CUDAでウェーブレットノイズ
をテンプレートにして作成
[
トップ
|
新規
|
一覧
|
検索
|
最終更新
|
ヘルプ
]
開始行:
----
#contents
----
((K. Perlin and E. M. Hoffert, "Hypertexture", SIGGRAPH '...
((R. L. Cook and T. DeRose, "Wavelet noise", SIGGRAPH 200...
|&ref(noise_perlin_fft_1.jpg);|&ref(noise_wavelet_fft_1.j...
#code(C){{
float *g_dNoiseTile = 0;
uint g_uNoiseTileSize = 0;
void CuSetWaveletTile(float *tile, int n)
{
if(!g_dNoiseTile && g_uNoiseTileSize != n){
int size = n*n*sizeof(float);
cutilSafeCall(cudaMalloc((void**)&g_dNoiseTile, size));
cutilSafeCall(cudaMemcpy((void*)g_dNoiseTile, (void*)ti...
g_uNoiseTileSize = n;
}
}
void CuWNoise2D(float *hW, int nx, int ny, int level)
{
printf("\n[CuWNoise2D]\n");
if(!g_dNoiseTile) return;
float *dW;
int size;
float w = powf(2.0f, (float)level);
size = nx*ny*sizeof(float);
cutilSafeCall(cudaMalloc((void**)&dW, size));
cutilSafeCall(cudaMemset((void*)dW, 0, size));
if(!hW) hW = new float[nx*ny];
dim3 block(BLOCK_SIZE, BLOCK_SIZE);
dim3 grid((nx+block.x-1)/block.x, (ny+block.y-1)/block.y);
wavelet_noise_2d<<< grid, block >>>(g_dNoiseTile, g_uNoi...
cutilCheckMsg("Kernel execution failed");
cutilSafeCall(cudaThreadSynchronize());
size = nx*ny*sizeof(float);
cutilSafeCall(cudaMemcpy(hW, dW, size, cudaMemcpyDeviceT...
cutilSafeCall(cudaFree(dW));
}
}}
#code(C){{
__device__
float wnoise2d(float *tile, int tile_n, float p[2])
{
int f[3], c[3]; // filter, noise coef. indices
int mid[3];
float w[2][3], t, result = 0;
for(int k = 0; k < 2; ++k){
mid[k] = ceil(p[k]-0.5);
t = mid[k]-(p[k]-0.5);
w[k][0] = t*t/2;
w[k][2] = (1-t)*(1-t)/2;
w[k][1] = 1-w[k][0]-w[k][2];
}
for(f[1] = -1; f[1] <= 1; ++f[1]){
for(f[0] = -1; f[0] <= 1; ++f[0]){
float weight = 1;
for(int k = 0; k < 2; ++k){
c[k] = fmodf(mid[k]+f[k], tile_n);
weight *= w[k][f[k]+1];
}
result += weight*tile[c[1]*tile_n+c[0]];
}
}
return result;
}
}}
#code(C){{
if(k == d){
w[k][0] = -t;
w[k][1] = 2*t-1;
w[k][2] = 1-t;
}
else{
w[k][0] = t*t/2;
w[k][2] = (1-t)*(1-t)/2;
w[k][1] = 1-w[k][0]-w[k][2];
}
}}
#code(C){{
__global__
void wavelet_noise_2d(float *tile, int tile_n, float *wt,...
{
int i = blockIdx.x*blockDim.x+threadIdx.x;
int j = blockIdx.y*blockDim.y+threadIdx.y;
if(i < nx && j < ny){
float p[2];
p[0] = (float)(i+0.5)*wpos/nx;
p[1] = (float)(j+0.5)*wpos/ny;
wt[i+j*nx] = wnoise2d(tile, tile_n, p);
}
}
}}
#code(C){{
__global__
void wavelet_multiband_noise_2d(float *tile, int tile_n, ...
{
int i = blockIdx.x*blockDim.x+threadIdx.x;
int j = blockIdx.y*blockDim.y+threadIdx.y;
if(i < nx && j < ny){
float p[2];
p[0] = (float)(i+0.5)/nx;
p[1] = (float)(j+0.5)/ny;
float result = 0;
float w = powf(2.0, (float)first);
float q[2];
float sigma_m = 0;
for(int b = 0; b < nbands; ++b){
q[0] = p[0]*w;
q[1] = p[1]*w;
result += wnoise2d(tile, tile_n, q)/w;
sigma_m += 1.0/w*w;
w *= 2.0;
}
float sigma_n = 0.265;
sigma_m = sqrtf(sigma_n*sigma_m);
if(sigma_m) result /= sigma_m;
wt[i+j*nx] = result;
}
}
}}
|CPU|32|178|
|GPU|9|17|
CENTER:&ref(multi_noise_wavelet.jpg);
#ref(wavelet_noise.zip)
終了行:
----
#contents
----
((K. Perlin and E. M. Hoffert, "Hypertexture", SIGGRAPH '...
((R. L. Cook and T. DeRose, "Wavelet noise", SIGGRAPH 200...
|&ref(noise_perlin_fft_1.jpg);|&ref(noise_wavelet_fft_1.j...
#code(C){{
float *g_dNoiseTile = 0;
uint g_uNoiseTileSize = 0;
void CuSetWaveletTile(float *tile, int n)
{
if(!g_dNoiseTile && g_uNoiseTileSize != n){
int size = n*n*sizeof(float);
cutilSafeCall(cudaMalloc((void**)&g_dNoiseTile, size));
cutilSafeCall(cudaMemcpy((void*)g_dNoiseTile, (void*)ti...
g_uNoiseTileSize = n;
}
}
void CuWNoise2D(float *hW, int nx, int ny, int level)
{
printf("\n[CuWNoise2D]\n");
if(!g_dNoiseTile) return;
float *dW;
int size;
float w = powf(2.0f, (float)level);
size = nx*ny*sizeof(float);
cutilSafeCall(cudaMalloc((void**)&dW, size));
cutilSafeCall(cudaMemset((void*)dW, 0, size));
if(!hW) hW = new float[nx*ny];
dim3 block(BLOCK_SIZE, BLOCK_SIZE);
dim3 grid((nx+block.x-1)/block.x, (ny+block.y-1)/block.y);
wavelet_noise_2d<<< grid, block >>>(g_dNoiseTile, g_uNoi...
cutilCheckMsg("Kernel execution failed");
cutilSafeCall(cudaThreadSynchronize());
size = nx*ny*sizeof(float);
cutilSafeCall(cudaMemcpy(hW, dW, size, cudaMemcpyDeviceT...
cutilSafeCall(cudaFree(dW));
}
}}
#code(C){{
__device__
float wnoise2d(float *tile, int tile_n, float p[2])
{
int f[3], c[3]; // filter, noise coef. indices
int mid[3];
float w[2][3], t, result = 0;
for(int k = 0; k < 2; ++k){
mid[k] = ceil(p[k]-0.5);
t = mid[k]-(p[k]-0.5);
w[k][0] = t*t/2;
w[k][2] = (1-t)*(1-t)/2;
w[k][1] = 1-w[k][0]-w[k][2];
}
for(f[1] = -1; f[1] <= 1; ++f[1]){
for(f[0] = -1; f[0] <= 1; ++f[0]){
float weight = 1;
for(int k = 0; k < 2; ++k){
c[k] = fmodf(mid[k]+f[k], tile_n);
weight *= w[k][f[k]+1];
}
result += weight*tile[c[1]*tile_n+c[0]];
}
}
return result;
}
}}
#code(C){{
if(k == d){
w[k][0] = -t;
w[k][1] = 2*t-1;
w[k][2] = 1-t;
}
else{
w[k][0] = t*t/2;
w[k][2] = (1-t)*(1-t)/2;
w[k][1] = 1-w[k][0]-w[k][2];
}
}}
#code(C){{
__global__
void wavelet_noise_2d(float *tile, int tile_n, float *wt,...
{
int i = blockIdx.x*blockDim.x+threadIdx.x;
int j = blockIdx.y*blockDim.y+threadIdx.y;
if(i < nx && j < ny){
float p[2];
p[0] = (float)(i+0.5)*wpos/nx;
p[1] = (float)(j+0.5)*wpos/ny;
wt[i+j*nx] = wnoise2d(tile, tile_n, p);
}
}
}}
#code(C){{
__global__
void wavelet_multiband_noise_2d(float *tile, int tile_n, ...
{
int i = blockIdx.x*blockDim.x+threadIdx.x;
int j = blockIdx.y*blockDim.y+threadIdx.y;
if(i < nx && j < ny){
float p[2];
p[0] = (float)(i+0.5)/nx;
p[1] = (float)(j+0.5)/ny;
float result = 0;
float w = powf(2.0, (float)first);
float q[2];
float sigma_m = 0;
for(int b = 0; b < nbands; ++b){
q[0] = p[0]*w;
q[1] = p[1]*w;
result += wnoise2d(tile, tile_n, q)/w;
sigma_m += 1.0/w*w;
w *= 2.0;
}
float sigma_n = 0.265;
sigma_m = sqrtf(sigma_n*sigma_m);
if(sigma_m) result /= sigma_m;
wt[i+j*nx] = result;
}
}
}}
|CPU|32|178|
|GPU|9|17|
CENTER:&ref(multi_noise_wavelet.jpg);
#ref(wavelet_noise.zip)
ページ名:
![[PukiWiki]](image/pukiwiki.png "[PukiWiki]")
