quick commit

src/A2Task1Solution/KernelDecomposition.cpp

@@ -50,36 +50,32 @@ void A2Task1SolutionKernelDecomposition::compute()
     vk::CommandBufferBeginInfo beginInfo(vk::CommandBufferUsageFlagBits::eOneTimeSubmit);
 
     cb.begin(beginInfo);
+    cb.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline);
 
-    // TODO: Implement reduction with kernel decomposition
-    // NOTE: make sure that activeBuffer points to the buffer with the final result in the end
-    // That buffer is read back for the correctness check
-    // (A2Task1SolutionKernelDecomposition::result())
-    // HINT: You can alternate between the two provided descriptor sets to implement ping-pong
-
-    uint blocksize = 512;
-    uint kernelCount = mpInput->size() / blocksize;
+    uint blocksize = workGroupSize * 2;
+    uint windowSize = mpInput->size();
     PushConstant p;
 
-    for (; kernelCount > 0; kernelCount /= 2)
+    activeBuffer = 1;
+
+    while (windowSize > 1)
     {
-        activeBuffer = activeBuffer == 0 ? 1 : 0;
+        activeBuffer = 1 - activeBuffer;
+        uint numGroups = (windowSize + blocksize - 1) / blocksize;
+
         cb.bindDescriptorSets(vk::PipelineBindPoint::eCompute, pipelineLayout, 0U, 1U, &descriptorSets[activeBuffer], 0U, nullptr);
-        p.size = blocksize;
+        p.offset = windowSize;
 
         cb.pushConstants(pipelineLayout, vk::ShaderStageFlagBits::eCompute, 0, sizeof(PushConstant), &p);
 
-        // for loop to call each kernel
-        for (int i = 0; i < kernelCount; i++)
-        {
-            cb.dispatch(i * blocksize, 0, 0);
-        }
+        cb.dispatch(numGroups, 1, 1);
 
         vk::MemoryBarrier memoryBarrier(vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eShaderRead);
         cb.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader, vk::PipelineStageFlagBits::eComputeShader, vk::DependencyFlags(), 1, &memoryBarrier, 0, nullptr, 0, nullptr);
     
-        std::cout << "deine mum \n" ;
+        windowSize = numGroups;
     }
+    // todo check which buffer is active and read from that one
     cb.end();
 
     vk::SubmitInfo submitInfo = vk::SubmitInfo(0, nullptr, nullptr, 1, &cb);
@@ -97,7 +93,7 @@ void A2Task1SolutionKernelDecomposition::compute()
 uint A2Task1SolutionKernelDecomposition::result() const
 {
     std::vector<uint> result(1, 0);
-    fillHostWithStagingBuffer<uint>(app.pDevice, app.device, app.transferCommandPool, app.transferQueue, buffers[activeBuffer], result);
+    fillHostWithStagingBuffer<uint>(app.pDevice, app.device, app.transferCommandPool, app.transferQueue, buffers[1 - activeBuffer], result);
     return result[0];
 }