drewcassidy/nvidia-texture-tools
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Update 2.0 project file.

Browse Source 
Fix backported cuda kernel.
This commit is contained in:
castano
2008-06-09 19:05:49 +00:00
parent 6e1a87129f
commit 0b17e3095a
2 changed files with 43 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
src/nvtt/cuda/CompressKernel.cu
View File

@ -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
////////////////////////////////////////////////////////////////////////////////