more small changes

Author: Erfan-Ahmadi

include/nbl/builtin/glsl/ext/FFT/default_compute_fft.comp

@@ -7,21 +7,9 @@
 #error "USE_SSBO_FOR_INPUT should be defined."
 #endif
 
-#ifndef _NBL_GLSL_EXT_FFT_BLOCK_SIZE_X_DEFINED_
-#define _NBL_GLSL_EXT_FFT_BLOCK_SIZE_X_DEFINED_ 256
-#endif
-
-#ifndef _NBL_GLSL_EXT_FFT_BLOCK_SIZE_Y_DEFINED_
-#define _NBL_GLSL_EXT_FFT_BLOCK_SIZE_Y_DEFINED_ 1
-#endif
-
-#ifndef _NBL_GLSL_EXT_FFT_BLOCK_SIZE_Z_DEFINED_
-#define _NBL_GLSL_EXT_FFT_BLOCK_SIZE_Z_DEFINED_ 1
-#endif
-
-#define _NBL_GLSL_WORKGROUP_SIZE_ (_NBL_GLSL_EXT_FFT_BLOCK_SIZE_X_DEFINED_*_NBL_GLSL_EXT_FFT_BLOCK_SIZE_Y_DEFINED_*_NBL_GLSL_EXT_FFT_BLOCK_SIZE_Z_DEFINED_)
+#define _NBL_GLSL_WORKGROUP_SIZE_ 256
 
-layout(local_size_x=_NBL_GLSL_EXT_FFT_BLOCK_SIZE_X_DEFINED_, local_size_y=_NBL_GLSL_EXT_FFT_BLOCK_SIZE_Y_DEFINED_, local_size_z=_NBL_GLSL_EXT_FFT_BLOCK_SIZE_Z_DEFINED_) in;
+layout(local_size_x=_NBL_GLSL_WORKGROUP_SIZE_, local_size_y=1, local_size_z=1) in;
 
 
 #define _NBL_GLSL_EXT_FFT_GET_DATA_DEFINED_

include/nbl/ext/FFT/FFT.h

@@ -57,7 +57,7 @@ class FFT : public core::TotalInterface
 			assert(core::isPoT(paddedInputDimensions.width) && core::isPoT(paddedInputDimensions.height) && core::isPoT(paddedInputDimensions.depth));
 			DispatchInfo_t ret = {};
 
-			ret.workGroupDims[0] = DEFAULT_WORK_GROUP_X_DIM;
+			ret.workGroupDims[0] = DEFAULT_WORK_GROUP_SIZE;
 			ret.workGroupDims[1] = 1;
 			ret.workGroupDims[2] = 1;
 
@@ -227,7 +227,7 @@ class FFT : public core::TotalInterface
 		FFT() = delete;
 		//~FFT() = delete;
 
-		_NBL_STATIC_INLINE_CONSTEXPR uint32_t DEFAULT_WORK_GROUP_X_DIM = 256u;
+		_NBL_STATIC_INLINE_CONSTEXPR uint32_t DEFAULT_WORK_GROUP_SIZE = 256u;
 
 		static void defaultBarrier();
 };

src/nbl/ext/FFT/FFT.cpp

@@ -116,13 +116,13 @@ R"===(#version 430 core
 
 	const size_t extraSize = 32 + 32 + 32 + 32;
 
-	const uint32_t maxItemsPerThread = (maxPaddedDimensionSize - 1u) / (DEFAULT_WORK_GROUP_X_DIM) + 1u;
+	const uint32_t maxItemsPerThread = (maxPaddedDimensionSize - 1u) / (DEFAULT_WORK_GROUP_SIZE) + 1u;
 	const uint32_t useSSBO = (DataType::SSBO == inputType) ? 1 : 0;
 	auto shader = core::make_smart_refctd_ptr<ICPUBuffer>(strlen(sourceFmt)+extraSize+1u);
 	snprintf(
 		reinterpret_cast<char*>(shader->getPointer()),shader->getSize(), sourceFmt,
 		useSSBO,
-		DEFAULT_WORK_GROUP_X_DIM,
+		DEFAULT_WORK_GROUP_SIZE,
 		maxPaddedDimensionSize,
 		maxItemsPerThread
 	);
@@ -271,7 +271,7 @@ void FFT::updateDescriptorSet_KernelNormalization(
 }
 
 void FFT::dispatchKernelNormalization(video::IVideoDriver* driver, asset::VkExtent3D const & paddedDimension, uint32_t numChannels) {
-		const uint32_t dispatchSizeX = core::ceil(float(paddedDimension.width * paddedDimension.height * paddedDimension.depth * numChannels) / DEFAULT_WORK_GROUP_X_DIM);
+		const uint32_t dispatchSizeX = core::ceil(float(paddedDimension.width * paddedDimension.height * paddedDimension.depth * numChannels) / DEFAULT_WORK_GROUP_SIZE);
 		driver->dispatch(dispatchSizeX, 1, 1);
 		defaultBarrier();
 }