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

Cleanup color rounding and expansion.

Browse Source
This commit is contained in:
castano 2009-01-19 10:42:31 +00:00
parent f753cc9702
commit 49409e9d92
1 changed files with 17 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
src/nvtt/cuda/CompressKernel.cu
View File

@ -305,7 +305,6 @@ __device__ void loadColorBlock(const uint * image, float2 colors[16], float2 sum
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
inline __device__ float3 roundAndExpand565(float3 v, ushort * w) inline __device__ float3 roundAndExpand565(float3 v, ushort * w)
{ {
#if 0
uint x = __float2uint_rn(__saturatef(v.x) * 31.0f); uint x = __float2uint_rn(__saturatef(v.x) * 31.0f);
uint y = __float2uint_rn(__saturatef(v.y) * 63.0f); uint y = __float2uint_rn(__saturatef(v.y) * 63.0f);
uint z = __float2uint_rn(__saturatef(v.z) * 31.0f); uint z = __float2uint_rn(__saturatef(v.z) * 31.0f);
@ -313,35 +312,26 @@ inline __device__ float3 roundAndExpand565(float3 v, ushort * w)
v.x = __uint2float_rn(x) * 1.0f / 31.0f; v.x = __uint2float_rn(x) * 1.0f / 31.0f;
v.y = __uint2float_rn(y) * 1.0f / 63.0f; v.y = __uint2float_rn(y) * 1.0f / 63.0f;
v.z = __uint2float_rn(z) * 1.0f / 31.0f; v.z = __uint2float_rn(z) * 1.0f / 31.0f;
#else
v.x = rintf(__saturatef(v.x) * 31.0f);
v.y = rintf(__saturatef(v.y) * 63.0f);
v.z = rintf(__saturatef(v.z) * 31.0f);
*w = ((ushort)v.x << 11) | ((ushort)v.y << 5) | (ushort)v.z;
v.x *= 1.0f / 31.0f;
v.y *= 1.0f / 63.0f;
v.z *= 1.0f / 31.0f;
#endif
return v; return v;
} }
inline __device__ float2 roundAndExpand56(float2 v, ushort * w) inline __device__ float2 roundAndExpand56(float2 v, ushort * w)
{ {
v.x = rintf(__saturatef(v.x) * 31.0f); uint x = __float2uint_rn(__saturatef(v.x) * 31.0f);
v.y = rintf(__saturatef(v.y) * 63.0f); uint y = __float2uint_rn(__saturatef(v.y) * 63.0f);
*w = ((ushort)v.x << 11) | ((ushort)v.y << 5); *w = (x << 11) | (y << 5);
v.x *= 1.0f / 31.0f; v.x = __uint2float_rn(x) * 1.0f / 31.0f;
v.y *= 1.0f / 63.0f; v.y = __uint2float_rn(y) * 1.0f / 63.0f;
return v; return v;
} }
inline __device__ float2 roundAndExpand88(float2 v, ushort * w) inline __device__ float2 roundAndExpand88(float2 v, ushort * w)
{ {
v.x = rintf(__saturatef(v.x) * 255.0f); uint x = __float2uint_rn(__saturatef(v.x) * 255.0f);
v.y = rintf(__saturatef(v.y) * 255.0f); uint y = __float2uint_rn(__saturatef(v.y) * 255.0f);
*w = ((ushort)v.x << 8) | ((ushort)v.y); *w = (x << 8) | y;
v.x *= 1.0f / 255.0f; v.x = __uint2float_rn(x) * 1.0f / 255.0f;
v.y *= 1.0f / 255.0f; v.y = __uint2float_rn(y) * 1.0f / 255.0f;
return v; return v;
} }
@ -1551,13 +1541,13 @@ __global__ void compressDXT5(const uint * permutations, const uint * image, uint
/*__device__ void evaluatePalette(uint alpha0, uint alpha1, uint alphas[8]) /*__device__ void evaluatePalette(uint alpha0, uint alpha1, uint alphas[8])
{ {
alpha[0] = alpha0; alpha[0] = alpha0;
alpha[1] = alpha1; alpha[1] = alpha1;
alpha[2] = (6 * alpha[0] + 1 * alpha[1]) / 7; // bit code 010 alpha[2] = (6 * alpha[0] + 1 * alpha[1]) / 7; // bit code 010
alpha[3] = (5 * alpha[0] + 2 * alpha[1]) / 7; // bit code 011 alpha[3] = (5 * alpha[0] + 2 * alpha[1]) / 7; // bit code 011
alpha[4] = (4 * alpha[0] + 3 * alpha[1]) / 7; // bit code 100 alpha[4] = (4 * alpha[0] + 3 * alpha[1]) / 7; // bit code 100
alpha[5] = (3 * alpha[0] + 4 * alpha[1]) / 7; // bit code 101 alpha[5] = (3 * alpha[0] + 4 * alpha[1]) / 7; // bit code 101
alpha[6] = (2 * alpha[0] + 5 * alpha[1]) / 7; // bit code 110 alpha[6] = (2 * alpha[0] + 5 * alpha[1]) / 7; // bit code 110
alpha[7] = (1 * alpha[0] + 6 * alpha[1]) / 7; // bit code 111 alpha[7] = (1 * alpha[0] + 6 * alpha[1]) / 7; // bit code 111
} }