virtualx-engine/thirdparty/cvtt/ConvectionKernels_BCCommon.h
K. S. Ernest (iFire) Lee 419b342a9a Faster CVTT by reducing quality.
Make BC6 and BC7 CVTT faster while still having better quality than DXT5.
2022-02-04 15:15:26 -08:00

104 lines
4.5 KiB
C++

#pragma once
#ifndef __CVTT_BCCOMMON_H__
#define __CVTT_BCCOMMON_H__
#include "ConvectionKernels_AggregatedError.h"
#include "ConvectionKernels_ParallelMath.h"
namespace cvtt
{
namespace Internal
{
class BCCommon
{
public:
typedef ParallelMath::Float MFloat;
typedef ParallelMath::UInt16 MUInt16;
typedef ParallelMath::UInt15 MUInt15;
typedef ParallelMath::AInt16 MAInt16;
typedef ParallelMath::SInt16 MSInt16;
typedef ParallelMath::SInt32 MSInt32;
static int TweakRoundsForRange(int range);
template<int TVectorSize>
static void ComputeErrorLDR(uint32_t flags, const MUInt15 reconstructed[TVectorSize], const MUInt15 original[TVectorSize], int numRealChannels, AggregatedError<TVectorSize> &aggError)
{
for (int ch = 0; ch < numRealChannels; ch++)
aggError.Add(ParallelMath::SqDiffUInt8(reconstructed[ch], original[ch]), ch);
}
template<int TVectorSize>
static void ComputeErrorLDR(uint32_t flags, const MUInt15 reconstructed[TVectorSize], const MUInt15 original[TVectorSize], AggregatedError<TVectorSize> &aggError)
{
ComputeErrorLDR<TVectorSize>(flags, reconstructed, original, TVectorSize, aggError);
}
template<int TVectorSize>
static MFloat ComputeErrorLDRSimple(uint32_t flags, const MUInt15 reconstructed[TVectorSize], const MUInt15 original[TVectorSize], int numRealChannels, const float *channelWeightsSq)
{
AggregatedError<TVectorSize> aggError;
ComputeErrorLDR<TVectorSize>(flags, reconstructed, original, numRealChannels, aggError);
return aggError.Finalize(flags, channelWeightsSq);
}
template<int TVectorSize>
static MFloat ComputeErrorHDRFast(uint32_t flags, const MSInt16 reconstructed[TVectorSize], const MSInt16 original[TVectorSize], const float channelWeightsSq[TVectorSize])
{
MFloat error = ParallelMath::MakeFloatZero();
if (flags & Flags::Uniform)
{
for (int ch = 0; ch < TVectorSize; ch++)
error = error + ParallelMath::SqDiffSInt16(reconstructed[ch], original[ch]);
}
else
{
for (int ch = 0; ch < TVectorSize; ch++)
error = error + ParallelMath::SqDiffSInt16(reconstructed[ch], original[ch]) * ParallelMath::MakeFloat(channelWeightsSq[ch]);
}
return error;
}
template<int TVectorSize>
static MFloat ComputeErrorHDRSlow(uint32_t flags, const MSInt16 reconstructed[TVectorSize], const MSInt16 original[TVectorSize], const float channelWeightsSq[TVectorSize])
{
MFloat error = ParallelMath::MakeFloatZero();
if (flags & Flags::Uniform)
{
for (int ch = 0; ch < TVectorSize; ch++)
error = error + ParallelMath::SqDiff2CL(reconstructed[ch], original[ch]);
}
else
{
for (int ch = 0; ch < TVectorSize; ch++)
error = error + ParallelMath::SqDiff2CL(reconstructed[ch], original[ch]) * ParallelMath::MakeFloat(channelWeightsSq[ch]);
}
return error;
}
template<int TChannelCount>
static void PreWeightPixelsLDR(MFloat preWeightedPixels[16][TChannelCount], const MUInt15 pixels[16][TChannelCount], const float channelWeights[TChannelCount])
{
for (int px = 0; px < 16; px++)
{
for (int ch = 0; ch < TChannelCount; ch++)
preWeightedPixels[px][ch] = ParallelMath::ToFloat(pixels[px][ch]) * channelWeights[ch];
}
}
template<int TChannelCount>
static void PreWeightPixelsHDR(MFloat preWeightedPixels[16][TChannelCount], const MSInt16 pixels[16][TChannelCount], const float channelWeights[TChannelCount])
{
for (int px = 0; px < 16; px++)
{
for (int ch = 0; ch < TChannelCount; ch++)
preWeightedPixels[px][ch] = ParallelMath::ToFloat(pixels[px][ch]) * channelWeights[ch];
}
}
};
}
}
#endif