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

I should be using branches instead of tags.

Browse Source
This commit is contained in:
castano
2008-02-28 07:49:13 +00:00
parent 7933aeea09
commit 21859216bc
17 changed files with 552 additions and 1652 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/nvtt/CMakeLists.txt
View File

@ -79,9 +79,6 @@ TARGET_LINK_LIBRARIES(nvassemble nvcore nvmath nvimage)
ADD_EXECUTABLE(filtertest tests/filtertest.cpp tools/cmdline.h)
TARGET_LINK_LIBRARIES(filtertest nvcore nvmath nvimage)
ADD_EXECUTABLE(stress tests/stress.cpp tools/cmdline.h)
TARGET_LINK_LIBRARIES(stress nvcore nvmath nvimage nvtt)
ADD_EXECUTABLE(nvzoom tools/resize.cpp tools/cmdline.h)
TARGET_LINK_LIBRARIES(nvzoom nvcore nvmath nvimage)

35
src/nvtt/Compressor.cpp
View File

@ -56,7 +56,7 @@ namespace
static int blockSize(Format format)
{
if (format == Format_DXT1 || format == Format_DXT1a || format == Format_DXT1n) {
if (format == Format_DXT1 || format == Format_DXT1a) {
return 8;
}
else if (format == Format_DXT3) {
@ -71,9 +71,6 @@ namespace
else if (format == Format_BC5) {
return 16;
}
else if (format == Format_CTX1) {
return 8;
}
return 0;
}
@ -336,7 +333,7 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
{
header.setLinearSize(computeImageSize(inputOptions.targetWidth, inputOptions.targetHeight, inputOptions.targetDepth, compressionOptions.bitcount, compressionOptions.format));
if (compressionOptions.format == Format_DXT1 || compressionOptions.format == Format_DXT1a || compressionOptions.format == Format_DXT1n) {
if (compressionOptions.format == Format_DXT1 || compressionOptions.format == Format_DXT1a) {
header.setFourCC('D', 'X', 'T', '1');
if (inputOptions.isNormalMap) header.setNormalFlag(true);
}
@ -357,10 +354,6 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
header.setFourCC('A', 'T', 'I', '2');
if (inputOptions.isNormalMap) header.setNormalFlag(true);
}
else if (compressionOptions.format == Format_CTX1) {
header.setFourCC('C', 'T', 'X', '1');
if (inputOptions.isNormalMap) header.setNormalFlag(true);
}
}
// Swap bytes if necessary.
@ -712,18 +705,6 @@ bool Compressor::Private::compressMipmap(const Mipmap & mipmap, const Compressio
}
}
}
else if (compressionOptions.format == Format_DXT1n)
{
if (cudaEnabled)
{
nvDebugCheck(cudaSupported);
cuda->compressDXT1n(image, outputOptions, compressionOptions);
}
else
{
if (outputOptions.errorHandler) outputOptions.errorHandler->error(Error_UnsupportedFeature);
}
}
else if (compressionOptions.format == Format_DXT3)
{
if (compressionOptions.quality == Quality_Fastest)
@ -781,18 +762,6 @@ bool Compressor::Private::compressMipmap(const Mipmap & mipmap, const Compressio
{
compressBC5(image, outputOptions, compressionOptions);
}
else if (compressionOptions.format == Format_CTX1)
{
if (cudaEnabled)
{
nvDebugCheck(cudaSupported);
cuda->compressCTX1(image, outputOptions, compressionOptions);
}
else
{
if (outputOptions.errorHandler) outputOptions.errorHandler->error(Error_UnsupportedFeature);
}
}
return true;
}

337
src/nvtt/cuda/CompressKernel.cu
View File

@ -205,43 +205,6 @@ __device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sum
}
}
__device__ void loadColorBlock(const uint * image, float2 colors[16], float2 sums[16], int xrefs[16])
{
const int bid = blockIdx.x;
const int idx = threadIdx.x;
__shared__ float dps[16];
if (idx < 16)
{
// Read color and copy to shared mem.
uint c = image[(bid) * 16 + idx];
colors[idx].y = ((c >> 8) & 0xFF) * (1.0f / 255.0f);
colors[idx].x = ((c >> 16) & 0xFF) * (1.0f / 255.0f);
// No need to synchronize, 16 < warp size.
#if __DEVICE_EMULATION__
} __debugsync(); if (idx < 16) {
#endif
// Sort colors along the best fit line.
colorSums(colors, sums);
float2 axis = bestFitLine(colors, sums[0]);
dps[idx] = dot(colors[idx], axis);
#if __DEVICE_EMULATION__
} __debugsync(); if (idx < 16) {
#endif
sortColors(dps, xrefs);
float2 tmp = colors[idx];
colors[xrefs[idx]] = tmp;
}
}
////////////////////////////////////////////////////////////////////////////////
// Round color to RGB565 and expand
@ -258,26 +221,6 @@ inline __device__ float3 roundAndExpand565(float3 v, ushort * w)
return v;
}
inline __device__ float2 roundAndExpand56(float2 v, ushort * w)
{
v.x = rintf(__saturatef(v.x) * 31.0f);
v.y = rintf(__saturatef(v.y) * 63.0f);
*w = ((ushort)v.x << 11) | ((ushort)v.y << 5);
v.x *= 0.03227752766457f; // approximate integer bit expansion.
v.y *= 0.01583151765563f;
return v;
}
inline __device__ float2 roundAndExpand88(float2 v, ushort * w)
{
v.x = rintf(__saturatef(v.x) * 255.0f);
v.y = rintf(__saturatef(v.y) * 255.0f);
*w = ((ushort)v.x << 8) | ((ushort)v.y);
v.x *= 1.0f / 255.0f;
v.y *= 1.0f / 255.0f;
return v;
}
////////////////////////////////////////////////////////////////////////////////
// Evaluate permutations
@ -521,114 +464,6 @@ __device__ float evalPermutation3(const float3 * colors, const float * weights,
}
*/
__device__ float evalPermutation4(const float2 * colors, float2 color_sum, uint permutation, ushort * start, ushort * end)
{
// Compute endpoints using least squares.
float2 alphax_sum = make_float2(0.0f, 0.0f);
uint akku = 0;
// Compute alpha & beta for this permutation.
#pragma unroll
for (int i = 0; i < 16; i++)
{
const uint bits = permutation >> (2*i);
alphax_sum += alphaTable4[bits & 3] * colors[i];
akku += prods4[bits & 3];
}
float alpha2_sum = float(akku >> 16);
float beta2_sum = float((akku >> 8) & 0xff);
float alphabeta_sum = float(akku & 0xff);
float2 betax_sum = 9.0f * color_sum - alphax_sum;
const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
float2 a = (alphax_sum * beta2_sum - betax_sum * alphabeta_sum) * factor;
float2 b = (betax_sum * alpha2_sum - alphax_sum * alphabeta_sum) * factor;
// Round a, b to the closest 5-6 color and expand...
a = roundAndExpand56(a, start);
b = roundAndExpand56(b, end);
// compute the error
float2 e = a * a * alpha2_sum + b * b * beta2_sum + 2.0f * (a * b * alphabeta_sum - a * alphax_sum - b * betax_sum);
return (1.0f / 9.0f) * (e.x + e.y);
}
__device__ float evalPermutation3(const float2 * colors, float2 color_sum, uint permutation, ushort * start, ushort * end)
{
// Compute endpoints using least squares.
float2 alphax_sum = make_float2(0.0f, 0.0f);
uint akku = 0;
// Compute alpha & beta for this permutation.
#pragma unroll
for (int i = 0; i < 16; i++)
{
const uint bits = permutation >> (2*i);
alphax_sum += alphaTable3[bits & 3] * colors[i];
akku += prods3[bits & 3];
}
float alpha2_sum = float(akku >> 16);
float beta2_sum = float((akku >> 8) & 0xff);
float alphabeta_sum = float(akku & 0xff);
float2 betax_sum = 4.0f * color_sum - alphax_sum;
const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
float2 a = (alphax_sum * beta2_sum - betax_sum * alphabeta_sum) * factor;
float2 b = (betax_sum * alpha2_sum - alphax_sum * alphabeta_sum) * factor;
// Round a, b to the closest 5-6 color and expand...
a = roundAndExpand56(a, start);
b = roundAndExpand56(b, end);
// compute the error
float2 e = a * a * alpha2_sum + b * b * beta2_sum + 2.0f * (a * b * alphabeta_sum - a * alphax_sum - b * betax_sum);
return (1.0f / 4.0f) * (e.x + e.y);
}
__device__ float evalPermutationCTX(const float2 * colors, float2 color_sum, uint permutation, ushort * start, ushort * end)
{
// Compute endpoints using least squares.
float2 alphax_sum = make_float2(0.0f, 0.0f);
uint akku = 0;
// Compute alpha & beta for this permutation.
#pragma unroll
for (int i = 0; i < 16; i++)
{
const uint bits = permutation >> (2*i);
alphax_sum += alphaTable4[bits & 3] * colors[i];
akku += prods4[bits & 3];
}
float alpha2_sum = float(akku >> 16);
float beta2_sum = float((akku >> 8) & 0xff);
float alphabeta_sum = float(akku & 0xff);
float2 betax_sum = 9.0f * color_sum - alphax_sum;
const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
float2 a = (alphax_sum * beta2_sum - betax_sum * alphabeta_sum) * factor;
float2 b = (betax_sum * alpha2_sum - alphax_sum * alphabeta_sum) * factor;
// Round a, b to the closest 8-8 color and expand...
a = roundAndExpand88(a, start);
b = roundAndExpand88(b, end);
// compute the error
float2 e = a * a * alpha2_sum + b * b * beta2_sum + 2.0f * (a * b * alphabeta_sum - a * alphax_sum - b * betax_sum);
return (1.0f / 9.0f) * (e.x + e.y);
}
////////////////////////////////////////////////////////////////////////////////
// Evaluate all permutations
@ -757,67 +592,6 @@ __device__ void evalAllPermutations(const float3 * colors, const float * weights
}
*/
__device__ void evalAllPermutations(const float2 * colors, float2 colorSum, const uint * permutations, ushort & bestStart, ushort & bestEnd, uint & bestPermutation, float * errors)
{
const int idx = threadIdx.x;
float bestError = FLT_MAX;
__shared__ uint s_permutations[160];
for(int i = 0; i < 16; i++)
{
int pidx = idx + NUM_THREADS * i;
if (pidx >= 992) break;
ushort start, end;
uint permutation = permutations[pidx];
if (pidx < 160) s_permutations[pidx] = permutation;
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.
}
for(int i = 0; i < 3; i++)
{
int pidx = idx + NUM_THREADS * i;
if (pidx >= 160) break;
ushort start, end;
uint permutation = s_permutations[pidx];
float error = evalPermutation3(colors, colorSum, permutation, &start, &end);
if (error < bestError)
{
bestError = error;
bestPermutation = permutation;
bestStart = start;
bestEnd = end;
if (bestStart > bestEnd)
{
swap(bestEnd, bestStart);
bestPermutation ^= (~bestPermutation >> 1) & 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;
@ -852,39 +626,6 @@ __device__ void evalLevel4Permutations(const float3 * colors, const float * weig
errors[idx] = bestError;
}
__device__ void evalAllPermutationsCTX(const float2 * colors, float2 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 = evalPermutationCTX(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;
}
////////////////////////////////////////////////////////////////////////////////
@ -996,11 +737,6 @@ __device__ void saveBlockDXT1(ushort start, ushort end, uint permutation, int xr
result[bid].y = indices;
}
__device__ void saveBlockCTX1(ushort start, ushort end, uint permutation, int xrefs[16], uint2 * result)
{
saveBlockDXT1(start, end, permutation, xrefs, result);
}
__device__ void saveSingleColorBlockDXT1(float3 color, uint2 * result)
{
const int bid = blockIdx.x;
@ -1012,10 +748,10 @@ __device__ void saveSingleColorBlockDXT1(float3 color, uint2 * result)
ushort color0 = (OMatch5[r][0] << 11) | (OMatch6[g][0] << 5) | OMatch5[b][0];
ushort color1 = (OMatch5[r][1] << 11) | (OMatch6[g][1] << 5) | OMatch5[b][1];
if (color0 < color1)
{
result[bid].x = (color0 << 16) | color1;
result[bid].y = 0xffffffff;
if (color0 < color1)
{
result[bid].x = (color0 << 16) | color1;
result[bid].y = 0xffffffff;
}
else
{
@ -1092,61 +828,6 @@ __global__ void compressWeightedDXT1(const uint * permutations, const uint * ima
}
__global__ void compressNormalDXT1(const uint * permutations, const uint * image, uint2 * result)
{
__shared__ float2 colors[16];
__shared__ float2 sums[16];
__shared__ int xrefs[16];
loadColorBlock(image, colors, sums, xrefs);
__syncthreads();
ushort bestStart, bestEnd;
uint bestPermutation;
__shared__ float errors[NUM_THREADS];
evalAllPermutations(colors, sums[0], permutations, bestStart, bestEnd, bestPermutation, errors);
// Use a parallel reduction to find minimum error.
const int minIdx = findMinError(errors);
// Only write the result of the winner thread.
if (threadIdx.x == minIdx)
{
saveBlockDXT1(bestStart, bestEnd, bestPermutation, xrefs, result);
}
}
__global__ void compressCTX1(const uint * permutations, const uint * image, uint2 * result)
{
__shared__ float2 colors[16];
__shared__ float2 sums[16];
__shared__ int xrefs[16];
loadColorBlock(image, colors, sums, xrefs);
__syncthreads();
ushort bestStart, bestEnd;
uint bestPermutation;
__shared__ float errors[NUM_THREADS];
evalAllPermutationsCTX(colors, sums[0], permutations, bestStart, bestEnd, bestPermutation, errors);
// Use a parallel reduction to find minimum error.
const int minIdx = findMinError(errors);
// Only write the result of the winner thread.
if (threadIdx.x == minIdx)
{
saveBlockCTX1(bestStart, bestEnd, bestPermutation, xrefs, result);
}
}
/*
__device__ float computeError(const float weights[16], uchar a0, uchar a1)
{
@ -1356,13 +1037,3 @@ extern "C" void compressWeightedKernelDXT1(uint blockNum, uint * d_data, uint *
{
compressWeightedDXT1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
}
extern "C" void compressNormalKernelDXT1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps)
{
compressNormalDXT1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
}
extern "C" void compressKernelCTX1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps)
{
compressCTX1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
}

1201
src/nvtt/cuda/CudaCompressDXT.cpp
View File

File diff suppressed because it is too large Load Diff

2
src/nvtt/cuda/CudaCompressDXT.h
View File

@ -40,8 +40,6 @@ namespace nv
void compressDXT1(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
void compressDXT3(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
void compressDXT5(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
void compressDXT1n(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
void compressCTX1(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
private:

142
src/nvtt/cuda/CudaMath.h
View File

@ -87,64 +87,6 @@ inline __device__ __host__ bool operator ==(float3 a, float3 b)
return a.x == b.x && a.y == b.y && a.z == b.z;
}
// float2 operators
inline __device__ __host__ float2 operator *(float2 a, float2 b)
{
return make_float2(a.x*b.x, a.y*b.y);
}
inline __device__ __host__ float2 operator *(float f, float2 v)
{
return make_float2(v.x*f, v.y*f);
}
inline __device__ __host__ float2 operator *(float2 v, float f)
{
return make_float2(v.x*f, v.y*f);
}
inline __device__ __host__ float2 operator +(float2 a, float2 b)
{
return make_float2(a.x+b.x, a.y+b.y);
}
inline __device__ __host__ void operator +=(float2 & b, float2 a)
{
b.x += a.x;
b.y += a.y;
}
inline __device__ __host__ float2 operator -(float2 a, float2 b)
{
return make_float2(a.x-b.x, a.y-b.y);
}
inline __device__ __host__ void operator -=(float2 & b, float2 a)
{
b.x -= a.x;
b.y -= a.y;
}
inline __device__ __host__ float2 operator /(float2 v, float f)
{
float inv = 1.0f / f;
return v * inv;
}
inline __device__ __host__ void operator /=(float2 & b, float f)
{
float inv = 1.0f / f;
b.x *= inv;
b.y *= inv;
}
inline __device__ __host__ float dot(float2 a, float2 b)
{
return a.x * b.x + a.y * b.y;
}
inline __device__ __host__ float dot(float3 a, float3 b)
{
return a.x * b.x + a.y * b.y + a.z * b.z;
@ -301,89 +243,5 @@ inline __device__ float3 bestFitLine(const float3 * colors, float3 color_sum, fl
return firstEigenVector(covariance);
}
// @@ For 2D this may not be the most efficient method. It's a quadratic equation, right?
inline __device__ __host__ float2 firstEigenVector2D( float matrix[3] )
{
// @@ 8 iterations is probably more than enough.
float2 v = make_float2(1.0f, 1.0f);
for(int i = 0; i < 8; i++) {
float x = v.x * matrix[0] + v.y * matrix[1];
float y = v.x * matrix[1] + v.y * matrix[2];
float m = max(x, y);
float iv = 1.0f / m;
if (m == 0.0f) iv = 0.0f;
v = make_float2(x*iv, y*iv);
}
return v;
}
inline __device__ void colorSums(const float2 * colors, float2 * sums)
{
#if __DEVICE_EMULATION__
float2 color_sum = make_float2(0.0f, 0.0f, 0.0f);
for (int i = 0; i < 16; i++)
{
color_sum += colors[i];
}
for (int i = 0; i < 16; i++)
{
sums[i] = color_sum;
}
#else
const int idx = threadIdx.x;
sums[idx] = colors[idx];
sums[idx] += sums[idx^8];
sums[idx] += sums[idx^4];
sums[idx] += sums[idx^2];
sums[idx] += sums[idx^1];
#endif
}
inline __device__ float2 bestFitLine(const float2 * colors, float2 color_sum)
{
// Compute covariance matrix of the given colors.
#if __DEVICE_EMULATION__
float covariance[3] = {0, 0, 0};
for (int i = 0; i < 16; i++)
{
float2 a = (colors[i] - color_sum * (1.0f / 16.0f));
covariance[0] += a.x * a.x;
covariance[1] += a.x * a.y;
covariance[3] += a.y * a.y;
}
#else
const int idx = threadIdx.x;
float2 diff = (colors[idx] - color_sum * (1.0f / 16.0f));
__shared__ float covariance[16*3];
covariance[3 * idx + 0] = diff.x * diff.x;
covariance[3 * idx + 1] = diff.x * diff.y;
covariance[3 * idx + 2] = diff.y * diff.y;
for(int d = 8; d > 0; d >>= 1)
{
if (idx < d)
{
covariance[3 * idx + 0] += covariance[3 * (idx+d) + 0];
covariance[3 * idx + 1] += covariance[3 * (idx+d) + 1];
covariance[3 * idx + 2] += covariance[3 * (idx+d) + 2];
}
}
#endif
// Compute first eigen vector.
return firstEigenVector2D(covariance);
}
#endif // CUDAMATH_H

3
src/nvtt/nvtt.h
View File

@ -75,9 +75,6 @@ namespace nvtt
Format_BC3n = Format_DXT5n,
Format_BC4, // ATI1
Format_BC5, // 3DC, ATI2
Format_DXT1n,
Format_CTX1,
};
/// Quality modes.

221
src/nvtt/tests/stress.cpp
View File

@ -1,221 +0,0 @@
// 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 <nvtt/nvtt.h>
#include <stdio.h> // printf
#include <stdlib.h> // rand
#include <time.h> // clock
#include <string.h> // memcpy, memcmp
#include <assert.h>
#define FRAME_COUNT 1000
#define WIDTH 2048
#define HEIGHT 2048
#define INPUT_SIZE (WIDTH*HEIGHT)
#define OUTPUT_SIZE (WIDTH*HEIGHT/16*2)
static int s_input[INPUT_SIZE];
static int s_reference[OUTPUT_SIZE];
static int s_output[OUTPUT_SIZE];
static int s_frame = 0;
struct MyOutputHandler : public nvtt::OutputHandler
{
MyOutputHandler() : m_ptr(NULL) {}
virtual void beginImage(int size, int width, int height, int depth, int face, int miplevel)
{
assert(size == sizeof(int) * OUTPUT_SIZE);
assert(width == WIDTH);
assert(height == HEIGHT);
assert(depth == 1);
assert(face == 0);
assert(miplevel == 0);
m_ptr = (unsigned char *)s_output;
if (s_frame == 1)
{
// Save first result as reference.
memcpy(s_reference, s_output, sizeof(int) * OUTPUT_SIZE);
}
else if (s_frame > 1)
{
// Compare against reference.
if (memcmp(s_output, s_reference, sizeof(int) * OUTPUT_SIZE) != 0)
{
printf("Compressed image different to original.\n");
exit(EXIT_FAILURE);
}
}
}
virtual bool writeData(const void * data, int size)
{
memcpy(m_ptr, data, size);
m_ptr += size;
return true;
}
unsigned char * m_ptr;
};
void precomp()
{
unsigned int bitmaps[1024];
int num = 0;
printf("{\n");
printf("\t%8X,\n", 0);
bitmaps[0] = 0;
num = 1;
for (int a = 1; a <= 15; a++)
{
for (int b = a; b <= 15; b++)
{
for (int c = b; c <= 15; c++)
{
int indices[16];
int i = 0;
for(; i < a; i++) {
indices[i] = 0;
}
for(; i < a+b; i++) {
indices[i] = 2;
}
for(; i < a+b+c; i++) {
indices[i] = 3;
}
for(; i < 16; i++) {
indices[i] = 1;
}
unsigned int bm = 0;
for(i = 0; i < 16; i++) {
bm |= indices[i] << (i * 2);
}
printf("\t0x%8X, // %d %d %d %d\n", bm, a-0, b-a, c-b, 16-c);
bitmaps[num] = bm;
num++;
}
}
}
printf("}\n");
printf("// num = %d\n", num);
/*
for( int i = imax; i >= 0; --i )
{
// second cluster [i,j) is one third along
for( int m = i; m < 16; ++m )
{
indices[m] = 2;
}
const int jmax = ( i == 0 ) ? 15 : 16;
for( int j = jmax; j >= i; --j )
{
// third cluster [j,k) is two thirds along
for( int m = j; m < 16; ++m )
{
indices[m] = 3;
}
int kmax = ( j == 0 ) ? 15 : 16;
for( int k = kmax; k >= j; --k )
{
// last cluster [k,n) is at the end
if( k < 16 )
{
indices[k] = 1;
}
uint bitmap = 0;
bool hasThree = false;
for(int p = 0; p < 16; p++) {
bitmap |= indices[p] << (p * 2);
}
bitmaps[num] = bitmap;
num++;
}
}
}
*/
}
int main(int argc, char *argv[])
{
//precomp();
nvtt::InputOptions inputOptions;
inputOptions.setTextureLayout(nvtt::TextureType_2D, WIDTH, HEIGHT);
for (int i = 0; i < INPUT_SIZE; i++)
{
s_input[i] = rand();
}
inputOptions.setMipmapData(s_input, WIDTH, HEIGHT);
inputOptions.setMipmapGeneration(false);
nvtt::CompressionOptions compressionOptions;
compressionOptions.setFormat(nvtt::Format_DXT1);
// compressionOptions.setFormat(nvtt::Format_DXT1n);
// compressionOptions.setFormat(nvtt::Format_CTX1);
nvtt::OutputOptions outputOptions;
outputOptions.setOutputHeader(false);
MyOutputHandler outputHandler;
outputOptions.setOutputHandler(&outputHandler);
nvtt::Compressor compressor;
for (s_frame = 0; s_frame < FRAME_COUNT; s_frame++)
{
clock_t start = clock();
printf("compressing frame %d:\n", s_frame);
compressor.process(inputOptions, compressionOptions, outputOptions);
clock_t end = clock();
printf("time taken: %.3f seconds\n", float(end-start) / CLOCKS_PER_SEC);
}
return EXIT_SUCCESS;
}