source/backend/opengl/glsl/maxpool.glsl (29 lines of code) (raw):
layout(FORMAT, binding=0) readonly uniform PRECISION image3D uInput;
layout(FORMAT, binding=1) writeonly uniform PRECISION image3D uOutput;
layout(location = 2) uniform ivec2 uKernel;
layout(location = 3) uniform ivec2 uStride;
layout(location = 4) uniform ivec2 uPad;
layout(location=10) uniform ivec3 uOutputSize;
layout(location=11) uniform ivec3 uInputSize;
layout (local_size_x = 2, local_size_y = 2, local_size_z = 16) in;
void main()
{
ivec3 pos = ivec3(gl_GlobalInvocationID);
ivec3 outputSize = uOutputSize;
ivec2 spos = pos.xy*uStride-uPad;
if (all(lessThan(pos, outputSize)))
{
ivec3 inputSize = uInputSize;
ivec2 sfxy = max(ivec2(0), -spos);
ivec2 efxy = min(uKernel, inputSize.xy-spos);
vec4 color = vec4(-100000.0);
for (int fy=sfxy.y; fy<efxy.y; ++fy)
{
for (int fx=sfxy.x; fx<efxy.x; ++fx)
{
color = max(color, imageLoad(uInput, ivec3(spos.x+fx, spos.y+fy, pos.z)));
}
}
imageStore(uOutput, pos, color);
}
}