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

Removed most of the __x86_64__ dependencies. Also removed _RT_NESTED_OPENMP

Browse Source
This commit is contained in:
heckflosse 2018-01-23 16:05:31 +01:00
parent 4046a3c9b9
commit 96bf06a950
28 changed files with 171 additions and 226 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
rtengine/EdgePreservingDecomposition.cc
View File

@ -291,7 +291,7 @@ bool MultiDiagonalSymmetricMatrix::LazySetEntry(float value, int row, int column
return true; return true;
} }
SSEFUNCTION void MultiDiagonalSymmetricMatrix::VectorProduct(float* RESTRICT Product, float* RESTRICT x) void MultiDiagonalSymmetricMatrix::VectorProduct(float* RESTRICT Product, float* RESTRICT x)
{ {
int srm = StartRows[m - 1]; int srm = StartRows[m - 1];
@ -693,7 +693,7 @@ EdgePreservingDecomposition::~EdgePreservingDecomposition()
delete A; delete A;
} }
SSEFUNCTION float *EdgePreservingDecomposition::CreateBlur(float *Source, float Scale, float EdgeStopping, int Iterates, float *Blur, bool UseBlurForEdgeStop) float *EdgePreservingDecomposition::CreateBlur(float *Source, float Scale, float EdgeStopping, int Iterates, float *Blur, bool UseBlurForEdgeStop)
{ {
if(Blur == nullptr) if(Blur == nullptr)
@ -884,7 +884,7 @@ float *EdgePreservingDecomposition::CreateIteratedBlur(float *Source, float Scal
return Blur; return Blur;
} }
SSEFUNCTION void EdgePreservingDecomposition::CompressDynamicRange(float *Source, float Scale, float EdgeStopping, float CompressionExponent, float DetailBoost, int Iterates, int Reweightings) void EdgePreservingDecomposition::CompressDynamicRange(float *Source, float Scale, float EdgeStopping, float CompressionExponent, float DetailBoost, int Iterates, int Reweightings)
{ {
if(w < 300 && h < 300) { // set number of Reweightings to zero for small images (thumbnails). We could try to find a better solution here. if(w < 300 && h < 300) { // set number of Reweightings to zero for small images (thumbnails). We could try to find a better solution here.
Reweightings = 0; Reweightings = 0;

80
rtengine/FTblockDN.cc
View File

@ -474,7 +474,7 @@ void ImProcFunctions::Tile_calc(int tilesize, int overlap, int kall, int imwidth
int denoiseNestedLevels = 1; int denoiseNestedLevels = 1;
enum nrquality {QUALITY_STANDARD, QUALITY_HIGH}; enum nrquality {QUALITY_STANDARD, QUALITY_HIGH};
SSEFUNCTION void ImProcFunctions::RGB_denoise(int kall, Imagefloat * src, Imagefloat * dst, Imagefloat * calclum, float * ch_M, float *max_r, float *max_b, bool isRAW, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, const NoiseCurve & noiseLCurve, const NoiseCurve & noiseCCurve, float &nresi, float &highresi) void ImProcFunctions::RGB_denoise(int kall, Imagefloat * src, Imagefloat * dst, Imagefloat * calclum, float * ch_M, float *max_r, float *max_b, bool isRAW, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, const NoiseCurve & noiseLCurve, const NoiseCurve & noiseCCurve, float &nresi, float &highresi)
{ {
BENCHFUN BENCHFUN
//#ifdef _DEBUG //#ifdef _DEBUG
@ -783,7 +783,7 @@ BENCHFUN
numthreads = MIN(numthreads, options.rgbDenoiseThreadLimit); numthreads = MIN(numthreads, options.rgbDenoiseThreadLimit);
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
denoiseNestedLevels = omp_get_max_threads() / numthreads; denoiseNestedLevels = omp_get_max_threads() / numthreads;
bool oldNested = omp_get_nested(); bool oldNested = omp_get_nested();
@ -916,7 +916,7 @@ BENCHFUN
if (!denoiseMethodRgb) { //lab mode if (!denoiseMethodRgb) { //lab mode
//modification Jacques feb 2013 and july 2014 //modification Jacques feb 2013 and july 2014
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
@ -964,7 +964,7 @@ BENCHFUN
} }
} }
} else {//RGB mode } else {//RGB mode
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
@ -999,7 +999,7 @@ BENCHFUN
} }
} }
} else {//image is not raw; use Lab parametrization } else {//image is not raw; use Lab parametrization
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
@ -1159,7 +1159,7 @@ BENCHFUN
if (!memoryAllocationFailed) { if (!memoryAllocationFailed) {
// precalculate madL, because it's used in adecomp and bdecomp // precalculate madL, because it's used in adecomp and bdecomp
int maxlvl = Ldecomp->maxlevel(); int maxlvl = Ldecomp->maxlevel();
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) collapse(2) num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel for schedule(dynamic) collapse(2) num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
@ -1284,7 +1284,7 @@ BENCHFUN
if (!memoryAllocationFailed) { if (!memoryAllocationFailed) {
// copy labdn->L to Lin before it gets modified by reconstruction // copy labdn->L to Lin before it gets modified by reconstruction
Lin = new array2D<float>(width, height); Lin = new array2D<float>(width, height);
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
@ -1337,14 +1337,14 @@ BENCHFUN
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
int masterThread = omp_get_thread_num(); int masterThread = omp_get_thread_num();
#endif #endif
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
{ {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
int subThread = masterThread * denoiseNestedLevels + omp_get_thread_num(); int subThread = masterThread * denoiseNestedLevels + omp_get_thread_num();
#else #else
int subThread = 0; int subThread = 0;
@ -1354,7 +1354,7 @@ BENCHFUN
float *fLblox = fLbloxArray[subThread]; float *fLblox = fLbloxArray[subThread];
float pBuf[width + TS + 2 * blkrad * offset] ALIGNED16; float pBuf[width + TS + 2 * blkrad * offset] ALIGNED16;
float nbrwt[TS * TS] ALIGNED64; float nbrwt[TS * TS] ALIGNED64;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -1454,7 +1454,7 @@ BENCHFUN
} }
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% //%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
@ -1591,7 +1591,7 @@ BENCHFUN
realred /= 100.f; realred /= 100.f;
realblue /= 100.f; realblue /= 100.f;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16) num_threads(denoiseNestedLevels) #pragma omp parallel for schedule(dynamic,16) num_threads(denoiseNestedLevels)
#endif #endif
@ -1641,7 +1641,7 @@ BENCHFUN
} }
} }
} else {//RGB mode } else {//RGB mode
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) #pragma omp parallel for num_threads(denoiseNestedLevels)
#endif #endif
@ -1681,7 +1681,7 @@ BENCHFUN
} }
} else { } else {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) #pragma omp parallel for num_threads(denoiseNestedLevels)
#endif #endif
@ -1748,7 +1748,7 @@ BENCHFUN
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
omp_set_nested(oldNested); omp_set_nested(oldNested);
#endif #endif
@ -2043,7 +2043,7 @@ BENCHFUN
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% //%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
SSEFUNCTION void ImProcFunctions::RGBtile_denoise(float * fLblox, int hblproc, float noisevar_Ldetail, float * nbrwt, float * blurbuffer) //for DCT void ImProcFunctions::RGBtile_denoise(float * fLblox, int hblproc, float noisevar_Ldetail, float * nbrwt, float * blurbuffer) //for DCT
{ {
int blkstart = hblproc * TS * TS; int blkstart = hblproc * TS * TS;
@ -2254,7 +2254,7 @@ void ImProcFunctions::Noise_residualAB(wavelet_decomposition &WaveletCoeffs_ab,
chmaxresid = maxresid; chmaxresid = maxresid;
} }
SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposition &WaveletCoeffs_L, float *noisevarlum, float madL[8][3]) bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposition &WaveletCoeffs_L, float *noisevarlum, float madL[8][3])
{ {
int maxlvl = min(WaveletCoeffs_L.maxlevel(), 5); int maxlvl = min(WaveletCoeffs_L.maxlevel(), 5);
const float eps = 0.01f; const float eps = 0.01f;
@ -2272,7 +2272,7 @@ SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposit
} }
bool memoryAllocationFailed = false; bool memoryAllocationFailed = false;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
{ {
@ -2287,7 +2287,7 @@ SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposit
if (!memoryAllocationFailed) { if (!memoryAllocationFailed) {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2387,7 +2387,7 @@ SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposit
return (!memoryAllocationFailed); return (!memoryAllocationFailed);
} }
SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposition &WaveletCoeffs_L, wavelet_decomposition &WaveletCoeffs_ab, bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposition &WaveletCoeffs_L, wavelet_decomposition &WaveletCoeffs_ab,
float *noisevarchrom, float madL[8][3], float noisevar_ab, const bool useNoiseCCurve, bool autoch, bool denoiseMethodRgb) float *noisevarchrom, float madL[8][3], float noisevar_ab, const bool useNoiseCCurve, bool autoch, bool denoiseMethodRgb)
{ {
int maxlvl = WaveletCoeffs_L.maxlevel(); int maxlvl = WaveletCoeffs_L.maxlevel();
@ -2411,7 +2411,7 @@ SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
} }
bool memoryAllocationFailed = false; bool memoryAllocationFailed = false;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
{ {
@ -2427,7 +2427,7 @@ SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
if (!memoryAllocationFailed) { if (!memoryAllocationFailed) {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2446,7 +2446,7 @@ SSEFUNCTION bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2550,7 +2550,7 @@ bool ImProcFunctions::WaveletDenoiseAllL(wavelet_decomposition &WaveletCoeffs_L,
} }
bool memoryAllocationFailed = false; bool memoryAllocationFailed = false;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
{ {
@ -2565,7 +2565,7 @@ bool ImProcFunctions::WaveletDenoiseAllL(wavelet_decomposition &WaveletCoeffs_L,
} }
if (!memoryAllocationFailed) { if (!memoryAllocationFailed) {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2605,7 +2605,7 @@ bool ImProcFunctions::WaveletDenoiseAllAB(wavelet_decomposition &WaveletCoeffs_L
} }
bool memoryAllocationFailed = false; bool memoryAllocationFailed = false;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1) #pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif #endif
{ {
@ -2619,7 +2619,7 @@ bool ImProcFunctions::WaveletDenoiseAllAB(wavelet_decomposition &WaveletCoeffs_L
} }
if (!memoryAllocationFailed) { if (!memoryAllocationFailed) {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2643,7 +2643,7 @@ bool ImProcFunctions::WaveletDenoiseAllAB(wavelet_decomposition &WaveletCoeffs_L
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% //%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
SSEFUNCTION void ImProcFunctions::ShrinkAllL(wavelet_decomposition &WaveletCoeffs_L, float **buffer, int level, int dir, void ImProcFunctions::ShrinkAllL(wavelet_decomposition &WaveletCoeffs_L, float **buffer, int level, int dir,
float *noisevarlum, float * madL, float * vari, int edge) float *noisevarlum, float * madL, float * vari, int edge)
{ {
@ -2733,7 +2733,7 @@ SSEFUNCTION void ImProcFunctions::ShrinkAllL(wavelet_decomposition &WaveletCoeff
} }
SSEFUNCTION void ImProcFunctions::ShrinkAllAB(wavelet_decomposition &WaveletCoeffs_L, wavelet_decomposition &WaveletCoeffs_ab, float **buffer, int level, int dir, void ImProcFunctions::ShrinkAllAB(wavelet_decomposition &WaveletCoeffs_L, wavelet_decomposition &WaveletCoeffs_ab, float **buffer, int level, int dir,
float *noisevarchrom, float noisevar_ab, const bool useNoiseCCurve, bool autoch, float *noisevarchrom, float noisevar_ab, const bool useNoiseCCurve, bool autoch,
bool denoiseMethodRgb, float * madL, float * madaab, bool madCalculated) bool denoiseMethodRgb, float * madL, float * madaab, bool madCalculated)
@ -2848,7 +2848,7 @@ SSEFUNCTION void ImProcFunctions::ShrinkAllAB(wavelet_decomposition &WaveletCoef
} }
SSEFUNCTION void ImProcFunctions::ShrinkAll_info(float ** WavCoeffs_a, float ** WavCoeffs_b, void ImProcFunctions::ShrinkAll_info(float ** WavCoeffs_a, float ** WavCoeffs_b,
int W_ab, int H_ab, float **noisevarlum, float **noisevarchrom, float **noisevarhue, float &chaut, int &Nb, float &redaut, float &blueaut, int W_ab, int H_ab, float **noisevarlum, float **noisevarchrom, float **noisevarhue, float &chaut, int &Nb, float &redaut, float &blueaut,
float &maxredaut, float &maxblueaut, float &minredaut, float &minblueaut, int schoice, int lvl, float &chromina, float &sigma, float &lumema, float &sigma_L, float &redyel, float &skinc, float &nsknc, float &maxredaut, float &maxblueaut, float &minredaut, float &minblueaut, int schoice, int lvl, float &chromina, float &sigma, float &lumema, float &sigma_L, float &redyel, float &skinc, float &nsknc,
float &maxchred, float &maxchblue, float &minchred, float &minchblue, int &nb, float &chau, float &chred, float &chblue, bool denoiseMethodRgb) float &maxchred, float &maxchblue, float &minchred, float &minchblue, int &nb, float &chau, float &chred, float &chblue, bool denoiseMethodRgb)
@ -2988,7 +2988,7 @@ void ImProcFunctions::WaveletDenoiseAll_info(int levwav, wavelet_decomposition &
} }
} }
SSEFUNCTION void ImProcFunctions::RGB_denoise_infoGamCurve(const procparams::DirPyrDenoiseParams & dnparams, bool isRAW, LUTf &gamcurve, float &gam, float &gamthresh, float &gamslope) void ImProcFunctions::RGB_denoise_infoGamCurve(const procparams::DirPyrDenoiseParams & dnparams, bool isRAW, LUTf &gamcurve, float &gam, float &gamthresh, float &gamslope)
{ {
gam = dnparams.gamma; gam = dnparams.gamma;
gamthresh = 0.001f; gamthresh = 0.001f;
@ -3153,7 +3153,7 @@ void ImProcFunctions::calcautodn_info(float &chaut, float &delta, int Nb, int le
} }
SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat * provicalc, const bool isRAW, LUTf &gamcurve, float gam, float gamthresh, float gamslope, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, float &chaut, int &Nb, float &redaut, float &blueaut, float &maxredaut, float &maxblueaut, float &minredaut, float &minblueaut, float &chromina, float &sigma, float &lumema, float &sigma_L, float &redyel, float &skinc, float &nsknc, bool multiThread) void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat * provicalc, const bool isRAW, LUTf &gamcurve, float gam, float gamthresh, float gamslope, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, float &chaut, int &Nb, float &redaut, float &blueaut, float &maxredaut, float &maxblueaut, float &minredaut, float &minblueaut, float &chromina, float &sigma, float &lumema, float &sigma_L, float &redyel, float &skinc, float &nsknc, bool multiThread)
{ {
if ((settings->leveldnautsimpl == 1 && dnparams.Cmethod == "MAN") || (settings->leveldnautsimpl == 0 && dnparams.C2method == "MANU")) { if ((settings->leveldnautsimpl == 1 && dnparams.Cmethod == "MAN") || (settings->leveldnautsimpl == 0 && dnparams.C2method == "MANU")) {
//nothing to do //nothing to do
@ -3192,7 +3192,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
bcalc[i] = new float[wid]; bcalc[i] = new float[wid];
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for if (multiThread) #pragma omp parallel for if (multiThread)
#endif #endif
@ -3315,7 +3315,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
//fill tile from image; convert RGB to "luma/chroma" //fill tile from image; convert RGB to "luma/chroma"
if (isRAW) {//image is raw; use channel differences for chroma channels if (isRAW) {//image is raw; use channel differences for chroma channels
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for if (multiThread) #pragma omp parallel for if (multiThread)
#endif #endif
@ -3368,7 +3368,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
#endif #endif
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for if (multiThread) #pragma omp parallel for if (multiThread)
#endif #endif
@ -3386,7 +3386,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
if (!denoiseMethodRgb) { //lab mode, modification Jacques feb 2013 and july 2014 if (!denoiseMethodRgb) { //lab mode, modification Jacques feb 2013 and july 2014
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for if (multiThread) #pragma omp parallel for if (multiThread)
#endif #endif
@ -3523,17 +3523,17 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
} }
const int levwav = 5; const int levwav = 5;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel sections if (multiThread) #pragma omp parallel sections if (multiThread)
#endif #endif
{ {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp section #pragma omp section
#endif #endif
{ {
adecomp = new wavelet_decomposition(labdn->data + datalen, labdn->W, labdn->H, levwav, 1); adecomp = new wavelet_decomposition(labdn->data + datalen, labdn->W, labdn->H, levwav, 1);
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp section #pragma omp section
#endif #endif
{ {

20
rtengine/LUT.h
View File

@ -104,7 +104,7 @@ protected:
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:
unsigned int owner; unsigned int owner;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat maxsv ALIGNED16; vfloat maxsv ALIGNED16;
vfloat sizev ALIGNED16; vfloat sizev ALIGNED16;
vint sizeiv ALIGNED16; vint sizeiv ALIGNED16;
@ -136,7 +136,7 @@ public:
maxs = size - 2; maxs = size - 2;
maxsf = (float)maxs; maxsf = (float)maxs;
maxIndexFloat = ((float)upperBound) - 1e-5; maxIndexFloat = ((float)upperBound) - 1e-5;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
maxsv = F2V( maxs ); maxsv = F2V( maxs );
sizeiv = _mm_set1_epi32( (int)(size - 1) ); sizeiv = _mm_set1_epi32( (int)(size - 1) );
sizev = F2V( size - 1 ); sizev = F2V( size - 1 );
@ -167,7 +167,7 @@ public:
maxs = size - 2; maxs = size - 2;
maxsf = (float)maxs; maxsf = (float)maxs;
maxIndexFloat = ((float)upperBound) - 1e-5; maxIndexFloat = ((float)upperBound) - 1e-5;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
maxsv = F2V( maxs ); maxsv = F2V( maxs );
sizeiv = _mm_set1_epi32( (int)(size - 1) ); sizeiv = _mm_set1_epi32( (int)(size - 1) );
sizev = F2V( size - 1 ); sizev = F2V( size - 1 );
@ -178,7 +178,7 @@ public:
{ {
data = nullptr; data = nullptr;
reset(); reset();
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
maxsv = ZEROV; maxsv = ZEROV;
sizev = ZEROV; sizev = ZEROV;
sizeiv = _mm_setzero_si128(); sizeiv = _mm_setzero_si128();
@ -243,7 +243,7 @@ public:
this->maxs = this->size - 2; this->maxs = this->size - 2;
this->maxsf = (float)this->maxs; this->maxsf = (float)this->maxs;
this->maxIndexFloat = ((float)this->upperBound) - 1e-5; this->maxIndexFloat = ((float)this->upperBound) - 1e-5;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
this->maxsv = F2V( this->size - 2); this->maxsv = F2V( this->size - 2);
this->sizeiv = _mm_set1_epi32( (int)(this->size - 1) ); this->sizeiv = _mm_set1_epi32( (int)(this->size - 1) );
this->sizev = F2V( this->size - 1 ); this->sizev = F2V( this->size - 1 );
@ -258,7 +258,7 @@ public:
LUT<T> & operator+=(LUT<T> &rhs) LUT<T> & operator+=(LUT<T> &rhs)
{ {
if (rhs.size == this->size) { if (rhs.size == this->size) {
#ifdef _RT_NESTED_OPENMP // temporary solution to fix Issue #3324 #ifdef _OPENMP
#pragma omp simd #pragma omp simd
#endif #endif
@ -274,7 +274,7 @@ public:
template<typename U = T, typename = typename std::enable_if<std::is_same<U, float>::value>::type> template<typename U = T, typename = typename std::enable_if<std::is_same<U, float>::value>::type>
LUT<float> & operator*=(float factor) LUT<float> & operator*=(float factor)
{ {
#ifdef _RT_NESTED_OPENMP // temporary solution to fix Issue #3324 #ifdef _OPENMP
#pragma omp simd #pragma omp simd
#endif #endif
@ -289,7 +289,7 @@ public:
template<typename U = T, typename = typename std::enable_if<std::is_same<U, float>::value>::type> template<typename U = T, typename = typename std::enable_if<std::is_same<U, float>::value>::type>
LUT<float> & operator/=(float divisor) LUT<float> & operator/=(float divisor)
{ {
#ifdef _RT_NESTED_OPENMP // temporary solution to fix Issue #3324 #ifdef _OPENMP
#pragma omp simd #pragma omp simd
#endif #endif
@ -307,7 +307,7 @@ public:
return data[ rtengine::LIM<int>(index, 0, upperBound) ]; return data[ rtengine::LIM<int>(index, 0, upperBound) ];
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
// NOTE: This function requires LUTs which clips only at lower bound // NOTE: This function requires LUTs which clips only at lower bound
@ -698,7 +698,7 @@ public:
maxs = size - 2; maxs = size - 2;
maxsf = (float)maxs; maxsf = (float)maxs;
maxIndexFloat = ((float)upperBound) - 1e-5; maxIndexFloat = ((float)upperBound) - 1e-5;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
maxsv = F2V( size - 2); maxsv = F2V( size - 2);
sizeiv = _mm_set1_epi32( (int)(size - 1) ); sizeiv = _mm_set1_epi32( (int)(size - 1) );
sizev = F2V( size - 1 ); sizev = F2V( size - 1 );

8
rtengine/PF_correct_RT.cc
View File

@ -43,7 +43,7 @@ namespace rtengine
{ {
extern const Settings* settings; extern const Settings* settings;
SSEFUNCTION void ImProcFunctions::PF_correct_RT(LabImage * src, LabImage * dst, double radius, int thresh) void ImProcFunctions::PF_correct_RT(LabImage * src, LabImage * dst, double radius, int thresh)
{ {
const int halfwin = ceil(2 * radius) + 1; const int halfwin = ceil(2 * radius) + 1;
@ -285,7 +285,7 @@ SSEFUNCTION void ImProcFunctions::PF_correct_RT(LabImage * src, LabImage * dst,
free(fringe); free(fringe);
} }
SSEFUNCTION void ImProcFunctions::PF_correct_RTcam(CieImage * src, CieImage * dst, double radius, int thresh) void ImProcFunctions::PF_correct_RTcam(CieImage * src, CieImage * dst, double radius, int thresh)
{ {
const int halfwin = ceil(2 * radius) + 1; const int halfwin = ceil(2 * radius) + 1;
@ -655,7 +655,7 @@ SSEFUNCTION void ImProcFunctions::PF_correct_RTcam(CieImage * src, CieImage * ds
free(fringe); free(fringe);
} }
SSEFUNCTION void ImProcFunctions::Badpixelscam(CieImage * src, CieImage * dst, double radius, int thresh, int mode, float skinprot, float chrom, int hotbad) void ImProcFunctions::Badpixelscam(CieImage * src, CieImage * dst, double radius, int thresh, int mode, float skinprot, float chrom, int hotbad)
{ {
const int halfwin = ceil(2 * radius) + 1; const int halfwin = ceil(2 * radius) + 1;
MyTime t1, t2; MyTime t1, t2;
@ -1263,7 +1263,7 @@ SSEFUNCTION void ImProcFunctions::Badpixelscam(CieImage * src, CieImage * dst, d
} }
SSEFUNCTION void ImProcFunctions::BadpixelsLab(LabImage * src, LabImage * dst, double radius, int thresh, int mode, float skinprot, float chrom) void ImProcFunctions::BadpixelsLab(LabImage * src, LabImage * dst, double radius, int thresh, int mode, float skinprot, float chrom)
{ {
const int halfwin = ceil(2 * radius) + 1; const int halfwin = ceil(2 * radius) + 1;
MyTime t1, t2; MyTime t1, t2;

2
rtengine/amaze_demosaic_RT.cc
View File

@ -38,7 +38,7 @@
namespace rtengine namespace rtengine
{ {
SSEFUNCTION void RawImageSource::amaze_demosaic_RT(int winx, int winy, int winw, int winh, array2D<float> &rawData, array2D<float> &red, array2D<float> &green, array2D<float> &blue) void RawImageSource::amaze_demosaic_RT(int winx, int winy, int winw, int winh, array2D<float> &rawData, array2D<float> &red, array2D<float> &green, array2D<float> &blue)
{ {
BENCHFUN BENCHFUN

6
rtengine/boxblur.h
View File

@ -121,7 +121,7 @@ template<class T, class A> void boxblur (T** src, A** dst, int radx, int rady, i
} }
template<class T, class A> SSEFUNCTION void boxblur (T** src, A** dst, T* buffer, int radx, int rady, int W, int H) template<class T, class A> void boxblur (T** src, A** dst, T* buffer, int radx, int rady, int W, int H)
{ {
//box blur image; box range = (radx,rady) //box blur image; box range = (radx,rady)
@ -309,7 +309,7 @@ template<class T, class A> SSEFUNCTION void boxblur (T** src, A** dst, T* buffer
} }
template<class T, class A> SSEFUNCTION void boxblur (T* src, A* dst, A* buffer, int radx, int rady, int W, int H) template<class T, class A> void boxblur (T* src, A* dst, A* buffer, int radx, int rady, int W, int H)
{ {
//box blur image; box range = (radx,rady) i.e. box size is (2*radx+1)x(2*rady+1) //box blur image; box range = (radx,rady) i.e. box size is (2*radx+1)x(2*rady+1)
@ -496,7 +496,7 @@ template<class T, class A> SSEFUNCTION void boxblur (T* src, A* dst, A* buffer,
} }
template<class T, class A> SSEFUNCTION void boxabsblur (T* src, A* dst, int radx, int rady, int W, int H, float * temp) template<class T, class A> void boxabsblur (T* src, A* dst, int radx, int rady, int W, int H, float * temp)
{ {
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% //%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

6
rtengine/color.cc
View File

@ -1725,7 +1725,7 @@ void Color::Lab2XYZ(vfloat L, vfloat a, vfloat b, vfloat &x, vfloat &y, vfloat &
void Color::RGB2Lab(float *R, float *G, float *B, float *L, float *a, float *b, const float wp[3][3], int width) void Color::RGB2Lab(float *R, float *G, float *B, float *L, float *a, float *b, const float wp[3][3], int width)
{ {
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat maxvalfv = F2V(MAXVALF); vfloat maxvalfv = F2V(MAXVALF);
vfloat c116v = F2V(116.f); vfloat c116v = F2V(116.f);
vfloat c5242d88v = F2V(5242.88f); vfloat c5242d88v = F2V(5242.88f);
@ -1733,7 +1733,7 @@ void Color::RGB2Lab(float *R, float *G, float *B, float *L, float *a, float *b,
vfloat c200v = F2V(200.f); vfloat c200v = F2V(200.f);
#endif #endif
int i = 0; int i = 0;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for(;i < width - 3; i+=4) { for(;i < width - 3; i+=4) {
const vfloat rv = LVFU(R[i]); const vfloat rv = LVFU(R[i]);
const vfloat gv = LVFU(G[i]); const vfloat gv = LVFU(G[i]);
@ -2606,7 +2606,7 @@ void Color::gamutLchonly (float2 sincosval, float &Lprov1, float &Chprov1, const
* const double wip[3][3]: matrix for working profile * const double wip[3][3]: matrix for working profile
* bool multiThread : parallelize the loop * bool multiThread : parallelize the loop
*/ */
SSEFUNCTION void Color::LabGamutMunsell(float *labL, float *laba, float *labb, const int N, bool corMunsell, bool lumaMuns, bool isHLEnabled, bool gamut, const double wip[3][3]) void Color::LabGamutMunsell(float *labL, float *laba, float *labb, const int N, bool corMunsell, bool lumaMuns, bool isHLEnabled, bool gamut, const double wip[3][3])
{ {
#ifdef _DEBUG #ifdef _DEBUG
MyTime t1e, t2e; MyTime t1e, t2e;

34
rtengine/cplx_wavelet_level.h
View File

@ -244,7 +244,7 @@ template<typename T> void wavelet_level<T>::SynthesisFilterHaarHorizontal (const
* Applies a Haar filter * Applies a Haar filter
* *
*/ */
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(numThreads) if(numThreads>1) #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
#endif #endif
@ -266,11 +266,11 @@ template<typename T> void wavelet_level<T>::SynthesisFilterHaarVertical (const T
* Applies a Haar filter * Applies a Haar filter
* *
*/ */
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(numThreads) if(numThreads>1) #pragma omp parallel num_threads(numThreads) if(numThreads>1)
#endif #endif
{ {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for nowait #pragma omp for nowait
#endif #endif
@ -281,7 +281,7 @@ template<typename T> void wavelet_level<T>::SynthesisFilterHaarVertical (const T
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -328,7 +328,7 @@ void wavelet_level<T>::AnalysisFilterSubsampHorizontal (T * RESTRICT srcbuffer,
} }
#ifdef __SSE2__ #ifdef __SSE2__
template<typename T> SSEFUNCTION void wavelet_level<T>::AnalysisFilterSubsampVertical (T * RESTRICT srcbuffer, T * RESTRICT dstLo, T * RESTRICT dstHi, float (* RESTRICT filterLo)[4], float (* RESTRICT filterHi)[4], template<typename T> void wavelet_level<T>::AnalysisFilterSubsampVertical (T * RESTRICT srcbuffer, T * RESTRICT dstLo, T * RESTRICT dstHi, float (* RESTRICT filterLo)[4], float (* RESTRICT filterHi)[4],
const int taps, const int offset, const int width, const int height, const int row) const int taps, const int offset, const int width, const int height, const int row)
{ {
@ -455,7 +455,7 @@ template<typename T> void wavelet_level<T>::SynthesisFilterSubsampHorizontal (T
// calculate coefficients // calculate coefficients
int shift = skip * (taps - offset - 1); //align filter with data int shift = skip * (taps - offset - 1); //align filter with data
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(numThreads) if(numThreads>1) #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
#endif #endif
@ -506,7 +506,7 @@ template<typename T> void wavelet_level<T>::SynthesisFilterSubsampHorizontal (T
} }
#ifdef __SSE2__ #ifdef __SSE2__
template<typename T> SSEFUNCTION void wavelet_level<T>::SynthesisFilterSubsampVertical (T * RESTRICT srcLo, T * RESTRICT srcHi, T * RESTRICT dst, float (* RESTRICT filterLo)[4], float (* RESTRICT filterHi)[4], const int taps, const int offset, const int width, const int srcheight, const int dstheight, const float blend) template<typename T> void wavelet_level<T>::SynthesisFilterSubsampVertical (T * RESTRICT srcLo, T * RESTRICT srcHi, T * RESTRICT dst, float (* RESTRICT filterLo)[4], float (* RESTRICT filterHi)[4], const int taps, const int offset, const int width, const int srcheight, const int dstheight, const float blend)
{ {
/* Basic convolution code /* Basic convolution code
@ -521,7 +521,7 @@ template<typename T> SSEFUNCTION void wavelet_level<T>::SynthesisFilterSubsampVe
__m128 fourv = _mm_set1_ps(4.f); __m128 fourv = _mm_set1_ps(4.f);
__m128 srcFactorv = _mm_set1_ps(srcFactor); __m128 srcFactorv = _mm_set1_ps(srcFactor);
__m128 dstFactorv = _mm_set1_ps(blend); __m128 dstFactorv = _mm_set1_ps(blend);
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(numThreads) if(numThreads>1) #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
#endif #endif
@ -594,7 +594,7 @@ template<typename T> void wavelet_level<T>::SynthesisFilterSubsampVertical (T *
// calculate coefficients // calculate coefficients
int shift = skip * (taps - offset - 1); //align filter with data int shift = skip * (taps - offset - 1); //align filter with data
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(numThreads) if(numThreads>1) #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
#endif #endif
@ -630,7 +630,7 @@ template<typename T> void wavelet_level<T>::SynthesisFilterSubsampVertical (T *
#endif #endif
#ifdef __SSE2__ #ifdef __SSE2__
template<typename T> template<typename E> SSEFUNCTION void wavelet_level<T>::decompose_level(E *src, E *dst, float *filterV, float *filterH, int taps, int offset) template<typename T> template<typename E> void wavelet_level<T>::decompose_level(E *src, E *dst, float *filterV, float *filterH, int taps, int offset)
{ {
/* filter along rows and columns */ /* filter along rows and columns */
@ -644,7 +644,7 @@ template<typename T> template<typename E> SSEFUNCTION void wavelet_level<T>::dec
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(numThreads) if(numThreads>1) #pragma omp parallel num_threads(numThreads) if(numThreads>1)
#endif #endif
{ {
@ -652,7 +652,7 @@ template<typename T> template<typename E> SSEFUNCTION void wavelet_level<T>::dec
T tmpHi[m_w] ALIGNED64; T tmpHi[m_w] ALIGNED64;
if(subsamp_out) { if(subsamp_out) {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -662,7 +662,7 @@ template<typename T> template<typename E> SSEFUNCTION void wavelet_level<T>::dec
AnalysisFilterSubsampHorizontal (tmpHi, wavcoeffs[2], wavcoeffs[3], filterH, filterH + taps, taps, offset, m_w, m_w2, row / 2); AnalysisFilterSubsampHorizontal (tmpHi, wavcoeffs[2], wavcoeffs[3], filterH, filterH + taps, taps, offset, m_w, m_w2, row / 2);
} }
} else { } else {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -678,7 +678,7 @@ template<typename T> template<typename E> SSEFUNCTION void wavelet_level<T>::dec
template<typename T> template<typename E> void wavelet_level<T>::decompose_level(E *src, E *dst, float *filterV, float *filterH, int taps, int offset) template<typename T> template<typename E> void wavelet_level<T>::decompose_level(E *src, E *dst, float *filterV, float *filterH, int taps, int offset)
{ {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(numThreads) if(numThreads>1) #pragma omp parallel num_threads(numThreads) if(numThreads>1)
#endif #endif
{ {
@ -687,7 +687,7 @@ template<typename T> template<typename E> void wavelet_level<T>::decompose_level
/* filter along rows and columns */ /* filter along rows and columns */
if(subsamp_out) if(subsamp_out)
{ {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -697,7 +697,7 @@ template<typename T> template<typename E> void wavelet_level<T>::decompose_level
AnalysisFilterSubsampHorizontal (tmpHi, wavcoeffs[2], wavcoeffs[3], filterH, filterH + taps, taps, offset, m_w, m_w2, row / 2); AnalysisFilterSubsampHorizontal (tmpHi, wavcoeffs[2], wavcoeffs[3], filterH, filterH + taps, taps, offset, m_w, m_w2, row / 2);
} }
} else { } else {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -714,7 +714,7 @@ template<typename T> template<typename E> void wavelet_level<T>::decompose_level
#ifdef __SSE2__ #ifdef __SSE2__
template<typename T> template<typename E> SSEFUNCTION void wavelet_level<T>::reconstruct_level(E* tmpLo, E* tmpHi, E * src, E *dst, float *filterV, float *filterH, int taps, int offset, const float blend) template<typename T> template<typename E> void wavelet_level<T>::reconstruct_level(E* tmpLo, E* tmpHi, E * src, E *dst, float *filterV, float *filterH, int taps, int offset, const float blend)
{ {
if(memoryAllocationFailed) { if(memoryAllocationFailed) {
return; return;

2
rtengine/curves.cc
View File

@ -476,7 +476,7 @@ void CurveFactory::complexsgnCurve (bool & autili, bool & butili, bool & ccutil
} }
SSEFUNCTION void CurveFactory::complexCurve (double ecomp, double black, double hlcompr, double hlcomprthresh, void CurveFactory::complexCurve (double ecomp, double black, double hlcompr, double hlcomprthresh,
double shcompr, double br, double contr, double shcompr, double br, double contr,
const std::vector<double>& curvePoints, const std::vector<double>& curvePoints,
const std::vector<double>& curvePoints2, const std::vector<double>& curvePoints2,

12
rtengine/curves.h
View File

@ -828,7 +828,7 @@ class WeightedStdToneCurve : public ToneCurve
{ {
private: private:
float Triangle(float refX, float refY, float X2) const; float Triangle(float refX, float refY, float X2) const;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat Triangle(vfloat refX, vfloat refY, vfloat X2) const; vfloat Triangle(vfloat refX, vfloat refY, vfloat X2) const;
#endif #endif
public: public:
@ -904,7 +904,7 @@ inline void StandardToneCurve::BatchApply(
// If we get to the end before getting to an aligned address, just return. // If we get to the end before getting to an aligned address, just return.
// (Or, for non-SSE mode, if we get to the end.) // (Or, for non-SSE mode, if we get to the end.)
return; return;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
} else if (reinterpret_cast<uintptr_t>(&r[i]) % 16 == 0) { } else if (reinterpret_cast<uintptr_t>(&r[i]) % 16 == 0) {
// Otherwise, we get to the first aligned address; go to the SSE part. // Otherwise, we get to the first aligned address; go to the SSE part.
break; break;
@ -916,7 +916,7 @@ inline void StandardToneCurve::BatchApply(
i++; i++;
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for (; i + 3 < end; i += 4) { for (; i + 3 < end; i += 4) {
__m128 r_val = LVF(r[i]); __m128 r_val = LVF(r[i]);
__m128 g_val = LVF(g[i]); __m128 g_val = LVF(g[i]);
@ -1007,7 +1007,7 @@ inline float WeightedStdToneCurve::Triangle(float a, float a1, float b) const
return a1; return a1;
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
inline vfloat WeightedStdToneCurve::Triangle(vfloat a, vfloat a1, vfloat b) const inline vfloat WeightedStdToneCurve::Triangle(vfloat a, vfloat a1, vfloat b) const
{ {
vfloat a2 = a1 - a; vfloat a2 = a1 - a;
@ -1061,7 +1061,7 @@ inline void WeightedStdToneCurve::BatchApply(const size_t start, const size_t en
// If we get to the end before getting to an aligned address, just return. // If we get to the end before getting to an aligned address, just return.
// (Or, for non-SSE mode, if we get to the end.) // (Or, for non-SSE mode, if we get to the end.)
return; return;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
} else if (reinterpret_cast<uintptr_t>(&r[i]) % 16 == 0) { } else if (reinterpret_cast<uintptr_t>(&r[i]) % 16 == 0) {
// Otherwise, we get to the first aligned address; go to the SSE part. // Otherwise, we get to the first aligned address; go to the SSE part.
break; break;
@ -1071,7 +1071,7 @@ inline void WeightedStdToneCurve::BatchApply(const size_t start, const size_t en
i++; i++;
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
const vfloat c65535v = F2V(65535.f); const vfloat c65535v = F2V(65535.f);
const vfloat zd5v = F2V(0.5f); const vfloat zd5v = F2V(0.5f);
const vfloat zd25v = F2V(0.25f); const vfloat zd25v = F2V(0.25f);

10
rtengine/demosaic_algos.cc
View File

@ -1314,7 +1314,7 @@ void RawImageSource::jdl_interpolate_omp() // from "Lassus"
// Adapted to RawTherapee by Jacques Desmis 3/2013 // Adapted to RawTherapee by Jacques Desmis 3/2013
// Improved speed and reduced memory consumption by Ingo Weyrich 2/2015 // Improved speed and reduced memory consumption by Ingo Weyrich 2/2015
//TODO Tiles to reduce memory consumption //TODO Tiles to reduce memory consumption
SSEFUNCTION void RawImageSource::lmmse_interpolate_omp(int winw, int winh, array2D<float> &rawData, array2D<float> &red, array2D<float> &green, array2D<float> &blue, int iterations) void RawImageSource::lmmse_interpolate_omp(int winw, int winh, array2D<float> &rawData, array2D<float> &red, array2D<float> &green, array2D<float> &blue, int iterations)
{ {
const int width = winw, height = winh; const int width = winw, height = winh;
const int ba = 10; const int ba = 10;
@ -1946,7 +1946,7 @@ SSEFUNCTION void RawImageSource::lmmse_interpolate_omp(int winw, int winh, array
// SSE version by Ingo Weyrich 5/2013 // SSE version by Ingo Weyrich 5/2013
#ifdef __SSE2__ #ifdef __SSE2__
#define CLIPV(a) LIMV(a,zerov,c65535v) #define CLIPV(a) LIMV(a,zerov,c65535v)
SSEFUNCTION void RawImageSource::igv_interpolate(int winw, int winh) void RawImageSource::igv_interpolate(int winw, int winh)
{ {
static const float eps = 1e-5f, epssq = 1e-5f; //mod epssq -10f =>-5f Jacques 3/2013 to prevent artifact (divide by zero) static const float eps = 1e-5f, epssq = 1e-5f; //mod epssq -10f =>-5f Jacques 3/2013 to prevent artifact (divide by zero)
@ -2873,7 +2873,7 @@ void RawImageSource::nodemosaic(bool bw)
#ifdef __SSE2__ #ifdef __SSE2__
#define CLIPV(a) LIMV(a,ZEROV,c65535v) #define CLIPV(a) LIMV(a,ZEROV,c65535v)
#endif #endif
SSEFUNCTION void RawImageSource::refinement(int PassCount) void RawImageSource::refinement(int PassCount)
{ {
MyTime t1e, t2e; MyTime t1e, t2e;
t1e.set(); t1e.set();
@ -3870,7 +3870,7 @@ void RawImageSource::cielab (const float (*rgb)[3], float* l, float* a, float *b
return; return;
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat zd5v = F2V(0.5f); vfloat zd5v = F2V(0.5f);
vfloat c116v = F2V(116.f); vfloat c116v = F2V(116.f);
vfloat c16v = F2V(16.f); vfloat c16v = F2V(16.f);
@ -3887,7 +3887,7 @@ void RawImageSource::cielab (const float (*rgb)[3], float* l, float* a, float *b
for(int i = 0; i < height; i++) { for(int i = 0; i < height; i++) {
int j = 0; int j = 0;
#if defined( __SSE2__ ) && defined( __x86_64__ ) // vectorized LUT access is restricted to __x86_64__ => we have to use the same restriction #ifdef __SSE2__
for(; j < labWidth - 3; j += 4) { for(; j < labWidth - 3; j += 4) {
vfloat redv, greenv, bluev; vfloat redv, greenv, bluev;

4
rtengine/dirpyr_equalizer.cc
View File

@ -40,7 +40,7 @@ extern const Settings* settings;
//sequence of scales //sequence of scales
SSEFUNCTION void ImProcFunctions :: dirpyr_equalizer(float ** src, float ** dst, int srcwidth, int srcheight, float ** l_a, float ** l_b, const double * mult, const double dirpyrThreshold, const double skinprot, float b_l, float t_l, float t_r, int scaleprev) void ImProcFunctions :: dirpyr_equalizer(float ** src, float ** dst, int srcwidth, int srcheight, float ** l_a, float ** l_b, const double * mult, const double dirpyrThreshold, const double skinprot, float b_l, float t_l, float t_r, int scaleprev)
{ {
int lastlevel = maxlevel; int lastlevel = maxlevel;
@ -380,7 +380,7 @@ void ImProcFunctions :: dirpyr_equalizercam (CieImage *ncie, float ** src, float
} }
} }
SSEFUNCTION void ImProcFunctions::dirpyr_channel(float ** data_fine, float ** data_coarse, int width, int height, int level, int scale) void ImProcFunctions::dirpyr_channel(float ** data_fine, float ** data_coarse, int width, int height, int level, int scale)
{ {
// scale is spacing of directional averaging weights // scale is spacing of directional averaging weights
// calculate weights, compute directionally weighted average // calculate weights, compute directionally weighted average

2
rtengine/fast_demo.cc
View File

@ -52,7 +52,7 @@ LUTf RawImageSource::initInvGrad()
#endif #endif
//LUTf RawImageSource::invGrad = RawImageSource::initInvGrad(); //LUTf RawImageSource::invGrad = RawImageSource::initInvGrad();
SSEFUNCTION void RawImageSource::fast_demosaic() void RawImageSource::fast_demosaic()
{ {
double progress = 0.0; double progress = 0.0;

10
rtengine/gauss.cc
View File

@ -229,7 +229,7 @@ template<class T> void gaussHorizontal3 (T** src, T** dst, int W, int H, const f
} }
#ifdef __SSE2__ #ifdef __SSE2__
template<class T> SSEFUNCTION void gaussVertical3 (T** src, T** dst, int W, int H, const float c0, const float c1) template<class T> void gaussVertical3 (T** src, T** dst, int W, int H, const float c0, const float c1)
{ {
vfloat Tv = F2V(0.f), Tm1v, Tp1v; vfloat Tv = F2V(0.f), Tm1v, Tp1v;
vfloat Tv1 = F2V(0.f), Tm1v1, Tp1v1; vfloat Tv1 = F2V(0.f), Tm1v1, Tp1v1;
@ -314,7 +314,7 @@ template<class T> void gaussVertical3 (T** src, T** dst, int W, int H, const flo
#ifdef __SSE2__ #ifdef __SSE2__
// fast gaussian approximation if the support window is large // fast gaussian approximation if the support window is large
template<class T> SSEFUNCTION void gaussHorizontalSse (T** src, T** dst, const int W, const int H, const float sigma) template<class T> void gaussHorizontalSse (T** src, T** dst, const int W, const int H, const float sigma)
{ {
double b1, b2, b3, B, M[3][3]; double b1, b2, b3, B, M[3][3];
calculateYvVFactors<double>(sigma, b1, b2, b3, B, M); calculateYvVFactors<double>(sigma, b1, b2, b3, B, M);
@ -474,7 +474,7 @@ template<class T> void gaussHorizontal (T** src, T** dst, const int W, const int
} }
#ifdef __SSE2__ #ifdef __SSE2__
template<class T> SSEFUNCTION void gaussVerticalSse (T** src, T** dst, const int W, const int H, const float sigma) template<class T> void gaussVerticalSse (T** src, T** dst, const int W, const int H, const float sigma)
{ {
double b1, b2, b3, B, M[3][3]; double b1, b2, b3, B, M[3][3];
calculateYvVFactors<double>(sigma, b1, b2, b3, B, M); calculateYvVFactors<double>(sigma, b1, b2, b3, B, M);
@ -617,7 +617,7 @@ template<class T> SSEFUNCTION void gaussVerticalSse (T** src, T** dst, const int
#endif #endif
#ifdef __SSE2__ #ifdef __SSE2__
template<class T> SSEFUNCTION void gaussVerticalSsemult (T** RESTRICT src, T** RESTRICT dst, const int W, const int H, const float sigma) template<class T> void gaussVerticalSsemult (T** RESTRICT src, T** RESTRICT dst, const int W, const int H, const float sigma)
{ {
double b1, b2, b3, B, M[3][3]; double b1, b2, b3, B, M[3][3];
calculateYvVFactors<double>(sigma, b1, b2, b3, B, M); calculateYvVFactors<double>(sigma, b1, b2, b3, B, M);
@ -758,7 +758,7 @@ template<class T> SSEFUNCTION void gaussVerticalSsemult (T** RESTRICT src, T** R
} }
} }
template<class T> SSEFUNCTION void gaussVerticalSsediv (T** RESTRICT src, T** RESTRICT dst, T** divBuffer, const int W, const int H, const float sigma) template<class T> void gaussVerticalSsediv (T** RESTRICT src, T** RESTRICT dst, T** divBuffer, const int W, const int H, const float sigma)
{ {
double b1, b2, b3, B, M[3][3]; double b1, b2, b3, B, M[3][3];
calculateYvVFactors<double>(sigma, b1, b2, b3, B, M); calculateYvVFactors<double>(sigma, b1, b2, b3, B, M);

20
rtengine/helpersse2.h
View File

@ -29,29 +29,13 @@ typedef __m128 vfloat;
typedef __m128i vint2; typedef __m128i vint2;
// //
#ifdef __GNUC__
#if (!defined(WIN32) || defined( __x86_64__ ))
#define LVF(x) _mm_load_ps((float*)&x) #define LVF(x) _mm_load_ps((float*)&x)
#define LVFU(x) _mm_loadu_ps(&x) #define LVFU(x) _mm_loadu_ps(&x)
#define STVF(x,y) _mm_store_ps(&x,y) #define STVF(x,y) _mm_store_ps(&x,y)
#define STVFU(x,y) _mm_storeu_ps(&x,y) #define STVFU(x,y) _mm_storeu_ps(&x,y)
#define LVI(x) _mm_load_si128((__m128i*)&x) #define LVI(x) _mm_load_si128((__m128i*)&x)
#else // there is a bug in gcc 4.7.x when using openmp and aligned memory and -O3, also need to map the aligned functions to unaligned functions for WIN32 builds
#define LVF(x) _mm_loadu_ps((float*)&x)
#define LVFU(x) _mm_loadu_ps(&x)
#define STVF(x,y) _mm_storeu_ps(&x,y)
#define STVFU(x,y) _mm_storeu_ps(&x,y)
#define LVI(x) _mm_loadu_si128((__m128i*)&x)
#endif
#else
#define LVF(x) _mm_load_ps((float*)&x)
#define LVFU(x) _mm_loadu_ps(&x)
#define STVF(x,y) _mm_store_ps(&x,y)
#define STVFU(x,y) _mm_storeu_ps(&x,y)
#define LVI(x) _mm_load_si128((__m128i*)&x)
#endif
#if defined(__x86_64__) && defined(__AVX__) #ifdef __AVX__
#define PERMUTEPS(a,mask) _mm_permute_ps(a,mask) #define PERMUTEPS(a,mask) _mm_permute_ps(a,mask)
#else #else
#define PERMUTEPS(a,mask) _mm_shuffle_ps(a,a,mask) #define PERMUTEPS(a,mask) _mm_shuffle_ps(a,a,mask)
@ -67,7 +51,7 @@ static INLINE vfloat LC2VFU(float &a)
// Store a vector of 4 floats in a[0],a[2],a[4] and a[6] // Store a vector of 4 floats in a[0],a[2],a[4] and a[6]
#if defined(__x86_64__) && defined(__SSE4_1__) #ifdef __SSE4_1__
// SSE4.1 => use _mm_blend_ps instead of _mm_set_epi32 and vself // SSE4.1 => use _mm_blend_ps instead of _mm_set_epi32 and vself
#define STC2VFU(a,v) {\ #define STC2VFU(a,v) {\
__m128 TST1V = _mm_loadu_ps(&a);\ __m128 TST1V = _mm_loadu_ps(&a);\

2
rtengine/hilite_recon.cc
View File

@ -35,7 +35,7 @@ extern const Settings* settings;
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% //%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% //%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
SSEFUNCTION void RawImageSource::boxblur2(float** src, float** dst, float** temp, int H, int W, int box ) void RawImageSource::boxblur2(float** src, float** dst, float** temp, int H, int W, int box )
{ {
//box blur image channel; box size = 2*box+1 //box blur image channel; box size = 2*box+1
//horizontal blur //horizontal blur

16
rtengine/improcfun.cc
View File

@ -53,7 +53,7 @@ using namespace rtengine;
// begin of helper function for rgbProc() // begin of helper function for rgbProc()
void shadowToneCurve(const LUTf &shtonecurve, float *rtemp, float *gtemp, float *btemp, int istart, int tH, int jstart, int tW, int tileSize) { void shadowToneCurve(const LUTf &shtonecurve, float *rtemp, float *gtemp, float *btemp, int istart, int tH, int jstart, int tW, int tileSize) {
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat cr = F2V(0.299f); vfloat cr = F2V(0.299f);
vfloat cg = F2V(0.587f); vfloat cg = F2V(0.587f);
vfloat cb = F2V(0.114f); vfloat cb = F2V(0.114f);
@ -61,7 +61,7 @@ void shadowToneCurve(const LUTf &shtonecurve, float *rtemp, float *gtemp, float
for (int i = istart, ti = 0; i < tH; i++, ti++) { for (int i = istart, ti = 0; i < tH; i++, ti++) {
int j = jstart, tj = 0; int j = jstart, tj = 0;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for (; j < tW - 3; j+=4, tj+=4) { for (; j < tW - 3; j+=4, tj+=4) {
vfloat rv = LVF(rtemp[ti * tileSize + tj]); vfloat rv = LVF(rtemp[ti * tileSize + tj]);
@ -94,14 +94,14 @@ void shadowToneCurve(const LUTf &shtonecurve, float *rtemp, float *gtemp, float
void highlightToneCurve(const LUTf &hltonecurve, float *rtemp, float *gtemp, float *btemp, int istart, int tH, int jstart, int tW, int tileSize, float exp_scale, float comp, float hlrange) { void highlightToneCurve(const LUTf &hltonecurve, float *rtemp, float *gtemp, float *btemp, int istart, int tH, int jstart, int tW, int tileSize, float exp_scale, float comp, float hlrange) {
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat threev = F2V(3.f); vfloat threev = F2V(3.f);
vfloat maxvalfv = F2V(MAXVALF); vfloat maxvalfv = F2V(MAXVALF);
#endif #endif
for (int i = istart, ti = 0; i < tH; i++, ti++) { for (int i = istart, ti = 0; i < tH; i++, ti++) {
int j = jstart, tj = 0; int j = jstart, tj = 0;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for (; j < tW - 3; j+=4, tj+=4) { for (; j < tW - 3; j+=4, tj+=4) {
vfloat rv = LVF(rtemp[ti * tileSize + tj]); vfloat rv = LVF(rtemp[ti * tileSize + tj]);
@ -158,7 +158,7 @@ void proPhotoBlue(float *rtemp, float *gtemp, float *btemp, int istart, int tH,
// this is a hack to avoid the blue=>black bug (Issue 2141) // this is a hack to avoid the blue=>black bug (Issue 2141)
for (int i = istart, ti = 0; i < tH; i++, ti++) { for (int i = istart, ti = 0; i < tH; i++, ti++) {
int j = jstart, tj = 0; int j = jstart, tj = 0;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for (; j < tW - 3; j+=4, tj+=4) { for (; j < tW - 3; j+=4, tj+=4) {
vfloat rv = LVF(rtemp[ti * tileSize + tj]); vfloat rv = LVF(rtemp[ti * tileSize + tj]);
vfloat gv = LVF(gtemp[ti * tileSize + tj]); vfloat gv = LVF(gtemp[ti * tileSize + tj]);
@ -3763,7 +3763,7 @@ void ImProcFunctions::rgbProc (Imagefloat* working, LabImage* lab, PipetteBuffer
} else { } else {
for (int i = istart, ti = 0; i < tH; i++, ti++) { for (int i = istart, ti = 0; i < tH; i++, ti++) {
int j = jstart, tj = 0; int j = jstart, tj = 0;
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for (; j < tW - 3; j+=4, tj+=4) { for (; j < tW - 3; j+=4, tj+=4) {
//brightness/contrast //brightness/contrast
STVF(rtemp[ti * TS + tj], tonecurve(LVF(rtemp[ti * TS + tj]))); STVF(rtemp[ti * TS + tj], tonecurve(LVF(rtemp[ti * TS + tj])));
@ -5391,7 +5391,7 @@ void ImProcFunctions::luminanceCurve (LabImage* lold, LabImage* lnew, LUTf & cur
SSEFUNCTION void ImProcFunctions::chromiLuminanceCurve (PipetteBuffer *pipetteBuffer, int pW, LabImage* lold, LabImage* lnew, LUTf & acurve, LUTf & bcurve, LUTf & satcurve, LUTf & lhskcurve, LUTf & clcurve, LUTf & curve, bool utili, bool autili, bool butili, bool ccutili, bool cclutili, bool clcutili, LUTu &histCCurve, LUTu &histLCurve) void ImProcFunctions::chromiLuminanceCurve (PipetteBuffer *pipetteBuffer, int pW, LabImage* lold, LabImage* lnew, LUTf & acurve, LUTf & bcurve, LUTf & satcurve, LUTf & lhskcurve, LUTf & clcurve, LUTf & curve, bool utili, bool autili, bool butili, bool ccutili, bool cclutili, bool clcutili, LUTu &histCCurve, LUTu &histLCurve)
{ {
if (!params->labCurve.enabled) { if (!params->labCurve.enabled) {
return; return;
@ -6939,7 +6939,7 @@ void ImProcFunctions::rgb2lab (const Imagefloat &src, LabImage &dst, const Glib:
} }
} }
SSEFUNCTION void ImProcFunctions::lab2rgb (const LabImage &src, Imagefloat &dst, const Glib::ustring &workingSpace) void ImProcFunctions::lab2rgb (const LabImage &src, Imagefloat &dst, const Glib::ustring &workingSpace)
{ {
TMatrix wiprof = ICCStore::getInstance()->workingSpaceInverseMatrix ( workingSpace ); TMatrix wiprof = ICCStore::getInstance()->workingSpaceInverseMatrix ( workingSpace );
const float wip[3][3] = { const float wip[3][3] = {

4
rtengine/impulse_denoise.cc
View File

@ -31,7 +31,7 @@ using namespace std;
namespace rtengine namespace rtengine
{ {
SSEFUNCTION void ImProcFunctions::impulse_nr (LabImage* lab, double thresh) void ImProcFunctions::impulse_nr (LabImage* lab, double thresh)
{ {
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% // %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// impulse noise removal // impulse noise removal
@ -260,7 +260,7 @@ SSEFUNCTION void ImProcFunctions::impulse_nr (LabImage* lab, double thresh)
} }
SSEFUNCTION void ImProcFunctions::impulse_nrcam (CieImage* ncie, double thresh, float **buffers[3]) void ImProcFunctions::impulse_nrcam (CieImage* ncie, double thresh, float **buffers[3])
{ {
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% // %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// impulse noise removal // impulse noise removal

2
rtengine/ipresize.cc
View File

@ -178,7 +178,7 @@ void ImProcFunctions::Lanczos (const Imagefloat* src, Imagefloat* dst, float sca
} }
SSEFUNCTION void ImProcFunctions::Lanczos (const LabImage* src, LabImage* dst, float scale) void ImProcFunctions::Lanczos (const LabImage* src, LabImage* dst, float scale)
{ {
const float delta = 1.0f / scale; const float delta = 1.0f / scale;
const float a = 3.0f; const float a = 3.0f;

2
rtengine/ipsharpen.cc
View File

@ -34,7 +34,7 @@ namespace rtengine
#define ABS(a) ((a)<0?-(a):(a)) #define ABS(a) ((a)<0?-(a):(a))
extern const Settings* settings; extern const Settings* settings;
SSEFUNCTION void ImProcFunctions::dcdamping (float** aI, float** aO, float damping, int W, int H) void ImProcFunctions::dcdamping (float** aI, float** aO, float damping, int W, int H)
{ {
const float dampingFac = -2.0 / (damping * damping); const float dampingFac = -2.0 / (damping * damping);

84
rtengine/ipwavelet.cc
View File

@ -140,7 +140,7 @@ struct cont_params {
int wavNestedLevels = 1; int wavNestedLevels = 1;
SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int kall, const procparams::WaveletParams & waparams, const WavCurve & wavCLVCcurve, const WavOpacityCurveRG & waOpacityCurveRG, const WavOpacityCurveBY & waOpacityCurveBY, const WavOpacityCurveW & waOpacityCurveW, const WavOpacityCurveWL & waOpacityCurveWL, LUTf &wavclCurve, int skip) void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int kall, const procparams::WaveletParams & waparams, const WavCurve & wavCLVCcurve, const WavOpacityCurveRG & waOpacityCurveRG, const WavOpacityCurveBY & waOpacityCurveBY, const WavOpacityCurveW & waOpacityCurveW, const WavOpacityCurveWL & waOpacityCurveWL, LUTf &wavclCurve, int skip)
{ {
@ -645,7 +645,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
numthreads = MIN(numthreads, maxnumberofthreadsforwavelet); numthreads = MIN(numthreads, maxnumberofthreadsforwavelet);
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
wavNestedLevels = omp_get_max_threads() / numthreads; wavNestedLevels = omp_get_max_threads() / numthreads;
bool oldNested = omp_get_nested(); bool oldNested = omp_get_nested();
@ -720,7 +720,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
Lold = lab->L; Lold = lab->L;
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@ -784,7 +784,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@ -868,7 +868,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
if(!Ldecomp->memoryAllocationFailed) { if(!Ldecomp->memoryAllocationFailed) {
float madL[8][3]; float madL[8][3];
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) collapse(2) num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for schedule(dynamic) collapse(2) num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@ -1076,7 +1076,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
bool highlight = params->toneCurve.hrenabled; bool highlight = params->toneCurve.hrenabled;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16) num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for schedule(dynamic,16) num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@ -1246,7 +1246,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
delete [] varchro; delete [] varchro;
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
omp_set_nested(oldNested); omp_set_nested(oldNested);
#endif #endif
@ -1276,12 +1276,12 @@ void ImProcFunctions::Aver( float * RESTRICT DataList, int datalen, float &aver
float thres = 5.f;//different fom zero to take into account only data large enough float thres = 5.f;//different fom zero to take into account only data large enough
max = 0.f; max = 0.f;
min = 0.f; min = 0.f;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
{ {
float lmax = 0.f, lmin = 0.f; float lmax = 0.f, lmin = 0.f;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for reduction(+:averaP,averaN,countP,countN) nowait #pragma omp for reduction(+:averaP,averaN,countP,countN) nowait
#endif #endif
@ -1305,7 +1305,7 @@ void ImProcFunctions::Aver( float * RESTRICT DataList, int datalen, float &aver
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp critical #pragma omp critical
#endif #endif
{ {
@ -1335,7 +1335,7 @@ void ImProcFunctions::Sigma( float * RESTRICT DataList, int datalen, float aver
float variP = 0.f, variN = 0.f; float variP = 0.f, variN = 0.f;
float thres = 5.f;//different fom zero to take into account only data large enough float thres = 5.f;//different fom zero to take into account only data large enough
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:variP,variN,countP,countN) num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for reduction(+:variP,variN,countP,countN) num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@ -1434,7 +1434,7 @@ float *ImProcFunctions::ContrastDR(float *Source, int W_L, int H_L, float *Contr
} }
memcpy(Contrast, Source, n * sizeof(float)); memcpy(Contrast, Source, n * sizeof(float));
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for #pragma omp parallel for
#endif #endif
@ -1445,19 +1445,19 @@ float *ImProcFunctions::ContrastDR(float *Source, int W_L, int H_L, float *Contr
return Contrast; return Contrast;
} }
SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int W_L, int H_L, float Compression, float DetailBoost, float *Compressed) float *ImProcFunctions::CompressDR(float *Source, int W_L, int H_L, float Compression, float DetailBoost, float *Compressed)
{ {
const float eps = 0.000001f; const float eps = 0.000001f;
int n = W_L * H_L; int n = W_L * H_L;
#ifdef __SSE2__ #ifdef __SSE2__
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel #pragma omp parallel
#endif #endif
{ {
__m128 epsv = _mm_set1_ps( eps ); __m128 epsv = _mm_set1_ps( eps );
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -1471,7 +1471,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int W_L, int H_L,
} }
#else #else
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for #pragma omp parallel for
#endif #endif
@ -1508,7 +1508,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int W_L, int H_L,
} }
#ifdef __SSE2__ #ifdef __SSE2__
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel #pragma omp parallel
#endif #endif
{ {
@ -1516,7 +1516,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int W_L, int H_L,
__m128 epsv = _mm_set1_ps( eps ); __m128 epsv = _mm_set1_ps( eps );
__m128 DetailBoostv = _mm_set1_ps( DetailBoost ); __m128 DetailBoostv = _mm_set1_ps( DetailBoost );
__m128 tempv = _mm_set1_ps( temp ); __m128 tempv = _mm_set1_ps( temp );
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -1537,7 +1537,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int W_L, int H_L,
} }
#else #else
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for #pragma omp parallel for
#endif #endif
@ -1572,7 +1572,7 @@ void ImProcFunctions::ContrastResid(float * WavCoeffs_L0, struct cont_params &cp
min0 = 0.0f; min0 = 0.0f;
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for #pragma omp parallel for
#endif #endif
@ -1592,7 +1592,7 @@ void ImProcFunctions::ContrastResid(float * WavCoeffs_L0, struct cont_params &cp
CompressDR(WavCoeffs_L0, W_L, H_L, Compression, DetailBoost, WavCoeffs_L0); CompressDR(WavCoeffs_L0, W_L, H_L, Compression, DetailBoost, WavCoeffs_L0);
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for // removed schedule(dynamic,10) #pragma omp parallel for // removed schedule(dynamic,10)
#endif #endif
@ -1624,7 +1624,7 @@ void ImProcFunctions::EPDToneMapResid(float * WavCoeffs_L0, unsigned int Iterat
} }
// max0=32768.f; // max0=32768.f;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for #pragma omp parallel for
#endif #endif
@ -1649,7 +1649,7 @@ void ImProcFunctions::EPDToneMapResid(float * WavCoeffs_L0, unsigned int Iterat
epd2.CompressDynamicRange(WavCoeffs_L0, (float)sca / skip, edgest, Compression, DetailBoost, Iterates, rew); epd2.CompressDynamicRange(WavCoeffs_L0, (float)sca / skip, edgest, Compression, DetailBoost, Iterates, rew);
//Restore past range, also desaturate a bit per Mantiuk's Color correction for tone mapping. //Restore past range, also desaturate a bit per Mantiuk's Color correction for tone mapping.
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for // removed schedule(dynamic,10) #pragma omp parallel for // removed schedule(dynamic,10)
#endif #endif
@ -1692,7 +1692,7 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
float min0 = FLT_MAX; float min0 = FLT_MAX;
if(contrast != 0.f || (cp.tonemap && cp.resena)) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step if(contrast != 0.f || (cp.tonemap && cp.resena)) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:avedbl) num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for reduction(+:avedbl) num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@ -1700,14 +1700,14 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
avedbl += WavCoeffs_L0[i]; avedbl += WavCoeffs_L0[i];
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
{ {
float lminL = FLT_MAX; float lminL = FLT_MAX;
float lmaxL = 0.f; float lmaxL = 0.f;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -1722,7 +1722,7 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp critical #pragma omp critical
#endif #endif
{ {
@ -1782,13 +1782,13 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
koeLi[j][i] = 0.f; koeLi[j][i] = 0.f;
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
{ {
if(contrast != 0.f && cp.resena) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step if(contrast != 0.f && cp.resena) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step
{ {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for #pragma omp for
#endif #endif
@ -1817,18 +1817,18 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
if(cp.tonemap && cp.contmet == 1 && cp.resena) { if(cp.tonemap && cp.contmet == 1 && cp.resena) {
float maxp = max0 * 256.f; float maxp = max0 * 256.f;
float minp = min0 * 256.f; float minp = min0 * 256.f;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp single #pragma omp single
#endif #endif
ContrastResid(WavCoeffs_L0, cp, W_L, H_L, maxp, minp); ContrastResid(WavCoeffs_L0, cp, W_L, H_L, maxp, minp);
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp barrier #pragma omp barrier
#endif #endif
if((cp.conres != 0.f || cp.conresH != 0.f) && cp.resena) { // cp.conres = 0.f and cp.comresH = 0.f means that all will be multiplied by 1.f, so we can skip this step if((cp.conres != 0.f || cp.conresH != 0.f) && cp.resena) { // cp.conres = 0.f and cp.comresH = 0.f means that all will be multiplied by 1.f, so we can skip this step
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for nowait #pragma omp for nowait
#endif #endif
@ -1890,7 +1890,7 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
tmC[i] = &tmCBuffer[i * W_L]; tmC[i] = &tmCBuffer[i * W_L];
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -1910,7 +1910,7 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
float aamp = 1.f + cp.eddetthrHi / 100.f; float aamp = 1.f + cp.eddetthrHi / 100.f;
for (int lvl = 0; lvl < 4; lvl++) { for (int lvl = 0; lvl < 4; lvl++) {
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic,16) #pragma omp for schedule(dynamic,16)
#endif #endif
@ -2013,7 +2013,7 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
// end // end
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2048,7 +2048,7 @@ void ImProcFunctions::WaveletAandBAllAB(wavelet_decomposition &WaveletCoeffs_a,
float * WavCoeffs_a0 = WaveletCoeffs_a.coeff0; float * WavCoeffs_a0 = WaveletCoeffs_a.coeff0;
float * WavCoeffs_b0 = WaveletCoeffs_b.coeff0; float * WavCoeffs_b0 = WaveletCoeffs_b.coeff0;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
{ {
@ -2056,7 +2056,7 @@ void ImProcFunctions::WaveletAandBAllAB(wavelet_decomposition &WaveletCoeffs_a,
float huebuffer[W_L] ALIGNED64; float huebuffer[W_L] ALIGNED64;
float chrbuffer[W_L] ALIGNED64; float chrbuffer[W_L] ALIGNED64;
#endif // __SSE2__ #endif // __SSE2__
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic,16) #pragma omp for schedule(dynamic,16)
#endif #endif
@ -2116,13 +2116,13 @@ void ImProcFunctions::WaveletcontAllAB(LabImage * labco, float ** varhue, float
float * WavCoeffs_ab0 = WaveletCoeffs_ab.coeff0; float * WavCoeffs_ab0 = WaveletCoeffs_ab.coeff0;
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
{ {
if(cp.chrores != 0.f && cp.resena) { // cp.chrores == 0.f means all will be multiplied by 1.f, so we can skip the processing of residual if(cp.chrores != 0.f && cp.resena) { // cp.chrores == 0.f means all will be multiplied by 1.f, so we can skip the processing of residual
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for nowait #pragma omp for nowait
#endif #endif
@ -2175,7 +2175,7 @@ void ImProcFunctions::WaveletcontAllAB(LabImage * labco, float ** varhue, float
if(cp.cbena && cp.resena) {//if user select Toning and color balance if(cp.cbena && cp.resena) {//if user select Toning and color balance
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for nowait #pragma omp for nowait
#endif #endif
@ -2230,7 +2230,7 @@ void ImProcFunctions::WaveletcontAllAB(LabImage * labco, float ** varhue, float
} }
} }
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) #pragma omp for schedule(dynamic) collapse(2)
#endif #endif
@ -2425,7 +2425,7 @@ void ImProcFunctions::finalContAllL (float ** WavCoeffs_L, float * WavCoeffs_L0,
float bsig = 0.5f - asig * mean[level]; float bsig = 0.5f - asig * mean[level];
float amean = 0.5f / mean[level]; float amean = 0.5f / mean[level];
#ifdef _RT_NESTED_OPENMP #ifdef _OPENMP
#pragma omp parallel for schedule(dynamic, W_L * 16) num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for schedule(dynamic, W_L * 16) num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif

4
rtengine/lcp.cc
View File

@ -1131,7 +1131,7 @@ void rtengine::LCPMapper::correctCA(double& x, double& y, int cx, int cy, int ch
y -= cy; y -= cy;
} }
SSEFUNCTION void rtengine::LCPMapper::processVignetteLine(int width, int y, float* line) const void rtengine::LCPMapper::processVignetteLine(int width, int y, float* line) const
{ {
// No need for swapXY, since vignette is in RAW and always before rotation // No need for swapXY, since vignette is in RAW and always before rotation
float yd = ((float)y - mc.y0) * mc.rfy; float yd = ((float)y - mc.y0) * mc.rfy;
@ -1169,7 +1169,7 @@ SSEFUNCTION void rtengine::LCPMapper::processVignetteLine(int width, int y, floa
} }
} }
SSEFUNCTION void rtengine::LCPMapper::processVignetteLine3Channels(int width, int y, float* line) const void rtengine::LCPMapper::processVignetteLine3Channels(int width, int y, float* line) const
{ {
// No need for swapXY, since vignette is in RAW and always before rotation // No need for swapXY, since vignette is in RAW and always before rotation
float yd = ((float)y - mc.y0) * mc.rfy; float yd = ((float)y - mc.y0) * mc.rfy;

37
rtengine/opthelper.h
View File

@ -26,44 +26,14 @@
#ifdef __SSE2__ #ifdef __SSE2__
#include "sleefsseavx.c" #include "sleefsseavx.c"
#ifdef __GNUC__
#if defined(WIN32) && !defined( __x86_64__ )
// needed for actual versions of GCC with 32-Bit Windows
#define SSEFUNCTION __attribute__((force_align_arg_pointer))
#else
#define SSEFUNCTION
#endif
#else
#define SSEFUNCTION
#endif
#else
#ifdef __SSE__
#ifdef __GNUC__
#if defined(WIN32) && !defined( __x86_64__ )
// needed for actual versions of GCC with 32-Bit Windows
#define SSEFUNCTION __attribute__((force_align_arg_pointer))
#else
#define SSEFUNCTION
#endif
#else
#define SSEFUNCTION
#endif
#else
#define SSEFUNCTION
#endif
#endif #endif
#ifdef __GNUC__ #ifdef __GNUC__
#define RESTRICT __restrict__ #define RESTRICT __restrict__
#define LIKELY(x) __builtin_expect (!!(x), 1) #define LIKELY(x) __builtin_expect (!!(x), 1)
#define UNLIKELY(x) __builtin_expect (!!(x), 0) #define UNLIKELY(x) __builtin_expect (!!(x), 0)
#if (!defined(WIN32) || defined( __x86_64__ )) #define ALIGNED64 __attribute__ ((aligned (64)))
#define ALIGNED64 __attribute__ ((aligned (64))) #define ALIGNED16 __attribute__ ((aligned (16)))
#define ALIGNED16 __attribute__ ((aligned (16)))
#else // there is a bug in gcc 4.7.x when using openmp and aligned memory and -O3, also needed for WIN32 builds
#define ALIGNED64
#define ALIGNED16
#endif
#else #else
#define RESTRICT #define RESTRICT
#define LIKELY(x) (x) #define LIKELY(x) (x)
@ -71,7 +41,4 @@
#define ALIGNED64 #define ALIGNED64
#define ALIGNED16 #define ALIGNED16
#endif #endif
#if defined _OPENMP
#define _RT_NESTED_OPENMP
#endif
#endif #endif

4
rtengine/rawimagesource.cc
View File

@ -1341,7 +1341,7 @@ int RawImageSource::interpolateBadPixelsXtrans( PixelsMap &bitmapBads )
* (Taken from Emil Martinec idea) * (Taken from Emil Martinec idea)
* (Optimized by Ingo Weyrich 2013 and 2015) * (Optimized by Ingo Weyrich 2013 and 2015)
*/ */
SSEFUNCTION int RawImageSource::findHotDeadPixels( PixelsMap &bpMap, float thresh, bool findHotPixels, bool findDeadPixels ) int RawImageSource::findHotDeadPixels( PixelsMap &bpMap, float thresh, bool findHotPixels, bool findDeadPixels )
{ {
float varthresh = (20.0 * (thresh / 100.0) + 1.0 ) / 24.f; float varthresh = (20.0 * (thresh / 100.0) + 1.0 ) / 24.f;
@ -3169,7 +3169,7 @@ void RawImageSource::copyOriginalPixels(const RAWParams &raw, RawImage *src, Raw
} }
} }
SSEFUNCTION void RawImageSource::cfaboxblur(RawImage *riFlatFile, float* cfablur, const int boxH, const int boxW) void RawImageSource::cfaboxblur(RawImage *riFlatFile, float* cfablur, const int boxH, const int boxW)
{ {
if(boxW == 0 && boxH == 0) { // nothing to blur if(boxW == 0 && boxH == 0) { // nothing to blur

10
rtengine/shmap.cc
View File

@ -358,7 +358,7 @@ void SHMap::forceStat (float max_, float min_, float avg_)
avg = avg_; avg = avg_;
} }
SSEFUNCTION void SHMap::dirpyr_shmap(float ** data_fine, float ** data_coarse, int width, int height, LUTf & rangefn, int level, int scale) void SHMap::dirpyr_shmap(float ** data_fine, float ** data_coarse, int width, int height, LUTf & rangefn, int level, int scale)
{ {
//scale is spacing of directional averaging weights //scale is spacing of directional averaging weights
@ -375,7 +375,7 @@ SSEFUNCTION void SHMap::dirpyr_shmap(float ** data_fine, float ** data_coarse, i
#pragma omp parallel #pragma omp parallel
#endif #endif
{ {
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat dirwtv, valv, normv, dftemp1v, dftemp2v; vfloat dirwtv, valv, normv, dftemp1v, dftemp2v;
#endif // __SSE2__ #endif // __SSE2__
int j; int j;
@ -402,7 +402,7 @@ SSEFUNCTION void SHMap::dirpyr_shmap(float ** data_fine, float ** data_coarse, i
data_coarse[i][j] = val / norm; // low pass filter data_coarse[i][j] = val / norm; // low pass filter
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
int inbrMin = max(i - scalewin, i % scale); int inbrMin = max(i - scalewin, i % scale);
for(; j < (width - scalewin) - 3; j += 4) { for(; j < (width - scalewin) - 3; j += 4) {
@ -482,7 +482,7 @@ SSEFUNCTION void SHMap::dirpyr_shmap(float ** data_fine, float ** data_coarse, i
#pragma omp parallel #pragma omp parallel
#endif #endif
{ {
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
vfloat dirwtv, valv, normv, dftemp1v, dftemp2v; vfloat dirwtv, valv, normv, dftemp1v, dftemp2v;
float domkerv[5][5][4] ALIGNED16 = {{{1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {2, 2, 2, 2}, {2, 2, 2, 2}, {2, 2, 2, 2}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {2, 2, 2, 2}, {2, 2, 2, 2}, {2, 2, 2, 2}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {2, 2, 2, 2}, {2, 2, 2, 2}, {2, 2, 2, 2}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}}}; float domkerv[5][5][4] ALIGNED16 = {{{1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {2, 2, 2, 2}, {2, 2, 2, 2}, {2, 2, 2, 2}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {2, 2, 2, 2}, {2, 2, 2, 2}, {2, 2, 2, 2}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {2, 2, 2, 2}, {2, 2, 2, 2}, {2, 2, 2, 2}, {1, 1, 1, 1}}, {{1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1}}};
@ -510,7 +510,7 @@ SSEFUNCTION void SHMap::dirpyr_shmap(float ** data_fine, float ** data_coarse, i
data_coarse[i][j] = val / norm; // low pass filter data_coarse[i][j] = val / norm; // low pass filter
} }
#if defined( __SSE2__ ) && defined( __x86_64__ ) #ifdef __SSE2__
for(; j < width - scalewin - 3; j += 4) { for(; j < width - scalewin - 3; j += 4) {
valv = _mm_setzero_ps(); valv = _mm_setzero_ps();

2
rtengine/sleefsseavx.c
View File

@ -906,7 +906,7 @@ typedef struct {
static INLINE vfloat vabsf(vfloat f) { return (vfloat)vandnotm((vmask)vcast_vf_f(-0.0f), (vmask)f); } static INLINE vfloat vabsf(vfloat f) { return (vfloat)vandnotm((vmask)vcast_vf_f(-0.0f), (vmask)f); }
static INLINE vfloat vnegf(vfloat f) { return (vfloat)vxorm((vmask)f, (vmask)vcast_vf_f(-0.0f)); } static INLINE vfloat vnegf(vfloat f) { return (vfloat)vxorm((vmask)f, (vmask)vcast_vf_f(-0.0f)); }
#if defined( __SSE4_1__ ) && defined( __x86_64__ ) #ifdef __SSE4_1__
// only one instruction when using SSE4.1 // only one instruction when using SSE4.1
static INLINE vfloat vself(vmask mask, vfloat x, vfloat y) { static INLINE vfloat vself(vmask mask, vfloat x, vfloat y) {
return _mm_blendv_ps(y,x,(vfloat)mask); return _mm_blendv_ps(y,x,(vfloat)mask);

6
rtgui/bqentryupdater.cc
View File

@ -73,14 +73,8 @@ void BatchQueueEntryUpdater::process (guint8* oimg, int ow, int oh, int newh, BQ
stopped = false; stopped = false;
tostop = false; tostop = false;
#if __GNUC__ == 4 && __GNUC_MINOR__ == 8 && defined( WIN32 ) && defined(__x86_64__)
#undef THREAD_PRIORITY_NORMAL
// See Issue 2384 comment #3
thread = Glib::Thread::create(sigc::mem_fun(*this, &BatchQueueEntryUpdater::processThread), (unsigned long int)0, true, true, Glib::THREAD_PRIORITY_NORMAL);
#else
#undef THREAD_PRIORITY_LOW #undef THREAD_PRIORITY_LOW
thread = Glib::Thread::create(sigc::mem_fun(*this, &BatchQueueEntryUpdater::processThread), (unsigned long int)0, true, true, Glib::THREAD_PRIORITY_LOW); thread = Glib::Thread::create(sigc::mem_fun(*this, &BatchQueueEntryUpdater::processThread), (unsigned long int)0, true, true, Glib::THREAD_PRIORITY_LOW);
#endif
} }
} }

2
rtgui/histogrampanel.cc
View File

@ -944,7 +944,7 @@ void HistogramArea::update (LUTu &histRed, LUTu &histGreen, LUTu &histBlue, LUTu
idle_register.add(func, haih); idle_register.add(func, haih);
} }
SSEFUNCTION void HistogramArea::updateBackBuffer () void HistogramArea::updateBackBuffer ()
{ {
if (!get_realized ()) { if (!get_realized ()) {