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

Reenable QuickCompress for DXT5n.

Browse Source 
Fix end of lines.
This commit is contained in:
castano 2008-02-01 02:33:51 +00:00
parent 05a760b2f6
commit d44b8e9358
4 changed files with 86 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/nvtt/CompressDXT.cpp
View File

@ -349,8 +349,7 @@ void nv::compressDXT5n(const Image * image, const OutputOptions::Private & outpu
} }
// Compress Y. // Compress Y.
// QuickCompress::compressDXT1G(rgba, &block.color); QuickCompress::compressDXT1G(rgba, &block.color);
compressGreenBlock_BruteForce(rgba, &block.color);
if (outputOptions.outputHandler != NULL) { if (outputOptions.outputHandler != NULL) {
outputOptions.outputHandler->writeData(&block, sizeof(block)); outputOptions.outputHandler->writeData(&block, sizeof(block));

22
src/nvtt/InputOptions.cpp
View File

@ -381,14 +381,14 @@ int InputOptions::Private::realMipmapCount() const
} }
const Image * InputOptions::Private::image(uint face, uint mipmap) const const Image * InputOptions::Private::image(uint face, uint mipmap) const
{ {
nvDebugCheck(face < faceCount); nvDebugCheck(face < faceCount);
nvDebugCheck(mipmap < mipmapCount); nvDebugCheck(mipmap < mipmapCount);
const InputImage & image = this->images[face * mipmapCount + mipmap]; const InputImage & image = this->images[face * mipmapCount + mipmap];
nvDebugCheck(image.face == face); nvDebugCheck(image.face == face);
nvDebugCheck(image.mipLevel == mipmap); nvDebugCheck(image.mipLevel == mipmap);
return image.data.ptr(); return image.data.ptr();
} }

8
src/nvtt/QuickCompressDXT.cpp
View File

@ -435,10 +435,10 @@ static int computeGreenError(const ColorBlock & rgba, const BlockDXT1 * block)
nvDebugCheck(block != NULL); nvDebugCheck(block != NULL);
int palette[4]; int palette[4];
palette[0] = (block->col0.g << 2) | (block->col0.g >> 4); palette[0] = (block->col0.g << 2) | (block->col0.g >> 4);
palette[1] = (block->col1.g << 2) | (block->col1.g >> 4); palette[1] = (block->col1.g << 2) | (block->col1.g >> 4);
palette[2] = (2 * palette[0] + palette[1]) / 3; palette[2] = (2 * palette[0] + palette[1]) / 3;
palette[3] = (2 * palette[1] + palette[0]) / 3; palette[3] = (2 * palette[1] + palette[0]) / 3;
int totalError = 0; int totalError = 0;

70
src/nvtt/cuda/ConvolveKernel.cu
View File

@ -45,6 +45,76 @@
#define __debugsync() #define __debugsync()
#endif #endif
#define TN 256
#define WARP_COUNT (TN / 32)
#define HWARP_COUNT (TN / 16)
// Window size
#define WS 20
struct WrapClamp
{
int operator()(int i, int h)
{
i = min(max(i, 0), h-1);
}
};
struct WrapRepeat
{
int operator()(int i, int h)
{
i = abs(i) % h; // :( Non power of two!
}
};
struct WrapMirror
{
int operator()(int i, int h)
{
i = abs(i);
while (i >= h) i = 2 * w - i - 2;
}
};
// Vertical convolution filter that processes vertical strips.
__global__ void convolveStrip(float * d_channel, float * d_kernel, int width, int height)
{
__shared__ float s_kernel[32 * WS];
// Preload kernel in shared memory.
for (int i = 0; i < 32 * WS / TN; i++)
{
int idx = i * TN + tid;
if (idx < 32 * WS) s_kernel[idx] = d_kernel[idx];
}
__shared__ float s_strip[32 * WS]; // TN/32
int wid = tid / 32 - WS/2;
Mirror wrap;
int row = wrap(wid);
// Preload image block.
for (int i = 0; i < 32 * WS / TN; i++)
{
}
// @@ Apply kernel to TN/32 rows.
// @@ Load
}
__constant__ float inputGamma, outputInverseGamma; __constant__ float inputGamma, outputInverseGamma;
__constant__ float kernel[MAX_KERNEL_WIDTH]; __constant__ float kernel[MAX_KERNEL_WIDTH];