diff --git a/Assets/Resources/FSR2.meta b/Assets/Resources/FSR2.meta new file mode 100644 index 0000000..6d47067 --- /dev/null +++ b/Assets/Resources/FSR2.meta @@ -0,0 +1,8 @@ +fileFormatVersion: 2 +guid: cad7d53fa2166a0449bec7a9b4f17d69 +folderAsset: yes +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.compute b/Assets/Resources/FSR2/ffx_fsr2_rcas_pass.compute similarity index 88% rename from Assets/Resources/Shaders/ffx_fsr2_rcas_pass.compute rename to Assets/Resources/FSR2/ffx_fsr2_rcas_pass.compute index f925fd2..68fc166 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.compute +++ b/Assets/Resources/FSR2/ffx_fsr2_rcas_pass.compute @@ -7,4 +7,4 @@ //#define FFX_FSR2_OPTION_HDR_COLOR_INPUT // This ought to be supported in the long run -#include "ffx_fsr2_rcas_pass.hlsl" +#include "shaders/ffx_fsr2_rcas_pass.hlsl" diff --git a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.compute.meta b/Assets/Resources/FSR2/ffx_fsr2_rcas_pass.compute.meta similarity index 100% rename from Assets/Resources/Shaders/ffx_fsr2_rcas_pass.compute.meta rename to Assets/Resources/FSR2/ffx_fsr2_rcas_pass.compute.meta diff --git a/Assets/Resources/FSR2/shaders.meta b/Assets/Resources/FSR2/shaders.meta new file mode 100644 index 0000000..676ff9a --- /dev/null +++ b/Assets/Resources/FSR2/shaders.meta @@ -0,0 +1,8 @@ +fileFormatVersion: 2 +guid: 4a24e63edc822264a871f58397325d51 +folderAsset: yes +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_common_types.h b/Assets/Resources/FSR2/shaders/ffx_common_types.h similarity index 100% rename from Assets/Resources/Shaders/ffx_common_types.h rename to Assets/Resources/FSR2/shaders/ffx_common_types.h diff --git a/Assets/Resources/Shaders/ffx_common_types.h.meta b/Assets/Resources/FSR2/shaders/ffx_common_types.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_common_types.h.meta rename to Assets/Resources/FSR2/shaders/ffx_common_types.h.meta index 18c0534..203c57a 100644 --- a/Assets/Resources/Shaders/ffx_common_types.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_common_types.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: a44469065afba1e40ab87f099c2c8848 +guid: c6da07d8aae05f04f87e4db20f84c73e PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/Shaders/ffx_core.h b/Assets/Resources/FSR2/shaders/ffx_core.h similarity index 100% rename from Assets/Resources/Shaders/ffx_core.h rename to Assets/Resources/FSR2/shaders/ffx_core.h diff --git a/Assets/Resources/Shaders/ffx_core.h.meta b/Assets/Resources/FSR2/shaders/ffx_core.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_core.h.meta rename to Assets/Resources/FSR2/shaders/ffx_core.h.meta index 3c4c9fd..db53534 100644 --- a/Assets/Resources/Shaders/ffx_core.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_core.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 35579ac40cf103a49a29966f8d991a61 +guid: b37eb663a0ae01b469b0b5a54365b301 PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/FSR2/shaders/ffx_core_cpu.h b/Assets/Resources/FSR2/shaders/ffx_core_cpu.h new file mode 100644 index 0000000..9bb9915 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_core_cpu.h @@ -0,0 +1,332 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// A define for a true value in a boolean expression. +/// +/// @ingroup CPU +#define FFX_TRUE (1) + +/// A define for a false value in a boolean expression. +/// +/// @ingroup CPU +#define FFX_FALSE (0) + +#if !defined(FFX_STATIC) +/// A define to abstract declaration of static variables and functions. +/// +/// @ingroup CPU +#define FFX_STATIC static +#endif // #if !defined(FFX_STATIC) + +#ifdef __clang__ +#pragma clang diagnostic ignored "-Wunused-variable" +#endif + +/// Interpret the bit layout of an IEEE-754 floating point value as an unsigned integer. +/// +/// @param [in] x A 32bit floating value. +/// +/// @returns +/// An unsigned 32bit integer value containing the bit pattern of x. +/// +/// @ingroup CPU +FFX_STATIC FfxUInt32 ffxAsUInt32(FfxFloat32 x) +{ + union + { + FfxFloat32 f; + FfxUInt32 u; + } bits; + + bits.f = x; + return bits.u; +} + +FFX_STATIC FfxFloat32 ffxDot2(FfxFloat32x2 a, FfxFloat32x2 b) +{ + return a[0] * b[0] + a[1] * b[1]; +} + +FFX_STATIC FfxFloat32 ffxDot3(FfxFloat32x3 a, FfxFloat32x3 b) +{ + return a[0] * b[0] + a[1] * b[1] + a[2] * b[2]; +} + +FFX_STATIC FfxFloat32 ffxDot4(FfxFloat32x4 a, FfxFloat32x4 b) +{ + return a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3]; +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxLerp(FfxFloat32 x, FfxFloat32 y, FfxFloat32 t) +{ + return y * t + (-x * t + x); +} + +/// Compute the reciprocal of a value. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxReciprocal(FfxFloat32 a) +{ + return 1.0f / a; +} + +/// Compute the square root of a value. +/// +/// @param [in] x The first value to compute the min of. +/// +/// @returns +/// The the square root of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxSqrt(FfxFloat32 x) +{ + return sqrt(x); +} + +FFX_STATIC FfxUInt32 AShrSU1(FfxUInt32 a, FfxUInt32 b) +{ + return FfxUInt32(FfxInt32(a) >> FfxInt32(b)); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxFract(FfxFloat32 a) +{ + return a - floor(a); +} + +/// Compute the reciprocal square root of a value. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 rsqrt(FfxFloat32 a) +{ + return ffxReciprocal(ffxSqrt(a)); +} + +FFX_STATIC FfxFloat32 ffxMin(FfxFloat32 x, FfxFloat32 y) +{ + return x < y ? x : y; +} + +FFX_STATIC FfxUInt32 ffxMin(FfxUInt32 x, FfxUInt32 y) +{ + return x < y ? x : y; +} + +FFX_STATIC FfxFloat32 ffxMax(FfxFloat32 x, FfxFloat32 y) +{ + return x > y ? x : y; +} + +FFX_STATIC FfxUInt32 ffxMax(FfxUInt32 x, FfxUInt32 y) +{ + return x > y ? x : y; +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup CPU +FFX_STATIC FfxFloat32 ffxSaturate(FfxFloat32 a) +{ + return ffxMin(1.0f, ffxMax(0.0f, a)); +} + +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +FFX_STATIC void opAAddOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b) +{ + d[0] = a[0] + b; + d[1] = a[1] + b; + d[2] = a[2] + b; + return; +} + +FFX_STATIC void opACpyF3(FfxFloat32x3 d, FfxFloat32x3 a) +{ + d[0] = a[0]; + d[1] = a[1]; + d[2] = a[2]; + return; +} + +FFX_STATIC void opAMulF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32x3 b) +{ + d[0] = a[0] * b[0]; + d[1] = a[1] * b[1]; + d[2] = a[2] * b[2]; + return; +} + +FFX_STATIC void opAMulOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b) +{ + d[0] = a[0] * b; + d[1] = a[1] * b; + d[2] = a[2] * b; + return; +} + +FFX_STATIC void opARcpF3(FfxFloat32x3 d, FfxFloat32x3 a) +{ + d[0] = ffxReciprocal(a[0]); + d[1] = ffxReciprocal(a[1]); + d[2] = ffxReciprocal(a[2]); + return; +} + +/// Convert FfxFloat32 to half (in lower 16-bits of output). +/// +/// This function implements the same fast technique that is documented here: ftp://ftp.fox-toolkit.org/pub/fasthalffloatconversion.pdf +/// +/// The function supports denormals. +/// +/// Some conversion rules are to make computations possibly "safer" on the GPU, +/// -INF & -NaN -> -65504 +/// +INF & +NaN -> +65504 +/// +/// @param [in] f The 32bit floating point value to convert. +/// +/// @returns +/// The closest 16bit floating point value to f. +/// +/// @ingroup CPU +FFX_STATIC FfxUInt32 f32tof16(FfxFloat32 f) +{ + static FfxUInt16 base[512] = { + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100, 0x0200, 0x0400, + 0x0800, 0x0c00, 0x1000, 0x1400, 0x1800, 0x1c00, 0x2000, 0x2400, 0x2800, 0x2c00, 0x3000, 0x3400, 0x3800, 0x3c00, 0x4000, 0x4400, 0x4800, 0x4c00, 0x5000, + 0x5400, 0x5800, 0x5c00, 0x6000, 0x6400, 0x6800, 0x6c00, 0x7000, 0x7400, 0x7800, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8001, 0x8002, + 0x8004, 0x8008, 0x8010, 0x8020, 0x8040, 0x8080, 0x8100, 0x8200, 0x8400, 0x8800, 0x8c00, 0x9000, 0x9400, 0x9800, 0x9c00, 0xa000, 0xa400, 0xa800, 0xac00, + 0xb000, 0xb400, 0xb800, 0xbc00, 0xc000, 0xc400, 0xc800, 0xcc00, 0xd000, 0xd400, 0xd800, 0xdc00, 0xe000, 0xe400, 0xe800, 0xec00, 0xf000, 0xf400, 0xf800, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff + }; + + static FfxUInt8 shift[512] = { + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18 + }; + + union + { + FfxFloat32 f; + FfxUInt32 u; + } bits; + + bits.f = f; + FfxUInt32 u = bits.u; + FfxUInt32 i = u >> 23; + return (FfxUInt32)(base[i]) + ((u & 0x7fffff) >> shift[i]); +} + +/// Pack 2x32-bit floating point values in a single 32bit value. +/// +/// This function first converts each component of value into their nearest 16-bit floating +/// point representation, and then stores the X and Y components in the lower and upper 16 bits of the +/// 32bit unsigned integer respectively. +/// +/// @param [in] value A 2-dimensional floating point value to convert and pack. +/// +/// @returns +/// A packed 32bit value containing 2 16bit floating point values. +/// +/// @ingroup CPU +FFX_STATIC FfxUInt32 packHalf2x16(FfxFloat32x2 a) +{ + return f32tof16(a[0]) + (f32tof16(a[1]) << 16); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_core_cpu.h.meta b/Assets/Resources/FSR2/shaders/ffx_core_cpu.h.meta new file mode 100644 index 0000000..704e2e3 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_core_cpu.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 9d1d6ed5c9da0c64b882f3ebc2bac307 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_core_glsl.h b/Assets/Resources/FSR2/shaders/ffx_core_glsl.h new file mode 100644 index 0000000..e419e39 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_core_glsl.h @@ -0,0 +1,1562 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +/// A define for abstracting shared memory between shading languages. +/// +/// @ingroup GPU +#define FFX_GROUPSHARED shared + +/// A define for abstracting compute memory barriers between shading languages. +/// +/// @ingroup GPU +#define FFX_GROUP_MEMORY_BARRIER() barrier() + +/// A define added to accept static markup on functions to aid CPU/GPU portability of code. +/// +/// @ingroup GPU +#define FFX_STATIC + +/// A define for abstracting loop unrolling between shading languages. +/// +/// @ingroup GPU +#define FFX_UNROLL + +/// A define for abstracting a 'greater than' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_GREATER_THAN(x, y) greaterThan(x, y) + +/// A define for abstracting a 'greater than or equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_GREATER_THAN_EQUAL(x, y) greaterThanEqual(x, y) + +/// A define for abstracting a 'less than' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_LESS_THAN(x, y) lessThan(x, y) + +/// A define for abstracting a 'less than or equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_LESS_THAN_EQUAL(x, y) lessThanEqual(x, y) + +/// A define for abstracting an 'equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_EQUAL(x, y) equal(x, y) + +/// A define for abstracting a 'not equal' comparison operator between two types. +/// +/// @ingroup GPU +#define FFX_NOT_EQUAL(x, y) notEqual(x, y) + +/// Broadcast a scalar value to a 1-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32(x) FfxFloat32(x) + +/// Broadcast a scalar value to a 2-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X2(x) FfxFloat32x2(FfxFloat32(x)) + +/// Broadcast a scalar value to a 3-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X3(x) FfxFloat32x3(FfxFloat32(x)) + +/// Broadcast a scalar value to a 4-dimensional floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_FLOAT32X4(x) FfxFloat32x4(FfxFloat32(x)) + +/// Broadcast a scalar value to a 1-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32(x) FfxUInt32(x) + +/// Broadcast a scalar value to a 2-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X2(x) FfxUInt32x2(FfxUInt32(x)) + +/// Broadcast a scalar value to a 3-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X3(x) FfxUInt32x3(FfxUInt32(x)) + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_UINT32X4(x) FfxUInt32x4(FfxUInt32(x)) + +/// Broadcast a scalar value to a 1-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32(x) FfxInt32(x) + +/// Broadcast a scalar value to a 2-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X2(x) FfxInt32x2(FfxInt32(x)) + +/// Broadcast a scalar value to a 3-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X3(x) FfxInt32x3(FfxInt32(x)) + +/// Broadcast a scalar value to a 4-dimensional signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_INT32X4(x) FfxInt32x4(FfxInt32(x)) + +/// Broadcast a scalar value to a 1-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16(x) FFX_MIN16_F(x) + +/// Broadcast a scalar value to a 2-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X2(x) FFX_MIN16_F2(FFX_MIN16_F(x)) + +/// Broadcast a scalar value to a 3-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X3(x) FFX_MIN16_F3(FFX_MIN16_F(x)) + +/// Broadcast a scalar value to a 4-dimensional half-precision floating point vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_FLOAT16X4(x) FFX_MIN16_F4(FFX_MIN16_F(x)) + +/// Broadcast a scalar value to a 1-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16(x) FFX_MIN16_U(x) + +/// Broadcast a scalar value to a 2-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X2(x) FFX_MIN16_U2(FFX_MIN16_U(x)) + +/// Broadcast a scalar value to a 3-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X3(x) FFX_MIN16_U3(FFX_MIN16_U(x)) + +/// Broadcast a scalar value to a 4-dimensional half-precision unsigned integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_UINT16X4(x) FFX_MIN16_U4(FFX_MIN16_U(x)) + +/// Broadcast a scalar value to a 1-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16(x) FFX_MIN16_I(x) + +/// Broadcast a scalar value to a 2-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X2(x) FFX_MIN16_I2(FFX_MIN16_I(x)) + +/// Broadcast a scalar value to a 3-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X3(x) FFX_MIN16_I3(FFX_MIN16_I(x)) + +/// Broadcast a scalar value to a 4-dimensional half-precision signed integer vector. +/// +/// @ingroup GPU +#define FFX_BROADCAST_MIN_INT16X4(x) FFX_MIN16_I4(FFX_MIN16_I(x)) + +#if !defined(FFX_SKIP_EXT) +#if FFX_HALF + #extension GL_EXT_shader_16bit_storage : require + #extension GL_EXT_shader_explicit_arithmetic_types : require +#endif // FFX_HALF + +#if defined(FFX_LONG) + #extension GL_ARB_gpu_shader_int64 : require + #extension GL_NV_shader_atomic_int64 : require +#endif // #if defined(FFX_LONG) + +#if defined(FFX_WAVE) + #extension GL_KHR_shader_subgroup_arithmetic : require + #extension GL_KHR_shader_subgroup_ballot : require + #extension GL_KHR_shader_subgroup_quad : require + #extension GL_KHR_shader_subgroup_shuffle : require +#endif // #if defined(FFX_WAVE) +#endif // #if !defined(FFX_SKIP_EXT) + +// Forward declarations +FfxFloat32 ffxSqrt(FfxFloat32 x); +FfxFloat32x2 ffxSqrt(FfxFloat32x2 x); +FfxFloat32x3 ffxSqrt(FfxFloat32x3 x); +FfxFloat32x4 ffxSqrt(FfxFloat32x4 x); + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32 ffxAsFloat(FfxUInt32 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxAsFloat(FfxUInt32x2 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxAsFloat(FfxUInt32x3 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as a floating-point number. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as a floating-point number. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxAsFloat(FfxUInt32x4 x) +{ + return uintBitsToFloat(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32 ffxAsUInt32(FfxFloat32 x) +{ + return floatBitsToUint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxAsUInt32(FfxFloat32x2 x) +{ + return floatBitsToUint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxAsUInt32(FfxFloat32x3 x) +{ + return floatBitsToUint(x); +} + +/// Interprets the bit pattern of x as an unsigned integer. +/// +/// @param [in] value The input value. +/// +/// @returns +/// The input interpreted as an unsigned integer. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxAsUInt32(FfxFloat32x4 x) +{ + return floatBitsToUint(x); +} + +/// Convert a 32bit IEEE 754 floating point value to its nearest 16bit equivalent. +/// +/// @param [in] value The value to convert. +/// +/// @returns +/// The nearest 16bit equivalent of value. +/// +/// @ingroup GLSL +FfxUInt32 f32tof16(FfxFloat32 value) +{ + return packHalf2x16(FfxFloat32x2(value, 0.0)); +} + +/// Broadcast a scalar value to a 2-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional floating point vector with value in each component. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxBroadcast2(FfxFloat32 value) +{ + return FfxFloat32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional floating point vector with value in each component. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxBroadcast3(FfxFloat32 value) +{ + return FfxFloat32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional floating point vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional floating point vector with value in each component. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxBroadcast4(FfxFloat32 value) +{ + return FfxFloat32x4(value, value, value, value); +} + +/// Broadcast a scalar value to a 2-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional signed integer vector with value in each component. +/// +/// @ingroup GLSL +FfxInt32x2 ffxBroadcast2(FfxInt32 value) +{ + return FfxInt32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional signed integer vector with value in each component. +/// +/// @ingroup GLSL +FfxInt32x3 ffxBroadcast3(FfxInt32 value) +{ + return FfxInt32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional signed integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional signed integer vector with value in each component. +/// +/// @ingroup GLSL +FfxInt32x4 ffxBroadcast4(FfxInt32 value) +{ + return FfxInt32x4(value, value, value, value); +} + +/// Broadcast a scalar value to a 2-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 2-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxBroadcast2(FfxUInt32 value) +{ + return FfxUInt32x2(value, value); +} + +/// Broadcast a scalar value to a 3-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 3-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxBroadcast3(FfxUInt32 value) +{ + return FfxUInt32x3(value, value, value); +} + +/// Broadcast a scalar value to a 4-dimensional unsigned integer vector. +/// +/// @param [in] value The value to to broadcast. +/// +/// @returns +/// A 4-dimensional unsigned integer vector with value in each component. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxBroadcast4(FfxUInt32 value) +{ + return FfxUInt32x4(value, value, value, value); +} + +/// +/// +/// @ingroup GLSL +FfxUInt32 bitfieldExtract(FfxUInt32 src, FfxUInt32 off, FfxUInt32 bits) +{ + return bitfieldExtract(src, FfxInt32(off), FfxInt32(bits)); +} + +/// +/// +/// @ingroup GLSL +FfxUInt32 bitfieldInsert(FfxUInt32 src, FfxUInt32 ins, FfxUInt32 mask) +{ + return (ins & mask) | (src & (~mask)); +} + +// Proxy for V_BFI_B32 where the 'mask' is set as 'bits', 'mask=(1<mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32 ffxLerp(FfxFloat32 x, FfxFloat32 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxLerp(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxLerp(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxLerp(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxLerp(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxLerp(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32 t) +{ + return mix(x, y, t); +} + +/// Compute the linear interopation between two values. +/// +/// Implemented by calling the GLSL mix instrinsic function. Implements the +/// following math: +/// +/// (1 - t) * x + t * y +/// +/// @param [in] x The first value to lerp between. +/// @param [in] y The second value to lerp between. +/// @param [in] t The value to determine how much of x and how much of y. +/// +/// @returns +/// A linearly interpolated value between x and y according to t. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxLerp(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 t) +{ + return mix(x, y, t); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32 ffxMax3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxMax3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxMax3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxMax3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32 ffxMax3(FfxUInt32 x, FfxUInt32 y, FfxUInt32 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN or RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxMax3(FfxUInt32x2 x, FfxUInt32x2 y, FfxUInt32x2 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxMax3(FfxUInt32x3 x, FfxUInt32x3 y, FfxUInt32x3 z) +{ + return max(x, max(y, z)); +} + +/// Compute the maximum of three values. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the max calculation. +/// @param [in] y The second value to include in the max calcuation. +/// @param [in] z The third value to include in the max calcuation. +/// +/// @returns +/// The maximum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxMax3(FfxUInt32x4 x, FfxUInt32x4 y, FfxUInt32x4 z) +{ + return max(x, max(y, z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32 ffxMed3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxMed3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxMed3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the median of three values. +/// +/// NOTE: This function should compile down to a single V_MED3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the median calculation. +/// @param [in] y The second value to include in the median calcuation. +/// @param [in] z The third value to include in the median calcuation. +/// +/// @returns +/// The median value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxMed3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return max(min(x, y), min(max(x, y), z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN and RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32 ffxMin3(FfxFloat32 x, FfxFloat32 y, FfxFloat32 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxMin3(FfxFloat32x2 x, FfxFloat32x2 y, FfxFloat32x2 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxMin3(FfxFloat32x3 x, FfxFloat32x3 y, FfxFloat32x3 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxMin3(FfxFloat32x4 x, FfxFloat32x4 y, FfxFloat32x4 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32 ffxMin3(FfxUInt32 x, FfxUInt32 y, FfxUInt32 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x2 ffxMin3(FfxUInt32x2 x, FfxUInt32x2 y, FfxUInt32x2 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x3 ffxMin3(FfxUInt32x3 x, FfxUInt32x3 y, FfxUInt32x3 z) +{ + return min(x, min(y, z)); +} + +/// Compute the minimum of three values. +/// +/// NOTE: This function should compile down to a single V_MIN3_F32 operation on +/// GCN/RDNA hardware. +/// +/// @param [in] x The first value to include in the min calculation. +/// @param [in] y The second value to include in the min calcuation. +/// @param [in] z The third value to include in the min calcuation. +/// +/// @returns +/// The minimum value of x, y, and z. +/// +/// @ingroup GLSL +FfxUInt32x4 ffxMin3(FfxUInt32x4 x, FfxUInt32x4 y, FfxUInt32x4 z) +{ + return min(x, min(y, z)); +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32 rcp(FfxFloat32 x) +{ + return FfxFloat32(1.0) / x; +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32x2 rcp(FfxFloat32x2 x) +{ + return ffxBroadcast2(1.0) / x; +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32x3 rcp(FfxFloat32x3 x) +{ + return ffxBroadcast3(1.0) / x; +} + +/// Compute the reciprocal of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rcp can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal value of x. +/// +/// @ingroup GLSL +FfxFloat32x4 rcp(FfxFloat32x4 x) +{ + return ffxBroadcast4(1.0) / x; +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32 rsqrt(FfxFloat32 x) +{ + return FfxFloat32(1.0) / ffxSqrt(x); +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32x2 rsqrt(FfxFloat32x2 x) +{ + return ffxBroadcast2(1.0) / ffxSqrt(x); +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32x3 rsqrt(FfxFloat32x3 x) +{ + return ffxBroadcast3(1.0) / ffxSqrt(x); +} + +/// Compute the reciprocal square root of a value. +/// +/// NOTE: This function is only provided for GLSL. In HLSL the intrinsic function rsqrt can be used. +/// +/// @param [in] x The value to compute the reciprocal for. +/// +/// @returns +/// The reciprocal square root value of x. +/// +/// @ingroup GLSL +FfxFloat32x4 rsqrt(FfxFloat32x4 x) +{ + return ffxBroadcast4(1.0) / ffxSqrt(x); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32 ffxSaturate(FfxFloat32 x) +{ + return clamp(x, FfxFloat32(0.0), FfxFloat32(1.0)); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32x2 ffxSaturate(FfxFloat32x2 x) +{ + return clamp(x, ffxBroadcast2(0.0), ffxBroadcast2(1.0)); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32x3 ffxSaturate(FfxFloat32x3 x) +{ + return clamp(x, ffxBroadcast3(0.0), ffxBroadcast3(1.0)); +} + +/// Clamp a value to a [0..1] range. +/// +/// @param [in] x The value to clamp to [0..1] range. +/// +/// @returns +/// The clamped version of x. +/// +/// @ingroup GLSL +FfxFloat32x4 ffxSaturate(FfxFloat32x4 x) +{ + return clamp(x, ffxBroadcast4(0.0), ffxBroadcast4(1.0)); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32 ffxFract(FfxFloat32 x) +{ + return fract(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x2 ffxFract(FfxFloat32x2 x) +{ + return fract(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x3 ffxFract(FfxFloat32x3 x) +{ + return fract(x); +} + +/// Compute the factional part of a decimal value. +/// +/// This function calculates x - floor(x). Where floor is the intrinsic HLSL function. +/// +/// NOTE: This function should compile down to a single V_MAX3_F32 operation on GCN/RDNA hardware. It is +/// worth further noting that this function is intentionally distinct from the HLSL frac intrinsic +/// function. +/// +/// @param [in] x The value to compute the fractional part from. +/// +/// @returns +/// The fractional part of x. +/// +/// @ingroup HLSL +FfxFloat32x4 ffxFract(FfxFloat32x4 x) +{ + return fract(x); +} + +FfxUInt32 AShrSU1(FfxUInt32 a, FfxUInt32 b) +{ + return FfxUInt32(FfxInt32(a) >> FfxInt32(b)); +} + +#if FFX_HALF + +#define FFX_UINT32_TO_FLOAT16X2(x) unpackFloat2x16(FfxUInt32(x)) + +FfxFloat16x4 ffxUint32x2ToFloat16x4(FfxUInt32x2 x) +{ + return FfxFloat16x4(unpackFloat2x16(x.x), unpackFloat2x16(x.y)); +} +#define FFX_UINT32X2_TO_FLOAT16X4(x) ffxUint32x2ToFloat16x4(FfxUInt32x2(x)) +#define FFX_UINT32_TO_UINT16X2(x) unpackUint2x16(FfxUInt32(x)) +#define FFX_UINT32X2_TO_UINT16X4(x) unpackUint4x16(pack64(FfxUInt32x2(x))) +//------------------------------------------------------------------------------------------------------------------------------ +#define FFX_FLOAT16X2_TO_UINT32(x) packFloat2x16(FfxFloat16x2(x)) +FfxUInt32x2 ffxFloat16x4ToUint32x2(FfxFloat16x4 x) +{ + return FfxUInt32x2(packFloat2x16(x.xy), packFloat2x16(x.zw)); +} +#define FFX_FLOAT16X4_TO_UINT32X2(x) ffxFloat16x4ToUint32x2(FfxFloat16x4(x)) +#define FFX_UINT16X2_TO_UINT32(x) packUint2x16(FfxUInt16x2(x)) +#define FFX_UINT16X4_TO_UINT32X2(x) unpack32(packUint4x16(FfxUInt16x4(x))) +//============================================================================================================================== +#define FFX_TO_UINT16(x) halfBitsToUint16(FfxFloat16(x)) +#define FFX_TO_UINT16X2(x) halfBitsToUint16(FfxFloat16x2(x)) +#define FFX_TO_UINT16X3(x) halfBitsToUint16(FfxFloat16x3(x)) +#define FFX_TO_UINT16X4(x) halfBitsToUint16(FfxFloat16x4(x)) +//------------------------------------------------------------------------------------------------------------------------------ +#define FFX_TO_FLOAT16(x) uint16BitsToHalf(FfxUInt16(x)) +#define FFX_TO_FLOAT16X2(x) uint16BitsToHalf(FfxUInt16x2(x)) +#define FFX_TO_FLOAT16X3(x) uint16BitsToHalf(FfxUInt16x3(x)) +#define FFX_TO_FLOAT16X4(x) uint16BitsToHalf(FfxUInt16x4(x)) +//============================================================================================================================== +FfxFloat16 ffxBroadcastFloat16(FfxFloat16 a) +{ + return FfxFloat16(a); +} +FfxFloat16x2 ffxBroadcastFloat16x2(FfxFloat16 a) +{ + return FfxFloat16x2(a, a); +} +FfxFloat16x3 ffxBroadcastFloat16x3(FfxFloat16 a) +{ + return FfxFloat16x3(a, a, a); +} +FfxFloat16x4 ffxBroadcastFloat16x4(FfxFloat16 a) +{ + return FfxFloat16x4(a, a, a, a); +} +#define FFX_BROADCAST_FLOAT16(a) FfxFloat16(a) +#define FFX_BROADCAST_FLOAT16X2(a) FfxFloat16x2(FfxFloat16(a)) +#define FFX_BROADCAST_FLOAT16X3(a) FfxFloat16x3(FfxFloat16(a)) +#define FFX_BROADCAST_FLOAT16X4(a) FfxFloat16x4(FfxFloat16(a)) +//------------------------------------------------------------------------------------------------------------------------------ +FfxInt16 ffxBroadcastInt16(FfxInt16 a) +{ + return FfxInt16(a); +} +FfxInt16x2 ffxBroadcastInt16x2(FfxInt16 a) +{ + return FfxInt16x2(a, a); +} +FfxInt16x3 ffxBroadcastInt16x3(FfxInt16 a) +{ + return FfxInt16x3(a, a, a); +} +FfxInt16x4 ffxBroadcastInt16x4(FfxInt16 a) +{ + return FfxInt16x4(a, a, a, a); +} +#define FFX_BROADCAST_INT16(a) FfxInt16(a) +#define FFX_BROADCAST_INT16X2(a) FfxInt16x2(FfxInt16(a)) +#define FFX_BROADCAST_INT16X3(a) FfxInt16x3(FfxInt16(a)) +#define FFX_BROADCAST_INT16X4(a) FfxInt16x4(FfxInt16(a)) +//------------------------------------------------------------------------------------------------------------------------------ +FfxUInt16 ffxBroadcastUInt16(FfxUInt16 a) +{ + return FfxUInt16(a); +} +FfxUInt16x2 ffxBroadcastUInt16x2(FfxUInt16 a) +{ + return FfxUInt16x2(a, a); +} +FfxUInt16x3 ffxBroadcastUInt16x3(FfxUInt16 a) +{ + return FfxUInt16x3(a, a, a); +} +FfxUInt16x4 ffxBroadcastUInt16x4(FfxUInt16 a) +{ + return FfxUInt16x4(a, a, a, a); +} +#define FFX_BROADCAST_UINT16(a) FfxUInt16(a) +#define FFX_BROADCAST_UINT16X2(a) FfxUInt16x2(FfxUInt16(a)) +#define FFX_BROADCAST_UINT16X3(a) FfxUInt16x3(FfxUInt16(a)) +#define FFX_BROADCAST_UINT16X4(a) FfxUInt16x4(FfxUInt16(a)) +//============================================================================================================================== +FfxUInt16 ffxAbsHalf(FfxUInt16 a) +{ + return FfxUInt16(abs(FfxInt16(a))); +} +FfxUInt16x2 ffxAbsHalf(FfxUInt16x2 a) +{ + return FfxUInt16x2(abs(FfxInt16x2(a))); +} +FfxUInt16x3 ffxAbsHalf(FfxUInt16x3 a) +{ + return FfxUInt16x3(abs(FfxInt16x3(a))); +} +FfxUInt16x4 ffxAbsHalf(FfxUInt16x4 a) +{ + return FfxUInt16x4(abs(FfxInt16x4(a))); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxClampHalf(FfxFloat16 x, FfxFloat16 n, FfxFloat16 m) +{ + return clamp(x, n, m); +} +FfxFloat16x2 ffxClampHalf(FfxFloat16x2 x, FfxFloat16x2 n, FfxFloat16x2 m) +{ + return clamp(x, n, m); +} +FfxFloat16x3 ffxClampHalf(FfxFloat16x3 x, FfxFloat16x3 n, FfxFloat16x3 m) +{ + return clamp(x, n, m); +} +FfxFloat16x4 ffxClampHalf(FfxFloat16x4 x, FfxFloat16x4 n, FfxFloat16x4 m) +{ + return clamp(x, n, m); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxFract(FfxFloat16 x) +{ + return fract(x); +} +FfxFloat16x2 ffxFract(FfxFloat16x2 x) +{ + return fract(x); +} +FfxFloat16x3 ffxFract(FfxFloat16x3 x) +{ + return fract(x); +} +FfxFloat16x4 ffxFract(FfxFloat16x4 x) +{ + return fract(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxLerp(FfxFloat16 x, FfxFloat16 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x2 ffxLerp(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x2 ffxLerp(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 a) +{ + return mix(x, y, a); +} +FfxFloat16x3 ffxLerp(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 a) +{ + return mix(x, y, a); +} +FfxFloat16x3 ffxLerp(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x4 ffxLerp(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16 a) +{ + return mix(x, y, a); +} +FfxFloat16x4 ffxLerp(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 a) +{ + return mix(x, y, a); +} +//------------------------------------------------------------------------------------------------------------------------------ +// No packed version of ffxMax3. +FfxFloat16 ffxMax3Half(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + return max(x, max(y, z)); +} +FfxFloat16x2 ffxMax3Half(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + return max(x, max(y, z)); +} +FfxFloat16x3 ffxMax3Half(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + return max(x, max(y, z)); +} +FfxFloat16x4 ffxMax3Half(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + return max(x, max(y, z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +// No packed version of ffxMin3. +FfxFloat16 ffxMin3Half(FfxFloat16 x, FfxFloat16 y, FfxFloat16 z) +{ + return min(x, min(y, z)); +} +FfxFloat16x2 ffxMin3Half(FfxFloat16x2 x, FfxFloat16x2 y, FfxFloat16x2 z) +{ + return min(x, min(y, z)); +} +FfxFloat16x3 ffxMin3Half(FfxFloat16x3 x, FfxFloat16x3 y, FfxFloat16x3 z) +{ + return min(x, min(y, z)); +} +FfxFloat16x4 ffxMin3Half(FfxFloat16x4 x, FfxFloat16x4 y, FfxFloat16x4 z) +{ + return min(x, min(y, z)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxReciprocalHalf(FfxFloat16 x) +{ + return FFX_BROADCAST_FLOAT16(1.0) / x; +} +FfxFloat16x2 ffxReciprocalHalf(FfxFloat16x2 x) +{ + return FFX_BROADCAST_FLOAT16X2(1.0) / x; +} +FfxFloat16x3 ffxReciprocalHalf(FfxFloat16x3 x) +{ + return FFX_BROADCAST_FLOAT16X3(1.0) / x; +} +FfxFloat16x4 ffxReciprocalHalf(FfxFloat16x4 x) +{ + return FFX_BROADCAST_FLOAT16X4(1.0) / x; +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxReciprocalSquareRootHalf(FfxFloat16 x) +{ + return FFX_BROADCAST_FLOAT16(1.0) / sqrt(x); +} +FfxFloat16x2 ffxReciprocalSquareRootHalf(FfxFloat16x2 x) +{ + return FFX_BROADCAST_FLOAT16X2(1.0) / sqrt(x); +} +FfxFloat16x3 ffxReciprocalSquareRootHalf(FfxFloat16x3 x) +{ + return FFX_BROADCAST_FLOAT16X3(1.0) / sqrt(x); +} +FfxFloat16x4 ffxReciprocalSquareRootHalf(FfxFloat16x4 x) +{ + return FFX_BROADCAST_FLOAT16X4(1.0) / sqrt(x); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxFloat16 ffxSaturate(FfxFloat16 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16(0.0), FFX_BROADCAST_FLOAT16(1.0)); +} +FfxFloat16x2 ffxSaturate(FfxFloat16x2 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16X2(0.0), FFX_BROADCAST_FLOAT16X2(1.0)); +} +FfxFloat16x3 ffxSaturate(FfxFloat16x3 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16X3(0.0), FFX_BROADCAST_FLOAT16X3(1.0)); +} +FfxFloat16x4 ffxSaturate(FfxFloat16x4 x) +{ + return clamp(x, FFX_BROADCAST_FLOAT16X4(0.0), FFX_BROADCAST_FLOAT16X4(1.0)); +} +//------------------------------------------------------------------------------------------------------------------------------ +FfxUInt16 ffxBitShiftRightHalf(FfxUInt16 a, FfxUInt16 b) +{ + return FfxUInt16(FfxInt16(a) >> FfxInt16(b)); +} +FfxUInt16x2 ffxBitShiftRightHalf(FfxUInt16x2 a, FfxUInt16x2 b) +{ + return FfxUInt16x2(FfxInt16x2(a) >> FfxInt16x2(b)); +} +FfxUInt16x3 ffxBitShiftRightHalf(FfxUInt16x3 a, FfxUInt16x3 b) +{ + return FfxUInt16x3(FfxInt16x3(a) >> FfxInt16x3(b)); +} +FfxUInt16x4 ffxBitShiftRightHalf(FfxUInt16x4 a, FfxUInt16x4 b) +{ + return FfxUInt16x4(FfxInt16x4(a) >> FfxInt16x4(b)); +} +#endif // FFX_HALF + +#if defined(FFX_WAVE) +// Where 'x' must be a compile time literal. +FfxFloat32 AWaveXorF1(FfxFloat32 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxFloat32x2 AWaveXorF2(FfxFloat32x2 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxFloat32x3 AWaveXorF3(FfxFloat32x3 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxFloat32x4 AWaveXorF4(FfxFloat32x4 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32 AWaveXorU1(FfxUInt32 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32x2 AWaveXorU2(FfxUInt32x2 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32x3 AWaveXorU3(FfxUInt32x3 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} +FfxUInt32x4 AWaveXorU4(FfxUInt32x4 v, FfxUInt32 x) +{ + return subgroupShuffleXor(v, x); +} + +//------------------------------------------------------------------------------------------------------------------------------ +#if FFX_HALF +FfxFloat16x2 ffxWaveXorFloat16x2(FfxFloat16x2 v, FfxUInt32 x) +{ + return FFX_UINT32_TO_FLOAT16X2(subgroupShuffleXor(FFX_FLOAT16X2_TO_UINT32(v), x)); +} +FfxFloat16x4 ffxWaveXorFloat16x4(FfxFloat16x4 v, FfxUInt32 x) +{ + return FFX_UINT32X2_TO_FLOAT16X4(subgroupShuffleXor(FFX_FLOAT16X4_TO_UINT32X2(v), x)); +} +FfxUInt16x2 ffxWaveXorUint16x2(FfxUInt16x2 v, FfxUInt32 x) +{ + return FFX_UINT32_TO_UINT16X2(subgroupShuffleXor(FFX_UINT16X2_TO_UINT32(v), x)); +} +FfxUInt16x4 ffxWaveXorUint16x4(FfxUInt16x4 v, FfxUInt32 x) +{ + return FFX_UINT32X2_TO_UINT16X4(subgroupShuffleXor(FFX_UINT16X4_TO_UINT32X2(v), x)); +} +#endif // FFX_HALF +#endif // #if defined(FFX_WAVE) diff --git a/Assets/Resources/FSR2/shaders/ffx_core_glsl.h.meta b/Assets/Resources/FSR2/shaders/ffx_core_glsl.h.meta new file mode 100644 index 0000000..0e1c0e0 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_core_glsl.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 8b1e08364d8608e4ea7f4372e5b2de67 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_core_gpu_common.h b/Assets/Resources/FSR2/shaders/ffx_core_gpu_common.h similarity index 100% rename from Assets/Resources/Shaders/ffx_core_gpu_common.h rename to Assets/Resources/FSR2/shaders/ffx_core_gpu_common.h diff --git a/Assets/Resources/Shaders/ffx_core_gpu_common.h.meta b/Assets/Resources/FSR2/shaders/ffx_core_gpu_common.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_core_gpu_common.h.meta rename to Assets/Resources/FSR2/shaders/ffx_core_gpu_common.h.meta index 43bd625..26832e5 100644 --- a/Assets/Resources/Shaders/ffx_core_gpu_common.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_core_gpu_common.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 041095789ff5b314fae434ff885e389d +guid: 402c509393f5bf647b41a962a48ed8e2 PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/Shaders/ffx_core_gpu_common_half.h b/Assets/Resources/FSR2/shaders/ffx_core_gpu_common_half.h similarity index 100% rename from Assets/Resources/Shaders/ffx_core_gpu_common_half.h rename to Assets/Resources/FSR2/shaders/ffx_core_gpu_common_half.h diff --git a/Assets/Resources/Shaders/ffx_core_gpu_common_half.h.meta b/Assets/Resources/FSR2/shaders/ffx_core_gpu_common_half.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_core_gpu_common_half.h.meta rename to Assets/Resources/FSR2/shaders/ffx_core_gpu_common_half.h.meta index addeb9c..22f9f75 100644 --- a/Assets/Resources/Shaders/ffx_core_gpu_common_half.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_core_gpu_common_half.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: d16a1a4953abf534caaa4bd8843de607 +guid: 142bf3947ada43541a0f31a328fdec07 PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/Shaders/ffx_core_hlsl.h b/Assets/Resources/FSR2/shaders/ffx_core_hlsl.h similarity index 100% rename from Assets/Resources/Shaders/ffx_core_hlsl.h rename to Assets/Resources/FSR2/shaders/ffx_core_hlsl.h diff --git a/Assets/Resources/Shaders/ffx_core_hlsl.h.meta b/Assets/Resources/FSR2/shaders/ffx_core_hlsl.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_core_hlsl.h.meta rename to Assets/Resources/FSR2/shaders/ffx_core_hlsl.h.meta index 240c8d2..996147a 100644 --- a/Assets/Resources/Shaders/ffx_core_hlsl.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_core_hlsl.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: d9d43e208984ec94facb71623986b1cf +guid: 89d6e02f97594f64ca2da4c8124df6cf PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/Shaders/ffx_core_portability.h b/Assets/Resources/FSR2/shaders/ffx_core_portability.h similarity index 100% rename from Assets/Resources/Shaders/ffx_core_portability.h rename to Assets/Resources/FSR2/shaders/ffx_core_portability.h diff --git a/Assets/Resources/Shaders/ffx_core_portability.h.meta b/Assets/Resources/FSR2/shaders/ffx_core_portability.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_core_portability.h.meta rename to Assets/Resources/FSR2/shaders/ffx_core_portability.h.meta index 85ea2c5..7cc75c1 100644 --- a/Assets/Resources/Shaders/ffx_core_portability.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_core_portability.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 99bc6885ecf87554aae6119a96eaaf6c +guid: 5d6e692075988194382122bac7819f02 PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/Shaders/ffx_fsr1.h b/Assets/Resources/FSR2/shaders/ffx_fsr1.h similarity index 100% rename from Assets/Resources/Shaders/ffx_fsr1.h rename to Assets/Resources/FSR2/shaders/ffx_fsr1.h diff --git a/Assets/Resources/Shaders/ffx_fsr1.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr1.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_fsr1.h.meta rename to Assets/Resources/FSR2/shaders/ffx_fsr1.h.meta index b675190..9691905 100644 --- a/Assets/Resources/Shaders/ffx_fsr1.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_fsr1.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 7d1bd748bf4775847ac7cec07191796a +guid: 7524e42f73f97f34bbeb414ea412a808 PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate.h new file mode 100644 index 0000000..d0c5eae --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate.h @@ -0,0 +1,263 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_ACCUMULATE_H +#define FFX_FSR2_ACCUMULATE_H + +#define FFX_FSR2_OPTION_GUARANTEE_UPSAMPLE_WEIGHT_ON_NEW_SAMPLES 1 + +FfxFloat32 GetPxHrVelocity(FfxFloat32x2 fMotionVector) +{ + return length(fMotionVector * DisplaySize()); +} +#if FFX_HALF +FFX_MIN16_F GetPxHrVelocity(FFX_MIN16_F2 fMotionVector) +{ + return length(fMotionVector * FFX_MIN16_F2(DisplaySize())); +} +#endif + +void Accumulate(FfxInt32x2 iPxHrPos, FFX_PARAMETER_INOUT FfxFloat32x4 fHistory, FFX_PARAMETER_IN FfxFloat32x4 fUpsampled, FFX_PARAMETER_IN FfxFloat32 fDepthClipFactor, FFX_PARAMETER_IN FfxFloat32 fHrVelocity) +{ + fHistory.w = fHistory.w + fUpsampled.w; + + fUpsampled.rgb = YCoCgToRGB(fUpsampled.rgb); + + const FfxFloat32 fAlpha = fUpsampled.w / fHistory.w; + fHistory.rgb = ffxLerp(fHistory.rgb, fUpsampled.rgb, fAlpha); + + FfxFloat32 fMaxAverageWeight = FfxFloat32(ffxLerp(MaxAccumulationWeight(), accumulationMaxOnMotion, ffxSaturate(fHrVelocity * 10.0f))); + fHistory.w = ffxMin(fHistory.w, fMaxAverageWeight); +} + +void RectifyHistory( + RectificationBoxData clippingBox, + inout FfxFloat32x4 fHistory, + FFX_PARAMETER_IN FfxFloat32x3 fLockStatus, + FFX_PARAMETER_IN FfxFloat32 fDepthClipFactor, + FFX_PARAMETER_IN FfxFloat32 fLumaStabilityFactor, + FFX_PARAMETER_IN FfxFloat32 fLuminanceDiff, + FFX_PARAMETER_IN FfxFloat32 fUpsampleWeight, + FFX_PARAMETER_IN FfxFloat32 fLockContributionThisFrame) +{ + FfxFloat32 fScaleFactorInfluence = FfxFloat32(1.0f / DownscaleFactor().x - 1); + FfxFloat32 fBoxScale = FfxFloat32(1.0f) + (FfxFloat32(0.5f) * fScaleFactorInfluence); + + FfxFloat32x3 fScaledBoxVec = clippingBox.boxVec * fBoxScale; + FfxFloat32x3 boxMin = clippingBox.boxCenter - fScaledBoxVec; + FfxFloat32x3 boxMax = clippingBox.boxCenter + fScaledBoxVec; + FfxFloat32x3 boxCenter = clippingBox.boxCenter; + FfxFloat32 boxVecSize = length(clippingBox.boxVec); + + boxMin = ffxMax(clippingBox.aabbMin, boxMin); + boxMax = ffxMin(clippingBox.aabbMax, boxMax); + + FfxFloat32x3 distToClampOutside = ffxMax(ffxMax(FfxFloat32x3(0, 0, 0), boxMin - fHistory.xyz), ffxMax(FfxFloat32x3(0, 0, 0), fHistory.xyz - boxMax)); + + if (any(FFX_GREATER_THAN(distToClampOutside, FfxFloat32x3(0, 0, 0)))) { + + const FfxFloat32x3 clampedHistorySample = clamp(fHistory.xyz, boxMin, boxMax); + + FfxFloat32x3 clippedHistoryToBoxCenter = abs(clampedHistorySample - boxCenter); + FfxFloat32x3 historyToBoxCenter = abs(fHistory.xyz - boxCenter); + FfxFloat32x3 HistoryColorWeight; + HistoryColorWeight.x = historyToBoxCenter.x > FfxFloat32(0) ? clippedHistoryToBoxCenter.x / historyToBoxCenter.x : FfxFloat32(0.0f); + HistoryColorWeight.y = historyToBoxCenter.y > FfxFloat32(0) ? clippedHistoryToBoxCenter.y / historyToBoxCenter.y : FfxFloat32(0.0f); + HistoryColorWeight.z = historyToBoxCenter.z > FfxFloat32(0) ? clippedHistoryToBoxCenter.z / historyToBoxCenter.z : FfxFloat32(0.0f); + + FfxFloat32x3 fHistoryContribution = HistoryColorWeight; + + // only lock luma + fHistoryContribution += ffxMax(fLockContributionThisFrame, fLumaStabilityFactor).xxx; + fHistoryContribution *= (fDepthClipFactor * fDepthClipFactor); + + fHistory.xyz = ffxLerp(clampedHistorySample.xyz, fHistory.xyz, ffxSaturate(fHistoryContribution)); + } +} + +void WriteUpscaledOutput(FfxInt32x2 iPxHrPos, FfxFloat32x3 fUpscaledColor) +{ + StoreUpscaledOutput(iPxHrPos, fUpscaledColor); +} + +FfxFloat32 GetLumaStabilityFactor(FfxFloat32x2 fHrUv, FfxFloat32 fHrVelocity) +{ + FfxFloat32 fLumaStabilityFactor = SampleLumaStabilityFactor(fHrUv); + + // Only apply on still, have to reproject luma history resource if we want it to work on motion + fLumaStabilityFactor *= FfxFloat32(fHrVelocity < 0.1f); + + return fLumaStabilityFactor; +} + +FfxFloat32 GetLockContributionThisFrame(FfxFloat32x2 fUvCoord, FfxFloat32 fAccumulationMask, FfxFloat32 fParticleMask, FfxFloat32x3 fLockStatus) +{ + const FfxFloat32 fNormalizedLockLifetime = GetNormalizedRemainingLockLifetime(fLockStatus); + + // Rectify on lock frame + FfxFloat32 fLockContributionThisFrame = ffxSaturate(fNormalizedLockLifetime * FfxFloat32(4)); + + return fLockContributionThisFrame; +} + +void FinalizeLockStatus(FfxInt32x2 iPxHrPos, FfxFloat32x3 fLockStatus, FfxFloat32 fUpsampledWeight) +{ + // Increase trust + const FfxFloat32 fTrustIncreaseLanczosMax = FfxFloat32(12); // same increase no matter the MaxAccumulationWeight() value. + const FfxFloat32 fTrustIncrease = FfxFloat32(fUpsampledWeight / fTrustIncreaseLanczosMax); + fLockStatus[LOCK_TRUST] = ffxMin(FfxFloat32(1), fLockStatus[LOCK_TRUST] + fTrustIncrease); + + // Decrease lock lifetime + const FfxFloat32 fLifetimeDecreaseLanczosMax = FfxFloat32(JitterSequenceLength()) * FfxFloat32(averageLanczosWeightPerFrame); + const FfxFloat32 fLifetimeDecrease = FfxFloat32(fUpsampledWeight / fLifetimeDecreaseLanczosMax); + fLockStatus[LOCK_LIFETIME_REMAINING] = ffxMax(FfxFloat32(0), fLockStatus[LOCK_LIFETIME_REMAINING] - fLifetimeDecrease); + + StoreLockStatus(iPxHrPos, fLockStatus); +} + +FfxFloat32 ComputeMaxAccumulationWeight(FfxFloat32 fHrVelocity, FfxFloat32 fReactiveMax, FfxFloat32 fDepthClipFactor, FfxFloat32 fLuminanceDiff, LockState lockState) { + + FfxFloat32 normalizedMinimum = FfxFloat32(accumulationMaxOnMotion) / FfxFloat32(MaxAccumulationWeight()); + + FfxFloat32 fReactiveMaxAccumulationWeight = FfxFloat32(1) - fReactiveMax; + FfxFloat32 fMotionMaxAccumulationWeight = ffxLerp(FfxFloat32(1), normalizedMinimum, ffxSaturate(fHrVelocity * FfxFloat32(10))); + FfxFloat32 fDepthClipMaxAccumulationWeight = fDepthClipFactor; + + FfxFloat32 fLuminanceDiffMaxAccumulationWeight = ffxSaturate(ffxMax(normalizedMinimum, FfxFloat32(1) - fLuminanceDiff)); + + FfxFloat32 maxAccumulation = FfxFloat32(MaxAccumulationWeight()) * ffxMin( + ffxMin(fReactiveMaxAccumulationWeight, fMotionMaxAccumulationWeight), + ffxMin(fDepthClipMaxAccumulationWeight, fLuminanceDiffMaxAccumulationWeight) + ); + + return (lockState.NewLock && !lockState.WasLockedPrevFrame) ? FfxFloat32(accumulationMaxOnMotion) : maxAccumulation; +} + +FfxFloat32x2 ComputeKernelWeight(in FfxFloat32 fHistoryWeight, in FfxFloat32 fDepthClipFactor, in FfxFloat32 fReactivityFactor) { + FfxFloat32 fKernelSizeBias = ffxSaturate(ffxMax(FfxFloat32(0), fHistoryWeight - FfxFloat32(0.5)) / FfxFloat32(3)); + + FfxFloat32 fOneMinusReactiveMax = FfxFloat32(1) - fReactivityFactor; + FfxFloat32x2 fKernelWeight = FfxFloat32(1) + (FfxFloat32(1.0f) / FfxFloat32x2(DownscaleFactor()) - FfxFloat32(1)) * FfxFloat32(fKernelSizeBias) * fOneMinusReactiveMax; + + //average value on disocclusion, to help decrease high value sample importance wait for accumulation to kick in + fKernelWeight *= FfxFloat32x2(0.5f, 0.5f) + fDepthClipFactor * FfxFloat32x2(0.5f, 0.5f); + + return ffxMin(FfxFloat32x2(1.99f, 1.99f), fKernelWeight); +} + +void Accumulate(FfxInt32x2 iPxHrPos) +{ + const FfxFloat32x2 fSamplePosHr = iPxHrPos + 0.5f; + const FfxFloat32x2 fPxLrPos = fSamplePosHr * DownscaleFactor(); // Source resolution output pixel center position + const FfxInt32x2 iPxLrPos = FfxInt32x2(floor(fPxLrPos)); // TODO: what about weird upscale factors... + + const FfxFloat32x2 fSamplePosUnjitterLr = (FfxFloat32x2(iPxLrPos) + FfxFloat32x2(0.5f, 0.5f)) - Jitter(); // This is the un-jittered position of the sample at offset 0,0 + + const FfxFloat32x2 fLrUvJittered = (fPxLrPos + Jitter()) / RenderSize(); + + const FfxFloat32x2 fHrUv = (iPxHrPos + 0.5f) / DisplaySize(); + const FfxFloat32x2 fMotionVector = GetMotionVector(iPxHrPos, fHrUv); + + const FfxFloat32 fHrVelocity = GetPxHrVelocity(fMotionVector); + const FfxFloat32 fDepthClipFactor = ffxSaturate(SampleDepthClip(fLrUvJittered)); + const FfxFloat32 fLumaStabilityFactor = GetLumaStabilityFactor(fHrUv, fHrVelocity); + const FfxFloat32x2 fDilatedReactiveMasks = SampleDilatedReactiveMasks(fLrUvJittered); + const FfxFloat32 fReactiveMax = fDilatedReactiveMasks.x; + const FfxFloat32 fAccumulationMask = fDilatedReactiveMasks.y; + const FfxBoolean bIsResetFrame = (0 == FrameIndex()); + + FfxFloat32x4 fHistoryColorAndWeight = FfxFloat32x4(0, 0, 0, 0); + FfxFloat32x3 fLockStatus; + InitializeNewLockSample(fLockStatus); + FfxBoolean bIsExistingSample = FFX_TRUE; + + FfxFloat32x2 fReprojectedHrUv = FfxFloat32x2(0, 0); + ComputeReprojectedUVs(iPxHrPos, fMotionVector, fReprojectedHrUv, bIsExistingSample); + + if (bIsExistingSample && !bIsResetFrame) { + ReprojectHistoryColor(iPxHrPos, fReprojectedHrUv, fHistoryColorAndWeight); + ReprojectHistoryLockStatus(iPxHrPos, fReprojectedHrUv, fLockStatus); + } + + FfxFloat32 fLuminanceDiff = FfxFloat32(0.0f); + + LockState lockState = PostProcessLockStatus(iPxHrPos, fLrUvJittered, FfxFloat32(fDepthClipFactor), fAccumulationMask, fHrVelocity, fHistoryColorAndWeight.w, fLockStatus, fLuminanceDiff); + + fHistoryColorAndWeight.w = ffxMin(fHistoryColorAndWeight.w, ComputeMaxAccumulationWeight( + FfxFloat32(fHrVelocity), fReactiveMax, FfxFloat32(fDepthClipFactor), FfxFloat32(fLuminanceDiff), lockState + )); + + const FfxFloat32 fNormalizedLockLifetime = GetNormalizedRemainingLockLifetime(fLockStatus); + + // Kill accumulation based on shading change + fHistoryColorAndWeight.w = ffxMin(fHistoryColorAndWeight.w, FfxFloat32(ffxMax(0.0f, MaxAccumulationWeight() * ffxPow(FfxFloat32(1) - fLuminanceDiff, 2.0f / 1.0f)))); + + // Load upsampled input color + RectificationBoxData clippingBox; + + FfxFloat32 fKernelBias = fAccumulationMask * ffxSaturate(ffxMax(0.0f, fHistoryColorAndWeight.w - 0.5f) / 3.0f); + + FfxFloat32 fReactiveWeighted = 0; + + // No trust in reactive areas + fLockStatus[LOCK_TRUST] = ffxMin(fLockStatus[LOCK_TRUST], FfxFloat32(1.0f) - FfxFloat32(pow(fReactiveMax, 1.0f / 3.0f))); + fLockStatus[LOCK_TRUST] = ffxMin(fLockStatus[LOCK_TRUST], FfxFloat32(fDepthClipFactor)); + + FfxFloat32x2 fKernelWeight = ComputeKernelWeight(fHistoryColorAndWeight.w, FfxFloat32(fDepthClipFactor), ffxMax((FfxFloat32(1) - fLockStatus[LOCK_TRUST]), fReactiveMax)); + + FfxFloat32x4 fUpsampledColorAndWeight = ComputeUpsampledColorAndWeight(iPxHrPos, fKernelWeight, clippingBox); + +#if FFX_FSR2_OPTION_GUARANTEE_UPSAMPLE_WEIGHT_ON_NEW_SAMPLES + // Make sure all samples have same weight on reset/first frame. Upsampled weight should never be 0.0f when history accumulation is 0.0f. + fUpsampledColorAndWeight.w = (fHistoryColorAndWeight.w == 0.0f) ? ffxMax(FSR2_EPSILON, fUpsampledColorAndWeight.w) : fUpsampledColorAndWeight.w; +#endif + + FfxFloat32 fLockContributionThisFrame = GetLockContributionThisFrame(fHrUv, fAccumulationMask, fReactiveMax, fLockStatus); + + // Update accumulation and rectify history + if (fHistoryColorAndWeight.w > FfxFloat32(0)) { + + RectifyHistory(clippingBox, fHistoryColorAndWeight, fLockStatus, FfxFloat32(fDepthClipFactor), FfxFloat32(fLumaStabilityFactor), FfxFloat32(fLuminanceDiff), fUpsampledColorAndWeight.w, fLockContributionThisFrame); + + fHistoryColorAndWeight.rgb = YCoCgToRGB(fHistoryColorAndWeight.rgb); + } + + Accumulate(iPxHrPos, fHistoryColorAndWeight, fUpsampledColorAndWeight, fDepthClipFactor, fHrVelocity); + + //Subtract accumulation weight in reactive areas + fHistoryColorAndWeight.w -= fUpsampledColorAndWeight.w * fReactiveMax; + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + fHistoryColorAndWeight.rgb = InverseTonemap(fHistoryColorAndWeight.rgb); +#endif + fHistoryColorAndWeight.rgb /= FfxFloat32(Exposure()); + + FinalizeLockStatus(iPxHrPos, fLockStatus, fUpsampledColorAndWeight.w); + + StoreInternalColorAndWeight(iPxHrPos, fHistoryColorAndWeight); + + // Output final color when RCAS is disabled +#if FFX_FSR2_OPTION_APPLY_SHARPENING == 0 + WriteUpscaledOutput(iPxHrPos, fHistoryColorAndWeight.rgb); +#endif +} + +#endif // FFX_FSR2_ACCUMULATE_H diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate.h.meta new file mode 100644 index 0000000..203a52f --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: a04cb2522aaff1045869a272ed129964 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.glsl new file mode 100644 index 0000000..e1ee116 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.glsl @@ -0,0 +1,99 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 5 +// SRV 4 : FSR2_Exposure : r_exposure +// SRV 6 : m_UpscaleTransparencyAndComposition : r_transparency_and_composition_mask +// SRV 8 : FSR2_DilatedVelocity : r_dilated_motion_vectors +// SRV 10 : FSR2_InternalUpscaled2 : r_internal_upscaled_color +// SRV 11 : FSR2_LockStatus2 : r_lock_status +// SRV 12 : FSR2_DepthClip : r_depth_clip +// SRV 13 : FSR2_PreparedInputColor : r_prepared_input_color +// SRV 14 : FSR2_LumaHistory : r_luma_history +// SRV 16 : FSR2_LanczosLutData : r_lanczos_lut +// SRV 26 : FSR2_MaximumUpsampleBias : r_upsample_maximum_bias_lut +// SRV 27 : FSR2_ReactiveMaskMax : r_reactive_max +// SRV 28 : FSR2_ExposureMips : r_imgMips +// UAV 10 : FSR2_InternalUpscaled1 : rw_internal_upscaled_color +// UAV 11 : FSR2_LockStatus1 : rw_lock_status +// UAV 18 : DisplayOutput : rw_upscaled_output +// CB 0 : cbFSR2 +// CB 1 : FSR2DispatchOffsets + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_EXPOSURE 0 +#define FSR2_BIND_SRV_DILATED_REACTIVE_MASKS 1 +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS +#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 2 +#else +#define FSR2_BIND_SRV_MOTION_VECTORS 2 +#endif +#define FSR2_BIND_SRV_INTERNAL_UPSCALED 3 +#define FSR2_BIND_SRV_LOCK_STATUS 4 +#define FSR2_BIND_SRV_DEPTH_CLIP 5 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 6 +#define FSR2_BIND_SRV_LUMA_HISTORY 7 +#define FSR2_BIND_SRV_LANCZOS_LUT 8 +#define FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT 9 +#define FSR2_BIND_SRV_EXPOSURE_MIPS 10 +#define FSR2_BIND_UAV_INTERNAL_UPSCALED 11 +#define FSR2_BIND_UAV_LOCK_STATUS 12 +#define FSR2_BIND_UAV_UPSCALED_OUTPUT 13 + +#define FSR2_BIND_CB_FSR2 14 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_upsample.h" +#include "ffx_fsr2_postprocess_lock_status.h" +#include "ffx_fsr2_reproject.h" +#include "ffx_fsr2_accumulate.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; + +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + uvec2 uGroupId = gl_WorkGroupID.xy; + const uint GroupRows = (uint(DisplaySize().y) + FFX_FSR2_THREAD_GROUP_HEIGHT - 1) / FFX_FSR2_THREAD_GROUP_HEIGHT; + uGroupId.y = GroupRows - uGroupId.y - 1; + + uvec2 uDispatchThreadId = uGroupId * uvec2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + gl_LocalInvocationID.xy; + + Accumulate(ivec2(uDispatchThreadId)); +} \ No newline at end of file diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.glsl.meta new file mode 100644 index 0000000..df1b03a --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: a47992654ac46784d976db0091b79aeb +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.hlsl new file mode 100644 index 0000000..4321f99 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.hlsl @@ -0,0 +1,93 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 5 +// SRV 4 : FSR2_Exposure : r_exposure +// SRV 6 : m_UpscaleTransparencyAndComposition : r_transparency_and_composition_mask +// SRV 8 : FSR2_DilatedVelocity : r_dilated_motion_vectors +// SRV 10 : FSR2_InternalUpscaled2 : r_internal_upscaled_color +// SRV 11 : FSR2_LockStatus2 : r_lock_status +// SRV 12 : FSR2_DepthClip : r_depth_clip +// SRV 13 : FSR2_PreparedInputColor : r_prepared_input_color +// SRV 14 : FSR2_LumaHistory : r_luma_history +// SRV 16 : FSR2_LanczosLutData : r_lanczos_lut +// SRV 26 : FSR2_MaximumUpsampleBias : r_upsample_maximum_bias_lut +// SRV 27 : FSR2_DilatedReactiveMasks : r_dilated_reactive_masks +// SRV 28 : FSR2_ExposureMips : r_imgMips +// UAV 10 : FSR2_InternalUpscaled1 : rw_internal_upscaled_color +// UAV 11 : FSR2_LockStatus1 : rw_lock_status +// UAV 18 : DisplayOutput : rw_upscaled_output +// CB 0 : cbFSR2 +// CB 1 : FSR2DispatchOffsets + +#define FSR2_BIND_SRV_EXPOSURE 0 +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS +#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 2 +#else +#define FSR2_BIND_SRV_MOTION_VECTORS 2 +#endif +#define FSR2_BIND_SRV_INTERNAL_UPSCALED 3 +#define FSR2_BIND_SRV_LOCK_STATUS 4 +#define FSR2_BIND_SRV_DEPTH_CLIP 5 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 6 +#define FSR2_BIND_SRV_LUMA_HISTORY 7 +#define FSR2_BIND_SRV_LANCZOS_LUT 8 +#define FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT 9 +#define FSR2_BIND_SRV_DILATED_REACTIVE_MASKS 10 +#define FSR2_BIND_SRV_EXPOSURE_MIPS 11 +#define FSR2_BIND_UAV_INTERNAL_UPSCALED 0 +#define FSR2_BIND_UAV_LOCK_STATUS 1 +#define FSR2_BIND_UAV_UPSCALED_OUTPUT 2 + +#define FSR2_BIND_CB_FSR2 0 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_upsample.h" +#include "ffx_fsr2_postprocess_lock_status.h" +#include "ffx_fsr2_reproject.h" +#include "ffx_fsr2_accumulate.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_ROOTSIG_CONTENT +void CS(uint2 uGroupId : SV_GroupID, uint2 uGroupThreadId : SV_GroupThreadID) +{ + const uint GroupRows = (uint(DisplaySize().y) + FFX_FSR2_THREAD_GROUP_HEIGHT - 1) / FFX_FSR2_THREAD_GROUP_HEIGHT; + uGroupId.y = GroupRows - uGroupId.y - 1; + + uint2 uDispatchThreadId = uGroupId * uint2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + uGroupThreadId; + + Accumulate(uDispatchThreadId); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.hlsl.meta new file mode 100644 index 0000000..cad5022 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_accumulate_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 356ec46d3f01672428b5a7a0de727548 +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.glsl new file mode 100644 index 0000000..b509eb0 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.glsl @@ -0,0 +1,91 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_PRE_ALPHA_COLOR 0 +#define FSR2_BIND_SRV_POST_ALPHA_COLOR 1 +#define FSR2_BIND_UAV_REACTIVE 2 +#define FSR2_BIND_CB_REACTIVE 3 +#define FSR2_BIND_CB_FSR2 4 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +layout (set = 1, binding = FSR2_BIND_SRV_PRE_ALPHA_COLOR) uniform texture2D r_input_color_pre_alpha; +layout (set = 1, binding = FSR2_BIND_SRV_POST_ALPHA_COLOR) uniform texture2D r_input_color_post_alpha; +layout (set = 1, binding = FSR2_BIND_UAV_REACTIVE, r8) uniform image2D rw_output_reactive_mask; + + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +layout (set = 1, binding = FSR2_BIND_CB_REACTIVE, std140) uniform cbGenerateReactive_t +{ + float scale; + float threshold; + float binaryValue; + uint flags; +} cbGenerateReactive; + +FFX_FSR2_NUM_THREADS +void main() +{ + FfxUInt32x2 uDispatchThreadId = gl_GlobalInvocationID.xy; + + FfxFloat32x3 ColorPreAlpha = texelFetch(r_input_color_pre_alpha, FfxInt32x2(uDispatchThreadId), 0).rgb; + FfxFloat32x3 ColorPostAlpha = texelFetch(r_input_color_post_alpha, FfxInt32x2(uDispatchThreadId), 0).rgb; + + if ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_TONEMAP) != 0) + { + ColorPreAlpha = Tonemap(ColorPreAlpha); + ColorPostAlpha = Tonemap(ColorPostAlpha); + } + + if ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_INVERSETONEMAP) != 0) + { + ColorPreAlpha = InverseTonemap(ColorPreAlpha); + ColorPostAlpha = InverseTonemap(ColorPostAlpha); + } + + FfxFloat32 out_reactive_value = 0.f; + FfxFloat32x3 delta = abs(ColorPostAlpha - ColorPreAlpha); + + out_reactive_value = ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_USE_COMPONENTS_MAX)!=0) ? max(delta.x, max(delta.y, delta.z)) : length(delta); + out_reactive_value *= cbGenerateReactive.scale; + + out_reactive_value = ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_THRESHOLD)!=0) ? ((out_reactive_value < cbGenerateReactive.threshold) ? 0 : cbGenerateReactive.binaryValue) : out_reactive_value; + + imageStore(rw_output_reactive_mask, FfxInt32x2(uDispatchThreadId), vec4(out_reactive_value)); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.glsl.meta new file mode 100644 index 0000000..47aaa1f --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 9b3697948343bfb42ac4fbc18d2fb8be +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.hlsl new file mode 100644 index 0000000..903ceae --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.hlsl @@ -0,0 +1,85 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#define FSR2_BIND_SRV_PRE_ALPHA_COLOR 0 +#define FSR2_BIND_SRV_POST_ALPHA_COLOR 1 +#define FSR2_BIND_UAV_REACTIVE 0 +#define FSR2_BIND_CB_FSR2 0 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" + +Texture2D r_input_color_pre_alpha : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_PRE_ALPHA_COLOR); +Texture2D r_input_color_post_alpha : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_POST_ALPHA_COLOR); +RWTexture2D rw_output_reactive_mask : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_REACTIVE); + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +cbuffer cbGenerateReactive : register(b0) +{ + float scale; + float threshold; + float binaryValue; + uint flags; +}; + +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_ROOTSIG_CONTENT +void CS(uint2 uGroupId : SV_GroupID, uint2 uGroupThreadId : SV_GroupThreadID) +{ + uint2 uDispatchThreadId = uGroupId * uint2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + uGroupThreadId; + + float3 ColorPreAlpha = r_input_color_pre_alpha[uDispatchThreadId].rgb; + float3 ColorPostAlpha = r_input_color_post_alpha[uDispatchThreadId].rgb; + + if (flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_TONEMAP) + { + ColorPreAlpha = Tonemap(ColorPreAlpha); + ColorPostAlpha = Tonemap(ColorPostAlpha); + } + + if (flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_INVERSETONEMAP) + { + ColorPreAlpha = InverseTonemap(ColorPreAlpha); + ColorPostAlpha = InverseTonemap(ColorPostAlpha); + } + + float out_reactive_value = 0.f; + float3 delta = abs(ColorPostAlpha - ColorPreAlpha); + + out_reactive_value = (flags & FFX_FSR2_AUTOREACTIVEFLAGS_USE_COMPONENTS_MAX) ? max(delta.x, max(delta.y, delta.z)) : length(delta); + out_reactive_value *= scale; + + out_reactive_value = (flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_THRESHOLD) ? (out_reactive_value < threshold ? 0 : binaryValue) : out_reactive_value; + + rw_output_reactive_mask[uDispatchThreadId] = out_reactive_value; +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.hlsl.meta new file mode 100644 index 0000000..2e93435 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_autogen_reactive_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: cc76bd6f46792f3418a56b79eb5c959b +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_glsl.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_glsl.h new file mode 100644 index 0000000..d598250 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_glsl.h @@ -0,0 +1,695 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +#include "ffx_fsr2_resources.h" + +#if defined(FFX_GPU) +#include "ffx_core.h" +#endif // #if defined(FFX_GPU) + +#if defined(FFX_GPU) +#ifndef FFX_FSR2_PREFER_WAVE64 +#define FFX_FSR2_PREFER_WAVE64 +#endif // #if defined(FFX_GPU) + +#if defined(FSR2_BIND_CB_FSR2) + layout (set = 1, binding = FSR2_BIND_CB_FSR2, std140) uniform cbFSR2_t + { + FfxInt32x2 iRenderSize; + FfxInt32x2 iDisplaySize; + FfxInt32x2 uLumaMipDimensions; + FfxInt32 uLumaMipLevelToUse; + FfxInt32 uFrameIndex; + FfxFloat32x2 fDisplaySizeRcp; + FfxFloat32x2 fJitter; + FfxFloat32x4 fDeviceToViewDepth; + FfxFloat32x2 depthclip_uv_scale; + FfxFloat32x2 postprocessed_lockstatus_uv_scale; + FfxFloat32x2 reactive_mask_dim_rcp; + FfxFloat32x2 MotionVectorScale; + FfxFloat32x2 fDownscaleFactor; + FfxFloat32 fPreExposure; + FfxFloat32 fTanHalfFOV; + FfxFloat32x2 fMotionVectorJitterCancellation; + FfxFloat32 fJitterSequenceLength; + FfxFloat32 fLockInitialLifetime; + FfxFloat32 fLockTickDelta; + FfxFloat32 fDeltaTime; + FfxFloat32 fDynamicResChangeFactor; + FfxFloat32 fLumaMipRcp; + } cbFSR2; +#endif + +FfxFloat32 LumaMipRcp() +{ + return cbFSR2.fLumaMipRcp; +} + +FfxInt32x2 LumaMipDimensions() +{ + return cbFSR2.uLumaMipDimensions; +} + +FfxInt32 LumaMipLevelToUse() +{ + return cbFSR2.uLumaMipLevelToUse; +} + +FfxFloat32x2 DownscaleFactor() +{ + return cbFSR2.fDownscaleFactor; +} + +FfxFloat32x2 Jitter() +{ + return cbFSR2.fJitter; +} + +FfxFloat32x2 MotionVectorJitterCancellation() +{ + return cbFSR2.fMotionVectorJitterCancellation; +} + +FfxInt32x2 RenderSize() +{ + return cbFSR2.iRenderSize; +} + +FfxInt32x2 DisplaySize() +{ + return cbFSR2.iDisplaySize; +} + +FfxFloat32x2 DisplaySizeRcp() +{ + return cbFSR2.fDisplaySizeRcp; +} + +FfxFloat32 JitterSequenceLength() +{ + return cbFSR2.fJitterSequenceLength; +} + +FfxFloat32 LockInitialLifetime() +{ + return cbFSR2.fLockInitialLifetime; +} + +FfxFloat32 LockTickDelta() +{ + return cbFSR2.fLockTickDelta; +} + +FfxFloat32 DeltaTime() +{ + return cbFSR2.fDeltaTime; +} + +FfxFloat32 MaxAccumulationWeight() +{ + const FfxFloat32 averageLanczosWeightPerFrame = 0.74f; // Average lanczos weight for jitter accumulated samples + + return 12; //32.0f * averageLanczosWeightPerFrame; +} + +FfxFloat32 DynamicResChangeFactor() +{ + return cbFSR2.fDynamicResChangeFactor; +} + +FfxInt32 FrameIndex() +{ + return cbFSR2.uFrameIndex; +} + +layout (set = 0, binding = 0) uniform sampler s_PointClamp; +layout (set = 0, binding = 1) uniform sampler s_LinearClamp; + +// SRVs +#if defined(FSR2_BIND_SRV_INPUT_COLOR) + layout (set = 1, binding = FSR2_BIND_SRV_INPUT_COLOR) uniform texture2D r_input_color_jittered; +#endif +#if defined(FSR2_BIND_SRV_MOTION_VECTORS) + layout (set = 1, binding = FSR2_BIND_SRV_MOTION_VECTORS) uniform texture2D r_motion_vectors; +#endif +#if defined(FSR2_BIND_SRV_DEPTH) + layout (set = 1, binding = FSR2_BIND_SRV_DEPTH) uniform texture2D r_depth; +#endif +#if defined(FSR2_BIND_SRV_EXPOSURE) + layout (set = 1, binding = FSR2_BIND_SRV_EXPOSURE) uniform texture2D r_exposure; +#endif +#if defined(FSR2_BIND_SRV_REACTIVE_MASK) + layout (set = 1, binding = FSR2_BIND_SRV_REACTIVE_MASK) uniform texture2D r_reactive_mask; +#endif +#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) + layout (set = 1, binding = FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) uniform texture2D r_transparency_and_composition_mask; +#endif +#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) + layout (set = 1, binding = FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) uniform utexture2D r_reconstructed_previous_nearest_depth; +#endif +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) + layout (set = 1, binding = FSR2_BIND_SRV_DILATED_MOTION_VECTORS) uniform texture2D r_dilated_motion_vectors; +#endif +#if defined(FSR2_BIND_SRV_DILATED_DEPTH) + layout (set = 1, binding = FSR2_BIND_SRV_DILATED_DEPTH) uniform texture2D r_dilatedDepth; +#endif +#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED) + layout (set = 1, binding = FSR2_BIND_SRV_INTERNAL_UPSCALED) uniform texture2D r_internal_upscaled_color; +#endif +#if defined(FSR2_BIND_SRV_LOCK_STATUS) + layout (set = 1, binding = FSR2_BIND_SRV_LOCK_STATUS) uniform texture2D r_lock_status; +#endif +#if defined(FSR2_BIND_SRV_DEPTH_CLIP) + layout (set = 1, binding = FSR2_BIND_SRV_DEPTH_CLIP) uniform texture2D r_depth_clip; +#endif +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) + layout (set = 1, binding = FSR2_BIND_SRV_PREPARED_INPUT_COLOR) uniform texture2D r_prepared_input_color; +#endif +#if defined(FSR2_BIND_SRV_LUMA_HISTORY) + layout (set = 1, binding = FSR2_BIND_SRV_LUMA_HISTORY) uniform texture2D r_luma_history; +#endif +#if defined(FSR2_BIND_SRV_RCAS_INPUT) + layout (set = 1, binding = FSR2_BIND_SRV_RCAS_INPUT) uniform texture2D r_rcas_input; +#endif +#if defined(FSR2_BIND_SRV_LANCZOS_LUT) + layout (set = 1, binding = FSR2_BIND_SRV_LANCZOS_LUT) uniform texture2D r_lanczos_lut; +#endif +#if defined(FSR2_BIND_SRV_EXPOSURE_MIPS) + layout (set = 1, binding = FSR2_BIND_SRV_EXPOSURE_MIPS) uniform texture2D r_imgMips; +#endif +#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) + layout (set = 1, binding = FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) uniform texture2D r_upsample_maximum_bias_lut; +#endif +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) + layout (set = 1, binding = FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) uniform texture2D r_dilated_reactive_masks; +#endif + +// UAV +#if defined FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH + layout (set = 1, binding = FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH, r32ui) uniform uimage2D rw_reconstructed_previous_nearest_depth; +#endif +#if defined FSR2_BIND_UAV_DILATED_MOTION_VECTORS + layout (set = 1, binding = FSR2_BIND_UAV_DILATED_MOTION_VECTORS, rg32f) uniform image2D rw_dilated_motion_vectors; +#endif +#if defined FSR2_BIND_UAV_DILATED_DEPTH + layout (set = 1, binding = FSR2_BIND_UAV_DILATED_DEPTH, r32f) uniform image2D rw_dilatedDepth; +#endif +#if defined FSR2_BIND_UAV_INTERNAL_UPSCALED + layout (set = 1, binding = FSR2_BIND_UAV_INTERNAL_UPSCALED, rgba32f) uniform image2D rw_internal_upscaled_color; +#endif +#if defined FSR2_BIND_UAV_LOCK_STATUS + layout (set = 1, binding = FSR2_BIND_UAV_LOCK_STATUS, r11f_g11f_b10f) uniform image2D rw_lock_status; +#endif +#if defined FSR2_BIND_UAV_DEPTH_CLIP + layout (set = 1, binding = FSR2_BIND_UAV_DEPTH_CLIP, r32f) uniform image2D rw_depth_clip; +#endif +#if defined FSR2_BIND_UAV_PREPARED_INPUT_COLOR + layout (set = 1, binding = FSR2_BIND_UAV_PREPARED_INPUT_COLOR, rgba16) uniform image2D rw_prepared_input_color; +#endif +#if defined FSR2_BIND_UAV_LUMA_HISTORY + layout (set = 1, binding = FSR2_BIND_UAV_LUMA_HISTORY, rgba32f) uniform image2D rw_luma_history; +#endif +#if defined FSR2_BIND_UAV_UPSCALED_OUTPUT + layout (set = 1, binding = FSR2_BIND_UAV_UPSCALED_OUTPUT, rgba32f) uniform image2D rw_upscaled_output; +#endif +#if defined FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE + layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE, r32f) coherent uniform image2D rw_img_mip_shading_change; +#endif +#if defined FSR2_BIND_UAV_EXPOSURE_MIP_5 + layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE_MIP_5, r32f) coherent uniform image2D rw_img_mip_5; +#endif +#if defined FSR2_BIND_UAV_DILATED_REACTIVE_MASKS + layout (set = 1, binding = FSR2_BIND_UAV_DILATED_REACTIVE_MASKS, rg32f) uniform image2D rw_dilated_reactive_masks; +#endif +#if defined FSR2_BIND_UAV_EXPOSURE + layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE, rg32f) uniform image2D rw_exposure; +#endif +#if defined FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC + layout (set = 1, binding = FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC, r32ui) coherent uniform uimage2D rw_spd_global_atomic; +#endif + +FfxFloat32 LoadMipLuma(FfxInt32x2 iPxPos, FfxInt32 mipLevel) +{ +#if defined(FSR2_BIND_SRV_EXPOSURE_MIPS) + return texelFetch(r_imgMips, iPxPos, FfxInt32(mipLevel)).r; +#else + return 0.f; +#endif +} + + +FfxFloat32 SampleMipLuma(FfxFloat32x2 fUV, FfxInt32 mipLevel) +{ +#if defined(FSR2_BIND_SRV_EXPOSURE_MIPS) + fUV *= cbFSR2.depthclip_uv_scale; + return textureLod(sampler2D(r_imgMips, s_LinearClamp), fUV, FfxFloat32(mipLevel)).r; +#else + return 0.f; +#endif +} + +// +// a 0 0 0 x +// 0 b 0 0 y +// 0 0 c d z +// 0 0 e 0 1 +// +// z' = (z*c+d)/(z*e) +// z' = (c/e) + d/(z*e) +// z' - (c/e) = d/(z*e) +// (z'e - c)/e = d/(z*e) +// e / (z'e - c) = (z*e)/d +// (e * d) / (z'e - c) = z*e +// z = d / (z'e - c) +FfxFloat32 ConvertFromDeviceDepthToViewSpace(FfxFloat32 fDeviceDepth) +{ + return -cbFSR2.fDeviceToViewDepth[2] / (fDeviceDepth * cbFSR2.fDeviceToViewDepth[1] - cbFSR2.fDeviceToViewDepth[0]); +} + +FfxFloat32 LoadInputDepth(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_DEPTH) + return texelFetch(r_depth, iPxPos, 0).r; +#else + return 0.f; +#endif +} + +FfxFloat32 LoadReactiveMask(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_REACTIVE_MASK) + return texelFetch(r_reactive_mask, FfxInt32x2(iPxPos), 0).r; +#else + return 0.f; +#endif +} + +FfxFloat32x4 GatherReactiveMask(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_REACTIVE_MASK) + return textureGather(sampler2D(r_reactive_mask, s_LinearClamp), FfxFloat32x2(iPxPos) * cbFSR2.reactive_mask_dim_rcp, 0); +#else + return FfxFloat32x4(0.f); +#endif +} + +FfxFloat32 LoadTransparencyAndCompositionMask(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) + return texelFetch(r_transparency_and_composition_mask, iPxPos, 0).r; +#else + return 0.f; +#endif +} + +FfxFloat32 SampleTransparencyAndCompositionMask(FfxFloat32x2 fUV) +{ +#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) + fUV *= cbFSR2.depthclip_uv_scale; + return textureLod(sampler2D(r_transparency_and_composition_mask, s_LinearClamp), fUV, 0.0f).x; +#else + return 0.f; +#endif +} + +FfxFloat32 PreExposure() +{ + return cbFSR2.fPreExposure; +} + +FfxFloat32x3 LoadInputColor(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_INPUT_COLOR) + return texelFetch(r_input_color_jittered, iPxPos, 0).rgb / PreExposure(); +#else + return FfxFloat32x3(0.f); +#endif +} + +FfxFloat32x3 LoadInputColorWithoutPreExposure(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_INPUT_COLOR) + return texelFetch(r_input_color_jittered, iPxPos, 0).rgb; +#else + return FfxFloat32x3(0.f); +#endif +} + +FfxFloat32x3 LoadPreparedInputColor(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) + return texelFetch(r_prepared_input_color, iPxPos, 0).rgb; +#else + return FfxFloat32x3(0.f); +#endif +} + +FfxFloat32 LoadPreparedInputColorLuma(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) + return texelFetch(r_prepared_input_color, iPxPos, 0).a; +#else + return 0.f; +#endif +} + +FfxFloat32x2 LoadInputMotionVector(FfxInt32x2 iPxDilatedMotionVectorPos) +{ +#if defined(FSR2_BIND_SRV_MOTION_VECTORS) + FfxFloat32x2 fSrcMotionVector = texelFetch(r_motion_vectors, iPxDilatedMotionVectorPos, 0).xy; +#else + FfxFloat32x2 fSrcMotionVector = FfxFloat32x2(0.f); +#endif + + FfxFloat32x2 fUvMotionVector = fSrcMotionVector * cbFSR2.MotionVectorScale; + +#if FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS + fUvMotionVector -= cbFSR2.fMotionVectorJitterCancellation; +#endif + + return fUvMotionVector; +} + +FfxFloat32x4 LoadHistory(FfxInt32x2 iPxHistory) +{ +#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED) + return texelFetch(r_internal_upscaled_color, iPxHistory, 0); +#else + return FfxFloat32x4(0.0f); +#endif +} + +FfxFloat32x4 LoadRwInternalUpscaledColorAndWeight(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) + return imageLoad(rw_internal_upscaled_color, iPxPos); +#else + return FfxFloat32x4(0.f); +#endif +} + +void StoreLumaHistory(FfxInt32x2 iPxPos, FfxFloat32x4 fLumaHistory) +{ +#if defined(FSR2_BIND_UAV_LUMA_HISTORY) + imageStore(rw_luma_history, FfxInt32x2(iPxPos), fLumaHistory); +#endif +} + +FfxFloat32x4 LoadRwLumaHistory(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_UAV_LUMA_HISTORY) + return imageLoad(rw_luma_history, FfxInt32x2(iPxPos)); +#else + return FfxFloat32x4(1.f); +#endif +} + +FfxFloat32 LoadLumaStabilityFactor(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_LUMA_HISTORY) + return texelFetch(r_luma_history, FfxInt32x2(iPxPos), 0).w; +#else + return 0.f; +#endif +} + +FfxFloat32 SampleLumaStabilityFactor(FfxFloat32x2 fUV) +{ +#if defined(FSR2_BIND_SRV_LUMA_HISTORY) + fUV *= cbFSR2.depthclip_uv_scale; + return textureLod(sampler2D(r_luma_history, s_LinearClamp), fUV, 0.0f).w; +#else + return 0.f; +#endif +} + +void StoreReprojectedHistory(FfxInt32x2 iPxHistory, FfxFloat32x4 fHistory) +{ +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) + imageStore(rw_internal_upscaled_color, iPxHistory, fHistory); +#endif +} + +void StoreInternalColorAndWeight(FfxInt32x2 iPxPos, FfxFloat32x4 fColorAndWeight) +{ +#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) + imageStore(rw_internal_upscaled_color, FfxInt32x2(iPxPos), fColorAndWeight); +#endif +} + +void StoreUpscaledOutput(FfxInt32x2 iPxPos, FfxFloat32x3 fColor) +{ +#if defined(FSR2_BIND_UAV_UPSCALED_OUTPUT) + imageStore(rw_upscaled_output, FfxInt32x2(iPxPos), FfxFloat32x4(fColor * PreExposure(), 1.f)); +#endif +} + +FfxFloat32x3 LoadLockStatus(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_LOCK_STATUS) + FfxFloat32x3 fLockStatus = texelFetch(r_lock_status, iPxPos, 0).rgb; + + fLockStatus[0] -= LockInitialLifetime() * 2.0f; + + return fLockStatus; +#else + return FfxFloat32x3(0.f); +#endif +} + +FfxFloat32x3 LoadRwLockStatus(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_UAV_LOCK_STATUS) + FfxFloat32x3 fLockStatus = imageLoad(rw_lock_status, iPxPos).rgb; + + fLockStatus[0] -= LockInitialLifetime() * 2.0f; + + return fLockStatus; +#else + return FfxFloat32x3(0.f); +#endif +} + +void StoreLockStatus(FfxInt32x2 iPxPos, FfxFloat32x3 fLockstatus) +{ +#if defined(FSR2_BIND_UAV_LOCK_STATUS) + fLockstatus[0] += LockInitialLifetime() * 2.0f; + + imageStore(rw_lock_status, iPxPos, vec4(fLockstatus, 0.0f)); +#endif +} + +void StorePreparedInputColor(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x4 fTonemapped) +{ +#if defined(FSR2_BIND_UAV_PREPARED_INPUT_COLOR) + imageStore(rw_prepared_input_color, iPxPos, fTonemapped); +#endif +} + +FfxBoolean IsResponsivePixel(FfxInt32x2 iPxPos) +{ + return FFX_FALSE; //not supported in prototype +} + +FfxFloat32 LoadDepthClip(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_DEPTH_CLIP) + return texelFetch(r_depth_clip, iPxPos, 0).r; +#else + return 0.f; +#endif +} + +FfxFloat32 SampleDepthClip(FfxFloat32x2 fUV) +{ +#if defined(FSR2_BIND_SRV_DEPTH_CLIP) + fUV *= cbFSR2.depthclip_uv_scale; + return textureLod(sampler2D(r_depth_clip, s_LinearClamp), fUV, 0.0f).r; +#else + return 0.f; +#endif +} + +FfxFloat32x3 SampleLockStatus(FfxFloat32x2 fUV) +{ +#if defined(FSR2_BIND_SRV_LOCK_STATUS) + fUV *= cbFSR2.postprocessed_lockstatus_uv_scale; + FfxFloat32x3 fLockStatus = textureLod(sampler2D(r_lock_status, s_LinearClamp), fUV, 0.0f).rgb; + fLockStatus[0] -= LockInitialLifetime() * 2.0f; + return fLockStatus; +#else + return FfxFloat32x3(0.f); +#endif +} + +void StoreDepthClip(FfxInt32x2 iPxPos, FfxFloat32 fClip) +{ +#if defined(FSR2_BIND_UAV_DEPTH_CLIP) + imageStore(rw_depth_clip, iPxPos, vec4(fClip, 0.0f, 0.0f, 0.0f)); +#endif +} + +FfxFloat32 TanHalfFoV() +{ + return cbFSR2.fTanHalfFOV; +} + +FfxFloat32 LoadSceneDepth(FfxInt32x2 iPxInput) +{ +#if defined(FSR2_BIND_SRV_DEPTH) + return texelFetch(r_depth, iPxInput, 0).r; +#else + return 0.f; +#endif +} + +FfxFloat32 LoadReconstructedPrevDepth(FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) + return uintBitsToFloat(texelFetch(r_reconstructed_previous_nearest_depth, iPxPos, 0).r); +#else + return 0.f; +#endif +} + +void StoreReconstructedDepth(FfxInt32x2 iPxSample, FfxFloat32 fDepth) +{ + FfxUInt32 uDepth = floatBitsToUint(fDepth); +#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) + #if FFX_FSR2_OPTION_INVERTED_DEPTH + imageAtomicMax(rw_reconstructed_previous_nearest_depth, iPxSample, uDepth); + #else + imageAtomicMin(rw_reconstructed_previous_nearest_depth, iPxSample, uDepth); // min for standard, max for inverted depth + #endif +#endif +} + +void SetReconstructedDepth(FfxInt32x2 iPxSample, FfxUInt32 uValue) +{ +#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) + imageStore(rw_reconstructed_previous_nearest_depth, iPxSample, uvec4(uValue, 0, 0, 0)); +#endif +} + +void StoreDilatedDepth(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32 fDepth) +{ +#if defined(FSR2_BIND_UAV_DILATED_DEPTH) + //FfxUInt32 uDepth = f32tof16(fDepth); + imageStore(rw_dilatedDepth, iPxPos, vec4(fDepth, 0.0f, 0.0f, 0.0f)); +#endif +} + +void StoreDilatedMotionVector(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fMotionVector) +{ +#if defined(FSR2_BIND_UAV_DILATED_MOTION_VECTORS) + imageStore(rw_dilated_motion_vectors, iPxPos, vec4(fMotionVector, 0.0f, 0.0f)); +#endif +} + +FfxFloat32x2 LoadDilatedMotionVector(FfxInt32x2 iPxInput) +{ +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) + return texelFetch(r_dilated_motion_vectors, iPxInput, 0).rg; +#else + return FfxFloat32x2(0.f); +#endif +} + +FfxFloat32x2 SampleDilatedMotionVector(FfxFloat32x2 fUV) +{ +#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) + fUV *= cbFSR2.depthclip_uv_scale; // TODO: assuming these are (RenderSize() / MaxRenderSize()) + return textureLod(sampler2D(r_dilated_motion_vectors, s_LinearClamp), fUV, 0.0f).rg; +#else + return FfxFloat32x2(0.f); +#endif +} + +FfxFloat32 LoadDilatedDepth(FfxInt32x2 iPxInput) +{ +#if defined(FSR2_BIND_SRV_DILATED_DEPTH) + return texelFetch(r_dilatedDepth, iPxInput, 0).r; +#else + return 0.f; +#endif +} + +FfxFloat32 Exposure() +{ + #if defined(FSR2_BIND_SRV_EXPOSURE) + FfxFloat32 exposure = texelFetch(r_exposure, FfxInt32x2(0,0), 0).x; + #else + FfxFloat32 exposure = 1.f; + #endif + + if (exposure == 0.0f) { + exposure = 1.0f; + } + + return exposure; +} + +FfxFloat32 SampleLanczos2Weight(FfxFloat32 x) +{ +#if defined(FSR2_BIND_SRV_LANCZOS_LUT) + return textureLod(sampler2D(r_lanczos_lut, s_LinearClamp), FfxFloat32x2(x / 2.0f, 0.5f), 0.0f).x; +#else + return 0.f; +#endif +} + +FfxFloat32 SampleUpsampleMaximumBias(FfxFloat32x2 uv) +{ +#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) + // Stored as a SNORM, so make sure to multiply by 2 to retrieve the actual expected range. + return FfxFloat32(2.0f) * FfxFloat32(textureLod(sampler2D(r_upsample_maximum_bias_lut, s_LinearClamp), abs(uv) * 2.0f, 0.0f).r); +#else + return FfxFloat32(0.f); +#endif +} + +FfxFloat32x2 SampleDilatedReactiveMasks(FfxFloat32x2 fUV) +{ +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) + fUV *= cbFSR2.depthclip_uv_scale; // TODO: assuming these are (RenderSize() / MaxRenderSize()) + return textureLod(sampler2D(r_dilated_reactive_masks, s_LinearClamp), fUV, 0.0f).rg; +#else + return FfxFloat32x2(0.f); +#endif +} + +FfxFloat32x2 LoadDilatedReactiveMasks(FFX_PARAMETER_IN FfxInt32x2 iPxPos) +{ +#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) + return texelFetch(r_dilated_reactive_masks, iPxPos, 0).rg; +#else + return FfxFloat32x2(0.f); +#endif +} + +void StoreDilatedReactiveMasks(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fDilatedReactiveMasks) +{ +#if defined(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS) + imageStore(rw_dilated_reactive_masks, iPxPos, vec4(fDilatedReactiveMasks, 0.0f, 0.0f)); +#endif +} + + +#endif // #if defined(FFX_GPU) diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_glsl.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_glsl.h.meta new file mode 100644 index 0000000..b3f8307 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_glsl.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 5cb9ad224bca34e4fbc3acf33bd82fbe +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_fsr2_callbacks_hlsl.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_hlsl.h similarity index 100% rename from Assets/Resources/Shaders/ffx_fsr2_callbacks_hlsl.h rename to Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_hlsl.h diff --git a/Assets/Resources/Shaders/ffx_fsr2_callbacks_hlsl.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_hlsl.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_fsr2_callbacks_hlsl.h.meta rename to Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_hlsl.h.meta index 0e963b5..1627fd4 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_callbacks_hlsl.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_callbacks_hlsl.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 5c9c193ec7dc3064097ab1f74232c98d +guid: eb121968296f9ba44b35d7e18d2b79df PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/Shaders/ffx_fsr2_common.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_common.h similarity index 100% rename from Assets/Resources/Shaders/ffx_fsr2_common.h rename to Assets/Resources/FSR2/shaders/ffx_fsr2_common.h diff --git a/Assets/Resources/Shaders/ffx_fsr2_common.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_common.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_fsr2_common.h.meta rename to Assets/Resources/FSR2/shaders/ffx_fsr2_common.h.meta index acb96ea..2264050 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_common.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_common.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 9cdba182d33eab646b8df7521a7ec649 +guid: 2176dca22b6e9604da8329c79abae68d PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid.h new file mode 100644 index 0000000..d5bbbcf --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid.h @@ -0,0 +1,188 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +FFX_GROUPSHARED FfxUInt32 spdCounter; + +#ifndef SPD_PACKED_ONLY +FFX_GROUPSHARED FfxFloat32 spdIntermediateR[16][16]; +FFX_GROUPSHARED FfxFloat32 spdIntermediateG[16][16]; +FFX_GROUPSHARED FfxFloat32 spdIntermediateB[16][16]; +FFX_GROUPSHARED FfxFloat32 spdIntermediateA[16][16]; + +FfxFloat32x4 SpdLoadSourceImage(FfxFloat32x2 tex, FfxUInt32 slice) +{ + FfxFloat32x3 fRgb = LoadInputColor(FfxInt32x2(tex)); + + FFX_STATIC const FfxFloat32x3 rgb2y = FfxFloat32x3(0.2126, 0.7152, 0.0722); + + //compute log luma + const FfxFloat32 fLogLuma = log(ffxMax(FSR2_EPSILON, dot(rgb2y, fRgb))); + + // Make sure out of screen pixels contribute no value to the end result + const FfxFloat32 result = all(FFX_LESS_THAN(tex, RenderSize())) ? fLogLuma : 0.0f; + + return FfxFloat32x4(result, 0, 0, 0); +} + +FfxFloat32x4 SpdLoad(FfxInt32x2 tex, FfxUInt32 slice) +{ + return SPD_LoadMipmap5(tex); +} + +void SpdStore(FfxInt32x2 pix, FfxFloat32x4 outValue, FfxUInt32 index, FfxUInt32 slice) +{ + if (index == LumaMipLevelToUse() || index == 5) + { + SPD_SetMipmap(pix, index, outValue.r); + } + + if (index == MipCount() - 1) { //accumulate on 1x1 level + + if (all(FFX_EQUAL(pix, FfxInt32x2(0, 0)))) + { + FfxFloat32 prev = SPD_LoadExposureBuffer().y; + FfxUInt32x2 renderSize = SPD_RenderSize(); + FfxFloat32 result = outValue.r / (renderSize.x * renderSize.y); + + if (prev < resetAutoExposureAverageSmoothing) // Compare Lavg, so small or negative values + { + FfxFloat32 rate = 1.0f; + result = prev + (result - prev) * (1 - exp(-DeltaTime() * rate)); + } + FfxFloat32x2 spdOutput = FfxFloat32x2(ComputeAutoExposureFromLavg(result), result); + SPD_SetExposureBuffer(spdOutput); + } + } +} + +void SpdIncreaseAtomicCounter(FfxUInt32 slice) +{ + SPD_IncreaseAtomicCounter(spdCounter); +} + +FfxUInt32 SpdGetAtomicCounter() +{ + return spdCounter; +} + +void SpdResetAtomicCounter(FfxUInt32 slice) +{ + SPD_ResetAtomicCounter(); +} + +FfxFloat32x4 SpdLoadIntermediate(FfxUInt32 x, FfxUInt32 y) +{ + return FfxFloat32x4( + spdIntermediateR[x][y], + spdIntermediateG[x][y], + spdIntermediateB[x][y], + spdIntermediateA[x][y]); +} +void SpdStoreIntermediate(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value) +{ + spdIntermediateR[x][y] = value.x; + spdIntermediateG[x][y] = value.y; + spdIntermediateB[x][y] = value.z; + spdIntermediateA[x][y] = value.w; +} +FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3) +{ + return (v0 + v1 + v2 + v3); +} +#endif + +// define fetch and store functions Packed +#if FFX_HALF +#error Callback must be implemented + +FFX_GROUPSHARED FfxFloat16x2 spdIntermediateRG[16][16]; +FFX_GROUPSHARED FfxFloat16x2 spdIntermediateBA[16][16]; + +FfxFloat16x4 SpdLoadSourceImageH(FfxFloat32x2 tex, FfxUInt32 slice) +{ + return FfxFloat16x4(imgDst[0][FfxFloat32x3(tex, slice)]); +} +FfxFloat16x4 SpdLoadH(FfxInt32x2 p, FfxUInt32 slice) +{ + return FfxFloat16x4(imgDst6[FfxUInt32x3(p, slice)]); +} +void SpdStoreH(FfxInt32x2 p, FfxFloat16x4 value, FfxUInt32 mip, FfxUInt32 slice) +{ + if (index == LumaMipLevelToUse() || index == 5) + { + imgDst6[FfxUInt32x3(p, slice)] = FfxFloat32x4(value); + return; + } + imgDst[mip + 1][FfxUInt32x3(p, slice)] = FfxFloat32x4(value); +} +void SpdIncreaseAtomicCounter(FfxUInt32 slice) +{ + InterlockedAdd(rw_spd_global_atomic[FfxInt16x2(0, 0)].counter[slice], 1, spdCounter); +} +FfxUInt32 SpdGetAtomicCounter() +{ + return spdCounter; +} +void SpdResetAtomicCounter(FfxUInt32 slice) +{ + rw_spd_global_atomic[FfxInt16x2(0, 0)].counter[slice] = 0; +} +FfxFloat16x4 SpdLoadIntermediateH(FfxUInt32 x, FfxUInt32 y) +{ + return FfxFloat16x4( + spdIntermediateRG[x][y].x, + spdIntermediateRG[x][y].y, + spdIntermediateBA[x][y].x, + spdIntermediateBA[x][y].y); +} +void SpdStoreIntermediateH(FfxUInt32 x, FfxUInt32 y, FfxFloat16x4 value) +{ + spdIntermediateRG[x][y] = value.xy; + spdIntermediateBA[x][y] = value.zw; +} +FfxFloat16x4 SpdReduce4H(FfxFloat16x4 v0, FfxFloat16x4 v1, FfxFloat16x4 v2, FfxFloat16x4 v3) +{ + return (v0 + v1 + v2 + v3) * FfxFloat16(0.25); +} +#endif + +#include "ffx_spd.h" + +void ComputeAutoExposure(FfxUInt32x3 WorkGroupId, FfxUInt32 LocalThreadIndex) +{ +#if FFX_HALF + SpdDownsampleH( + FfxUInt32x2(WorkGroupId.xy), + FfxUInt32(LocalThreadIndex), + FfxUInt32(MipCount()), + FfxUInt32(NumWorkGroups()), + FfxUInt32(WorkGroupId.z), + FfxUInt32x2(WorkGroupOffset())); +#else + SpdDownsample( + FfxUInt32x2(WorkGroupId.xy), + FfxUInt32(LocalThreadIndex), + FfxUInt32(MipCount()), + FfxUInt32(NumWorkGroups()), + FfxUInt32(WorkGroupId.z), + FfxUInt32x2(WorkGroupOffset())); +#endif +} \ No newline at end of file diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid.h.meta new file mode 100644 index 0000000..08c002c --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: dbcdb6dfb36311a49aa7b05bc5054280 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl new file mode 100644 index 0000000..9a6a329 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl @@ -0,0 +1,171 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_INPUT_COLOR 0 +#define FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC 1 +#define FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE 2 +#define FSR2_BIND_UAV_EXPOSURE_MIP_5 3 +#define FSR2_BIND_UAV_EXPOSURE 4 +#define FSR2_BIND_CB_FSR2 5 +#define FSR2_BIND_CB_SPD 6 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +#if defined(FSR2_BIND_CB_SPD) + layout (set = 1, binding = FSR2_BIND_CB_SPD, std140) uniform cbSPD_t + { + uint mips; + uint numWorkGroups; + uvec2 workGroupOffset; + uvec2 renderSize; + } cbSPD; + + uint MipCount() + { + return cbSPD.mips; + } + + uint NumWorkGroups() + { + return cbSPD.numWorkGroups; + } + + uvec2 WorkGroupOffset() + { + return cbSPD.workGroupOffset; + } + + uvec2 SPD_RenderSize() + { + return cbSPD.renderSize; + } +#else + uint MipCount() + { + return 0; + } + + uint NumWorkGroups() + { + return 0; + } + + uvec2 WorkGroupOffset() + { + return uvec2(0); + } + + uvec2 SPD_RenderSize() + { + return uvec2(0); + } +#endif + +vec2 SPD_LoadExposureBuffer() +{ +#if defined(FSR2_BIND_UAV_EXPOSURE) + return imageLoad(rw_exposure, ivec2(0,0)).xy; +#else + return vec2(0); +#endif +} + +void SPD_SetExposureBuffer(vec2 value) +{ +#if defined(FSR2_BIND_UAV_EXPOSURE) + imageStore(rw_exposure, ivec2(0,0), vec4(value, 0.0f, 0.0f)); +#endif +} + +vec4 SPD_LoadMipmap5(ivec2 iPxPos) +{ +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) + return vec4(imageLoad(rw_img_mip_5, iPxPos).x, 0.0f, 0.0f, 0.0f); +#else + return vec4(0); +#endif +} + +void SPD_SetMipmap(ivec2 iPxPos, uint slice, float value) +{ + switch (slice) + { +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE) + case FFX_FSR2_SHADING_CHANGE_MIP_LEVEL: + imageStore(rw_img_mip_shading_change, iPxPos, vec4(value, 0.0f, 0.0f, 0.0f)); + break; +#endif +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) + case 5: + imageStore(rw_img_mip_5, iPxPos, vec4(value, 0.0f, 0.0f, 0.0f)); + break; +#endif + default: + // avoid flattened side effect +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE) + imageStore(rw_img_mip_shading_change, iPxPos, vec4(imageLoad(rw_img_mip_shading_change, iPxPos).x, 0.0f, 0.0f, 0.0f)); +#elif defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) + imageStore(rw_img_mip_5, iPxPos, vec4(imageLoad(rw_img_mip_5, iPxPos).x, 0.0f, 0.0f, 0.0f)); +#endif + break; + } +} + +void SPD_IncreaseAtomicCounter(inout uint spdCounter) +{ +#if defined(FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC) + spdCounter = imageAtomicAdd(rw_spd_global_atomic, ivec2(0,0), 1); +#endif +} + +void SPD_ResetAtomicCounter() +{ +#if defined(FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC) + imageStore(rw_spd_global_atomic, ivec2(0,0), uvec4(0)); +#endif +} + +#include "ffx_fsr2_compute_luminance_pyramid.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 256 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + ComputeAutoExposure(gl_WorkGroupID.xyz, gl_LocalInvocationIndex); +} \ No newline at end of file diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl.meta new file mode 100644 index 0000000..d6ba893 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 43b69b4a78d09164aa834576507445c1 +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.hlsl new file mode 100644 index 0000000..07a097a --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.hlsl @@ -0,0 +1,164 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#define FSR2_BIND_SRV_INPUT_COLOR 0 +#define FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC 0 +#define FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE 1 +#define FSR2_BIND_UAV_EXPOSURE_MIP_5 2 +#define FSR2_BIND_UAV_EXPOSURE 3 +#define FSR2_BIND_CB_FSR2 0 +#define FSR2_BIND_CB_SPD 1 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" + +#if defined(FSR2_BIND_CB_SPD) + cbuffer cbSPD : FFX_FSR2_DECLARE_CB(FSR2_BIND_CB_SPD) { + + uint mips; + uint numWorkGroups; + uint2 workGroupOffset; + uint2 renderSize; + }; + + uint MipCount() + { + return mips; + } + + uint NumWorkGroups() + { + return numWorkGroups; + } + + uint2 WorkGroupOffset() + { + return workGroupOffset; + } + + uint2 SPD_RenderSize() + { + return renderSize; + } +#else + uint MipCount() + { + return 0; + } + + uint NumWorkGroups() + { + return 0; + } + + uint2 WorkGroupOffset() + { + return uint2(0, 0); + } + + uint2 SPD_RenderSize() + { + return uint2(0, 0); + } +#endif + + +float2 SPD_LoadExposureBuffer() +{ +#if defined(FSR2_BIND_UAV_EXPOSURE) || defined(FFX_INTERNAL) + return rw_exposure[min16int2(0,0)]; +#else + return 0; +#endif +} + +void SPD_SetExposureBuffer(float2 value) +{ +#if defined(FSR2_BIND_UAV_EXPOSURE) || defined(FFX_INTERNAL) + rw_exposure[min16int2(0,0)] = value; +#endif +} + +float4 SPD_LoadMipmap5(int2 iPxPos) +{ +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) || defined(FFX_INTERNAL) + return float4(rw_img_mip_5[iPxPos], 0, 0, 0); +#else + return 0; +#endif +} + +void SPD_SetMipmap(int2 iPxPos, int slice, float value) +{ + switch (slice) + { +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE) || defined(FFX_INTERNAL) + case FFX_FSR2_SHADING_CHANGE_MIP_LEVEL: + rw_img_mip_shading_change[iPxPos] = value; + break; +#endif +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) || defined(FFX_INTERNAL) + case 5: + rw_img_mip_5[iPxPos] = value; + break; +#endif + default: + // avoid flattened side effect +#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE) || defined(FFX_INTERNAL) + rw_img_mip_shading_change[iPxPos] = rw_img_mip_shading_change[iPxPos]; +#elif defined(FSR2_BIND_UAV_EXPOSURE_MIP_5) || defined(FFX_INTERNAL) + rw_img_mip_5[iPxPos] = rw_img_mip_5[iPxPos]; +#endif + break; + } +} + +void SPD_IncreaseAtomicCounter(inout uint spdCounter) +{ + InterlockedAdd(rw_spd_global_atomic[min16int2(0,0)], 1, spdCounter); +} + +void SPD_ResetAtomicCounter() +{ + rw_spd_global_atomic[min16int2(0,0)] = 0; +} + +#include "ffx_fsr2_compute_luminance_pyramid.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 256 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_CB2_ROOTSIG_CONTENT +void CS(uint3 WorkGroupId : SV_GroupID, uint LocalThreadIndex : SV_GroupIndex) +{ + ComputeAutoExposure(WorkGroupId, LocalThreadIndex); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.hlsl.meta new file mode 100644 index 0000000..ac39806 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 19dfb00afb70c3144b43ec2dc05ecdd9 +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip.h new file mode 100644 index 0000000..81db737 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip.h @@ -0,0 +1,98 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_DEPTH_CLIP_H +#define FFX_FSR2_DEPTH_CLIP_H + +FFX_STATIC const FfxFloat32 DepthClipBaseScale = 4.0f; + +FfxFloat32 ComputeSampleDepthClip(FfxInt32x2 iPxSamplePos, FfxFloat32 fPreviousDepth, FfxFloat32 fPreviousDepthBilinearWeight, FfxFloat32 fCurrentDepthViewSpace) +{ + FfxFloat32 fPrevNearestDepthViewSpace = abs(ConvertFromDeviceDepthToViewSpace(fPreviousDepth)); + + // Depth separation logic ref: See "Minimum Triangle Separation for Correct Z-Buffer Occlusion" + // Intention: worst case of formula in Figure4 combined with Ksep factor in Section 4 + // TODO: check intention and improve, some banding visible + const FfxFloat32 fHalfViewportWidth = RenderSize().x * 0.5f; + FfxFloat32 fDepthThreshold = ffxMin(fCurrentDepthViewSpace, fPrevNearestDepthViewSpace); + + // WARNING: Ksep only works with reversed-z with infinite projection. + const FfxFloat32 Ksep = 1.37e-05f; + FfxFloat32 fRequiredDepthSeparation = Ksep * fDepthThreshold * TanHalfFoV() * fHalfViewportWidth; + FfxFloat32 fDepthDiff = fCurrentDepthViewSpace - fPrevNearestDepthViewSpace; + + FfxFloat32 fDepthClipFactor = (fDepthDiff > 0) ? ffxSaturate(fRequiredDepthSeparation / fDepthDiff) : 1.0f; + +#ifdef _DEBUG + rw_debug_out[iPxSamplePos] = FfxFloat32x4(fCurrentDepthViewSpace, fPrevNearestDepthViewSpace, fDepthDiff, fDepthClipFactor); +#endif + + return fPreviousDepthBilinearWeight * fDepthClipFactor * ffxLerp(1.0f, DepthClipBaseScale, ffxSaturate(fDepthDiff * fDepthDiff)); +} + +FfxFloat32 ComputeDepthClip(FfxFloat32x2 fUvSample, FfxFloat32 fCurrentDepthViewSpace) +{ + FfxFloat32x2 fPxSample = fUvSample * RenderSize() - 0.5f; + FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); + FfxFloat32x2 fPxFrac = ffxFract(fPxSample); + + const FfxFloat32 fBilinearWeights[2][2] = { + { + (1 - fPxFrac.x) * (1 - fPxFrac.y), + (fPxFrac.x) * (1 - fPxFrac.y) + }, + { + (1 - fPxFrac.x) * (fPxFrac.y), + (fPxFrac.x) * (fPxFrac.y) + } + }; + + FfxFloat32 fDepth = 0.0f; + FfxFloat32 fWeightSum = 0.0f; + for (FfxInt32 y = 0; y <= 1; ++y) { + for (FfxInt32 x = 0; x <= 1; ++x) { + FfxInt32x2 iSamplePos = iPxSample + FfxInt32x2(x, y); + if (IsOnScreen(iSamplePos, RenderSize())) { + FfxFloat32 fBilinearWeight = fBilinearWeights[y][x]; + if (fBilinearWeight > reconstructedDepthBilinearWeightThreshold) { + fDepth += ComputeSampleDepthClip(iSamplePos, LoadReconstructedPrevDepth(iSamplePos), fBilinearWeight, fCurrentDepthViewSpace); + fWeightSum += fBilinearWeight; + } + } + } + } + + return (fWeightSum > 0) ? fDepth / fWeightSum : DepthClipBaseScale; +} + +void DepthClip(FfxInt32x2 iPxPos) +{ + FfxFloat32x2 fDepthUv = (iPxPos + 0.5f) / RenderSize(); + FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos); + FfxFloat32x2 fDilatedUv = fDepthUv + fMotionVector; + FfxFloat32 fCurrentDepthViewSpace = abs(ConvertFromDeviceDepthToViewSpace(LoadDilatedDepth(iPxPos))); + + FfxFloat32 fDepthClip = ComputeDepthClip(fDilatedUv, fCurrentDepthViewSpace); + + StoreDepthClip(iPxPos, fDepthClip); +} + +#endif //!defined( FFX_FSR2_DEPTH_CLIPH ) \ No newline at end of file diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip.h.meta new file mode 100644 index 0000000..eca1b58 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: f7c16477aeb3a9b4f94f4ef818d10d9b +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.glsl new file mode 100644 index 0000000..7233ec6 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.glsl @@ -0,0 +1,62 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 3 +// SRV 7 : FSR2_ReconstructedPrevNearestDepth : r_reconstructed_previous_nearest_depth +// SRV 8 : FSR2_DilatedVelocity : r_dilated_motion_vectors +// SRV 9 : FSR2_DilatedDepth : r_dilatedDepth +// UAV 12 : FSR2_DepthClip : rw_depth_clip +// CB 0 : cbFSR2 + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH 0 +#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 1 +#define FSR2_BIND_SRV_DILATED_DEPTH 2 +#define FSR2_BIND_UAV_DEPTH_CLIP 3 +#define FSR2_BIND_CB_FSR2 4 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_depth_clip.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + DepthClip(ivec2(gl_GlobalInvocationID.xy)); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.glsl.meta new file mode 100644 index 0000000..a0459e8 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: b02126743783b1942a0777c5e9b7526a +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.hlsl new file mode 100644 index 0000000..8433734 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.hlsl @@ -0,0 +1,63 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 3 +// SRV 7 : FSR2_ReconstructedPrevNearestDepth : r_reconstructed_previous_nearest_depth +// SRV 8 : FSR2_DilatedVelocity : r_dilated_motion_vectors +// SRV 9 : FSR2_DilatedDepth : r_dilatedDepth +// UAV 12 : FSR2_DepthClip : rw_depth_clip +// CB 0 : cbFSR2 + +#define FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH 0 +#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 1 +#define FSR2_BIND_SRV_DILATED_DEPTH 2 +#define FSR2_BIND_UAV_DEPTH_CLIP 0 +#define FSR2_BIND_CB_FSR2 0 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_depth_clip.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_PREFER_WAVE64 +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_ROOTSIG_CONTENT +void CS( + int2 iGroupId : SV_GroupID, + int2 iDispatchThreadId : SV_DispatchThreadID, + int2 iGroupThreadId : SV_GroupThreadID, + int iGroupIndex : SV_GroupIndex) +{ + DepthClip(iDispatchThreadId); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.hlsl.meta new file mode 100644 index 0000000..d7ce66d --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_depth_clip_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 7981b48622ddaa944909ebf209284d83 +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_lock.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock.h new file mode 100644 index 0000000..b2266b7 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock.h @@ -0,0 +1,126 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_LOCK_H +#define FFX_FSR2_LOCK_H + +FfxFloat32 GetLuma(FfxInt32x2 pos) +{ + //add some bias to avoid locking dark areas + return FfxFloat32(LoadPreparedInputColorLuma(pos)); +} + +FfxFloat32 ComputeThinFeatureConfidence(FfxInt32x2 pos) +{ + const FfxInt32 RADIUS = 1; + + FfxFloat32 fNucleus = GetLuma(pos); + + FfxFloat32 similar_threshold = 1.05f; + FfxFloat32 dissimilarLumaMin = FSR2_FLT_MAX; + FfxFloat32 dissimilarLumaMax = 0; + + /* + 0 1 2 + 3 4 5 + 6 7 8 + */ + + #define SETBIT(x) (1U << x) + + FfxUInt32 mask = SETBIT(4); //flag fNucleus as similar + + const FfxUInt32 rejectionMasks[4] = { + SETBIT(0) | SETBIT(1) | SETBIT(3) | SETBIT(4), //Upper left + SETBIT(1) | SETBIT(2) | SETBIT(4) | SETBIT(5), //Upper right + SETBIT(3) | SETBIT(4) | SETBIT(6) | SETBIT(7), //Lower left + SETBIT(4) | SETBIT(5) | SETBIT(7) | SETBIT(8), //Lower right + }; + + FfxInt32 idx = 0; + FFX_UNROLL + for (FfxInt32 y = -RADIUS; y <= RADIUS; y++) { + FFX_UNROLL + for (FfxInt32 x = -RADIUS; x <= RADIUS; x++, idx++) { + if (x == 0 && y == 0) continue; + + FfxInt32x2 samplePos = ClampLoad(pos, FfxInt32x2(x, y), FfxInt32x2(RenderSize())); + + FfxFloat32 sampleLuma = GetLuma(samplePos); + FfxFloat32 difference = ffxMax(sampleLuma, fNucleus) / ffxMin(sampleLuma, fNucleus); + + if (difference > 0 && (difference < similar_threshold)) { + mask |= SETBIT(idx); + } else { + dissimilarLumaMin = ffxMin(dissimilarLumaMin, sampleLuma); + dissimilarLumaMax = ffxMax(dissimilarLumaMax, sampleLuma); + } + } + } + + FfxBoolean isRidge = fNucleus > dissimilarLumaMax || fNucleus < dissimilarLumaMin; + + if (FFX_FALSE == isRidge) { + + return 0; + } + + FFX_UNROLL + for (FfxInt32 i = 0; i < 4; i++) { + + if ((mask & rejectionMasks[i]) == rejectionMasks[i]) { + return 0; + } + } + + return 1; +} + +FFX_STATIC FfxBoolean s_bLockUpdated = FFX_FALSE; + +FfxFloat32x3 ComputeLockStatus(FfxInt32x2 iPxLrPos, FfxFloat32x3 fLockStatus) +{ + FfxFloat32 fConfidenceOfThinFeature = ComputeThinFeatureConfidence(iPxLrPos); + + s_bLockUpdated = FFX_FALSE; + if (fConfidenceOfThinFeature > 0.0f) + { + //put to negative on new lock + fLockStatus[LOCK_LIFETIME_REMAINING] = (fLockStatus[LOCK_LIFETIME_REMAINING] == FfxFloat32(0.0f)) ? FfxFloat32(-LockInitialLifetime()) : FfxFloat32(-(LockInitialLifetime() * 2)); + + s_bLockUpdated = FFX_TRUE; + } + + return fLockStatus; +} + +void ComputeLock(FfxInt32x2 iPxLrPos) +{ + FfxInt32x2 iPxHrPos = ComputeHrPosFromLrPos(iPxLrPos); + + FfxFloat32x3 fLockStatus = ComputeLockStatus(iPxLrPos, LoadLockStatus(iPxHrPos)); + + if ((s_bLockUpdated)) { + StoreLockStatus(iPxHrPos, fLockStatus); + } +} + +#endif // FFX_FSR2_LOCK_H diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_lock.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock.h.meta new file mode 100644 index 0000000..6d83098 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 9b4cdc5f81194ac4fa946c31b86234ed +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.glsl new file mode 100644 index 0000000..9c37774 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.glsl @@ -0,0 +1,65 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 4 +// SRV 5 : m_UpscaleReactive : r_reactive_mask +// SRV 11 : FSR2_LockStatus2 : r_lock_status +// SRV 13 : FSR2_PreparedInputColor : r_prepared_input_color +// UAV 11 : FSR2_LockStatus1 : rw_lock_status +// UAV 27 : FSR2_ReactiveMaskMax : rw_reactive_max +// CB 0 : cbFSR2 +// CB 1 : FSR2DispatchOffsets + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_LOCK_STATUS 0 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 1 +#define FSR2_BIND_UAV_LOCK_STATUS 2 +#define FSR2_BIND_CB_FSR2 3 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_lock.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + uvec2 uDispatchThreadId = gl_WorkGroupID.xy * uvec2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + gl_LocalInvocationID.xy; + + ComputeLock(ivec2(uDispatchThreadId)); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.glsl.meta new file mode 100644 index 0000000..982c1d4 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 5f63384878e8f1c428f5716f3b8e1065 +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.hlsl new file mode 100644 index 0000000..492965c --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.hlsl @@ -0,0 +1,60 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 4 +// SRV 5 : m_UpscaleReactive : r_reactive_mask +// SRV 11 : FSR2_LockStatus2 : r_lock_status +// SRV 13 : FSR2_PreparedInputColor : r_prepared_input_color +// UAV 11 : FSR2_LockStatus1 : rw_lock_status +// CB 0 : cbFSR2 + +#define FSR2_BIND_SRV_LOCK_STATUS 1 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 2 +#define FSR2_BIND_UAV_LOCK_STATUS 0 +#define FSR2_BIND_CB_FSR2 0 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_lock.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_PREFER_WAVE64 +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_ROOTSIG_CONTENT +void CS(uint2 uGroupId : SV_GroupID, uint2 uGroupThreadId : SV_GroupThreadID) +{ + uint2 uDispatchThreadId = uGroupId * uint2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + uGroupThreadId; + + ComputeLock(uDispatchThreadId); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.hlsl.meta new file mode 100644 index 0000000..4ded638 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_lock_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 471a3f7a033c72f4fa737d4f8238a9bd +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_postprocess_lock_status.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_postprocess_lock_status.h new file mode 100644 index 0000000..959031b --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_postprocess_lock_status.h @@ -0,0 +1,98 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_POSTPROCESS_LOCK_STATUS_H +#define FFX_FSR2_POSTPROCESS_LOCK_STATUS_H + +FfxFloat32x4 WrapShadingChangeLuma(FfxInt32x2 iPxSample) +{ + return FfxFloat32x4(LoadMipLuma(iPxSample, LumaMipLevelToUse()), 0, 0, 0); +} + +#if FFX_HALF +FFX_MIN16_F4 WrapShadingChangeLuma(FFX_MIN16_I2 iPxSample) +{ + return FFX_MIN16_F4(LoadMipLuma(iPxSample, LumaMipLevelToUse()), 0, 0, 0); +} +#endif + +#if FFX_FSR2_OPTION_POSTPROCESSLOCKSTATUS_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBilinearSamplesMin16(FetchShadingChangeLumaSamples, WrapShadingChangeLuma) +#else +DeclareCustomFetchBilinearSamples(FetchShadingChangeLumaSamples, WrapShadingChangeLuma) +#endif +DeclareCustomTextureSample(ShadingChangeLumaSample, Bilinear, FetchShadingChangeLumaSamples) + +FfxFloat32 GetShadingChangeLuma(FfxFloat32x2 fUvCoord) +{ + // const FfxFloat32 fShadingChangeLuma = exp(ShadingChangeLumaSample(fUvCoord, LumaMipDimensions()) * LumaMipRcp()); + const FfxFloat32 fShadingChangeLuma = FfxFloat32(exp(SampleMipLuma(fUvCoord, LumaMipLevelToUse()) * FfxFloat32(LumaMipRcp()))); + return fShadingChangeLuma; +} + +LockState GetLockState(FfxFloat32x3 fLockStatus) +{ + LockState state = { FFX_FALSE, FFX_FALSE }; + + //Check if this is a new or refreshed lock + state.NewLock = fLockStatus[LOCK_LIFETIME_REMAINING] < FfxFloat32(0.0f); + + //For a non-refreshed lock, the lifetime is set to LockInitialLifetime() + state.WasLockedPrevFrame = fLockStatus[LOCK_TRUST] != FfxFloat32(0.0f); + + return state; +} + +LockState PostProcessLockStatus(FfxInt32x2 iPxHrPos, FFX_PARAMETER_IN FfxFloat32x2 fLrUvJittered, FFX_PARAMETER_IN FfxFloat32 fDepthClipFactor, const FfxFloat32 fAccumulationMask, FFX_PARAMETER_IN FfxFloat32 fHrVelocity, + FFX_PARAMETER_INOUT FfxFloat32 fAccumulationTotalWeight, FFX_PARAMETER_INOUT FfxFloat32x3 fLockStatus, FFX_PARAMETER_OUT FfxFloat32 fLuminanceDiff) { + + const LockState state = GetLockState(fLockStatus); + + fLockStatus[LOCK_LIFETIME_REMAINING] = abs(fLockStatus[LOCK_LIFETIME_REMAINING]); + + FfxFloat32 fShadingChangeLuma = GetShadingChangeLuma(fLrUvJittered); + + //init temporal shading change factor, init to -1 or so in reproject to know if "true new"? + fLockStatus[LOCK_TEMPORAL_LUMA] = (fLockStatus[LOCK_TEMPORAL_LUMA] == FfxFloat32(0.0f)) ? fShadingChangeLuma : fLockStatus[LOCK_TEMPORAL_LUMA]; + + FfxFloat32 fPreviousShadingChangeLuma = fLockStatus[LOCK_TEMPORAL_LUMA]; + fLockStatus[LOCK_TEMPORAL_LUMA] = ffxLerp(fLockStatus[LOCK_TEMPORAL_LUMA], FfxFloat32(fShadingChangeLuma), FfxFloat32(0.5f)); + fLuminanceDiff = FfxFloat32(1) - MinDividedByMax(fPreviousShadingChangeLuma, fShadingChangeLuma); + + if (fLuminanceDiff > FfxFloat32(0.2f)) { + KillLock(fLockStatus); + } + + if (!state.NewLock && fLockStatus[LOCK_LIFETIME_REMAINING] >= FfxFloat32(0)) + { + fLockStatus[LOCK_LIFETIME_REMAINING] *= (1.0f - fAccumulationMask); + + const FfxFloat32 depthClipThreshold = FfxFloat32(0.99f); + if (fDepthClipFactor < depthClipThreshold) + { + KillLock(fLockStatus); + } + } + + return state; +} + +#endif //!defined( FFX_FSR2_POSTPROCESS_LOCK_STATUS_H ) diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_postprocess_lock_status.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_postprocess_lock_status.h.meta new file mode 100644 index 0000000..ab7b1dd --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_postprocess_lock_status.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 3945c3cfd2cc1a64cb0513864d88d8ca +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color.h new file mode 100644 index 0000000..a773cda --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color.h @@ -0,0 +1,88 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_PREPARE_INPUT_COLOR_H +#define FFX_FSR2_PREPARE_INPUT_COLOR_H + +//TODO: Move to common location & share with Accumulate +void ClearResourcesForNextFrame(in FfxInt32x2 iPxHrPos) +{ + if (all(FFX_LESS_THAN(iPxHrPos, FfxInt32x2(RenderSize())))) + { +#if FFX_FSR2_OPTION_INVERTED_DEPTH + const FfxUInt32 farZ = 0x0; +#else + const FfxUInt32 farZ = 0x3f800000; +#endif + SetReconstructedDepth(iPxHrPos, farZ); + } +} + +void ComputeLumaStabilityFactor(FfxInt32x2 iPxLrPos, FfxFloat32 fCurrentFrameLuma) +{ + FfxFloat32x4 fCurrentFrameLumaHistory = LoadRwLumaHistory(iPxLrPos); + + fCurrentFrameLumaHistory.a = FfxFloat32(0); + + if (FrameIndex() > 3) { + FfxFloat32 fDiffs0 = MinDividedByMax(fCurrentFrameLumaHistory[2], fCurrentFrameLuma); + FfxFloat32 fDiffs1 = ffxMax(MinDividedByMax(fCurrentFrameLumaHistory[0], fCurrentFrameLuma), MinDividedByMax(fCurrentFrameLumaHistory[1], fCurrentFrameLuma)); + + fCurrentFrameLumaHistory.a = ffxSaturate(fDiffs1 - fDiffs0); + } + + //move history + fCurrentFrameLumaHistory[0] = fCurrentFrameLumaHistory[1]; + fCurrentFrameLumaHistory[1] = fCurrentFrameLumaHistory[2]; + fCurrentFrameLumaHistory[2] = fCurrentFrameLuma; + + StoreLumaHistory(iPxLrPos, fCurrentFrameLumaHistory); +} + +void PrepareInputColor(FfxInt32x2 iPxLrPos) +{ + //We assume linear data. if non-linear input (sRGB, ...), + //then we should convert to linear first and back to sRGB on output. + + FfxFloat32x3 fRgb = ffxMax(FfxFloat32x3(0, 0, 0), LoadInputColor(iPxLrPos)); + + fRgb *= Exposure(); + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + // Tonemap color, used in lockstatus and luma stability computations + fRgb = Tonemap(fRgb); +#endif + + FfxFloat32x4 fYCoCg; + + fYCoCg.xyz = RGBToYCoCg(fRgb); + + const FfxFloat32 fPerceivedLuma = RGBToPerceivedLuma(fRgb); + ComputeLumaStabilityFactor(iPxLrPos, fPerceivedLuma); + + //compute luma used to lock pixels, if used elsewhere the ffxPow must be moved! + fYCoCg.w = ffxPow(fPerceivedLuma, FfxFloat32(1.0 / 6.0)); + + StorePreparedInputColor(iPxLrPos, fYCoCg); + ClearResourcesForNextFrame(iPxLrPos); +} + +#endif // FFX_FSR2_PREPARE_INPUT_COLOR_H diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color.h.meta new file mode 100644 index 0000000..0a5ca2d --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 778b4088d62d52f49acf76d0c4d97922 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.glsl new file mode 100644 index 0000000..d37e0af --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.glsl @@ -0,0 +1,62 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +// FSR2 pass 1 +// SRV 1 : m_HDR : r_input_color_jittered +// SRV 4 : FSR2_Exposure : r_exposure +// UAV 7 : FSR2_ReconstructedPrevNearestDepth : rw_reconstructed_previous_nearest_depth +// UAV 13 : FSR2_PreparedInputColor : rw_prepared_input_color +// UAV 14 : FSR2_LumaHistory : rw_luma_history +// CB 0 : cbFSR2 + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_INPUT_COLOR 0 +#define FSR2_BIND_SRV_EXPOSURE 1 +#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 2 +#define FSR2_BIND_UAV_PREPARED_INPUT_COLOR 3 +#define FSR2_BIND_UAV_LUMA_HISTORY 4 +#define FSR2_BIND_CB_FSR2 5 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_prepare_input_color.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + PrepareInputColor(ivec2(gl_GlobalInvocationID.xy)); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.glsl.meta new file mode 100644 index 0000000..b4f7fb5 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 130af27aabe0f3347bf8fcfea8d5de84 +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.hlsl new file mode 100644 index 0000000..bed086f --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.hlsl @@ -0,0 +1,64 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 1 +// SRV 1 : m_HDR : r_input_color_jittered +// SRV 4 : FSR2_Exposure : r_exposure +// UAV 7 : FSR2_ReconstructedPrevNearestDepth : rw_reconstructed_previous_nearest_depth +// UAV 13 : FSR2_PreparedInputColor : rw_prepared_input_color +// UAV 14 : FSR2_LumaHistory : rw_luma_history +// CB 0 : cbFSR2 + +#define FSR2_BIND_SRV_INPUT_COLOR 0 +#define FSR2_BIND_SRV_EXPOSURE 1 +#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 0 +#define FSR2_BIND_UAV_PREPARED_INPUT_COLOR 1 +#define FSR2_BIND_UAV_LUMA_HISTORY 2 +#define FSR2_BIND_CB_FSR2 0 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_prepare_input_color.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_ROOTSIG_CONTENT +void CS( + uint2 uGroupId : SV_GroupID, + uint2 uDispatchThreadId : SV_DispatchThreadID, + uint2 uGroupThreadId : SV_GroupThreadID, + uint uGroupIndex : SV_GroupIndex +) +{ + PrepareInputColor(uDispatchThreadId); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.hlsl.meta new file mode 100644 index 0000000..1926d04 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_prepare_input_color_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 0710c8cc5e741b34882a94dddc5a1a6b +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_fsr2_rcas.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas.h similarity index 100% rename from Assets/Resources/Shaders/ffx_fsr2_rcas.h rename to Assets/Resources/FSR2/shaders/ffx_fsr2_rcas.h diff --git a/Assets/Resources/Shaders/ffx_fsr2_rcas.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_fsr2_rcas.h.meta rename to Assets/Resources/FSR2/shaders/ffx_fsr2_rcas.h.meta index 9418e79..5df489b 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_rcas.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 755780b5a13da4047a06cae98bbcc0de +guid: 7bd7d4eb34c626342966cb9b3fe00363 PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.glsl new file mode 100644 index 0000000..1097faf --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.glsl @@ -0,0 +1,92 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 6 +// SRV 4 : m_Exposure : r_exposure +// SRV 19 : FSR2_InternalUpscaled1 : r_rcas_input +// UAV 18 : DisplayOutput : rw_upscaled_output +// CB 0 : cbFSR2 +// CB 1 : cbRCAS + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_EXPOSURE 0 +#define FSR2_BIND_SRV_RCAS_INPUT 1 +#define FSR2_BIND_UAV_UPSCALED_OUTPUT 2 +#define FSR2_BIND_CB_FSR2 3 +#define FSR2_BIND_CB_RCAS 4 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" + +//Move to prototype shader! +#if defined(FSR2_BIND_CB_RCAS) + layout (set = 1, binding = FSR2_BIND_CB_RCAS, std140) uniform cbRCAS_t + { + uvec4 rcasConfig; + } cbRCAS; + + uvec4 RCASConfig() + { + return cbRCAS.rcasConfig; + } +#else + uvec4 RCASConfig() + { + return uvec4(0); + } +#endif + +#if FFX_HALF +vec4 LoadRCAS_Input(FfxInt16x2 iPxPos) +{ + return texelFetch(r_rcas_input, iPxPos, 0); +} +#else +vec4 LoadRCAS_Input(FfxInt32x2 iPxPos) +{ + return texelFetch(r_rcas_input, iPxPos, 0); +} +#endif + +#include "ffx_fsr2_rcas.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 64 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + RCAS(gl_LocalInvocationID.xyz, gl_WorkGroupID.xyz, gl_GlobalInvocationID.xyz); +} \ No newline at end of file diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.glsl.meta new file mode 100644 index 0000000..160f665 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 6e70536fee1272645bab8093b1c715e7 +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.hlsl similarity index 93% rename from Assets/Resources/Shaders/ffx_fsr2_rcas_pass.hlsl rename to Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.hlsl index 7138281..ea6b35a 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.hlsl +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.hlsl @@ -49,8 +49,7 @@ #else uint4 RCASConfig() { - // These are not your typical config parameters... these are packed floats-as-ints, i.e. complete nonsense values - return uint4(1061290752, 974141968, 0, 0); // TODO: this needs to be a constant buffer value + return 0; } #endif diff --git a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.hlsl.meta similarity index 75% rename from Assets/Resources/Shaders/ffx_fsr2_rcas_pass.hlsl.meta rename to Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.hlsl.meta index f67bad0..b7e089b 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_rcas_pass.hlsl.meta +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_rcas_pass.hlsl.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: 77fd29b4fc373da40881622c212f2505 +guid: 871ca1938c701d64f94ef8ec00ef06f4 ShaderIncludeImporter: externalObjects: {} userData: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h new file mode 100644 index 0000000..aad1992 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h @@ -0,0 +1,202 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H +#define FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H + +void ReconstructPrevDepth(FfxInt32x2 iPxPos, FfxFloat32 fDepth, FfxFloat32x2 fMotionVector, FfxInt32x2 iPxDepthSize) +{ + FfxFloat32x2 fDepthUv = (iPxPos + FfxFloat32(0.5)) / iPxDepthSize; + FfxFloat32x2 fPxPrevPos = (fDepthUv + fMotionVector) * iPxDepthSize - FfxFloat32x2(0.5, 0.5); + FfxInt32x2 iPxPrevPos = FfxInt32x2(floor(fPxPrevPos)); + FfxFloat32x2 fPxFrac = ffxFract(fPxPrevPos); + + const FfxFloat32 bilinearWeights[2][2] = { + { + (1 - fPxFrac.x) * (1 - fPxFrac.y), + (fPxFrac.x) * (1 - fPxFrac.y) + }, + { + (1 - fPxFrac.x) * (fPxFrac.y), + (fPxFrac.x) * (fPxFrac.y) + } + }; + + // Project current depth into previous frame locations. + // Push to all pixels having some contribution if reprojection is using bilinear logic. + for (FfxInt32 y = 0; y <= 1; ++y) { + for (FfxInt32 x = 0; x <= 1; ++x) { + + FfxInt32x2 offset = FfxInt32x2(x, y); + FfxFloat32 w = bilinearWeights[y][x]; + + if (w > reconstructedDepthBilinearWeightThreshold) { + + FfxInt32x2 storePos = iPxPrevPos + offset; + if (IsOnScreen(storePos, iPxDepthSize)) { + StoreReconstructedDepth(storePos, fDepth); + } + } + } + } +} + +void FindNearestDepth(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxInt32x2 iPxSize, FFX_PARAMETER_OUT FfxFloat32 fNearestDepth, FFX_PARAMETER_OUT FfxInt32x2 fNearestDepthCoord) +{ + const FfxInt32 iSampleCount = 9; + const FfxInt32x2 iSampleOffsets[iSampleCount] = { + FfxInt32x2(+0, +0), + FfxInt32x2(+1, +0), + FfxInt32x2(+0, +1), + FfxInt32x2(+0, -1), + FfxInt32x2(-1, +0), + FfxInt32x2(-1, +1), + FfxInt32x2(+1, +1), + FfxInt32x2(-1, -1), + FfxInt32x2(+1, -1), + }; + + // pull out the depth loads to allow SC to batch them + FfxFloat32 depth[9]; + FfxInt32 iSampleIndex = 0; + FFX_UNROLL + for (iSampleIndex = 0; iSampleIndex < iSampleCount; ++iSampleIndex) { + + FfxInt32x2 iPos = iPxPos + iSampleOffsets[iSampleIndex]; + depth[iSampleIndex] = LoadInputDepth(iPos); + } + + // find closest depth + fNearestDepthCoord = iPxPos; + fNearestDepth = depth[0]; + FFX_UNROLL + for (iSampleIndex = 1; iSampleIndex < iSampleCount; ++iSampleIndex) { + + FfxInt32x2 iPos = iPxPos + iSampleOffsets[iSampleIndex]; + if (IsOnScreen(iPos, iPxSize)) { + + FfxFloat32 fNdDepth = depth[iSampleIndex]; +#if FFX_FSR2_OPTION_INVERTED_DEPTH + if (fNdDepth > fNearestDepth) { +#else + if (fNdDepth < fNearestDepth) { +#endif + fNearestDepthCoord = iPos; + fNearestDepth = fNdDepth; + } + } + } +} + +FfxFloat32 ComputeMotionDivergence(FfxInt32x2 iPxPos, FfxInt32x2 iPxInputMotionVectorSize) +{ + FfxFloat32 minconvergence = 1.0f; + + FfxFloat32x2 fMotionVectorNucleus = LoadInputMotionVector(iPxPos) * RenderSize(); + FfxFloat32 fNucleusVelocity = length(fMotionVectorNucleus); + + const FfxFloat32 MotionVectorVelocityEpsilon = 1e-02f; + + if (fNucleusVelocity > MotionVectorVelocityEpsilon) { + for (FfxInt32 y = -1; y <= 1; ++y) { + for (FfxInt32 x = -1; x <= 1; ++x) { + + FfxInt32x2 sp = ClampLoad(iPxPos, FfxInt32x2(x, y), iPxInputMotionVectorSize); + + FfxFloat32x2 fMotionVector = LoadInputMotionVector(sp) * RenderSize(); + FfxFloat32 fVelocity = length(fMotionVector); + + fVelocity = ffxMax(fVelocity, fNucleusVelocity); + minconvergence = ffxMin(minconvergence, dot(fMotionVector / fVelocity, fMotionVectorNucleus / fVelocity)); + } + } + } + + return ffxSaturate(1.0f - minconvergence); +} + + +void PreProcessReactiveMasks(FfxInt32x2 iPxLrPos, FfxFloat32 fMotionDivergence) +{ + // Compensate for bilinear sampling in accumulation pass + + FfxFloat32x3 fReferenceColor = LoadPreparedInputColor(iPxLrPos); + FfxFloat32x2 fReactiveFactor = FfxFloat32x2(0.0f, fMotionDivergence); + + for (int y = -1; y < 2; y++) { + for (int x = -1; x < 2; x++) { + + const FfxInt32x2 sampleCoord = ClampLoad(iPxLrPos, FfxInt32x2(x, y), FfxInt32x2(RenderSize())); + + FfxFloat32x3 fColorSample = LoadPreparedInputColor(sampleCoord); + FfxFloat32 fReactiveSample = LoadReactiveMask(sampleCoord); + FfxFloat32 fTransparencyAndCompositionSample = LoadTransparencyAndCompositionMask(sampleCoord); + + const FfxFloat32 fColorSimilarity = dot(normalize(fReferenceColor), normalize(fColorSample)); + const FfxFloat32 fVelocitySimilarity = 1.0f - abs(length(fReferenceColor) - length(fColorSample)); + const FfxFloat32 fSimilarity = fColorSimilarity * fVelocitySimilarity; + + // Increase power for non-similar samples + const FfxFloat32 fPowerBiasMax = 6.0f; + const FfxFloat32 fSimilarityPower = 1.0f + (fPowerBiasMax - fSimilarity * fPowerBiasMax); + const FfxFloat32 fWeightedReactiveSample = ffxPow(fReactiveSample, fSimilarityPower); + const FfxFloat32 fWeightedTransparencyAndCompositionSample = ffxPow(fTransparencyAndCompositionSample, fSimilarityPower); + + fReactiveFactor = ffxMax(fReactiveFactor, FfxFloat32x2(fWeightedReactiveSample, fWeightedTransparencyAndCompositionSample)); + } + } + + StoreDilatedReactiveMasks(iPxLrPos, fReactiveFactor); +} + +void ReconstructAndDilate(FfxInt32x2 iPxLrPos) +{ + FfxFloat32 fDilatedDepth; + FfxInt32x2 iNearestDepthCoord; + + FindNearestDepth(iPxLrPos, RenderSize(), fDilatedDepth, iNearestDepthCoord); + +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS + FfxInt32x2 iSamplePos = iPxLrPos; + FfxInt32x2 iMotionVectorPos = iNearestDepthCoord; +#else + FfxInt32x2 iSamplePos = ComputeHrPosFromLrPos(iPxLrPos); + FfxInt32x2 iMotionVectorPos = ComputeHrPosFromLrPos(iNearestDepthCoord); +#endif + + FfxFloat32x2 fDilatedMotionVector = LoadInputMotionVector(iMotionVectorPos); + + StoreDilatedDepth(iPxLrPos, fDilatedDepth); + StoreDilatedMotionVector(iPxLrPos, fDilatedMotionVector); + + ReconstructPrevDepth(iPxLrPos, fDilatedDepth, fDilatedMotionVector, RenderSize()); + +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS + FfxFloat32 fMotionDivergence = ComputeMotionDivergence(iSamplePos, RenderSize()); +#else + FfxFloat32 fMotionDivergence = ComputeMotionDivergence(iSamplePos, DisplaySize()); +#endif + + PreProcessReactiveMasks(iPxLrPos, fMotionDivergence); +} + + +#endif //!defined( FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H ) diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h.meta new file mode 100644 index 0000000..9af47d5 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 13cd33c3d34a317409049dfd939e64ef +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl new file mode 100644 index 0000000..96d1383 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl @@ -0,0 +1,68 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 2 +// SRV 2 : m_MotionVector : r_motion_vectors +// SRV 3 : m_depthbuffer : r_depth +// UAV 7 : FSR2_ReconstructedPrevNearestDepth : rw_reconstructed_previous_nearest_depth +// UAV 8 : FSR2_DilatedVelocity : rw_dilated_motion_vectors +// UAV 9 : FSR2_DilatedDepth : rw_dilatedDepth +// CB 0 : cbFSR2 + +#version 450 + +#extension GL_GOOGLE_include_directive : require +#extension GL_EXT_samplerless_texture_functions : require + +#define FSR2_BIND_SRV_MOTION_VECTORS 0 +#define FSR2_BIND_SRV_DEPTH 1 +#define FSR2_BIND_SRV_REACTIVE_MASK 2 +#define FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK 3 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 4 +#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 5 +#define FSR2_BIND_UAV_DILATED_MOTION_VECTORS 6 +#define FSR2_BIND_UAV_DILATED_DEPTH 7 +#define FSR2_BIND_UAV_DILATED_REACTIVE_MASKS 8 +#define FSR2_BIND_CB_FSR2 9 + +#include "ffx_fsr2_callbacks_glsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in; +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_NUM_THREADS +void main() +{ + ReconstructAndDilate(FFX_MIN16_I2(gl_GlobalInvocationID.xy)); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl.meta new file mode 100644 index 0000000..b76404d --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: e3f9c69c87e772f4387b1bd3363a8688 +DefaultImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.hlsl b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.hlsl new file mode 100644 index 0000000..57f3f49 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.hlsl @@ -0,0 +1,70 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +// FSR2 pass 2 +// SRV 2 : m_MotionVector : r_motion_vectors +// SRV 3 : m_depthbuffer : r_depth +// UAV 7 : FSR2_ReconstructedPrevNearestDepth : rw_reconstructed_previous_nearest_depth +// UAV 8 : FSR2_DilatedVelocity : rw_dilated_motion_vectors +// UAV 9 : FSR2_DilatedDepth : rw_dilatedDepth +// CB 0 : cbFSR2 + +#define FSR2_BIND_SRV_MOTION_VECTORS 0 +#define FSR2_BIND_SRV_DEPTH 1 +#define FSR2_BIND_SRV_REACTIVE_MASK 2 +#define FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK 3 +#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 4 +#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 0 +#define FSR2_BIND_UAV_DILATED_MOTION_VECTORS 1 +#define FSR2_BIND_UAV_DILATED_DEPTH 2 +#define FSR2_BIND_UAV_DILATED_REACTIVE_MASKS 3 +#define FSR2_BIND_CB_FSR2 0 + +#include "ffx_fsr2_callbacks_hlsl.h" +#include "ffx_fsr2_common.h" +#include "ffx_fsr2_sample.h" +#include "ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h" + +#ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#define FFX_FSR2_THREAD_GROUP_WIDTH 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH +#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#define FFX_FSR2_THREAD_GROUP_HEIGHT 8 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT +#ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#define FFX_FSR2_THREAD_GROUP_DEPTH 1 +#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH +#ifndef FFX_FSR2_NUM_THREADS +#define FFX_FSR2_NUM_THREADS [numthreads(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT, FFX_FSR2_THREAD_GROUP_DEPTH)] +#endif // #ifndef FFX_FSR2_NUM_THREADS + +FFX_FSR2_PREFER_WAVE64 +FFX_FSR2_NUM_THREADS +FFX_FSR2_EMBED_ROOTSIG_CONTENT +void CS( + int2 iGroupId : SV_GroupID, + int2 iDispatchThreadId : SV_DispatchThreadID, + int2 iGroupThreadId : SV_GroupThreadID, + int iGroupIndex : SV_GroupIndex +) +{ + ReconstructAndDilate(iDispatchThreadId); +} diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.hlsl.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.hlsl.meta new file mode 100644 index 0000000..4f5036c --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.hlsl.meta @@ -0,0 +1,7 @@ +fileFormatVersion: 2 +guid: 9355c255c8505ae48ae89af286943747 +ShaderIncludeImporter: + externalObjects: {} + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reproject.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_reproject.h new file mode 100644 index 0000000..5ae962d --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reproject.h @@ -0,0 +1,125 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_REPROJECT_H +#define FFX_FSR2_REPROJECT_H + +#ifndef FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE +#define FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE 1 // Approximate +#endif + +FfxFloat32x4 WrapHistory(FfxInt32x2 iPxSample) +{ + return LoadHistory(iPxSample); +} + +#if FFX_HALF +FFX_MIN16_F4 WrapHistory(FFX_MIN16_I2 iPxSample) +{ + return FFX_MIN16_F4(LoadHistory(iPxSample)); +} +#endif + + +#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBicubicSamplesMin16(FetchHistorySamples, WrapHistory) +DeclareCustomTextureSampleMin16(HistorySample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchHistorySamples) +#else +DeclareCustomFetchBicubicSamples(FetchHistorySamples, WrapHistory) +DeclareCustomTextureSample(HistorySample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchHistorySamples) +#endif + +FfxFloat32x4 WrapLockStatus(FfxInt32x2 iPxSample) +{ + return FfxFloat32x4(LoadLockStatus(iPxSample), 0.0f); +} + +#if FFX_HALF +FFX_MIN16_F4 WrapLockStatus(FFX_MIN16_I2 iPxSample) +{ + return FFX_MIN16_F4(LoadLockStatus(iPxSample), 0.0f); +} +#endif + +#if 1 +#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBilinearSamplesMin16(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSampleMin16(LockStatusSample, Bilinear, FetchLockStatusSamples) +#else +DeclareCustomFetchBilinearSamples(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSample(LockStatusSample, Bilinear, FetchLockStatusSamples) +#endif +#else +#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF +DeclareCustomFetchBicubicSamplesMin16(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSampleMin16(LockStatusSample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchLockStatusSamples) +#else +DeclareCustomFetchBicubicSamples(FetchLockStatusSamples, WrapLockStatus) +DeclareCustomTextureSample(LockStatusSample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchLockStatusSamples) +#endif +#endif + +FfxFloat32x2 GetMotionVector(FfxInt32x2 iPxHrPos, FfxFloat32x2 fHrUv) +{ +#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS + FfxFloat32x2 fDilatedMotionVector = LoadDilatedMotionVector(FFX_MIN16_I2(fHrUv * RenderSize())); +#else + FfxFloat32x2 fDilatedMotionVector = LoadInputMotionVector(iPxHrPos); +#endif + + return fDilatedMotionVector; +} + +void ComputeReprojectedUVs(FfxInt32x2 iPxHrPos, FfxFloat32x2 fMotionVector, FFX_PARAMETER_OUT FfxFloat32x2 fReprojectedHrUv, FFX_PARAMETER_OUT FfxBoolean bIsExistingSample) +{ + FfxFloat32x2 fHrUv = (iPxHrPos + 0.5f) / DisplaySize(); + fReprojectedHrUv = fHrUv + fMotionVector; + + bIsExistingSample = (fReprojectedHrUv.x >= 0.0f && fReprojectedHrUv.x <= 1.0f) && + (fReprojectedHrUv.y >= 0.0f && fReprojectedHrUv.y <= 1.0f); +} + +void ReprojectHistoryColor(FfxInt32x2 iPxHrPos, FfxFloat32x2 fReprojectedHrUv, FFX_PARAMETER_OUT FfxFloat32x4 fHistoryColorAndWeight) +{ + fHistoryColorAndWeight = HistorySample(fReprojectedHrUv, DisplaySize()); + fHistoryColorAndWeight.rgb *= Exposure(); + +#if FFX_FSR2_OPTION_HDR_COLOR_INPUT + fHistoryColorAndWeight.rgb = Tonemap(fHistoryColorAndWeight.rgb); +#endif + + fHistoryColorAndWeight.rgb = RGBToYCoCg(fHistoryColorAndWeight.rgb); +} + +void ReprojectHistoryLockStatus(FfxInt32x2 iPxHrPos, FfxFloat32x2 fReprojectedHrUv, FFX_PARAMETER_OUT FfxFloat32x3 fReprojectedLockStatus) +{ + // If function is called from Accumulate pass, we need to treat locks differently + FfxFloat32 fInPlaceLockLifetime = LoadRwLockStatus(iPxHrPos)[LOCK_LIFETIME_REMAINING]; + + fReprojectedLockStatus = SampleLockStatus(fReprojectedHrUv); + + // Keep lifetime if new lock + if (fInPlaceLockLifetime < 0.0f) { + fReprojectedLockStatus[LOCK_LIFETIME_REMAINING] = fInPlaceLockLifetime; + } +} + +#endif //!defined( FFX_FSR2_REPROJECT_H ) diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_reproject.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_reproject.h.meta new file mode 100644 index 0000000..1071a12 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_reproject.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 63fc917ca6895cb4aac237ea35edb838 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/Shaders/ffx_fsr2_resources.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_resources.h similarity index 100% rename from Assets/Resources/Shaders/ffx_fsr2_resources.h rename to Assets/Resources/FSR2/shaders/ffx_fsr2_resources.h diff --git a/Assets/Resources/Shaders/ffx_fsr2_resources.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_resources.h.meta similarity index 92% rename from Assets/Resources/Shaders/ffx_fsr2_resources.h.meta rename to Assets/Resources/FSR2/shaders/ffx_fsr2_resources.h.meta index 27709d0..9ee0942 100644 --- a/Assets/Resources/Shaders/ffx_fsr2_resources.h.meta +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_resources.h.meta @@ -1,5 +1,5 @@ fileFormatVersion: 2 -guid: f54305f51c7d4314e8f18b6470ca188e +guid: fd5bed2bf4ba07444ae815390168a15d PluginImporter: externalObjects: {} serializedVersion: 2 diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_sample.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_sample.h new file mode 100644 index 0000000..cfa9db8 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_sample.h @@ -0,0 +1,602 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_SAMPLE_H +#define FFX_FSR2_SAMPLE_H + +// suppress warnings +#ifdef FFX_HLSL +#pragma warning(disable: 4008) // potentially divide by zero +#endif //FFX_HLSL + +struct FetchedBilinearSamples { + + FfxFloat32x4 fColor00; + FfxFloat32x4 fColor10; + + FfxFloat32x4 fColor01; + FfxFloat32x4 fColor11; +}; + +struct FetchedBicubicSamples { + + FfxFloat32x4 fColor00; + FfxFloat32x4 fColor10; + FfxFloat32x4 fColor20; + FfxFloat32x4 fColor30; + + FfxFloat32x4 fColor01; + FfxFloat32x4 fColor11; + FfxFloat32x4 fColor21; + FfxFloat32x4 fColor31; + + FfxFloat32x4 fColor02; + FfxFloat32x4 fColor12; + FfxFloat32x4 fColor22; + FfxFloat32x4 fColor32; + + FfxFloat32x4 fColor03; + FfxFloat32x4 fColor13; + FfxFloat32x4 fColor23; + FfxFloat32x4 fColor33; +}; + +#if FFX_HALF +struct FetchedBilinearSamplesMin16 { + + FFX_MIN16_F4 fColor00; + FFX_MIN16_F4 fColor10; + + FFX_MIN16_F4 fColor01; + FFX_MIN16_F4 fColor11; +}; + +struct FetchedBicubicSamplesMin16 { + + FFX_MIN16_F4 fColor00; + FFX_MIN16_F4 fColor10; + FFX_MIN16_F4 fColor20; + FFX_MIN16_F4 fColor30; + + FFX_MIN16_F4 fColor01; + FFX_MIN16_F4 fColor11; + FFX_MIN16_F4 fColor21; + FFX_MIN16_F4 fColor31; + + FFX_MIN16_F4 fColor02; + FFX_MIN16_F4 fColor12; + FFX_MIN16_F4 fColor22; + FFX_MIN16_F4 fColor32; + + FFX_MIN16_F4 fColor03; + FFX_MIN16_F4 fColor13; + FFX_MIN16_F4 fColor23; + FFX_MIN16_F4 fColor33; +}; +#else //FFX_HALF +#define FetchedBicubicSamplesMin16 FetchedBicubicSamples +#define FetchedBilinearSamplesMin16 FetchedBilinearSamples +#endif //FFX_HALF + +FfxFloat32x4 Linear(FfxFloat32x4 A, FfxFloat32x4 B, FfxFloat32 t) +{ + return A + (B - A) * t; +} + +FfxFloat32x4 Bilinear(FetchedBilinearSamples BilinearSamples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Linear(BilinearSamples.fColor00, BilinearSamples.fColor10, fPxFrac.x); + FfxFloat32x4 fColorX1 = Linear(BilinearSamples.fColor01, BilinearSamples.fColor11, fPxFrac.x); + FfxFloat32x4 fColorXY = Linear(fColorX0, fColorX1, fPxFrac.y); + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Linear(FFX_MIN16_F4 A, FFX_MIN16_F4 B, FFX_MIN16_F t) +{ + return A + (B - A) * t; +} + +FFX_MIN16_F4 Bilinear(FetchedBilinearSamplesMin16 BilinearSamples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Linear(BilinearSamples.fColor00, BilinearSamples.fColor10, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Linear(BilinearSamples.fColor01, BilinearSamples.fColor11, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Linear(fColorX0, fColorX1, fPxFrac.y); + return fColorXY; +} +#endif + +FfxFloat32 Lanczos2NoClamp(FfxFloat32 x) +{ + const FfxFloat32 PI = 3.141592653589793f; // TODO: share SDK constants + return abs(x) < FSR2_EPSILON ? 1.f : (sin(PI * x) / (PI * x)) * (sin(0.5f * PI * x) / (0.5f * PI * x)); +} + +FfxFloat32 Lanczos2(FfxFloat32 x) +{ + x = ffxMin(abs(x), 2.0f); + return Lanczos2NoClamp(x); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2NoClamp(FFX_MIN16_F x) +{ + const FFX_MIN16_F PI = FFX_MIN16_F(3.141592653589793f); // TODO: share SDK constants + return abs(x) < FFX_MIN16_F(FSR2_EPSILON) ? FFX_MIN16_F(1.f) : (sin(PI * x) / (PI * x)) * (sin(FFX_MIN16_F(0.5f) * PI * x) / (FFX_MIN16_F(0.5f) * PI * x)); +} + +FFX_MIN16_F Lanczos2(FFX_MIN16_F x) +{ + x = ffxMin(abs(x), FFX_MIN16_F(2.0f)); + return Lanczos2NoClamp(x); +} +#endif //FFX_HALF + +// FSR1 lanczos approximation. Input is x*x and must be <= 4. +FfxFloat32 Lanczos2ApproxSqNoClamp(FfxFloat32 x2) +{ + FfxFloat32 a = (2.0f / 5.0f) * x2 - 1; + FfxFloat32 b = (1.0f / 4.0f) * x2 - 1; + return ((25.0f / 16.0f) * a * a - (25.0f / 16.0f - 1)) * (b * b); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2ApproxSqNoClamp(FFX_MIN16_F x2) +{ + FFX_MIN16_F a = FFX_MIN16_F(2.0f / 5.0f) * x2 - FFX_MIN16_F(1); + FFX_MIN16_F b = FFX_MIN16_F(1.0f / 4.0f) * x2 - FFX_MIN16_F(1); + return (FFX_MIN16_F(25.0f / 16.0f) * a * a - FFX_MIN16_F(25.0f / 16.0f - 1)) * (b * b); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2ApproxSq(FfxFloat32 x2) +{ + x2 = ffxMin(x2, 4.0f); + return Lanczos2ApproxSqNoClamp(x2); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2ApproxSq(FFX_MIN16_F x2) +{ + x2 = ffxMin(x2, FFX_MIN16_F(4.0f)); + return Lanczos2ApproxSqNoClamp(x2); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2ApproxNoClamp(FfxFloat32 x) +{ + return Lanczos2ApproxSqNoClamp(x * x); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2ApproxNoClamp(FFX_MIN16_F x) +{ + return Lanczos2ApproxSqNoClamp(x * x); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2Approx(FfxFloat32 x) +{ + return Lanczos2ApproxSq(x * x); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2Approx(FFX_MIN16_F x) +{ + return Lanczos2ApproxSq(x * x); +} +#endif //FFX_HALF + +FfxFloat32 Lanczos2_UseLUT(FfxFloat32 x) +{ + return SampleLanczos2Weight(abs(x)); +} + +#if FFX_HALF +FFX_MIN16_F Lanczos2_UseLUT(FFX_MIN16_F x) +{ + return FFX_MIN16_F(SampleLanczos2Weight(abs(x))); +} +#endif //FFX_HALF + +FfxFloat32x4 Lanczos2_UseLUT(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t) +{ + FfxFloat32 fWeight0 = Lanczos2_UseLUT(-1.f - t); + FfxFloat32 fWeight1 = Lanczos2_UseLUT(-0.f - t); + FfxFloat32 fWeight2 = Lanczos2_UseLUT(+1.f - t); + FfxFloat32 fWeight3 = Lanczos2_UseLUT(+2.f - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} +#if FFX_HALF +FFX_MIN16_F4 Lanczos2_UseLUT(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t) +{ + FFX_MIN16_F fWeight0 = Lanczos2_UseLUT(FFX_MIN16_F(-1.f) - t); + FFX_MIN16_F fWeight1 = Lanczos2_UseLUT(FFX_MIN16_F(-0.f) - t); + FFX_MIN16_F fWeight2 = Lanczos2_UseLUT(FFX_MIN16_F(+1.f) - t); + FFX_MIN16_F fWeight3 = Lanczos2_UseLUT(FFX_MIN16_F(+2.f) - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} +#endif + +FfxFloat32x4 Lanczos2(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t) +{ + FfxFloat32 fWeight0 = Lanczos2(-1.f - t); + FfxFloat32 fWeight1 = Lanczos2(-0.f - t); + FfxFloat32 fWeight2 = Lanczos2(+1.f - t); + FfxFloat32 fWeight3 = Lanczos2(+2.f - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} + +FfxFloat32x4 Lanczos2(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Lanczos2(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FfxFloat32x4 fColorX1 = Lanczos2(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FfxFloat32x4 fColorX2 = Lanczos2(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FfxFloat32x4 fColorX3 = Lanczos2(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FfxFloat32x4 fColorXY = Lanczos2(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FfxFloat32x4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FfxFloat32x4 fDeringingMin = fDeringingSamples[0]; + FfxFloat32x4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) { + + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t) +{ + FFX_MIN16_F fWeight0 = Lanczos2(FFX_MIN16_F(-1.f) - t); + FFX_MIN16_F fWeight1 = Lanczos2(FFX_MIN16_F(-0.f) - t); + FFX_MIN16_F fWeight2 = Lanczos2(FFX_MIN16_F(+1.f) - t); + FFX_MIN16_F fWeight3 = Lanczos2(FFX_MIN16_F(+2.f) - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} + +FFX_MIN16_F4 Lanczos2(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Lanczos2(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Lanczos2(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FFX_MIN16_F4 fColorX2 = Lanczos2(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FFX_MIN16_F4 fColorX3 = Lanczos2(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Lanczos2(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FFX_MIN16_F4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0]; + FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} +#endif //FFX_HALF + + +FfxFloat32x4 Lanczos2LUT(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Lanczos2_UseLUT(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FfxFloat32x4 fColorX1 = Lanczos2_UseLUT(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FfxFloat32x4 fColorX2 = Lanczos2_UseLUT(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FfxFloat32x4 fColorX3 = Lanczos2_UseLUT(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FfxFloat32x4 fColorXY = Lanczos2_UseLUT(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FfxFloat32x4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FfxFloat32x4 fDeringingMin = fDeringingSamples[0]; + FfxFloat32x4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) { + + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2LUT(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Lanczos2_UseLUT(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Lanczos2_UseLUT(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FFX_MIN16_F4 fColorX2 = Lanczos2_UseLUT(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FFX_MIN16_F4 fColorX3 = Lanczos2_UseLUT(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Lanczos2_UseLUT(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FFX_MIN16_F4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0]; + FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} +#endif //FFX_HALF + + + +FfxFloat32x4 Lanczos2Approx(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t) +{ + FfxFloat32 fWeight0 = Lanczos2ApproxNoClamp(-1.f - t); + FfxFloat32 fWeight1 = Lanczos2ApproxNoClamp(-0.f - t); + FfxFloat32 fWeight2 = Lanczos2ApproxNoClamp(+1.f - t); + FfxFloat32 fWeight3 = Lanczos2ApproxNoClamp(+2.f - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2Approx(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t) +{ + FFX_MIN16_F fWeight0 = Lanczos2ApproxNoClamp(FFX_MIN16_F(-1.f) - t); + FFX_MIN16_F fWeight1 = Lanczos2ApproxNoClamp(FFX_MIN16_F(-0.f) - t); + FFX_MIN16_F fWeight2 = Lanczos2ApproxNoClamp(FFX_MIN16_F(+1.f) - t); + FFX_MIN16_F fWeight3 = Lanczos2ApproxNoClamp(FFX_MIN16_F(+2.f) - t); + return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3); +} +#endif //FFX_HALF + +FfxFloat32x4 Lanczos2Approx(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac) +{ + FfxFloat32x4 fColorX0 = Lanczos2Approx(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FfxFloat32x4 fColorX1 = Lanczos2Approx(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FfxFloat32x4 fColorX2 = Lanczos2Approx(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FfxFloat32x4 fColorX3 = Lanczos2Approx(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FfxFloat32x4 fColorXY = Lanczos2Approx(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FfxFloat32x4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FfxFloat32x4 fDeringingMin = fDeringingSamples[0]; + FfxFloat32x4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} + +#if FFX_HALF +FFX_MIN16_F4 Lanczos2Approx(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac) +{ + FFX_MIN16_F4 fColorX0 = Lanczos2Approx(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x); + FFX_MIN16_F4 fColorX1 = Lanczos2Approx(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x); + FFX_MIN16_F4 fColorX2 = Lanczos2Approx(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x); + FFX_MIN16_F4 fColorX3 = Lanczos2Approx(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x); + FFX_MIN16_F4 fColorXY = Lanczos2Approx(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y); + + // Deringing + + // TODO: only use 4 by checking jitter + const FfxInt32 iDeringingSampleCount = 4; + const FFX_MIN16_F4 fDeringingSamples[4] = { + Samples.fColor11, + Samples.fColor21, + Samples.fColor12, + Samples.fColor22, + }; + + FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0]; + FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0]; + + FFX_UNROLL + for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) + { + fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]); + fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]); + } + + fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax); + + return fColorXY; +} +#endif + +// Clamp by offset direction. Assuming iPxSample is already in range and iPxOffset is compile time constant. +FfxInt32x2 ClampCoord(FfxInt32x2 iPxSample, FfxInt32x2 iPxOffset, FfxInt32x2 iTextureSize) +{ + FfxInt32x2 result = iPxSample + iPxOffset; + result.x = (iPxOffset.x < 0) ? ffxMax(result.x, 0) : result.x; + result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - 1) : result.x; + result.y = (iPxOffset.y < 0) ? ffxMax(result.y, 0) : result.y; + result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - 1) : result.y; + return result; +} +#if FFX_HALF +FFX_MIN16_I2 ClampCoord(FFX_MIN16_I2 iPxSample, FFX_MIN16_I2 iPxOffset, FFX_MIN16_I2 iTextureSize) +{ + FFX_MIN16_I2 result = iPxSample + iPxOffset; + result.x = (iPxOffset.x < FFX_MIN16_I(0)) ? ffxMax(result.x, FFX_MIN16_I(0)) : result.x; + result.x = (iPxOffset.x > FFX_MIN16_I(0)) ? ffxMin(result.x, iTextureSize.x - FFX_MIN16_I(1)) : result.x; + result.y = (iPxOffset.y < FFX_MIN16_I(0)) ? ffxMax(result.y, FFX_MIN16_I(0)) : result.y; + result.y = (iPxOffset.y > FFX_MIN16_I(0)) ? ffxMin(result.y, iTextureSize.y - FFX_MIN16_I(1)) : result.y; + return result; +} +#endif //FFX_HALF + + +#define DeclareCustomFetchBicubicSamplesWithType(SampleType, TextureType, AddrType, Name, LoadTexture) \ + SampleType Name(AddrType iPxSample, AddrType iTextureSize) \ + { \ + SampleType Samples; \ + \ + Samples.fColor00 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, -1), iTextureSize))); \ + Samples.fColor10 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, -1), iTextureSize))); \ + Samples.fColor20 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, -1), iTextureSize))); \ + Samples.fColor30 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, -1), iTextureSize))); \ + \ + Samples.fColor01 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +0), iTextureSize))); \ + Samples.fColor11 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +0), iTextureSize))); \ + Samples.fColor21 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +0), iTextureSize))); \ + Samples.fColor31 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +0), iTextureSize))); \ + \ + Samples.fColor02 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +1), iTextureSize))); \ + Samples.fColor12 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +1), iTextureSize))); \ + Samples.fColor22 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +1), iTextureSize))); \ + Samples.fColor32 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +1), iTextureSize))); \ + \ + Samples.fColor03 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +2), iTextureSize))); \ + Samples.fColor13 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +2), iTextureSize))); \ + Samples.fColor23 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +2), iTextureSize))); \ + Samples.fColor33 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +2), iTextureSize))); \ + \ + return Samples; \ + } + +#define DeclareCustomFetchBicubicSamples(Name, LoadTexture) \ + DeclareCustomFetchBicubicSamplesWithType(FetchedBicubicSamples, FfxFloat32x4, FfxInt32x2, Name, LoadTexture) + +#define DeclareCustomFetchBicubicSamplesMin16(Name, LoadTexture) \ + DeclareCustomFetchBicubicSamplesWithType(FetchedBicubicSamplesMin16, FFX_MIN16_F4, FfxInt32x2, Name, LoadTexture) + +#define DeclareCustomFetchBilinearSamplesWithType(SampleType, TextureType,AddrType, Name, LoadTexture) \ + SampleType Name(AddrType iPxSample, AddrType iTextureSize) \ + { \ + SampleType Samples; \ + Samples.fColor00 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +0), iTextureSize))); \ + Samples.fColor10 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +0), iTextureSize))); \ + Samples.fColor01 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +1), iTextureSize))); \ + Samples.fColor11 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +1), iTextureSize))); \ + return Samples; \ + } + +#define DeclareCustomFetchBilinearSamples(Name, LoadTexture) \ + DeclareCustomFetchBilinearSamplesWithType(FetchedBilinearSamples, FfxFloat32x4, FfxInt32x2, Name, LoadTexture) + +#define DeclareCustomFetchBilinearSamplesMin16(Name, LoadTexture) \ + DeclareCustomFetchBilinearSamplesWithType(FetchedBilinearSamplesMin16, FFX_MIN16_F4, FfxInt32x2, Name, LoadTexture) + +// BE CAREFUL: there is some precision issues and (3253, 125) leading to (3252.9989778, 125.001102) +// is common, so iPxSample can "jitter" +#define DeclareCustomTextureSample(Name, InterpolateSamples, FetchSamples) \ + FfxFloat32x4 Name(FfxFloat32x2 fUvSample, FfxInt32x2 iTextureSize) \ + { \ + FfxFloat32x2 fPxSample = fUvSample * FfxFloat32x2(iTextureSize) - FfxFloat32x2(0.5f, 0.5f); \ + FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); \ + /* Clamp base coords */ \ + iPxSample.x = ffxMax(0, ffxMin(iPxSample.x, iTextureSize.x - 1)); \ + iPxSample.y = ffxMax(0, ffxMin(iPxSample.y, iTextureSize.y - 1)); \ + /* */ \ + FfxFloat32x2 fPxFrac = ffxFract(fPxSample); \ + FfxFloat32x4 fColorXY = FfxFloat32x4(InterpolateSamples(FetchSamples(iPxSample, iTextureSize), fPxFrac)); \ + return fColorXY; \ + } + +#define DeclareCustomTextureSampleMin16(Name, InterpolateSamples, FetchSamples) \ + FFX_MIN16_F4 Name(FfxFloat32x2 fUvSample, FfxInt32x2 iTextureSize) \ + { \ + FfxFloat32x2 fPxSample = fUvSample * FfxFloat32x2(iTextureSize) - FfxFloat32x2(0.5f, 0.5f); \ + FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); \ + /* Clamp base coords */ \ + iPxSample.x = ffxMax(0, ffxMin(iPxSample.x, iTextureSize.x - 1)); \ + iPxSample.y = ffxMax(0, ffxMin(iPxSample.y, iTextureSize.y - 1)); \ + /* */ \ + FFX_MIN16_F2 fPxFrac = FFX_MIN16_F2(ffxFract(fPxSample)); \ + FFX_MIN16_F4 fColorXY = FFX_MIN16_F4(InterpolateSamples(FetchSamples(iPxSample, iTextureSize), fPxFrac)); \ + return fColorXY; \ + } + +#define FFX_FSR2_CONCAT_ID(x, y) x ## y +#define FFX_FSR2_CONCAT(x, y) FFX_FSR2_CONCAT_ID(x, y) +#define FFX_FSR2_SAMPLER_1D_0 Lanczos2 +#define FFX_FSR2_SAMPLER_1D_1 Lanczos2LUT +#define FFX_FSR2_SAMPLER_1D_2 Lanczos2Approx + +#define FFX_FSR2_GET_LANCZOS_SAMPLER1D(x) FFX_FSR2_CONCAT(FFX_FSR2_SAMPLER_1D_, x) + +#endif //!defined( FFX_FSR2_SAMPLE_H ) diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_sample.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_sample.h.meta new file mode 100644 index 0000000..8977d12 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_sample.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: ba3ee190167b6c240aaeb1f8f4dbcb67 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_upsample.h b/Assets/Resources/FSR2/shaders/ffx_fsr2_upsample.h new file mode 100644 index 0000000..80524d4 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_upsample.h @@ -0,0 +1,214 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifndef FFX_FSR2_UPSAMPLE_H +#define FFX_FSR2_UPSAMPLE_H + +#define FFX_FSR2_OPTION_GUARANTEE_POSITIVE_UPSAMPLE_WEIGHT 0 + +FFX_STATIC const FfxUInt32 iLanczos2SampleCount = 16; + +void Deringing(RectificationBoxData clippingBox, FFX_PARAMETER_INOUT FfxFloat32x3 fColor) +{ + fColor = clamp(fColor, clippingBox.aabbMin, clippingBox.aabbMax); +} +#if FFX_HALF +void Deringing(RectificationBoxDataMin16 clippingBox, FFX_PARAMETER_INOUT FFX_MIN16_F3 fColor) +{ + fColor = clamp(fColor, clippingBox.aabbMin, clippingBox.aabbMax); +} +#endif + +#ifndef FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE +#define FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE 1 // Approximate +#endif + +FfxFloat32 GetUpsampleLanczosWeight(FfxFloat32x2 fSrcSampleOffset, FfxFloat32x2 fKernelWeight) +{ + FfxFloat32x2 fSrcSampleOffsetBiased = fSrcSampleOffset * fKernelWeight; +#if FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 0 // LANCZOS_TYPE_REFERENCE + FfxFloat32 fSampleWeight = Lanczos2(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 1 // LANCZOS_TYPE_LUT + FfxFloat32 fSampleWeight = Lanczos2_UseLUT(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 2 // LANCZOS_TYPE_APPROXIMATE + FfxFloat32 fSampleWeight = Lanczos2ApproxSq(dot(fSrcSampleOffsetBiased, fSrcSampleOffsetBiased)); +#else +#error "Invalid Lanczos type" +#endif + return fSampleWeight; +} + +#if FFX_HALF +FFX_MIN16_F GetUpsampleLanczosWeight(FFX_MIN16_F2 fSrcSampleOffset, FFX_MIN16_F2 fKernelWeight) +{ + FFX_MIN16_F2 fSrcSampleOffsetBiased = fSrcSampleOffset * fKernelWeight; +#if FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 0 // LANCZOS_TYPE_REFERENCE + FFX_MIN16_F fSampleWeight = Lanczos2(length(fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 1 // LANCZOS_TYPE_APPROXIMATE + FFX_MIN16_F fSampleWeight = Lanczos2ApproxSq(dot(fSrcSampleOffsetBiased, fSrcSampleOffsetBiased)); +#elif FFX_FSR2_OPTION_UPSAMPLE_USE_LANCZOS_TYPE == 2 // LANCZOS_TYPE_LUT + FFX_MIN16_F fSampleWeight = Lanczos2_UseLUT(length(fSrcSampleOffsetBiased)); + // To Test: Save reciproqual sqrt compute + // FfxFloat32 fSampleWeight = Lanczos2Sq_UseLUT(dot(fSrcSampleOffsetBiased, fSrcSampleOffsetBiased)); +#else +#error "Invalid Lanczos type" +#endif + return fSampleWeight; +} +#endif + +FfxFloat32 Pow3(FfxFloat32 x) +{ + return x * x * x; +} + +#if FX_HALF +FFX_MIN16_F Pow3(FFX_MIN16_F x) +{ + return x * x * x; +} +#endif + +FfxFloat32x4 ComputeUpsampledColorAndWeight(FfxInt32x2 iPxHrPos, FfxFloat32x2 fKernelWeight, FFX_PARAMETER_INOUT RectificationBoxData clippingBox) +{ +#if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF +#include "ffx_fsr2_force16_begin.h" +#endif + // We compute a sliced lanczos filter with 2 lobes (other slices are accumulated temporaly) + FfxFloat32x2 fDstOutputPos = FfxFloat32x2(iPxHrPos) + FFX_BROADCAST_FLOAT32X2(0.5f); // Destination resolution output pixel center position + FfxFloat32x2 fSrcOutputPos = fDstOutputPos * DownscaleFactor(); // Source resolution output pixel center position + FfxInt32x2 iSrcInputPos = FfxInt32x2(floor(fSrcOutputPos)); // TODO: what about weird upscale factors... + +#if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF +#include "ffx_fsr2_force16_end.h" +#endif + +#if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF +#include "ffx_fsr2_force16_begin.h" + RectificationBoxMin16 fRectificationBox; +#else + RectificationBox fRectificationBox; +#endif + + FfxFloat32x3 fSamples[iLanczos2SampleCount]; + + + FfxFloat32x2 fSrcUnjitteredPos = (FfxFloat32x2(iSrcInputPos) + FfxFloat32x2(0.5f, 0.5f)) - Jitter(); // This is the un-jittered position of the sample at offset 0,0 + + FfxInt32x2 offsetTL; + offsetTL.x = (fSrcUnjitteredPos.x > fSrcOutputPos.x) ? FfxInt32(-2) : FfxInt32(-1); + offsetTL.y = (fSrcUnjitteredPos.y > fSrcOutputPos.y) ? FfxInt32(-2) : FfxInt32(-1); + + //Load samples + // If fSrcUnjitteredPos.y > fSrcOutputPos.y, indicates offsetTL.y = -2, sample offset Y will be [-2, 1], clipbox will be rows [1, 3]. + // Flip row# for sampling offset in this case, so first 0~2 rows in the sampled array can always be used for computing the clipbox. + // This reduces branch or cmove on sampled colors, but moving this overhead to sample position / weight calculation time which apply to less values. + const FfxBoolean bFlipRow = fSrcUnjitteredPos.y > fSrcOutputPos.y; + const FfxBoolean bFlipCol = fSrcUnjitteredPos.x > fSrcOutputPos.x; + + FfxFloat32x2 fOffsetTL = FfxFloat32x2(offsetTL); + + FFX_UNROLL + for (FfxInt32 row = 0; row < 4; row++) { + + FFX_UNROLL + for (FfxInt32 col = 0; col < 4; col++) { + FfxInt32 iSampleIndex = col + (row << 2); + + FfxInt32x2 sampleColRow = FfxInt32x2(bFlipCol ? (3 - col) : col, bFlipRow ? (3 - row) : row); + FfxInt32x2 iSrcSamplePos = FfxInt32x2(iSrcInputPos) + offsetTL + sampleColRow; + + const FfxInt32x2 sampleCoord = ClampLoad(iSrcSamplePos, FfxInt32x2(0, 0), FfxInt32x2(RenderSize())); + + fSamples[iSampleIndex] = LoadPreparedInputColor(FfxInt32x2(sampleCoord)); + } + } + + RectificationBoxReset(fRectificationBox, fSamples[0]); + + FfxFloat32x3 fColor = FfxFloat32x3(0.f, 0.f, 0.f); + FfxFloat32 fWeight = FfxFloat32(0.f); + FfxFloat32x2 fBaseSampleOffset = FfxFloat32x2(fSrcUnjitteredPos - fSrcOutputPos); + + FFX_UNROLL + for (FfxInt32 row = 0; row < 3; row++) { + + FFX_UNROLL + for (FfxInt32 col = 0; col < 3; col++) { + FfxInt32 iSampleIndex = col + (row << 2); + + const FfxInt32x2 sampleColRow = FfxInt32x2(bFlipCol ? (3 - col) : col, bFlipRow ? (3 - row) : row); + const FfxFloat32x2 fOffset = fOffsetTL + FfxFloat32x2(sampleColRow); + FfxFloat32x2 fSrcSampleOffset = fBaseSampleOffset + fOffset; + + FfxInt32x2 iSrcSamplePos = FfxInt32x2(iSrcInputPos) + FfxInt32x2(offsetTL) + sampleColRow; + + FfxFloat32 fSampleWeight = FfxFloat32(IsOnScreen(FfxInt32x2(iSrcSamplePos), FfxInt32x2(RenderSize()))) * GetUpsampleLanczosWeight(fSrcSampleOffset, fKernelWeight); + + // Update rectification box + const FfxFloat32 fSrcSampleOffsetSq = dot(fSrcSampleOffset, fSrcSampleOffset); + FfxFloat32 fBoxSampleWeight = FfxFloat32(1) - ffxSaturate(fSrcSampleOffsetSq / FfxFloat32(3)); + fBoxSampleWeight *= fBoxSampleWeight; + RectificationBoxAddSample(fRectificationBox, fSamples[iSampleIndex], fBoxSampleWeight); + + fWeight += fSampleWeight; + fColor += fSampleWeight * fSamples[iSampleIndex]; + } + } + // Normalize for deringing (we need to compare colors) + fColor = fColor / (abs(fWeight) > FSR2_EPSILON ? fWeight : FfxFloat32(1.f)); + + RectificationBoxComputeVarianceBoxData(fRectificationBox); +#if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF + RectificationBoxDataMin16 rectificationData = RectificationBoxGetData(fRectificationBox); + clippingBox.aabbMax = rectificationData.aabbMax; + clippingBox.aabbMin = rectificationData.aabbMin; + clippingBox.boxCenter = rectificationData.boxCenter; + clippingBox.boxVec = rectificationData.boxVec; +#else + RectificationBoxData rectificationData = RectificationBoxGetData(fRectificationBox); + clippingBox = rectificationData; +#endif + + Deringing(rectificationData, fColor); + +#if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF + clippingBox.aabbMax = rectificationData.aabbMax; + clippingBox.aabbMin = rectificationData.aabbMin; + clippingBox.boxCenter = rectificationData.boxCenter; + clippingBox.boxVec = rectificationData.boxVec; +#endif + + if (any(FFX_LESS_THAN(fKernelWeight, FfxFloat32x2(1, 1)))) { + fWeight = FfxFloat32(averageLanczosWeightPerFrame); + } +#if FFX_FSR2_OPTION_UPSAMPLE_SAMPLERS_USE_DATA_HALF && FFX_HALF +#include "ffx_fsr2_force16_end.h" +#endif + +#if FFX_FSR2_OPTION_GUARANTEE_POSITIVE_UPSAMPLE_WEIGHT + return FfxFloat32x4(fColor, ffxMax(FfxFloat32(FSR2_EPSILON), fWeight)); +#else + return FfxFloat32x4(fColor, ffxMax(FfxFloat32(0), fWeight)); +#endif +} + +#endif //!defined( FFX_FSR2_UPSAMPLE_H ) diff --git a/Assets/Resources/FSR2/shaders/ffx_fsr2_upsample.h.meta b/Assets/Resources/FSR2/shaders/ffx_fsr2_upsample.h.meta new file mode 100644 index 0000000..778092e --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_fsr2_upsample.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 1ff3a385cfe07db4387e4d7b457238f8 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Resources/FSR2/shaders/ffx_spd.h b/Assets/Resources/FSR2/shaders/ffx_spd.h new file mode 100644 index 0000000..5a27a84 --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_spd.h @@ -0,0 +1,936 @@ +// This file is part of the FidelityFX SDK. +// +// Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#ifdef FFX_CPU +FFX_STATIC void SpdSetup(FfxUInt32x2 dispatchThreadGroupCountXY, // CPU side: dispatch thread group count xy + FfxUInt32x2 workGroupOffset, // GPU side: pass in as constant + FfxUInt32x2 numWorkGroupsAndMips, // GPU side: pass in as constant + FfxUInt32x4 rectInfo, // left, top, width, height + FfxInt32 mips) // optional: if -1, calculate based on rect width and height +{ + workGroupOffset[0] = rectInfo[0] / 64; // rectInfo[0] = left + workGroupOffset[1] = rectInfo[1] / 64; // rectInfo[1] = top + + FfxUInt32 endIndexX = (rectInfo[0] + rectInfo[2] - 1) / 64; // rectInfo[0] = left, rectInfo[2] = width + FfxUInt32 endIndexY = (rectInfo[1] + rectInfo[3] - 1) / 64; // rectInfo[1] = top, rectInfo[3] = height + + dispatchThreadGroupCountXY[0] = endIndexX + 1 - workGroupOffset[0]; + dispatchThreadGroupCountXY[1] = endIndexY + 1 - workGroupOffset[1]; + + numWorkGroupsAndMips[0] = (dispatchThreadGroupCountXY[0]) * (dispatchThreadGroupCountXY[1]); + + if (mips >= 0) + { + numWorkGroupsAndMips[1] = FfxUInt32(mips); + } + else + { + // calculate based on rect width and height + FfxUInt32 resolution = ffxMax(rectInfo[2], rectInfo[3]); + numWorkGroupsAndMips[1] = FfxUInt32((ffxMin(floor(log2(FfxFloat32(resolution))), FfxFloat32(12)))); + } +} + +FFX_STATIC void SpdSetup(FfxUInt32x2 dispatchThreadGroupCountXY, // CPU side: dispatch thread group count xy + FfxUInt32x2 workGroupOffset, // GPU side: pass in as constant + FfxUInt32x2 numWorkGroupsAndMips, // GPU side: pass in as constant + FfxUInt32x4 rectInfo) // left, top, width, height +{ + SpdSetup(dispatchThreadGroupCountXY, workGroupOffset, numWorkGroupsAndMips, rectInfo, -1); +} +#endif // #ifdef FFX_CPU + + +//============================================================================================================================== +// NON-PACKED VERSION +//============================================================================================================================== +#ifdef FFX_GPU +#ifdef SPD_PACKED_ONLY +// Avoid compiler error +FfxFloat32x4 SpdLoadSourceImage(FfxInt32x2 p, FfxUInt32 slice) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} + +FfxFloat32x4 SpdLoad(FfxInt32x2 p, FfxUInt32 slice) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} +void SpdStore(FfxInt32x2 p, FfxFloat32x4 value, FfxUInt32 mip, FfxUInt32 slice) +{ +} +FfxFloat32x4 SpdLoadIntermediate(FfxUInt32 x, FfxUInt32 y) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} +void SpdStoreIntermediate(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value) +{ +} +FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3) +{ + return FfxFloat32x4(0.0, 0.0, 0.0, 0.0); +} +#endif // #ifdef SPD_PACKED_ONLY + +//_____________________________________________________________/\_______________________________________________________________ +#if defined(FFX_GLSL) && !defined(SPD_NO_WAVE_OPERATIONS) +#extension GL_KHR_shader_subgroup_quad:require +#endif + +void SpdWorkgroupShuffleBarrier() +{ +#ifdef FFX_GLSL + barrier(); +#endif +#ifdef FFX_HLSL + GroupMemoryBarrierWithGroupSync(); +#endif +} + +// Only last active workgroup should proceed +bool SpdExitWorkgroup(FfxUInt32 numWorkGroups, FfxUInt32 localInvocationIndex, FfxUInt32 slice) +{ + // global atomic counter + if (localInvocationIndex == 0) + { + SpdIncreaseAtomicCounter(slice); + } + + SpdWorkgroupShuffleBarrier(); + return (SpdGetAtomicCounter() != (numWorkGroups - 1)); +} + +// User defined: FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3); +FfxFloat32x4 SpdReduceQuad(FfxFloat32x4 v) +{ +#if defined(FFX_GLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + + FfxFloat32x4 v0 = v; + FfxFloat32x4 v1 = subgroupQuadSwapHorizontal(v); + FfxFloat32x4 v2 = subgroupQuadSwapVertical(v); + FfxFloat32x4 v3 = subgroupQuadSwapDiagonal(v); + return SpdReduce4(v0, v1, v2, v3); + +#elif defined(FFX_HLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + + // requires SM6.0 + FfxUInt32 quad = WaveGetLaneIndex() & (~0x3); + FfxFloat32x4 v0 = v; + FfxFloat32x4 v1 = WaveReadLaneAt(v, quad | 1); + FfxFloat32x4 v2 = WaveReadLaneAt(v, quad | 2); + FfxFloat32x4 v3 = WaveReadLaneAt(v, quad | 3); + return SpdReduce4(v0, v1, v2, v3); +/* + // if SM6.0 is not available, you can use the AMD shader intrinsics + // the AMD shader intrinsics are available in AMD GPU Services (AGS) library: + // https://gpuopen.com/amd-gpu-services-ags-library/ + // works for DX11 + FfxFloat32x4 v0 = v; + FfxFloat32x4 v1; + v1.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + FfxFloat32x4 v2; + v2.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + FfxFloat32x4 v3; + v3.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + return SpdReduce4(v0, v1, v2, v3); + */ +#endif + return v; +} + +FfxFloat32x4 SpdReduceIntermediate(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3) +{ + FfxFloat32x4 v0 = SpdLoadIntermediate(i0.x, i0.y); + FfxFloat32x4 v1 = SpdLoadIntermediate(i1.x, i1.y); + FfxFloat32x4 v2 = SpdLoadIntermediate(i2.x, i2.y); + FfxFloat32x4 v3 = SpdLoadIntermediate(i3.x, i3.y); + return SpdReduce4(v0, v1, v2, v3); +} + +FfxFloat32x4 SpdReduceLoad4(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat32x4 v0 = SpdLoad(FfxInt32x2(i0), slice); + FfxFloat32x4 v1 = SpdLoad(FfxInt32x2(i1), slice); + FfxFloat32x4 v2 = SpdLoad(FfxInt32x2(i2), slice); + FfxFloat32x4 v3 = SpdLoad(FfxInt32x2(i3), slice); + return SpdReduce4(v0, v1, v2, v3); +} + +FfxFloat32x4 SpdReduceLoad4(FfxUInt32x2 base, FfxUInt32 slice) +{ + return SpdReduceLoad4(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +} + +FfxFloat32x4 SpdReduceLoadSourceImage4(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat32x4 v0 = SpdLoadSourceImage(FfxInt32x2(i0), slice); + FfxFloat32x4 v1 = SpdLoadSourceImage(FfxInt32x2(i1), slice); + FfxFloat32x4 v2 = SpdLoadSourceImage(FfxInt32x2(i2), slice); + FfxFloat32x4 v3 = SpdLoadSourceImage(FfxInt32x2(i3), slice); + return SpdReduce4(v0, v1, v2, v3); +} + +FfxFloat32x4 SpdReduceLoadSourceImage(FfxUInt32x2 base, FfxUInt32 slice) +{ +#ifdef SPD_LINEAR_SAMPLER + return SpdLoadSourceImage(FfxInt32x2(base), slice); +#else + return SpdReduceLoadSourceImage4(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +#endif +} + +void SpdDownsampleMips_0_1_Intrinsics(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ + FfxFloat32x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[3], 0, slice); + + if (mip <= 1) + return; + + v[0] = SpdReduceQuad(v[0]); + v[1] = SpdReduceQuad(v[1]); + v[2] = SpdReduceQuad(v[2]); + v[3] = SpdReduceQuad(v[3]); + + if ((localInvocationIndex % 4) == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2), v[0], 1, slice); + SpdStoreIntermediate(x / 2, y / 2, v[0]); + + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2), v[1], 1, slice); + SpdStoreIntermediate(x / 2 + 8, y / 2, v[1]); + + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2 + 8), v[2], 1, slice); + SpdStoreIntermediate(x / 2, y / 2 + 8, v[2]); + + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2 + 8), v[3], 1, slice); + SpdStoreIntermediate(x / 2 + 8, y / 2 + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1_LDS(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ + FfxFloat32x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImage(tex, slice); + SpdStore(pix, v[3], 0, slice); + + if (mip <= 1) + return; + + for (FfxUInt32 i = 0; i < 4; i++) + { + SpdStoreIntermediate(x, y, v[i]); + SpdWorkgroupShuffleBarrier(); + if (localInvocationIndex < 64) + { + v[i] = SpdReduceIntermediate(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStore(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x + (i % 2) * 8, y + (i / 2) * 8), v[i], 1, slice); + } + SpdWorkgroupShuffleBarrier(); + } + + if (localInvocationIndex < 64) + { + SpdStoreIntermediate(x + 0, y + 0, v[0]); + SpdStoreIntermediate(x + 8, y + 0, v[1]); + SpdStoreIntermediate(x + 0, y + 8, v[2]); + SpdStoreIntermediate(x + 8, y + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + SpdDownsampleMips_0_1_LDS(x, y, workGroupID, localInvocationIndex, mip, slice); +#else + SpdDownsampleMips_0_1_Intrinsics(x, y, workGroupID, localInvocationIndex, mip, slice); +#endif +} + + +void SpdDownsampleMip_2(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 64) + { + FfxFloat32x4 v = SpdReduceIntermediate(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStore(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS, try to reduce bank conflicts + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 x + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // ... + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + SpdStoreIntermediate(x * 2 + y % 2, y * 2, v); + } +#else + FfxFloat32x4 v = SpdLoadIntermediate(x, y); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediate(x + (y / 2) % 2, y, v); + } +#endif +} + +void SpdDownsampleMip_3(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 16) + { + // x 0 x 0 + // 0 0 0 0 + // 0 x 0 x + // 0 0 0 0 + FfxFloat32x4 v = + SpdReduceIntermediate(FfxUInt32x2(x * 4 + 0 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 2 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 0 + 1, y * 4 + 2), FfxUInt32x2(x * 4 + 2 + 1, y * 4 + 2)); + SpdStore(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 + // ... + // 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 + // ... + // 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x + // ... + SpdStoreIntermediate(x * 4 + y, y * 4, v); + } +#else + if (localInvocationIndex < 64) + { + FfxFloat32x4 v = SpdLoadIntermediate(x * 2 + y % 2, y * 2); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediate(x * 2 + y / 2, y * 2, v); + } + } +#endif +} + +void SpdDownsampleMip_4(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 4) + { + // x 0 0 0 x 0 0 0 + // ... + // 0 x 0 0 0 x 0 0 + FfxFloat32x4 v = SpdReduceIntermediate(FfxUInt32x2(x * 8 + 0 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 4 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 0 + 1 + y * 2, y * 8 + 4), + FfxUInt32x2(x * 8 + 4 + 1 + y * 2, y * 8 + 4)); + SpdStore(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x x x x 0 ... + // 0 ... + SpdStoreIntermediate(x + y * 2, 0, v); + } +#else + if (localInvocationIndex < 16) + { + FfxFloat32x4 v = SpdLoadIntermediate(x * 4 + y, y * 4); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediate(x / 2 + y, 0, v); + } + } +#endif +} + +void SpdDownsampleMip_5(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 1) + { + // x x x x 0 ... + // 0 ... + FfxFloat32x4 v = SpdReduceIntermediate(FfxUInt32x2(0, 0), FfxUInt32x2(1, 0), FfxUInt32x2(2, 0), FfxUInt32x2(3, 0)); + SpdStore(FfxInt32x2(workGroupID.xy), v, mip, slice); + } +#else + if (localInvocationIndex < 4) + { + FfxFloat32x4 v = SpdLoadIntermediate(localInvocationIndex, 0); + v = SpdReduceQuad(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStore(FfxInt32x2(workGroupID.xy), v, mip, slice); + } + } +#endif +} + +void SpdDownsampleMips_6_7(FfxUInt32 x, FfxUInt32 y, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxInt32x2 tex = FfxInt32x2(x * 4 + 0, y * 4 + 0); + FfxInt32x2 pix = FfxInt32x2(x * 2 + 0, y * 2 + 0); + FfxFloat32x4 v0 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v0, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 0); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 0); + FfxFloat32x4 v1 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v1, 6, slice); + + tex = FfxInt32x2(x * 4 + 0, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 0, y * 2 + 1); + FfxFloat32x4 v2 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v2, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 1); + FfxFloat32x4 v3 = SpdReduceLoad4(tex, slice); + SpdStore(pix, v3, 6, slice); + + if (mips <= 7) + return; + // no barrier needed, working on values only from the same thread + + FfxFloat32x4 v = SpdReduce4(v0, v1, v2, v3); + SpdStore(FfxInt32x2(x, y), v, 7, slice); + SpdStoreIntermediate(x, y, v); +} + +void SpdDownsampleNextFour(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 baseMip, FfxUInt32 mips, FfxUInt32 slice) +{ + if (mips <= baseMip) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_2(x, y, workGroupID, localInvocationIndex, baseMip, slice); + + if (mips <= baseMip + 1) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_3(x, y, workGroupID, localInvocationIndex, baseMip + 1, slice); + + if (mips <= baseMip + 2) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_4(x, y, workGroupID, localInvocationIndex, baseMip + 2, slice); + + if (mips <= baseMip + 3) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_5(workGroupID, localInvocationIndex, baseMip + 3, slice); +} + +void SpdDownsample(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice) +{ + FfxUInt32x2 sub_xy = ffxRemapForWaveReduction(localInvocationIndex % 64); + FfxUInt32 x = sub_xy.x + 8 * ((localInvocationIndex >> 6) % 2); + FfxUInt32 y = sub_xy.y + 8 * ((localInvocationIndex >> 7)); + SpdDownsampleMips_0_1(x, y, workGroupID, localInvocationIndex, mips, slice); + + SpdDownsampleNextFour(x, y, workGroupID, localInvocationIndex, 2, mips, slice); + + if (mips <= 6) + return; + + if (SpdExitWorkgroup(numWorkGroups, localInvocationIndex, slice)) + return; + + SpdResetAtomicCounter(slice); + + // After mip 6 there is only a single workgroup left that downsamples the remaining up to 64x64 texels. + SpdDownsampleMips_6_7(x, y, mips, slice); + + SpdDownsampleNextFour(x, y, FfxUInt32x2(0, 0), localInvocationIndex, 8, mips, slice); +} + +void SpdDownsample(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice, FfxUInt32x2 workGroupOffset) +{ + SpdDownsample(workGroupID + workGroupOffset, localInvocationIndex, mips, numWorkGroups, slice); +} + +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +//============================================================================================================================== +// PACKED VERSION +//============================================================================================================================== + +#if FFX_HALF + +#ifdef FFX_GLSL +#extension GL_EXT_shader_subgroup_extended_types_float16:require +#endif + +FfxFloat16x4 SpdReduceQuadH(FfxFloat16x4 v) +{ +#if defined(FFX_GLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + FfxFloat16x4 v0 = v; + FfxFloat16x4 v1 = subgroupQuadSwapHorizontal(v); + FfxFloat16x4 v2 = subgroupQuadSwapVertical(v); + FfxFloat16x4 v3 = subgroupQuadSwapDiagonal(v); + return SpdReduce4H(v0, v1, v2, v3); +#elif defined(FFX_HLSL) && !defined(SPD_NO_WAVE_OPERATIONS) + // requires SM6.0 + FfxUInt32 quad = WaveGetLaneIndex() & (~0x3); + FfxFloat16x4 v0 = v; + FfxFloat16x4 v1 = WaveReadLaneAt(v, quad | 1); + FfxFloat16x4 v2 = WaveReadLaneAt(v, quad | 2); + FfxFloat16x4 v3 = WaveReadLaneAt(v, quad | 3); + return SpdReduce4H(v0, v1, v2, v3); +/* + // if SM6.0 is not available, you can use the AMD shader intrinsics + // the AMD shader intrinsics are available in AMD GPU Services (AGS) library: + // https://gpuopen.com/amd-gpu-services-ags-library/ + // works for DX11 + FfxFloat16x4 v0 = v; + FfxFloat16x4 v1; + v1.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + v1.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX1); + FfxFloat16x4 v2; + v2.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + v2.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_SwapX2); + FfxFloat16x4 v3; + v3.x = AmdExtD3DShaderIntrinsics_SwizzleF(v.x, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.y = AmdExtD3DShaderIntrinsics_SwizzleF(v.y, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.z = AmdExtD3DShaderIntrinsics_SwizzleF(v.z, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + v3.w = AmdExtD3DShaderIntrinsics_SwizzleF(v.w, AmdExtD3DShaderIntrinsicsSwizzle_ReverseX4); + return SpdReduce4H(v0, v1, v2, v3); + */ +#endif + return FfxFloat16x4(0.0, 0.0, 0.0, 0.0); +} + +FfxFloat16x4 SpdReduceIntermediateH(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3) +{ + FfxFloat16x4 v0 = SpdLoadIntermediateH(i0.x, i0.y); + FfxFloat16x4 v1 = SpdLoadIntermediateH(i1.x, i1.y); + FfxFloat16x4 v2 = SpdLoadIntermediateH(i2.x, i2.y); + FfxFloat16x4 v3 = SpdLoadIntermediateH(i3.x, i3.y); + return SpdReduce4H(v0, v1, v2, v3); +} + +FfxFloat16x4 SpdReduceLoad4H(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat16x4 v0 = SpdLoadH(FfxInt32x2(i0), slice); + FfxFloat16x4 v1 = SpdLoadH(FfxInt32x2(i1), slice); + FfxFloat16x4 v2 = SpdLoadH(FfxInt32x2(i2), slice); + FfxFloat16x4 v3 = SpdLoadH(FfxInt32x2(i3), slice); + return SpdReduce4H(v0, v1, v2, v3); +} + +FfxFloat16x4 SpdReduceLoad4H(FfxUInt32x2 base, FfxUInt32 slice) +{ + return SpdReduceLoad4H(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +} + +FfxFloat16x4 SpdReduceLoadSourceImage4H(FfxUInt32x2 i0, FfxUInt32x2 i1, FfxUInt32x2 i2, FfxUInt32x2 i3, FfxUInt32 slice) +{ + FfxFloat16x4 v0 = SpdLoadSourceImageH(FfxInt32x2(i0), slice); + FfxFloat16x4 v1 = SpdLoadSourceImageH(FfxInt32x2(i1), slice); + FfxFloat16x4 v2 = SpdLoadSourceImageH(FfxInt32x2(i2), slice); + FfxFloat16x4 v3 = SpdLoadSourceImageH(FfxInt32x2(i3), slice); + return SpdReduce4H(v0, v1, v2, v3); +} + +FfxFloat16x4 SpdReduceLoadSourceImageH(FfxUInt32x2 base, FfxUInt32 slice) +{ +#ifdef SPD_LINEAR_SAMPLER + return SpdLoadSourceImageH(FfxInt32x2(base), slice); +#else + return SpdReduceLoadSourceImage4H(FfxUInt32x2(base + FfxUInt32x2(0, 0)), FfxUInt32x2(base + FfxUInt32x2(0, 1)), FfxUInt32x2(base + FfxUInt32x2(1, 0)), FfxUInt32x2(base + FfxUInt32x2(1, 1)), slice); +#endif +} + +void SpdDownsampleMips_0_1_IntrinsicsH(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxFloat16x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[3], 0, slice); + + if (mips <= 1) + return; + + v[0] = SpdReduceQuadH(v[0]); + v[1] = SpdReduceQuadH(v[1]); + v[2] = SpdReduceQuadH(v[2]); + v[3] = SpdReduceQuadH(v[3]); + + if ((localInvocationIndex % 4) == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2), v[0], 1, slice); + SpdStoreIntermediateH(x / 2, y / 2, v[0]); + + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2), v[1], 1, slice); + SpdStoreIntermediateH(x / 2 + 8, y / 2, v[1]); + + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2, y / 2 + 8), v[2], 1, slice); + SpdStoreIntermediateH(x / 2, y / 2 + 8, v[2]); + + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x / 2 + 8, y / 2 + 8), v[3], 1, slice); + SpdStoreIntermediateH(x / 2 + 8, y / 2 + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1_LDSH(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxFloat16x4 v[4]; + + FfxInt32x2 tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2); + FfxInt32x2 pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y); + v[0] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[0], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y); + v[1] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[1], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x, y + 16); + v[2] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[2], 0, slice); + + tex = FfxInt32x2(workGroupID.xy * 64) + FfxInt32x2(x * 2 + 32, y * 2 + 32); + pix = FfxInt32x2(workGroupID.xy * 32) + FfxInt32x2(x + 16, y + 16); + v[3] = SpdReduceLoadSourceImageH(tex, slice); + SpdStoreH(pix, v[3], 0, slice); + + if (mips <= 1) + return; + + for (FfxInt32 i = 0; i < 4; i++) + { + SpdStoreIntermediateH(x, y, v[i]); + SpdWorkgroupShuffleBarrier(); + if (localInvocationIndex < 64) + { + v[i] = SpdReduceIntermediateH(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 16) + FfxInt32x2(x + (i % 2) * 8, y + (i / 2) * 8), v[i], 1, slice); + } + SpdWorkgroupShuffleBarrier(); + } + + if (localInvocationIndex < 64) + { + SpdStoreIntermediateH(x + 0, y + 0, v[0]); + SpdStoreIntermediateH(x + 8, y + 0, v[1]); + SpdStoreIntermediateH(x + 0, y + 8, v[2]); + SpdStoreIntermediateH(x + 8, y + 8, v[3]); + } +} + +void SpdDownsampleMips_0_1H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + SpdDownsampleMips_0_1_LDSH(x, y, workGroupID, localInvocationIndex, mips, slice); +#else + SpdDownsampleMips_0_1_IntrinsicsH(x, y, workGroupID, localInvocationIndex, mips, slice); +#endif +} + + +void SpdDownsampleMip_2H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 64) + { + FfxFloat16x4 v = SpdReduceIntermediateH(FfxUInt32x2(x * 2 + 0, y * 2 + 0), FfxUInt32x2(x * 2 + 1, y * 2 + 0), FfxUInt32x2(x * 2 + 0, y * 2 + 1), FfxUInt32x2(x * 2 + 1, y * 2 + 1)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS, try to reduce bank conflicts + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 x + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + // ... + // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 + SpdStoreIntermediateH(x * 2 + y % 2, y * 2, v); + } +#else + FfxFloat16x4 v = SpdLoadIntermediateH(x, y); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 8) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediateH(x + (y / 2) % 2, y, v); + } +#endif +} + +void SpdDownsampleMip_3H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 16) + { + // x 0 x 0 + // 0 0 0 0 + // 0 x 0 x + // 0 0 0 0 + FfxFloat16x4 v = + SpdReduceIntermediateH(FfxUInt32x2(x * 4 + 0 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 2 + 0, y * 4 + 0), FfxUInt32x2(x * 4 + 0 + 1, y * 4 + 2), FfxUInt32x2(x * 4 + 2 + 1, y * 4 + 2)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 + // 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 + // ... + // 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 + // ... + // 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x + // ... + SpdStoreIntermediateH(x * 4 + y, y * 4, v); + } +#else + if (localInvocationIndex < 64) + { + FfxFloat16x4 v = SpdLoadIntermediateH(x * 2 + y % 2, y * 2); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 4) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediateH(x * 2 + y / 2, y * 2, v); + } + } +#endif +} + +void SpdDownsampleMip_4H(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 4) + { + // x 0 0 0 x 0 0 0 + // ... + // 0 x 0 0 0 x 0 0 + FfxFloat16x4 v = SpdReduceIntermediateH(FfxUInt32x2(x * 8 + 0 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 4 + 0 + y * 2, y * 8 + 0), + FfxUInt32x2(x * 8 + 0 + 1 + y * 2, y * 8 + 4), + FfxUInt32x2(x * 8 + 4 + 1 + y * 2, y * 8 + 4)); + SpdStoreH(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x, y), v, mip, slice); + // store to LDS + // x x x x 0 ... + // 0 ... + SpdStoreIntermediateH(x + y * 2, 0, v); + } +#else + if (localInvocationIndex < 16) + { + FfxFloat16x4 v = SpdLoadIntermediateH(x * 4 + y, y * 4); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy * 2) + FfxInt32x2(x / 2, y / 2), v, mip, slice); + SpdStoreIntermediateH(x / 2 + y, 0, v); + } + } +#endif +} + +void SpdDownsampleMip_5H(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mip, FfxUInt32 slice) +{ +#ifdef SPD_NO_WAVE_OPERATIONS + if (localInvocationIndex < 1) + { + // x x x x 0 ... + // 0 ... + FfxFloat16x4 v = SpdReduceIntermediateH(FfxUInt32x2(0, 0), FfxUInt32x2(1, 0), FfxUInt32x2(2, 0), FfxUInt32x2(3, 0)); + SpdStoreH(FfxInt32x2(workGroupID.xy), v, mip, slice); + } +#else + if (localInvocationIndex < 4) + { + FfxFloat16x4 v = SpdLoadIntermediateH(localInvocationIndex, 0); + v = SpdReduceQuadH(v); + // quad index 0 stores result + if (localInvocationIndex % 4 == 0) + { + SpdStoreH(FfxInt32x2(workGroupID.xy), v, mip, slice); + } + } +#endif +} + +void SpdDownsampleMips_6_7H(FfxUInt32 x, FfxUInt32 y, FfxUInt32 mips, FfxUInt32 slice) +{ + FfxInt32x2 tex = FfxInt32x2(x * 4 + 0, y * 4 + 0); + FfxInt32x2 pix = FfxInt32x2(x * 2 + 0, y * 2 + 0); + FfxFloat16x4 v0 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v0, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 0); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 0); + FfxFloat16x4 v1 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v1, 6, slice); + + tex = FfxInt32x2(x * 4 + 0, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 0, y * 2 + 1); + FfxFloat16x4 v2 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v2, 6, slice); + + tex = FfxInt32x2(x * 4 + 2, y * 4 + 2); + pix = FfxInt32x2(x * 2 + 1, y * 2 + 1); + FfxFloat16x4 v3 = SpdReduceLoad4H(tex, slice); + SpdStoreH(pix, v3, 6, slice); + + if (mips < 8) + return; + // no barrier needed, working on values only from the same thread + + FfxFloat16x4 v = SpdReduce4H(v0, v1, v2, v3); + SpdStoreH(FfxInt32x2(x, y), v, 7, slice); + SpdStoreIntermediateH(x, y, v); +} + +void SpdDownsampleNextFourH(FfxUInt32 x, FfxUInt32 y, FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 baseMip, FfxUInt32 mips, FfxUInt32 slice) +{ + if (mips <= baseMip) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_2H(x, y, workGroupID, localInvocationIndex, baseMip, slice); + + if (mips <= baseMip + 1) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_3H(x, y, workGroupID, localInvocationIndex, baseMip + 1, slice); + + if (mips <= baseMip + 2) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_4H(x, y, workGroupID, localInvocationIndex, baseMip + 2, slice); + + if (mips <= baseMip + 3) + return; + SpdWorkgroupShuffleBarrier(); + SpdDownsampleMip_5H(workGroupID, localInvocationIndex, baseMip + 3, slice); +} + +void SpdDownsampleH(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice) +{ + FfxUInt32x2 sub_xy = ffxRemapForWaveReduction(localInvocationIndex % 64); + FfxUInt32 x = sub_xy.x + 8 * ((localInvocationIndex >> 6) % 2); + FfxUInt32 y = sub_xy.y + 8 * ((localInvocationIndex >> 7)); + + SpdDownsampleMips_0_1H(x, y, workGroupID, localInvocationIndex, mips, slice); + + SpdDownsampleNextFourH(x, y, workGroupID, localInvocationIndex, 2, mips, slice); + + if (mips < 7) + return; + + if (SpdExitWorkgroup(numWorkGroups, localInvocationIndex, slice)) + return; + + SpdResetAtomicCounter(slice); + + // After mip 6 there is only a single workgroup left that downsamples the remaining up to 64x64 texels. + SpdDownsampleMips_6_7H(x, y, mips, slice); + + SpdDownsampleNextFourH(x, y, FfxUInt32x2(0, 0), localInvocationIndex, 8, mips, slice); +} + +void SpdDownsampleH(FfxUInt32x2 workGroupID, FfxUInt32 localInvocationIndex, FfxUInt32 mips, FfxUInt32 numWorkGroups, FfxUInt32 slice, FfxUInt32x2 workGroupOffset) +{ + SpdDownsampleH(workGroupID + workGroupOffset, localInvocationIndex, mips, numWorkGroups, slice); +} + +#endif // #if FFX_HALF +#endif // #ifdef FFX_GPU diff --git a/Assets/Resources/FSR2/shaders/ffx_spd.h.meta b/Assets/Resources/FSR2/shaders/ffx_spd.h.meta new file mode 100644 index 0000000..e17851b --- /dev/null +++ b/Assets/Resources/FSR2/shaders/ffx_spd.h.meta @@ -0,0 +1,27 @@ +fileFormatVersion: 2 +guid: 3ef69a900a925bb498651c10581e0979 +PluginImporter: + externalObjects: {} + serializedVersion: 2 + iconMap: {} + executionOrder: {} + defineConstraints: [] + isPreloaded: 0 + isOverridable: 0 + isExplicitlyReferenced: 0 + validateReferences: 1 + platformData: + - first: + Any: + second: + enabled: 1 + settings: {} + - first: + Editor: Editor + second: + enabled: 0 + settings: + DefaultValueInitialized: true + userData: + assetBundleName: + assetBundleVariant: diff --git a/Assets/Scripts/FSR2Thing.cs b/Assets/Scripts/FSR2Thing.cs index 90fb35f..a3ee6e6 100644 --- a/Assets/Scripts/FSR2Thing.cs +++ b/Assets/Scripts/FSR2Thing.cs @@ -52,7 +52,7 @@ public class FSR2Thing : MonoBehaviour { // TODO: this is nasty, I don't like this. How do we manage/bind compute shaders best? => make a Callbacks class/interface with overridable implementation // GPUInstancer used Resources, we modified that to load stuff from asset bundles instead. Maybe provide a custom loader callback interface? - _rcasComputeShader = Resources.Load("Shaders/ffx_fsr2_rcas_pass"); + _rcasComputeShader = Resources.Load("FSR2/ffx_fsr2_rcas_pass"); } return _rcasComputeShader;