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

Speedup for fast demosaic, Issue 2037

Browse Source
This commit is contained in:
Ingo
2014-03-14 16:53:43 +01:00
parent f7c45246c9
commit 3101f299a7
1 changed files with 257 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

362
rtengine/fast_demo.cc
View File

@@ -26,17 +26,15 @@
#include "rawimagesource.h"
#include "../rtgui/multilangmgr.h"
#include "procparams.h"
/*#ifdef _OPENMP
#include <omp.h>
#endif*/
#include "opthelper.h"
using namespace std;
using namespace rtengine;
#define TS 224
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
/*
LUTf RawImageSource::initInvGrad()
{
LUTf invGrad (0x10000);
@@ -47,46 +45,74 @@ LUTf RawImageSource::initInvGrad()
return invGrad;
}
*/
#define INVGRAD(i) (16.0f/SQR(4.0f+i))
#ifdef __SSE2__
#define INVGRADV(i) (c16v*_mm_rcp_ps(SQRV(fourv+i)))
#endif
//LUTf RawImageSource::invGrad = RawImageSource::initInvGrad();
LUTf RawImageSource::invGrad = RawImageSource::initInvGrad();
SSEFUNCTION void RawImageSource::fast_demosaic(int winx, int winy, int winw, int winh) {
double progress = 0.0;
const bool plistenerActive = plistener;
void RawImageSource::fast_demosaic(int winx, int winy, int winw, int winh) {
//int winx=0, winy=0;
//int winw=W, winh=H;
if (plistener) {
plistener->setProgressStr (Glib::ustring::compose(M("TP_RAW_DMETHOD_PROGRESSBAR"), RAWParams::methodstring[RAWParams::fast]));
plistener->setProgress (0.0);
plistener->setProgress (progress);
}
float progress = 0.0;
const int bord=4;
const int bord=5;
int clip_pt = 4*65535*initialGain;
float clip_pt = 4*65535*initialGain;
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#ifdef _OPENMP
#pragma omp parallel
#endif
{
char (*buffer);
float (*greentile);
float (*redtile);
float (*bluetile);
#define CLF 1
// assign working space
buffer = (char *) malloc(3*sizeof(float)*TS*TS + 3*CLF*64 + 63);
char *data;
data = (char*)( ( uintptr_t(buffer) + uintptr_t(63)) / 64 * 64);
greentile = (float (*)) data; //pointers to array
redtile = (float (*)) ((char*)greentile + sizeof(float)*TS*TS + CLF*64);
bluetile = (float (*)) ((char*)redtile + sizeof(float)*TS*TS + CLF*64);
#ifdef _OPENMP
#pragma omp for
#pragma omp sections
#endif
{
#ifdef _OPENMP
#pragma omp section
#endif
{
//first, interpolate borders using bilinear
for (int i=0; i<H; i++) {
float sum[6];
int imin = max(0,i-1);
int imax = min(i+2,H);
for (int j=0; j<bord; j++) {//first few columns
for (int c=0; c<6; c++) sum[c]=0;
for (int i1=i-1; i1<i+2; i1++)
for (int j1=j-1; j1<j+2; j1++) {
if ((i1 > -1) && (i1 < H) && (j1 > -1)) {
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int jmin = max(0,j-1);
for (int i1=imin; i1<imax; i1++)
for (int j1=jmin; j1<j+2; j1++) {
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int c=FC(i,j);
if (c==1) {
@@ -107,13 +133,12 @@ void RawImageSource::fast_demosaic(int winx, int winy, int winw, int winh) {
for (int j=W-bord; j<W; j++) {//last few columns
for (int c=0; c<6; c++) sum[c]=0;
for (int i1=i-1; i1<i+2; i1++)
for (int j1=j-1; j1<j+2; j1++) {
if ((i1 > -1) && (i1 < H ) && (j1 < W)) {
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int jmax = min(j+2,W);
for (int i1=imin; i1<imax; i1++)
for (int j1=j-1; j1<jmax; j1++) {
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int c=FC(i,j);
if (c==1) {
@@ -132,23 +157,24 @@ void RawImageSource::fast_demosaic(int winx, int winy, int winw, int winh) {
}
}//j
}//i
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#ifdef _OPENMP
#pragma omp for
#pragma omp section
#endif
{
for (int j=bord; j<W-bord; j++) {
float sum[6];
for (int i=0; i<bord; i++) {//first few rows
for (int c=0; c<6; c++) sum[c]=0;
for (int i1=i-1; i1<i+2; i1++)
for (int i1=max(0,i-1); i1<i+2; i1++)
for (int j1=j-1; j1<j+2; j1++) {
if ((j1 > -1) && (j1 < W) && (i1 > -1)) {
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int c=FC(i,j);
if (c==1) {
@@ -169,13 +195,11 @@ void RawImageSource::fast_demosaic(int winx, int winy, int winw, int winh) {
for (int i=H-bord; i<H; i++) {//last few rows
for (int c=0; c<6; c++) sum[c]=0;
for (int i1=i-1; i1<i+2; i1++)
for (int i1=i-1; i1<min(i+2,H); i1++)
for (int j1=j-1; j1<j+2; j1++) {
if ((j1 > -1) && (j1 < W) && (i1 < H)) {
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int c = FC(i1,j1);
sum[c] += rawData[i1][j1];
sum[c+3]++;
}
int c=FC(i,j);
if (c==1) {
@@ -194,91 +218,219 @@ void RawImageSource::fast_demosaic(int winx, int winy, int winw, int winh) {
}
}//i
}//j
if(plistener) plistener->setProgress(0.05);
}
}
#ifdef _OPENMP
#pragma omp single
#endif
{
if(plistenerActive) {
progress += 0.1;
plistener->setProgress(progress);
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
int progressCounter = 0;
const double progressInc = 16.0*(1.0-progress)/((H*W)/((TS-4)*(TS-4)));
#ifdef _OPENMP
#pragma omp for
#pragma omp for nowait
#endif
for (int top=bord-2; top < H-bord+2; top += TS-(4))
for (int left=bord-2; left < W-bord+2; left += TS-(4)) {
int bottom = min(top+TS, H-bord+2);
int right = min(left+TS, W-bord+2);
#ifdef __SSE2__
int j,cc;
__m128 wtuv, wtdv, wtlv, wtrv;
__m128 greenv,tempv,absv,abs2v;
__m128 onev = _mm_set1_ps( 1.0f );
__m128 c16v = _mm_set1_ps( 16.0f );
__m128 fourv = _mm_set1_ps( 4.0f );
vmask selmask;
vmask andmask = _mm_set_epi32( 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff );
if(FC(top,left) == 1)
selmask = _mm_set_epi32( 0, 0xffffffff, 0, 0xffffffff );
else
selmask = _mm_set_epi32( 0xffffffff, 0, 0xffffffff, 0 );
#endif
// interpolate G using gradient weights
for (int i=bord; i< H-bord; i++) {
for (int i=top,rr=0; i< bottom; i++,rr++) {
float wtu, wtd, wtl, wtr;
for (int j=bord; j < W-bord; j++) {
#ifdef __SSE2__
selmask = (vmask)_mm_andnot_ps( (__m128)selmask, (__m128)andmask);
for (j=left,cc=0; j < right-3; j+=4,cc+=4) {
tempv = LVFU(rawData[i][j]);
absv = vabsf(LVFU(rawData[i-1][j])-LVFU(rawData[i+1][j]));
wtuv = INVGRADV(absv+vabsf(tempv-LVFU(rawData[i-2][j]))+vabsf(LVFU(rawData[i-1][j])-LVFU(rawData[i-3][j])));
wtdv = INVGRADV(absv+vabsf(tempv-LVFU(rawData[i+2][j]))+vabsf(LVFU(rawData[i+1][j])-LVFU(rawData[i+3][j])));
abs2v = vabsf(LVFU(rawData[i][j-1])-LVFU(rawData[i][j+1]));
wtlv = INVGRADV(abs2v+vabsf(tempv-LVFU(rawData[i][j-2]))+vabsf(LVFU(rawData[i][j-1])-LVFU(rawData[i][j-3])));
wtrv = INVGRADV(abs2v+vabsf(tempv-LVFU(rawData[i][j+2]))+vabsf(LVFU(rawData[i][j+1])-LVFU(rawData[i][j+3])));
greenv = (wtuv*LVFU(rawData[i-1][j])+wtdv*LVFU(rawData[i+1][j])+wtlv*LVFU(rawData[i][j-1])+wtrv*LVFU(rawData[i][j+1])) / (wtuv+wtdv+wtlv+wtrv);
_mm_store_ps(&greentile[rr*TS+cc],vself(selmask, greenv, tempv));
_mm_store_ps(&redtile[rr*TS+cc],tempv);
_mm_store_ps(&bluetile[rr*TS+cc],tempv);
}
for (; j < right; j++,cc++) {
if (FC(i,j)==1) {
green[i][j] = rawData[i][j];
//red[i][j] = green[i][j];
//blue[i][j] = green[i][j];
greentile[rr*TS+cc] = rawData[i][j];
} else {
} else {
//compute directional weights using image gradients
wtu=invGrad[(abs(rawData[i+1][j]-rawData[i-1][j])+abs(rawData[i][j]-rawData[i-2][j])+abs(rawData[i-1][j]-rawData[i-3][j])) /4];
wtd=invGrad[(abs(rawData[i-1][j]-rawData[i+1][j])+abs(rawData[i][j]-rawData[i+2][j])+abs(rawData[i+1][j]-rawData[i+3][j])) /4];
wtl=invGrad[(abs(rawData[i][j+1]-rawData[i][j-1])+abs(rawData[i][j]-rawData[i][j-2])+abs(rawData[i][j-1]-rawData[i][j-3])) /4];
wtr=invGrad[(abs(rawData[i][j-1]-rawData[i][j+1])+abs(rawData[i][j]-rawData[i][j+2])+abs(rawData[i][j+1]-rawData[i][j+3])) /4];
wtu=INVGRAD((abs(rawData[i+1][j]-rawData[i-1][j])+abs(rawData[i][j]-rawData[i-2][j])+abs(rawData[i-1][j]-rawData[i-3][j])));
wtd=INVGRAD((abs(rawData[i-1][j]-rawData[i+1][j])+abs(rawData[i][j]-rawData[i+2][j])+abs(rawData[i+1][j]-rawData[i+3][j])));
wtl=INVGRAD((abs(rawData[i][j+1]-rawData[i][j-1])+abs(rawData[i][j]-rawData[i][j-2])+abs(rawData[i][j-1]-rawData[i][j-3])));
wtr=INVGRAD((abs(rawData[i][j-1]-rawData[i][j+1])+abs(rawData[i][j]-rawData[i][j+2])+abs(rawData[i][j+1]-rawData[i][j+3])));
//store in rgb array the interpolated G value at R/B grid points using directional weighted average
green[i][j]=(wtu*rawData[i-1][j]+wtd*rawData[i+1][j]+wtl*rawData[i][j-1]+wtr*rawData[i][j+1]) / (wtu+wtd+wtl+wtr);
//red[i][j] = green[i][j];
//blue[i][j] = green[i][j];
greentile[rr*TS+cc]=(wtu*rawData[i-1][j]+wtd*rawData[i+1][j]+wtl*rawData[i][j-1]+wtr*rawData[i][j+1]) / (wtu+wtd+wtl+wtr);
}
redtile[rr*TS+cc] = rawData[i][j];
bluetile[rr*TS+cc] = rawData[i][j];
}
//progress+=(double)0.33/(H);
//if(plistener) plistener->setProgress(progress);
}
if(plistener) plistener->setProgress(0.4);
#else
for (int j=left,cc=0; j < right; j++,cc++) {
if (FC(i,j)==1) {
greentile[rr*TS+cc] = rawData[i][j];
} else {
//compute directional weights using image gradients
wtu=INVGRAD((abs(rawData[i+1][j]-rawData[i-1][j])+abs(rawData[i][j]-rawData[i-2][j])+abs(rawData[i-1][j]-rawData[i-3][j])));
wtd=INVGRAD((abs(rawData[i-1][j]-rawData[i+1][j])+abs(rawData[i][j]-rawData[i+2][j])+abs(rawData[i+1][j]-rawData[i+3][j])));
wtl=INVGRAD((abs(rawData[i][j+1]-rawData[i][j-1])+abs(rawData[i][j]-rawData[i][j-2])+abs(rawData[i][j-1]-rawData[i][j-3])));
wtr=INVGRAD((abs(rawData[i][j-1]-rawData[i][j+1])+abs(rawData[i][j]-rawData[i][j+2])+abs(rawData[i][j+1]-rawData[i][j+3])));
#ifdef _OPENMP
#pragma omp for
#endif
for (int i=bord; i< H-bord; i++) {
for (int j=bord+(FC(i,2)&1); j < W-bord; j+=2) {
int c=FC(i,j);
//interpolate B/R colors at R/B sites
if (c==0) {//R site
red[i][j] = rawData[i][j];
blue[i][j] = green[i][j] - 0.25f*((green[i-1][j-1]+green[i-1][j+1]+green[i+1][j+1]+green[i+1][j-1]) -
min(static_cast<float>(clip_pt),rawData[i-1][j-1]+rawData[i-1][j+1]+rawData[i+1][j+1]+rawData[i+1][j-1]));
} else {//B site
red[i][j] = green[i][j] - 0.25f*((green[i-1][j-1]+green[i-1][j+1]+green[i+1][j+1]+green[i+1][j-1]) -
min(static_cast<float>(clip_pt),rawData[i-1][j-1]+rawData[i-1][j+1]+rawData[i+1][j+1]+rawData[i+1][j-1]));
blue[i][j] = rawData[i][j];
//store in rgb array the interpolated G value at R/B grid points using directional weighted average
greentile[rr*TS+cc]=(wtu*rawData[i-1][j]+wtd*rawData[i+1][j]+wtl*rawData[i][j-1]+wtr*rawData[i][j+1]) / (wtu+wtd+wtl+wtr);
}
redtile[rr*TS+cc] = rawData[i][j];
bluetile[rr*TS+cc] = rawData[i][j];
}
//progress+=(double)0.33/(H);
//if(plistener) plistener->setProgress(progress);
}
if(plistener) plistener->setProgress(0.7);
#ifdef _OPENMP
#pragma omp barrier
#endif
}
#ifdef _OPENMP
#pragma omp for
#ifdef __SSE2__
__m128 zd25v = _mm_set1_ps(0.25f);
__m128 clip_ptv = _mm_set1_ps( clip_pt );
#endif
for (int i=top+1,rr=1; i< bottom-1; i++,rr++) {
if (FC(i,left+(FC(i,2)&1)+1)==0)
#ifdef __SSE2__
for (int j=left+1,cc=1; j < right-1; j+=4,cc+=4) {
//interpolate B/R colors at R/B sites
_mm_storeu_ps(&bluetile[rr*TS+cc], LVFU(greentile[rr*TS+cc]) - zd25v*((LVFU(greentile[(rr-1)*TS+(cc-1)])+LVFU(greentile[(rr-1)*TS+(cc+1)])+LVFU(greentile[(rr+1)*TS+cc+1])+LVFU(greentile[(rr+1)*TS+cc-1])) -
_mm_min_ps(clip_ptv,LVFU(rawData[i-1][j-1])+LVFU(rawData[i-1][j+1])+LVFU(rawData[i+1][j+1])+LVFU(rawData[i+1][j-1]))));
}
#else
for (int cc=(FC(i,2)&1)+1, j=left+cc; j < right-1; j+=2, cc+=2) {
//interpolate B/R colors at R/B sites
bluetile[rr*TS+cc] = greentile[rr*TS+cc] - 0.25f*((greentile[(rr-1)*TS+(cc-1)]+greentile[(rr-1)*TS+(cc+1)]+greentile[(rr+1)*TS+cc+1]+greentile[(rr+1)*TS+cc-1]) -
min(clip_pt,rawData[i-1][j-1]+rawData[i-1][j+1]+rawData[i+1][j+1]+rawData[i+1][j-1]));
}
#endif
else
#ifdef __SSE2__
for (int j=left+1,cc=1; j < right-1; j+=4,cc+=4) {
//interpolate B/R colors at R/B sites
_mm_storeu_ps(&redtile[rr*TS+cc], LVFU(greentile[rr*TS+cc]) - zd25v*((LVFU(greentile[(rr-1)*TS+cc-1])+LVFU(greentile[(rr-1)*TS+cc+1])+LVFU(greentile[(rr+1)*TS+cc+1])+LVFU(greentile[(rr+1)*TS+cc-1])) -
_mm_min_ps(clip_ptv,LVFU(rawData[i-1][j-1])+LVFU(rawData[i-1][j+1])+LVFU(rawData[i+1][j+1])+LVFU(rawData[i+1][j-1]))));
}
#else
for (int cc=(FC(i,2)&1)+1, j=left+cc; j < right-1; j+=2,cc+=2) {
//interpolate B/R colors at R/B sites
redtile[rr*TS+cc] = greentile[rr*TS+cc] - 0.25f*((greentile[(rr-1)*TS+cc-1]+greentile[(rr-1)*TS+cc+1]+greentile[(rr+1)*TS+cc+1]+greentile[(rr+1)*TS+cc-1]) -
min(clip_pt,rawData[i-1][j-1]+rawData[i-1][j+1]+rawData[i+1][j+1]+rawData[i+1][j-1]));
}
#endif
}
#ifdef __SSE2__
__m128 temp1v,temp2v,greensumv;
selmask = _mm_set_epi32( 0xffffffff, 0, 0xffffffff, 0 );
#endif
// interpolate R/B using color differences
for (int i=bord; i< H-bord; i++) {
for (int j=bord+1-(FC(i,2)&1); j < W-bord; j+=2) {
//interpolate R and B colors at G sites
red[i][j] = green[i][j] - 0.25f*((green[i-1][j]-red[i-1][j])+(green[i+1][j]-red[i+1][j])+
(green[i][j-1]-red[i][j-1])+(green[i][j+1]-red[i][j+1]));
blue[i][j] = green[i][j] - 0.25f*((green[i-1][j]-blue[i-1][j])+(green[i+1][j]-blue[i+1][j])+
(green[i][j-1]-blue[i][j-1])+(green[i][j+1]-blue[i][j+1]));
for (int i=top+2, rr=2; i< bottom-2; i++,rr++) {
#ifdef __SSE2__
for (int cc=2+(FC(i,2)&1), j=left+cc; j < right-2; j+=4,cc+=4) {
// no need to take care about the borders of the tile. There's enough free space.
//interpolate R and B colors at G sites
greenv = LVFU(greentile[rr*TS+cc]);
greensumv = LVFU(greentile[(rr-1)*TS+cc]) + LVFU(greentile[(rr+1)*TS+cc]) + LVFU(greentile[rr*TS+cc-1]) + LVFU(greentile[rr*TS+cc+1]);
temp1v = LVFU(redtile[rr*TS+cc]);
temp2v = greenv - zd25v*(greensumv - LVFU(redtile[(rr-1)*TS+cc]) - LVFU(redtile[(rr+1)*TS+cc]) - LVFU(redtile[rr*TS+cc-1]) - LVFU(redtile[rr*TS+cc+1]));
// temp2v = greenv - zd25v*((LVFU(greentile[(rr-1)*TS+cc])-LVFU(redtile[(rr-1)*TS+cc]))+(LVFU(greentile[(rr+1)*TS+cc])-LVFU(redtile[(rr+1)*TS+cc]))+
// (LVFU(greentile[rr*TS+cc-1])-LVFU(redtile[rr*TS+cc-1]))+(LVFU(greentile[rr*TS+cc+1])-LVFU(redtile[rr*TS+cc+1])));
_mm_storeu_ps( &redtile[rr*TS+cc], vself(selmask, temp1v, temp2v));
temp1v = LVFU(bluetile[rr*TS+cc]);
temp2v = greenv - zd25v*(greensumv - LVFU(bluetile[(rr-1)*TS+cc]) - LVFU(bluetile[(rr+1)*TS+cc]) - LVFU(bluetile[rr*TS+cc-1]) - LVFU(bluetile[rr*TS+cc+1]));
// temp2v = greenv - zd25v*((LVFU(greentile[(rr-1)*TS+cc])-LVFU(bluetile[(rr-1)*TS+cc]))+(LVFU(greentile[(rr+1)*TS+cc])-LVFU(bluetile[(rr+1)*TS+cc]))+
// (LVFU(greentile[rr*TS+cc-1])-LVFU(bluetile[rr*TS+cc-1]))+(LVFU(greentile[rr*TS+cc+1])-LVFU(bluetile[rr*TS+cc+1])));
_mm_storeu_ps( &bluetile[rr*TS+cc], vself(selmask, temp1v, temp2v));
}
#else
for (int cc=2+(FC(i,2)&1), j=left+cc; j < right-2; j+=2,cc+=2) {
//interpolate R and B colors at G sites
redtile[rr*TS+cc] = greentile[rr*TS+cc] - 0.25f*((greentile[(rr-1)*TS+cc]-redtile[(rr-1)*TS+cc])+(greentile[(rr+1)*TS+cc]-redtile[(rr+1)*TS+cc])+
(greentile[rr*TS+cc-1]-redtile[rr*TS+cc-1])+(greentile[rr*TS+cc+1]-redtile[rr*TS+cc+1]));
bluetile[rr*TS+cc] = greentile[rr*TS+cc] - 0.25f*((greentile[(rr-1)*TS+cc]-bluetile[(rr-1)*TS+cc])+(greentile[(rr+1)*TS+cc]-bluetile[(rr+1)*TS+cc])+
(greentile[rr*TS+cc-1]-bluetile[rr*TS+cc-1])+(greentile[rr*TS+cc+1]-bluetile[rr*TS+cc+1]));
}
#endif
}
progress+=(double)0.33/(H);
//if(plistener) plistener->setProgress(progress);
}
if(plistener) plistener->setProgress(0.99);
} // End of parallelization
#undef bord
for (int i=top+2, rr=2; i< bottom-2; i++,rr++) {
#ifdef __SSE2__
for (j=left+2, cc=2; j< right-5; j+=4,cc+=4) {
_mm_storeu_ps(&red[i][j], LVFU(redtile[rr*TS+cc]));
_mm_storeu_ps(&green[i][j], LVFU(greentile[rr*TS+cc]));
_mm_storeu_ps(&blue[i][j], LVFU(bluetile[rr*TS+cc]));
}
for (; j< right-2; j++,cc++) {
red[i][j] = redtile[rr*TS+cc];
green[i][j] = greentile[rr*TS+cc];
blue[i][j] = bluetile[rr*TS+cc];
}
#else
for (int j=left+2, cc=2; j< right-2; j++,cc++) {
red[i][j] = redtile[rr*TS+cc];
green[i][j] = greentile[rr*TS+cc];
blue[i][j] = bluetile[rr*TS+cc];
}
#endif
}
if(plistenerActive && ((++progressCounter) % 16 == 0)) {
#ifdef _OPENMP
#pragma omp critical (updateprogress)
#endif
{
progress += progressInc;
progress = min(1.0,progress);
plistener->setProgress (progress);
}
}
}
free(buffer);
} // End of parallelization
if(plistenerActive) plistener->setProgress(1.00);
}
#undef TS
#undef CLF