godot/thirdparty/cvtt/ConvectionKernels_UnfinishedEndpoints.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

122 lines
4.5 KiB
C++

#pragma once
#include "ConvectionKernels_Util.h"
namespace cvtt
{
namespace Internal
{
template<int TVectorSize>
class UnfinishedEndpoints
{
public:
typedef ParallelMath::Float MFloat;
typedef ParallelMath::UInt16 MUInt16;
typedef ParallelMath::UInt15 MUInt15;
typedef ParallelMath::SInt16 MSInt16;
typedef ParallelMath::SInt32 MSInt32;
UnfinishedEndpoints()
{
}
UnfinishedEndpoints(const MFloat *base, const MFloat *offset)
{
for (int ch = 0; ch < TVectorSize; ch++)
m_base[ch] = base[ch];
for (int ch = 0; ch < TVectorSize; ch++)
m_offset[ch] = offset[ch];
}
UnfinishedEndpoints(const UnfinishedEndpoints& other)
{
for (int ch = 0; ch < TVectorSize; ch++)
m_base[ch] = other.m_base[ch];
for (int ch = 0; ch < TVectorSize; ch++)
m_offset[ch] = other.m_offset[ch];
}
void FinishHDRUnsigned(int tweak, int range, MSInt16 *outEP0, MSInt16 *outEP1, ParallelMath::RoundTowardNearestForScope *roundingMode)
{
float tweakFactors[2];
Util::ComputeTweakFactors(tweak, range, tweakFactors);
for (int ch = 0; ch < TVectorSize; ch++)
{
MUInt15 channelEPs[2];
for (int epi = 0; epi < 2; epi++)
{
MFloat f = ParallelMath::Clamp(m_base[ch] + m_offset[ch] * tweakFactors[epi], 0.0f, 31743.0f);
channelEPs[epi] = ParallelMath::RoundAndConvertToU15(f, roundingMode);
}
outEP0[ch] = ParallelMath::LosslessCast<MSInt16>::Cast(channelEPs[0]);
outEP1[ch] = ParallelMath::LosslessCast<MSInt16>::Cast(channelEPs[1]);
}
}
void FinishHDRSigned(int tweak, int range, MSInt16* outEP0, MSInt16* outEP1, ParallelMath::RoundTowardNearestForScope* roundingMode)
{
float tweakFactors[2];
Util::ComputeTweakFactors(tweak, range, tweakFactors);
for (int ch = 0; ch < TVectorSize; ch++)
{
MSInt16 channelEPs[2];
for (int epi = 0; epi < 2; epi++)
{
MFloat f = ParallelMath::Clamp(m_base[ch] + m_offset[ch] * tweakFactors[epi], -31743.0f, 31743.0f);
channelEPs[epi] = ParallelMath::RoundAndConvertToS16(f, roundingMode);
}
outEP0[ch] = channelEPs[0];
outEP1[ch] = channelEPs[1];
}
}
void FinishLDR(int tweak, int range, MUInt15* outEP0, MUInt15* outEP1)
{
ParallelMath::RoundTowardNearestForScope roundingMode;
float tweakFactors[2];
Util::ComputeTweakFactors(tweak, range, tweakFactors);
for (int ch = 0; ch < TVectorSize; ch++)
{
MFloat ep0f = ParallelMath::Clamp(m_base[ch] + m_offset[ch] * tweakFactors[0], 0.0f, 255.0f);
MFloat ep1f = ParallelMath::Clamp(m_base[ch] + m_offset[ch] * tweakFactors[1], 0.0f, 255.0f);
outEP0[ch] = ParallelMath::RoundAndConvertToU15(ep0f, &roundingMode);
outEP1[ch] = ParallelMath::RoundAndConvertToU15(ep1f, &roundingMode);
}
}
template<int TNewVectorSize>
UnfinishedEndpoints<TNewVectorSize> ExpandTo(float filler)
{
MFloat newBase[TNewVectorSize];
MFloat newOffset[TNewVectorSize];
for (int ch = 0; ch < TNewVectorSize && ch < TVectorSize; ch++)
{
newBase[ch] = m_base[ch];
newOffset[ch] = m_offset[ch];
}
MFloat fillerV = ParallelMath::MakeFloat(filler);
for (int ch = TVectorSize; ch < TNewVectorSize; ch++)
{
newBase[ch] = fillerV;
newOffset[ch] = ParallelMath::MakeFloatZero();
}
return UnfinishedEndpoints<TNewVectorSize>(newBase, newOffset);
}
private:
MFloat m_base[TVectorSize];
MFloat m_offset[TVectorSize];
};
}
}