liz/rawTherapee
liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix broken non SSE build and add SSE code for one loop in wavcont()

Browse Source
This commit is contained in:
Ingo Weyrich 2020-07-04 11:35:37 +02:00
parent af4b17ac84
commit 8728aee0ce
2 changed files with 30 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
rtengine/LUT.h
View File

@ -99,8 +99,8 @@ protected:
unsigned int size; unsigned int size;
unsigned int upperBound; // always equals size-1, parameter created for performance reason unsigned int upperBound; // always equals size-1, parameter created for performance reason
private: private:
#ifdef __SSE2__
unsigned int owner; unsigned int owner;
#ifdef __SSE2__
alignas(16) vfloat maxsv; alignas(16) vfloat maxsv;
alignas(16) vfloat sizev; alignas(16) vfloat sizev;
alignas(16) vint sizeiv; alignas(16) vint sizeiv;

44
rtengine/iplocallab.cc
View File

@ -7184,16 +7184,6 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
float lutFactor; float lutFactor;
const float inVals[] = {0.05f, 0.2f, 0.7f, 1.f, 1.f, 0.8f, 0.65f, 0.5f, 0.4f, 0.25f, 0.1f}; const float inVals[] = {0.05f, 0.2f, 0.7f, 1.f, 1.f, 0.8f, 0.65f, 0.5f, 0.4f, 0.25f, 0.1f};
const auto meaLut = buildMeaLut(inVals, mea, lutFactor); const auto meaLut = buildMeaLut(inVals, mea, lutFactor);
float klev = (loccomprewavCurve[level * 55.5f] - 0.75f);
if (klev < 0.f) {
klev *= 2.6666f;//compression increase contraste
} else {
klev *= 4.f;//dilatation reduce contraste - detailattenuator
}
const float compression = expf(-klev);
const float detailattenuator = std::max(klev, 0.f);
const auto wav_L = wdspot.level_coeffs(level)[dir]; const auto wav_L = wdspot.level_coeffs(level)[dir];
#ifdef _OPENMP #ifdef _OPENMP
@ -7206,14 +7196,38 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
} }
} }
float klev = (loccomprewavCurve[level * 55.5f] - 0.75f);
if (klev < 0.f) {
klev *= 2.6666f;//compression increase contraste
} else {
klev *= 4.f;//dilatation reduce contraste - detailattenuator
}
const float compression = expf(-klev);
const float detailattenuator = std::max(klev, 0.f);
Compresslevels(templevel, W_L, H_L, compression, detailattenuator, thres, mean[level], MaxP[level], meanN[level], MaxN[level], madL[level][dir - 1]); Compresslevels(templevel, W_L, H_L, compression, detailattenuator, thres, mean[level], MaxP[level], meanN[level], MaxN[level], madL[level][dir - 1]);
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for if (multiThread) #pragma omp parallel if (multiThread)
#endif #endif
for (int y = 0; y < H_L; y++) { {
for (int x = 0; x < W_L; x++) { #ifdef __SSE2__
int j = y * W_L + x; const vfloat lutFactorv = F2V(lutFactor);
wav_L[j] = intp((*meaLut)[std::fabs(wav_L[j]) * lutFactor], templevel[y][x], wav_L[j]); #endif
#ifdef _OPENMP
#pragma omp for
#endif
for (int y = 0; y < H_L; y++) {
int x = 0;
int j = y * W_L;
#ifdef __SSE2__
for (; x < W_L - 3; x += 4, j += 4) {
const vfloat valv = LVFU(wav_L[j]);
STVFU(wav_L[j], intp((*meaLut)[vabsf(valv) * lutFactorv], LVFU(templevel[y][x]), valv));
}
#endif
for (; x < W_L; x++, j++) {
wav_L[j] = intp((*meaLut)[std::fabs(wav_L[j]) * lutFactor], templevel[y][x], wav_L[j]);
}
} }
} }
} }