optimize normalization states and not make them process garbage

Author: devshgraphicsprogramming

include/nbl/asset/filters/NormalizationStates.h

@@ -1,7 +1,6 @@
-// Copyright (C) 2018-2020 - DevSH Graphics Programming Sp. z O.O.
+// Copyright (C) 2018-2023 - DevSH Graphics Programming Sp. z O.O.
 // This file is part of the "Nabla Engine".
 // For conditions of distribution and use, see copyright notice in nabla.h
-
 #ifndef _NBL_ASSET_NORMALIZATION_STATES_H_INCLUDED_
 #define _NBL_ASSET_NORMALIZATION_STATES_H_INCLUDED_
 
@@ -40,7 +39,7 @@ class CGlobalNormalizationState
 		void prepass(Tenc* encodeBuffer, const core::vectorSIMDu32& position, uint32_t blockX, uint32_t blockY, uint8_t channels)
 		{
 			static_assert(std::is_floating_point_v<Tenc>, "Integer decode not supported yet!");
-			for (uint8_t channel=0u; channel<4u; ++channel)
+			for (uint8_t channel=0u; channel<channels; ++channel)
 			{
 				const auto val = encodeBuffer[channel];
 				if constexpr (std::is_floating_point_v<Tenc>)
@@ -112,6 +111,23 @@ namespace impl
 
 class CDerivativeMapNormalizationStateBase
 {
+	protected:
+		// we only care about R and G
+		static inline constexpr uint32_t kChannels = 2;
+
+		template<bool isSignedFormat, typename Tenc>
+		void impl(Tenc* encodeBuffer, const core::vectorSIMDu32& position, uint32_t blockX, uint32_t blockY, uint8_t channels) const
+		{
+			static_assert(std::is_floating_point_v<Tenc>, "Encode types must be double or float!");
+
+			if constexpr (isSignedFormat)
+				for (uint8_t channel = 0; channel < kChannels; ++channel)
+					encodeBuffer[channel] = encodeBuffer[channel]/maxAbsPerChannel[channel];
+			else
+				for (uint8_t channel = 0; channel < kChannels; ++channel)
+					encodeBuffer[channel] = encodeBuffer[channel]*0.5f/maxAbsPerChannel[channel]+0.5f;
+		}
+
 	public:
 		inline bool validate() const {return true;}
 
@@ -120,15 +136,15 @@ class CDerivativeMapNormalizationStateBase
 		inline void initialize()
 		{
 			static_assert(std::is_floating_point_v<Tenc>, "Integer encode not supported yet!");
-			std::fill_n(maxAbsPerChannel,4,0.f);
+			std::fill_n(maxAbsPerChannel,kChannels,0.f);
 		}
 
 		//
 		template<typename Tenc>
 		void prepass(Tenc* encodeBuffer, const core::vectorSIMDu32& position, uint32_t blockX, uint32_t blockY, uint8_t channels)
 		{
 			static_assert(std::is_floating_point_v<Tenc>, "Integer encode not supported yet!");
-			for (uint8_t channel=0u; channel<4u; ++channel)
+			for (uint8_t channel=0u; channel<kChannels; ++channel)
 				core::atomic_fetch_max(maxAbsPerChannel+channel,core::abs(encodeBuffer[channel]));
 		}
 
@@ -144,6 +160,7 @@ class CDerivativeMapNormalizationStateBase
 		template<typename Tenc>
 		void operator()(E_FORMAT format, Tenc* encodeBuffer, const core::vectorSIMDu32& position, uint32_t blockX, uint32_t blockY, uint8_t channels) const
 		{
+			assert(channels>=kChannels);
 			#ifdef _NBL_DEBUG
 			bool status = isFloatingPointFormat(format)||isNormalizedFormat(format);
 			assert(status);
@@ -155,20 +172,7 @@ class CDerivativeMapNormalizationStateBase
 				impl<false,Tenc>(encodeBuffer,position,blockX,blockY,channels);
 		}
 
-		core::atomic<float> maxAbsPerChannel[4];
-	protected:
-		template<bool isSignedFormat, typename Tenc>
-		void impl(Tenc* encodeBuffer, const core::vectorSIMDu32& position, uint32_t blockX, uint32_t blockY, uint8_t channels) const
-		{
-			static_assert(std::is_floating_point_v<Tenc>, "Encode types must be double or float!");
-
-			if constexpr (isSignedFormat)
-				for (uint8_t channel = 0; channel < channels; ++channel)
-					encodeBuffer[channel] = encodeBuffer[channel]/maxAbsPerChannel[channel];
-			else
-				for (uint8_t channel = 0; channel < channels; ++channel)
-					encodeBuffer[channel] = encodeBuffer[channel]*0.5f/maxAbsPerChannel[channel]+0.5f;
-		}
+		core::atomic<float> maxAbsPerChannel[kChannels];
 };
 
 }
@@ -183,8 +187,8 @@ class CDerivativeMapNormalizationState : public impl::CDerivativeMapNormalizatio
 			static_assert(std::is_floating_point_v<Tenc>, "Integer encode types not supported yet!");
 			if constexpr (isotropic)
 			{
-				const float isotropicMax = core::max(core::max(maxAbsPerChannel[0],maxAbsPerChannel[1]),core::max(maxAbsPerChannel[2],maxAbsPerChannel[3]));
-				for (auto i=0u; i<4u; i++)
+				const float isotropicMax = core::max<float>(maxAbsPerChannel[0],maxAbsPerChannel[1]);
+				for (auto i=0u; i<kChannels; i++)
 					maxAbsPerChannel[i] = isotropicMax;
 			}
 		}