coalesces writes > coalesced loads

Author: devshgraphicsprogramming

examples_tests/49.ComputeFFT/fft_convolve_ifft.comp

@@ -2,15 +2,7 @@ layout(local_size_x=_NBL_GLSL_WORKGROUP_SIZE_, local_size_y=1, local_size_z=1) i
 
 #include <nbl/builtin/glsl/ext/FFT/types.glsl>
 
-// Input and Output Descriptor
-layout(set=0, binding=0) buffer restrict InputOutputBuffer
-{
-	nbl_glsl_ext_FFT_storage_t inoutData[];
-};
-#define _NBL_GLSL_EXT_FFT_INPUT_DESCRIPTOR_DEFINED_
-#define _NBL_GLSL_EXT_FFT_OUTPUT_DESCRIPTOR_DEFINED_
-
-layout(set=0, binding=1) uniform sampler2D NormalizedKernel[3];
+layout(set=0, binding=2) uniform sampler2D NormalizedKernel[3];
 
 #include "convolve_parameters.glsl"
 layout(push_constant) uniform PushConstants
@@ -25,9 +17,6 @@ nbl_glsl_ext_FFT_Parameters_t nbl_glsl_ext_FFT_getParameters()
 }
 #define _NBL_GLSL_EXT_FFT_GET_PARAMETERS_DEFINED_
 
-// do I need to unrestrict these buffers?
-#define inData inoutData
-#define outData inoutData
 #define _NBL_GLSL_EXT_FFT_MAIN_DEFINED_
 #include "nbl/builtin/glsl/ext/FFT/default_compute_fft.comp"
 

examples_tests/49.ComputeFFT/main.cpp

@@ -71,35 +71,42 @@ R"===(#version 430 core
 inline void updateDescriptorSet_Convolution (
 	video::IVideoDriver * driver,
 	video::IGPUDescriptorSet * set,
-	core::smart_refctd_ptr<video::IGPUBuffer> inputOutputBufferDescriptor,
+	core::smart_refctd_ptr<video::IGPUBuffer> inputBufferDescriptor,
+	core::smart_refctd_ptr<video::IGPUBuffer> outputBufferDescriptor,
 	const core::smart_refctd_ptr<video::IGPUImageView>* kernelNormalizedSpectrumImageDescriptors)
 {
-	constexpr uint32_t descCount = 2u;
-	video::IGPUDescriptorSet::SDescriptorInfo pInfos[1u+channelCountOverride];
+	constexpr uint32_t descCount = 3u;
+	video::IGPUDescriptorSet::SDescriptorInfo pInfos[descCount-1u+channelCountOverride];
 	video::IGPUDescriptorSet::SWriteDescriptorSet pWrites[descCount];
 
 	for (auto i = 0; i < descCount; i++)
 	{
+		pWrites[i].binding = i;
 		pWrites[i].dstSet = set;
 		pWrites[i].arrayElement = 0u;
 		pWrites[i].info = pInfos+i;
 	}
 
-	// InputOutput Buffer 
-	pWrites[0].binding = 0;
+	// Input Buffer 
 	pWrites[0].descriptorType = asset::EDT_STORAGE_BUFFER;
 	pWrites[0].count = 1;
-	pInfos[0].desc = inputOutputBufferDescriptor;
-	pInfos[0].buffer.size = inputOutputBufferDescriptor->getSize();
+	pInfos[0].desc = inputBufferDescriptor;
+	pInfos[0].buffer.size = inputBufferDescriptor->getSize();
 	pInfos[0].buffer.offset = 0u;
+	
+	//
+	pWrites[1].descriptorType = asset::EDT_STORAGE_BUFFER;
+	pWrites[1].count = 1;
+	pInfos[1].desc = outputBufferDescriptor;
+	pInfos[1].buffer.size = outputBufferDescriptor->getSize();
+	pInfos[1].buffer.offset = 0u;
 
 	// Kernel Buffer 
-	pWrites[1].binding = 1;
-	pWrites[1].descriptorType = asset::EDT_COMBINED_IMAGE_SAMPLER;
-	pWrites[1].count = channelCountOverride;
+	pWrites[2].descriptorType = asset::EDT_COMBINED_IMAGE_SAMPLER;
+	pWrites[2].count = channelCountOverride;
 	for (uint32_t i=0u; i<channelCountOverride; i++)
 	{
-		auto& info = pInfos[1u+i];
+		auto& info = pInfos[2u+i];
 		info.desc = kernelNormalizedSpectrumImageDescriptors[i];
 		//info.image.imageLayout = ;
 		info.image.sampler = nullptr;
@@ -330,6 +337,13 @@ int main()
 			},
 			{
 				1u,
+				EDT_STORAGE_BUFFER,
+				1u,
+				ISpecializedShader::ESS_COMPUTE,
+				nullptr
+			},
+			{
+				2u,
 				EDT_COMBINED_IMAGE_SAMPLER,
 				channelCountOverride,
 				ISpecializedShader::ESS_COMPUTE,
@@ -579,11 +593,11 @@ int main()
 
 	// Convolution
 	auto convolveDescriptorSet = driver->createGPUDescriptorSet(core::smart_refctd_ptr<const IGPUDescriptorSetLayout>(convolvePipeline->getLayout()->getDescriptorSetLayout(0u)));
-	updateDescriptorSet_Convolution(driver, convolveDescriptorSet.get(), fftOutputBuffer_0, kernelNormalizedSpectrums);
+	updateDescriptorSet_Convolution(driver, convolveDescriptorSet.get(), fftOutputBuffer_0, fftOutputBuffer_1, kernelNormalizedSpectrums);
 
 	// Last IFFTX 
 	auto lastFFTDescriptorSet = driver->createGPUDescriptorSet(core::smart_refctd_ptr<const IGPUDescriptorSetLayout>(lastFFTPipeline->getLayout()->getDescriptorSetLayout(0u)));
-	updateDescriptorSet_LastFFT(driver, lastFFTDescriptorSet.get(), fftOutputBuffer_0, outImgView);
+	updateDescriptorSet_LastFFT(driver, lastFFTDescriptorSet.get(), fftOutputBuffer_1, outImgView);
 
 	uint32_t outBufferIx = 0u;
 	auto lastPresentStamp = std::chrono::high_resolution_clock::now();
@@ -601,7 +615,9 @@ int main()
 	const auto passes = FFTClass::buildParameters(false,srcNumChannels,srcDim,fftPushConstants,fftDispatchInfo,fftPadding,paddedSrcDim);
 	{
 		fftPushConstants[1].input_dimensions.x = 2048u;
-		fftPushConstants[1].output_strides = fftPushConstants[1].input_strides;
+		fftPushConstants[1].input_strides = fftPushConstants[0].output_strides;
+		fftPushConstants[1].output_strides.x = 2048u;
+		fftPushConstants[1].output_strides.y = 1u;
 		fftPushConstants[2] = fftPushConstants[0];
 		fftPushConstants[2].input_dimensions.x = 2048u;
 		fftPushConstants[2].input_dimensions.y = 2048u;

include/nbl/builtin/glsl/workgroup/fft.glsl

@@ -105,16 +105,5 @@ void nbl_glsl_workgroupFFT(in bool is_inverse, inout nbl_glsl_complex lo, inout
     }
 }
 
-#if 0 // TODO
-// Computes Forward FFT of two real signals
-void nbl_glsl_workgroupRealFFT(in bool is_inverse, in float sequenceALo, in float sequenceAHi, in float sequenceBLo, in float sequenceBHi)
-{
-    nbl_glsl_complex lo = nbl_glsl_complex(sequenceALo,sequenceBLo);
-    nbl_glsl_complex hi = nbl_glsl_complex(sequenceAHi,sequenceBHi);
-    nbl_glsl_workgroupFFT(false,lo,hi);
-    // extract aDFT and bDFT by using sorensens method
-}
-#endif
-
 
 #endif