Use push constants for convoluton

Author: devshgraphicsprogramming

examples_tests/49.ComputeFFT/convolve_parameters.glsl

@@ -5,7 +5,6 @@
 #include "nbl/builtin/glsl/ext/FFT/parameters_struct.glsl"
 struct convolve_parameters_t
 {
-    nbl_glsl_ext_FFT_Parameters_t fft_params;
-    vec2    bitreversed_to_normalized;
+    nbl_glsl_ext_FFT_Parameters_t fft;
     vec2    kernel_half_pixel_size;
 };

examples_tests/49.ComputeFFT/fft_convolve_ifft.comp

@@ -13,6 +13,17 @@ layout(set=0, binding=0) buffer restrict InputOutputBuffer
 layout(set=0, binding=1) uniform sampler2D NormalizedKernel[3];
 
 #include "convolve_parameters.glsl"
+layout(push_constant) uniform PushConstants
+{
+	convolve_parameters_t params;
+} pc;
+#define _NBL_GLSL_EXT_FFT_PUSH_CONSTANTS_DEFINED_
+
+nbl_glsl_ext_FFT_Parameters_t nbl_glsl_ext_FFT_getParameters()
+{
+	return pc.params.fft;
+}
+#define _NBL_GLSL_EXT_FFT_GET_PARAMETERS_DEFINED_
 
 #define inData inoutData
 #define outData inoutData
@@ -26,14 +37,11 @@ void convolve(in uint item_per_thread_count, in uint ch)
 	{
 		const uint tid = _NBL_GLSL_WORKGROUP_SIZE_*t+gl_LocalInvocationIndex;
 
-		// TODO: do push constants here
-		uvec3 coords = nbl_glsl_ext_FFT_getCoordinates(tid);
-		const uvec3 log2_size = uvec3(11u, 10u, 0u);
-        coords = bitfieldReverse(coords)>>(uvec3(32u)-log2_size); // reverse_shifts
-
 		nbl_glsl_complex sourceSpectrum = nbl_glsl_ext_FFT_impl_values[t];
-
-        vec2 uv = (vec2(coords.xy))/vec2(uvec2(1u)<<log2_size.xy)+vec2(0.5f)/vec2(textureSize(NormalizedKernel[ch],0)); //kernel_half_pixel_size
+		
+		// TODO: do push constants here
+		const uvec3 coords = nbl_glsl_ext_FFT_getCoordinates(tid);
+        vec2 uv = vec2(bitfieldReverse(coords.xy))/vec2(4294967296.f)+pc.params.kernel_half_pixel_size;
 		//
 		nbl_glsl_complex convSpectrum = textureLod(NormalizedKernel[ch],uv,0).xy;
 		nbl_glsl_ext_FFT_impl_values[t] = nbl_glsl_complex_mul(sourceSpectrum,convSpectrum);

examples_tests/49.ComputeFFT/main.cpp

@@ -339,12 +339,11 @@ int main()
 			}
 		};
 
-		using FFTClass = ext::FFT::FFT;
-		core::SRange<const asset::SPushConstantRange> pcRange = FFTClass::getDefaultPushConstantRanges();
+		const asset::SPushConstantRange pcRange = {ISpecializedShader::ESS_COMPUTE,0u,sizeof(convolve_parameters_t)};
 		core::SRange<const video::IGPUDescriptorSetLayout::SBinding> bindings = {bnd,bnd+sizeof(bnd)/sizeof(IGPUDescriptorSetLayout::SBinding)};
 
 		return driver->createGPUPipelineLayout(
-			pcRange.begin(),pcRange.end(),
+			&pcRange,&pcRange+1,
 			driver->createGPUDescriptorSetLayout(bindings.begin(),bindings.end()),nullptr,nullptr,nullptr
 		);
 	}();
@@ -618,6 +617,13 @@ int main()
 		// Src Image FFT Y + Convolution + Convolved IFFT Y
 		driver->bindComputePipeline(convolvePipeline.get());
 		driver->bindDescriptorSets(EPBP_COMPUTE, convolvePipeline->getLayout(), 0u, 1u, &convolveDescriptorSet.get(), nullptr);
+		{
+			const auto& kernelImgExtent = kernelNormalizedSpectrums[0]->getCreationParameters().image->getCreationParameters().extent;
+			vec2 kernel_half_pixel_size{0.5f,0.5f};
+			kernel_half_pixel_size.x /= kernelImgExtent.width;
+			kernel_half_pixel_size.y /= kernelImgExtent.height;
+			driver->pushConstants(convolvePipeline->getLayout(),ISpecializedShader::ESS_COMPUTE,offsetof(convolve_parameters_t,kernel_half_pixel_size),sizeof(convolve_parameters_t::kernel_half_pixel_size),&kernel_half_pixel_size);
+		}
 		FFTClass::dispatchHelper(driver, convolvePipeline->getLayout(), fftPushConstants[1], fftDispatchInfo[1]);
 
 		// Last FFT Padding and Copy to GPU Image