Fixed more filter stuff, offsetDifference needs to be measured in blocks

Author: Erfan-Ahmadi

include/nbl/asset/filters/CCopyImageFilter.h

@@ -65,9 +65,8 @@ class NBL_API CCopyImageFilter : public CImageFilter<CCopyImageFilter>, public C
 				const auto blockDims = asset::getBlockDimensions(commonExecuteData.inFormat);
 				auto copy = [&commonExecuteData,&blockDims](uint32_t readBlockArrayOffset, core::vectorSIMDu32 readBlockPos) -> void
 				{
-					const auto localOutPos = readBlockPos+commonExecuteData.offsetDifference;
-					const auto writeOffset = commonExecuteData.oit->getByteOffset(localOutPos,commonExecuteData.outByteStrides);
-					memcpy(commonExecuteData.outData+writeOffset,commonExecuteData.inData+readBlockArrayOffset,commonExecuteData.outBlockByteSize);
+					const auto localOutPos = readBlockPos + commonExecuteData.offsetDifference;
+					memcpy(commonExecuteData.outData + commonExecuteData.oit->getByteOffset(localOutPos, commonExecuteData.outBlockByteStrides), commonExecuteData.inData + readBlockArrayOffset, commonExecuteData.outBlockByteSize);
 				};
 				CBasicImageFilterCommon::executePerRegion<ExecutionPolicy>(policy,commonExecuteData.inImg,copy,commonExecuteData.inRegions.begin(),commonExecuteData.inRegions.end(),clip);
 

include/nbl/asset/filters/CMatchedSizeInOutImageFilterCommon.h

@@ -160,7 +160,8 @@ class NBL_API CMatchedSizeInOutImageFilterCommon : public CBasicImageFilterCommo
 			const core::SRange<const IImage::SBufferCopy> inRegions;
 			const core::SRange<const IImage::SBufferCopy> outRegions;
 			const IImage::SBufferCopy* oit;									//!< oit is a current output handled region by commonExecute lambda. Notice that the lambda may execute executePerRegion a few times with different oits data since regions may overlap in a certain mipmap in an image!
-			core::vectorSIMDu32 offsetDifference, outByteStrides; 
+			core::vectorSIMDu32 offsetDifference;
+			core::vectorSIMDu32 outBlockByteStrides;
 		};
 		template<typename PerOutputFunctor>
 		static inline bool commonExecute(state_type* state, PerOutputFunctor& perOutput)
@@ -190,17 +191,22 @@ class NBL_API CMatchedSizeInOutImageFilterCommon : public CBasicImageFilterCommo
 				outRegions.begin(), {}, {}
 			};
 
+			const asset::TexelBlockInfo srcImageTexelBlockInfo(inParams.format);
+			const asset::TexelBlockInfo dstImageTexelBlockInfo(outParams.format);
+
 			// iterate over output regions, then input cause read cache miss is faster
 			for (; commonExecuteData.oit!=commonExecuteData.outRegions.end(); commonExecuteData.oit++)
 			{
 				IImage::SSubresourceLayers subresource = {static_cast<IImage::E_ASPECT_FLAGS>(0u),state->inMipLevel,state->inBaseLayer,state->layerCount};
 				state_type::TexelRange range = {state->inOffset,state->extent};
 				CBasicImageFilterCommon::clip_region_functor_t clip(subresource,range,commonExecuteData.inFormat);
 				// setup convert state
-				// I know my two's complement wraparound well enough to make this work
 				const auto& outRegionOffset = commonExecuteData.oit->imageOffset;
-				commonExecuteData.offsetDifference = state->outOffsetBaseLayer - (core::vectorSIMDu32(outRegionOffset.x, outRegionOffset.y, outRegionOffset.z, commonExecuteData.oit->imageSubresource.baseArrayLayer) + state->inOffsetBaseLayer);
-				commonExecuteData.outByteStrides = commonExecuteData.oit->getByteStrides(TexelBlockInfo(commonExecuteData.outFormat));
+				const auto& inOffset = (core::vectorSIMDu32(outRegionOffset.x, outRegionOffset.y, outRegionOffset.z, commonExecuteData.oit->imageSubresource.baseArrayLayer) + state->inOffsetBaseLayer);
+
+				// offsetDifference types are uint but I know my two's complement wraparound well enough to make this work
+				commonExecuteData.offsetDifference = dstImageTexelBlockInfo.convertTexelsToBlocks(state->outOffsetBaseLayer) - srcImageTexelBlockInfo.convertTexelsToBlocks(inOffset);
+				commonExecuteData.outBlockByteStrides = commonExecuteData.oit->getByteStrides(TexelBlockInfo(commonExecuteData.outFormat));
 				if (!perOutput(commonExecuteData,clip))
 					return false;
 			}

include/nbl/asset/filters/CSwizzleAndConvertImageFilter.h

@@ -77,7 +77,7 @@ class NBL_API CSwizzleAndConvertImageFilterBase : public CSwizzleableAndDitherab
 								base_t::template onDecode<kInFormat,encodeBufferType>(state, srcPix, decodeBuffer, encodeBuffer, blockX, blockY);
 							else
 								base_t::template onDecode<encodeBufferType>(rInFormat, state, srcPix, decodeBuffer, encodeBuffer, blockX, blockY);
-							state->normalization.prepass(encodeBuffer,readBlockPos*blockDims+commonExecuteData.offsetDifference,blockX,blockY,4u/*TODO: figure this out*/);
+							state->normalization.prepass(encodeBuffer,readBlockPos+commonExecuteData.offsetDifference,blockX,blockY,4u/*TODO: figure this out*/);
 						}
 					};
 					CBasicImageFilterCommon::executePerRegion(policy, commonExecuteData.inImg, normalizePrepass, commonExecuteData.inRegions.begin(), commonExecuteData.inRegions.end(), clip);
@@ -146,8 +146,8 @@ class NBL_API CSwizzleAndConvertImageFilter : public CImageFilter<CSwizzleAndCon
 					for (auto blockY=0u; blockY<blockDims.y; blockY++)
 					for (auto blockX=0u; blockX<blockDims.x; blockX++)
 					{
-						auto localOutPos = readBlockPos*blockDims+commonExecuteData.offsetDifference;
-						uint8_t* dstPix = commonExecuteData.outData+commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY),commonExecuteData.outByteStrides);
+						auto localOutPos = readBlockPos+commonExecuteData.offsetDifference;
+						uint8_t* dstPix = commonExecuteData.outData+commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY),commonExecuteData.outBlockByteStrides);
 
 						constexpr auto maxChannels = 4;
 						decodeBufferType decodeBuffer[maxChannels] = {};
@@ -212,8 +212,8 @@ class NBL_API CSwizzleAndConvertImageFilter<EF_UNKNOWN,EF_UNKNOWN,Swizzle,Dither
 					for (auto blockY=0u; blockY<blockDims.y; blockY++)
 					for (auto blockX=0u; blockX<blockDims.x; blockX++)
 					{
-						auto localOutPos = readBlockPos*blockDims+commonExecuteData.offsetDifference;
-						uint8_t* dstPix = commonExecuteData.outData+commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY),commonExecuteData.outByteStrides);
+						auto localOutPos = readBlockPos+commonExecuteData.offsetDifference;
+						uint8_t* dstPix = commonExecuteData.outData+commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY),commonExecuteData.outBlockByteStrides);
 
 						constexpr auto maxChannels = 4;
 						double decodeBuffer[maxChannels] = {};
@@ -287,8 +287,8 @@ class NBL_API CSwizzleAndConvertImageFilter<EF_UNKNOWN,outFormat,Swizzle,Dither,
 					for (auto blockY = 0u; blockY < blockDims.y; blockY++)
 					for (auto blockX = 0u; blockX < blockDims.x; blockX++)
 					{
-						auto localOutPos = readBlockPos * blockDims + commonExecuteData.offsetDifference;
-						uint8_t* dstPix = commonExecuteData.outData + commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY), commonExecuteData.outByteStrides);
+						auto localOutPos = readBlockPos + commonExecuteData.offsetDifference;
+						uint8_t* dstPix = commonExecuteData.outData + commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY), commonExecuteData.outBlockByteStrides);
 
 						constexpr auto maxChannels = 4;
 						double decodeBuffer[maxChannels] = {};
@@ -363,8 +363,8 @@ class NBL_API CSwizzleAndConvertImageFilter<inFormat,EF_UNKNOWN,Swizzle,Dither,N
 					for (auto blockY = 0u; blockY < blockDims.y; blockY++)
 						for (auto blockX = 0u; blockX < blockDims.x; blockX++)
 						{
-							auto localOutPos = readBlockPos * blockDims + commonExecuteData.offsetDifference;
-							uint8_t* dstPix = commonExecuteData.outData + commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY), commonExecuteData.outByteStrides);
+							auto localOutPos = readBlockPos + commonExecuteData.offsetDifference;
+							uint8_t* dstPix = commonExecuteData.outData + commonExecuteData.oit->getByteOffset(localOutPos + core::vectorSIMDu32(blockX, blockY), commonExecuteData.outBlockByteStrides);
 
 							constexpr auto maxChannels = 4;
 							decodeBufferType decodeBuffer[maxChannels] = {};