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

hphd demosaic: small speedup and own compilation unit, #5159

Browse Source
This commit is contained in:
heckflosse
2019-01-31 22:20:23 +01:00
parent a4acc1dca3
commit ac152ac423
4 changed files with 353 additions and 286 deletions
Download Patch File Download Diff File Expand all files Collapse all files

283
rtengine/demosaic_algos.cc
View File

@@ -20,16 +20,9 @@
#include <cassert>
#include "rawimagesource.h"
#include "rawimagesource_i.h"
#include "jaggedarray.h"
#include "rawimage.h"
#include "mytime.h"
#include "iccmatrices.h"
#include "iccstore.h"
#include "image8.h"
#include "curves.h"
#include "dfmanager.h"
#include "slicer.h"
#include "rt_math.h"
#include "color.h"
#include "../rtgui/multilangmgr.h"
@@ -60,282 +53,6 @@ extern const Settings* settings;
#define x00625(a) xdivf(a, 4)
#define x0125(a) xdivf(a, 3)
void RawImageSource::hphd_vertical (float** hpmap, int col_from, int col_to)
{
// BENCHFUN
constexpr int numCols = 8;
JaggedArray<float> temp(numCols, H, true);
JaggedArray<float> avg(numCols, H, true);
JaggedArray<float> dev(numCols, H, true);
int k = col_from;
for (; k < col_to - 7; k += numCols) {
for (int i = 5; i < H - 5; i++) {
#pragma omp simd
for(int h = 0; h < numCols; ++h) {
temp[i][h] = std::fabs((rawData[i - 5][k + h] - rawData[i + 5][k + h]) - 8 * (rawData[i - 4][k + h] - rawData[i + 4][k + h]) + 27 * (rawData[i - 3][k + h] - rawData[i + 3][k + h]) - 48 * (rawData[i - 2][k + h] - rawData[i + 2][k + h]) + 42 * (rawData[i - 1][k + h] - rawData[i - 1][k + h]));
}
}
for (int j = 4; j < H - 4; j++) {
#pragma omp simd
for(int h = 0; h < numCols; ++h) {
const float avgL = (temp[j - 4][h] + temp[j - 3][h] + temp[j - 2][h] + temp[j - 1][h] + temp[j][h] + temp[j + 1][h] + temp[j + 2][h] + temp[j + 3][h] + temp[j + 4][h]) / 9.f;
avg[j][h] = avgL;
dev[j][h] = std::max(0.001f, SQR(temp[j - 4][h] - avgL) + SQR(temp[j - 3][h] - avgL) + SQR(temp[j - 2][h] - avgL) + SQR(temp[j - 1][h] - avgL) + SQR(temp[j][h] - avgL) + SQR(temp[j + 1][h] - avgL) + SQR(temp[j + 2][h] - avgL) + SQR(temp[j + 3][h] - avgL) + SQR(temp[j + 4][h] - avgL));
}
}
for (int j = 5; j < H - 5; j++) {
#pragma omp simd
for(int h = 0; h < numCols; ++h) {
const float avgL = avg[j - 1][h];
const float avgR = avg[j + 1][h];
const float devL = dev[j - 1][h];
const float devR = dev[j + 1][h];
hpmap[j][k + h] = avgL + (avgR - avgL) * devL / (devL + devR);
}
}
}
for (; k < col_to; k++) {
for (int i = 5; i < H - 5; i++) {
temp[i][0] = std::fabs((rawData[i - 5][k] - rawData[i + 5][k]) - 8 * (rawData[i - 4][k] - rawData[i + 4][k]) + 27 * (rawData[i - 3][k] - rawData[i + 3][k]) - 48 * (rawData[i - 2][k] - rawData[i + 2][k]) + 42 * (rawData[i - 1][k] -rawData[i + 1][k]));
}
for (int j = 4; j < H - 4; j++) {
const float avgL = (temp[j - 4][0] + temp[j - 3][0] + temp[j - 2][0] + temp[j - 1][0] + temp[j][0] + temp[j + 1][0] + temp[j + 2][0] + temp[j + 3][0] + temp[j + 4][0]) / 9.f;
avg[j][0] = avgL;
dev[j][0] = std::max(0.001f, SQR(temp[j - 4][0] - avgL) + SQR(temp[j - 3][0] - avgL) + SQR(temp[j - 2][0] - avgL) + SQR(temp[j - 1][0] - avgL) + SQR(temp[j][0] - avgL) + SQR(temp[j + 1][0] - avgL) + SQR(temp[j + 2][0] - avgL) + SQR(temp[j + 3][0] - avgL) + SQR(temp[j + 4][0] - avgL));
}
for (int j = 5; j < H - 5; j++) {
const float avgL = avg[j - 1][0];
const float avgR = avg[j + 1][0];
const float devL = dev[j - 1][0];
const float devR = dev[j + 1][0];
hpmap[j][k] = avgL + (avgR - avgL) * devL / (devL + devR);
}
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
void RawImageSource::hphd_horizontal (float** hpmap, int row_from, int row_to)
{
// BENCHFUN
float* temp = new float[W];
float* avg = new float[W];
float* dev = new float[W];
memset(temp, 0, W * sizeof(float));
memset(avg, 0, W * sizeof(float));
memset(dev, 0, W * sizeof(float));
for (int i = row_from; i < row_to; i++) {
#pragma omp simd
for (int j = 5; j < W - 5; j++) {
temp[j] = std::fabs((rawData[i][j - 5] - rawData[i][j + 5]) - 8 * (rawData[i][j - 4] - rawData[i][j + 4]) + 27 * (rawData[i][j - 3] - rawData[i][j + 3]) - 48 * (rawData[i][j - 2] - rawData[i][j + 2]) + 42 * (rawData[i][j - 1] - rawData[i][j + 1]));
}
#pragma omp simd
for (int j = 4; j < W - 4; j++) {
const float avgL = ((temp[j - 4] + temp[j - 3]) + (temp[j - 2] + temp[j - 1]) + (temp[j] + temp[j + 1]) + (temp[j + 2] + temp[j + 3]) + temp[j + 4]) / 9.f;
avg[j] = avgL;
dev[j] = std::max(0.001f, SQR(temp[j - 4] - avgL) + SQR(temp[j - 3] - avgL) + SQR(temp[j - 2] - avgL) + SQR(temp[j - 1] - avgL) + SQR(temp[j] - avgL) + SQR(temp[j + 1] - avgL) + SQR(temp[j + 2] - avgL) + SQR(temp[j + 3] - avgL) + SQR(temp[j + 4] - avgL));
}
#pragma omp simd
for (int j = 5; j < W - 5; j++) {
const float avgL = avg[j - 1];
const float avgR = avg[j + 1];
const float devL = dev[j - 1];
const float devR = dev[j + 1];
const float hpv = avgL + (avgR - avgL) * devL / (devL + devR);
if (hpmap[i][j] < 0.8f * hpv) {
hpmap[i][j] = 2;
} else if (hpv < 0.8f * hpmap[i][j]) {
hpmap[i][j] = 1;
} else {
hpmap[i][j] = 0;
}
}
}
delete [] temp;
delete [] avg;
delete [] dev;
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
void RawImageSource::hphd_green (float** hpmap)
{
// BENCHFUN
constexpr float eps = 0.001f;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic, 16)
#endif
for (int i = 3; i < H - 3; i++) {
for (int j = 3; j < W - 3; j++) {
if (ri->ISGREEN(i, j)) {
green[i][j] = rawData[i][j];
} else {
if (hpmap[i][j] == 1) {
const float g2 = rawData[i][j + 1] + (rawData[i][j] - rawData[i][j + 2]) * 0.5f;
const float g4 = rawData[i][j - 1] + (rawData[i][j] - rawData[i][j - 2]) * 0.5f;
const float dx = eps + std::fabs(rawData[i][j + 1] - rawData[i][j - 1]);
float d1 = rawData[i][j + 3] - rawData[i][j + 1];
float d2 = rawData[i][j + 2] - rawData[i][j];
float d3 = rawData[i - 1][j + 2] - rawData[i - 1][j];
float d4 = rawData[i + 1][j + 2] - rawData[i + 1][j];
const float e2 = 1.f / (dx + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
d1 = rawData[i][j - 3] - rawData[i][j - 1];
d2 = rawData[i][j - 2] - rawData[i][j];
d3 = rawData[i - 1][j - 2] - rawData[i - 1][j];
d4 = rawData[i + 1][j - 2] - rawData[i + 1][j];
const float e4 = 1.f / (dx + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
green[i][j] = (e2 * g2 + e4 * g4) / (e2 + e4);
} else if (hpmap[i][j] == 2) {
const float g1 = rawData[i - 1][j] + (rawData[i][j] - rawData[i - 2][j]) * 0.5f;
const float g3 = rawData[i + 1][j] + (rawData[i][j] - rawData[i + 2][j]) * 0.5f;
const float dy = eps + std::fabs(rawData[i + 1][j] - rawData[i - 1][j]);
float d1 = rawData[i - 1][j] - rawData[i - 3][j];
float d2 = rawData[i][j] - rawData[i - 2][j];
float d3 = rawData[i][j - 1] - rawData[i - 2][j - 1];
float d4 = rawData[i][j + 1] - rawData[i - 2][j + 1];
const float e1 = 1.f / (dy + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
d1 = rawData[i + 1][j] - rawData[i + 3][j];
d2 = rawData[i][j] - rawData[i + 2][j];
d3 = rawData[i][j - 1] - rawData[i + 2][j - 1];
d4 = rawData[i][j + 1] - rawData[i + 2][j + 1];
const float e3 = 1.f / (dy + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
green[i][j] = (e1 * g1 + e3 * g3) / (e1 + e3);
} else {
const float g1 = rawData[i - 1][j] + (rawData[i][j] - rawData[i - 2][j]) * 0.5f;
const float g2 = rawData[i][j + 1] + (rawData[i][j] - rawData[i][j + 2]) * 0.5f;
const float g3 = rawData[i + 1][j] + (rawData[i][j] - rawData[i + 2][j]) * 0.5f;
const float g4 = rawData[i][j - 1] + (rawData[i][j] - rawData[i][j - 2]) * 0.5f;
const float dx = eps + std::fabs(rawData[i][j + 1] - rawData[i][j - 1]);
const float dy = eps + std::fabs(rawData[i + 1][j] - rawData[i - 1][j]);
float d1 = rawData[i - 1][j] - rawData[i - 3][j];
float d2 = rawData[i][j] - rawData[i - 2][j];
float d3 = rawData[i][j - 1] - rawData[i - 2][j - 1];
float d4 = rawData[i][j + 1] - rawData[i - 2][j + 1];
const float e1 = 1.f / (dy + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
d1 = rawData[i][j + 3] - rawData[i][j + 1];
d2 = rawData[i][j + 2] - rawData[i][j];
d3 = rawData[i - 1][j + 2] - rawData[i - 1][j];
d4 = rawData[i + 1][j + 2] - rawData[i + 1][j];
const float e2 = 1.f / (dx + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
d1 = rawData[i + 1][j] - rawData[i + 3][j];
d2 = rawData[i][j] - rawData[i + 2][j];
d3 = rawData[i][j - 1] - rawData[i + 2][j - 1];
d4 = rawData[i][j + 1] - rawData[i + 2][j + 1];
const float e3 = 1.f / (dy + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
d1 = rawData[i][j - 3] - rawData[i][j - 1];
d2 = rawData[i][j - 2] - rawData[i][j];
d3 = rawData[i - 1][j - 2] - rawData[i - 1][j];
d4 = rawData[i + 1][j - 2] - rawData[i + 1][j];
const float e4 = 1.f / (dx + std::fabs(d1) + std::fabs(d2) + (std::fabs(d3) + std::fabs(d4)) * 0.5f);
green[i][j] = (e1 * g1 + e2 * g2 + e3 * g3 + e4 * g4) / (e1 + e2 + e3 + e4);
}
}
}
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
void RawImageSource::hphd_demosaic ()
{
BENCHFUN
if (plistener) {
plistener->setProgressStr(Glib::ustring::compose(M("TP_RAW_DMETHOD_PROGRESSBAR"), RAWParams::BayerSensor::getMethodString(RAWParams::BayerSensor::Method::HPHD)));
plistener->setProgress(0.0);
}
JaggedArray<float> hpmap(W, H, true);
#ifdef _OPENMP
#pragma omp parallel
{
int tid = omp_get_thread_num();
int nthreads = omp_get_num_threads();
int blk = W / nthreads;
if (tid < nthreads - 1) {
hphd_vertical(hpmap, tid * blk, (tid + 1)*blk);
} else {
hphd_vertical(hpmap, tid * blk, W);
}
}
#else
hphd_vertical(hpmap, 0, W);
#endif
if (plistener) {
plistener->setProgress(0.33);
}
#ifdef _OPENMP
#pragma omp parallel
{
int tid = omp_get_thread_num();
int nthreads = omp_get_num_threads();
int blk = H / nthreads;
if (tid < nthreads - 1) {
hphd_horizontal(hpmap, tid * blk, (tid + 1)*blk);
} else {
hphd_horizontal(hpmap, tid * blk, H);
}
}
#else
hphd_horizontal(hpmap, 0, H);
#endif
hphd_green(hpmap);
if (plistener) {
plistener->setProgress(0.66);
}
#pragma omp parallel for
for (int i = 4; i < H - 4; i++) {
interpolate_row_rb_mul_pp(rawData, red[i], blue[i], green[i - 1], green[i], green[i + 1], i, 1.0, 1.0, 1.0, 0, W, 1);
}
border_interpolate2(W, H, 4, rawData, red, green, blue);
if (plistener) {
plistener->setProgress(1.0);
}
}
#undef fc
#define fc(row,col) \
(ri->get_filters() >> ((((row) << 1 & 14) + ((col) & 1)) << 1) & 3)