Merge PVRTC compressor into split library.

BPTCEncoder/CMakeLists.txt

@@ -150,4 +150,4 @@ ADD_LIBRARY( BPTCEncoder
   ${SOURCES}
 )
 
-TARGET_LINK_LIBRARIES( BPTCEncoder FasTCBase )
+TARGET_LINK_LIBRARIES( BPTCEncoder FasTCBase )

Base/include/Image.h

@@ -49,17 +49,29 @@
 class Image {
 
  public:
-  Image(uint32 width, uint32 height, const uint32 *pixels);
+  Image(uint32 width, uint32 height,
+        const uint32 *pixels,
+        bool bBlockStreamOrder = false);
   Image(const Image &);
   Image &operator=(const Image &);
   virtual ~Image();
-  
+
+  virtual Image *Clone() const {
+    return new Image(*this);
+  };
+
   const uint8 *RawData() const { return m_Data; }
 
   uint32 GetWidth() const { return m_Width; }
   uint32 GetHeight() const { return m_Height; }
 
-  void SetBlockStreamOrder(bool flag) { m_bBlockStreamOrder = flag; }
+  void SetBlockStreamOrder(bool flag) {
+    if(flag) {
+      ConvertToBlockStreamOrder();
+    } else {
+      ConvertFromBlockStreamOrder();
+    }
+  }
   bool GetBlockStreamOrder() const { return m_bBlockStreamOrder; }
 
   double ComputePSNR(Image *other);
@@ -77,6 +89,9 @@ class Image {
 
  protected:
   uint8 *m_Data;
+
+  void ConvertToBlockStreamOrder();
+  void ConvertFromBlockStreamOrder();
 };
 
 #endif // __TEXCOMP_IMAGE_H__

Base/src/Image.cpp

@@ -67,10 +67,10 @@ Image::Image(const Image &other)
   }
 }
 
-Image::Image(uint32 width, uint32 height, const uint32 *pixels) 
+Image::Image(uint32 width, uint32 height, const uint32 *pixels, bool bBlockStreamOrder)
   : m_Width(width)
   , m_Height(height)
-  , m_bBlockStreamOrder(false)
+  , m_bBlockStreamOrder(bBlockStreamOrder)
 {
   if(pixels) {
     m_Data = new uint8[4 * m_Width * m_Height];
@@ -80,6 +80,12 @@ Image::Image(uint32 width, uint32 height, const uint32 *pixels)
   }
 }
 
+Image::~Image() {
+  if(m_Data) {
+    delete [] m_Data;
+    m_Data = 0;
+  }
+}
 
 Image &Image::operator=(const Image &other) {
   
@@ -105,13 +111,6 @@ Image &Image::operator=(const Image &other) {
   return *this;
 }
 
-Image::~Image() {
-  if(m_Data) {
-    delete [] m_Data;
-    m_Data = 0;
-  }
-}
-
 double Image::ComputePSNR(Image *other) {
   if(!other)
     return -1.0;
@@ -130,42 +129,88 @@ double Image::ComputePSNR(Image *other) {
   const uint8 *otherData =
     reinterpret_cast<const uint8 *>(other->GetRGBA());
 
-  const double wr = 1.0;
+  //  const double w[3] = { 0.2126, 0.7152, 0.0722 };
-  const double wg = 1.0;
+  const double w[3] = { 1.0, 1.0, 1.0 };
-  const double wb = 1.0;
     
-  double MSE = 0.0;
+  double mse = 0.0;
   const uint32 imageSz = GetWidth() * GetHeight() * 4;
   for(uint32 i = 0; i < imageSz; i+=4) {
 
-    const unsigned char *ourPixel = ourData + i;
+    const unsigned char *pixelDataRaw = ourData + i;
-    const unsigned char *otherPixel = otherData + i;
+    const unsigned char *pixelDataUncomp = otherData + i;
 
-    double ourAlphaScale = double(ourPixel[3]) / 255.0;
+    float r[4], u[4];
-    double otherAlphaScale = double(otherPixel[3]) / 255.0;
+    for(uint32 c = 0; c < 4; c++) {
-    double dr = double(sad(ourAlphaScale * ourPixel[0],
+      if(c == 3) {
-                           otherAlphaScale * otherPixel[0])) * wr;
+        r[c] = pixelDataRaw[c] / 255.0;
-    double dg = double(sad(ourAlphaScale * ourPixel[1],
+        u[c] = pixelDataUncomp[c] / 255.0;
-                           otherAlphaScale * otherPixel[1])) * wg;
+      } else {
-    double db = double(sad(ourAlphaScale * ourPixel[2],
+        r[c] = static_cast<double>(pixelDataRaw[c]) * w[c];
-                           otherAlphaScale * otherPixel[2])) * wb;
+        u[c] = static_cast<double>(pixelDataUncomp[c]) * w[c];
-    
+      }
-    const double pixelMSE = 
+    }
-      (double(dr) * double(dr)) + 
+
-      (double(dg) * double(dg)) + 
+    for(uint32 c = 0; c < 3; c++) {
-      (double(db) * double(db));
+      double diff = (r[3] * r[c] - u[3] * u[c]);
-    
+      mse += diff * diff;
-    //fprintf(stderr, "Pixel MSE: %f\n", pixelMSE);
+    }
-    MSE += pixelMSE;
   }
 
-  MSE /= (double(GetWidth()) * double(GetHeight()));
+  mse /= GetWidth() * GetHeight();
 
-  double MAXI = 
+  const double C = 255.0 * 255.0;
-    (255.0 * wr) * (255.0 * wr) + 
+  double maxi = (w[0]*w[0] + w[1]*w[1] + w[2]*w[2]) * C;
-    (255.0 * wg) * (255.0 * wg) + 
+  return 10 * log10(maxi/mse);
-    (255.0 * wb) * (255.0 * wb);
+}
-
+
-  double PSNR = 10 * log10(MAXI/MSE);
+void Image::ConvertToBlockStreamOrder() {
-  return PSNR;
+  if(m_bBlockStreamOrder || !m_Data)
+    return;
+
+  uint32 *newPixelData = new uint32[GetWidth() * GetHeight() * 4];
+  for(uint32 j = 0; j < GetHeight(); j+=4) {
+    for(uint32 i = 0; i < GetWidth(); i+=4) {
+      uint32 blockX = i / 4;
+      uint32 blockY = j / 4;
+      uint32 blockIdx = blockY * (GetWidth() / 4) + blockX;
+
+      uint32 offset = blockIdx * 4 * 4;
+      for(uint32 t = 0; t < 16; t++) {
+        uint32 x = i + t % 4;
+        uint32 y = j + t / 4;
+        newPixelData[offset + t] =
+          reinterpret_cast<uint32 *>(m_Data)[y*GetWidth() + x];
+      }
+    }
+  }
+
+  delete m_Data;
+  m_Data = reinterpret_cast<uint8 *>(newPixelData);
+  m_bBlockStreamOrder = true;
+}
+
+void Image::ConvertFromBlockStreamOrder() {
+  if(!m_bBlockStreamOrder || !m_Data)
+    return;
+
+  uint32 *newPixelData = new uint32[GetWidth() * GetHeight() * 4];
+  for(uint32 j = 0; j < GetHeight(); j+=4) {
+    for(uint32 i = 0; i < GetWidth(); i+=4) {
+      uint32 blockX = i / 4;
+      uint32 blockY = j / 4;
+      uint32 blockIdx = blockY * (GetWidth() / 4) + blockX;
+
+      uint32 offset = blockIdx * 4 * 4;
+      for(uint32 t = 0; t < 16; t++) {
+        uint32 x = i + t % 4;
+        uint32 y = j + t / 4;
+        newPixelData[y*GetWidth() + x] =
+          reinterpret_cast<uint32 *>(m_Data)[offset + t];
+      }
+    }
+  }
+
+  delete m_Data;
+  m_Data = reinterpret_cast<uint8 *>(newPixelData);
+  m_bBlockStreamOrder = false;
 }

CLTool/src/clunix.cpp

@@ -53,6 +53,7 @@
 void PrintUsage() {
   fprintf(stderr, "Usage: tc [OPTIONS] imagefile\n");
   fprintf(stderr, "\n");
+  fprintf(stderr, "\t-f\t\tFormat to use. Either \"BPTC\" or \"PVRTC\". Default: BPTC\n");
   fprintf(stderr, "\t-l\t\tSave an output log.\n");
   fprintf(stderr, "\t-q <quality>\tSet compression quality level. Default: 50\n");
   fprintf(stderr, "\t-n <num>\tCompress the image num times and give the average time and PSNR. Default: 1\n");
@@ -92,7 +93,8 @@ int main(int argc, char **argv) {
   bool bUseSIMD = false;
   bool bSaveLog = false;
   bool bUseAtomics = false;
-  
+  ECompressionFormat format = eCompressionFormat_BPTC;
+
   bool knowArg = false;
   do {
     knowArg = false;
@@ -110,6 +112,23 @@ int main(int argc, char **argv) {
       continue;
     }
 
+    if(strcmp(argv[fileArg], "-f") == 0) {
+      fileArg++;
+
+      if(fileArg == argc) {
+        PrintUsage();
+        exit(1);
+      } else {
+        if(!strcmp(argv[fileArg], "PVRTC")) {
+          format = eCompressionFormat_PVRTC;
+        }
+      }
+
+      fileArg++;
+      knowArg = true;
+      continue;
+    }
+
     if(strcmp(argv[fileArg], "-l") == 0) {
       fileArg++;
       bSaveLog = true;
@@ -187,6 +206,9 @@ int main(int argc, char **argv) {
   }
 
   Image img = Image(*file.GetImage());
+  if(format == eCompressionFormat_PVRTC) {
+    img.SetBlockStreamOrder(false);
+  }
 
   int numBlocks = (img.GetWidth() * img.GetHeight())/16;
   BlockStatManager *statManager = NULL;
@@ -195,6 +217,7 @@ int main(int argc, char **argv) {
   }
   
   SCompressionSettings settings;
+  settings.format = format;
   settings.bUseSIMD = bUseSIMD;
   settings.bUseAtomics = bUseAtomics;
   settings.iNumThreads = numThreads;
@@ -223,8 +246,13 @@ int main(int argc, char **argv) {
     statManager->ToFile(logname);
   }
 
-  Image cImg(*ci);
+  if(format == eCompressionFormat_BPTC) {
-  ImageFile cImgFile (strcat(basename, "-bc7.png"), eFileFormat_PNG, cImg);
+    strcat(basename, "-bc7.png");
+  } else if(format == eCompressionFormat_PVRTC) {
+    strcat(basename, "-pvrtc.png");
+  }
+
+  ImageFile cImgFile (basename, eFileFormat_PNG, *ci);
   cImgFile.Write();
 
   // Cleanup 

CLTool/src/clwin32.cpp

@@ -54,6 +54,7 @@
 void PrintUsage() {
   fprintf(stderr, "Usage: tc [OPTIONS] imagefile\n");
   fprintf(stderr, "\n");
+  fprintf(stderr, "\t-f\t\tFormat to use. Either \"BPTC\" or \"PVRTC\". Default: BPTC\n");
   fprintf(stderr, "\t-l\t\tSave an output log.\n");
   fprintf(stderr, "\t-q <quality>\tSet compression quality level. Default: 50\n");
   fprintf(stderr, "\t-n <num>\tCompress the image num times and give the average time and PSNR. Default: 1\n");
@@ -93,11 +94,29 @@ int _tmain(int argc, _TCHAR* argv[])
   bool bUseSIMD = false;
   bool bSaveLog = false;
   bool bUseAtomics = false;
+  ECompressionFormat format = eCompressionFormat_BPTC;
   
   bool knowArg = false;
   do {
     knowArg = false;
 
+    if(strcmp(argv[fileArg], "-f") == 0) {
+      fileArg++;
+
+      if(fileArg == argc) {
+        PrintUsage();
+        exit(1);
+      } else {
+        if(!strcmp(argv[fileArg], "PVRTC")) {
+          format = eCompressionFormat_PVRTC;
+        }
+      }
+
+      fileArg++;
+      knowArg = true;
+      continue;
+    }
+
     if(strcmp(argv[fileArg], "-n") == 0) {
       fileArg++;
 
@@ -192,15 +211,19 @@ int _tmain(int argc, _TCHAR* argv[])
     return 1;
   }
 
-  const Image *img = file.GetImage();
+  Image img (*file.GetImage())
+  if(format == eCompressionFormat_PVRTC) {
+    img.SetBlockStreamOrder(false);
+  }
 
-  int numBlocks = (img->GetWidth() * img->GetHeight())/16;
+  int numBlocks = (img.GetWidth() * img.GetHeight())/16;
   BlockStatManager *statManager = NULL;
   if(bSaveLog) {
     statManager = new BlockStatManager(numBlocks);
   }
   
   SCompressionSettings settings;
+  settings.format = format;
   settings.bUseSIMD = bUseSIMD;
   settings.bUseAtomics = bUseAtomics;
   settings.iNumThreads = numThreads;
@@ -226,11 +249,16 @@ int _tmain(int argc, _TCHAR* argv[])
   if(bSaveLog) {
     strcat_s(basename, ".log");
     statManager->ToFile(basename);
-  basename[strlen(basename) - 4] = '\0';
+    basename[strlen(basename) - 4] = '\0';
   }
-  strcat_s(basename, "-bc7.png");
+
-  Image cImg (*ci);
+  if(format == eCompressionFormat_BPTC) {
-  ImageFile cImgFile (basename, eFileFormat_PNG, cImg);
+    strcat_s(basename, "-bc7.png");
+  } else if(format == eCompressionFormat_PVRTC) {
+    strcat_s(basename, "-pvrtc.png");
+  }
+
+  ImageFile cImgFile (basename, eFileFormat_PNG, *ci);
   cImgFile.Write();
 
   // Cleanup 

CMakeModules/FindPVRTexLib.cmake

@@ -56,30 +56,51 @@
 #  PVRTEXLIB_LIBRARIES - The libraries needed to use PVRTexLib
 
 IF (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
+  SET( PVRTEXLIB_ROOT "/Applications/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library" )
   find_path(
     PVRTEXLIB_INCLUDE_DIR PVRTexture.h
-    PATHS "/Applications/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/Include"
+    PATHS ${PVRTEXLIB_ROOT}/Include
   )
 
   find_library(PVRTEXLIB_LIB PVRTexLib
-    PATHS "/Applications/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/OSX_x86/Static"
+    PATHS ${PVRTEXLIB_ROOT}/OSX_x86/Static
-          "/Applications/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/OSX_x86/Dynamic"
+          ${PVRTEXLIB_ROOT}/OSX_x86/Dynamic
   )
-ELSEIF(MSVC)
+ELSEIF (${CMAKE_SYSTEM_NAME} MATCHES "Linux")
+  SET( PVRTEXLIB_ROOT "/opt/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library" )
   find_path(
     PVRTEXLIB_INCLUDE_DIR PVRTexture.h
-    PATHS "C:/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/Include"
+    PATHS ${PVRTEXLIB_ROOT}/Include
+  )
+
+  IF(${CMAKE_SIZEOF_VOID_P} EQUAL 8)
+    find_library(PVRTEXLIB_LIB PVRTexLib
+      PATHS ${PVRTEXLIB_ROOT}/Linux_x86_64/Static
+            ${PVRTEXLIB_ROOT}/Linux_x86_64/Dynamic
+    )
+  ELSE()
+    find_library(PVRTEXLIB_LIB PVRTexLib
+      PATHS ${PVRTEXLIB_ROOT}/Linux_x86_32/Static
+            ${PVRTEXLIB_ROOT}/Linux_x86_32/Dynamic
+    )
+  ENDIF()
+
+ELSEIF(MSVC)
+  SET( PVRTEXLIB_ROOT "C:/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library" )
+  find_path(
+    PVRTEXLIB_INCLUDE_DIR PVRTexture.h
+    PATHS ${PVRTEXLIB_ROOT}/Include
   )
 
   IF(${CMAKE_GENERATOR} MATCHES Win64)
     find_library(PVRTEXLIB_LIB PVRTexLib
-      PATHS "C:/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/Windows_x86_64/Static"
+      PATHS ${PVRTEXLIB_ROOT}/Windows_x86_64/Static
-            "C:/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/Windows_x86_64/Dynamic"
+            ${PVRTEXLIB_ROOT}/Windows_x86_64/Dynamic
     )
   ELSE()
     find_library(PVRTEXLIB_LIB PVRTexLib
-      PATHS "C:/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/Windows_x86_32/Static"
+      PATHS ${PVRTEXLIB_ROOT}/Windows_x86_32/Static
-            "C:/Imagination/PowerVR/GraphicsSDK/PVRTexTool/Library/Windows_x86_32/Dynamic"
+            ${PVRTEXLIB_ROOT}/Windows_x86_32/Dynamic
     )
   ENDIF()
 ENDIF()
@@ -93,4 +114,4 @@ include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(PVRTexLib  DEFAULT_MSG
                                   PVRTEXLIB_LIB PVRTEXLIB_INCLUDE_DIR)
 
-mark_as_advanced(PVRTEXLIB_INCLUDE_DIR PVRTEXLIB_LIB )
+mark_as_advanced( PVRTEXLIB_ROOT PVRTEXLIB_INCLUDE_DIR PVRTEXLIB_LIB )

Core/CMakeLists.txt

@@ -77,6 +77,8 @@ ELSE()
 ENDIF()
 
 INCLUDE_DIRECTORIES( ${FasTC_SOURCE_DIR}/Base/include )
+INCLUDE_DIRECTORIES( ${FasTC_SOURCE_DIR} )
+INCLUDE_DIRECTORIES( ${FasTC_SOURCE_DIR}/PVRTCEncoder/include )
 
 INCLUDE_DIRECTORIES( ${FasTC_SOURCE_DIR}/BPTCEncoder/include )
 INCLUDE_DIRECTORIES( ${FasTC_BINARY_DIR}/BPTCEncoder/include )
@@ -157,6 +159,7 @@ ADD_LIBRARY( FasTCCore
 TARGET_LINK_LIBRARIES( FasTCCore FasTCBase )
 TARGET_LINK_LIBRARIES( FasTCCore FasTCIO )
 TARGET_LINK_LIBRARIES( FasTCCore BPTCEncoder )
+TARGET_LINK_LIBRARIES( FasTCCore PVRTCEncoder )
 
 IF( THREAD_API MATCHES "Boost" )
 	TARGET_LINK_LIBRARIES( FasTCCore ${Boost_LIBRARIES} )

Core/include/CompressedImage.h

@@ -44,10 +44,13 @@
 #ifndef _COMPRESSED_IMAGE_H_
 #define _COMPRESSED_IMAGE_H_
 
+#include "TexCompTypes.h"
+
 enum ECompressionFormat {
   eCompressionFormat_DXT1,
   eCompressionFormat_DXT5,
   eCompressionFormat_BPTC,
+  eCompressionFormat_PVRTC,
 
   kNumCompressionFormats
 };
@@ -55,37 +58,47 @@ enum ECompressionFormat {
 #include "Image.h"
 
 class CompressedImage : public Image {
-
  private:
   ECompressionFormat m_Format;
   uint32 *m_RGBAData;
   uint32 m_DataSz;
 
-  void InitData(const unsigned char *withData);
  public:
-  CompressedImage();
+  CompressedImage(const CompressedImage &);
+  CompressedImage &operator=(const CompressedImage &);
 
   // Create a compressed image from the given data according to
   // the passed format. The size of the data is expected to conform
   // to the width, height, and format specified.
   CompressedImage(
-    const unsigned int width, 
+    const uint32 width, 
-    const unsigned int height, 
+    const uint32 height, 
     const ECompressionFormat format, 
-    const unsigned char *data
+    const uint8 *data
   );
 
-  CompressedImage( const CompressedImage &other );
   virtual ~CompressedImage();
 
+  virtual Image *Clone() const {
+    return new CompressedImage(*this);
+  }
+
   virtual void ComputeRGBA();
   virtual const uint32 *GetRGBA() const { return m_RGBAData; }
 
+  static uint32 GetCompressedSize(uint32 uncompressedSize, ECompressionFormat format);
+  static uint32 GetUncompressedSize(uint32 compressedSize, ECompressionFormat format) {
+    uint32 cmp = GetCompressedSize(compressedSize, format);
+    return compressedSize * (compressedSize / cmp);
+  }
+
   // Decompress the compressed image data into outBuf. outBufSz is expected
   // to be the proper size determined by the width, height, and format.
   // !FIXME! We should have a function to explicitly return the in/out buf
   // size for a given compressed image.
-  bool DecompressImage(unsigned char *outBuf, unsigned int outBufSz) const;
+  bool DecompressImage(uint8 *outBuf, uint32 outBufSz) const;
+
+  ECompressionFormat GetFormat() const { return m_Format; }
 };
 
 #endif // _COMPRESSED_IMAGE_H_

Core/include/TexComp.h

@@ -96,7 +96,8 @@ extern CompressedImage *CompressImage(Image *img, const SCompressionSettings &se
 
 extern bool CompressImageData(
   const unsigned char *data,
-  const unsigned int dataSz,
+  const unsigned int width,
+  const unsigned int height,
   unsigned char *cmpData,
   const unsigned int cmpDataSz,
   const SCompressionSettings &settings

Core/src/CompressedImage.cpp

@@ -50,14 +50,18 @@
 
 #include "TexCompTypes.h"
 #include "BC7Compressor.h"
+#include "PVRTCCompressor.h"
 
 CompressedImage::CompressedImage( const CompressedImage &other )
   : Image(other)
   , m_Format(other.m_Format)
   , m_RGBAData(0)
-  , m_DataSz(0)
+  , m_DataSz(other.m_DataSz)
 {
-  InitData(other.m_Data);
+  if(other.m_RGBAData) {
+    m_RGBAData = new uint32[GetWidth() * GetHeight()];
+    memcpy(m_RGBAData, other.m_RGBAData, sizeof(uint32) * GetWidth() * GetHeight());
+  }
 }
 
 CompressedImage::CompressedImage(
@@ -69,46 +73,36 @@ CompressedImage::CompressedImage(
   : Image(width, height, NULL)
   , m_Format(format)
   , m_RGBAData(0)
-  , m_DataSz(0)
+  , m_DataSz(GetCompressedSize(GetWidth() * GetHeight() * 4, m_Format))
 {
-  InitData(data);
+  if(m_DataSz > 0) {
+    assert(!m_Data);
+    m_Data = new unsigned char[m_DataSz];
+    memcpy(m_Data, data, m_DataSz);
+  }
 }
 
-void CompressedImage::InitData(const unsigned char *withData) {
+CompressedImage &CompressedImage::operator=(const CompressedImage &other) {
-  m_DataSz = 0;
+  Image::operator=(other);
-  int uncompDataSz = GetWidth() * GetHeight() * 4;
+  m_Format = other.m_Format;
-
+  m_DataSz = other.m_DataSz;
-  switch(m_Format) {
+  if(other.m_RGBAData) {
-    default: assert(!"Not implemented!"); // Fall through V
+    m_RGBAData = new uint32[GetWidth() * GetHeight()];
-    case eCompressionFormat_DXT1: m_DataSz = uncompDataSz / 8; break;
+    memcpy(m_RGBAData, other.m_RGBAData, sizeof(uint32) * GetWidth() * GetHeight());
-    case eCompressionFormat_DXT5: m_DataSz = uncompDataSz / 4; break;
-    case eCompressionFormat_BPTC: m_DataSz = uncompDataSz / 4; break;
-  }
-
-  if(m_DataSz > 0) {
-    m_Data = new unsigned char[m_DataSz];
-    memcpy(m_Data, withData, m_DataSz);
   }
 }
 
 CompressedImage::~CompressedImage() {
-  if(m_Data) {
+  if(m_RGBAData) {
-    delete [] m_Data;
+    delete m_RGBAData;
-    m_Data = NULL;
+    m_RGBAData = NULL;
   }
 }
 
 bool CompressedImage::DecompressImage(unsigned char *outBuf, unsigned int outBufSz) const {
 
   // First make sure that we have enough data
-  uint32 dataSz = 0;
+  uint32 dataSz = GetUncompressedSize(m_DataSz, m_Format);
-  switch(m_Format) {
-    default: assert(!"Not implemented!"); // Fall through V
-    case eCompressionFormat_DXT1: dataSz = m_DataSz * 8; break;
-    case eCompressionFormat_DXT5: dataSz = m_DataSz * 4; break;
-    case eCompressionFormat_BPTC: dataSz = m_DataSz * 4; break;
-  }
-
   if(dataSz > outBufSz) {
     fprintf(stderr, "Not enough space to store entire decompressed image! "
                     "Got %d bytes, but need %d!\n", outBufSz, dataSz);
@@ -118,6 +112,12 @@ bool CompressedImage::DecompressImage(unsigned char *outBuf, unsigned int outBuf
 
   DecompressionJob dj (m_Data, outBuf, GetWidth(), GetHeight());
   switch(m_Format) {
+    case eCompressionFormat_PVRTC:
+    {
+      PVRTCC::Decompress(dj);
+    }
+    break;
+
     case eCompressionFormat_BPTC: 
     { 
       BC7C::Decompress(dj);
@@ -146,3 +146,26 @@ void CompressedImage::ComputeRGBA() {
   uint8 *pixelData = reinterpret_cast<uint8 *>(m_RGBAData);
   DecompressImage(pixelData, GetWidth() * GetHeight() * 4);
 }
+
+uint32 CompressedImage::GetCompressedSize(uint32 uncompressedSize, ECompressionFormat format) {
+  assert(uncompressedSize % 8 == 0);
+
+  uint32 cmpDataSzNeeded = 0;
+  switch(format) {
+  default:
+    assert(!"Not implemented!");
+    // Fall through V
+  case eCompressionFormat_DXT1:
+  case eCompressionFormat_PVRTC:
+    cmpDataSzNeeded = uncompressedSize / 8;
+    break;
+
+  case eCompressionFormat_DXT5:
+  case eCompressionFormat_BPTC:
+    cmpDataSzNeeded = uncompressedSize / 4;
+    break;
+  }
+
+  return cmpDataSzNeeded;
+}
+

Core/src/TexComp.cpp

@@ -50,14 +50,14 @@
 #include <cassert>
 #include <iostream>
 
-#include "CompressionFuncs.h"
 #include "BC7Compressor.h"
-#include "Thread.h"
+#include "CompressionFuncs.h"
-#include "WorkerQueue.h"
-#include "ThreadGroup.h"
-
-#include "ImageFile.h"
 #include "Image.h"
+#include "ImageFile.h"
+#include "PVRTCCompressor.h"
+#include "Thread.h"
+#include "ThreadGroup.h"
+#include "WorkerQueue.h"
 
 template <typename T>
 static void clamp(T &x, const T &minX, const T &maxX) {
@@ -69,6 +69,10 @@ static inline T sad(const T &a, const T &b) {
   return (a > b)? a - b : b - a;
 }
 
+static void CompressPVRTC(const CompressionJob &cj) {
+  PVRTCC::Compress(cj);
+}
+
 SCompressionSettings:: SCompressionSettings()
   : format(eCompressionFormat_BPTC)
   , bUseSIMD(false)
@@ -87,6 +91,12 @@ static  CompressionFuncWithStats ChooseFuncFromSettingsWithStats(const SCompress
     }
     break;
 
+    case eCompressionFormat_PVRTC:
+    {
+      // !FIXME! actually implement one of these methods...
+      return NULL;
+    }
+
     default:
     {
       assert(!"Not implemented!");
@@ -110,6 +120,11 @@ static CompressionFunc ChooseFuncFromSettings(const SCompressionSettings &s) {
     }
     break;
 
+    case eCompressionFormat_PVRTC:
+    {
+      return CompressPVRTC;
+    }
+
     default:
     {
       assert(!"Not implemented!");
@@ -124,8 +139,9 @@ static void ReportError(const char *msg) {
 }
 
 static double CompressImageInSerial(
-  const unsigned char *imgData,
+  const uint8 *imgData,
-  const unsigned int imgDataSz,
+  const uint32 imgWidth,
+  const uint32 imgHeight,
   const SCompressionSettings &settings,
   unsigned char *outBuf
 ) {
@@ -134,14 +150,14 @@ static double CompressImageInSerial(
 
   double cmpTimeTotal = 0.0;
 
+  StopWatch stopWatch = StopWatch();
   for(int i = 0; i < settings.iNumCompressions; i++) {
 
-    StopWatch stopWatch = StopWatch();
     stopWatch.Reset();
     stopWatch.Start();
 
     // !FIXME! We're assuming that we have 4x4 blocks here...
-    CompressionJob cj (imgData, outBuf, imgDataSz / 16, 4);
+    CompressionJob cj (imgData, outBuf, imgWidth, imgHeight);
     if(fStats && settings.pStatManager) {
       // !FIXME! Actually use the stat manager...
       //(*fStats)(cj, *(settings.pStatManager));
@@ -347,20 +363,14 @@ CompressedImage *CompressImage(
   assert(dataSz > 0);
 
   // Allocate data based on the compression method
-  int cmpDataSz = 0;
+  uint32 cmpDataSz = CompressedImage::GetCompressedSize(dataSz, settings.format);
-  switch(settings.format) {
-    default: assert(!"Not implemented!"); // Fall Through V
-    case eCompressionFormat_DXT1: cmpDataSz = dataSz / 8; break;
-    case eCompressionFormat_DXT5: cmpDataSz = dataSz / 4; break;
-    case eCompressionFormat_BPTC: cmpDataSz = dataSz / 4; break;
-  }
 
   // Make sure that we have RGBA data...
   img->ComputeRGBA();
 
   unsigned char *cmpData = new unsigned char[cmpDataSz];
   const uint8 *pixelData = reinterpret_cast<const uint8 *>(img->GetRGBA());
-  CompressImageData(pixelData, dataSz, cmpData, cmpDataSz, settings);
+  CompressImageData(pixelData, w, h, cmpData, cmpDataSz, settings);
 
   outImg = new CompressedImage(w, h, settings.format, cmpData);
   
@@ -370,12 +380,15 @@ CompressedImage *CompressImage(
 
 bool CompressImageData(
   const unsigned char *data, 
-  const unsigned int dataSz,
+  const unsigned int width,
+  const unsigned int height,
   unsigned char *cmpData,
   const unsigned int cmpDataSz,
   const SCompressionSettings &settings
 ) { 
 
+  uint32 dataSz = width * height * 4;
+
   // Make sure that platform supports SSE if they chose this
   // option...
   #ifndef HAS_SSE_41
@@ -390,15 +403,23 @@ bool CompressImageData(
     return false;
   }
 
-  // Allocate data based on the compression method
+  uint32 numThreads = settings.iNumThreads;
-  uint32 cmpDataSzNeeded = 0;
+  if(settings.format == eCompressionFormat_PVRTC &&
-  switch(settings.format) {
+     (settings.iNumThreads > 1 || settings.pStatManager)) {
-    default: assert(!"Not implemented!"); // Fall through V
+    if(settings.iNumThreads > 1) {
-    case eCompressionFormat_DXT1: cmpDataSzNeeded = dataSz / 8; break;
+      ReportError("WARNING - PVRTC compressor does not support multithreading.");
-    case eCompressionFormat_DXT5: cmpDataSzNeeded = dataSz / 4; break;
+      numThreads = 1;
-    case eCompressionFormat_BPTC: cmpDataSzNeeded = dataSz / 4; break;
+    }
+
+    if(settings.pStatManager) {
+      ReportError("WARNING - PVRTC compressor does not support stat collection.");
+    }
   }
 
+  // Allocate data based on the compression method
+  uint32 cmpDataSzNeeded =
+    CompressedImage::GetCompressedSize(dataSz, settings.format);
+
   if(cmpDataSzNeeded == 0) {
     ReportError("Unknown compression format");
     return false;
@@ -413,21 +434,17 @@ bool CompressImageData(
 
     double cmpMSTime = 0.0;
 
-    if(settings.iNumThreads > 1) {
+    if(numThreads > 1) {
       if(settings.bUseAtomics) {
-        //!KLUDGE!
-        unsigned int height = 4;
-        unsigned int width = dataSz / 16;
-
         cmpMSTime = CompressImageWithAtomics(data, width, height, settings, cmpData);
-      }
+      } else if(settings.iJobSize > 0) {
-      else if(settings.iJobSize > 0)
         cmpMSTime = CompressImageWithWorkerQueue(data, dataSz, settings, cmpData);
-      else
+      } else {
         cmpMSTime = CompressImageWithThreads(data, dataSz, settings, cmpData);
+      }
     }
     else {
-      cmpMSTime = CompressImageInSerial(data, dataSz, settings, cmpData);
+      cmpMSTime = CompressImageInSerial(data, width, height, settings, cmpData);
     }
 
     // Report compression time

IO/src/ImageFile.cpp

@@ -107,7 +107,7 @@ ImageFile::ImageFile(const CHAR *filename, EImageFileFormat format)
 
 ImageFile::ImageFile(const char *filename, EImageFileFormat format, const Image &image)
   : m_FileFormat(format)
-  , m_Image(new Image(image))
+  , m_Image(image.Clone())
 {
   strncpy(m_Filename, filename, kMaxFilenameSz);
 }
@@ -207,7 +207,7 @@ Image *ImageFile::LoadImage(const unsigned char *rawImageData) const {
   }
 
   uint32 *pixels = reinterpret_cast<uint32 *>(pixelData);
-  Image *i = new Image(loader->GetWidth(), loader->GetHeight(), pixels);
+  Image *i = new Image(loader->GetWidth(), loader->GetHeight(), pixels, true);
 
   // Cleanup
   delete loader;

IO/src/ImageWriterPNG.cpp

@@ -84,11 +84,13 @@ public:
 
 };
 
-ImageWriterPNG::ImageWriterPNG(const Image &im)
+ImageWriterPNG::ImageWriterPNG(Image &im)
   : ImageWriter(im.GetWidth(), im.GetHeight(), im.RawData())
   , m_bBlockStreamOrder(im.GetBlockStreamOrder())
   , m_StreamPosition(0)
 {
+  im.ComputeRGBA();
+  m_PixelData = reinterpret_cast<const uint8 *>(im.GetRGBA());
 }
 
 bool ImageWriterPNG::WriteImage() {
@@ -111,7 +113,6 @@ bool ImageWriterPNG::WriteImage() {
   }
     
   /* Set image attributes. */
-
   png_set_IHDR (png_ptr,
                 info_ptr,
                 m_Width,
@@ -123,7 +124,6 @@ bool ImageWriterPNG::WriteImage() {
                 PNG_FILTER_TYPE_DEFAULT);
     
   /* Initialize rows of PNG. */
-
   row_pointers = (png_byte **)png_malloc (png_ptr, m_Height * sizeof (png_byte *));
   for (uint32 y = 0; y < m_Height; ++y) {
     png_byte *row = (png_byte *)png_malloc (png_ptr, sizeof (uint8) * m_Width * pixel_size);
@@ -136,8 +136,8 @@ bool ImageWriterPNG::WriteImage() {
           *row++ = GetChannelForPixel(x, y, ch);
         }
       } else {
-        *(reinterpret_cast<uint32 *>(row) + x) =
+        reinterpret_cast<uint32 *>(row)[x] =
-          (reinterpret_cast<const uint32 *>(m_PixelData))[y * m_Width + x];
+          reinterpret_cast<const uint32 *>(m_PixelData)[y * m_Width + x];
       }
     }
   }

IO/src/ImageWriterPNG.h

@@ -50,7 +50,7 @@
 class Image;
 class ImageWriterPNG : public ImageWriter {
  public:
-  ImageWriterPNG(const Image &);
+  ImageWriterPNG(Image &);
   virtual ~ImageWriterPNG() { }
 
   virtual bool WriteImage();

PVRTCEncoder/CMakeLists.txt

@@ -62,6 +62,7 @@ SET( HEADERS
 )
 
 SET( SOURCES
+  src/Compressor.cpp
   src/Decompressor.cpp
   src/Pixel.cpp
   src/Block.cpp

PVRTCEncoder/include/PVRTCCompressor.h

@@ -70,9 +70,18 @@ namespace PVRTCC {
   // decompress the data.
   void Decompress(const DecompressionJob &,
                   bool bTwoBitMode = false,
-                  const EWrapMode wrapMode = eWrapMode_Clamp,
+                  const EWrapMode wrapMode = eWrapMode_Wrap,
                   bool bDebugImages = false);
 
+  // Takes a stream of uncompressed RGBA8 data and compresses it into PVRTC
+  // version one. The width and height must be specified in order to properly
+  // decompress the data.
+  void Compress(const CompressionJob &,
+                bool bTwoBitMode = false,
+                const EWrapMode wrapMode = eWrapMode_Wrap);
+
+  static const uint32 kBlockSize = sizeof(uint64);
+
 }  // namespace PVRTCC
 
 #endif  // PVRTCENCODER_INCLUDE_PVRTCCOMPRESSOR_H_

PVRTCEncoder/src/Block.cpp

@@ -83,6 +83,42 @@ namespace PVRTCC {
     return m_ColorA;
   }
 
+  Pixel Block::SetColor(const Pixel &c, bool transparent,
+                        const uint8 (&tbd)[4], const uint8 (&obd)[4]) {
+    uint8 cDepth[4];
+    c.GetBitDepth(cDepth);
+
+    Pixel final = c;
+    if(transparent) {
+      final.ChangeBitDepth(tbd);
+
+      // If we went effectively transparent, then just switch over to opaque...
+      if(final.A() == 0x7) {
+        return SetColor(c, false, tbd, obd);
+      }
+
+    } else {
+      final.A() = 255;
+      final.ChangeBitDepth(obd);
+    }
+
+    return final;
+  }
+
+  void Block::SetColorA(const Pixel &c, bool transparent) {
+    const uint8 transparentBitDepth[4] = { 3, 4, 4, 4 };
+    const uint8 opaqueBitDepth[4] = { 0, 5, 5, 5 };
+    m_ColorA = SetColor(c, transparent, transparentBitDepth, opaqueBitDepth);
+    m_ColorACached = true;
+  }
+
+  void Block::SetColorB(const Pixel &c, bool transparent) {
+    const uint8 transparentBitDepth[4] = { 3, 4, 4, 3 };
+    const uint8 opaqueBitDepth[4] = { 0, 5, 5, 4 };
+    m_ColorB = SetColor(c, transparent, transparentBitDepth, opaqueBitDepth);
+    m_ColorBCached = true;
+  }
+
   Pixel Block::GetColorB() {
     if(m_ColorBCached) {
       return m_ColorB;
@@ -109,6 +145,17 @@ namespace PVRTCC {
     return (m_LongData >> (texelIdx * 2)) & 0x3;
   }
 
+  void Block::SetLerpValue(uint32 texelIdx, uint8 lerpVal) {
+    assert(texelIdx >= 0);
+    assert(texelIdx <= 15);
+
+    assert(lerpVal >= 0);
+    assert(lerpVal < 4);
+
+    m_LongData &= ~(static_cast<uint64>(0x3) << (texelIdx * 2));
+    m_LongData |= static_cast<uint64>(lerpVal & 0x3) << (texelIdx * 2);
+  }
+
   Block::E2BPPSubMode Block::Get2BPPSubMode() const {
     uint8 first = GetLerpValue(0);
     if(!(first & 0x1)) {
@@ -145,4 +192,65 @@ namespace PVRTCC {
 
     return ret;
   }
+
+  uint64 Block::Pack() {
+    assert(m_ColorACached);
+    assert(m_ColorBCached);
+
+#ifndef NDEBUG
+    uint8 bitDepthA[4];
+    m_ColorA.GetBitDepth(bitDepthA);
+
+    uint32 sumA = 0;
+    for(int i = 0; i < 4; i++) {
+      sumA += bitDepthA[i];
+    }
+    assert(sumA == 15);
+#endif
+
+#ifndef NDEBUG
+    uint8 bitDepthB[4];
+    m_ColorB.GetBitDepth(bitDepthB);
+
+    uint32 sumB = 0;
+    for(int i = 0; i < 4; i++) {
+      sumB += bitDepthB[i];
+    }
+    assert(sumB == 14);
+#endif
+
+    uint8 aBits[2], bBits[2];
+    memset(aBits, 0, sizeof(aBits));
+    memset(bBits, 0, sizeof(bBits));
+
+    m_ColorA.ToBits(aBits, 2);
+    m_ColorB.ToBits(bBits, 2, 1);
+
+    if(m_ColorA.A() == 0xFF) {
+      m_ByteData[7] |= 0x80;
+    } else {
+      m_ByteData[7] &= 0x7f;
+    }
+    m_ByteData[7] = aBits[1];
+    m_ByteData[6] = aBits[0];
+
+    bool modeBit = GetModeBit();
+    m_ByteData[5] = bBits[1];
+    m_ByteData[4] = bBits[0];
+    if(m_ColorB.A() == 0xFF) {
+      m_ByteData[5] |= 0x80;
+    } else {
+      m_ByteData[5] &= 0x7f;
+    }
+
+    if(modeBit) {
+      m_ByteData[4] |= 0x1;
+    } else {
+      m_ByteData[4] &= 0xFE;
+    }
+
+    // Modulation data should have already been set...
+    return m_LongData;
+  }
+
 }  // namespace PVRTCC

PVRTCEncoder/src/Block.h

@@ -60,15 +60,29 @@ namespace PVRTCC {
 
 class Block {
  public:
+  Block(): m_LongData(0) { }
   explicit Block(const uint8 *data);
 
+  // Accessors for the A and B colors of the block.
   Pixel GetColorA();
+  void SetColorA(const Pixel &, bool transparent=false);
+
   Pixel GetColorB();
+  void SetColorB(const Pixel &, bool transparent=false);
 
   bool GetModeBit() const {
     return static_cast<bool>((m_LongData >> 32) & 0x1);
   }
 
+  void SetModeBit(bool flag) {
+    const uint64 bit = 0x100000000L;
+    if(flag) {
+      m_LongData |= bit;
+    } else {
+      m_LongData &= ~bit;
+    }
+  }
+
   // For 2BPP PVRTC, if the mode bit is set, then we use the modulation data
   // as 2 bits for every other texel in the 8x4 block in a checkerboard pattern.
   // The interleaved texel data is decided by averaging nearby texel modulation
@@ -100,6 +114,11 @@ class Block {
   // 12 13 14 15
   uint8 GetLerpValue(uint32 texelIdx) const;
 
+  // Sets the values in the data for this block according to the texel and
+  // modulation value passed. This happens immediately (i.e. a call to Pack()
+  // will reflect these changes).
+  void SetLerpValue(uint32 texelIdx, uint8 lerpVal);
+
   // This returns the modulation value for the texel in the block interpreted as
   // 2BPP. If the modulation bit is not set, then it expects a number from 0-31
   // and does the same operation as GetLerpValue. If the modulation bit is set,
@@ -109,6 +128,12 @@ class Block {
   // global information.
   uint8 Get2BPPLerpValue(uint32 texelIdx) const;
 
+  // Returns the 64-bit word that represents this block. This function packs the
+  // A and B colors based on their bit depths and preserves the corresponding mode
+  // bits. The color modes are determined by whether or not the alpha channel of
+  // each block is fully opaque or not.
+  uint64 Pack();
+
  private:
   union {
     uint8 m_ByteData[8];
@@ -120,6 +145,11 @@ class Block {
 
   bool m_ColorBCached;
   Pixel m_ColorB;
+
+  // tbd -- transparent bit depth
+  // obd -- opaque bit depth
+  static Pixel SetColor(const Pixel &c, bool transparent,
+                        const uint8 (&tbd)[4], const uint8 (&obd)[4]);
 };
 
 }  // namespace PVRTCC

PVRTCEncoder/src/Compressor.cpp

@@ -0,0 +1,290 @@
+/* FasTC
+ * Copyright (c) 2013 University of North Carolina at Chapel Hill.
+ * All rights reserved.
+ *
+ * Permission to use, copy, modify, and distribute this software and its
+ * documentation for educational, research, and non-profit purposes, without
+ * fee, and without a written agreement is hereby granted, provided that the
+ * above copyright notice, this paragraph, and the following four paragraphs
+ * appear in all copies.
+ *
+ * Permission to incorporate this software into commercial products may be
+ * obtained by contacting the authors or the Office of Technology Development
+ * at the University of North Carolina at Chapel Hill <otd@unc.edu>.
+ *
+ * This software program and documentation are copyrighted by the University of
+ * North Carolina at Chapel Hill. The software program and documentation are
+ * supplied "as is," without any accompanying services from the University of
+ * North Carolina at Chapel Hill or the authors. The University of North
+ * Carolina at Chapel Hill and the authors do not warrant that the operation of
+ * the program will be uninterrupted or error-free. The end-user understands
+ * that the program was developed for research purposes and is advised not to
+ * rely exclusively on the program for any reason.
+ *
+ * IN NO EVENT SHALL THE UNIVERSITY OF NORTH CAROLINA AT CHAPEL HILL OR THE
+ * AUTHORS BE LIABLE TO ANY PARTY FOR DIRECT, INDIRECT, SPECIAL, INCIDENTAL,
+ * OR CONSEQUENTIAL DAMAGES, INCLUDING LOST PROFITS, ARISING OUT OF THE USE OF
+ * THIS SOFTWARE AND ITS DOCUMENTATION, EVEN IF THE UNIVERSITY OF NORTH CAROLINA
+ * AT CHAPEL HILL OR THE AUTHORS HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ *
+ * THE UNIVERSITY OF NORTH CAROLINA AT CHAPEL HILL AND THE AUTHORS SPECIFICALLY
+ * DISCLAIM ANY WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE AND ANY 
+ * STATUTORY WARRANTY OF NON-INFRINGEMENT. THE SOFTWARE PROVIDED HEREUNDER IS ON
+ * AN "AS IS" BASIS, AND THE UNIVERSITY  OF NORTH CAROLINA AT CHAPEL HILL AND
+ * THE AUTHORS HAVE NO OBLIGATIONS TO PROVIDE MAINTENANCE, SUPPORT, UPDATES, 
+ * ENHANCEMENTS, OR MODIFICATIONS.
+ *
+ * Please send all BUG REPORTS to <pavel@cs.unc.edu>.
+ *
+ * The authors may be contacted via:
+ *
+ * Pavel Krajcevski
+ * Dept of Computer Science
+ * 201 S Columbia St
+ * Frederick P. Brooks, Jr. Computer Science Bldg
+ * Chapel Hill, NC 27599-3175
+ * USA
+ * 
+ * <http://gamma.cs.unc.edu/FasTC/>
+ */
+
+#include "PVRTCCompressor.h"
+
+#include <algorithm>
+#include <cassert>
+#include <iostream>
+#include <vector>
+
+#include "Pixel.h"
+#include "Image.h"
+#include "Block.h"
+
+namespace PVRTCC {
+
+  static uint32 Interleave(uint16 inx, uint16 iny) {
+    // Taken from:
+    // http://graphics.stanford.edu/~seander/bithacks.html#InterleaveBMN
+
+    static const uint32 B[] = {0x55555555, 0x33333333, 0x0F0F0F0F, 0x00FF00FF};
+    static const uint32 S[] = {1, 2, 4, 8};
+
+    uint32 x = static_cast<uint32>(inx);
+    uint32 y = static_cast<uint32>(iny);
+
+    x = (x | (x << S[3])) & B[3];
+    x = (x | (x << S[2])) & B[2];
+    x = (x | (x << S[1])) & B[1];
+    x = (x | (x << S[0])) & B[0];
+
+    y = (y | (y << S[3])) & B[3];
+    y = (y | (y << S[2])) & B[2];
+    y = (y | (y << S[1])) & B[1];
+    y = (y | (y << S[0])) & B[0];
+
+    return x | (y << 1);
+  }
+
+  template <typename T>
+  static T Clamp(const T &v, const T &low, const T &high) {
+    return ::std::min(::std::max(low, v), high);
+  }
+
+  static const Pixel &Lookup(const Image &img,
+                             int32 x, int32 y,
+                             uint32 width, uint32 height,
+                             const EWrapMode wrapMode) {
+    while(x >= width) {
+      if(wrapMode == eWrapMode_Wrap) {
+        x -= width;
+      } else {
+        x = width - 1;
+      }
+    }
+
+    while(x < 0) {
+      if(wrapMode == eWrapMode_Wrap) {
+        x += width;
+      } else {
+        x = 0;
+      }
+    }
+
+    while(y >= height) {
+      if(wrapMode == eWrapMode_Wrap) {
+        y -= height;
+      } else {
+        y = height - 1;
+      }
+    }
+
+    while(y < 0) {
+      if(wrapMode == eWrapMode_Wrap) {
+        y += height;
+      } else {
+        y = 0;
+      }
+    }
+
+    return img(x, y);
+  }
+
+  void Compress(const CompressionJob &dcj,
+                bool bTwoBitMode,
+                const EWrapMode wrapMode) {
+    Image img(dcj.height, dcj.width);
+    uint32 nPixels = dcj.height * dcj.width;
+    for(uint32 i = 0; i < nPixels; i++) {
+      uint32 x = i % dcj.width;
+      uint32 y = i / dcj.width;
+
+      const uint32 *pixels = reinterpret_cast<const uint32 *>(dcj.inBuf);
+      img(x, y).UnpackRGBA(pixels[i]);      
+    }
+
+    Image original = img;
+    img.DebugOutput("Original");
+
+    // Downscale it using anisotropic diffusion based scheme in order to preserve
+    // image features, then reupscale and compute deltas. Use deltas to generate
+    // initial A & B images followed by modulation data.
+    img.ContentAwareDownscale(1, 1, eWrapMode_Wrap, true);
+    img.ContentAwareDownscale(1, 1, eWrapMode_Wrap, false);
+
+    Image downscaled = img;
+
+    // Upscale it again
+    img.BilinearUpscale(2, 2, eWrapMode_Wrap);
+
+    img.DebugOutput("Reconstruction");
+
+    // Compute difference...
+    ::std::vector<int16> difference;
+    difference.resize(dcj.height * dcj.width * 4);
+    for(uint32 j = 0; j < dcj.height; j++) {
+      for(uint32 i = 0; i < dcj.width; i++) {
+        for(uint32 c = 0; c < 4; c++) {
+          int16 o = original(i, j).Component(c);
+          int16 n = img(i, j).Component(c);
+          difference[j*dcj.width*4 + i*4 + c] = o - n;
+        }
+      }
+    }
+
+    const uint32 blocksW = dcj.width / 4;
+    const uint32 blocksH = dcj.height / 4;
+
+    // Go over the 7x7 texel blocks and extract bounding box diagonals for each
+    // block. We should be able to choose which diagonal we want...
+    const uint32 kKernelSz = 7;
+
+    Image imgA = downscaled;
+    Image imgB = downscaled;
+    for(uint32 j = 0; j < blocksH; j++) {
+      for(uint32 i = 0; i < blocksW; i++) {
+        int32 startX = i*4 + 2 - (kKernelSz / 2);
+        int32 startY = j*4 + 2 - (kKernelSz / 2);
+        for(int32 y = startY; y < startY + kKernelSz; y++) {
+          for(int32 x = startX; x < startX + kKernelSz; x++) {
+            const Pixel &po = Lookup(original, x, y, dcj.width, dcj.height, wrapMode);
+            Pixel &pa = imgA(i, j);
+            Pixel &pb = imgB(i, j);
+            for(uint32 c = 0; c < 4; c++) {
+              pa.Component(c) = ::std::max(po.Component(c), pa.Component(c));
+              pb.Component(c) = ::std::min(po.Component(c), pb.Component(c));
+            }
+          }
+        }
+      }
+    }
+
+    imgA.DebugOutput("ImageA");
+    imgB.DebugOutput("ImageB");
+
+    // Determine modulation values...
+    Image upA = imgA;
+    Image upB = imgB;
+
+    upA.BilinearUpscale(2, 2, wrapMode);
+    upB.BilinearUpscale(2, 2, wrapMode);
+
+    assert(upA.GetHeight() == dcj.height && upA.GetWidth() == dcj.width);
+    assert(upB.GetHeight() == dcj.height && upB.GetWidth() == dcj.width);
+
+    upA.DebugOutput("UpscaledA");
+    upB.DebugOutput("UpscaledB");
+
+    // Choose the most appropriate modulation values for the two images...
+    ::std::vector<uint8> modValues;
+    modValues.resize(dcj.width * dcj.height);
+    for(uint32 j = 0; j < dcj.height; j++) {
+      for(uint32 i = 0; i < dcj.width; i++) {
+        uint8 &mv = modValues[j * dcj.width + i];
+
+        const Pixel pa = upA(i, j);
+        const Pixel pb = upB(i, j);
+        const Pixel po = original(i, j);
+        
+        // !FIXME! there are two modulation modes... we're only using one.
+        uint8 modSteps[4] = { 8, 5, 3, 0 };
+        uint8 bestMod = 0;
+        uint32 bestError = 0xFFFFFFFF;
+        for(uint32 s = 0; s < 4; s++) {
+          uint32 error = 0;
+          for(uint32 c = 0; c < 4; c++) {
+            uint16 va = static_cast<uint16>(pa.Component(c));
+            uint16 vb = static_cast<uint16>(pb.Component(c));
+            uint16 vo = static_cast<uint16>(po.Component(c));
+
+            uint16 lerpVal = modSteps[s];
+            uint16 res = (va * (8 - lerpVal) + vb * lerpVal) / 8;
+            uint16 e = (res > vo)? res - vo : vo - res;
+            error += e * e;
+          }
+
+          if(error < bestError) {
+            bestError = error;
+            bestMod = s;
+          }
+        }
+
+        mv = bestMod;
+      }
+    }
+
+    // Pack everything into a PVRTC blocks.
+    assert(imgA.GetHeight() == blocksH);
+    assert(imgA.GetWidth() == blocksW);
+
+    std::vector<uint64> blocks;
+    blocks.reserve(blocksW * blocksH);
+    for(uint32 j = 0; j < blocksH; j++) {
+      for(uint32 i = 0; i < blocksW; i++) {
+        Block b;
+        b.SetColorA(imgA(i, j), true);
+        b.SetColorB(imgB(i, j), true);
+        for(uint32 t = 0; t < 16; t++) {
+          uint32 x = i*4 + (t%4);
+          uint32 y = j*4 + (t/4);
+          b.SetLerpValue(t, modValues[y*dcj.width + x]);
+        }
+        blocks.push_back(b.Pack());
+      }
+    }
+
+    // Spit out the blocks...
+    for(uint32 j = 0; j < blocksH; j++) {
+      for(uint32 i = 0; i < blocksW; i++) {
+
+        // The blocks are initially arranged in morton order. Let's
+        // linearize them...
+        uint32 idx = Interleave(j, i);
+
+        uint64 *outPtr = reinterpret_cast<uint64 *>(dcj.outBuf);
+        outPtr[idx] = blocks[j*blocksW + i];
+      }
+    }
+  }
+
+}  // namespace PVRTCC

PVRTCEncoder/src/Decompressor.cpp

@@ -304,11 +304,10 @@ namespace PVRTCC {
 
     // First, extract all of the block information...
     std::vector<Block> blocks;
-    blocks.reserve(w * h);
 
     const uint32 blocksW = bTwoBitMode? (w / 8) : (w / 4);
     const uint32 blocksH = h / 4;
-    const uint32 blockSz = 8;
+    blocks.reserve(blocksW * blocksH);
 
     for(uint32 j = 0; j < blocksH; j++) {
       for(uint32 i = 0; i < blocksW; i++) {
@@ -317,7 +316,7 @@ namespace PVRTCC {
         // linearize them...
         uint32 idx = Interleave(j, i);
 
-        uint32 offset = idx * blockSz;
+        uint32 offset = idx * kBlockSize;
         blocks.push_back( Block(dcj.inBuf + offset) );
       }
     }

PVRTCEncoder/src/Image.cpp

@@ -52,15 +52,32 @@
 
 #include "Image.h"
 
+#include <algorithm>
 #include <cassert>
 #include <cstring>
 #include <cstdio>
+#include <cmath>
+
+#if _MSC_VER
+#  define _CRT_SECURE_NO_WARNINGS
+#  define snprintf _snprintf
+#endif
 
 #include "Pixel.h"
 
 #include "../../Base/include/Image.h"
 #include "../../IO/include/ImageFile.h"
 
+template <typename T>
+inline T Clamp(const T &v, const T &a, const T &b) {
+  return ::std::min(::std::max(a, v), b);
+}
+
+static float ConvertChannelToFloat(uint8 channel, uint8 bitDepth) {
+  float denominator = static_cast<float>((1 << bitDepth) - 1);
+  return static_cast<float>(channel) / denominator;
+}
+
 namespace PVRTCC {
 
 Image::Image(uint32 height, uint32 width)
@@ -83,27 +100,27 @@ Image::Image(uint32 height, uint32 width, const Pixel *pixels)
 }
 
 Image::Image(const Image &other)
-  : m_Width(other.m_Width)
+  : m_Width(other.GetWidth())
-  , m_Height(other.m_Height)
+  , m_Height(other.GetHeight())
-  , m_Pixels(new Pixel[other.m_Width * other.m_Height])
+  , m_Pixels(new Pixel[other.GetWidth() * other.GetHeight()])
-  , m_FractionalPixels(new Pixel[other.m_Width * other.m_Height]) {
+  , m_FractionalPixels(new Pixel[other.GetWidth() * other.GetHeight()]) {
-  memcpy(m_Pixels, other.m_Pixels, m_Width * m_Height * sizeof(Pixel));
+  memcpy(m_Pixels, other.m_Pixels, GetWidth() * GetHeight() * sizeof(Pixel));
 }
 
 Image &Image::operator=(const Image &other) {
-  m_Width = other.m_Width;
+  m_Width = other.GetWidth();
-  m_Height = other.m_Height;
+  m_Height = other.GetHeight();
 
   assert(m_Pixels);
   delete m_Pixels;
-  m_Pixels = new Pixel[other.m_Width * other.m_Height];
+  m_Pixels = new Pixel[other.GetWidth() * other.GetHeight()];
-  memcpy(m_Pixels, other.m_Pixels, m_Width * m_Height * sizeof(Pixel));
+  memcpy(m_Pixels, other.m_Pixels, GetWidth() * GetHeight() * sizeof(Pixel));
 
   assert(m_FractionalPixels);
   delete m_FractionalPixels;
-  m_FractionalPixels = new Pixel[other.m_Width * other.m_Height];
+  m_FractionalPixels = new Pixel[other.GetWidth() * other.GetHeight()];
   memcpy(m_FractionalPixels, other.m_FractionalPixels,
-         m_Width * m_Height * sizeof(Pixel));
+         GetWidth() * GetHeight() * sizeof(Pixel));
 
   return *this;
 }
@@ -129,8 +146,8 @@ static bool CompareBitDepths(const uint8 (&depth1)[4],
 
 void Image::BilinearUpscale(uint32 xtimes, uint32 ytimes,
                             EWrapMode wrapMode) {
-  const uint32 newWidth = m_Width << xtimes;
+  const uint32 newWidth = GetWidth() << xtimes;
-  const uint32 newHeight = m_Height << ytimes;
+  const uint32 newHeight = GetHeight() << ytimes;
 
   const uint32 xscale = 1 << xtimes;
   const uint32 xoffset = xscale >> 1;
@@ -213,10 +230,152 @@ void Image::BilinearUpscale(uint32 xtimes, uint32 ytimes,
   m_Height = newHeight;
 }
 
+void Image::ContentAwareDownscale(uint32 xtimes, uint32 ytimes,
+                                  EWrapMode wrapMode, bool bOffsetNewPixels) {
+  const uint32 w = GetWidth();
+  const uint32 h = GetHeight();
+
+  const uint32 newWidth = w >> xtimes;
+  const uint32 newHeight = h >> ytimes;
+
+  Pixel *downscaledPixels = new Pixel[newWidth * newHeight];
+  const uint32 numDownscaledPixels = newWidth * newHeight;
+
+  uint8 bitDepth[4];
+  m_Pixels[0].GetBitDepth(bitDepth);
+
+  for(uint32 i = 0; i < numDownscaledPixels; i++) {
+    downscaledPixels[i].ChangeBitDepth(bitDepth);
+  }
+
+  // Allocate memory
+  float *imgData = new float[19 * w * h];
+  float *I = imgData;
+  float *Ix[5] = {
+    imgData + (w * h),
+    imgData + (2 * w * h),
+    imgData + (3 * w * h),
+    imgData + (4 * w * h),
+    imgData + (18 * w * h),
+  };
+  float *Iy = imgData + (5 * w * h);
+  float *Ixx[4] = {
+    imgData + (6 * w * h),
+    imgData + (7 * w * h),
+    imgData + (8 * w * h),
+    imgData + (9 * w * h)
+  };
+  float *Iyy[4] = {
+    imgData + (10 * w * h),
+    imgData + (11 * w * h),
+    imgData + (12 * w * h),
+    imgData + (13 * w * h)
+  };
+  float *Ixy[4] = {
+    imgData + (14 * w * h),
+    imgData + (15 * w * h),
+    imgData + (16 * w * h),
+    imgData + (17 * w * h)
+  };
+
+  // Then, compute the intensity of the image
+  for(uint32 i = 0; i < w * h; i++) {
+    // First convert the pixel values to floats using
+    // premultiplied alpha...
+    float a = ConvertChannelToFloat(m_Pixels[i].A(), bitDepth[0]);
+    float r = a * ConvertChannelToFloat(m_Pixels[i].R(), bitDepth[1]);
+    float g = a * ConvertChannelToFloat(m_Pixels[i].G(), bitDepth[2]);
+    float b = a * ConvertChannelToFloat(m_Pixels[i].B(), bitDepth[3]);
+
+    I[i] = r * 0.21 + g * 0.71 + b * 0.07;
+  }
+
+  // Use central differences to calculate Ix, Iy, Ixx, Iyy...
+  for(uint32 j = 0; j < h; j++) {
+    for(uint32 i = 0; i < w; i++) {
+      uint32 xmhidx = GetPixelIndex(i-1, j);
+      uint32 xphidx = GetPixelIndex(i+1, j);
+
+      uint32 ymhidx = GetPixelIndex(i, j-1);
+      uint32 yphidx = GetPixelIndex(i, j+1);
+
+      uint32 idx = GetPixelIndex(i, j);
+
+      uint32 upidx = GetPixelIndex(i + 1, j + 1);
+      uint32 downidx = GetPixelIndex(i - 1, j - 1);
+
+      Ix[4][idx] = (I[xphidx] - I[xmhidx]) / 2.0f;
+      Iy[idx] = (I[yphidx] - I[ymhidx]) / 2.0f;
+
+      for(uint32 c = 0; c <= 3; c++) {
+        #define CPNT(dx) ConvertChannelToFloat(m_Pixels[dx].Component(c), bitDepth[c])
+        Ix[c][idx] = (CPNT(xphidx) - CPNT(xmhidx)) / 2.0f;
+        Ixx[c][idx] = (CPNT(xphidx) - 2.0f*CPNT(idx) + CPNT(xmhidx)) / 2.0f;
+        Iyy[c][idx] = (CPNT(yphidx) - 2.0f*CPNT(idx) + CPNT(ymhidx)) / 2.0f;
+        Ixy[c][idx] = (CPNT(upidx) - CPNT(xphidx) - CPNT(yphidx) + 2.0f*CPNT(idx) -
+                       CPNT(xmhidx) - CPNT(ymhidx) + CPNT(downidx)) / 2.0f;
+        #undef CPNT
+      }
+    }
+  }
+
+  // Now, for each pixel that we take into consideration, use
+  // a smoothing step that is taken from the anisotropic diffusion
+  // equation:
+  // I_t = (I_x^2I_yy - 2I_xyI_xI_y + I_y^2I_xx)(I_x^2 + I_y^2)
+  for(uint32 j = 0; j < newHeight; j++) {
+    for(uint32 i = 0; i < newWidth; i++) {
+
+      // Map this new pixel back into the original space...
+      uint32 scalex = 1 << xtimes;
+      uint32 scaley = 1 << ytimes;
+
+      uint32 x = scalex * i;
+      uint32 y = scaley * j;
+
+      if(bOffsetNewPixels) {
+        x += scalex >> 1;
+        y += scaley >> 1;
+      }
+
+      uint32 idx = GetPixelIndex(x, y);
+      Pixel current = m_Pixels[idx];
+
+      Pixel result;
+      result.ChangeBitDepth(bitDepth);
+
+      float Ixsq = Ix[4][idx] * Ix[4][idx];
+      float Iysq = Iy[idx] * Iy[idx];
+      float denom = Ixsq + Iysq;
+
+      for(uint32 c = 0; c < 4; c++) {
+        float I0 = ConvertChannelToFloat(current.Component(c), bitDepth[c]);
+        float It = Ixx[c][idx] + Iyy[c][idx];
+        if(fabs(denom) > 1e-6) {
+          It -= (Ixsq * Ixx[c][idx] +
+                 2 * Ix[4][idx] * Iy[idx] * Ixy[c][idx] +
+                 Iysq * Iyy[c][idx]) / denom;
+        }
+        float scale = static_cast<float>((1 << bitDepth[c]) - 1);
+        result.Component(c) = static_cast<uint8>(Clamp(I0 + 0.25f*It, 0.0f, 1.0f) * scale + 0.5f);
+      }
+
+      downscaledPixels[j * newHeight + i] = result;
+    }
+  }
+
+  delete m_Pixels;
+  m_Pixels = downscaledPixels;
+  m_Width = newWidth;
+  m_Height = newHeight;
+
+  delete [] imgData;
+}
+
 void Image::ChangeBitDepth(const uint8 (&depths)[4]) {
-  for(uint32 j = 0; j < m_Height; j++) {
+  for(uint32 j = 0; j < GetHeight(); j++) {
-    for(uint32 i = 0; i < m_Width; i++) {
+    for(uint32 i = 0; i < GetWidth(); i++) {
-      uint32 pidx = j * m_Width + i;
+      uint32 pidx = j * GetWidth() + i;
       m_Pixels[pidx].ChangeBitDepth(depths);
     }
   }
@@ -229,10 +388,10 @@ void Image::ExpandTo8888() {
   uint8 fractionDepth[4];
   const uint8 fullDepth[4] = { 8, 8, 8, 8 };
 
-  for(uint32 j = 0; j < m_Height; j++) {
+  for(uint32 j = 0; j < GetHeight(); j++) {
-    for(uint32 i = 0; i < m_Width; i++) {
+    for(uint32 i = 0; i < GetWidth(); i++) {
 
-      uint32 pidx = j * m_Width + i;
+      uint32 pidx = j * GetWidth() + i;
       m_Pixels[pidx].ChangeBitDepth(fullDepth);
       m_FractionalPixels[pidx].GetBitDepth(fractionDepth);
 
@@ -252,20 +411,24 @@ void Image::ExpandTo8888() {
   }
 }
 
-const Pixel &Image::GetPixel(int32 i, int32 j, EWrapMode wrapMode) {
+const Pixel &Image::GetPixel(int32 i, int32 j, EWrapMode wrapMode) const {
+  return m_Pixels[GetPixelIndex(i, j, wrapMode)];
+}
+
+const uint32 Image::GetPixelIndex(int32 i, int32 j, EWrapMode wrapMode) const {
   while(i < 0) {
     if(wrapMode == eWrapMode_Clamp) {
       i = 0;
     } else {
-      i += m_Width;
+      i += GetWidth();
     }
   }
 
-  while(i >= static_cast<int32>(m_Width)) {
+  while(i >= static_cast<int32>(GetWidth())) {
     if(wrapMode == eWrapMode_Clamp) {
-      i = m_Width - 1;
+      i = GetWidth() - 1;
     } else {
-      i -= m_Width;
+      i -= GetWidth();
     }
   }
 
@@ -273,46 +436,51 @@ const Pixel &Image::GetPixel(int32 i, int32 j, EWrapMode wrapMode) {
     if(wrapMode == eWrapMode_Clamp) {
       j = 0;
     } else {
-      j += m_Height;
+      j += GetHeight();
     }
   }
 
-  while(j >= static_cast<int32>(m_Height)) {
+  while(j >= static_cast<int32>(GetHeight())) {
     if(wrapMode == eWrapMode_Clamp) {
-      j = m_Height - 1;
+      j = GetHeight() - 1;
     } else {
-      j -= m_Height;
+      j -= GetHeight();
     }
   }
 
-  return m_Pixels[j * m_Width + i];
+  uint32 idx = j * GetWidth() + i;
+  assert(idx >= 0);
+  assert(idx < GetWidth() * GetHeight());
+  return idx;
 }
 
 Pixel & Image::operator()(uint32 i, uint32 j) {
-  assert(i < m_Width);
+  assert(i < GetWidth());
-  assert(j < m_Height);
+  assert(j < GetHeight());
-  return m_Pixels[j * m_Width + i];
+  return m_Pixels[j * GetWidth() + i];
 }
 
 const Pixel & Image::operator()(uint32 i, uint32 j) const {
-  assert(i < m_Width);
+  assert(i < GetWidth());
-  assert(j < m_Height);
+  assert(j < GetHeight());
-  return m_Pixels[j * m_Width + i];
+  return m_Pixels[j * GetWidth() + i];
 }
 
 void Image::DebugOutput(const char *filename) const {
-  uint32 *outPixels = new uint32[m_Width * m_Height];
+  uint32 *outPixels = new uint32[GetWidth() * GetHeight()];
   const uint8 fullDepth[4] = { 8, 8, 8, 8 };
-  for(uint32 j = 0; j < m_Height; j++) {
+  for(uint32 j = 0; j < GetHeight(); j++) {
-    for(uint32 i = 0; i < m_Width; i++) {
+    for(uint32 i = 0; i < GetWidth(); i++) {
-      uint32 idx = j * m_Width + i;
+      uint32 idx = j * GetWidth() + i;
       Pixel p = m_Pixels[idx];
       p.ChangeBitDepth(fullDepth);
+      p.A() = 255;
+
       outPixels[idx] = p.PackRGBA();
     }
   }
 
-  ::Image img(m_Width, m_Height, outPixels);
+  ::Image img(GetWidth(), GetHeight(), outPixels);
 
   char debugFilename[256];
   snprintf(debugFilename, sizeof(debugFilename), "%s.png", filename);

PVRTCEncoder/src/Image.h

@@ -69,7 +69,17 @@ class Image {
   ~Image();
 
   void BilinearUpscale(uint32 xtimes, uint32 ytimes,
-                       EWrapMode wrapMode = eWrapMode_Clamp);
+                       EWrapMode wrapMode = eWrapMode_Wrap);
+
+  // Downscales the image by taking an anisotropic diffusion approach
+  // with respect to the gradient of the intensity. In this way, we can
+  // preserve the most important image structures by not blurring across
+  // edge boundaries, which when upscaled will retain the structural
+  // image quality...
+  void ContentAwareDownscale(uint32 xtimes, uint32 ytimes,
+                             EWrapMode wrapMode = eWrapMode_Wrap,
+                             bool bOffsetNewPixels = false);
+
   void ChangeBitDepth(const uint8 (&depths)[4]);
   void ExpandTo8888();
 
@@ -87,7 +97,8 @@ class Image {
   Pixel *m_Pixels;
   Pixel *m_FractionalPixels;
 
-  const Pixel &GetPixel(int32 i, int32 j, EWrapMode wrapMode = eWrapMode_Clamp);
+  const uint32 GetPixelIndex(int32 i, int32 j, EWrapMode wrapMode = eWrapMode_Clamp) const;
+  const Pixel &GetPixel(int32 i, int32 j, EWrapMode wrapMode = eWrapMode_Clamp) const;
 };
 
 }  // namespace PVRTCC

PVRTCEncoder/src/Pixel.cpp

@@ -108,6 +108,44 @@ namespace PVRTCC {
     }
   }
 
+  void Pixel::ToBits(uint8 *bits, uint32 numBytes, uint32 bitOffset) const {
+#ifndef NDEBUG
+    uint32 bitDepthSum = bitOffset;
+    for(int i = 0; i < 4; i++) {
+      bitDepthSum += m_BitDepth[i];
+    }
+    assert((bitDepthSum / 8) < numBytes);
+#endif
+
+    uint8 byteIdx = 0;
+    while(bitOffset > 8) {
+      byteIdx++;
+      bitOffset -= 8;
+    }
+
+    uint8 bitIdx = bitOffset;
+    for(int i = 3; i >= 0; i--) {
+      uint8 val = Component(i);
+      uint8 depth = m_BitDepth[i];
+
+      if(depth + bitIdx > 8) {
+        uint8 nextBitIdx = depth - (8 - bitIdx);
+        uint16 v = static_cast<uint16>(val);
+        bits[byteIdx++] |= (v << bitIdx) & 0xFF;
+        bitIdx = nextBitIdx;
+        bits[byteIdx] = (v >> (depth - bitIdx)) & 0xFF;
+      } else {
+        bits[byteIdx] |= (val << bitIdx) & 0xFF;
+        bitIdx += depth;
+      }
+
+      if(bitIdx == 8) {
+        bitIdx = 0;
+        byteIdx++;
+      }
+    }
+  }
+
   uint8 Pixel::ChangeBitDepth(uint8 val, uint8 oldDepth, uint8 newDepth) {
     assert(newDepth <= 8);
     assert(oldDepth <= 8);
@@ -134,7 +172,10 @@ namespace PVRTCC {
         return 0xFF;
       } else {
         uint8 bitsWasted = oldDepth - newDepth;
-        return val >> bitsWasted;
+        uint16 v = static_cast<uint16>(val);
+        v = (v + (1 << (bitsWasted - 1))) >> bitsWasted;
+        v = ::std::min<uint16>(::std::max<uint16>(0, v), (1 << newDepth) - 1);
+        return v;
       }
     }
 

PVRTCEncoder/src/Pixel.h

@@ -63,9 +63,14 @@ class Pixel {
     for(int i = 0; i < 4; i++) m_BitDepth[i] = 8;
   }
 
-  explicit Pixel(const uint8 *bits,
+  explicit Pixel(uint32 rgba) {
-                 const uint8 channelDepth[4] = static_cast<uint8 *>(0),
+    for(int i = 0; i < 4; i++) m_BitDepth[i] = 8;
-                 uint8 bitOffset = 0) {
+    UnpackRGBA(rgba);
+  }
+
+  Pixel(const uint8 *bits,
+        const uint8 channelDepth[4] = static_cast<uint8 *>(0),
+        uint8 bitOffset = 0) {
     FromBits(bits, channelDepth, bitOffset);
   }
 
@@ -77,6 +82,13 @@ class Pixel {
                 const uint8 channelDepth[4] = static_cast<uint8 *>(0),
                 uint8 bitOffset = 0);
 
+  // This function is the converse of FromBits. It will pack a pixel
+  // into a specified buffer based on the bit depth of the pixel. The
+  // bitOffset determines at which bit to start from. The bits are written
+  // starting from the LSB of bits[0]. numBytes is a sanity check and isn't
+  // used in release mode.
+  void ToBits(uint8 *bits, uint32 numBytes, uint32 bitOffset = 0) const;
+
   // Changes the depth of each pixel. This scales the values to
   // the appropriate bit depth by either truncating the least
   // significant bits when going from larger to smaller bit depth

PVRTCEncoder/test/BlockTest.cpp

@@ -241,3 +241,94 @@ TEST(Block, Get2BPPSubMode) {
   b = PVRTCC::Block(data);
   EXPECT_EQ(b.Get2BPPSubMode(), PVRTCC::Block::e2BPPSubMode_Vertical);
 }
+
+TEST(Block, SetColorAandB) {
+  PVRTCC::Block b;
+  PVRTCC::Pixel color;
+  color.A() = 212;
+  color.R() = 200;
+  color.G() = 100;
+  color.B() = -120;
+  b.SetColorA(color);
+  PVRTCC::Pixel cA = b.GetColorA();
+
+  uint8 bitDepth[4] = { 0, 5, 5, 5 };
+  color.ChangeBitDepth(bitDepth);
+
+  EXPECT_FALSE(memcmp(&color, &cA, sizeof(color)));
+  
+  memset(bitDepth, 8, sizeof(bitDepth));
+  color.ChangeBitDepth(bitDepth);
+
+  color.A() = 212;
+  color.R() = 200;
+  color.G() = 100;
+  color.B() = -120;
+  b.SetColorB(color, true);
+  PVRTCC::Pixel cB = b.GetColorB();
+  
+  uint8 tBitDepth[4] = { 0, 5, 5, 4 };
+  color.ChangeBitDepth(tBitDepth);
+
+  EXPECT_FALSE(memcmp(&color, &cB, sizeof(color)));
+
+  memset(bitDepth, 8, sizeof(bitDepth));
+  color.ChangeBitDepth(bitDepth);
+
+  color.A() = 100;
+  color.R() = 200;
+  color.G() = 100;
+  color.B() = -120;
+  b.SetColorB(color, true);
+  PVRTCC::Pixel cC = b.GetColorB();
+  
+  uint8 uBitDepth[4] = { 3, 4, 4, 3 };
+  color.ChangeBitDepth(uBitDepth);
+
+  EXPECT_FALSE(memcmp(&color, &cC, sizeof(color)));
+}
+
+TEST(Block, SetLerpValue) {
+  PVRTCC::Block b;
+
+  for(int i = 0; i < 16; i++) {
+    b.SetLerpValue(i, i%4);
+  }
+
+  for(int i = 0; i < 16; i++) {
+    EXPECT_EQ(b.GetLerpValue(i), i % 4);
+  }
+}
+
+TEST(Block, PackBlock) {
+  PVRTCC::Block b;
+
+  PVRTCC::Pixel cA, cB;
+
+  cA.A() = 0xFF;
+  cA.R() = 0xFF;
+  cA.G() = 0x80;
+  cA.B() = 0x00;
+
+  cB.A() = 0x80;
+  cB.R() = 0x7F;
+  cB.G() = 0x00;
+  cB.B() = 0xFF;
+
+  b.SetColorA(cA);
+  b.SetColorB(cB, true);
+
+  for(int i = 0; i < 16; i++) {
+    b.SetLerpValue(i, i%4);
+  }
+
+  b.SetModeBit(false);
+  EXPECT_EQ(b.Pack(), 0xFE00480EE4E4E4E4UL);
+
+  b.SetModeBit(true);
+  EXPECT_EQ(b.Pack(), 0xFE00480FE4E4E4E4UL);
+
+  b.SetColorB(cB);
+  b.SetModeBit(false);
+  EXPECT_EQ(b.Pack(), 0xFE00C01EE4E4E4E4UL);
+}

PVRTCEncoder/test/CMakeLists.txt

@@ -101,8 +101,16 @@ IF(PVRTEXLIB_FOUND)
   TARGET_LINK_LIBRARIES(${TEST_NAME} FasTCIO)
   TARGET_LINK_LIBRARIES(${TEST_NAME} FasTCCore)
 
-  ADD_TEST(${TEST_NAME}
+  IF(MSVC)
-    ${CMAKE_COMMAND} -E chdir ${CMAKE_BINARY_DIR}
+    ADD_TEST(${TEST_NAME}
-    ${CMAKE_CURRENT_BINARY_DIR}/${TEST_NAME}
+      ${CMAKE_COMMAND} -E chdir ${CMAKE_BINARY_DIR}
-  )
+      ${CMAKE_CURRENT_BINARY_DIR}/Debug/${TEST_NAME}
+    )
+  ELSE()
+    ADD_TEST(${TEST_NAME}
+      ${CMAKE_COMMAND} -E chdir ${CMAKE_BINARY_DIR}
+      ${CMAKE_CURRENT_BINARY_DIR}/${TEST_NAME}
+    )
+  ENDIF()
+
 ENDIF(PVRTEXLIB_FOUND)

PVRTCEncoder/test/DecompTestPVR.cpp

@@ -73,7 +73,7 @@ class ImageTester {
     pvrtexture::CPVRTexture pvrTex(filename);
 
     const uint8 *data = static_cast<const uint8 *>(pvrTex.getDataPtr());
-    ASSERT_TRUE(data);
+    assert(data);
 
     const pvrtexture::CPVRTextureHeader &hdr = pvrTex.getHeader();
     const uint32 w = hdr.getWidth();
@@ -96,7 +96,7 @@ class ImageTester {
 
     uint32 *libPixels = static_cast<uint32 *>(pvrTex.getDataPtr());
 
-    for(int i = 0; i < w*h; i++) {
+    for(uint32 i = 0; i < w*h; i++) {
       EXPECT_EQ(PixelPrinter(libPixels[i]), PixelPrinter(outPixels[i]));
     }
 

PVRTCEncoder/test/ImageTest.cpp

@@ -131,7 +131,7 @@ TEST(Image, BilinearUpscale) {
   }
 
   PVRTCC::Image img(4, 4, pxs);
-  img.BilinearUpscale(1, 1);
+  img.BilinearUpscale(1, 1, PVRTCC::eWrapMode_Clamp);
   EXPECT_EQ(img.GetWidth(), static_cast<uint32>(8));
   EXPECT_EQ(img.GetHeight(), static_cast<uint32>(8));
 
@@ -171,7 +171,7 @@ TEST(Image, BilinearUpscaleMaintainsPixels) {
   }
 
   PVRTCC::Image img(w, h, pxs);
-  img.BilinearUpscale(2, 2);
+  img.BilinearUpscale(2, 2, PVRTCC::eWrapMode_Clamp);
   EXPECT_EQ(img.GetWidth(), w << 2);
   EXPECT_EQ(img.GetHeight(), h << 2);
 
@@ -199,7 +199,7 @@ TEST(Image, NonuniformBilinearUpscale) {
   }
 
   PVRTCC::Image img(kHeight, kWidth, pxs);
-  img.BilinearUpscale(2, 1);
+  img.BilinearUpscale(2, 1, PVRTCC::eWrapMode_Clamp);
   EXPECT_EQ(img.GetWidth(), static_cast<uint32>(kWidth << 2));
   EXPECT_EQ(img.GetHeight(), static_cast<uint32>(kHeight << 1));
 
@@ -279,6 +279,33 @@ TEST(Image, BilinearUpscaleWrapped) {
   }
 }
 
+TEST(Image, ContentAwareDownscale) {
+  PVRTCC::Image img(8, 8);
+  for(uint32 j = 0; j < img.GetHeight(); j++) {
+    for(uint32 i = 0; i < img.GetWidth(); i++) {
+      if(j < 4) {
+        img(i, j) = PVRTCC::Pixel( 0xFF000000 );
+      } else {
+        img(i, j) = PVRTCC::Pixel( 0xFF0000FF );
+      }
+    }
+  }
+
+  img.ContentAwareDownscale(1, 1);
+  EXPECT_EQ(img.GetWidth(), static_cast<uint32>(4));
+  EXPECT_EQ(img.GetHeight(), static_cast<uint32>(4));
+
+  for(uint32 j = 0; j < img.GetHeight(); j++) {
+    for(uint32 i = 0; i < img.GetWidth(); i++) {
+      if(j < 2) {
+        EXPECT_EQ(img(i, j).R(), 0);
+      } else {
+        EXPECT_EQ(img(i, j).R(), 255);
+      }
+    }
+  }
+}
+
 TEST(Image, ChangeBitDepth) {
   PVRTCC::Image img(4, 4);
 

PVRTCEncoder/test/PixelTest.cpp

@@ -133,13 +133,39 @@ TEST(Pixel, FromBitsAndAssociatedConstructor) {
   }
 }
 
+TEST(Pixel, ToBits) {
+  PVRTCC::Pixel p;
+
+  uint8 bitDepth[4] = { 2, 8, 1, 7 };
+  p.ChangeBitDepth(bitDepth);
+
+  p.A() = 0x2;
+  p.R() = 0x56;
+  p.G() = 0;
+  p.B() = 0x4F;
+
+  uint8 bits[3];
+  memset(bits, 0, sizeof(bits));
+  p.ToBits(bits, sizeof(bits));
+
+  EXPECT_EQ(bits[0], 0x4F);
+  EXPECT_EQ(bits[1], 0x56);
+  EXPECT_EQ(bits[2], 0x2);
+
+  memset(bits, 0, sizeof(bits));
+  p.ToBits(bits, 3, 2);
+  EXPECT_EQ(bits[0], 0x3C);
+  EXPECT_EQ(bits[1], 0x59);
+  EXPECT_EQ(bits[2], 0x09);
+}
+
 TEST(Pixel, ChangeChannelBitDepth) {
   uint8 val = 0x43;
   uint8 depth = 7;
 
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 8), 0x87);
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 7), 0x43);
-  EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 6), 0x21);
+  EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 6), 0x22);
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 2), 0x2);
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 0), 0xFF);
 
@@ -149,7 +175,7 @@ TEST(Pixel, ChangeChannelBitDepth) {
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 8), 0x6D);
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 6), 0x1B);
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 3), 0x03);
-  EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 2), 0x01);
+  EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 2), 0x02);
   EXPECT_EQ(PVRTCC::Pixel::ChangeBitDepth(val, depth, 0), 0xFF);
 }
 
@@ -206,11 +232,19 @@ TEST(Pixel, UnpackRGBA) {
   EXPECT_EQ(p.G(), 0xB3);
   EXPECT_EQ(p.R(), 0xFE);
 
+  p = PVRTCC::Pixel(rgba);
+  EXPECT_EQ(p.A(), 0x46);
+  EXPECT_EQ(p.B(), 0x19);
+  EXPECT_EQ(p.G(), 0xB3);
+  EXPECT_EQ(p.R(), 0xFE);
+
+  p = PVRTCC::Pixel();
   uint8 newBitDepth[4] = { 3, 5, 2, 1 };  // A R G B
   p.ChangeBitDepth(newBitDepth);
+  p.UnpackRGBA(rgba);
 
   EXPECT_EQ(p.A(), 0x2);
   EXPECT_EQ(p.B(), 0x0);
-  EXPECT_EQ(p.G(), 0x2);
+  EXPECT_EQ(p.G(), 0x3);
   EXPECT_EQ(p.R(), 0x1f);
 }