Deal with this bug once and for all. If we have an image in block stream order, then explicitly reorder it before doing any work. Then keep it that way. I probably could have fixed this in the amount of time I've wasted on it. -____-

CLTool/src/clunix.cpp

@@ -206,6 +206,9 @@ int main(int argc, char **argv) {
   }
 
   const Image *img = file.GetImage();
+  if(format == eCompressionFormat_PVRTC) {
+    const_cast<Image *>(img)->SetBlockStreamOrder(false);
+  }
 
   int numBlocks = (img->GetWidth() * img->GetHeight())/16;
   BlockStatManager *statManager = NULL;
@@ -247,7 +250,6 @@ int main(int argc, char **argv) {
   if(format == eCompressionFormat_BPTC) {
     strcat(basename, "-bc7.png");
   } else if(format == eCompressionFormat_PVRTC) {
-    cImg.SetBlockStreamOrder(false);
     strcat(basename, "-pvrtc.png");
   }
 

CLTool/src/clwin32.cpp

@@ -212,6 +212,9 @@ int _tmain(int argc, _TCHAR* argv[])
   }
 
   const Image *img = file.GetImage();
+  if(format == eCompressionFormat_PVRTC) {
+    const_cast<Image *>(img)->SetBlockStreamOrder(false);
+  }
 
   int numBlocks = (img->GetWidth() * img->GetHeight())/16;
   BlockStatManager *statManager = NULL;
@@ -253,7 +256,6 @@ int _tmain(int argc, _TCHAR* argv[])
   if(format == eCompressionFormat_BPTC) {
     strcat_s(basename, "-bc7.png");
   } else if(format == eCompressionFormat_PVRTC) {
-    cImg.SetBlockStreamOrder(false);
     strcat_s(basename, "-pvrtc.png");
   }
 

Core/include/CompressedImage.h

@@ -97,6 +97,8 @@ class CompressedImage {
   // !FIXME! We should have a function to explicitly return the in/out buf
   // size for a given compressed image.
   bool DecompressImage(uint8 *outBuf, uint32 outBufSz) const;
+
+  ECompressionFormat GetFormat() const { return m_Format; }
 };
 
 #endif // _COMPRESSED_IMAGE_H_

Core/include/Image.h

@@ -70,7 +70,13 @@ class Image {
   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; }
 
  private:
@@ -80,6 +86,9 @@ class Image {
   bool m_bBlockStreamOrder;
 
   uint8 *m_PixelData;
+
+  void ConvertToBlockStreamOrder();
+  void ConvertFromBlockStreamOrder();
 };
 
 #endif // __TEXCOMP_IMAGE_H__

Core/src/Image.cpp

@@ -117,6 +117,11 @@ Image::Image(const CompressedImage &ci)
     fprintf(stderr, "Error decompressing image!\n");
     return;
   }
+
+  // !HACK!
+  if(ci.GetFormat() == eCompressionFormat_PVRTC) {
+    m_bBlockStreamOrder = false;
+  }
 }
 
 Image::Image(const ImageLoader &loader) 
@@ -207,3 +212,55 @@ double Image::ComputePSNR(const CompressedImage &ci) const {
   delete unCompData;
   return PSNR;
 }
+
+void Image::ConvertToBlockStreamOrder() {
+  if(m_bBlockStreamOrder || !m_PixelData)
+    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_PixelData)[y*GetWidth() + x];
+      }
+    }
+  }
+
+  delete m_PixelData;
+  m_PixelData = reinterpret_cast<uint8 *>(newPixelData);
+  m_bBlockStreamOrder = true;
+}
+
+void Image::ConvertFromBlockStreamOrder() {
+  if(!m_bBlockStreamOrder || !m_PixelData)
+    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_PixelData)[offset + t];
+      }
+    }
+  }
+
+  delete m_PixelData;
+  m_PixelData = reinterpret_cast<uint8 *>(newPixelData);
+  m_bBlockStreamOrder = false;
+}

PVRTCEncoder/src/Compressor.cpp

@@ -137,14 +137,8 @@ namespace PVRTCC {
     Image img(dcj.height, dcj.width);
     uint32 nPixels = dcj.height * dcj.width;
     for(uint32 i = 0; i < nPixels; i++) {
-      // Assume block stream order (whyyyy)
-      uint32 blockIdx = i / 16;
-      uint32 blockWidth = dcj.width / 4;
-      uint32 blockX = blockIdx % blockWidth;
-      uint32 blockY = blockIdx / blockWidth;
-
-      uint32 x = blockX * 4 + (i % 4);
-      uint32 y = blockY * 4 + (i % 16) / 4;
+      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]);