Fix wrapping

A) Nvidia has a driver bug with fallthroughs in switches
B) proper % operator does not exist in GLSL

Author: devshgraphicsprogramming

examples_tests/49.ComputeFFT/fft_convolve_ifft.comp

@@ -50,7 +50,7 @@ void main()
 		{
 			const uint tid = (t<<_NBL_GLSL_WORKGROUP_SIZE_LOG2_)|gl_LocalInvocationIndex;
 			const uint trueDim = nbl_glsl_ext_FFT_Parameters_t_getDimensions()[nbl_glsl_ext_FFT_Parameters_t_getDirection()];
-			nbl_glsl_ext_FFT_impl_values[t] = nbl_glsl_ext_FFT_getPaddedData(ivec3(nbl_glsl_ext_FFT_getCoordinates(tid)),ch);
+			nbl_glsl_ext_FFT_impl_values[t] = nbl_glsl_ext_FFT_getPaddedData(nbl_glsl_ext_FFT_getPaddedCoordinates(tid,log2FFTSize,trueDim),ch);
 		}
 		nbl_glsl_ext_FFT_preloaded(false,log2FFTSize);
 		barrier();

examples_tests/49.ComputeFFT/last_fft.comp

@@ -8,7 +8,7 @@ layout(set=0, binding=1, rgba16f) uniform image2D outImage;
 void nbl_glsl_ext_FFT_setData(in uvec3 coordinate, in uint channel, in nbl_glsl_complex complex_value)
 {
 // TODO PC
-    const ivec2 padding = imageSize(outImage).x!=512u ? ivec2(384,/*664*/0):ivec2(0);
+    const ivec2 padding = imageSize(outImage).x!=512u ? ivec2(384,664):ivec2(0);
 	const ivec2 coords = ivec2(coordinate.xy)-padding;
 
     if (all(lessThanEqual(ivec2(0),coords))&&all(greaterThan(imageSize(outImage),coords)))

examples_tests/49.ComputeFFT/main.cpp

@@ -615,12 +615,14 @@ int main()
 	const auto passes = FFTClass::buildParameters(false,srcNumChannels,srcDim,fftPushConstants,fftDispatchInfo,fftPadding,paddedSrcDim);
 	{
 		fftPushConstants[2].input_dimensions = fftPushConstants[1].input_dimensions;
+		fftPushConstants[2].input_dimensions.y = 2048u;
 		{
 			fftPushConstants[2].input_dimensions.w = fftPushConstants[0].input_dimensions.w^0x80000000u;
 			fftPushConstants[2].input_strides = fftPushConstants[1].output_strides;
 			fftPushConstants[2].output_strides = fftPushConstants[0].input_strides;
 		}
 		fftDispatchInfo[2] = fftDispatchInfo[0];
+		fftDispatchInfo[2].workGroupCount[1] = 2048;
 	}
 	assert(passes==2);
 

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

@@ -78,25 +78,33 @@ nbl_glsl_complex nbl_glsl_ext_FFT_getPaddedData(ivec3 coordinate, in uint channe
 	const ivec3 max_coord = dimensions-ivec3(1u);
 
 	const ivec3 clamped_coord = clamp(coordinate,ivec3(0),max_coord);
-
+	
 	const uint paddingType = nbl_glsl_ext_FFT_Parameters_t_getPaddingType();
 	switch (paddingType)
 	{
-		case _NBL_GLSL_EXT_FFT_PAD_CLAMP_TO_EDGE_:
-			coordinate = clamped_coord;
-			break;
 		case _NBL_GLSL_EXT_FFT_PAD_CLAMP_TO_BORDER_:
 			if (any(bvec3(coordinate!=clamped_coord)))
 				return nbl_glsl_complex(0,0);
 			coordinate = clamped_coord;
 			break;
-		case _NBL_GLSL_EXT_FFT_PAD_MIRROR_:
-			const bvec3 flip = bvec3((coordinate/dimensions)&0x1u);
-			coordinate %= dimensions;
-			coordinate = mix(coordinate,max_coord-coordinate,flip);
+		case _NBL_GLSL_EXT_FFT_PAD_CLAMP_TO_EDGE_:
+			coordinate = clamped_coord;
 			break;
-		default:
-			coordinate %= dimensions;
+		case _NBL_GLSL_EXT_FFT_PAD_REPEAT_:
+			{
+				const ivec3 negMask = coordinate>>31u;
+				const ivec3 d = ((coordinate^negMask)/dimensions)^negMask;
+				coordinate = coordinate-d*dimensions;
+			}
+			break;
+		case _NBL_GLSL_EXT_FFT_PAD_MIRROR_:
+			{
+				const ivec3 negMask = coordinate>>31u;
+				const ivec3 d = ((coordinate^negMask)/dimensions)^negMask;
+				coordinate = coordinate-d*dimensions;
+				const bvec3 flip = bvec3(d&0x1);
+				coordinate = mix(coordinate,max_coord-coordinate,flip);
+			}
 			break;
 	}