You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
648 lines
18 KiB
C++
648 lines
18 KiB
C++
// Copyright NVIDIA Corporation 2007 -- Ignacio Castano <icastano@nvidia.com>
|
|
//
|
|
// Permission is hereby granted, free of charge, to any person
|
|
// obtaining a copy of this software and associated documentation
|
|
// files (the "Software"), to deal in the Software without
|
|
// restriction, including without limitation the rights to use,
|
|
// copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
// copies of the Software, and to permit persons to whom the
|
|
// Software is furnished to do so, subject to the following
|
|
// conditions:
|
|
//
|
|
// The above copyright notice and this permission notice shall be
|
|
// included in all copies or substantial portions of the Software.
|
|
//
|
|
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
|
|
// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
|
|
// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
|
|
// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
|
|
// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
|
|
// OTHER DEALINGS IN THE SOFTWARE.
|
|
|
|
#include <nvcore/Debug.h>
|
|
#include <nvcore/Containers.h>
|
|
#include <nvmath/Color.h>
|
|
#include <nvimage/Image.h>
|
|
#include <nvimage/ColorBlock.h>
|
|
#include <nvimage/BlockDXT.h>
|
|
#include <nvtt/CompressionOptions.h>
|
|
#include <nvtt/OutputOptions.h>
|
|
#include <nvtt/QuickCompressDXT.h>
|
|
#include <nvtt/OptimalCompressDXT.h>
|
|
|
|
#include "CudaCompressDXT.h"
|
|
#include "CudaUtils.h"
|
|
|
|
|
|
#if defined HAVE_CUDA
|
|
#include <cuda_runtime_api.h>
|
|
#endif
|
|
|
|
#include <time.h>
|
|
#include <stdio.h>
|
|
|
|
using namespace nv;
|
|
using namespace nvtt;
|
|
|
|
#if defined HAVE_CUDA
|
|
|
|
#define MAX_BLOCKS 8192U // 32768, 65535
|
|
|
|
|
|
extern "C" void setupCompressKernel(const float weights[3]);
|
|
extern "C" void bindTextureToArray(cudaArray * d_data);
|
|
|
|
extern "C" void compressKernelDXT1(uint firstBlock, uint blockNum, uint w, uint * d_result, uint * d_bitmaps);
|
|
extern "C" void compressKernelDXT1_Level4(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps);
|
|
extern "C" void compressWeightedKernelDXT1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps);
|
|
extern "C" void compressKernelDXT3(uint firstBlock, uint blockNum, uint w, uint * d_result, uint * d_bitmaps);
|
|
//extern "C" void compressNormalKernelDXT1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps);
|
|
//extern "C" void compressKernelCTX1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps);
|
|
|
|
|
|
#pragma message(NV_FILE_LINE "TODO: Rename Bitmaps.h to BitmapTable.h")
|
|
#include "Bitmaps.h"
|
|
|
|
/*
|
|
// Convert linear image to block linear.
|
|
static void convertToBlockLinear(const Image * image, uint * blockLinearImage)
|
|
{
|
|
const uint w = (image->width() + 3) / 4;
|
|
const uint h = (image->height() + 3) / 4;
|
|
|
|
for(uint by = 0; by < h; by++) {
|
|
for(uint bx = 0; bx < w; bx++) {
|
|
const uint bw = min(image->width() - bx * 4, 4U);
|
|
const uint bh = min(image->height() - by * 4, 4U);
|
|
|
|
for (uint i = 0; i < 16; i++) {
|
|
const int x = (i % 4) % bw;
|
|
const int y = (i / 4) % bh;
|
|
blockLinearImage[(by * w + bx) * 16 + i] = image->pixel(bx * 4 + x, by * 4 + y).u;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
*/
|
|
|
|
#endif
|
|
|
|
|
|
CudaContext::CudaContext() :
|
|
bitmapTable(NULL),
|
|
bitmapTableCTX(NULL),
|
|
data(NULL),
|
|
result(NULL)
|
|
{
|
|
#if defined HAVE_CUDA
|
|
// Allocate and upload bitmaps.
|
|
cudaMalloc((void**) &bitmapTable, 992 * sizeof(uint));
|
|
if (bitmapTable != NULL)
|
|
{
|
|
cudaMemcpy(bitmapTable, s_bitmapTable, 992 * sizeof(uint), cudaMemcpyHostToDevice);
|
|
}
|
|
|
|
cudaMalloc((void**) &bitmapTableCTX, 704 * sizeof(uint));
|
|
if (bitmapTableCTX != NULL)
|
|
{
|
|
cudaMemcpy(bitmapTableCTX, s_bitmapTableCTX, 704 * sizeof(uint), cudaMemcpyHostToDevice);
|
|
}
|
|
|
|
// Allocate scratch buffers.
|
|
cudaMalloc((void**) &data, MAX_BLOCKS * 64U);
|
|
cudaMalloc((void**) &result, MAX_BLOCKS * 8U);
|
|
#endif
|
|
}
|
|
|
|
CudaContext::~CudaContext()
|
|
{
|
|
#if defined HAVE_CUDA
|
|
// Free device mem allocations.
|
|
cudaFree(bitmapTableCTX);
|
|
cudaFree(bitmapTable);
|
|
cudaFree(data);
|
|
cudaFree(result);
|
|
#endif
|
|
}
|
|
|
|
bool CudaContext::isValid() const
|
|
{
|
|
#if defined HAVE_CUDA
|
|
cudaError_t err = cudaGetLastError();
|
|
if (err != cudaSuccess)
|
|
{
|
|
nvDebug("*** CUDA Error: %s\n", cudaGetErrorString(err));
|
|
return false;
|
|
}
|
|
#endif
|
|
return bitmapTable != NULL && bitmapTableCTX != NULL && data != NULL && result != NULL;
|
|
}
|
|
|
|
|
|
|
|
CudaCompressor::CudaCompressor(CudaContext & ctx) : m_ctx(ctx)
|
|
{
|
|
|
|
}
|
|
|
|
void CudaCompressor::compress(nvtt::InputFormat inputFormat, nvtt::AlphaMode alphaMode, uint w, uint h, void * data, const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions)
|
|
{
|
|
nvDebugCheck(cuda::isHardwarePresent());
|
|
|
|
#if defined HAVE_CUDA
|
|
|
|
// Allocate image as a cuda array.
|
|
cudaArray * d_image;
|
|
if (inputFormat == nvtt::InputFormat_BGRA_8UB)
|
|
{
|
|
cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(8, 8, 8, 8, cudaChannelFormatKindUnsigned);
|
|
cudaMallocArray(&d_image, &channelDesc, w, h);
|
|
|
|
const int imageSize = w * h * sizeof(uint);
|
|
cudaMemcpyToArray(d_image, 0, 0, data, imageSize, cudaMemcpyHostToDevice);
|
|
}
|
|
else
|
|
{
|
|
#pragma message(NV_FILE_LINE "FIXME: Floating point textures not really supported by CUDA compressors.")
|
|
cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 32, 32, 32, cudaChannelFormatKindFloat);
|
|
cudaMallocArray(&d_image, &channelDesc, w, h);
|
|
|
|
const int imageSize = w * h * sizeof(uint);
|
|
cudaMemcpyToArray(d_image, 0, 0, data, imageSize, cudaMemcpyHostToDevice);
|
|
}
|
|
|
|
// Image size in blocks.
|
|
const uint bw = (w + 3) / 4;
|
|
const uint bh = (h + 3) / 4;
|
|
const uint bs = blockSize();
|
|
const uint blockNum = bw * bh;
|
|
const uint compressedSize = blockNum * bs;
|
|
|
|
void * h_result = malloc(min(blockNum, MAX_BLOCKS) * bs);
|
|
|
|
setup(d_image, compressionOptions);
|
|
|
|
// Timer timer;
|
|
// timer.start();
|
|
|
|
uint bn = 0;
|
|
while(bn != blockNum)
|
|
{
|
|
uint count = min(blockNum - bn, MAX_BLOCKS);
|
|
|
|
compressBlocks(bn, count, w, h, alphaMode, compressionOptions, h_result);
|
|
|
|
// Check for errors.
|
|
cudaError_t err = cudaGetLastError();
|
|
if (err != cudaSuccess)
|
|
{
|
|
//nvDebug("CUDA Error: %s\n", cudaGetErrorString(err));
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
}
|
|
|
|
// Output result.
|
|
if (outputOptions.outputHandler != NULL)
|
|
{
|
|
outputOptions.outputHandler->writeData(h_result, count * bs);
|
|
}
|
|
|
|
bn += count;
|
|
}
|
|
|
|
//timer.stop();
|
|
//printf("\rCUDA time taken: %.3f seconds\n", timer.elapsed() / CLOCKS_PER_SEC);
|
|
|
|
free(h_result);
|
|
cudaFreeArray(d_image);
|
|
|
|
#else
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
void CudaCompressorDXT1::setup(cudaArray * image, const nvtt::CompressionOptions::Private & compressionOptions)
|
|
{
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
bindTextureToArray(image);
|
|
}
|
|
|
|
void CudaCompressorDXT1::compressBlocks(uint first, uint count, uint w, uint h, nvtt::AlphaMode alphaMode, const nvtt::CompressionOptions::Private & compressionOptions, void * output)
|
|
{
|
|
// Launch kernel.
|
|
compressKernelDXT1(first, count, w, m_ctx.result, m_ctx.bitmapTable);
|
|
|
|
// Copy result to host.
|
|
cudaMemcpy(output, m_ctx.result, count * 8, cudaMemcpyDeviceToHost);
|
|
}
|
|
|
|
|
|
void CudaCompressorDXT3::setup(cudaArray * image, const nvtt::CompressionOptions::Private & compressionOptions)
|
|
{
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
bindTextureToArray(image);
|
|
}
|
|
|
|
void CudaCompressorDXT3::compressBlocks(uint first, uint count, uint w, uint h, nvtt::AlphaMode alphaMode, const nvtt::CompressionOptions::Private & compressionOptions, void * output)
|
|
{
|
|
// Launch kernel.
|
|
compressKernelDXT3(first, count, w, m_ctx.result, m_ctx.bitmapTable);
|
|
|
|
// Copy result to host.
|
|
cudaMemcpy(output, m_ctx.result, count * 16, cudaMemcpyDeviceToHost);
|
|
}
|
|
|
|
|
|
void CudaCompressorDXT5::setup(cudaArray * image, const nvtt::CompressionOptions::Private & compressionOptions)
|
|
{
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
bindTextureToArray(image);
|
|
}
|
|
|
|
void CudaCompressorDXT5::compressBlocks(uint first, uint count, uint w, uint h, nvtt::AlphaMode alphaMode, const nvtt::CompressionOptions::Private & compressionOptions, void * output)
|
|
{
|
|
/*// Launch kernel.
|
|
compressKernelDXT5(first, count, w, m_ctx.result, m_ctx.bitmapTable);
|
|
|
|
// Copy result to host.
|
|
cudaMemcpy(output, m_ctx.result, count * 16, cudaMemcpyDeviceToHost);*/
|
|
|
|
// Launch kernel.
|
|
if (alphaMode == AlphaMode_Transparency)
|
|
{
|
|
// compressWeightedKernelDXT1(first, count, w, m_ctx.result, m_ctx.bitmapTable);
|
|
}
|
|
else
|
|
{
|
|
// compressKernelDXT1_Level4(first, count, w, m_ctx.result, m_ctx.bitmapTable);
|
|
}
|
|
|
|
// Compress alpha in parallel with the GPU.
|
|
for (uint i = 0; i < count; i++)
|
|
{
|
|
//ColorBlock rgba(blockLinearImage + (first + i) * 16);
|
|
//OptimalCompress::compressDXT3A(rgba, alphaBlocks + i);
|
|
}
|
|
|
|
// Copy result to host.
|
|
cudaMemcpy(output, m_ctx.result, count * 8, cudaMemcpyDeviceToHost);
|
|
|
|
// @@ Interleave color and alpha blocks.
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
// @@ This code is very repetitive and needs to be cleaned up.
|
|
|
|
#if 0
|
|
|
|
/// Compress image using CUDA.
|
|
void CudaCompressor::compressDXT3(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
|
|
{
|
|
nvDebugCheck(cuda::isHardwarePresent());
|
|
#if defined HAVE_CUDA
|
|
|
|
// Image size in blocks.
|
|
const uint w = (m_image->width() + 3) / 4;
|
|
const uint h = (m_image->height() + 3) / 4;
|
|
|
|
uint imageSize = w * h * 16 * sizeof(Color32);
|
|
uint * blockLinearImage = (uint *) malloc(imageSize);
|
|
convertToBlockLinear(m_image, blockLinearImage);
|
|
|
|
const uint blockNum = w * h;
|
|
const uint compressedSize = blockNum * 8;
|
|
|
|
AlphaBlockDXT3 * alphaBlocks = NULL;
|
|
alphaBlocks = (AlphaBlockDXT3 *)malloc(min(compressedSize, MAX_BLOCKS * 8U));
|
|
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
|
|
clock_t start = clock();
|
|
|
|
uint bn = 0;
|
|
while(bn != blockNum)
|
|
{
|
|
uint count = min(blockNum - bn, MAX_BLOCKS);
|
|
|
|
cudaMemcpy(m_ctx.data, blockLinearImage + bn * 16, count * 64, cudaMemcpyHostToDevice);
|
|
|
|
// Launch kernel.
|
|
if (m_alphaMode == AlphaMode_Transparency)
|
|
{
|
|
compressWeightedKernelDXT1(count, m_ctx.data, m_ctx.result, m_ctx.bitmapTable);
|
|
}
|
|
else
|
|
{
|
|
compressKernelDXT1_Level4(count, m_ctx.data, m_ctx.result, m_ctx.bitmapTable);
|
|
}
|
|
|
|
// Compress alpha in parallel with the GPU.
|
|
for (uint i = 0; i < count; i++)
|
|
{
|
|
ColorBlock rgba(blockLinearImage + (bn + i) * 16);
|
|
OptimalCompress::compressDXT3A(rgba, alphaBlocks + i);
|
|
}
|
|
|
|
// Check for errors.
|
|
cudaError_t err = cudaGetLastError();
|
|
if (err != cudaSuccess)
|
|
{
|
|
nvDebug("CUDA Error: %s\n", cudaGetErrorString(err));
|
|
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
}
|
|
|
|
// Copy result to host, overwrite swizzled image.
|
|
cudaMemcpy(blockLinearImage, m_ctx.result, count * 8, cudaMemcpyDeviceToHost);
|
|
|
|
// Output result.
|
|
if (outputOptions.outputHandler != NULL)
|
|
{
|
|
for (uint i = 0; i < count; i++)
|
|
{
|
|
outputOptions.outputHandler->writeData(alphaBlocks + i, 8);
|
|
outputOptions.outputHandler->writeData(blockLinearImage + i * 2, 8);
|
|
}
|
|
}
|
|
|
|
bn += count;
|
|
}
|
|
|
|
clock_t end = clock();
|
|
//printf("\rCUDA time taken: %.3f seconds\n", float(end-start) / CLOCKS_PER_SEC);
|
|
|
|
free(alphaBlocks);
|
|
free(blockLinearImage);
|
|
|
|
#else
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
|
|
/// Compress image using CUDA.
|
|
void CudaCompressor::compressDXT5(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
|
|
{
|
|
nvDebugCheck(cuda::isHardwarePresent());
|
|
#if defined HAVE_CUDA
|
|
|
|
// Image size in blocks.
|
|
const uint w = (m_image->width() + 3) / 4;
|
|
const uint h = (m_image->height() + 3) / 4;
|
|
|
|
uint imageSize = w * h * 16 * sizeof(Color32);
|
|
uint * blockLinearImage = (uint *) malloc(imageSize);
|
|
convertToBlockLinear(m_image, blockLinearImage);
|
|
|
|
const uint blockNum = w * h;
|
|
const uint compressedSize = blockNum * 8;
|
|
|
|
AlphaBlockDXT5 * alphaBlocks = NULL;
|
|
alphaBlocks = (AlphaBlockDXT5 *)malloc(min(compressedSize, MAX_BLOCKS * 8U));
|
|
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
|
|
clock_t start = clock();
|
|
|
|
uint bn = 0;
|
|
while(bn != blockNum)
|
|
{
|
|
uint count = min(blockNum - bn, MAX_BLOCKS);
|
|
|
|
cudaMemcpy(m_ctx.data, blockLinearImage + bn * 16, count * 64, cudaMemcpyHostToDevice);
|
|
|
|
// Launch kernel.
|
|
if (m_alphaMode == AlphaMode_Transparency)
|
|
{
|
|
compressWeightedKernelDXT1(count, m_ctx.data, m_ctx.result, m_ctx.bitmapTable);
|
|
}
|
|
else
|
|
{
|
|
compressKernelDXT1_Level4(count, m_ctx.data, m_ctx.result, m_ctx.bitmapTable);
|
|
}
|
|
|
|
// Compress alpha in parallel with the GPU.
|
|
for (uint i = 0; i < count; i++)
|
|
{
|
|
ColorBlock rgba(blockLinearImage + (bn + i) * 16);
|
|
QuickCompress::compressDXT5A(rgba, alphaBlocks + i);
|
|
}
|
|
|
|
// Check for errors.
|
|
cudaError_t err = cudaGetLastError();
|
|
if (err != cudaSuccess)
|
|
{
|
|
nvDebug("CUDA Error: %s\n", cudaGetErrorString(err));
|
|
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
}
|
|
|
|
// Copy result to host, overwrite swizzled image.
|
|
cudaMemcpy(blockLinearImage, m_ctx.result, count * 8, cudaMemcpyDeviceToHost);
|
|
|
|
// Output result.
|
|
if (outputOptions.outputHandler != NULL)
|
|
{
|
|
for (uint i = 0; i < count; i++)
|
|
{
|
|
outputOptions.outputHandler->writeData(alphaBlocks + i, 8);
|
|
outputOptions.outputHandler->writeData(blockLinearImage + i * 2, 8);
|
|
}
|
|
}
|
|
|
|
bn += count;
|
|
}
|
|
|
|
clock_t end = clock();
|
|
//printf("\rCUDA time taken: %.3f seconds\n", float(end-start) / CLOCKS_PER_SEC);
|
|
|
|
free(alphaBlocks);
|
|
free(blockLinearImage);
|
|
|
|
#else
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
|
|
void CudaCompressor::compressDXT1n(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions)
|
|
{
|
|
nvDebugCheck(cuda::isHardwarePresent());
|
|
#if defined HAVE_CUDA
|
|
|
|
// Image size in blocks.
|
|
const uint w = (m_image->width() + 3) / 4;
|
|
const uint h = (m_image->height() + 3) / 4;
|
|
|
|
uint imageSize = w * h * 16 * sizeof(Color32);
|
|
uint * blockLinearImage = (uint *) malloc(imageSize);
|
|
convertToBlockLinear(m_image, blockLinearImage); // @@ Do this in parallel with the GPU, or in the GPU!
|
|
|
|
const uint blockNum = w * h;
|
|
const uint compressedSize = blockNum * 8;
|
|
|
|
clock_t start = clock();
|
|
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
|
|
// TODO: Add support for multiple GPUs.
|
|
uint bn = 0;
|
|
while(bn != blockNum)
|
|
{
|
|
uint count = min(blockNum - bn, MAX_BLOCKS);
|
|
|
|
cudaMemcpy(m_ctx.data, blockLinearImage + bn * 16, count * 64, cudaMemcpyHostToDevice);
|
|
|
|
// Launch kernel.
|
|
compressNormalKernelDXT1(count, m_ctx.data, m_ctx.result, m_ctx.bitmapTable);
|
|
|
|
// Check for errors.
|
|
cudaError_t err = cudaGetLastError();
|
|
if (err != cudaSuccess)
|
|
{
|
|
nvDebug("CUDA Error: %s\n", cudaGetErrorString(err));
|
|
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
}
|
|
|
|
// Copy result to host, overwrite swizzled image.
|
|
cudaMemcpy(blockLinearImage, m_ctx.result, count * 8, cudaMemcpyDeviceToHost);
|
|
|
|
// Output result.
|
|
if (outputOptions.outputHandler != NULL)
|
|
{
|
|
outputOptions.outputHandler->writeData(blockLinearImage, count * 8);
|
|
}
|
|
|
|
bn += count;
|
|
}
|
|
|
|
clock_t end = clock();
|
|
//printf("\rCUDA time taken: %.3f seconds\n", float(end-start) / CLOCKS_PER_SEC);
|
|
|
|
free(blockLinearImage);
|
|
|
|
#else
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
|
|
void CudaCompressor::compressCTX1(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions)
|
|
{
|
|
nvDebugCheck(cuda::isHardwarePresent());
|
|
#if defined HAVE_CUDA
|
|
|
|
// Image size in blocks.
|
|
const uint w = (m_image->width() + 3) / 4;
|
|
const uint h = (m_image->height() + 3) / 4;
|
|
|
|
uint imageSize = w * h * 16 * sizeof(Color32);
|
|
uint * blockLinearImage = (uint *) malloc(imageSize);
|
|
convertToBlockLinear(m_image, blockLinearImage); // @@ Do this in parallel with the GPU, or in the GPU!
|
|
|
|
const uint blockNum = w * h;
|
|
const uint compressedSize = blockNum * 8;
|
|
|
|
clock_t start = clock();
|
|
|
|
setupCompressKernel(compressionOptions.colorWeight.ptr());
|
|
|
|
// TODO: Add support for multiple GPUs.
|
|
uint bn = 0;
|
|
while(bn != blockNum)
|
|
{
|
|
uint count = min(blockNum - bn, MAX_BLOCKS);
|
|
|
|
cudaMemcpy(m_ctx.data, blockLinearImage + bn * 16, count * 64, cudaMemcpyHostToDevice);
|
|
|
|
// Launch kernel.
|
|
compressKernelCTX1(count, m_ctx.data, m_ctx.result, m_ctx.bitmapTableCTX);
|
|
|
|
// Check for errors.
|
|
cudaError_t err = cudaGetLastError();
|
|
if (err != cudaSuccess)
|
|
{
|
|
nvDebug("CUDA Error: %s\n", cudaGetErrorString(err));
|
|
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
}
|
|
|
|
// Copy result to host, overwrite swizzled image.
|
|
cudaMemcpy(blockLinearImage, m_ctx.result, count * 8, cudaMemcpyDeviceToHost);
|
|
|
|
// Output result.
|
|
if (outputOptions.outputHandler != NULL)
|
|
{
|
|
outputOptions.outputHandler->writeData(blockLinearImage, count * 8);
|
|
}
|
|
|
|
bn += count;
|
|
}
|
|
|
|
clock_t end = clock();
|
|
//printf("\rCUDA time taken: %.3f seconds\n", float(end-start) / CLOCKS_PER_SEC);
|
|
|
|
free(blockLinearImage);
|
|
|
|
#else
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
|
|
void CudaCompressor::compressDXT5n(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions)
|
|
{
|
|
nvDebugCheck(cuda::isHardwarePresent());
|
|
#if defined HAVE_CUDA
|
|
|
|
// @@ TODO
|
|
|
|
#else
|
|
if (outputOptions.errorHandler != NULL)
|
|
{
|
|
outputOptions.errorHandler->error(Error_CudaError);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
#endif // 0
|