Add option to select decompression algorithm to test suite.

src/nvimage/BlockDXT.cpp

@@ -39,9 +39,9 @@ using namespace nv;
 uint BlockDXT1::evaluatePalette(Color32 color_array[4]) const
 {
 	// Does bit expansion before interpolation.
-	color_array[0].b = (col0.b << 3) | (col0.b >> 2);
-	color_array[0].g = (col0.g << 2) | (col0.g >> 4);
 	color_array[0].r = (col0.r << 3) | (col0.r >> 2);
+	color_array[0].g = (col0.g << 2) | (col0.g >> 4);
+	color_array[0].b = (col0.b << 3) | (col0.b >> 2);
 	color_array[0].a = 0xFF;
 	
 	// @@ Same as above, but faster?
@@ -97,14 +97,14 @@ uint BlockDXT1::evaluatePalette(Color32 color_array[4]) const
 uint BlockDXT1::evaluatePaletteNV5x(Color32 color_array[4]) const
 {
 	// Does bit expansion before interpolation.
-	color_array[0].b = col0.b * 22 / 8;
+	color_array[0].r = (3 * col0.r * 22) / 8;
 	color_array[0].g = (col0.g << 2) | (col0.g >> 4);
-	color_array[0].r = col0.r * 22 / 8;
+	color_array[0].b = (3 * col0.b * 22) / 8;
 	color_array[0].a = 0xFF;
-	
+
-	color_array[1].r = col1.r * 22 / 8;
+	color_array[1].r = (3 * col1.r * 22) / 8;
 	color_array[1].g = (col1.g << 2) | (col1.g >> 4);
-	color_array[1].b = col1.b * 22 / 8;
+	color_array[1].b = (3 * col1.b * 22) / 8;
 	color_array[1].a = 0xFF;
 	
 	if( col0.u > col1.u ) {
@@ -117,8 +117,8 @@ uint BlockDXT1::evaluatePaletteNV5x(Color32 color_array[4]) const
 		color_array[3].r = (2 * col1.r + col0.r) * 22 / 8;
 		color_array[3].g = (256 * color_array[1].g + (color_array[0].g - color_array[1].g)/4 + 128 + (color_array[0].g - color_array[1].g) * 80) / 256;
 		color_array[3].b = (2 * col1.b + col0.b) * 22 / 8;
-		color_array[3].a = 0xFF;
 		
+		color_array[3].a = 0xFF;
 		return 4;
 	}
 	else {
@@ -207,6 +207,25 @@ void BlockDXT1::decodeBlock(ColorBlock * block) const
 	}	
 }
 
+void BlockDXT1::decodeBlockNV5x(ColorBlock * block) const
+{
+	nvDebugCheck(block != NULL);
+
+	// Decode color block.
+	Color32 color_array[4];
+	evaluatePalette(color_array);
+	evaluatePaletteNV5x(color_array);
+
+	// Write color block.
+	for( uint j = 0; j < 4; j++ ) {
+		for( uint i = 0; i < 4; i++ ) {
+			uint idx = (row[j] >> (2 * i)) & 3;
+			block->color(i, j) = color_array[idx];
+		}
+	}
+}
+
+
 void BlockDXT1::setIndices(int * idx)
 {
 	indices = 0;

src/nvimage/BlockDXT.h

@@ -53,6 +53,7 @@ namespace nv
 		void evaluatePalette4(Color32 color_array[4]) const;
 		
 		void decodeBlock(ColorBlock * block) const;
+		void decodeBlockNV5x(ColorBlock * block) const;
 		
 		void setIndices(int * idx);
 

src/nvtt/tests/testsuite.cpp

@@ -119,6 +119,11 @@ static ImageSet s_imageSets[] = {
 };
 const int s_imageSetCount = sizeof(s_imageSets)/sizeof(s_imageSets[0]);
 
+enum Decoder
+{
+	Decoder_Reference,
+	Decoder_NVIDIA,
+};
 
 struct MyOutputHandler : public nvtt::OutputHandler
 {
@@ -145,7 +150,7 @@ struct MyOutputHandler : public nvtt::OutputHandler
 		return true;
 	}
 
-	Image * decompress(nvtt::Format format)
+	Image * decompress(nvtt::Format format, Decoder decoder)
 	{
 		int bw = (m_width + 3) / 4;
 		int bh = (m_height + 3) / 4;
@@ -162,7 +167,12 @@ struct MyOutputHandler : public nvtt::OutputHandler
 				for (int x = 0; x < bw; x++)
 				{
 					ColorBlock colors;
-					block->decodeBlock(&colors);
+					if (decoder == Decoder_Reference) {
+						block->decodeBlock(&colors);
+					}
+					else if (decoder == Decoder_NVIDIA) {
+						block->decodeBlockNV5x(&colors);
+					}
 
 					for (int yy = 0; yy < 4; yy++)
 					{
@@ -235,6 +245,7 @@ int main(int argc, char *argv[])
 	bool fast = false;
 	bool nocuda = false;
 	bool showHelp = false;
+	Decoder decoder = Decoder_Reference;
 	const char * basePath = "";
 	const char * outPath = "output";
 	const char * regressPath = NULL;
@@ -249,6 +260,13 @@ int main(int argc, char *argv[])
 				i++;
 			}
 		}
+		else if (strcmp("-dec", argv[i]) == 0)
+		{
+			if (i+1 < argc && argv[i+1][0] != '-') {
+				decoder = (Decoder)atoi(argv[i+1]);
+				i++;
+			}
+		}
 		else if (strcmp("-fast", argv[i]) == 0)
 		{
 			fast = true;
@@ -289,9 +307,16 @@ int main(int argc, char *argv[])
 		printf("usage: nvtestsuite [options]\n\n");
 		
 		printf("Input options:\n");
-		printf("  -path <path>\tInput image path.\n");
+		printf("  -path <path>   \tInput image path.\n");
-		printf("  -set [0:2]\tImage set.\n");
 		printf("  -regress <path>\tRegression directory.\n");
+		printf("  -set [0:2]     \tImage set.\n");
+		printf("    0:           \tKodak.\n");
+		printf("    1:           \tWaterloo.\n");
+		printf("    2:           \tEpic.\n");
+		printf("    3:           \tFarbrausch.\n");
+		printf("  -dec x         \tDecompressor.\n");
+		printf("    0:           \tReference.\n");
+		printf("    1:           \tNVIDIA.\n");
 
 		printf("Compression options:\n");
 		printf("  -fast          \tFast compression.\n");
@@ -365,7 +390,7 @@ int main(int argc, char *argv[])
 		printf("  Time: \t%.3f sec\n", float(end-start) / CLOCKS_PER_SEC);
 		totalTime += float(end-start);
 
-		AutoPtr<Image> img_out( outputHandler.decompress(nvtt::Format_BC1) );
+		AutoPtr<Image> img_out( outputHandler.decompress(nvtt::Format_BC1, decoder) );
 
 		Path outputFileName;
 		outputFileName.format("%s/%s", outPath, fileNames[i]);