Update 2.0 project file.

Fix backported cuda kernel.

project/vc8/nvtt/nvtt.vcproj

@@ -711,7 +711,7 @@
 					>
 					<Tool
 						Name="VCCustomBuildTool"
-						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -m32 -ccbin &quot;$(VCInstallDir)bin&quot; -c -DNDEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MD -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu"
+						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -m32 -ccbin &quot;$(VCInstallDir)bin&quot; -c -DNDEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MD -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
 						AdditionalDependencies="CudaMath.h"
 						Outputs="$(IntDir)\$(InputName).obj"
 					/>
@@ -849,10 +849,6 @@
 				RelativePath="..\..\..\src\nvtt\cuda\CudaUtils.cpp"
 				>
 			</File>
-			<File
-				RelativePath="..\..\..\src\nvtt\FastCompressDXT.cpp"
-				>
-			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\InputOptions.cpp"
 				>
@@ -865,6 +861,10 @@
 				RelativePath="..\..\..\src\nvtt\nvtt_wrapper.cpp"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvtt\OptimalCompressDXT.cpp"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\OutputOptions.cpp"
 				>
@@ -911,10 +911,6 @@
 				RelativePath="..\..\..\src\nvtt\cuda\CudaUtils.h"
 				>
 			</File>
-			<File
-				RelativePath="..\..\..\src\nvtt\FastCompressDXT.h"
-				>
-			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\InputOptions.h"
 				>
@@ -927,6 +923,10 @@
 				RelativePath="..\..\..\src\nvtt\nvtt_wrapper.h"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvtt\OptimalCompressDXT.h"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\OutputOptions.h"
 				>

src/nvtt/cuda/CompressKernel.cu

@@ -594,6 +594,40 @@ __device__ void evalAllPermutations(const float3 * colors, const float * weights
 }
 */
 
+__device__ void evalLevel4Permutations(const float3 * colors, float3 colorSum, const uint * permutations, ushort & bestStart, ushort & bestEnd, uint & bestPermutation, float * errors)
+{
+	const int idx = threadIdx.x;
+	
+	float bestError = FLT_MAX;
+	
+	for(int i = 0; i < 16; i++)
+	{
+		int pidx = idx + NUM_THREADS * i;
+		if (pidx >= 992) break;
+		
+		ushort start, end;
+		uint permutation = permutations[pidx];
+
+		float error = evalPermutation4(colors, colorSum, permutation, &start, &end);
+		
+		if (error < bestError)
+		{
+			bestError = error;
+			bestPermutation = permutation;
+			bestStart = start;
+			bestEnd = end;
+		}
+	}
+
+	if (bestStart < bestEnd)
+	{
+		swap(bestEnd, bestStart);
+		bestPermutation ^= 0x55555555;	// Flip indices.
+	}
+
+	errors[idx] = bestError;
+}
+
 __device__ void evalLevel4Permutations(const float3 * colors, const float * weights, float3 colorSum, const uint * permutations, ushort & bestStart, ushort & bestEnd, uint & bestPermutation, float * errors)
 {
 	const int idx = threadIdx.x;
@@ -629,7 +663,6 @@ __device__ void evalLevel4Permutations(const float3 * colors, const float * weig
 }
 
 
-
 ////////////////////////////////////////////////////////////////////////////////
 // Find index with minimum error
 ////////////////////////////////////////////////////////////////////////////////