arif/gpgpu-two
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

stop being dumb dumb

Browse Source
This commit is contained in:
Arif Hasanic
2025-12-27 22:01:42 +01:00
parent c957b833d3
commit 1884daea1f
4 changed files with 12 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
shaders/A2Task1Interleaved.comp
View File

@@ -17,7 +17,11 @@ layout(push_constant) uniform PushStruct {
layout(binding = 0) buffer inoutBufer {uint v[];};
void main() {
uint idx = gl_GlobalInvocationID.x;
uint idx = gl_GlobalInvocationID.x * 2 * p.stride;
if (idx + p.stride >= p.size) {
return;
}
v[idx] += v[idx + p.stride];
}

5
shaders/A2Task1Sequential.comp
View File

@@ -18,9 +18,10 @@ layout(binding = 0) buffer inoutBufer { uint v[]; };
void main() {
uint gIDx = gl_GlobalInvocationID.x;
if (gIDx >= p.size) {
if (gIDx + p.offset >= p.size) {
return;
}
v[gIDx] = v[gIDx] + v[gIDx + p.offset];
v[gIDx] += v[gIDx + p.offset];
}

4
src/A2Task1Solution/Interleaved.cpp
View File

@@ -56,12 +56,12 @@ void A2Task1SolutionInterleaved::compute()
PushConstant pc;
pc.size = inputSize;
for (uint stride = inputSize / 2; stride > 0; stride >>= 1)
for (uint stride = 1; stride < inputSize; stride <<= 1)
{
pc.stride = stride;
cb.pushConstants(pipelineLayout, vk::ShaderStageFlagBits::eCompute, 0, sizeof(PushConstant), &pc);
cb.dispatch((stride + workGroupSize - 1) / workGroupSize, 1, 1);
cb.dispatch((inputSize / (stride * 2) + workGroupSize - 1) / workGroupSize, 1, 1);
vk::MemoryBarrier memoryBarrier(vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eShaderRead);
cb.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader, vk::PipelineStageFlagBits::eComputeShader,

2
src/A2Task1Solution/Sequential.cpp
View File

@@ -62,7 +62,7 @@ void A2Task1SolutionSequential::compute()
{
cb.pushConstants(pipelineLayout, vk::ShaderStageFlagBits::eCompute, 0, sizeof(PushConstant), &pc);
uint groupCount = (pc.size + pc.offset + workGroupSize - 1) / workGroupSize;
uint groupCount = (pc.offset + workGroupSize - 1) / workGroupSize;
cb.dispatch(groupCount, 1, 1);
vk::MemoryBarrier memoryBarrier(vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eShaderRead);