eb45954684
The filters are ~3.5 time faster with CS.
37 lines
1.1 KiB
Plaintext
37 lines
1.1 KiB
Plaintext
// From http://http.developer.nvidia.com/GPUGems3/gpugems3_ch40.html
|
|
|
|
uniform layout(size1x16) restrict readonly image2D source;
|
|
uniform layout(size1x16) volatile restrict writeonly image2D dest;
|
|
uniform float sigma = 5.;
|
|
|
|
layout (local_size_x = 8, local_size_y = 8) in;
|
|
|
|
shared float local_src[8][8 + 2 * 8];
|
|
|
|
void main()
|
|
{
|
|
int x = int(gl_LocalInvocationID.x), y = int(gl_LocalInvocationID.y);
|
|
ivec2 uv = ivec2(gl_GlobalInvocationID.x, gl_GlobalInvocationID.y);
|
|
local_src[x][y] = imageLoad(source, ivec2(uv) - ivec2(0, 8)).x;
|
|
local_src[x][y + 8] = imageLoad(source, ivec2(uv)).x;
|
|
local_src[x][y + 16] = imageLoad(source, ivec2(uv) + ivec2(0, 8)).x;
|
|
|
|
barrier();
|
|
|
|
float g0, g1, g2;
|
|
g0 = 1.0 / (sqrt(2.0 * 3.14) * sigma);
|
|
g1 = exp(-0.5 / (sigma * sigma));
|
|
g2 = g1 * g1;
|
|
float sum = local_src[x][y + 8] * g0;
|
|
g0 *= g1;
|
|
g1 *= g2;
|
|
for (int j = 1; j < 8; j++) {
|
|
sum += local_src[x][y + 8 + j] * g0;
|
|
sum += local_src[x][y + 8 - j] * g0;
|
|
g0 *= g1;
|
|
g1 *= g2;
|
|
}
|
|
imageStore(dest, ivec2(uv), vec4(sum));
|
|
}
|
|
|