virtualx-engine/thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip.h

258 lines
11 KiB
C++
Raw Normal View History

// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 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 ComputeDepthClip(FfxFloat32x2 fUvSample, FfxFloat32 fCurrentDepthSample)
{
FfxFloat32 fCurrentDepthViewSpace = GetViewSpaceDepth(fCurrentDepthSample);
BilinearSamplingData bilinearInfo = GetBilinearSamplingData(fUvSample, RenderSize());
FfxFloat32 fDilatedSum = 0.0f;
FfxFloat32 fDepth = 0.0f;
FfxFloat32 fWeightSum = 0.0f;
for (FfxInt32 iSampleIndex = 0; iSampleIndex < 4; iSampleIndex++) {
const FfxInt32x2 iOffset = bilinearInfo.iOffsets[iSampleIndex];
const FfxInt32x2 iSamplePos = bilinearInfo.iBasePos + iOffset;
if (IsOnScreen(iSamplePos, RenderSize())) {
const FfxFloat32 fWeight = bilinearInfo.fWeights[iSampleIndex];
if (fWeight > fReconstructedDepthBilinearWeightThreshold) {
const FfxFloat32 fPrevDepthSample = LoadReconstructedPrevDepth(iSamplePos);
const FfxFloat32 fPrevNearestDepthViewSpace = GetViewSpaceDepth(fPrevDepthSample);
const FfxFloat32 fDepthDiff = fCurrentDepthViewSpace - fPrevNearestDepthViewSpace;
if (fDepthDiff > 0.0f) {
#if FFX_FSR2_OPTION_INVERTED_DEPTH
const FfxFloat32 fPlaneDepth = ffxMin(fPrevDepthSample, fCurrentDepthSample);
#else
const FfxFloat32 fPlaneDepth = ffxMax(fPrevDepthSample, fCurrentDepthSample);
#endif
const FfxFloat32x3 fCenter = GetViewSpacePosition(FfxInt32x2(RenderSize() * 0.5f), RenderSize(), fPlaneDepth);
const FfxFloat32x3 fCorner = GetViewSpacePosition(FfxInt32x2(0, 0), RenderSize(), fPlaneDepth);
const FfxFloat32 fHalfViewportWidth = length(FfxFloat32x2(RenderSize()));
const FfxFloat32 fDepthThreshold = ffxMax(fCurrentDepthViewSpace, fPrevNearestDepthViewSpace);
const FfxFloat32 Ksep = 1.37e-05f;
const FfxFloat32 Kfov = length(fCorner) / length(fCenter);
const FfxFloat32 fRequiredDepthSeparation = Ksep * Kfov * fHalfViewportWidth * fDepthThreshold;
const FfxFloat32 fResolutionFactor = ffxSaturate(length(FfxFloat32x2(RenderSize())) / length(FfxFloat32x2(1920.0f, 1080.0f)));
const FfxFloat32 fPower = ffxLerp(1.0f, 3.0f, fResolutionFactor);
fDepth += ffxPow(ffxSaturate(FfxFloat32(fRequiredDepthSeparation / fDepthDiff)), fPower) * fWeight;
fWeightSum += fWeight;
}
}
}
}
return (fWeightSum > 0) ? ffxSaturate(1.0f - fDepth / fWeightSum) : 0.0f;
}
FfxFloat32 ComputeMotionDivergence(FfxInt32x2 iPxPos, FfxInt32x2 iPxInputMotionVectorSize)
{
FfxFloat32 minconvergence = 1.0f;
FfxFloat32x2 fMotionVectorNucleus = LoadInputMotionVector(iPxPos);
FfxFloat32 fNucleusVelocityLr = length(fMotionVectorNucleus * RenderSize());
FfxFloat32 fMaxVelocityUv = length(fMotionVectorNucleus);
const FfxFloat32 MotionVectorVelocityEpsilon = 1e-02f;
if (fNucleusVelocityLr > 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);
FfxFloat32 fVelocityUv = length(fMotionVector);
fMaxVelocityUv = ffxMax(fVelocityUv, fMaxVelocityUv);
fVelocityUv = ffxMax(fVelocityUv, fMaxVelocityUv);
minconvergence = ffxMin(minconvergence, dot(fMotionVector / fVelocityUv, fMotionVectorNucleus / fVelocityUv));
}
}
}
return ffxSaturate(1.0f - minconvergence) * ffxSaturate(fMaxVelocityUv / 0.01f);
}
FfxFloat32 ComputeDepthDivergence(FfxInt32x2 iPxPos)
{
const FfxFloat32 fMaxDistInMeters = GetMaxDistanceInMeters();
FfxFloat32 fDepthMax = 0.0f;
FfxFloat32 fDepthMin = fMaxDistInMeters;
FfxInt32 iMaxDistFound = 0;
for (FfxInt32 y = -1; y < 2; y++) {
for (FfxInt32 x = -1; x < 2; x++) {
const FfxInt32x2 iOffset = FfxInt32x2(x, y);
const FfxInt32x2 iSamplePos = iPxPos + iOffset;
const FfxFloat32 fOnScreenFactor = IsOnScreen(iSamplePos, RenderSize()) ? 1.0f : 0.0f;
FfxFloat32 fDepth = GetViewSpaceDepthInMeters(LoadDilatedDepth(iSamplePos)) * fOnScreenFactor;
iMaxDistFound |= FfxInt32(fMaxDistInMeters == fDepth);
fDepthMin = ffxMin(fDepthMin, fDepth);
fDepthMax = ffxMax(fDepthMax, fDepth);
}
}
return (1.0f - fDepthMin / fDepthMax) * (FfxBoolean(iMaxDistFound) ? 0.0f : 1.0f);
}
FfxFloat32 ComputeTemporalMotionDivergence(FfxInt32x2 iPxPos)
{
const FfxFloat32x2 fUv = FfxFloat32x2(iPxPos + 0.5f) / RenderSize();
FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos);
FfxFloat32x2 fReprojectedUv = fUv + fMotionVector;
fReprojectedUv = ClampUv(fReprojectedUv, RenderSize(), MaxRenderSize());
FfxFloat32x2 fPrevMotionVector = SamplePreviousDilatedMotionVector(fReprojectedUv);
float fPxDistance = length(fMotionVector * DisplaySize());
return fPxDistance > 1.0f ? ffxLerp(0.0f, 1.0f - ffxSaturate(length(fPrevMotionVector) / length(fMotionVector)), ffxSaturate(ffxPow(fPxDistance / 20.0f, 3.0f))) : 0;
}
void PreProcessReactiveMasks(FfxInt32x2 iPxLrPos, FfxFloat32 fMotionDivergence)
{
// Compensate for bilinear sampling in accumulation pass
FfxFloat32x3 fReferenceColor = LoadInputColor(iPxLrPos).xyz;
FfxFloat32x2 fReactiveFactor = FfxFloat32x2(0.0f, fMotionDivergence);
float fMasksSum = 0.0f;
FfxFloat32x3 fColorSamples[9];
FfxFloat32 fReactiveSamples[9];
FfxFloat32 fTransparencyAndCompositionSamples[9];
FFX_UNROLL
for (FfxInt32 y = -1; y < 2; y++) {
FFX_UNROLL
for (FfxInt32 x = -1; x < 2; x++) {
const FfxInt32x2 sampleCoord = ClampLoad(iPxLrPos, FfxInt32x2(x, y), FfxInt32x2(RenderSize()));
FfxInt32 sampleIdx = (y + 1) * 3 + x + 1;
FfxFloat32x3 fColorSample = LoadInputColor(sampleCoord).xyz;
FfxFloat32 fReactiveSample = LoadReactiveMask(sampleCoord);
FfxFloat32 fTransparencyAndCompositionSample = LoadTransparencyAndCompositionMask(sampleCoord);
fColorSamples[sampleIdx] = fColorSample;
fReactiveSamples[sampleIdx] = fReactiveSample;
fTransparencyAndCompositionSamples[sampleIdx] = fTransparencyAndCompositionSample;
fMasksSum += (fReactiveSample + fTransparencyAndCompositionSample);
}
}
if (fMasksSum > 0)
{
for (FfxInt32 sampleIdx = 0; sampleIdx < 9; sampleIdx++)
{
FfxFloat32x3 fColorSample = fColorSamples[sampleIdx];
FfxFloat32 fReactiveSample = fReactiveSamples[sampleIdx];
FfxFloat32 fTransparencyAndCompositionSample = fTransparencyAndCompositionSamples[sampleIdx];
const FfxFloat32 fMaxLenSq = ffxMax(dot(fReferenceColor, fReferenceColor), dot(fColorSample, fColorSample));
const FfxFloat32 fSimilarity = dot(fReferenceColor, fColorSample) / fMaxLenSq;
// 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);
}
FfxFloat32x3 ComputePreparedInputColor(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 = PrepareRgb(fRgb, Exposure(), PreExposure());
const FfxFloat32x3 fPreparedYCoCg = RGBToYCoCg(fRgb);
return fPreparedYCoCg;
}
FfxFloat32 EvaluateSurface(FfxInt32x2 iPxPos, FfxFloat32x2 fMotionVector)
{
FfxFloat32 d0 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, -1)));
FfxFloat32 d1 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, 0)));
FfxFloat32 d2 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, 1)));
return 1.0f - FfxFloat32(((d0 - d1) > (d1 * 0.01f)) && ((d1 - d2) > (d2 * 0.01f)));
}
void DepthClip(FfxInt32x2 iPxPos)
{
FfxFloat32x2 fDepthUv = (iPxPos + 0.5f) / RenderSize();
FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos);
// Discard tiny mvs
fMotionVector *= FfxFloat32(length(fMotionVector * DisplaySize()) > 0.01f);
const FfxFloat32x2 fDilatedUv = fDepthUv + fMotionVector;
const FfxFloat32 fDilatedDepth = LoadDilatedDepth(iPxPos);
const FfxFloat32 fCurrentDepthViewSpace = GetViewSpaceDepth(LoadInputDepth(iPxPos));
// Compute prepared input color and depth clip
FfxFloat32 fDepthClip = ComputeDepthClip(fDilatedUv, fDilatedDepth) * EvaluateSurface(iPxPos, fMotionVector);
FfxFloat32x3 fPreparedYCoCg = ComputePreparedInputColor(iPxPos);
StorePreparedInputColor(iPxPos, FfxFloat32x4(fPreparedYCoCg, fDepthClip));
// Compute dilated reactive mask
#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
FfxInt32x2 iSamplePos = iPxPos;
#else
FfxInt32x2 iSamplePos = ComputeHrPosFromLrPos(iPxPos);
#endif
FfxFloat32 fMotionDivergence = ComputeMotionDivergence(iSamplePos, RenderSize());
FfxFloat32 fTemporalMotionDifference = ffxSaturate(ComputeTemporalMotionDivergence(iPxPos) - ComputeDepthDivergence(iPxPos));
PreProcessReactiveMasks(iPxPos, ffxMax(fTemporalMotionDifference, fMotionDivergence));
}
#endif //!defined( FFX_FSR2_DEPTH_CLIPH )