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

wavcont(): use per level/dir beta values for all methods, speedup process 2, reduce memory usage, fix a race

Browse Source
This commit is contained in:
Ingo Weyrich 2020-06-21 11:39:58 +02:00
parent 227a2ad150
commit 2dd2313e02
2 changed files with 105 additions and 296 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
rtengine/improcfun.h
View File

@ -338,7 +338,7 @@ public:
const LocwavCurve & locedgwavCurve, bool locedgwavutili, const LocwavCurve & locedgwavCurve, bool locedgwavutili,
float sigm, float offs,int & maxlvl, float fatdet, float fatanch, float chromalev, float chromablu, bool blurlc, bool blurena, bool levelena, bool comprena, bool compreena, float compress, float thres); float sigm, float offs,int & maxlvl, float fatdet, float fatanch, float chromalev, float chromablu, bool blurlc, bool blurena, bool levelena, bool comprena, bool compreena, float compress, float thres);
void wavcont(const struct local_params& lp, float ** tmp, wavelet_decomposition &wdspot, float ****templevel, int level_bl, int maxlvl, void wavcont(const struct local_params& lp, float ** tmp, wavelet_decomposition &wdspot, int level_bl, int maxlvl,
const LocwavCurve & loclevwavCurve, bool loclevwavutili, const LocwavCurve & loclevwavCurve, bool loclevwavutili,
const LocwavCurve & loccompwavCurve, bool loccompwavutili, const LocwavCurve & loccompwavCurve, bool loccompwavutili,
const LocwavCurve & loccomprewavCurve, bool loccomprewavutili, const LocwavCurve & loccomprewavCurve, bool loccomprewavutili,
@ -355,7 +355,7 @@ public:
void Exclude_Local(float **deltaso, float hueref, float chromaref, float lumaref, float sobelref, float meansobel, const struct local_params & lp, const LabImage * original, LabImage * transformed, const LabImage * rsv, const LabImage * reserv, int cx, int cy, int sk); void Exclude_Local(float **deltaso, float hueref, float chromaref, float lumaref, float sobelref, float meansobel, const struct local_params & lp, const LabImage * original, LabImage * transformed, const LabImage * rsv, const LabImage * reserv, int cx, int cy, int sk);
void DeNoise_Local(int call, const struct local_params& lp, LabImage* originalmask, int levred, float hueref, float lumaref, float chromaref, LabImage* original, LabImage* transformed, const LabImage &tmp1, int cx, int cy, int sk); void DeNoise_Local(int call, const struct local_params& lp, LabImage* originalmask, int levred, float hueref, float lumaref, float chromaref, LabImage* original, LabImage* transformed, const LabImage &tmp1, int cx, int cy, int sk);
void DeNoise(int call, int del, float * slidL, float * slida, float * slidb, int aut, bool noiscfactiv, struct local_params& lp, LabImage* originalmaskbl, int levred, float huerefblur, float lumarefblur, float chromarefblur, LabImage* original, LabImage* transformed, int cx, int cy, int sk); void DeNoise(int call, int del, float * slidL, float * slida, float * slidb, int aut, bool noiscfactiv, const struct local_params& lp, LabImage* originalmaskbl, int levred, float huerefblur, float lumarefblur, float chromarefblur, LabImage* original, LabImage* transformed, int cx, int cy, int sk);
void fftw_denoise(int GW, int GH, int max_numblox_W, int min_numblox_W, float **tmp1, array2D<float> *Lin, int numThreads, const struct local_params & lp, int chrom); void fftw_denoise(int GW, int GH, int max_numblox_W, int min_numblox_W, float **tmp1, array2D<float> *Lin, int numThreads, const struct local_params & lp, int chrom);

279
rtengine/iplocallab.cc
View File

@ -6951,7 +6951,7 @@ void ImProcFunctions::Compresslevels(float **Source, int W_L, int H_L, float com
} }
} }
void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavelet_decomposition& wdspot, float ****templevel, int level_bl, int maxlvl, void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavelet_decomposition& wdspot, int level_bl, int maxlvl,
const LocwavCurve & loclevwavCurve, bool loclevwavutili, const LocwavCurve & loclevwavCurve, bool loclevwavutili,
const LocwavCurve & loccompwavCurve, bool loccompwavutili, const LocwavCurve & loccompwavCurve, bool loccompwavutili,
const LocwavCurve & loccomprewavCurve, bool loccomprewavutili, const LocwavCurve & loccomprewavCurve, bool loccomprewavutili,
@ -6963,7 +6963,11 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
const int H_L = wdspot.level_H(0); const int H_L = wdspot.level_H(0);
const std::unique_ptr<float[]> beta(new float[W_L * H_L]); const std::unique_ptr<float[]> beta(new float[W_L * H_L]);
array2D<float> templevel;
if (process == 1 || process == 3) {
templevel(W_L, H_L);
}
#ifdef _OPENMP #ifdef _OPENMP
const int numThreads = omp_get_max_threads(); const int numThreads = omp_get_max_threads();
#else #else
@ -6979,13 +6983,12 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
float MaxN[10]; float MaxN[10];
Evaluate2(wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN, numThreads); Evaluate2(wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN, numThreads);
for (int dir = 3; dir >= 1; --dir) { for (int dir = 1; dir < 4; ++dir) {
for (int level = maxlvl - 1; level >= level_bl; --level) { for (int level = level_bl; level < maxlvl; ++level) {
const int W_L = wdspot.level_W(level); const int W_L = wdspot.level_W(level);
const int H_L = wdspot.level_H(level); const int H_L = wdspot.level_H(level);
const auto WavL = wdspot.level_coeffs(level)[dir]; const auto WavL = wdspot.level_coeffs(level)[dir];
if (dir == 3 && level + 1 == maxlvl) {
const float effect = lp.sigmabl; const float effect = lp.sigmabl;
constexpr float offs = 1.f; constexpr float offs = 1.f;
float mea[10]; float mea[10];
@ -7021,7 +7024,6 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
beta[co] = 0.05f; beta[co] = 0.05f;
} }
} }
}
const float klev = 0.25f * loclevwavCurve[level * 55.5f]; const float klev = 0.25f * loclevwavCurve[level * 55.5f];
float* src[H_L]; float* src[H_L];
@ -7032,7 +7034,7 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
#pragma omp parallel if (multiThread) #pragma omp parallel if (multiThread)
#endif #endif
{ {
gaussianBlur(src, templevel[0][0], W_L, H_L, radlevblur * klev * chromablu); gaussianBlur(src, templevel, W_L, H_L, radlevblur * klev * chromablu);
} }
#ifdef _OPENMP #ifdef _OPENMP
@ -7041,28 +7043,12 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
for (int y = 0; y < H_L; y++) { for (int y = 0; y < H_L; y++) {
for (int x = 0; x < W_L; x++) { for (int x = 0; x < W_L; x++) {
int j = y * W_L + x; int j = y * W_L + x;
WavL[j] = intp(beta[j], templevel[0][0][y][x], WavL[j]); WavL[j] = intp(beta[j], templevel[y][x], WavL[j]);
} }
} }
} }
} }
} else if (process == 2) { //Directional contrast } else if (process == 2) { //Directional contrast
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) collapse(2) if (multiThread)
#endif
for (int dir = 1; dir < 4; dir++) {
for (int level = level_bl; level < maxlvl; ++level) {
const int W_L = wdspot.level_W(level);
const int H_L = wdspot.level_H(level);
const auto wav_L = wdspot.level_coeffs(level)[dir];
for (int y = 0; y < H_L; y++) {
for (int x = 0; x < W_L; x++) {
float val = wav_L[y * W_L + x];
templevel[dir - 1][level][y][x] = val;
}
}
}
}
float mean[10]; float mean[10];
float meanN[10]; float meanN[10];
float sigma[10]; float sigma[10];
@ -7071,21 +7057,21 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
float MaxN[10]; float MaxN[10];
Evaluate2(wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN, numThreads); Evaluate2(wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN, numThreads);
#ifdef _OPENMP for (int dir = 1; dir < 4; ++dir) {
#pragma omp parallel for schedule(dynamic) collapse(2) if (multiThread)
#endif
for (int dir = 1; dir < 4; dir++) {
for (int level = level_bl; level < maxlvl; ++level) { for (int level = level_bl; level < maxlvl; ++level) {
int W_L = wdspot.level_W(level); const int W_L = wdspot.level_W(level);
int H_L = wdspot.level_H(level); const int H_L = wdspot.level_H(level);
float effect = sigmadc; auto WavL = wdspot.level_coeffs(level)[dir];
float offs = 1.f;
const float effect = sigmadc;
constexpr float offs = 1.f;
float mea[10]; float mea[10];
calceffect(level, mean, sigma, mea, effect, offs); calceffect(level, mean, sigma, mea, effect, offs);
auto WavL = wdspot.level_coeffs(level); #ifdef _OPENMP
#pragma omp parallel for if (multiThread)
#endif
for (int co = 0; co < H_L * W_L; co++) { for (int co = 0; co < H_L * W_L; co++) {
const float WavCL = std::fabs(WavL[dir][co]); const float WavCL = std::fabs(WavL[co]);
if (WavCL < mea[0]) { if (WavCL < mea[0]) {
beta[co] = 0.05f; beta[co] = 0.05f;
@ -7112,10 +7098,10 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
} }
} }
int iteration = deltad; const int iteration = deltad;
int itplus = 7 + iteration; const int itplus = 7 + iteration;
int itmoins = 7 - iteration; const int itmoins = 7 - iteration;
int med = maxlvl / 2; const int med = maxlvl / 2;
int it; int it;
if (level < med) { if (level < med) {
@ -7125,44 +7111,18 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
} else { } else {
it = itplus; it = itplus;
} }
if (loccompwavCurve && loccompwavutili) { if (loccompwavCurve && loccompwavutili) {
const float itf = it;
for (int j = 0; j < it; j++) {
float kba = 1.f;
for (int i = 0; i < W_L * H_L; i++) {
int ii = i / W_L;
int jj = i - ii * W_L;
float LL100 = tmp[ii * 2][jj * 2] / 327.68f;
float k1 = 0.3f * (loccompwavCurve[6.f * LL100] - 0.5f); //k1 between 0 and 0.5 0.5==> 1/6=0.16
float k2 = k1 * 2.f;
if (dir < 3) {
kba = 1.f + k1;
}
if (dir == 3) {
kba = 1.f - k2;
}
templevel[dir - 1][level][ii][jj] *= (1.f + (kba - 1.f) * beta[i]);
}
}
}
}
}
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) collapse(2) if (multiThread) #pragma omp parallel for if (multiThread)
#endif #endif
for (int dir = 1; dir < 4; dir++) { for (int i = 0; i < H_L; ++i) {
for (int level = level_bl; level < maxlvl; ++level) { for (int j = 0; j < W_L; ++j) {
const int W_L = wdspot.level_W(level); const float LL100 = tmp[i * 2][j * 2] / 327.68f;
const int H_L = wdspot.level_H(level); const float kba = (dir < 3 ? 0.3f : -0.6f) * (loccompwavCurve[6.f * LL100] - 0.5f); //k1 between 0 and 0.5 0.5==> 1/6=0.16
const auto wav_L = wdspot.level_coeffs(level)[dir]; WavL[i * W_L + j] *= pow_F(1.f + kba * beta[i * W_L + j], itf);
}
for (int y = 0; y < H_L; y++) {
for (int x = 0; x < W_L; x++) {
int j = y * W_L + x;
wav_L[j] = templevel[dir - 1][level][y][x];
} }
} }
} }
@ -7188,12 +7148,11 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
float MaxN[10]; float MaxN[10];
Evaluate2(wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN, numThreads); Evaluate2(wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN, numThreads);
for (int dir = 3; dir >= 1; --dir) { for (int dir = 1; dir < 4; ++dir) {
for (int level = maxlvl - 1; level >= level_bl; --level) { for (int level = level_bl; level < maxlvl; ++level) {
int W_L = wdspot.level_W(level); int W_L = wdspot.level_W(level);
int H_L = wdspot.level_H(level); int H_L = wdspot.level_H(level);
if (dir == 3 && level + 1 == maxlvl) {
const float effect = lp.sigmadr; const float effect = lp.sigmadr;
constexpr float offs = 1.f; constexpr float offs = 1.f;
float mea[10]; float mea[10];
@ -7230,7 +7189,6 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
beta[co] = 0.1f; beta[co] = 0.1f;
} }
} }
}
float klev = (loccomprewavCurve[level * 55.5f] - 0.75f); float klev = (loccomprewavCurve[level * 55.5f] - 0.75f);
if (klev < 0.f) { if (klev < 0.f) {
@ -7249,18 +7207,18 @@ void ImProcFunctions::wavcont(const struct local_params& lp, float ** tmp, wavel
for (int y = 0; y < H_L; y++) { for (int y = 0; y < H_L; y++) {
for (int x = 0; x < W_L; x++) { for (int x = 0; x < W_L; x++) {
int j = y * W_L + x; int j = y * W_L + x;
templevel[0][0][y][x] = wav_L[j]; templevel[y][x] = wav_L[j];
} }
} }
Compresslevels(templevel[0][0], 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 for if (multiThread)
#endif #endif
for (int y = 0; y < H_L; y++) { for (int y = 0; y < H_L; y++) {
for (int x = 0; x < W_L; x++) { for (int x = 0; x < W_L; x++) {
int j = y * W_L + x; int j = y * W_L + x;
wav_L[j] = intp(beta[j], templevel[0][0][y][x], wav_L[j]); wav_L[j] = intp(beta[j], templevel[y][x], wav_L[j]);
} }
} }
} }
@ -7562,84 +7520,17 @@ BENCHFUN
bhigh = -ahigh * level_br; bhigh = -ahigh * level_br;
} }
int dir = 3;
int leve = maxlvl;
float ****templevel = nullptr;
float ****templevela = nullptr;
float ****templevelb = nullptr;
if (wavcurvelev || wavcurvecomp || wavcurvecompre) {//compress dynamic and blur if (wavcurvelev || wavcurvecomp || wavcurvecompre) {//compress dynamic and blur
/*
float mean[10];
float meanN[10];
float sigma[10];
float sigmaN[10];
float MaxP[10];
float MaxN[10];
Evaluate2(*wdspot, mean, meanN, sigma, sigmaN, MaxP, MaxN);
*/
// fatParams.enabled = wavcurvecomp;
templevel = new float***[dir];
//allocate memory for 3 DIR n levels, H_L, W_L
for (int d = 0; d < dir; d++) {
templevel[d] = new float**[leve];
for (int k = 0; k < leve; k++) {
templevel[d][k] = new float*[H_L];
for (int i = 0; i < H_L; i++) {
templevel[d][k][i] = new float[W_L];
}
}
}
if (templevel == nullptr) {
fprintf(stderr, "allocation error\n");
return;
}
if (wavcurvelev && radlevblur > 0.f && blurena) { if (wavcurvelev && radlevblur > 0.f && blurena) {
wavcont(lp, tmp, *wdspot, templevel, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 1, 1.f, 0.f, 0.f, 0.f); wavcont(lp, tmp, *wdspot, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 1, 1.f, 0.f, 0.f, 0.f);
} }
if (wavcurvecomp && comprena) { if (wavcurvecomp && comprena) {
wavcont(lp, tmp, *wdspot, templevel, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 2, 1.f, 0.f, sigmadc, deltad); wavcont(lp, tmp, *wdspot, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 2, 1.f, 0.f, sigmadc, deltad);
} }
if (wavcurvecompre && compreena) { if (wavcurvecompre && compreena) {
wavcont(lp, tmp, *wdspot, templevel, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 3, 1.f, thres, 0.f, 0.f); wavcont(lp, tmp, *wdspot, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 3, 1.f, thres, 0.f, 0.f);
}
//free memory templevel
if (wavcurvelev || wavcurvecomp || wavcurvecompre) {
for (int i = 0; i < dir; i++) {
for (int j = 0; j < leve; j++) {
for (int l = 0; l < H_L; l++) {
delete [] templevel[i][j][l];
}
}
}
for (int i = 0; i < dir; i++) {
for (int j = 0; j < leve; j++) {
delete [] templevel[i][j];
}
}
for (int i = 0; i < dir; i++) {
delete [] templevel[i];
}
delete [] templevel;
} }
} }
@ -8071,53 +7962,13 @@ BENCHFUN
return; return;
} }
templevela = new float***[dir];
for (int d = 0; d < dir; d++) {
templevela[d] = new float**[leve];
for (int k = 0; k < leve; k++) {
templevela[d][k] = new float*[H_L];
for (int i = 0; i < H_L; i++) {
templevela[d][k][i] = new float[W_L];
}
}
}
if (templevela == nullptr) {
fprintf(stderr, "allocation error\n");
return;
}
if (radlevblur > 0.f && chromablu > 0.f) { if (radlevblur > 0.f && chromablu > 0.f) {
wavcont(lp, tmp, *wdspota, templevela, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 1, chromablu, 0.f, 0.f, 0.f); wavcont(lp, tmp, *wdspota, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 1, chromablu, 0.f, 0.f, 0.f);
} }
wdspota->reconstruct(tmpa[0], 1.f); wdspota->reconstruct(tmpa[0], 1.f);
delete wdspota; delete wdspota;
for (int i = 0; i < dir; i++) {
for (int j = 0; j < leve; j++) {
for (int l = 0; l < H_L; l++) {
delete [] templevela[i][j][l];
}
}
}
for (int i = 0; i < dir; i++) {
for (int j = 0; j < leve; j++) {
delete [] templevela[i][j];
}
}
for (int i = 0; i < dir; i++) {
delete [] templevela[i];
}
delete [] templevela;
//b //b
wdspotb = new wavelet_decomposition(tmpb[0], bfw, bfh, maxlvl, 1, sk, numThreads, lp.daubLen); wdspotb = new wavelet_decomposition(tmpb[0], bfw, bfh, maxlvl, 1, sk, numThreads, lp.daubLen);
@ -8125,54 +7976,12 @@ BENCHFUN
return; return;
} }
templevelb = new float***[dir];
for (int d = 0; d < dir; d++) {
templevelb[d] = new float**[leve];
for (int k = 0; k < leve; k++) {
templevelb[d][k] = new float*[H_L];
for (int i = 0; i < H_L; i++) {
templevelb[d][k][i] = new float[W_L];
}
}
}
if (templevelb == nullptr) {
fprintf(stderr, "allocation error\n");
return;
}
if (radlevblur > 0.f && chromablu > 0.f) { if (radlevblur > 0.f && chromablu > 0.f) {
wavcont(lp, tmp, *wdspotb, templevelb, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 1, chromablu, 0.f, 0.f, 0.f); wavcont(lp, tmp, *wdspotb, level_bl, maxlvl, loclevwavCurve, loclevwavutili, loccompwavCurve, loccompwavutili, loccomprewavCurve, loccomprewavutili, radlevblur, 1, chromablu, 0.f, 0.f, 0.f);
} }
wdspotb->reconstruct(tmpb[0], 1.f); wdspotb->reconstruct(tmpb[0], 1.f);
delete wdspotb; delete wdspotb;
for (int i = 0; i < dir; i++) {
for (int j = 0; j < leve; j++) {
for (int l = 0; l < H_L; l++) {
delete [] templevelb[i][j][l];
}
}
}
for (int i = 0; i < dir; i++) {
for (int j = 0; j < leve; j++) {
delete [] templevelb[i][j];
}
}
for (int i = 0; i < dir; i++) {
delete [] templevelb[i];
}
delete [] templevelb;
} }
} }
@ -8427,7 +8236,7 @@ void ImProcFunctions::fftw_denoise(int GW, int GH, int max_numblox_W, int min_nu
} }
void ImProcFunctions::DeNoise(int call, int del, float * slidL, float * slida, float * slidb, int aut, bool noiscfactiv, struct local_params & lp, LabImage * originalmaskbl, int levred, float huerefblur, float lumarefblur, float chromarefblur, LabImage * original, LabImage * transformed, int cx, int cy, int sk) void ImProcFunctions::DeNoise(int call, int del, float * slidL, float * slida, float * slidb, int aut, bool noiscfactiv, const struct local_params & lp, LabImage * originalmaskbl, int levred, float huerefblur, float lumarefblur, float chromarefblur, LabImage * original, LabImage * transformed, int cx, int cy, int sk)
{ {
//local denoise //local denoise