * 64);
+    else
+        m_pSample_buf = (uint8_t *)alloc(m_max_blocks_per_row * 64);
+
+    m_total_lines_left = m_image_y_size;
+    m_mcu_lines_left = 0;
+    create_look_ups();
+}
+
+
+// The coeff_buf series of methods originally stored the coefficients
+// into a "virtual" file which was located in EMS, XMS, or a disk file. A cache
+// was used to make this process more efficient. Now, we can store the entire
+// thing in RAM.
+jpeg_decoder::coeff_buf* jpeg_decoder::coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y)
+{
+    coeff_buf* cb = (coeff_buf*)alloc(sizeof(coeff_buf));
+    cb->block_num_x = block_num_x;
+    cb->block_num_y = block_num_y;
+    cb->block_len_x = block_len_x;
+    cb->block_len_y = block_len_y;
+    cb->block_size = (block_len_x * block_len_y) * sizeof(jpgd_block_t);
+    cb->pData = (uint8_t *)alloc(cb->block_size * block_num_x * block_num_y, true);
+    return cb;
+}
+
+
+inline jpgd_block_t *jpeg_decoder::coeff_buf_getp(coeff_buf *cb, int block_x, int block_y)
+{
+    JPGD_ASSERT((block_x < cb->block_num_x) && (block_y < cb->block_num_y));
+    return (jpgd_block_t *)(cb->pData + block_x * cb->block_size + block_y * (cb->block_size * cb->block_num_x));
+}
+
+
+// The following methods decode the various types of m_blocks encountered
+// in progressively encoded images.
+void jpeg_decoder::decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y)
+{
+    int s, r;
+    jpgd_block_t *p = pD->coeff_buf_getp(pD->m_dc_coeffs[component_id], block_x, block_y);
+
+    if ((s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_dc_tab[component_id]])) != 0) {
+        r = pD->get_bits_no_markers(s);
+        s = JPGD_HUFF_EXTEND(r, s);
+    }
+    pD->m_last_dc_val[component_id] = (s += pD->m_last_dc_val[component_id]);
+    p[0] = static_cast<jpgd_block_t>(static_cast<unsigned int>(s) << pD->m_successive_low);
+}
+
+
+void jpeg_decoder::decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y)
+{
+    if (pD->get_bits_no_markers(1)) {
+        jpgd_block_t *p = pD->coeff_buf_getp(pD->m_dc_coeffs[component_id], block_x, block_y);
+        p[0] |= (1 << pD->m_successive_low);
+    }
+}
+
+
+void jpeg_decoder::decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y)
+{
+    int k, s, r;
+
+    if (pD->m_eob_run) {
+        pD->m_eob_run--;
+        return;
+    }
+    jpgd_block_t *p = pD->coeff_buf_getp(pD->m_ac_coeffs[component_id], block_x, block_y);
+
+    for (k = pD->m_spectral_start; k <= pD->m_spectral_end; k++) {
+        s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_ac_tab[component_id]]);
+        r = s >> 4;
+        s &= 15;
+        if (s) {
+            if ((k += r) > 63) pD->stop_decoding(JPGD_DECODE_ERROR);
+            r = pD->get_bits_no_markers(s);
+            s = JPGD_HUFF_EXTEND(r, s);
+            p[g_ZAG[k]] = static_cast<jpgd_block_t>(static_cast<unsigned int>(s) << pD->m_successive_low);
+        } else {
+            if (r == 15) {
+                if ((k += 15) > 63) pD->stop_decoding(JPGD_DECODE_ERROR);
+            } else {
+                pD->m_eob_run = 1 << r;
+                if (r) pD->m_eob_run += pD->get_bits_no_markers(r);
+                pD->m_eob_run--;
+                break;
+            }
+        }
+    }
+}
+
+
+void jpeg_decoder::decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y)
+{
+    int s, k, r;
+    int p1 = 1 << pD->m_successive_low;
+    int m1 = static_cast<unsigned int>(-1) << pD->m_successive_low;
+    jpgd_block_t *p = pD->coeff_buf_getp(pD->m_ac_coeffs[component_id], block_x, block_y);
+
+    JPGD_ASSERT(pD->m_spectral_end <= 63);
+
+    k = pD->m_spectral_start;
+
+    if (pD->m_eob_run == 0) {
+        for ( ; k <= pD->m_spectral_end; k++) {
+            s = pD->huff_decode(pD->m_pHuff_tabs[pD->m_comp_ac_tab[component_id]]);
+            r = s >> 4;
+            s &= 15;
+            if (s) {
+                if (s != 1) pD->stop_decoding(JPGD_DECODE_ERROR);
+                if (pD->get_bits_no_markers(1)) s = p1;
+                else s = m1;
+            } else {
+                if (r != 15) {
+                    pD->m_eob_run = 1 << r;
+                    if (r) pD->m_eob_run += pD->get_bits_no_markers(r);
+                    break;
+                }
+            }
+
+            do {
+                jpgd_block_t *this_coef = p + g_ZAG[k & 63];
+
+                if (*this_coef != 0) {
+                    if (pD->get_bits_no_markers(1)) {
+                        if ((*this_coef & p1) == 0) {
+                            if (*this_coef >= 0) *this_coef = static_cast<jpgd_block_t>(*this_coef + p1);
+                            else *this_coef = static_cast<jpgd_block_t>(*this_coef + m1);
+                        }
+                    }
+                } else {
+                    if (--r < 0) break;
+                }
+                k++;
+            } while (k <= pD->m_spectral_end);
+
+            if ((s) && (k < 64)) {
+              p[g_ZAG[k]] = static_cast<jpgd_block_t>(s);
+            }
+        }
+    }
+
+    if (pD->m_eob_run > 0) {
+        for ( ; k <= pD->m_spectral_end; k++) {
+            jpgd_block_t *this_coef = p + g_ZAG[k & 63]; // logical AND to shut up static code analysis
+
+            if (*this_coef != 0) {
+                if (pD->get_bits_no_markers(1)) {
+                    if ((*this_coef & p1) == 0) {
+                        if (*this_coef >= 0) *this_coef = static_cast<jpgd_block_t>(*this_coef + p1);
+                        else *this_coef = static_cast<jpgd_block_t>(*this_coef + m1);
+                    }
+                }
+            }
+        }
+        pD->m_eob_run--;
+    }
+}
+
+
+// Decode a scan in a progressively encoded image.
+void jpeg_decoder::decode_scan(pDecode_block_func decode_block_func)
+{
+    int mcu_row, mcu_col, mcu_block;
+    int block_x_mcu[JPGD_MAX_COMPONENTS], m_block_y_mcu[JPGD_MAX_COMPONENTS];
+
+    memset(m_block_y_mcu, 0, sizeof(m_block_y_mcu));
+
+    for (mcu_col = 0; mcu_col < m_mcus_per_col; mcu_col++) {
+        int component_num, component_id;
+        memset(block_x_mcu, 0, sizeof(block_x_mcu));
+
+        for (mcu_row = 0; mcu_row < m_mcus_per_row; mcu_row++) {
+            int block_x_mcu_ofs = 0, block_y_mcu_ofs = 0;
+
+            if ((m_restart_interval) && (m_restarts_left == 0)) process_restart();
+
+            for (mcu_block = 0; mcu_block < m_blocks_per_mcu; mcu_block++) {
+                component_id = m_mcu_org[mcu_block];
+                decode_block_func(this, component_id, block_x_mcu[component_id] + block_x_mcu_ofs, m_block_y_mcu[component_id] + block_y_mcu_ofs);
+
+                if (m_comps_in_scan == 1) block_x_mcu[component_id]++;
+                else {
+                    if (++block_x_mcu_ofs == m_comp_h_samp[component_id]) {
+                        block_x_mcu_ofs = 0;
+
+                        if (++block_y_mcu_ofs == m_comp_v_samp[component_id]) {
+                            block_y_mcu_ofs = 0;
+                            block_x_mcu[component_id] += m_comp_h_samp[component_id];
+                        }
+                    }
+                }
+            }
+            m_restarts_left--;
+        }
+
+        if (m_comps_in_scan == 1) m_block_y_mcu[m_comp_list[0]]++;
+        else {
+            for (component_num = 0; component_num < m_comps_in_scan; component_num++) {
+                component_id = m_comp_list[component_num];
+                m_block_y_mcu[component_id] += m_comp_v_samp[component_id];
+            }
+        }
+    }
+}
+
+
+// Decode a progressively encoded image.
+void jpeg_decoder::init_progressive()
+{
+    int i;
+
+    if (m_comps_in_frame == 4) stop_decoding(JPGD_UNSUPPORTED_COLORSPACE);
+
+    // Allocate the coefficient buffers.
+    for (i = 0; i < m_comps_in_frame; i++) {
+        m_dc_coeffs[i] = coeff_buf_open(m_max_mcus_per_row * m_comp_h_samp[i], m_max_mcus_per_col * m_comp_v_samp[i], 1, 1);
+        m_ac_coeffs[i] = coeff_buf_open(m_max_mcus_per_row * m_comp_h_samp[i], m_max_mcus_per_col * m_comp_v_samp[i], 8, 8);
+    }
+
+    while (true) {
+        int dc_only_scan, refinement_scan;
+        pDecode_block_func decode_block_func;
+
+        if (!init_scan()) break;
+
+        dc_only_scan = (m_spectral_start == 0);
+        refinement_scan = (m_successive_high != 0);
+
+        if ((m_spectral_start > m_spectral_end) || (m_spectral_end > 63)) stop_decoding(JPGD_BAD_SOS_SPECTRAL);
+
+        if (dc_only_scan) {
+            if (m_spectral_end) stop_decoding(JPGD_BAD_SOS_SPECTRAL);
+        } else if (m_comps_in_scan != 1) {  /* AC scans can only contain one component */
+            stop_decoding(JPGD_BAD_SOS_SPECTRAL);
+        }
+
+        if ((refinement_scan) && (m_successive_low != m_successive_high - 1)) stop_decoding(JPGD_BAD_SOS_SUCCESSIVE);
+
+        if (dc_only_scan) {
+            if (refinement_scan) decode_block_func = decode_block_dc_refine;
+            else decode_block_func = decode_block_dc_first;
+        } else {
+            if (refinement_scan) decode_block_func = decode_block_ac_refine;
+            else decode_block_func = decode_block_ac_first;
+        }
+        decode_scan(decode_block_func);
+        m_bits_left = 16;
+        get_bits(16);
+        get_bits(16);
+    }
+
+    m_comps_in_scan = m_comps_in_frame;
+
+    for (i = 0; i < m_comps_in_frame; i++) {
+        m_comp_list[i] = i;
+    }
+
+    calc_mcu_block_order();
+}
+
+
+void jpeg_decoder::init_sequential()
+{
+    if (!init_scan()) stop_decoding(JPGD_UNEXPECTED_MARKER);
+}
+
+
+void jpeg_decoder::decode_start()
+{
+    init_frame();
+    if (m_progressive_flag) init_progressive();
+    else init_sequential();
+}
+
+
+void jpeg_decoder::decode_init(jpeg_decoder_stream *pStream)
+{
+    init(pStream);
+    locate_sof_marker();
+}
+
+
+jpeg_decoder::jpeg_decoder(jpeg_decoder_stream *pStream)
+{
+    if (setjmp(m_jmp_state)) return;
+    decode_init(pStream);
+}
+
+
+int jpeg_decoder::begin_decoding()
+{
+    if (m_ready_flag) return JPGD_SUCCESS;
+    if (m_error_code) return JPGD_FAILED;
+    if (setjmp(m_jmp_state)) return JPGD_FAILED;
+
+    decode_start();
+    m_ready_flag = true;
+
+    return JPGD_SUCCESS;
+}
+
+
+jpeg_decoder::~jpeg_decoder()
+{
+    free_all_blocks();
+}
+
+
+void jpeg_decoder_file_stream::close()
+{
+    if (m_pFile) {
+        fclose(m_pFile);
+        m_pFile = nullptr;
+    }
+    m_eof_flag = false;
+    m_error_flag = false;
+}
+
+
+jpeg_decoder_file_stream::~jpeg_decoder_file_stream()
+{
+    close();
+}
+
+
+bool jpeg_decoder_file_stream::open(const char *Pfilename)
+{
+    close();
+
+    m_eof_flag = false;
+    m_error_flag = false;
+
+#if defined(_MSC_VER)
+    m_pFile = nullptr;
+    fopen_s(&m_pFile, Pfilename, "rb");
+#else
+    m_pFile = fopen(Pfilename, "rb");
+#endif
+    return m_pFile != nullptr;
+}
+
+
+int jpeg_decoder_file_stream::read(uint8_t *pBuf, int max_bytes_to_read, bool *pEOF_flag)
+{
+    if (!m_pFile) return -1;
+
+    if (m_eof_flag) {
+        *pEOF_flag = true;
+        return 0;
+    }
+
+    if (m_error_flag) return -1;
+
+    int bytes_read = static_cast<int>(fread(pBuf, 1, max_bytes_to_read, m_pFile));
+    if (bytes_read < max_bytes_to_read) {
+        if (ferror(m_pFile)) {
+            m_error_flag = true;
+            return -1;
+        }
+        m_eof_flag = true;
+        *pEOF_flag = true;
+    }
+    return bytes_read;
+}
+
+
+bool jpeg_decoder_mem_stream::open(const uint8_t *pSrc_data, uint32_t size)
+{
+    close();
+    m_pSrc_data = pSrc_data;
+    m_ofs = 0;
+    m_size = size;
+    return true;
+}
+
+
+int jpeg_decoder_mem_stream::read(uint8_t *pBuf, int max_bytes_to_read, bool *pEOF_flag)
+{
+    *pEOF_flag = false;
+    if (!m_pSrc_data) return -1;
+
+    uint32_t bytes_remaining = m_size - m_ofs;
+    if ((uint32_t)max_bytes_to_read > bytes_remaining) {
+        max_bytes_to_read = bytes_remaining;
+        *pEOF_flag = true;
+    }
+    memcpy(pBuf, m_pSrc_data + m_ofs, max_bytes_to_read);
+    m_ofs += max_bytes_to_read;
+
+    return max_bytes_to_read;
+}
+
+
+/************************************************************************/
+/* External Class Implementation                                        */
+/************************************************************************/
+
+
+jpeg_decoder* jpgdHeader(const char* data, int size, int* width, int* height)
+{
+    auto decoder = new jpeg_decoder(new jpeg_decoder_mem_stream((const uint8_t*)data, size));
+    if (decoder->get_error_code() != JPGD_SUCCESS) {
+        delete(decoder);
+        return nullptr;
+    }
+
+    if (width) *width = decoder->get_width();
+    if (height) *height = decoder->get_height();
+
+    return decoder;
+}
+
+
+jpeg_decoder* jpgdHeader(const char* filename, int* width, int* height)
+{
+    auto fileStream = new jpeg_decoder_file_stream();
+    if (!fileStream->open(filename)) {
+        delete(fileStream);
+        return nullptr;
+    }
+
+    auto decoder = new jpeg_decoder(fileStream);
+    if (decoder->get_error_code() != JPGD_SUCCESS) {
+        delete(fileStream);
+        delete(decoder);
+        return nullptr;
+    }
+
+    if (width) *width = decoder->get_width();
+    if (height) *height = decoder->get_height();
+
+    return decoder;
+}
+
+
+void jpgdDelete(jpeg_decoder* decoder)
+{
+    delete(decoder);
+}
+
+
+unsigned char* jpgdDecompress(jpeg_decoder* decoder)
+{
+    if (!decoder) return nullptr;
+
+    int req_comps = 4;  //TODO: fixed 4 channel components now?
+    if ((req_comps != 1) && (req_comps != 3) && (req_comps != 4)) return nullptr;
+
+    auto image_width = decoder->get_width();
+    auto image_height = decoder->get_height();
+    //auto actual_comps = decoder->get_num_components();
+
+    if (decoder->begin_decoding() != JPGD_SUCCESS) return nullptr;
+
+    const int dst_bpl = image_width * req_comps;
+    uint8_t *pImage_data = (uint8_t*)malloc(dst_bpl * image_height);
+    if (!pImage_data) return nullptr;
+
+    for (int y = 0; y < image_height; y++) {
+        const uint8_t* pScan_line;
+        uint32_t scan_line_len;
+        if (decoder->decode((const void**)&pScan_line, &scan_line_len) != JPGD_SUCCESS) {
+            free(pImage_data);
+            return nullptr;
+        }
+
+        uint8_t *pDst = pImage_data + y * dst_bpl;
+
+        //Return as BGRA
+        if ((req_comps == 4) && (decoder->get_num_components() == 3)) {
+            for (int x = 0; x < image_width; x++) {
+                pDst[0] = pScan_line[x*4+2];
+                pDst[1] = pScan_line[x*4+1];
+                pDst[2] = pScan_line[x*4+0];
+                pDst[3] = 255;
+                pDst += 4;
+            }
+        } else if (((req_comps == 1) && (decoder->get_num_components() == 1)) || ((req_comps == 4) && (decoder->get_num_components() == 3))) {
+            memcpy(pDst, pScan_line, dst_bpl);
+        } else if (decoder->get_num_components() == 1) {
+            if (req_comps == 3) {
+                for (int x = 0; x < image_width; x++) {
+                    uint8_t luma = pScan_line[x];
+                    pDst[0] = luma;
+                    pDst[1] = luma;
+                    pDst[2] = luma;
+                    pDst += 3;
+                }
+            } else {
+                for (int x = 0; x < image_width; x++) {
+                    uint8_t luma = pScan_line[x];
+                    pDst[0] = luma;
+                    pDst[1] = luma;
+                    pDst[2] = luma;
+                    pDst[3] = 255;
+                    pDst += 4;
+                }
+            }
+        } else if (decoder->get_num_components() == 3) {
+            if (req_comps == 1) {
+                const int YR = 19595, YG = 38470, YB = 7471;
+                for (int x = 0; x < image_width; x++) {
+                    int r = pScan_line[x*4+0];
+                    int g = pScan_line[x*4+1];
+                    int b = pScan_line[x*4+2];
+                    *pDst++ = static_cast<uint8_t>((r * YR + g * YG + b * YB + 32768) >> 16);
+                }
+            } else {
+                for (int x = 0; x < image_width; x++) {
+                    pDst[0] = pScan_line[x*4+0];
+                    pDst[1] = pScan_line[x*4+1];
+                    pDst[2] = pScan_line[x*4+2];
+                    pDst += 3;
+                }
+            }
+        }
+    }
+    return pImage_data;
+}

+ 35 - 0
thirdparty/thorvg/src/loaders/jpg/tvgJpgd.h

@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2021 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+// jpgd.h - C++ class for JPEG decompression.
+// Public domain, Rich Geldreich <[email protected]>
+#ifndef _TVG_JPGD_H_
+#define _TVG_JPGD_H_
+
+class jpeg_decoder;
+
+jpeg_decoder* jpgdHeader(const char* data, int size, int* width, int* height);
+jpeg_decoder* jpgdHeader(const char* filename, int* width, int* height);
+unsigned char* jpgdDecompress(jpeg_decoder* decoder);
+void jpgdDelete(jpeg_decoder* decoder);
+
+#endif //_TVG_JPGD_H_

+ 2628 - 0
thirdparty/thorvg/src/loaders/png/tvgLodePng.cpp

@@ -0,0 +1,2628 @@
+/*
+ * Copyright (c) 2020 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+/*
+  LodePNG version 20200306
+
+  Copyright (c) 2005-2020 Lode Vandevenne
+
+  This software is provided 'as-is', without any express or implied
+  warranty. In no event will the authors be held liable for any damages
+  arising from the use of this software.
+
+  Permission is granted to anyone to use this software for any purpose,
+  including commercial applications, and to alter it and redistribute it
+  freely, subject to the following restrictions:
+
+    1. The origin of this software must not be misrepresented; you must not
+    claim that you wrote the original software. If you use this software
+    in a product, an acknowledgment in the product documentation would be
+    appreciated but is not required.
+
+    2. Altered source versions must be plainly marked as such, and must not be
+    misrepresented as being the original software.
+
+    3. This notice may not be removed or altered from any sourcedistribution.
+*/
+
+#include <cstdlib>
+#include "tvgLodePng.h"
+
+
+/************************************************************************/
+/* Internal Class Implementation                                        */
+/************************************************************************/
+
+#if defined(_MSC_VER) && (_MSC_VER >= 1310) /*Visual Studio: A few warning types are not desired here.*/
+    #pragma warning( disable : 4244 ) /*implicit conversions: not warned by gcc -Wall -Wextra and requires too much casts*/
+    #pragma warning( disable : 4996 ) /*VS does not like fopen, but fopen_s is not standard C so unusable here*/
+#endif /*_MSC_VER */
+
+
+/* convince the compiler to inline a function, for use when this measurably improves performance */
+/* inline is not available in C90, but use it when supported by the compiler */
+#if (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) || (defined(__cplusplus) && (__cplusplus >= 199711L))
+    #define LODEPNG_INLINE inline
+#else
+    #define LODEPNG_INLINE /* not available */
+#endif
+
+/* restrict is not available in C90, but use it when supported by the compiler */
+#if (defined(__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) ||\
+    (defined(_MSC_VER) && (_MSC_VER >= 1400)) || \
+    (defined(__WATCOMC__) && (__WATCOMC__ >= 1250) && !defined(__cplusplus))
+    #define LODEPNG_RESTRICT __restrict
+#else
+    #define LODEPNG_RESTRICT /* not available */
+#endif
+
+#define LODEPNG_MAX(a, b) (((a) > (b)) ? (a) : (b))
+#define LODEPNG_MIN(a, b) (((a) < (b)) ? (a) : (b))
+#define LODEPNG_ABS(x) ((x) < 0 ? -(x) : (x))
+
+
+/* Replacements for C library functions such as memcpy and strlen, to support platforms
+where a full C library is not available. The compiler can recognize them and compile
+to something as fast. */
+
+static void lodepng_memcpy(void* LODEPNG_RESTRICT dst, const void* LODEPNG_RESTRICT src, size_t size)
+{
+    size_t i;
+    for (i = 0; i < size; i++) ((char*)dst)[i] = ((const char*)src)[i];
+}
+
+
+static void lodepng_memset(void* LODEPNG_RESTRICT dst, int value, size_t num)
+{
+    size_t i;
+    for (i = 0; i < num; i++) ((char*)dst)[i] = (char)value;
+}
+
+
+/* does not check memory out of bounds, do not use on untrusted data */
+static size_t lodepng_strlen(const char* a)
+{
+    const char* orig = a;
+    /* avoid warning about unused function in case of disabled COMPILE... macros */
+    (void)(&lodepng_strlen);
+    while (*a) a++;
+    return (size_t)(a - orig);
+}
+
+
+/* Safely check if adding two integers will overflow (no undefined
+behavior, compiler removing the code, etc...) and output result. */
+static int lodepng_addofl(size_t a, size_t b, size_t* result)
+{
+    *result = a + b; /* Unsigned addition is well defined and safe in C90 */
+    return *result < a;
+}
+
+
+/* Safely check if multiplying two integers will overflow (no undefined
+behavior, compiler removing the code, etc...) and output result. */
+static int lodepng_mulofl(size_t a, size_t b, size_t* result)
+{
+    *result = a * b; /* Unsigned multiplication is well defined and safe in C90 */
+    return (a != 0 && *result / a != b);
+}
+
+
+/* Safely check if a + b > c, even if overflow could happen. */
+static int lodepng_gtofl(size_t a, size_t b, size_t c)
+{
+    size_t d;
+    if (lodepng_addofl(a, b, &d)) return 1;
+    return d > c;
+}
+
+
+/*
+    Often in case of an error a value is assigned to a variable and then it breaks
+    out of a loop (to go to the cleanup phase of a function). This macro does that.
+    It makes the error handling code shorter and more readable.
+
+    Example: if(!uivector_resize(&lz77_encoded, datasize)) ERROR_BREAK(83);
+*/
+#define CERROR_BREAK(errorvar, code){\
+  errorvar = code;\
+  break;\
+}
+
+/* version of CERROR_BREAK that assumes the common case where the error variable is named "error" */
+#define ERROR_BREAK(code) CERROR_BREAK(error, code)
+
+/* Set error var to the error code, and return it.*/
+#define CERROR_RETURN_ERROR(errorvar, code){\
+  errorvar = code;\
+  return code;\
+}
+
+/* Try the code, if it returns error, also return the error. */
+#define CERROR_TRY_RETURN(call){\
+  unsigned error = call;\
+  if(error) return error;\
+}
+
+/* Set error var to the error code, and return from the void function. */
+#define CERROR_RETURN(errorvar, code){\
+  errorvar = code;\
+  return;\
+}
+
+
+/* dynamic vector of unsigned chars */
+struct ucvector
+{
+    unsigned char* data;
+    size_t size; /*used size*/
+    size_t allocsize; /*allocated size*/
+};
+
+
+/* returns 1 if success, 0 if failure ==> nothing done */
+static unsigned ucvector_resize(ucvector* p, size_t size)
+{
+    if (size > p->allocsize) {
+        size_t newsize = size + (p->allocsize >> 1u);
+        void* data = realloc(p->data, newsize);
+        if(data) {
+            p->allocsize = newsize;
+            p->data = (unsigned char*)data;
+        }
+        else return 0; /*error: not enough memory*/
+    }
+    p->size = size;
+    return 1; /*success*/
+}
+
+
+static ucvector ucvector_init(unsigned char* buffer, size_t size)
+{
+    ucvector v;
+    v.data = buffer;
+    v.allocsize = v.size = size;
+    return v;
+}
+
+
+static unsigned lodepng_read32bitInt(const unsigned char* buffer)
+{
+    return (((unsigned)buffer[0] << 24u) | ((unsigned)buffer[1] << 16u) | ((unsigned)buffer[2] << 8u) | (unsigned)buffer[3]);
+}
+
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // End of common code and tools. Begin of Zlib related code.            // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+struct LodePNGBitReader
+{
+    const unsigned char* data;
+    size_t size; /*size of data in bytes*/
+    size_t bitsize; /*size of data in bits, end of valid bp values, should be 8*size*/
+    size_t bp;
+    unsigned buffer; /*buffer for reading bits. NOTE: 'unsigned' must support at least 32 bits*/
+};
+
+
+/* data size argument is in bytes. Returns error if size too large causing overflow */
+static unsigned LodePNGBitReader_init(LodePNGBitReader* reader, const unsigned char* data, size_t size)
+{
+    size_t temp;
+    reader->data = data;
+    reader->size = size;
+    /* size in bits, return error if overflow (if size_t is 32 bit this supports up to 500MB)  */
+    if (lodepng_mulofl(size, 8u, &reader->bitsize)) return 105;
+    /*ensure incremented bp can be compared to bitsize without overflow even when it would be incremented 32 too much and
+    trying to ensure 32 more bits*/
+    if (lodepng_addofl(reader->bitsize, 64u, &temp)) return 105;
+    reader->bp = 0;
+    reader->buffer = 0;
+    return 0; /*ok*/
+  }
+
+/*
+  ensureBits functions:
+  Ensures the reader can at least read nbits bits in one or more readBits calls,
+  safely even if not enough bits are available.
+  Returns 1 if there are enough bits available, 0 if not.
+*/
+
+/*See ensureBits documentation above. This one ensures exactly 1 bit */
+/*static unsigned ensureBits1(LodePNGBitReader* reader) {
+  if(reader->bp >= reader->bitsize) return 0;
+  reader->buffer = (unsigned)reader->data[reader->bp >> 3u] >> (reader->bp & 7u);
+  return 1;
+}*/
+
+/*See ensureBits documentation above. This one ensures up to 9 bits */
+static unsigned ensureBits9(LodePNGBitReader* reader, size_t nbits)
+{
+    size_t start = reader->bp >> 3u;
+    size_t size = reader->size;
+    if (start + 1u < size) {
+        reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u);
+        reader->buffer >>= (reader->bp & 7u);
+        return 1;
+    } else {
+        reader->buffer = 0;
+        if (start + 0u < size) reader->buffer |= reader->data[start + 0];
+        reader->buffer >>= (reader->bp & 7u);
+        return reader->bp + nbits <= reader->bitsize;
+    }
+}
+
+
+/*See ensureBits documentation above. This one ensures up to 17 bits */
+static unsigned ensureBits17(LodePNGBitReader* reader, size_t nbits)
+{
+    size_t start = reader->bp >> 3u;
+    size_t size = reader->size;
+    if (start + 2u < size) {
+        reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u) | ((unsigned)reader->data[start + 2] << 16u);
+        reader->buffer >>= (reader->bp & 7u);
+        return 1;
+    } else {
+        reader->buffer = 0;
+        if (start + 0u < size) reader->buffer |= reader->data[start + 0];
+        if (start + 1u < size) reader->buffer |= ((unsigned)reader->data[start + 1] << 8u);
+        reader->buffer >>= (reader->bp & 7u);
+        return reader->bp + nbits <= reader->bitsize;
+    }
+}
+
+
+/*See ensureBits documentation above. This one ensures up to 25 bits */
+static LODEPNG_INLINE unsigned ensureBits25(LodePNGBitReader* reader, size_t nbits)
+{
+    size_t start = reader->bp >> 3u;
+    size_t size = reader->size;
+    if (start + 3u < size) {
+        reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u) |  ((unsigned)reader->data[start + 2] << 16u) | ((unsigned)reader->data[start + 3] << 24u);
+        reader->buffer >>= (reader->bp & 7u);
+        return 1;
+    } else {
+        reader->buffer = 0;
+        if (start + 0u < size) reader->buffer |= reader->data[start + 0];
+        if (start + 1u < size) reader->buffer |= ((unsigned)reader->data[start + 1] << 8u);
+        if (start + 2u < size) reader->buffer |= ((unsigned)reader->data[start + 2] << 16u);
+        reader->buffer >>= (reader->bp & 7u);
+        return reader->bp + nbits <= reader->bitsize;
+    }
+}
+
+
+/*See ensureBits documentation above. This one ensures up to 32 bits */
+static LODEPNG_INLINE unsigned ensureBits32(LodePNGBitReader* reader, size_t nbits)
+{
+    size_t start = reader->bp >> 3u;
+    size_t size = reader->size;
+    if(start + 4u < size) {
+        reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u) | ((unsigned)reader->data[start + 2] << 16u) | ((unsigned)reader->data[start + 3] << 24u);
+        reader->buffer >>= (reader->bp & 7u);
+        reader->buffer |= (((unsigned)reader->data[start + 4] << 24u) << (8u - (reader->bp & 7u)));
+        return 1;
+    } else {
+        reader->buffer = 0;
+        if (start + 0u < size) reader->buffer |= reader->data[start + 0];
+        if (start + 1u < size) reader->buffer |= ((unsigned)reader->data[start + 1] << 8u);
+        if (start + 2u < size) reader->buffer |= ((unsigned)reader->data[start + 2] << 16u);
+        if (start + 3u < size) reader->buffer |= ((unsigned)reader->data[start + 3] << 24u);
+        reader->buffer >>= (reader->bp & 7u);
+        return reader->bp + nbits <= reader->bitsize;
+    }
+}
+
+
+/* Get bits without advancing the bit pointer. Must have enough bits available with ensureBits. Max nbits is 31. */
+static unsigned peekBits(LodePNGBitReader* reader, size_t nbits)
+{
+    /* The shift allows nbits to be only up to 31. */
+    return reader->buffer & ((1u << nbits) - 1u);
+}
+
+
+/* Must have enough bits available with ensureBits */
+static void advanceBits(LodePNGBitReader* reader, size_t nbits)
+{
+    reader->buffer >>= nbits;
+    reader->bp += nbits;
+}
+
+
+/* Must have enough bits available with ensureBits */
+static unsigned readBits(LodePNGBitReader* reader, size_t nbits)
+{
+    unsigned result = peekBits(reader, nbits);
+    advanceBits(reader, nbits);
+    return result;
+}
+
+
+/* Public for testing only. steps and result must have numsteps values. */
+unsigned lode_png_test_bitreader(const unsigned char* data, size_t size, size_t numsteps, const size_t* steps, unsigned* result)
+{
+    size_t i;
+    LodePNGBitReader reader;
+    unsigned error = LodePNGBitReader_init(&reader, data, size);
+    if (error) return 0;
+    for (i = 0; i < numsteps; i++) {
+        size_t step = steps[i];
+        unsigned ok;
+        if (step > 25) ok = ensureBits32(&reader, step);
+        else if (step > 17) ok = ensureBits25(&reader, step);
+        else if (step > 9) ok = ensureBits17(&reader, step);
+        else ok = ensureBits9(&reader, step);
+        if (!ok) return 0;
+        result[i] = readBits(&reader, step);
+    }
+    return 1;
+}
+
+
+static unsigned reverseBits(unsigned bits, unsigned num)
+{
+    /*TODO: implement faster lookup table based version when needed*/
+    unsigned i, result = 0;
+    for (i = 0; i < num; i++) result |= ((bits >> (num - i - 1u)) & 1u) << i;
+    return result;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Deflate - Huffman                                                      / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#define FIRST_LENGTH_CODE_INDEX 257
+#define LAST_LENGTH_CODE_INDEX 285
+/*256 literals, the end code, some length codes, and 2 unused codes*/
+#define NUM_DEFLATE_CODE_SYMBOLS 288
+/*the distance codes have their own symbols, 30 used, 2 unused*/
+#define NUM_DISTANCE_SYMBOLS 32
+/*the code length codes. 0-15: code lengths, 16: copy previous 3-6 times, 17: 3-10 zeros, 18: 11-138 zeros*/
+#define NUM_CODE_LENGTH_CODES 19
+
+/*the base lengths represented by codes 257-285*/
+static const unsigned LENGTHBASE[29]
+  = {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};
+
+/*the extra bits used by codes 257-285 (added to base length)*/
+static const unsigned LENGTHEXTRA[29]
+  = {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};
+
+/*the base backwards distances (the bits of distance codes appear after length codes and use their own huffman tree)*/
+static const unsigned DISTANCEBASE[30]
+  = {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};
+
+/*the extra bits of backwards distances (added to base)*/
+static const unsigned DISTANCEEXTRA[30]
+  = {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};
+
+/*the order in which "code length alphabet code lengths" are stored as specified by deflate, out of this the huffman
+tree of the dynamic huffman tree lengths is generated*/
+static const unsigned CLCL_ORDER[NUM_CODE_LENGTH_CODES]
+  = {16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15};
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*
+Huffman tree struct, containing multiple representations of the tree
+*/
+struct HuffmanTree
+{
+    unsigned* codes; /*the huffman codes (bit patterns representing the symbols)*/
+    unsigned* lengths; /*the lengths of the huffman codes*/
+    unsigned maxbitlen; /*maximum number of bits a single code can get*/
+    unsigned numcodes; /*number of symbols in the alphabet = number of codes*/
+    /* for reading only */
+    unsigned char* table_len; /*length of symbol from lookup table, or max length if secondary lookup needed*/
+    unsigned short* table_value; /*value of symbol from lookup table, or pointer to secondary table if needed*/
+};
+
+
+static void HuffmanTree_init(HuffmanTree* tree)
+{
+    tree->codes = 0;
+    tree->lengths = 0;
+    tree->table_len = 0;
+    tree->table_value = 0;
+}
+
+
+static void HuffmanTree_cleanup(HuffmanTree* tree)
+{
+    free(tree->codes);
+    free(tree->lengths);
+    free(tree->table_len);
+    free(tree->table_value);
+}
+
+
+/* amount of bits for first huffman table lookup (aka root bits), see HuffmanTree_makeTable and huffmanDecodeSymbol.*/
+/* values 8u and 9u work the fastest */
+#define FIRSTBITS 9u
+
+/* a symbol value too big to represent any valid symbol, to indicate reading disallowed huffman bits combination,
+which is possible in case of only 0 or 1 present symbols. */
+#define INVALIDSYMBOL 65535u
+
+/* make table for huffman decoding */
+static unsigned HuffmanTree_makeTable(HuffmanTree* tree)
+{
+    static const unsigned headsize = 1u << FIRSTBITS; /*size of the first table*/
+    static const unsigned mask = (1u << FIRSTBITS) /*headsize*/ - 1u;
+    size_t i, numpresent, pointer, size; /*total table size*/
+    unsigned* maxlens = (unsigned*)malloc(headsize * sizeof(unsigned));
+    if (!maxlens) return 83; /*alloc fail*/
+
+    /* compute maxlens: max total bit length of symbols sharing prefix in the first table*/
+    lodepng_memset(maxlens, 0, headsize * sizeof(*maxlens));
+    for (i = 0; i < tree->numcodes; i++) {
+        unsigned symbol = tree->codes[i];
+        unsigned l = tree->lengths[i];
+        unsigned index;
+        if(l <= FIRSTBITS) continue; /*symbols that fit in first table don't increase secondary table size*/
+        /*get the FIRSTBITS MSBs, the MSBs of the symbol are encoded first. See later comment about the reversing*/
+        index = reverseBits(symbol >> (l - FIRSTBITS), FIRSTBITS);
+        maxlens[index] = LODEPNG_MAX(maxlens[index], l);
+    }
+    /* compute total table size: size of first table plus all secondary tables for symbols longer than FIRSTBITS */
+    size = headsize;
+    for (i = 0; i < headsize; ++i) {
+        unsigned l = maxlens[i];
+        if (l > FIRSTBITS) size += (1u << (l - FIRSTBITS));
+    }
+    tree->table_len = (unsigned char*)malloc(size * sizeof(*tree->table_len));
+    tree->table_value = (unsigned short*)malloc(size * sizeof(*tree->table_value));
+    if (!tree->table_len || !tree->table_value) {
+        free(maxlens);
+        /* freeing tree->table values is done at a higher scope */
+        return 83; /*alloc fail*/
+    }
+    /*initialize with an invalid length to indicate unused entries*/
+    for (i = 0; i < size; ++i) tree->table_len[i] = 16;
+
+    /*fill in the first table for long symbols: max prefix size and pointer to secondary tables*/
+    pointer = headsize;
+    for (i = 0; i < headsize; ++i) {
+        unsigned l = maxlens[i];
+        if(l <= FIRSTBITS) continue;
+        tree->table_len[i] = l;
+        tree->table_value[i] = pointer;
+        pointer += (1u << (l - FIRSTBITS));
+    }
+    free(maxlens);
+
+    /*fill in the first table for short symbols, or secondary table for long symbols*/
+    numpresent = 0;
+    for (i = 0; i < tree->numcodes; ++i) {
+        unsigned l = tree->lengths[i];
+        unsigned symbol = tree->codes[i]; /*the huffman bit pattern. i itself is the value.*/
+        /*reverse bits, because the huffman bits are given in MSB first order but the bit reader reads LSB first*/
+        unsigned reverse = reverseBits(symbol, l);
+        if (l == 0) continue;
+        numpresent++;
+
+        if (l <= FIRSTBITS) {
+            /*short symbol, fully in first table, replicated num times if l < FIRSTBITS*/
+            unsigned num = 1u << (FIRSTBITS - l);
+            unsigned j;
+            for (j = 0; j < num; ++j) {
+                /*bit reader will read the l bits of symbol first, the remaining FIRSTBITS - l bits go to the MSB's*/
+                unsigned index = reverse | (j << l);
+                if(tree->table_len[index] != 16) return 55; /*invalid tree: long symbol shares prefix with short symbol*/
+                tree->table_len[index] = l;
+                tree->table_value[index] = i;
+            }
+        } else {
+            /*long symbol, shares prefix with other long symbols in first lookup table, needs second lookup*/
+            /*the FIRSTBITS MSBs of the symbol are the first table index*/
+            unsigned index = reverse & mask;
+            unsigned maxlen = tree->table_len[index];
+            /*log2 of secondary table length, should be >= l - FIRSTBITS*/
+            unsigned tablelen = maxlen - FIRSTBITS;
+            unsigned start = tree->table_value[index]; /*starting index in secondary table*/
+            unsigned num = 1u << (tablelen - (l - FIRSTBITS)); /*amount of entries of this symbol in secondary table*/
+            unsigned j;
+            if (maxlen < l) return 55; /*invalid tree: long symbol shares prefix with short symbol*/
+            for (j = 0; j < num; ++j) {
+                unsigned reverse2 = reverse >> FIRSTBITS; /* l - FIRSTBITS bits */
+                unsigned index2 = start + (reverse2 | (j << (l - FIRSTBITS)));
+                tree->table_len[index2] = l;
+                tree->table_value[index2] = i;
+            }
+        }
+    }
+
+    if (numpresent < 2) {
+        /* In case of exactly 1 symbol, in theory the huffman symbol needs 0 bits,
+        but deflate uses 1 bit instead. In case of 0 symbols, no symbols can
+        appear at all, but such huffman tree could still exist (e.g. if distance
+        codes are never used). In both cases, not all symbols of the table will be
+        filled in. Fill them in with an invalid symbol value so returning them from
+        huffmanDecodeSymbol will cause error. */
+        for (i = 0; i < size; ++i) {
+            if (tree->table_len[i] == 16) {
+                /* As length, use a value smaller than FIRSTBITS for the head table,
+                and a value larger than FIRSTBITS for the secondary table, to ensure
+                valid behavior for advanceBits when reading this symbol. */
+                tree->table_len[i] = (i < headsize) ? 1 : (FIRSTBITS + 1);
+                tree->table_value[i] = INVALIDSYMBOL;
+            }
+        }
+    } else {
+        /* A good huffman tree has N * 2 - 1 nodes, of which N - 1 are internal nodes.
+        If that is not the case (due to too long length codes), the table will not
+        have been fully used, and this is an error (not all bit combinations can be
+        decoded): an oversubscribed huffman tree, indicated by error 55. */
+        for (i = 0; i < size; ++i) {
+            if (tree->table_len[i] == 16) return 55;
+        }
+    }
+    return 0;
+}
+
+
+/*
+  Second step for the ...makeFromLengths and ...makeFromFrequencies functions.
+  numcodes, lengths and maxbitlen must already be filled in correctly. return
+  value is error.
+*/
+static unsigned HuffmanTree_makeFromLengths2(HuffmanTree* tree)
+{
+    unsigned* blcount;
+    unsigned* nextcode;
+    unsigned error = 0;
+    unsigned bits, n;
+
+    tree->codes = (unsigned*)malloc(tree->numcodes * sizeof(unsigned));
+    blcount = (unsigned*)malloc((tree->maxbitlen + 1) * sizeof(unsigned));
+    nextcode = (unsigned*)malloc((tree->maxbitlen + 1) * sizeof(unsigned));
+    if (!tree->codes || !blcount || !nextcode) error = 83; /*alloc fail*/
+
+    if (!error) {
+        for (n = 0; n != tree->maxbitlen + 1; n++) blcount[n] = nextcode[n] = 0;
+        /*step 1: count number of instances of each code length*/
+        for (bits = 0; bits != tree->numcodes; ++bits) ++blcount[tree->lengths[bits]];
+        /*step 2: generate the nextcode values*/
+        for(bits = 1; bits <= tree->maxbitlen; ++bits) {
+            nextcode[bits] = (nextcode[bits - 1] + blcount[bits - 1]) << 1u;
+        }
+        /*step 3: generate all the codes*/
+        for (n = 0; n != tree->numcodes; ++n) {
+            if (tree->lengths[n] != 0) {
+                tree->codes[n] = nextcode[tree->lengths[n]]++;
+                /*remove superfluous bits from the code*/
+                tree->codes[n] &= ((1u << tree->lengths[n]) - 1u);
+            }
+        }
+    }
+
+    free(blcount);
+    free(nextcode);
+
+    if (!error) error = HuffmanTree_makeTable(tree);
+    return error;
+}
+
+
+/*
+  given the code lengths (as stored in the PNG file), generate the tree as defined
+  by Deflate. maxbitlen is the maximum bits that a code in the tree can have.
+  return value is error.
+*/
+static unsigned HuffmanTree_makeFromLengths(HuffmanTree* tree, const unsigned* bitlen, size_t numcodes, unsigned maxbitlen)
+{
+    unsigned i;
+    tree->lengths = (unsigned*)malloc(numcodes * sizeof(unsigned));
+    if (!tree->lengths) return 83; /*alloc fail*/
+    for (i = 0; i != numcodes; ++i) tree->lengths[i] = bitlen[i];
+    tree->numcodes = (unsigned)numcodes; /*number of symbols*/
+    tree->maxbitlen = maxbitlen;
+    return HuffmanTree_makeFromLengths2(tree);
+}
+
+
+/*get the literal and length code tree of a deflated block with fixed tree, as per the deflate specification*/
+static unsigned generateFixedLitLenTree(HuffmanTree* tree)
+{
+    unsigned i, error = 0;
+    unsigned* bitlen = (unsigned*)malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(unsigned));
+    if (!bitlen) return 83; /*alloc fail*/
+
+    /*288 possible codes: 0-255=literals, 256=endcode, 257-285=lengthcodes, 286-287=unused*/
+    for (i =   0; i <= 143; ++i) bitlen[i] = 8;
+    for (i = 144; i <= 255; ++i) bitlen[i] = 9;
+    for (i = 256; i <= 279; ++i) bitlen[i] = 7;
+    for (i = 280; i <= 287; ++i) bitlen[i] = 8;
+
+    error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DEFLATE_CODE_SYMBOLS, 15);
+
+    free(bitlen);
+    return error;
+}
+
+
+/*get the distance code tree of a deflated block with fixed tree, as specified in the deflate specification*/
+static unsigned generateFixedDistanceTree(HuffmanTree* tree)
+{
+    unsigned i, error = 0;
+    unsigned* bitlen = (unsigned*)malloc(NUM_DISTANCE_SYMBOLS * sizeof(unsigned));
+    if (!bitlen) return 83; /*alloc fail*/
+
+    /*there are 32 distance codes, but 30-31 are unused*/
+    for (i = 0; i != NUM_DISTANCE_SYMBOLS; ++i) bitlen[i] = 5;
+    error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DISTANCE_SYMBOLS, 15);
+
+    free(bitlen);
+    return error;
+}
+
+
+/*
+  returns the code. The bit reader must already have been ensured at least 15 bits
+*/
+static unsigned huffmanDecodeSymbol(LodePNGBitReader* reader, const HuffmanTree* codetree)
+{
+    unsigned short code = peekBits(reader, FIRSTBITS);
+    unsigned short l = codetree->table_len[code];
+    unsigned short value = codetree->table_value[code];
+    if (l <= FIRSTBITS) {
+        advanceBits(reader, l);
+        return value;
+    } else {
+        unsigned index2;
+        advanceBits(reader, FIRSTBITS);
+        index2 = value + peekBits(reader, l - FIRSTBITS);
+        advanceBits(reader, codetree->table_len[index2] - FIRSTBITS);
+        return codetree->table_value[index2];
+    }
+}
+
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Inflator (Decompressor)                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*get the tree of a deflated block with fixed tree, as specified in the deflate specification
+Returns error code.*/
+static unsigned getTreeInflateFixed(HuffmanTree* tree_ll, HuffmanTree* tree_d)
+{
+    unsigned error = generateFixedLitLenTree(tree_ll);
+    if (error) return error;
+    return generateFixedDistanceTree(tree_d);
+}
+
+
+/*get the tree of a deflated block with dynamic tree, the tree itself is also Huffman compressed with a known tree*/
+static unsigned getTreeInflateDynamic(HuffmanTree* tree_ll, HuffmanTree* tree_d, LodePNGBitReader* reader)
+{
+    /*make sure that length values that aren't filled in will be 0, or a wrong tree will be generated*/
+    unsigned error = 0;
+    unsigned n, HLIT, HDIST, HCLEN, i;
+
+    /*see comments in deflateDynamic for explanation of the context and these variables, it is analogous*/
+    unsigned* bitlen_ll = 0; /*lit,len code lengths*/
+    unsigned* bitlen_d = 0; /*dist code lengths*/
+    /*code length code lengths ("clcl"), the bit lengths of the huffman tree used to compress bitlen_ll and bitlen_d*/
+    unsigned* bitlen_cl = 0;
+    HuffmanTree tree_cl; /*the code tree for code length codes (the huffman tree for compressed huffman trees)*/
+
+    if (!ensureBits17(reader, 14)) return 49; /*error: the bit pointer is or will go past the memory*/
+
+    /*number of literal/length codes + 257. Unlike the spec, the value 257 is added to it here already*/
+    HLIT =  readBits(reader, 5) + 257;
+    /*number of distance codes. Unlike the spec, the value 1 is added to it here already*/
+    HDIST = readBits(reader, 5) + 1;
+    /*number of code length codes. Unlike the spec, the value 4 is added to it here already*/
+    HCLEN = readBits(reader, 4) + 4;
+
+    bitlen_cl = (unsigned*)malloc(NUM_CODE_LENGTH_CODES * sizeof(unsigned));
+    if(!bitlen_cl) return 83 /*alloc fail*/;
+
+    HuffmanTree_init(&tree_cl);
+
+    while (!error) {
+        /*read the code length codes out of 3 * (amount of code length codes) bits*/
+        if (lodepng_gtofl(reader->bp, HCLEN * 3, reader->bitsize)) {
+            ERROR_BREAK(50); /*error: the bit pointer is or will go past the memory*/
+        }
+        for (i = 0; i != HCLEN; ++i) {
+            ensureBits9(reader, 3); /*out of bounds already checked above */
+            bitlen_cl[CLCL_ORDER[i]] = readBits(reader, 3);
+        }
+        for (i = HCLEN; i != NUM_CODE_LENGTH_CODES; ++i) {
+            bitlen_cl[CLCL_ORDER[i]] = 0;
+        }
+
+        error = HuffmanTree_makeFromLengths(&tree_cl, bitlen_cl, NUM_CODE_LENGTH_CODES, 7);
+        if(error) break;
+
+        /*now we can use this tree to read the lengths for the tree that this function will return*/
+        bitlen_ll = (unsigned*)malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(unsigned));
+        bitlen_d = (unsigned*)malloc(NUM_DISTANCE_SYMBOLS * sizeof(unsigned));
+        if (!bitlen_ll || !bitlen_d) ERROR_BREAK(83 /*alloc fail*/);
+        lodepng_memset(bitlen_ll, 0, NUM_DEFLATE_CODE_SYMBOLS * sizeof(*bitlen_ll));
+        lodepng_memset(bitlen_d, 0, NUM_DISTANCE_SYMBOLS * sizeof(*bitlen_d));
+
+        /*i is the current symbol we're reading in the part that contains the code lengths of lit/len and dist codes*/
+        i = 0;
+        while (i < HLIT + HDIST) {
+            unsigned code;
+            ensureBits25(reader, 22); /* up to 15 bits for huffman code, up to 7 extra bits below*/
+            code = huffmanDecodeSymbol(reader, &tree_cl);
+            if (code <= 15) /*a length code*/ {
+                if (i < HLIT) bitlen_ll[i] = code;
+                else bitlen_d[i - HLIT] = code;
+                ++i;
+            } else if (code == 16) /*repeat previous*/ {
+                unsigned replength = 3; /*read in the 2 bits that indicate repeat length (3-6)*/
+                unsigned value; /*set value to the previous code*/
+
+                if (i == 0) ERROR_BREAK(54); /*can't repeat previous if i is 0*/
+
+                replength += readBits(reader, 2);
+
+                if (i < HLIT + 1) value = bitlen_ll[i - 1];
+                else value = bitlen_d[i - HLIT - 1];
+                /*repeat this value in the next lengths*/
+                for (n = 0; n < replength; ++n) {
+                    if (i >= HLIT + HDIST) ERROR_BREAK(13); /*error: i is larger than the amount of codes*/
+                    if (i < HLIT) bitlen_ll[i] = value;
+                    else bitlen_d[i - HLIT] = value;
+                    ++i;
+                }
+            } else if(code == 17) /*repeat "0" 3-10 times*/ {
+                unsigned replength = 3; /*read in the bits that indicate repeat length*/
+                replength += readBits(reader, 3);
+
+                /*repeat this value in the next lengths*/
+                for (n = 0; n < replength; ++n) {
+                    if (i >= HLIT + HDIST) ERROR_BREAK(14); /*error: i is larger than the amount of codes*/
+
+                    if (i < HLIT) bitlen_ll[i] = 0;
+                    else bitlen_d[i - HLIT] = 0;
+                    ++i;
+                }
+            } else if(code == 18) /*repeat "0" 11-138 times*/ {
+                unsigned replength = 11; /*read in the bits that indicate repeat length*/
+                replength += readBits(reader, 7);
+
+                /*repeat this value in the next lengths*/
+                for (n = 0; n < replength; ++n) {
+                    if(i >= HLIT + HDIST) ERROR_BREAK(15); /*error: i is larger than the amount of codes*/
+
+                    if(i < HLIT) bitlen_ll[i] = 0;
+                    else bitlen_d[i - HLIT] = 0;
+                    ++i;
+                }
+            } else /*if(code == INVALIDSYMBOL)*/ {
+                ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+            }
+            /*check if any of the ensureBits above went out of bounds*/
+            if (reader->bp > reader->bitsize) {
+                /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol
+                (10=no endcode, 11=wrong jump outside of tree)*/
+                /* TODO: revise error codes 10,11,50: the above comment is no longer valid */
+                ERROR_BREAK(50); /*error, bit pointer jumps past memory*/
+            }
+        }
+        if (error) break;
+
+        if (bitlen_ll[256] == 0) ERROR_BREAK(64); /*the length of the end code 256 must be larger than 0*/
+
+        /*now we've finally got HLIT and HDIST, so generate the code trees, and the function is done*/
+        error = HuffmanTree_makeFromLengths(tree_ll, bitlen_ll, NUM_DEFLATE_CODE_SYMBOLS, 15);
+        if (error) break;
+        error = HuffmanTree_makeFromLengths(tree_d, bitlen_d, NUM_DISTANCE_SYMBOLS, 15);
+
+        break; /*end of error-while*/
+    }
+
+    free(bitlen_cl);
+    free(bitlen_ll);
+    free(bitlen_d);
+    HuffmanTree_cleanup(&tree_cl);
+
+    return error;
+}
+
+
+/*inflate a block with dynamic of fixed Huffman tree. btype must be 1 or 2.*/
+static unsigned inflateHuffmanBlock(ucvector* out, LodePNGBitReader* reader, unsigned btype)
+{
+    unsigned error = 0;
+    HuffmanTree tree_ll; /*the huffman tree for literal and length codes*/
+    HuffmanTree tree_d; /*the huffman tree for distance codes*/
+
+    HuffmanTree_init(&tree_ll);
+    HuffmanTree_init(&tree_d);
+
+    if (btype == 1) error = getTreeInflateFixed(&tree_ll, &tree_d);
+    else /*if(btype == 2)*/ error = getTreeInflateDynamic(&tree_ll, &tree_d, reader);
+
+    while (!error) /*decode all symbols until end reached, breaks at end code*/ {
+        /*code_ll is literal, length or end code*/
+        unsigned code_ll;
+        ensureBits25(reader, 20); /* up to 15 for the huffman symbol, up to 5 for the length extra bits */
+        code_ll = huffmanDecodeSymbol(reader, &tree_ll);
+        if (code_ll <= 255) /*literal symbol*/ {
+            if (!ucvector_resize(out, out->size + 1)) ERROR_BREAK(83 /*alloc fail*/);
+            out->data[out->size - 1] = (unsigned char)code_ll;
+        } else if (code_ll >= FIRST_LENGTH_CODE_INDEX && code_ll <= LAST_LENGTH_CODE_INDEX) /*length code*/ {
+            unsigned code_d, distance;
+            unsigned numextrabits_l, numextrabits_d; /*extra bits for length and distance*/
+            size_t start, backward, length;
+
+            /*part 1: get length base*/
+            length = LENGTHBASE[code_ll - FIRST_LENGTH_CODE_INDEX];
+
+            /*part 2: get extra bits and add the value of that to length*/
+            numextrabits_l = LENGTHEXTRA[code_ll - FIRST_LENGTH_CODE_INDEX];
+            if (numextrabits_l != 0) {
+                /* bits already ensured above */
+                length += readBits(reader, numextrabits_l);
+            }
+
+            /*part 3: get distance code*/
+            ensureBits32(reader, 28); /* up to 15 for the huffman symbol, up to 13 for the extra bits */
+            code_d = huffmanDecodeSymbol(reader, &tree_d);
+            if (code_d > 29) {
+                if (code_d <= 31) {
+                    ERROR_BREAK(18); /*error: invalid distance code (30-31 are never used)*/
+                } else /* if(code_d == INVALIDSYMBOL) */{
+                    ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+                }
+            }
+            distance = DISTANCEBASE[code_d];
+
+            /*part 4: get extra bits from distance*/
+            numextrabits_d = DISTANCEEXTRA[code_d];
+            if (numextrabits_d != 0) {
+                /* bits already ensured above */
+                distance += readBits(reader, numextrabits_d);
+            }
+
+            /*part 5: fill in all the out[n] values based on the length and dist*/
+            start = out->size;
+            if (distance > start) ERROR_BREAK(52); /*too long backward distance*/
+            backward = start - distance;
+
+            if (!ucvector_resize(out, out->size + length)) ERROR_BREAK(83 /*alloc fail*/);
+            if (distance < length) {
+                size_t forward;
+                lodepng_memcpy(out->data + start, out->data + backward, distance);
+                start += distance;
+                for (forward = distance; forward < length; ++forward) {
+                  out->data[start++] = out->data[backward++];
+                }
+            } else {
+                lodepng_memcpy(out->data + start, out->data + backward, length);
+            }
+        } else if (code_ll == 256) {
+            break; /*end code, break the loop*/
+        } else /*if(code_ll == INVALIDSYMBOL)*/ {
+            ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+        }
+        /*check if any of the ensureBits above went out of bounds*/
+        if (reader->bp > reader->bitsize) {
+          /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol
+          (10=no endcode, 11=wrong jump outside of tree)*/
+          /* TODO: revise error codes 10,11,50: the above comment is no longer valid */
+          ERROR_BREAK(51); /*error, bit pointer jumps past memory*/
+        }
+    }
+
+    HuffmanTree_cleanup(&tree_ll);
+    HuffmanTree_cleanup(&tree_d);
+
+    return error;
+}
+
+
+static unsigned inflateNoCompression(ucvector* out, LodePNGBitReader* reader, const LodePNGDecompressSettings* settings)
+{
+    size_t bytepos;
+    size_t size = reader->size;
+    unsigned LEN, NLEN, error = 0;
+
+    /*go to first boundary of byte*/
+    bytepos = (reader->bp + 7u) >> 3u;
+
+    /*read LEN (2 bytes) and NLEN (2 bytes)*/
+    if (bytepos + 4 >= size) return 52; /*error, bit pointer will jump past memory*/
+    LEN = (unsigned)reader->data[bytepos] + ((unsigned)reader->data[bytepos + 1] << 8u); bytepos += 2;
+    NLEN = (unsigned)reader->data[bytepos] + ((unsigned)reader->data[bytepos + 1] << 8u); bytepos += 2;
+
+    /*check if 16-bit NLEN is really the one's complement of LEN*/
+    if (!settings->ignore_nlen && LEN + NLEN != 65535) {
+        return 21; /*error: NLEN is not one's complement of LEN*/
+    }
+
+    if (!ucvector_resize(out, out->size + LEN)) return 83; /*alloc fail*/
+
+    /*read the literal data: LEN bytes are now stored in the out buffer*/
+    if (bytepos + LEN > size) return 23; /*error: reading outside of in buffer*/
+
+    lodepng_memcpy(out->data + out->size - LEN, reader->data + bytepos, LEN);
+    bytepos += LEN;
+
+    reader->bp = bytepos << 3u;
+
+    return error;
+}
+
+
+static unsigned lodepng_inflatev(ucvector* out, const unsigned char* in, size_t insize, const LodePNGDecompressSettings* settings)
+{
+    unsigned BFINAL = 0;
+    LodePNGBitReader reader;
+    unsigned error = LodePNGBitReader_init(&reader, in, insize);
+
+    if (error) return error;
+
+    while (!BFINAL) {
+        unsigned BTYPE;
+        if (!ensureBits9(&reader, 3)) return 52; /*error, bit pointer will jump past memory*/
+        BFINAL = readBits(&reader, 1);
+        BTYPE = readBits(&reader, 2);
+
+        if (BTYPE == 3) return 20; /*error: invalid BTYPE*/
+        else if (BTYPE == 0) error = inflateNoCompression(out, &reader, settings); /*no compression*/
+        else error = inflateHuffmanBlock(out, &reader, BTYPE); /*compression, BTYPE 01 or 10*/
+
+        if (error) return error;
+    }
+
+    return error;
+}
+
+
+static unsigned inflatev(ucvector* out, const unsigned char* in, size_t insize, const LodePNGDecompressSettings* settings)
+{
+    if (settings->custom_inflate) {
+        unsigned error = settings->custom_inflate(&out->data, &out->size, in, insize, settings);
+        out->allocsize = out->size;
+        return error;
+    } else {
+        return lodepng_inflatev(out, in, insize, settings);
+    }
+}
+
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Adler32                                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static unsigned update_adler32(unsigned adler, const unsigned char* data, unsigned len)
+{
+    unsigned s1 = adler & 0xffffu;
+    unsigned s2 = (adler >> 16u) & 0xffffu;
+
+    while (len != 0u) {
+        unsigned i;
+        /*at least 5552 sums can be done before the sums overflow, saving a lot of module divisions*/
+        unsigned amount = len > 5552u ? 5552u : len;
+        len -= amount;
+        for (i = 0; i != amount; ++i) {
+            s1 += (*data++);
+            s2 += s1;
+        }
+        s1 %= 65521u;
+        s2 %= 65521u;
+    }
+
+    return (s2 << 16u) | s1;
+}
+
+/*Return the adler32 of the bytes data[0..len-1]*/
+static unsigned adler32(const unsigned char* data, unsigned len)
+{
+    return update_adler32(1u, data, len);
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Zlib                                                                   / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static unsigned lodepng_zlib_decompressv(ucvector* out, const unsigned char* in, size_t insize, const LodePNGDecompressSettings* settings)
+{
+    unsigned error = 0;
+    unsigned CM, CINFO, FDICT;
+
+    if (insize < 2) return 53; /*error, size of zlib data too small*/
+    /*read information from zlib header*/
+    if ((in[0] * 256 + in[1]) % 31 != 0) {
+        /*error: 256 * in[0] + in[1] must be a multiple of 31, the FCHECK value is supposed to be made that way*/
+        return 24;
+    }
+
+    CM = in[0] & 15;
+    CINFO = (in[0] >> 4) & 15;
+    /*FCHECK = in[1] & 31;*/ /*FCHECK is already tested above*/
+    FDICT = (in[1] >> 5) & 1;
+    /*FLEVEL = (in[1] >> 6) & 3;*/ /*FLEVEL is not used here*/
+
+    if (CM != 8 || CINFO > 7) {
+        /*error: only compression method 8: inflate with sliding window of 32k is supported by the PNG spec*/
+        return 25;
+    }
+    if (FDICT != 0) {
+        /*error: the specification of PNG says about the zlib stream:
+          "The additional flags shall not specify a preset dictionary."*/
+        return 26;
+    }
+
+    error = inflatev(out, in + 2, insize - 2, settings);
+    if (error) return error;
+
+    if (!settings->ignore_adler32) {
+        unsigned ADLER32 = lodepng_read32bitInt(&in[insize - 4]);
+        unsigned checksum = adler32(out->data, (unsigned)(out->size));
+        if(checksum != ADLER32) return 58; /*error, adler checksum not correct, data must be corrupted*/
+    }
+
+    return 0; /*no error*/
+}
+
+
+/*expected_size is expected output size, to avoid intermediate allocations. Set to 0 if not known. */
+static unsigned zlib_decompress(unsigned char** out, size_t* outsize, size_t expected_size, const unsigned char* in, size_t insize, const LodePNGDecompressSettings* settings)
+{
+    if(settings->custom_zlib) {
+        return settings->custom_zlib(out, outsize, in, insize, settings);
+    } else {
+        unsigned error;
+        ucvector v = ucvector_init(*out, *outsize);
+        if (expected_size) {
+            /*reserve the memory to avoid intermediate reallocations*/
+            ucvector_resize(&v, *outsize + expected_size);
+            v.size = *outsize;
+        }
+        error = lodepng_zlib_decompressv(&v, in, insize, settings);
+        *out = v.data;
+        *outsize = v.size;
+        return error;
+    }
+}
+
+
+static void lodepng_decompress_settings_init(LodePNGDecompressSettings* settings)
+{
+    settings->ignore_adler32 = 0;
+    settings->ignore_nlen = 0;
+    settings->custom_zlib = 0;
+    settings->custom_inflate = 0;
+    settings->custom_context = 0;
+}
+
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // End of Zlib related code. Begin of PNG related code.                 // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+
+#if 0 //thorvg don't use crc
+/* CRC polynomial: 0xedb88320 */
+static unsigned lodepng_crc32_table[256] = {
+            0u, 1996959894u, 3993919788u, 2567524794u,  124634137u, 1886057615u, 3915621685u, 2657392035u,
+    249268274u, 2044508324u, 3772115230u, 2547177864u,  162941995u, 2125561021u, 3887607047u, 2428444049u,
+    498536548u, 1789927666u, 4089016648u, 2227061214u,  450548861u, 1843258603u, 4107580753u, 2211677639u,
+    325883990u, 1684777152u, 4251122042u, 2321926636u,  335633487u, 1661365465u, 4195302755u, 2366115317u,
+    997073096u, 1281953886u, 3579855332u, 2724688242u, 1006888145u, 1258607687u, 3524101629u, 2768942443u,
+    901097722u, 1119000684u, 3686517206u, 2898065728u,  853044451u, 1172266101u, 3705015759u, 2882616665u,
+    651767980u, 1373503546u, 3369554304u, 3218104598u,  565507253u, 1454621731u, 3485111705u, 3099436303u,
+    671266974u, 1594198024u, 3322730930u, 2970347812u,  795835527u, 1483230225u, 3244367275u, 3060149565u,
+    1994146192u,   31158534u, 2563907772u, 4023717930u, 1907459465u,  112637215u, 2680153253u, 3904427059u,
+    2013776290u,  251722036u, 2517215374u, 3775830040u, 2137656763u,  141376813u, 2439277719u, 3865271297u,
+    1802195444u,  476864866u, 2238001368u, 4066508878u, 1812370925u,  453092731u, 2181625025u, 4111451223u,
+    1706088902u,  314042704u, 2344532202u, 4240017532u, 1658658271u,  366619977u, 2362670323u, 4224994405u,
+    1303535960u,  984961486u, 2747007092u, 3569037538u, 1256170817u, 1037604311u, 2765210733u, 3554079995u,
+    1131014506u,  879679996u, 2909243462u, 3663771856u, 1141124467u,  855842277u, 2852801631u, 3708648649u,
+    1342533948u,  654459306u, 3188396048u, 3373015174u, 1466479909u,  544179635u, 3110523913u, 3462522015u,
+    1591671054u,  702138776u, 2966460450u, 3352799412u, 1504918807u,  783551873u, 3082640443u, 3233442989u,
+    3988292384u, 2596254646u,   62317068u, 1957810842u, 3939845945u, 2647816111u,   81470997u, 1943803523u,
+    3814918930u, 2489596804u,  225274430u, 2053790376u, 3826175755u, 2466906013u,  167816743u, 2097651377u,
+    4027552580u, 2265490386u,  503444072u, 1762050814u, 4150417245u, 2154129355u,  426522225u, 1852507879u,
+    4275313526u, 2312317920u,  282753626u, 1742555852u, 4189708143u, 2394877945u,  397917763u, 1622183637u,
+    3604390888u, 2714866558u,  953729732u, 1340076626u, 3518719985u, 2797360999u, 1068828381u, 1219638859u,
+    3624741850u, 2936675148u,  906185462u, 1090812512u, 3747672003u, 2825379669u,  829329135u, 1181335161u,
+    3412177804u, 3160834842u,  628085408u, 1382605366u, 3423369109u, 3138078467u,  570562233u, 1426400815u,
+    3317316542u, 2998733608u,  733239954u, 1555261956u, 3268935591u, 3050360625u,  752459403u, 1541320221u,
+    2607071920u, 3965973030u, 1969922972u,   40735498u, 2617837225u, 3943577151u, 1913087877u,   83908371u,
+    2512341634u, 3803740692u, 2075208622u,  213261112u, 2463272603u, 3855990285u, 2094854071u,  198958881u,
+    2262029012u, 4057260610u, 1759359992u,  534414190u, 2176718541u, 4139329115u, 1873836001u,  414664567u,
+    2282248934u, 4279200368u, 1711684554u,  285281116u, 2405801727u, 4167216745u, 1634467795u,  376229701u,
+    2685067896u, 3608007406u, 1308918612u,  956543938u, 2808555105u, 3495958263u, 1231636301u, 1047427035u,
+    2932959818u, 3654703836u, 1088359270u,  936918000u, 2847714899u, 3736837829u, 1202900863u,  817233897u,
+    3183342108u, 3401237130u, 1404277552u,  615818150u, 3134207493u, 3453421203u, 1423857449u,  601450431u,
+    3009837614u, 3294710456u, 1567103746u,  711928724u, 3020668471u, 3272380065u, 1510334235u,  755167117u
+};
+
+
+/* Calculate CRC32 of buffer
+   Return the CRC of the bytes buf[0..len-1]. */
+static unsigned lodepng_crc32(const unsigned char* data, size_t length)
+{
+    unsigned r = 0xffffffffu;
+    size_t i;
+    for (i = 0; i < length; ++i) {
+        r = lodepng_crc32_table[(r ^ data[i]) & 0xffu] ^ (r >> 8u);
+    }
+    return r ^ 0xffffffffu;
+}
+#endif
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Reading and writing PNG color channel bits                             / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/* The color channel bits of less-than-8-bit pixels are read with the MSB of bytes first,
+so LodePNGBitWriter and LodePNGBitReader can't be used for those. */
+
+static unsigned char readBitFromReversedStream(size_t* bitpointer, const unsigned char* bitstream)
+{
+    unsigned char result = (unsigned char)((bitstream[(*bitpointer) >> 3] >> (7 - ((*bitpointer) & 0x7))) & 1);
+    ++(*bitpointer);
+    return result;
+}
+
+
+/* TODO: make this faster */
+static unsigned readBitsFromReversedStream(size_t* bitpointer, const unsigned char* bitstream, size_t nbits)
+{
+    unsigned result = 0;
+    size_t i;
+    for (i = 0 ; i < nbits; ++i) {
+        result <<= 1u;
+        result |= (unsigned)readBitFromReversedStream(bitpointer, bitstream);
+    }
+    return result;
+}
+
+
+static void setBitOfReversedStream(size_t* bitpointer, unsigned char* bitstream, unsigned char bit)
+{
+    /*the current bit in bitstream may be 0 or 1 for this to work*/
+    if (bit == 0) bitstream[(*bitpointer) >> 3u] &=  (unsigned char)(~(1u << (7u - ((*bitpointer) & 7u))));
+    else bitstream[(*bitpointer) >> 3u] |=  (1u << (7u - ((*bitpointer) & 7u)));
+    ++(*bitpointer);
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG chunks                                                             / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*
+  The lodepng_chunk functions are normally not needed, except to traverse the
+  unknown chunks stored in the LodePNGInfo struct, or add new ones to it.
+  It also allows traversing the chunks of an encoded PNG file yourself.
+
+  The chunk pointer always points to the beginning of the chunk itself, that is
+  the first byte of the 4 length bytes.
+
+  In the PNG file format, chunks have the following format:
+  -4 bytes length: length of the data of the chunk in bytes (chunk itself is 12 bytes longer)
+  -4 bytes chunk type (ASCII a-z,A-Z only, see below)
+  -length bytes of data (may be 0 bytes if length was 0)
+  -4 bytes of CRC, computed on chunk name + data
+
+  The first chunk starts at the 8th byte of the PNG file, the entire rest of the file
+  exists out of concatenated chunks with the above format.
+
+  PNG standard chunk ASCII naming conventions:
+  -First byte: uppercase = critical, lowercase = ancillary
+  -Second byte: uppercase = public, lowercase = private
+  -Third byte: must be uppercase
+  -Fourth byte: uppercase = unsafe to copy, lowercase = safe to copy
+*/
+
+
+/*
+  Gets the length of the data of the chunk. Total chunk length has 12 bytes more.
+  There must be at least 4 bytes to read from. If the result value is too large,
+  it may be corrupt data.
+*/
+static unsigned lodepng_chunk_length(const unsigned char* chunk)
+{
+    return lodepng_read32bitInt(&chunk[0]);
+}
+
+
+/* check if the type is the given type */
+static unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type)
+{
+    if (lodepng_strlen(type) != 4) return 0;
+    return (chunk[4] == type[0] && chunk[5] == type[1] && chunk[6] == type[2] && chunk[7] == type[3]);
+}
+
+
+/* 0: it's one of the critical chunk types, 1: it's an ancillary chunk (see PNG standard) */
+static unsigned char lodepng_chunk_ancillary(const unsigned char* chunk)
+{
+    return ((chunk[4] & 32) != 0);
+}
+
+
+static const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk)
+{
+    return &chunk[8];
+}
+
+#if 0 //thorvg don't use crc
+/* returns 0 if the crc is correct, 1 if it's incorrect (0 for OK as usual!) */
+static unsigned lodepng_chunk_check_crc(const unsigned char* chunk)
+{
+    unsigned length = lodepng_chunk_length(chunk);
+    unsigned CRC = lodepng_read32bitInt(&chunk[length + 8]);
+    /*the CRC is taken of the data and the 4 chunk type letters, not the length*/
+    unsigned checksum = lodepng_crc32(&chunk[4], length + 4);
+    if (CRC != checksum) return 1;
+    else return 0;
+}
+#endif
+
+static const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk, const unsigned char* end)
+{
+    if (chunk >= end || end - chunk < 12) return end; /*too small to contain a chunk*/
+    if (chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47
+        && chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) {
+        /* Is PNG magic header at start of PNG file. Jump to first actual chunk. */
+        return chunk + 8;
+    } else {
+        size_t total_chunk_length;
+        const unsigned char* result;
+        if (lodepng_addofl(lodepng_chunk_length(chunk), 12, &total_chunk_length)) return end;
+        result = chunk + total_chunk_length;
+        if (result < chunk) return end; /*pointer overflow*/
+        return result;
+    }
+}
+
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Color types, channels, bits                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*checks if the colortype is valid and the bitdepth bd is allowed for this colortype.
+Return value is a LodePNG error code.*/
+static unsigned checkColorValidity(LodePNGColorType colortype, unsigned bd)
+{
+    switch(colortype) {
+        case LCT_GREY:       if(!(bd == 1 || bd == 2 || bd == 4 || bd == 8 || bd == 16)) return 37; break;
+        case LCT_RGB:        if(!(                                 bd == 8 || bd == 16)) return 37; break;
+        case LCT_PALETTE:    if(!(bd == 1 || bd == 2 || bd == 4 || bd == 8            )) return 37; break;
+        case LCT_GREY_ALPHA: if(!(                                 bd == 8 || bd == 16)) return 37; break;
+        case LCT_RGBA:       if(!(                                 bd == 8 || bd == 16)) return 37; break;
+        case LCT_MAX_OCTET_VALUE: return 31; /* invalid color type */
+        default: return 31; /* invalid color type */
+    }
+    return 0; /*allowed color type / bits combination*/
+}
+
+
+static unsigned getNumColorChannels(LodePNGColorType colortype)
+{
+    switch(colortype) {
+        case LCT_GREY: return 1;
+        case LCT_RGB: return 3;
+        case LCT_PALETTE: return 1;
+        case LCT_GREY_ALPHA: return 2;
+        case LCT_RGBA: return 4;
+        case LCT_MAX_OCTET_VALUE: return 0; /* invalid color type */
+        default: return 0; /*invalid color type*/
+    }
+}
+
+
+static unsigned lodepng_get_bpp_lct(LodePNGColorType colortype, unsigned bitdepth)
+{
+    /*bits per pixel is amount of channels * bits per channel*/
+    return getNumColorChannels(colortype) * bitdepth;
+}
+
+
+static void lodepng_color_mode_init(LodePNGColorMode* info)
+{
+    info->key_defined = 0;
+    info->key_r = info->key_g = info->key_b = 0;
+    info->colortype = LCT_RGBA;
+    info->bitdepth = 8;
+    info->palette = 0;
+    info->palettesize = 0;
+}
+
+
+/*allocates palette memory if needed, and initializes all colors to black*/
+static void lodepng_color_mode_alloc_palette(LodePNGColorMode* info)
+{
+    size_t i;
+    /*if the palette is already allocated, it will have size 1024 so no reallocation needed in that case*/
+    /*the palette must have room for up to 256 colors with 4 bytes each.*/
+    if (!info->palette) info->palette = (unsigned char*)malloc(1024);
+    if (!info->palette) return; /*alloc fail*/
+    for (i = 0; i != 256; ++i) {
+        /*Initialize all unused colors with black, the value used for invalid palette indices.
+        This is an error according to the PNG spec, but common PNG decoders make it black instead.
+        That makes color conversion slightly faster due to no error handling needed.*/
+        info->palette[i * 4 + 0] = 0;
+        info->palette[i * 4 + 1] = 0;
+        info->palette[i * 4 + 2] = 0;
+        info->palette[i * 4 + 3] = 255;
+    }
+}
+
+static void lodepng_palette_clear(LodePNGColorMode* info)
+{
+    if (info->palette) free(info->palette);
+    info->palette = 0;
+    info->palettesize = 0;
+}
+
+
+static void lodepng_color_mode_cleanup(LodePNGColorMode* info)
+{
+    lodepng_palette_clear(info);
+}
+
+
+/*return value is error code (0 means no error)*/
+static unsigned lodepng_color_mode_copy(LodePNGColorMode* dest, const LodePNGColorMode* source)
+{
+    lodepng_color_mode_cleanup(dest);
+    lodepng_memcpy(dest, source, sizeof(LodePNGColorMode));
+    if (source->palette) {
+        dest->palette = (unsigned char*)malloc(1024);
+        if (!dest->palette && source->palettesize) return 83; /*alloc fail*/
+        lodepng_memcpy(dest->palette, source->palette, source->palettesize * 4);
+    }
+    return 0;
+}
+
+
+static int lodepng_color_mode_equal(const LodePNGColorMode* a, const LodePNGColorMode* b)
+{
+    size_t i;
+    if (a->colortype != b->colortype) return 0;
+    if (a->bitdepth != b->bitdepth) return 0;
+    if (a->key_defined != b->key_defined) return 0;
+    if (a->key_defined) {
+        if(a->key_r != b->key_r) return 0;
+        if(a->key_g != b->key_g) return 0;
+        if(a->key_b != b->key_b) return 0;
+    }
+    if (a->palettesize != b->palettesize) return 0;
+    for (i = 0; i != a->palettesize * 4; ++i) {
+        if (a->palette[i] != b->palette[i]) return 0;
+    }
+    return 1;
+}
+
+
+static size_t lodepng_get_raw_size_lct(unsigned w, unsigned h, LodePNGColorType colortype, unsigned bitdepth)
+{
+    size_t bpp = lodepng_get_bpp_lct(colortype, bitdepth);
+    size_t n = (size_t)w * (size_t)h;
+    return ((n / 8u) * bpp) + ((n & 7u) * bpp + 7u) / 8u;
+}
+
+
+/* Returns the byte size of a raw image buffer with given width, height and color mode */
+static size_t lodepng_get_raw_size(unsigned w, unsigned h, const LodePNGColorMode* color)
+{
+    return lodepng_get_raw_size_lct(w, h, color->colortype, color->bitdepth);
+}
+
+
+/*in an idat chunk, each scanline is a multiple of 8 bits, unlike the lodepng output buffer,
+and in addition has one extra byte per line: the filter byte. So this gives a larger
+result than lodepng_get_raw_size. Set h to 1 to get the size of 1 row including filter byte. */
+static size_t lodepng_get_raw_size_idat(unsigned w, unsigned h, unsigned bpp)
+{
+    /* + 1 for the filter byte, and possibly plus padding bits per line. */
+    /* Ignoring casts, the expression is equal to (w * bpp + 7) / 8 + 1, but avoids overflow of w * bpp */
+    size_t line = ((size_t)(w / 8u) * bpp) + 1u + ((w & 7u) * bpp + 7u) / 8u;
+    return (size_t)h * line;
+}
+
+
+/* Safely checks whether size_t overflow can be caused due to amount of pixels.
+   This check is overcautious rather than precise. If this check indicates no overflow,
+   you can safely compute in a size_t (but not an unsigned):
+   -(size_t)w * (size_t)h * 8
+   -amount of bytes in IDAT (including filter, padding and Adam7 bytes)
+   -amount of bytes in raw color model
+   Returns 1 if overflow possible, 0 if not. */
+static int lodepng_pixel_overflow(unsigned w, unsigned h, const LodePNGColorMode* pngcolor, const LodePNGColorMode* rawcolor)
+{
+    size_t bpp = LODEPNG_MAX(lodepng_get_bpp_lct(pngcolor->colortype, pngcolor->bitdepth), lodepng_get_bpp_lct(rawcolor->colortype, rawcolor->bitdepth));
+    size_t numpixels, total;
+    size_t line; /* bytes per line in worst case */
+
+    if (lodepng_mulofl((size_t)w, (size_t)h, &numpixels)) return 1;
+    if (lodepng_mulofl(numpixels, 8, &total)) return 1; /* bit pointer with 8-bit color, or 8 bytes per channel color */
+
+    /* Bytes per scanline with the expression "(w / 8u) * bpp) + ((w & 7u) * bpp + 7u) / 8u" */
+    if (lodepng_mulofl((size_t)(w / 8u), bpp, &line)) return 1;
+    if (lodepng_addofl(line, ((w & 7u) * bpp + 7u) / 8u, &line)) return 1;
+
+    if (lodepng_addofl(line, 5, &line)) return 1; /* 5 bytes overhead per line: 1 filterbyte, 4 for Adam7 worst case */
+    if (lodepng_mulofl(line, h, &total)) return 1; /* Total bytes in worst case */
+
+    return 0; /* no overflow */
+}
+
+
+static void lodepng_info_init(LodePNGInfo* info)
+{
+    lodepng_color_mode_init(&info->color);
+    info->interlace_method = 0;
+    info->compression_method = 0;
+    info->filter_method = 0;
+}
+
+
+static void lodepng_info_cleanup(LodePNGInfo* info)
+{
+    lodepng_color_mode_cleanup(&info->color);
+}
+
+
+/* index: bitgroup index, bits: bitgroup size(1, 2 or 4), in: bitgroup value, out: octet array to add bits to */
+static void addColorBits(unsigned char* out, size_t index, unsigned bits, unsigned in)
+{
+    unsigned m = bits == 1 ? 7 : bits == 2 ? 3 : 1; /*8 / bits - 1*/
+    /*p = the partial index in the byte, e.g. with 4 palettebits it is 0 for first half or 1 for second half*/
+    unsigned p = index & m;
+    in &= (1u << bits) - 1u; /*filter out any other bits of the input value*/
+    in = in << (bits * (m - p));
+    if(p == 0) out[index * bits / 8u] = in;
+    else out[index * bits / 8u] |= in;
+}
+
+/*
+    One node of a color tree
+    This is the data structure used to count the number of unique colors and to get a palette
+    index for a color. It's like an octree, but because the alpha channel is used too, each
+    node has 16 instead of 8 children.
+*/
+struct ColorTree
+{
+    ColorTree* children[16]; /* up to 16 pointers to ColorTree of next level */
+    int index; /* the payload. Only has a meaningful value if this is in the last level */
+};
+
+static void color_tree_init(ColorTree* tree)
+{
+    lodepng_memset(tree->children, 0, 16 * sizeof(*tree->children));
+    tree->index = -1;
+}
+
+static void color_tree_cleanup(ColorTree* tree)
+{
+    int i;
+    for (i = 0; i != 16; ++i) {
+        if(tree->children[i]) {
+            color_tree_cleanup(tree->children[i]);
+            free(tree->children[i]);
+        }
+    }
+}
+
+
+/* returns -1 if color not present, its index otherwise */
+static int color_tree_get(ColorTree* tree, unsigned char r, unsigned char g, unsigned char b, unsigned char a)
+{
+    int bit = 0;
+    for (bit = 0; bit < 8; ++bit) {
+        int i = 8 * ((r >> bit) & 1) + 4 * ((g >> bit) & 1) + 2 * ((b >> bit) & 1) + 1 * ((a >> bit) & 1);
+        if (!tree->children[i]) return -1;
+        else tree = tree->children[i];
+    }
+    return tree ? tree->index : -1;
+}
+
+
+/* color is not allowed to already exist.
+   Index should be >= 0 (it's signed to be compatible with using -1 for "doesn't exist")
+   Returns error code, or 0 if ok */
+static unsigned color_tree_add(ColorTree* tree, unsigned char r, unsigned char g, unsigned char b, unsigned char a, unsigned index)
+{
+    int bit;
+    for (bit = 0; bit < 8; ++bit) {
+        int i = 8 * ((r >> bit) & 1) + 4 * ((g >> bit) & 1) + 2 * ((b >> bit) & 1) + 1 * ((a >> bit) & 1);
+        if (!tree->children[i]) {
+            tree->children[i] = (ColorTree*)malloc(sizeof(ColorTree));
+            if (!tree->children[i]) return 83; /*alloc fail*/
+            color_tree_init(tree->children[i]);
+        }
+        tree = tree->children[i];
+    }
+    tree->index = (int)index;
+    return 0;
+}
+
+/* put a pixel, given its RGBA color, into image of any color type */
+static unsigned rgba8ToPixel(unsigned char* out, size_t i, const LodePNGColorMode* mode, ColorTree* tree /*for palette*/, unsigned char r, unsigned char g, unsigned char b, unsigned char a)
+{
+    if (mode->colortype == LCT_GREY) {
+        unsigned char gray = r; /*((unsigned short)r + g + b) / 3u;*/
+        if (mode->bitdepth == 8) out[i] = gray;
+        else if (mode->bitdepth == 16) out[i * 2 + 0] = out[i * 2 + 1] = gray;
+        else {
+            /*take the most significant bits of gray*/
+            gray = ((unsigned)gray >> (8u - mode->bitdepth)) & ((1u << mode->bitdepth) - 1u);
+            addColorBits(out, i, mode->bitdepth, gray);
+        }
+    } else if (mode->colortype == LCT_RGB) {
+        if (mode->bitdepth == 8) {
+            out[i * 3 + 0] = r;
+            out[i * 3 + 1] = g;
+            out[i * 3 + 2] = b;
+        } else {
+            out[i * 6 + 0] = out[i * 6 + 1] = r;
+            out[i * 6 + 2] = out[i * 6 + 3] = g;
+            out[i * 6 + 4] = out[i * 6 + 5] = b;
+        }
+    } else if(mode->colortype == LCT_PALETTE) {
+        int index = color_tree_get(tree, r, g, b, a);
+        if (index < 0) return 82; /*color not in palette*/
+        if (mode->bitdepth == 8) out[i] = index;
+        else addColorBits(out, i, mode->bitdepth, (unsigned)index);
+    } else if (mode->colortype == LCT_GREY_ALPHA) {
+        unsigned char gray = r; /*((unsigned short)r + g + b) / 3u;*/
+        if (mode->bitdepth == 8) {
+            out[i * 2 + 0] = gray;
+            out[i * 2 + 1] = a;
+        } else if (mode->bitdepth == 16) {
+            out[i * 4 + 0] = out[i * 4 + 1] = gray;
+            out[i * 4 + 2] = out[i * 4 + 3] = a;
+        }
+    } else if (mode->colortype == LCT_RGBA) {
+        if (mode->bitdepth == 8) {
+            out[i * 4 + 0] = r;
+            out[i * 4 + 1] = g;
+            out[i * 4 + 2] = b;
+            out[i * 4 + 3] = a;
+        } else {
+            out[i * 8 + 0] = out[i * 8 + 1] = r;
+            out[i * 8 + 2] = out[i * 8 + 3] = g;
+            out[i * 8 + 4] = out[i * 8 + 5] = b;
+            out[i * 8 + 6] = out[i * 8 + 7] = a;
+        }
+    }
+    return 0; /*no error*/
+}
+
+
+/* put a pixel, given its RGBA16 color, into image of any color 16-bitdepth type */
+static void rgba16ToPixel(unsigned char* out, size_t i, const LodePNGColorMode* mode, unsigned short r, unsigned short g, unsigned short b, unsigned short a)
+{
+    if (mode->colortype == LCT_GREY) {
+        unsigned short gray = r; /*((unsigned)r + g + b) / 3u;*/
+        out[i * 2 + 0] = (gray >> 8) & 255;
+        out[i * 2 + 1] = gray & 255;
+    } else if (mode->colortype == LCT_RGB) {
+        out[i * 6 + 0] = (r >> 8) & 255;
+        out[i * 6 + 1] = r & 255;
+        out[i * 6 + 2] = (g >> 8) & 255;
+        out[i * 6 + 3] = g & 255;
+        out[i * 6 + 4] = (b >> 8) & 255;
+        out[i * 6 + 5] = b & 255;
+    } else if (mode->colortype == LCT_GREY_ALPHA) {
+        unsigned short gray = r; /*((unsigned)r + g + b) / 3u;*/
+        out[i * 4 + 0] = (gray >> 8) & 255;
+        out[i * 4 + 1] = gray & 255;
+        out[i * 4 + 2] = (a >> 8) & 255;
+        out[i * 4 + 3] = a & 255;
+    } else if (mode->colortype == LCT_RGBA) {
+        out[i * 8 + 0] = (r >> 8) & 255;
+        out[i * 8 + 1] = r & 255;
+        out[i * 8 + 2] = (g >> 8) & 255;
+        out[i * 8 + 3] = g & 255;
+        out[i * 8 + 4] = (b >> 8) & 255;
+        out[i * 8 + 5] = b & 255;
+        out[i * 8 + 6] = (a >> 8) & 255;
+        out[i * 8 + 7] = a & 255;
+    }
+}
+
+
+/* Get RGBA8 color of pixel with index i (y * width + x) from the raw image with given color type. */
+static void getPixelColorRGBA8(unsigned char* r, unsigned char* g, unsigned char* b, unsigned char* a, const unsigned char* in, size_t i, const LodePNGColorMode* mode)
+{
+    if (mode->colortype == LCT_GREY) {
+        if (mode->bitdepth == 8) {
+            *r = *g = *b = in[i];
+            if (mode->key_defined && *r == mode->key_r) *a = 0;
+            else *a = 255;
+        } else if (mode->bitdepth == 16) {
+            *r = *g = *b = in[i * 2 + 0];
+            if (mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r) *a = 0;
+            else *a = 255;
+        } else {
+            unsigned highest = ((1U << mode->bitdepth) - 1U); /* highest possible value for this bit depth */
+            size_t j = i * mode->bitdepth;
+            unsigned value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+            *r = *g = *b = (value * 255) / highest;
+            if (mode->key_defined && value == mode->key_r) *a = 0;
+            else *a = 255;
+        }
+    } else if (mode->colortype == LCT_RGB) {
+        if (mode->bitdepth == 8) {
+            *r = in[i * 3 + 0]; *g = in[i * 3 + 1]; *b = in[i * 3 + 2];
+            if (mode->key_defined && *r == mode->key_r && *g == mode->key_g && *b == mode->key_b) *a = 0;
+            else *a = 255;
+        } else {
+            *r = in[i * 6 + 0];
+            *g = in[i * 6 + 2];
+            *b = in[i * 6 + 4];
+            if (mode->key_defined && 256U * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r
+              && 256U * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g
+              && 256U * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b) *a = 0;
+            else *a = 255;
+        }
+    } else if (mode->colortype == LCT_PALETTE) {
+        unsigned index;
+        if (mode->bitdepth == 8) index = in[i];
+        else {
+            size_t j = i * mode->bitdepth;
+            index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+        }
+        /* out of bounds of palette not checked: see lodepng_color_mode_alloc_palette. */
+        *r = mode->palette[index * 4 + 0];
+        *g = mode->palette[index * 4 + 1];
+        *b = mode->palette[index * 4 + 2];
+        *a = mode->palette[index * 4 + 3];
+    } else if (mode->colortype == LCT_GREY_ALPHA) {
+        if (mode->bitdepth == 8) {
+            *r = *g = *b = in[i * 2 + 0];
+            *a = in[i * 2 + 1];
+        } else {
+            *r = *g = *b = in[i * 4 + 0];
+            *a = in[i * 4 + 2];
+        }
+    } else if (mode->colortype == LCT_RGBA) {
+        if (mode->bitdepth == 8) {
+            *r = in[i * 4 + 0];
+            *g = in[i * 4 + 1];
+            *b = in[i * 4 + 2];
+            *a = in[i * 4 + 3];
+        } else {
+            *r = in[i * 8 + 0];
+            *g = in[i * 8 + 2];
+            *b = in[i * 8 + 4];
+            *a = in[i * 8 + 6];
+        }
+    }
+}
+
+
+/* Similar to getPixelColorRGBA8, but with all the for loops inside of the color
+   mode test cases, optimized to convert the colors much faster, when converting
+   to the common case of RGBA with 8 bit per channel. buffer must be RGBA with
+   enough memory.*/
+static void getPixelColorsRGBA8(unsigned char* LODEPNG_RESTRICT buffer, size_t numpixels, const unsigned char* LODEPNG_RESTRICT in, const LodePNGColorMode* mode)
+{
+    unsigned num_channels = 4;
+    size_t i;
+    if (mode->colortype == LCT_GREY) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = buffer[1] = buffer[2] = in[i];
+                buffer[3] = 255;
+            }
+            if (mode->key_defined) {
+                buffer -= numpixels * num_channels;
+                for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                    if(buffer[0] == mode->key_r) buffer[3] = 0;
+                }
+            }
+        } else if (mode->bitdepth == 16) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = buffer[1] = buffer[2] = in[i * 2];
+                buffer[3] = mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r ? 0 : 255;
+            }
+        } else {
+            unsigned highest = ((1U << mode->bitdepth) - 1U); /* highest possible value for this bit depth */
+            size_t j = 0;
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                unsigned value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+                buffer[0] = buffer[1] = buffer[2] = (value * 255) / highest;
+                buffer[3] = mode->key_defined && value == mode->key_r ? 0 : 255;
+            }
+        }
+    } else if (mode->colortype == LCT_RGB) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                lodepng_memcpy(buffer, &in[i * 3], 3);
+                buffer[3] = 255;
+            }
+            if (mode->key_defined) {
+                buffer -= numpixels * num_channels;
+                for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                    if (buffer[0] == mode->key_r && buffer[1]== mode->key_g && buffer[2] == mode->key_b) buffer[3] = 0;
+                }
+            }
+        } else {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = in[i * 6 + 0];
+                buffer[1] = in[i * 6 + 2];
+                buffer[2] = in[i * 6 + 4];
+                buffer[3] = mode->key_defined
+                  && 256U * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r
+                  && 256U * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g
+                  && 256U * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b ? 0 : 255;
+            }
+        }
+    } else if (mode->colortype == LCT_PALETTE) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+              unsigned index = in[i];
+              /* out of bounds of palette not checked: see lodepng_color_mode_alloc_palette. */
+              lodepng_memcpy(buffer, &mode->palette[index * 4], 4);
+            }
+        } else {
+            size_t j = 0;
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+              unsigned index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+              /* out of bounds of palette not checked: see lodepng_color_mode_alloc_palette. */
+              lodepng_memcpy(buffer, &mode->palette[index * 4], 4);
+            }
+        }
+    } else if (mode->colortype == LCT_GREY_ALPHA) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+              buffer[0] = buffer[1] = buffer[2] = in[i * 2 + 0];
+              buffer[3] = in[i * 2 + 1];
+            }
+        } else {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+              buffer[0] = buffer[1] = buffer[2] = in[i * 4 + 0];
+              buffer[3] = in[i * 4 + 2];
+            }
+        }
+    } else if (mode->colortype == LCT_RGBA) {
+        if (mode->bitdepth == 8) {
+            lodepng_memcpy(buffer, in, numpixels * 4);
+        } else {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = in[i * 8 + 0];
+                buffer[1] = in[i * 8 + 2];
+                buffer[2] = in[i * 8 + 4];
+                buffer[3] = in[i * 8 + 6];
+            }
+        }
+    }
+}
+
+
+/* Similar to getPixelColorsRGBA8, but with 3-channel RGB output. */
+static void getPixelColorsRGB8(unsigned char* LODEPNG_RESTRICT buffer, size_t numpixels, const unsigned char* LODEPNG_RESTRICT in, const LodePNGColorMode* mode)
+{
+    const unsigned num_channels = 3;
+    size_t i;
+    if (mode->colortype == LCT_GREY) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+               buffer[0] = buffer[1] = buffer[2] = in[i];
+            }
+        } else if (mode->bitdepth == 16) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = buffer[1] = buffer[2] = in[i * 2];
+            }
+        } else {
+            unsigned highest = ((1U << mode->bitdepth) - 1U); /* highest possible value for this bit depth */
+            size_t j = 0;
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                unsigned value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+                buffer[0] = buffer[1] = buffer[2] = (value * 255) / highest;
+            }
+        }
+    } else if (mode->colortype == LCT_RGB) {
+        if (mode->bitdepth == 8) {
+           lodepng_memcpy(buffer, in, numpixels * 3);
+        } else {
+            for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = in[i * 6 + 0];
+                buffer[1] = in[i * 6 + 2];
+                buffer[2] = in[i * 6 + 4];
+            }
+        }
+    } else if (mode->colortype == LCT_PALETTE) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                unsigned index = in[i];
+                /* out of bounds of palette not checked: see lodepng_color_mode_alloc_palette. */
+                lodepng_memcpy(buffer, &mode->palette[index * 4], 3);
+            }
+        } else {
+            size_t j = 0;
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                unsigned index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+                /* out of bounds of palette not checked: see lodepng_color_mode_alloc_palette. */
+                lodepng_memcpy(buffer, &mode->palette[index * 4], 3);
+            }
+        }
+    } else if (mode->colortype == LCT_GREY_ALPHA) {
+        if (mode->bitdepth == 8) {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = buffer[1] = buffer[2] = in[i * 2 + 0];
+            }
+        } else {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = buffer[1] = buffer[2] = in[i * 4 + 0];
+            }
+        }
+    } else if (mode->colortype == LCT_RGBA) {
+        if (mode->bitdepth == 8) {
+            for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+                lodepng_memcpy(buffer, &in[i * 4], 3);
+            }
+        } else {
+            for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+                buffer[0] = in[i * 8 + 0];
+                buffer[1] = in[i * 8 + 2];
+                buffer[2] = in[i * 8 + 4];
+            }
+        }
+    }
+}
+
+
+/* Get RGBA16 color of pixel with index i (y * width + x) from the raw image with
+   given color type, but the given color type must be 16-bit itself. */
+static void getPixelColorRGBA16(unsigned short* r, unsigned short* g, unsigned short* b, unsigned short* a, const unsigned char* in, size_t i, const LodePNGColorMode* mode)
+{
+    if (mode->colortype == LCT_GREY) {
+        *r = *g = *b = 256 * in[i * 2 + 0] + in[i * 2 + 1];
+        if (mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r) *a = 0;
+        else *a = 65535;
+    } else if (mode->colortype == LCT_RGB) {
+        *r = 256u * in[i * 6 + 0] + in[i * 6 + 1];
+        *g = 256u * in[i * 6 + 2] + in[i * 6 + 3];
+        *b = 256u * in[i * 6 + 4] + in[i * 6 + 5];
+        if (mode->key_defined
+          && 256u * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r
+          && 256u * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g
+          && 256u * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b) *a = 0;
+        else *a = 65535;
+    } else if (mode->colortype == LCT_GREY_ALPHA) {
+        *r = *g = *b = 256u * in[i * 4 + 0] + in[i * 4 + 1];
+        *a = 256u * in[i * 4 + 2] + in[i * 4 + 3];
+    } else if (mode->colortype == LCT_RGBA) {
+        *r = 256u * in[i * 8 + 0] + in[i * 8 + 1];
+        *g = 256u * in[i * 8 + 2] + in[i * 8 + 3];
+        *b = 256u * in[i * 8 + 4] + in[i * 8 + 5];
+        *a = 256u * in[i * 8 + 6] + in[i * 8 + 7];
+    }
+}
+
+/*
+  Converts raw buffer from one color type to another color type, based on
+  LodePNGColorMode structs to describe the input and output color type.
+  See the reference manual at the end of this header file to see which color conversions are supported.
+  return value = LodePNG error code (0 if all went ok, an error if the conversion isn't supported)
+  The out buffer must have size (w * h * bpp + 7) / 8, where bpp is the bits per pixel
+  of the output color type (lodepng_get_bpp).
+  For < 8 bpp images, there should not be padding bits at the end of scanlines.
+  For 16-bit per channel colors, uses big endian format like PNG does.
+  Return value is LodePNG error code
+*/
+static unsigned lodepng_convert(unsigned char* out, const unsigned char* in, const LodePNGColorMode* mode_out, const LodePNGColorMode* mode_in, unsigned w, unsigned h)
+{
+    size_t i;
+    ColorTree tree;
+    size_t numpixels = (size_t)w * (size_t)h;
+    unsigned error = 0;
+
+    if (mode_in->colortype == LCT_PALETTE && !mode_in->palette) {
+        return 107; /* error: must provide palette if input mode is palette */
+    }
+
+    if (lodepng_color_mode_equal(mode_out, mode_in)) {
+        size_t numbytes = lodepng_get_raw_size(w, h, mode_in);
+        lodepng_memcpy(out, in, numbytes);
+        return 0;
+    }
+
+    if (mode_out->colortype == LCT_PALETTE) {
+        size_t palettesize = mode_out->palettesize;
+        const unsigned char* palette = mode_out->palette;
+        size_t palsize = (size_t)1u << mode_out->bitdepth;
+        /* if the user specified output palette but did not give the values, assume
+           they want the values of the input color type (assuming that one is palette).
+           Note that we never create a new palette ourselves.*/
+        if (palettesize == 0) {
+            palettesize = mode_in->palettesize;
+            palette = mode_in->palette;
+            /* if the input was also palette with same bitdepth, then the color types are also
+               equal, so copy literally. This to preserve the exact indices that were in the PNG
+               even in case there are duplicate colors in the palette.*/
+            if (mode_in->colortype == LCT_PALETTE && mode_in->bitdepth == mode_out->bitdepth) {
+                size_t numbytes = lodepng_get_raw_size(w, h, mode_in);
+                lodepng_memcpy(out, in, numbytes);
+                return 0;
+            }
+        }
+        if (palettesize < palsize) palsize = palettesize;
+        color_tree_init(&tree);
+        for (i = 0; i != palsize; ++i) {
+            const unsigned char* p = &palette[i * 4];
+            error = color_tree_add(&tree, p[0], p[1], p[2], p[3], (unsigned)i);
+            if (error) break;
+        }
+    }
+
+    if (!error) {
+        if (mode_in->bitdepth == 16 && mode_out->bitdepth == 16) {
+            for (i = 0; i != numpixels; ++i) {
+                unsigned short r = 0, g = 0, b = 0, a = 0;
+                getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+                rgba16ToPixel(out, i, mode_out, r, g, b, a);
+            }
+        } else if (mode_out->bitdepth == 8 && mode_out->colortype == LCT_RGBA) {
+            getPixelColorsRGBA8(out, numpixels, in, mode_in);
+        } else if(mode_out->bitdepth == 8 && mode_out->colortype == LCT_RGB) {
+            getPixelColorsRGB8(out, numpixels, in, mode_in);
+        } else {
+            unsigned char r = 0, g = 0, b = 0, a = 0;
+            for (i = 0; i != numpixels; ++i) {
+                getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+                error = rgba8ToPixel(out, i, mode_out, &tree, r, g, b, a);
+                if (error) break;
+            }
+        }
+    }
+
+    if (mode_out->colortype == LCT_PALETTE) {
+        color_tree_cleanup(&tree);
+    }
+
+    return error;
+}
+
+
+/* Paeth predictor, used by PNG filter type 4
+   The parameters are of type short, but should come from unsigned chars, the shorts
+   are only needed to make the paeth calculation correct.
+*/
+static unsigned char paethPredictor(short a, short b, short c)
+{
+    short pa = LODEPNG_ABS(b - c);
+    short pb = LODEPNG_ABS(a - c);
+    short pc = LODEPNG_ABS(a + b - c - c);
+    /* return input value associated with smallest of pa, pb, pc (with certain priority if equal) */
+    if (pb < pa) { a = b; pa = pb; }
+    return (pc < pa) ? c : a;
+}
+
+
+/*shared values used by multiple Adam7 related functions*/
+static const unsigned ADAM7_IX[7] = { 0, 4, 0, 2, 0, 1, 0 }; /*x start values*/
+static const unsigned ADAM7_IY[7] = { 0, 0, 4, 0, 2, 0, 1 }; /*y start values*/
+static const unsigned ADAM7_DX[7] = { 8, 8, 4, 4, 2, 2, 1 }; /*x delta values*/
+static const unsigned ADAM7_DY[7] = { 8, 8, 8, 4, 4, 2, 2 }; /*y delta values*/
+
+/* Outputs various dimensions and positions in the image related to the Adam7 reduced images.
+   passw: output containing the width of the 7 passes
+   passh: output containing the height of the 7 passes
+   filter_passstart: output containing the index of the start and end of each
+   reduced image with filter bytes
+   padded_passstart output containing the index of the start and end of each
+   reduced image when without filter bytes but with padded scanlines
+   passstart: output containing the index of the start and end of each reduced
+   image without padding between scanlines, but still padding between the images
+   w, h: width and height of non-interlaced image
+   bpp: bits per pixel
+   "padded" is only relevant if bpp is less than 8 and a scanline or image does not
+   end at a full byte */
+static void Adam7_getpassvalues(unsigned passw[7], unsigned passh[7], size_t filter_passstart[8], size_t padded_passstart[8], size_t passstart[8], unsigned w, unsigned h, unsigned bpp)
+{
+    /* the passstart values have 8 values: the 8th one indicates the byte after the end of the 7th (= last) pass */
+    unsigned i;
+
+    /* calculate width and height in pixels of each pass */
+    for (i = 0; i != 7; ++i) {
+        passw[i] = (w + ADAM7_DX[i] - ADAM7_IX[i] - 1) / ADAM7_DX[i];
+        passh[i] = (h + ADAM7_DY[i] - ADAM7_IY[i] - 1) / ADAM7_DY[i];
+        if(passw[i] == 0) passh[i] = 0;
+        if(passh[i] == 0) passw[i] = 0;
+    }
+
+    filter_passstart[0] = padded_passstart[0] = passstart[0] = 0;
+    for (i = 0; i != 7; ++i) {
+        /* if passw[i] is 0, it's 0 bytes, not 1 (no filtertype-byte) */
+        filter_passstart[i + 1] = filter_passstart[i]
+                                + ((passw[i] && passh[i]) ? passh[i] * (1u + (passw[i] * bpp + 7u) / 8u) : 0);
+        /* bits padded if needed to fill full byte at end of each scanline */
+        padded_passstart[i + 1] = padded_passstart[i] + passh[i] * ((passw[i] * bpp + 7u) / 8u);
+        /* only padded at end of reduced image */
+        passstart[i + 1] = passstart[i] + (passh[i] * passw[i] * bpp + 7u) / 8u;
+    }
+}
+
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG Decoder                                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static unsigned unfilterScanline(unsigned char* recon, const unsigned char* scanline, const unsigned char* precon, size_t bytewidth, unsigned char filterType, size_t length)
+{
+    /* For PNG filter method 0
+       unfilter a PNG image scanline by scanline. when the pixels are smaller than 1 byte,
+       the filter works byte per byte (bytewidth = 1)
+       precon is the previous unfiltered scanline, recon the result, scanline the current one
+       the incoming scanlines do NOT include the filtertype byte, that one is given in the parameter filterType instead
+       recon and scanline MAY be the same memory address! precon must be disjoint. */
+
+    size_t i;
+    switch (filterType) {
+        case 0:
+            for (i = 0; i != length; ++i) recon[i] = scanline[i];
+            break;
+        case 1:
+            for (i = 0; i != bytewidth; ++i) recon[i] = scanline[i];
+            for (i = bytewidth; i < length; ++i) recon[i] = scanline[i] + recon[i - bytewidth];
+            break;
+        case 2:
+            if (precon) {
+                for(i = 0; i != length; ++i) recon[i] = scanline[i] + precon[i];
+            } else {
+                for(i = 0; i != length; ++i) recon[i] = scanline[i];
+            }
+            break;
+        case 3:
+          if (precon) {
+              for (i = 0; i != bytewidth; ++i) recon[i] = scanline[i] + (precon[i] >> 1u);
+              for (i = bytewidth; i < length; ++i) recon[i] = scanline[i] + ((recon[i - bytewidth] + precon[i]) >> 1u);
+          } else {
+              for (i = 0; i != bytewidth; ++i) recon[i] = scanline[i];
+              for (i = bytewidth; i < length; ++i) recon[i] = scanline[i] + (recon[i - bytewidth] >> 1u);
+          }
+          break;
+        case 4:
+            if (precon) {
+                for (i = 0; i != bytewidth; ++i) {
+                    recon[i] = (scanline[i] + precon[i]); /*paethPredictor(0, precon[i], 0) is always precon[i]*/
+                }
+
+                /* Unroll independent paths of the paeth predictor. A 6x and 8x version would also be possible but that
+                   adds too much code. Whether this actually speeds anything up at all depends on compiler and settings. */
+                if (bytewidth >= 4) {
+                    for (; i + 3 < length; i += 4) {
+                        size_t j = i - bytewidth;
+                        unsigned char s0 = scanline[i + 0], s1 = scanline[i + 1], s2 = scanline[i + 2], s3 = scanline[i + 3];
+                        unsigned char r0 = recon[j + 0], r1 = recon[j + 1], r2 = recon[j + 2], r3 = recon[j + 3];
+                        unsigned char p0 = precon[i + 0], p1 = precon[i + 1], p2 = precon[i + 2], p3 = precon[i + 3];
+                        unsigned char q0 = precon[j + 0], q1 = precon[j + 1], q2 = precon[j + 2], q3 = precon[j + 3];
+                        recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+                        recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+                        recon[i + 2] = s2 + paethPredictor(r2, p2, q2);
+                        recon[i + 3] = s3 + paethPredictor(r3, p3, q3);
+                    }
+                } else if (bytewidth >= 3) {
+                    for (; i + 2 < length; i += 3) {
+                        size_t j = i - bytewidth;
+                        unsigned char s0 = scanline[i + 0], s1 = scanline[i + 1], s2 = scanline[i + 2];
+                        unsigned char r0 = recon[j + 0], r1 = recon[j + 1], r2 = recon[j + 2];
+                        unsigned char p0 = precon[i + 0], p1 = precon[i + 1], p2 = precon[i + 2];
+                        unsigned char q0 = precon[j + 0], q1 = precon[j + 1], q2 = precon[j + 2];
+                        recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+                        recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+                        recon[i + 2] = s2 + paethPredictor(r2, p2, q2);
+                    }
+                } else if (bytewidth >= 2) {
+                    for (; i + 1 < length; i += 2) {
+                        size_t j = i - bytewidth;
+                        unsigned char s0 = scanline[i + 0], s1 = scanline[i + 1];
+                        unsigned char r0 = recon[j + 0], r1 = recon[j + 1];
+                        unsigned char p0 = precon[i + 0], p1 = precon[i + 1];
+                        unsigned char q0 = precon[j + 0], q1 = precon[j + 1];
+                        recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+                        recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+                    }
+                }
+
+                for (; i != length; ++i) {
+                    recon[i] = (scanline[i] + paethPredictor(recon[i - bytewidth], precon[i], precon[i - bytewidth]));
+                }
+            } else {
+                for (i = 0; i != bytewidth; ++i) {
+                    recon[i] = scanline[i];
+                }
+                for (i = bytewidth; i < length; ++i) {
+                    /* paethPredictor(recon[i - bytewidth], 0, 0) is always recon[i - bytewidth] */
+                    recon[i] = (scanline[i] + recon[i - bytewidth]);
+                }
+            }
+            break;
+        default: return 36; /* error: invalid filter type given */
+    }
+    return 0;
+}
+
+
+static unsigned unfilter(unsigned char* out, const unsigned char* in, unsigned w, unsigned h, unsigned bpp)
+{
+    /* For PNG filter method 0
+       this function unfilters a single image (e.g. without interlacing this is called once, with Adam7 seven times)
+       out must have enough bytes allocated already, in must have the scanlines + 1 filtertype byte per scanline
+       w and h are image dimensions or dimensions of reduced image, bpp is bits per pixel
+       in and out are allowed to be the same memory address (but aren't the same size since in has the extra filter bytes) */
+
+    unsigned y;
+    unsigned char* prevline = 0;
+
+    /* bytewidth is used for filtering, is 1 when bpp < 8, number of bytes per pixel otherwise */
+    size_t bytewidth = (bpp + 7u) / 8u;
+    /* the width of a scanline in bytes, not including the filter type */
+    size_t linebytes = lodepng_get_raw_size_idat(w, 1, bpp) - 1u;
+
+    for (y = 0; y < h; ++y) {
+        size_t outindex = linebytes * y;
+        size_t inindex = (1 + linebytes) * y; /* the extra filterbyte added to each row */
+        unsigned char filterType = in[inindex];
+        CERROR_TRY_RETURN(unfilterScanline(&out[outindex], &in[inindex + 1], prevline, bytewidth, filterType, linebytes));
+        prevline = &out[outindex];
+    }
+
+    return 0;
+}
+
+/* in: Adam7 interlaced image, with no padding bits between scanlines, but between
+   reduced images so that each reduced image starts at a byte.
+   out: the same pixels, but re-ordered so that they're now a non-interlaced image with size w*h
+   bpp: bits per pixel
+   out has the following size in bits: w * h * bpp.
+   in is possibly bigger due to padding bits between reduced images.
+   out must be big enough AND must be 0 everywhere if bpp < 8 in the current implementation
+   (because that's likely a little bit faster)
+   NOTE: comments about padding bits are only relevant if bpp < 8 */
+static void Adam7_deinterlace(unsigned char* out, const unsigned char* in, unsigned w, unsigned h, unsigned bpp)
+{
+    unsigned passw[7], passh[7];
+    size_t filter_passstart[8], padded_passstart[8], passstart[8];
+    unsigned i;
+
+    Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+    if (bpp >= 8) {
+        for(i = 0; i != 7; ++i) {
+            unsigned x, y, b;
+            size_t bytewidth = bpp / 8u;
+            for (y = 0; y < passh[i]; ++y)
+            for (x = 0; x < passw[i]; ++x) {
+                size_t pixelinstart = passstart[i] + (y * passw[i] + x) * bytewidth;
+                size_t pixeloutstart = ((ADAM7_IY[i] + (size_t)y * ADAM7_DY[i]) * (size_t)w + ADAM7_IX[i] + (size_t)x * ADAM7_DX[i]) * bytewidth;
+                for (b = 0; b < bytewidth; ++b) {
+                    out[pixeloutstart + b] = in[pixelinstart + b];
+                }
+            }
+        }
+    } else /* bpp < 8: Adam7 with pixels < 8 bit is a bit trickier: with bit pointers */ {
+        for (i = 0; i != 7; ++i) {
+            unsigned x, y, b;
+            unsigned ilinebits = bpp * passw[i];
+            unsigned olinebits = bpp * w;
+            size_t obp, ibp; /* bit pointers (for out and in buffer) */
+            for (y = 0; y < passh[i]; ++y)
+            for (x = 0; x < passw[i]; ++x) {
+                ibp = (8 * passstart[i]) + (y * ilinebits + x * bpp);
+                obp = (ADAM7_IY[i] + (size_t)y * ADAM7_DY[i]) * olinebits + (ADAM7_IX[i] + (size_t)x * ADAM7_DX[i]) * bpp;
+                for (b = 0; b < bpp; ++b) {
+                    unsigned char bit = readBitFromReversedStream(&ibp, in);
+                    setBitOfReversedStream(&obp, out, bit);
+                }
+            }
+        }
+    }
+}
+
+
+static void removePaddingBits(unsigned char* out, const unsigned char* in, size_t olinebits, size_t ilinebits, unsigned h)
+{
+    /* After filtering there are still padding bits if scanlines have non multiple of 8 bit amounts. They need
+       to be removed (except at last scanline of (Adam7-reduced) image) before working with pure image buffers
+       for the Adam7 code, the color convert code and the output to the user.
+       in and out are allowed to be the same buffer, in may also be higher but still overlapping; in must
+       have >= ilinebits*h bits, out must have >= olinebits*h bits, olinebits must be <= ilinebits
+       also used to move bits after earlier such operations happened, e.g. in a sequence of reduced images from Adam7
+       only useful if (ilinebits - olinebits) is a value in the range 1..7 */
+    unsigned y;
+    size_t diff = ilinebits - olinebits;
+    size_t ibp = 0, obp = 0; /*input and output bit pointers*/
+    for (y = 0; y < h; ++y) {
+        size_t x;
+        for (x = 0; x < olinebits; ++x) {
+            unsigned char bit = readBitFromReversedStream(&ibp, in);
+            setBitOfReversedStream(&obp, out, bit);
+        }
+        ibp += diff;
+    }
+}
+
+
+/* out must be buffer big enough to contain full image, and in must contain the full decompressed data from
+   the IDAT chunks (with filter index bytes and possible padding bits)
+   return value is error */
+static unsigned postProcessScanlines(unsigned char* out, unsigned char* in, unsigned w, unsigned h, const LodePNGInfo* info_png)
+{
+    /* This function converts the filtered-padded-interlaced data into pure 2D image buffer with the PNG's colortype.
+       Steps:
+       *) if no Adam7: 1) unfilter 2) remove padding bits (= possible extra bits per scanline if bpp < 8)
+       *) if adam7: 1) 7x unfilter 2) 7x remove padding bits 3) Adam7_deinterlace
+       NOTE: the in buffer will be overwritten with intermediate data! */
+    unsigned bpp = lodepng_get_bpp_lct(info_png->color.colortype, info_png->color.bitdepth);
+    if (bpp == 0) return 31; /* error: invalid colortype */
+
+    if (info_png->interlace_method == 0) {
+        if (bpp < 8 && w * bpp != ((w * bpp + 7u) / 8u) * 8u) {
+            CERROR_TRY_RETURN(unfilter(in, in, w, h, bpp));
+            removePaddingBits(out, in, w * bpp, ((w * bpp + 7u) / 8u) * 8u, h);
+        }
+        /* we can immediately filter into the out buffer, no other steps needed */
+        else CERROR_TRY_RETURN(unfilter(out, in, w, h, bpp));
+    } else /* interlace_method is 1 (Adam7) */ {
+        unsigned passw[7], passh[7]; size_t filter_passstart[8], padded_passstart[8], passstart[8];
+        unsigned i;
+
+        Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+        for (i = 0; i != 7; ++i) {
+            CERROR_TRY_RETURN(unfilter(&in[padded_passstart[i]], &in[filter_passstart[i]], passw[i], passh[i], bpp));
+            /* TODO: possible efficiency improvement: if in this reduced image the bits fit nicely in 1 scanline,
+               move bytes instead of bits or move not at all */
+            if (bpp < 8) {
+              /* remove padding bits in scanlines; after this there still may be padding
+                 bits between the different reduced images: each reduced image still starts nicely at a byte */
+              removePaddingBits(&in[passstart[i]], &in[padded_passstart[i]], passw[i] * bpp, ((passw[i] * bpp + 7u) / 8u) * 8u, passh[i]);
+            }
+        }
+        Adam7_deinterlace(out, in, w, h, bpp);
+    }
+    return 0;
+}
+
+
+static unsigned readChunk_PLTE(LodePNGColorMode* color, const unsigned char* data, size_t chunkLength)
+{
+    unsigned pos = 0, i;
+    color->palettesize = chunkLength / 3u;
+    if (color->palettesize == 0 || color->palettesize > 256) return 38; /* error: palette too small or big */
+    lodepng_color_mode_alloc_palette(color);
+    if (!color->palette && color->palettesize) {
+        color->palettesize = 0;
+        return 83; /* alloc fail */
+    }
+
+    for (i = 0; i != color->palettesize; ++i) {
+        color->palette[4 * i + 0] = data[pos++]; /*R*/
+        color->palette[4 * i + 1] = data[pos++]; /*G*/
+        color->palette[4 * i + 2] = data[pos++]; /*B*/
+        color->palette[4 * i + 3] = 255; /*alpha*/
+    }
+
+    return 0; /* OK */
+}
+
+
+static unsigned readChunk_tRNS(LodePNGColorMode* color, const unsigned char* data, size_t chunkLength)
+{
+    unsigned i;
+    if (color->colortype == LCT_PALETTE) {
+        /* error: more alpha values given than there are palette entries */
+        if (chunkLength > color->palettesize) return 39;
+
+        for (i = 0; i != chunkLength; ++i) color->palette[4 * i + 3] = data[i];
+    } else if (color->colortype == LCT_GREY) {
+        /* error: this chunk must be 2 bytes for grayscale image */
+        if (chunkLength != 2) return 30;
+
+        color->key_defined = 1;
+        color->key_r = color->key_g = color->key_b = 256u * data[0] + data[1];
+    } else if (color->colortype == LCT_RGB) {
+        /* error: this chunk must be 6 bytes for RGB image */
+        if (chunkLength != 6) return 41;
+
+        color->key_defined = 1;
+        color->key_r = 256u * data[0] + data[1];
+        color->key_g = 256u * data[2] + data[3];
+        color->key_b = 256u * data[4] + data[5];
+    }
+    else return 42; /* error: tRNS chunk not allowed for other color models */
+
+    return 0; /* OK */
+}
+
+
+/* read a PNG, the result will be in the same color type as the PNG (hence "generic") */
+static void decodeGeneric(unsigned char** out, unsigned* w, unsigned* h, LodePNGState* state, const unsigned char* in, size_t insize)
+{
+    unsigned char IEND = 0;
+    const unsigned char* chunk;
+    unsigned char* idat; /*the data from idat chunks, zlib compressed*/
+    size_t idatsize = 0;
+    unsigned char* scanlines = 0;
+    size_t scanlines_size = 0, expected_size = 0;
+    size_t outsize = 0;
+
+    /* safe output values in case error happens */
+    *out = 0;
+    *w = *h = 0;
+
+    state->error = lodepng_inspect(w, h, state, in, insize); /*reads header and resets other parameters in state->info_png*/
+    if (state->error) return;
+
+    if (lodepng_pixel_overflow(*w, *h, &state->info_png.color, &state->info_raw)) {
+        CERROR_RETURN(state->error, 92); /*overflow possible due to amount of pixels*/
+    }
+
+    /*the input filesize is a safe upper bound for the sum of idat chunks size*/
+    idat = (unsigned char*)malloc(insize);
+    if (!idat) CERROR_RETURN(state->error, 83); /*alloc fail*/
+
+    chunk = &in[33]; /*first byte of the first chunk after the header*/
+
+    /*loop through the chunks, ignoring unknown chunks and stopping at IEND chunk.
+    IDAT data is put at the start of the in buffer*/
+    while (!IEND && !state->error) {
+        unsigned chunkLength;
+        const unsigned char* data; /*the data in the chunk*/
+
+        /*error: size of the in buffer too small to contain next chunk*/
+        if ((size_t)((chunk - in) + 12) > insize || chunk < in) {
+            if (state->decoder.ignore_end) break; /*other errors may still happen though*/
+            CERROR_BREAK(state->error, 30);
+        }
+
+        /*length of the data of the chunk, excluding the length bytes, chunk type and CRC bytes*/
+        chunkLength = lodepng_chunk_length(chunk);
+        /*error: chunk length larger than the max PNG chunk size*/
+        if (chunkLength > 2147483647) {
+            if (state->decoder.ignore_end) break; /*other errors may still happen though*/
+            CERROR_BREAK(state->error, 63);
+        }
+
+        if ((size_t)((chunk - in) + chunkLength + 12) > insize || (chunk + chunkLength + 12) < in) {
+            CERROR_BREAK(state->error, 64); /*error: size of the in buffer too small to contain next chunk*/
+        }
+
+        data = lodepng_chunk_data_const(chunk);
+
+        /*for unknown chunk order*/
+        //unsigned unknown = 0;
+
+        /*IDAT chunk, containing compressed image data*/
+        if (lodepng_chunk_type_equals(chunk, "IDAT")) {
+            size_t newsize;
+            if (lodepng_addofl(idatsize, chunkLength, &newsize)) CERROR_BREAK(state->error, 95);
+            if (newsize > insize) CERROR_BREAK(state->error, 95);
+            lodepng_memcpy(idat + idatsize, data, chunkLength);
+            idatsize += chunkLength;
+        } else if (lodepng_chunk_type_equals(chunk, "IEND")) {
+            /*IEND chunk*/
+            IEND = 1;
+        } else if (lodepng_chunk_type_equals(chunk, "PLTE")) {
+            /*palette chunk (PLTE)*/
+            state->error = readChunk_PLTE(&state->info_png.color, data, chunkLength);
+            if (state->error) break;
+        } else if (lodepng_chunk_type_equals(chunk, "tRNS")) {
+            /*palette transparency chunk (tRNS). Even though this one is an ancillary chunk , it is still compiled
+            in without 'LODEPNG_COMPILE_ANCILLARY_CHUNKS' because it contains essential color information that
+            affects the alpha channel of pixels. */
+            state->error = readChunk_tRNS(&state->info_png.color, data, chunkLength);
+            if (state->error) break;
+        } else /*it's not an implemented chunk type, so ignore it: skip over the data*/ {
+            /*error: unknown critical chunk (5th bit of first byte of chunk type is 0)*/
+            if (!state->decoder.ignore_critical && !lodepng_chunk_ancillary(chunk)) {
+                CERROR_BREAK(state->error, 69);
+            }
+            //unknown = 1;
+        }
+
+#if 0 //We don't use CRC
+        if (!state->decoder.ignore_crc && !unknown) /*check CRC if wanted, only on known chunk types*/ {
+            if (lodepng_chunk_check_crc(chunk)) CERROR_BREAK(state->error, 57); /*invalid CRC*/
+        }
+#endif
+        if (!IEND) chunk = lodepng_chunk_next_const(chunk, in + insize);
+    }
+
+    if (state->info_png.color.colortype == LCT_PALETTE && !state->info_png.color.palette) {
+        state->error = 106; /* error: PNG file must have PLTE chunk if color type is palette */
+    }
+
+    if (!state->error) {
+        /*predict output size, to allocate exact size for output buffer to avoid more dynamic allocation.
+        If the decompressed size does not match the prediction, the image must be corrupt.*/
+        if (state->info_png.interlace_method == 0) {
+            size_t bpp = lodepng_get_bpp_lct(state->info_png.color.colortype, state->info_png.color.bitdepth);
+            expected_size = lodepng_get_raw_size_idat(*w, *h, bpp);
+        } else {
+            size_t bpp = lodepng_get_bpp_lct(state->info_png.color.colortype, state->info_png.color.bitdepth);
+            /*Adam-7 interlaced: expected size is the sum of the 7 sub-images sizes*/
+            expected_size = 0;
+            expected_size += lodepng_get_raw_size_idat((*w + 7) >> 3, (*h + 7) >> 3, bpp);
+            if (*w > 4) expected_size += lodepng_get_raw_size_idat((*w + 3) >> 3, (*h + 7) >> 3, bpp);
+            expected_size += lodepng_get_raw_size_idat((*w + 3) >> 2, (*h + 3) >> 3, bpp);
+            if (*w > 2) expected_size += lodepng_get_raw_size_idat((*w + 1) >> 2, (*h + 3) >> 2, bpp);
+            expected_size += lodepng_get_raw_size_idat((*w + 1) >> 1, (*h + 1) >> 2, bpp);
+            if (*w > 1) expected_size += lodepng_get_raw_size_idat((*w + 0) >> 1, (*h + 1) >> 1, bpp);
+            expected_size += lodepng_get_raw_size_idat((*w + 0), (*h + 0) >> 1, bpp);
+        }
+        state->error = zlib_decompress(&scanlines, &scanlines_size, expected_size, idat, idatsize, &state->decoder.zlibsettings);
+    }
+
+    if (!state->error && scanlines_size != expected_size) state->error = 91; /*decompressed size doesn't match prediction*/
+    free(idat);
+
+    if (!state->error) {
+        outsize = lodepng_get_raw_size(*w, *h, &state->info_png.color);
+        *out = (unsigned char*)malloc(outsize);
+        if (!*out) state->error = 83; /*alloc fail*/
+    }
+    if (!state->error) {
+        lodepng_memset(*out, 0, outsize);
+        state->error = postProcessScanlines(*out, scanlines, *w, *h, &state->info_png);
+    }
+    free(scanlines);
+}
+
+
+static void lodepng_decoder_settings_init(LodePNGDecoderSettings* settings)
+{
+    settings->color_convert = 1;
+    settings->ignore_crc = 0;
+    settings->ignore_critical = 0;
+    settings->ignore_end = 0;
+    lodepng_decompress_settings_init(&settings->zlibsettings);
+}
+
+
+/************************************************************************/
+/* External Class Implementation                                        */
+/************************************************************************/
+
+/*read the information from the header and store it in the LodePNGInfo. return value is error*/
+unsigned lodepng_inspect(unsigned* w, unsigned* h, LodePNGState* state, const unsigned char* in, size_t insize)
+{
+    unsigned width, height;
+    LodePNGInfo* info = &state->info_png;
+    if (insize == 0 || in == 0) {
+        CERROR_RETURN_ERROR(state->error, 48); /*error: the given data is empty*/
+    }
+    if (insize < 33) {
+        CERROR_RETURN_ERROR(state->error, 27); /*error: the data length is smaller than the length of a PNG header*/
+    }
+
+    /* when decoding a new PNG image, make sure all parameters created after previous decoding are reset */
+    /* TODO: remove this. One should use a new LodePNGState for new sessions */
+    lodepng_info_cleanup(info);
+    lodepng_info_init(info);
+
+    if (in[0] != 137 || in[1] != 80 || in[2] != 78 || in[3] != 71 || in[4] != 13 || in[5] != 10 || in[6] != 26 || in[7] != 10) {
+        CERROR_RETURN_ERROR(state->error, 28); /*error: the first 8 bytes are not the correct PNG signature*/
+    }
+    if (lodepng_chunk_length(in + 8) != 13) {
+        CERROR_RETURN_ERROR(state->error, 94); /*error: header size must be 13 bytes*/
+    }
+    if (!lodepng_chunk_type_equals(in + 8, "IHDR")) {
+        CERROR_RETURN_ERROR(state->error, 29); /*error: it doesn't start with a IHDR chunk!*/
+    }
+
+    /*read the values given in the header*/
+    width = lodepng_read32bitInt(&in[16]);
+    height = lodepng_read32bitInt(&in[20]);
+    /*TODO: remove the undocumented feature that allows to give null pointers to width or height*/
+    if (w) *w = width;
+    if (h) *h = height;
+    info->color.bitdepth = in[24];
+    info->color.colortype = (LodePNGColorType)in[25];
+    info->compression_method = in[26];
+    info->filter_method = in[27];
+    info->interlace_method = in[28];
+
+    /*errors returned only after the parsing so other values are still output*/
+
+    /*error: invalid image size*/
+    if (width == 0 || height == 0) CERROR_RETURN_ERROR(state->error, 93);
+    /*error: invalid colortype or bitdepth combination*/
+    state->error = checkColorValidity(info->color.colortype, info->color.bitdepth);
+    if (state->error) return state->error;
+    /*error: only compression method 0 is allowed in the specification*/
+    if (info->compression_method != 0) CERROR_RETURN_ERROR(state->error, 32);
+    /*error: only filter method 0 is allowed in the specification*/
+    if (info->filter_method != 0) CERROR_RETURN_ERROR(state->error, 33);
+    /*error: only interlace methods 0 and 1 exist in the specification*/
+    if (info->interlace_method > 1) CERROR_RETURN_ERROR(state->error, 34);
+
+#if 0 //thorvg don't use crc
+    if (!state->decoder.ignore_crc) {
+        unsigned CRC = lodepng_read32bitInt(&in[29]);
+        unsigned checksum = lodepng_crc32(&in[12], 17);
+        if (CRC != checksum) {
+          CERROR_RETURN_ERROR(state->error, 57); /*invalid CRC*/
+        }
+    }
+#endif
+    return state->error;
+}
+
+
+unsigned lodepng_decode(unsigned char** out, unsigned* w, unsigned* h, LodePNGState* state, const unsigned char* in, size_t insize)
+{
+    *out = 0;
+    decodeGeneric(out, w, h, state, in, insize);
+    if (state->error) return state->error;
+    if (!state->decoder.color_convert || lodepng_color_mode_equal(&state->info_raw, &state->info_png.color)) {
+        /*same color type, no copying or converting of data needed*/
+        /*store the info_png color settings on the info_raw so that the info_raw still reflects what colortype
+        the raw image has to the end user*/
+        if (!state->decoder.color_convert) {
+            state->error = lodepng_color_mode_copy(&state->info_raw, &state->info_png.color);
+            if (state->error) return state->error;
+        }
+    } else { /*color conversion needed*/
+        unsigned char* data = *out;
+        size_t outsize;
+
+        /*TODO: check if this works according to the statement in the documentation: "The converter can convert
+        from grayscale input color type, to 8-bit grayscale or grayscale with alpha"*/
+        if (!(state->info_raw.colortype == LCT_RGB || state->info_raw.colortype == LCT_RGBA) && !(state->info_raw.bitdepth == 8)) {
+            return 56; /*unsupported color mode conversion*/
+        }
+
+        outsize = lodepng_get_raw_size(*w, *h, &state->info_raw);
+        *out = (unsigned char*)malloc(outsize);
+        if (!(*out)) {
+            state->error = 83; /*alloc fail*/
+        }
+        else state->error = lodepng_convert(*out, data, &state->info_raw, &state->info_png.color, *w, *h);
+        free(data);
+    }
+    return state->error;
+}
+
+
+void lodepng_state_init(LodePNGState* state)
+{
+    lodepng_decoder_settings_init(&state->decoder);
+    lodepng_color_mode_init(&state->info_raw);
+    lodepng_info_init(&state->info_png);
+    state->error = 1;
+}
+
+
+void lodepng_state_cleanup(LodePNGState* state)
+{
+    lodepng_color_mode_cleanup(&state->info_raw);
+    lodepng_info_cleanup(&state->info_png);
+}

+ 174 - 0
thirdparty/thorvg/src/loaders/png/tvgLodePng.h

@@ -0,0 +1,174 @@
+/*
+ * Copyright (c) 2020 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+/*
+  LodePNG version 20200306
+
+  Copyright (c) 2005-2020 Lode Vandevenne
+
+  This software is provided 'as-is', without any express or implied
+  warranty. In no event will the authors be held liable for any damages
+  arising from the use of this software.
+
+  Permission is granted to anyone to use this software for any purpose,
+  including commercial applications, and to alter it and redistribute it
+  freely, subject to the following restrictions:
+
+      1. The origin of this software must not be misrepresented; you must not
+      claim that you wrote the original software. If you use this software
+      in a product, an acknowledgment in the product documentation would be
+      appreciated but is not required.
+
+      2. Altered source versions must be plainly marked as such, and must not be
+      misrepresented as being the original software.
+
+      3. This notice may not be removed or altered from any source
+      distribution.
+*/
+
+#ifndef _TVG_LODEPNG_H_
+#define _TVG_LODEPNG_H_
+
+#include <stddef.h>
+
+/*The PNG color types (also used for raw image).*/
+enum LodePNGColorType
+{
+    LCT_GREY = 0, /*grayscale: 1,2,4,8,16 bit*/
+    LCT_RGB = 2, /*RGB: 8,16 bit*/
+    LCT_PALETTE = 3, /*palette: 1,2,4,8 bit*/
+    LCT_GREY_ALPHA = 4, /*grayscale with alpha: 8,16 bit*/
+    LCT_RGBA = 6, /*RGB with alpha: 8,16 bit*/
+    /*LCT_MAX_OCTET_VALUE lets the compiler allow this enum to represent any invalid
+    byte value from 0 to 255 that could be present in an invalid PNG file header. Do
+    not use, compare with or set the name LCT_MAX_OCTET_VALUE, instead either use
+    the valid color type names above, or numeric values like 1 or 7 when checking for
+    particular disallowed color type byte values, or cast to integer to print it.*/
+    LCT_MAX_OCTET_VALUE = 255
+};
+
+/*Settings for zlib decompression*/
+struct LodePNGDecompressSettings
+{
+    /* Check LodePNGDecoderSettings for more ignorable errors such as ignore_crc */
+    unsigned ignore_adler32; /*if 1, continue and don't give an error message if the Adler32 checksum is corrupted*/
+    unsigned ignore_nlen; /*ignore complement of len checksum in uncompressed blocks*/
+
+    /*use custom zlib decoder instead of built in one (default: null)*/
+    unsigned (*custom_zlib)(unsigned char**, size_t*, const unsigned char*, size_t, const LodePNGDecompressSettings*);
+    /*use custom deflate decoder instead of built in one (default: null) if custom_zlib is not null, custom_inflate is ignored (the zlib format uses deflate)*/
+    unsigned (*custom_inflate)(unsigned char**, size_t*, const unsigned char*, size_t, const LodePNGDecompressSettings*);
+
+    const void* custom_context; /*optional custom settings for custom functions*/
+};
+
+/*
+  Color mode of an image. Contains all information required to decode the pixel
+  bits to RGBA colors. This information is the same as used in the PNG file
+  format, and is used both for PNG and raw image data in LodePNG.
+*/
+struct LodePNGColorMode
+{
+    /*header (IHDR)*/
+    LodePNGColorType colortype; /*color type, see PNG standard or documentation further in this header file*/
+    unsigned bitdepth;  /*bits per sample, see PNG standard or documentation further in this header file*/
+
+    /*
+      palette (PLTE and tRNS)
+
+      Dynamically allocated with the colors of the palette, including alpha.
+      This field may not be allocated directly, use lodepng_color_mode_init first,
+      then lodepng_palette_add per color to correctly initialize it (to ensure size
+      of exactly 1024 bytes).
+
+      The alpha channels must be set as well, set them to 255 for opaque images.
+
+      When decoding, by default you can ignore this palette, since LodePNG already
+      fills the palette colors in the pixels of the raw RGBA output.
+
+      The palette is only supported for color type 3.
+    */
+    unsigned char* palette; /*palette in RGBARGBA... order. Must be either 0, or when allocated must have 1024 bytes*/
+    size_t palettesize; /*palette size in number of colors (amount of used bytes is 4 * palettesize)*/
+
+    /*
+      transparent color key (tRNS)
+
+      This color uses the same bit depth as the bitdepth value in this struct, which can be 1-bit to 16-bit.
+      For grayscale PNGs, r, g and b will all 3 be set to the same.
+
+      When decoding, by default you can ignore this information, since LodePNG sets
+      pixels with this key to transparent already in the raw RGBA output.
+
+      The color key is only supported for color types 0 and 2.
+    */
+    unsigned key_defined; /*is a transparent color key given? 0 = false, 1 = true*/
+    unsigned key_r;       /*red/grayscale component of color key*/
+    unsigned key_g;       /*green component of color key*/
+    unsigned key_b;       /*blue component of color key*/
+};
+
+/*Information about the PNG image, except pixels, width and height.*/
+struct LodePNGInfo
+{
+    /*header (IHDR), palette (PLTE) and transparency (tRNS) chunks*/
+    unsigned compression_method;/*compression method of the original file. Always 0.*/
+    unsigned filter_method;     /*filter method of the original file*/
+    unsigned interlace_method;  /*interlace method of the original file: 0=none, 1=Adam7*/
+    LodePNGColorMode color;     /*color type and bits, palette and transparency of the PNG file*/
+};
+
+/*
+  Settings for the decoder. This contains settings for the PNG and the Zlib
+  decoder, but not the Info settings from the Info structs.
+*/
+struct LodePNGDecoderSettings
+{
+    LodePNGDecompressSettings zlibsettings; /*in here is the setting to ignore Adler32 checksums*/
+
+    /* Check LodePNGDecompressSettings for more ignorable errors such as ignore_adler32 */
+    unsigned ignore_crc; /*ignore CRC checksums*/
+    unsigned ignore_critical; /*ignore unknown critical chunks*/
+    unsigned ignore_end; /*ignore issues at end of file if possible (missing IEND chunk, too large chunk, ...)*/
+    /* TODO: make a system involving warnings with levels and a strict mode instead. Other potentially recoverable
+       errors: srgb rendering intent value, size of content of ancillary chunks, more than 79 characters for some
+       strings, placement/combination rules for ancillary chunks, crc of unknown chunks, allowed characters
+       in string keys, etc... */
+
+    unsigned color_convert; /*whether to convert the PNG to the color type you want. Default: yes*/
+};
+
+/*The settings, state and information for extended encoding and decoding.*/
+struct LodePNGState
+{
+    LodePNGDecoderSettings decoder; /*the decoding settings*/
+    LodePNGColorMode info_raw; /*specifies the format in which you would like to get the raw pixel buffer*/
+    LodePNGInfo info_png; /*info of the PNG image obtained after decoding*/
+    unsigned error;
+};
+
+void lodepng_state_init(LodePNGState* state);
+void lodepng_state_cleanup(LodePNGState* state);
+unsigned lodepng_decode(unsigned char** out, unsigned* w, unsigned* h, LodePNGState* state, const unsigned char* in, size_t insize);
+unsigned lodepng_inspect(unsigned* w, unsigned* h, LodePNGState* state, const unsigned char* in, size_t insize);
+
+#endif //_TVG_LODEPNG_H_

+ 185 - 0
thirdparty/thorvg/src/loaders/png/tvgPngLoader.cpp

@@ -0,0 +1,185 @@
+/*
+ * Copyright (c) 2021 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+#include <memory.h>
+#include "tvgLoader.h"
+#include "tvgPngLoader.h"
+
+
+/************************************************************************/
+/* Internal Class Implementation                                        */
+/************************************************************************/
+
+void PngLoader::clear()
+{
+    lodepng_state_cleanup(&state);
+
+    if (freeData) free(data);
+    data = nullptr;
+    size = 0;
+    freeData = false;
+}
+
+
+/************************************************************************/
+/* External Class Implementation                                        */
+/************************************************************************/
+
+PngLoader::PngLoader()
+{
+    lodepng_state_init(&state);
+}
+
+
+PngLoader::~PngLoader()
+{
+    if (freeData) free(data);
+    free(image);
+}
+
+
+bool PngLoader::open(const string& path)
+{
+    clear();
+
+    auto pngFile = fopen(path.c_str(), "rb");
+    if (!pngFile) return false;
+
+    auto ret = false;
+
+    //determine size
+    if (fseek(pngFile, 0, SEEK_END) < 0) goto finalize;
+    if (((size = ftell(pngFile)) < 1)) goto finalize;
+    if (fseek(pngFile, 0, SEEK_SET)) goto finalize;
+
+    data = (unsigned char *) malloc(size);
+    if (!data) goto finalize;
+
+    freeData = true;
+
+    if (fread(data, size, 1, pngFile) < 1) goto failure;
+
+    lodepng_state_init(&state);
+
+    unsigned int width, height;
+    if (lodepng_inspect(&width, &height, &state, data, size) > 0) goto failure;
+
+    w = static_cast<float>(width);
+    h = static_cast<float>(height);
+
+    if (state.info_png.color.colortype == LCT_RGBA) cs = ColorSpace::ABGR8888;
+    else cs = ColorSpace::ARGB8888;
+
+    ret = true;
+
+    goto finalize;
+
+failure:
+    clear();
+
+finalize:
+    fclose(pngFile);
+    return ret;
+}
+
+
+bool PngLoader::open(const char* data, uint32_t size, bool copy)
+{
+    clear();
+
+    lodepng_state_init(&state);
+
+    unsigned int width, height;
+    if (lodepng_inspect(&width, &height, &state, (unsigned char*)(data), size) > 0) return false;
+
+    if (copy) {
+        this->data = (unsigned char *) malloc(size);
+        if (!this->data) return false;
+        memcpy((unsigned char *)this->data, data, size);
+        freeData = true;
+    } else {
+        this->data = (unsigned char *) data;
+        freeData = false;
+    }
+
+    w = static_cast<float>(width);
+    h = static_cast<float>(height);
+    this->size = size;
+
+    cs = ColorSpace::ABGR8888;
+
+    return true;
+}
+
+
+bool PngLoader::read()
+{
+    if (!data || w <= 0 || h <= 0) return false;
+
+    TaskScheduler::request(this);
+
+    return true;
+}
+
+
+bool PngLoader::close()
+{
+    this->done();
+    clear();
+    return true;
+}
+
+
+unique_ptr<Surface> PngLoader::bitmap()
+{
+    this->done();
+
+    if (!image) return nullptr;
+
+    //TODO: It's better to keep this surface instance in the loader side
+    auto surface = new Surface;
+    surface->buf8 = image;
+    surface->stride = static_cast<uint32_t>(w);
+    surface->w = static_cast<uint32_t>(w);
+    surface->h = static_cast<uint32_t>(h);
+    surface->cs = cs;
+    surface->channelSize = sizeof(uint32_t);
+    surface->premultiplied = false;
+    surface->owner = true;
+
+    return unique_ptr<Surface>(surface);
+}
+
+
+void PngLoader::run(unsigned tid)
+{
+    if (image) {
+        free(image);
+        image = nullptr;
+    }
+    auto width = static_cast<unsigned>(w);
+    auto height = static_cast<unsigned>(h);
+
+    if (lodepng_decode(&image, &width, &height, &state, data, size)) {
+        TVGERR("PNG", "Failed to decode image");
+    }
+}

+ 55 - 0
thirdparty/thorvg/src/loaders/png/tvgPngLoader.h

@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2021 - 2023 the ThorVG project. All rights reserved.
+
+ * 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.
+ */
+
+#ifndef _TVG_PNG_LOADER_H_
+#define _TVG_PNG_LOADER_H_
+
+#include "tvgLodePng.h"
+#include "tvgTaskScheduler.h"
+
+
+class PngLoader : public LoadModule, public Task
+{
+private:
+    LodePNGState state;
+    unsigned char* data = nullptr;
+    unsigned char *image = nullptr;
+    unsigned long size = 0;
+    bool freeData = false;
+
+    void clear();
+
+public:
+    PngLoader();
+    ~PngLoader();
+
+    using LoadModule::open;
+    bool open(const string& path) override;
+    bool open(const char* data, uint32_t size, bool copy) override;
+    bool read() override;
+    bool close() override;
+
+    unique_ptr<Surface> bitmap() override;
+    void run(unsigned tid) override;
+};
+
+#endif //_TVG_PNG_LOADER_H_

+ 6 - 2
thirdparty/thorvg/src/loaders/svg/tvgSvgSceneBuilder.cpp

@@ -284,8 +284,12 @@ static void _applyComposition(SvgLoaderData& loaderData, Paint* paint, const Svg
 
             bool isMaskWhite = true;
             if (auto comp = _sceneBuildHelper(loaderData, compNode, vBox, svgPath, true, 0, &isMaskWhite)) {
-                Matrix finalTransform = _compositionTransform(paint, node, compNode, SvgNodeType::Mask);
-                comp->transform(finalTransform);
+                if (!compNode->node.mask.userSpace) {
+                    Matrix finalTransform = _compositionTransform(paint, node, compNode, SvgNodeType::Mask);
+                    comp->transform(finalTransform);
+                } else {
+                    if (node->transform) comp->transform(*node->transform);
+                }
 
                 if (compNode->node.mask.type == SvgMaskType::Luminance && !isMaskWhite) {
                     paint->composite(std::move(comp), CompositeMethod::LumaMask);

+ 1 - 0
thirdparty/thorvg/src/renderer/sw_engine/tvgSwCommon.h

@@ -205,6 +205,7 @@ struct SwDashStroke
     float* pattern = nullptr;
     uint32_t cnt = 0;
     bool curOpGap = false;
+    bool move = true;
 };
 
 struct SwShape

+ 42 - 16
thirdparty/thorvg/src/renderer/sw_engine/tvgSwShape.cpp

@@ -121,7 +121,10 @@ static void _dashLineTo(SwDashStroke& dash, const Point* to, const Matrix* trans
     if (len < dash.curLen) {
         dash.curLen -= len;
         if (!dash.curOpGap) {
-            _outlineMoveTo(*dash.outline, &dash.ptCur, transform);
+            if (dash.move) {
+                _outlineMoveTo(*dash.outline, &dash.ptCur, transform);
+                dash.move = false;
+            }
             _outlineLineTo(*dash.outline, to, transform);
         }
     } else {
@@ -131,7 +134,10 @@ static void _dashLineTo(SwDashStroke& dash, const Point* to, const Matrix* trans
                 len -= dash.curLen;
                 _lineSplitAt(cur, dash.curLen, left, right);
                 if (!dash.curOpGap) {
-                    _outlineMoveTo(*dash.outline, &left.pt1, transform);
+                    if (dash.move || dash.pattern[dash.curIdx] - dash.curLen < FLT_EPSILON) {
+                        _outlineMoveTo(*dash.outline, &left.pt1, transform);
+                        dash.move = false;
+                    }
                     _outlineLineTo(*dash.outline, &left.pt2, transform);
                 }
             } else {
@@ -142,11 +148,15 @@ static void _dashLineTo(SwDashStroke& dash, const Point* to, const Matrix* trans
             dash.curOpGap = !dash.curOpGap;
             cur = right;
             dash.ptCur = cur.pt1;
+            dash.move = true;
         }
         //leftovers
         dash.curLen -= len;
         if (!dash.curOpGap) {
-            _outlineMoveTo(*dash.outline, &cur.pt1, transform);
+            if (dash.move) {
+                _outlineMoveTo(*dash.outline, &cur.pt1, transform);
+                dash.move = false;
+            }
             _outlineLineTo(*dash.outline, &cur.pt2, transform);
         }
         if (dash.curLen < 1 && TO_SWCOORD(len) > 1) {
@@ -168,21 +178,22 @@ static void _dashCubicTo(SwDashStroke& dash, const Point* ctrl1, const Point* ct
     if (len < dash.curLen) {
         dash.curLen -= len;
         if (!dash.curOpGap) {
-            _outlineMoveTo(*dash.outline, &dash.ptCur, transform);
+            if (dash.move) {
+                _outlineMoveTo(*dash.outline, &dash.ptCur, transform);
+                dash.move = false;
+            }
             _outlineCubicTo(*dash.outline, ctrl1, ctrl2, to, transform);
         }
     } else {
-        bool begin = true;          //starting with move_to
         while (len > dash.curLen) {
             Bezier left, right;
             if (dash.curLen > 0) {
                 len -= dash.curLen;
                 bezSplitAt(cur, dash.curLen, left, right);
                 if (!dash.curOpGap) {
-                    // leftovers from a previous command don't require moveTo
-                    if (begin || dash.pattern[dash.curIdx] - dash.curLen < FLT_EPSILON) {
+                    if (dash.move || dash.pattern[dash.curIdx] - dash.curLen < FLT_EPSILON) {
                         _outlineMoveTo(*dash.outline, &left.start, transform);
-                        begin = false;
+                        dash.move = false;
                     }
                     _outlineCubicTo(*dash.outline, &left.ctrl1, &left.ctrl2, &left.end, transform);
                 }
@@ -194,11 +205,15 @@ static void _dashCubicTo(SwDashStroke& dash, const Point* ctrl1, const Point* ct
             dash.curOpGap = !dash.curOpGap;
             cur = right;
             dash.ptCur = right.start;
+            dash.move = true;
         }
         //leftovers
         dash.curLen -= len;
         if (!dash.curOpGap) {
-            _outlineMoveTo(*dash.outline, &cur.start, transform);
+            if (dash.move) {
+                _outlineMoveTo(*dash.outline, &cur.start, transform);
+                dash.move = false;
+            }
             _outlineCubicTo(*dash.outline, &cur.ctrl1, &cur.ctrl2, &cur.end, transform);
         }
         if (dash.curLen < 1 && TO_SWCOORD(len) > 1) {
@@ -212,6 +227,22 @@ static void _dashCubicTo(SwDashStroke& dash, const Point* ctrl1, const Point* ct
 }
 
 
+static void _dashClose(SwDashStroke& dash, const Matrix* transform)
+{
+    _dashLineTo(dash, &dash.ptStart, transform);
+}
+
+
+static void _dashMoveTo(SwDashStroke& dash, uint32_t offIdx, float offset, const Point* pts, const Matrix* transform)
+{
+    dash.curIdx = offIdx % dash.cnt;
+    dash.curLen = dash.pattern[dash.curIdx] - offset;
+    dash.curOpGap = offIdx % 2;
+    dash.ptStart = dash.ptCur = *pts;
+    dash.move = true;
+}
+
+
 static SwOutline* _genDashOutline(const RenderShape* rshape, const Matrix* transform, float length, SwMpool* mpool, unsigned tid)
 {
     const PathCommand* cmds = rshape->path.cmds.data;
@@ -261,7 +292,6 @@ static SwOutline* _genDashOutline(const RenderShape* rshape, const Matrix* trans
         trimmed = true;
     //just a dasy style.
     } else {
-
         if (dash.cnt == 0) return nullptr;
     }
 
@@ -303,15 +333,11 @@ static SwOutline* _genDashOutline(const RenderShape* rshape, const Matrix* trans
     while (cmdCnt-- > 0) {
         switch (*cmds) {
             case PathCommand::Close: {
-                _dashLineTo(dash, &dash.ptStart, transform);
+                _dashClose(dash, transform);
                 break;
             }
             case PathCommand::MoveTo: {
-                //reset the dash
-                dash.curIdx = offIdx % dash.cnt;
-                dash.curLen = dash.pattern[dash.curIdx] - offset;
-                dash.curOpGap = offIdx % 2;
-                dash.ptStart = dash.ptCur = *pts;
+                _dashMoveTo(dash, offIdx, offset, pts, transform);
                 ++pts;
                 break;
             }

+ 0 - 1
thirdparty/thorvg/src/renderer/sw_engine/tvgSwStroke.cpp

@@ -689,7 +689,6 @@ static void _endSubPath(SwStroke& stroke)
 
         //No specific corner processing is required if the turn is 0
         if (turn != 0) {
-
             //when we turn to the right, the inside is 0
             int32_t inside = 0;
 

+ 7 - 2
thirdparty/thorvg/update-thorvg.sh

@@ -1,6 +1,6 @@
 #!/bin/bash -e
 
-VERSION=0.11.0
+VERSION=0.11.1
 
 cd thirdparty/thorvg/ || true
 rm -rf AUTHORS LICENSE inc/ src/ *.zip *.tar.gz tmp/
@@ -36,6 +36,8 @@ cat << EOF > ../inc/config.h
 
 #define THORVG_SW_RASTER_SUPPORT
 #define THORVG_SVG_LOADER_SUPPORT
+#define THORVG_PNG_LOADER_SUPPORT
+#define THORVG_JPG_LOADER_SUPPORT
 
 // For internal debugging:
 //#define THORVG_LOG_ENABLED
@@ -51,9 +53,12 @@ cp -rv src/renderer ../src/
 # Only sw_engine is enabled.
 rm -rfv ../src/renderer/gl_engine
 
-# Only svg (+raw) loader is enabled.
+# Enabled embedded loaders: raw, JPEG, PNG.
 mkdir ../src/loaders
 cp -rv src/loaders/svg src/loaders/raw  ../src/loaders/
+cp -rv src/loaders/svg src/loaders/jpg  ../src/loaders/
+cp -rv src/loaders/svg src/loaders/png  ../src/loaders/
+cp -rv src/loaders/svg src/loaders/external_png  ../src/loaders/
 
 popd
 rm -rf tmp