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

eliminate AlignedBufferMP, also eliminate some unsused functions in boxblur.h

Browse Source
This commit is contained in:
heckflosse
2017-10-22 21:06:24 +02:00
parent 8bd9f174dc
commit 8de8775264
4 changed files with 1 additions and 547 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
rtengine/FTblockDN.cc
View File

@@ -3261,7 +3261,6 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
realblue = 0.001f;
}
//TODO: implement using AlignedBufferMP
//fill tile from image; convert RGB to "luma/chroma"
if (isRAW) {//image is raw; use channel differences for chroma channels

52
rtengine/alignedbuffer.h
View File

@@ -18,12 +18,8 @@
*/
#ifndef _ALIGNEDBUFFER_
#define _ALIGNEDBUFFER_
#include <cstdint>
#include <cstdlib>
#include <vector>
#include <utility>
#include <glibmm.h>
#include "../rtgui/threadutils.h"
// Aligned buffer that should be faster
template <class T> class AlignedBuffer
@@ -111,7 +107,6 @@ public:
}
if (real) {
//data = (T*)( (uintptr_t)real + (alignment-((uintptr_t)real)%alignment) );
data = (T*)( ( uintptr_t(real) + uintptr_t(alignment - 1)) / alignment * alignment);
inUse = true;
} else {
@@ -142,51 +137,4 @@ public:
}
};
// Multi processor version, use with OpenMP
template <class T> class AlignedBufferMP
{
private:
MyMutex mtx;
std::vector<AlignedBuffer<T>*> buffers;
size_t size;
public:
explicit AlignedBufferMP(size_t sizeP)
{
size = sizeP;
}
~AlignedBufferMP()
{
for (size_t i = 0; i < buffers.size(); i++) {
delete buffers[i];
}
}
AlignedBuffer<T>* acquire()
{
MyMutex::MyLock lock(mtx);
// Find available buffer
for (size_t i = 0; i < buffers.size(); i++) {
if (!buffers[i]->inUse) {
buffers[i]->inUse = true;
return buffers[i];
}
}
// Add new buffer if nothing is free
AlignedBuffer<T>* buffer = new AlignedBuffer<T>(size);
buffers.push_back(buffer);
return buffer;
}
void release(AlignedBuffer<T>* buffer)
{
MyMutex::MyLock lock(mtx);
buffer->inUse = false;
}
};
#endif

494
rtengine/boxblur.h
View File

@@ -34,8 +34,6 @@ namespace rtengine
template<class T, class A> void boxblur (T** src, A** dst, int radx, int rady, int W, int H)
{
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//box blur image; box range = (radx,rady)
AlignedBuffer<float>* buffer = new AlignedBuffer<float> (W * H);
@@ -125,8 +123,6 @@ 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)
{
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//box blur image; box range = (radx,rady)
float* temp = buffer;
@@ -313,13 +309,8 @@ 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)
{
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//box blur image; box range = (radx,rady) i.e. box size is (2*radx+1)x(2*rady+1)
float* temp = buffer;
@@ -505,489 +496,6 @@ template<class T, class A> SSEFUNCTION void boxblur (T* src, A* dst, A* buffer,
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
template<typename T> void boxvar (T* src, T* dst, int radx, int rady, int W, int H)
{
AlignedBuffer<float> buffer1(W * H);
AlignedBuffer<float> buffer2(W * H);
float* tempave = buffer1.data;
float* tempsqave = buffer2.data;
AlignedBufferMP<float> buffer3(H);
//float image_ave = 0;
//box blur image channel; box size = 2*box+1
//horizontal blur
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++) {
int len = radx + 1;
tempave[row * W + 0] = src[row * W + 0] / len;
tempsqave[row * W + 0] = SQR(src[row * W + 0]) / len;
for (int j = 1; j <= radx; j++) {
tempave[row * W + 0] += src[row * W + j] / len;
tempsqave[row * W + 0] += SQR(src[row * W + j]) / len;
}
for (int col = 1; col <= radx; col++) {
tempave[row * W + col] = (tempave[row * W + col - 1] * len + src[row * W + col + radx]) / (len + 1);
tempsqave[row * W + col] = (tempsqave[row * W + col - 1] * len + SQR(src[row * W + col + radx])) / (len + 1);
len ++;
}
for (int col = radx + 1; col < W - radx; col++) {
tempave[row * W + col] = tempave[row * W + col - 1] + (src[row * W + col + radx] - src[row * W + col - radx - 1]) / len;
tempsqave[row * W + col] = tempsqave[row * W + col - 1] + (SQR(src[row * W + col + radx]) - SQR(src[row * W + col - radx - 1])) / len;
}
for (int col = W - radx; col < W; col++) {
tempave[row * W + col] = (tempave[row * W + col - 1] * len - src[row * W + col - radx - 1]) / (len - 1);
tempsqave[row * W + col] = (tempsqave[row * W + col - 1] * len - SQR(src[row * W + col - radx - 1])) / (len - 1);
len --;
}
}
//vertical blur
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int col = 0; col < W; col++) {
AlignedBuffer<float>* pBuf3 = buffer3.acquire();
T* tempave2 = (T*)pBuf3->data;
int len = rady + 1;
tempave2[0] = tempave[0 * W + col] / len;
dst[0 * W + col] = tempsqave[0 * W + col] / len;
for (int i = 1; i <= rady; i++) {
tempave2[0] += tempave[i * W + col] / len;
dst[0 * W + col] += tempsqave[i * W + col] / len;
}
for (int row = 1; row <= rady; row++) {
tempave2[row] = (tempave2[(row - 1)] * len + tempave[(row + rady) * W + col]) / (len + 1);
dst[row * W + col] = (dst[(row - 1) * W + col] * len + tempsqave[(row + rady) * W + col]) / (len + 1);
len ++;
}
for (int row = rady + 1; row < H - rady; row++) {
tempave2[row] = tempave2[(row - 1)] + (tempave[(row + rady) * W + col] - tempave[(row - rady - 1) * W + col]) / len;
dst[row * W + col] = dst[(row - 1) * W + col] + (tempsqave[(row + rady) * W + col] - tempsqave[(row - rady - 1) * W + col]) / len;
}
for (int row = H - rady; row < H; row++) {
tempave2[row] = (tempave2[(row - 1)] * len - tempave[(row - rady - 1) * W + col]) / (len - 1);
dst[row * W + col] = (dst[(row - 1) * W + col] * len - tempsqave[(row - rady - 1) * W + col]) / (len - 1);
len --;
}
//now finish off
for (int row = 0; row < H; row++) {
dst[row * W + col] = fabs(dst[row * W + col] - SQR(tempave2[row]));
//image_ave += src[row*W+col];
}
buffer3.release(pBuf3);
}
//image_ave /= (W*H);
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
template<typename T> void boxdev (T* src, T* dst, 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)
AlignedBuffer<float>* buffer1 = new AlignedBuffer<float> (W * H);
float* temp = buffer1->data;
AlignedBuffer<float>* buffer2 = new AlignedBuffer<float> (W * H);
float* tempave = buffer2->data;
if (radx == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++)
for (int col = 0; col < W; col++) {
temp[row * W + col] = src[row * W + col];
}
} else {
//horizontal blur
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++) {
int len = radx + 1;
temp[row * W + 0] = (float)src[row * W + 0] / len;
for (int j = 1; j <= radx; j++) {
temp[row * W + 0] += (float)src[row * W + j] / len;
}
for (int col = 1; col <= radx; col++) {
temp[row * W + col] = (temp[row * W + col - 1] * len + src[row * W + col + radx]) / (len + 1);
len ++;
}
for (int col = radx + 1; col < W - radx; col++) {
temp[row * W + col] = temp[row * W + col - 1] + ((float)(src[row * W + col + radx] - src[row * W + col - radx - 1])) / len;
}
for (int col = W - radx; col < W; col++) {
temp[row * W + col] = (temp[row * W + col - 1] * len - src[row * W + col - radx - 1]) / (len - 1);
len --;
}
}
}
if (rady == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++) {
for (int col = 0; col < W; col++) {
tempave[row * W + col] = temp[row * W + col];
}
}
} else {
//vertical blur
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int col = 0; col < W; col++) {
int len = rady + 1;
tempave[0 * W + col] = temp[0 * W + col] / len;
for (int i = 1; i <= rady; i++) {
tempave[0 * W + col] += temp[i * W + col] / len;
}
for (int row = 1; row <= rady; row++) {
tempave[row * W + col] = (tempave[(row - 1) * W + col] * len + temp[(row + rady) * W + col]) / (len + 1);
len ++;
}
for (int row = rady + 1; row < H - rady; row++) {
tempave[row * W + col] = tempave[(row - 1) * W + col] + (temp[(row + rady) * W + col] - temp[(row - rady - 1) * W + col]) / len;
}
for (int row = H - rady; row < H; row++) {
tempave[row * W + col] = (tempave[(row - 1) * W + col] * len - temp[(row - rady - 1) * W + col]) / (len - 1);
len --;
}
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//box blur absolute deviation
if (radx == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++)
for (int col = 0; col < W; col++) {
temp[row * W + col] = fabs(src[row * W + col] - tempave[row * W + col]);
}
} else {
//horizontal blur
//OpenMP here
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++) {
int len = radx + 1;
temp[row * W + 0] = fabs(src[row * W + 0] - tempave[row * W + 0]) / len;
for (int j = 1; j <= radx; j++) {
temp[row * W + 0] += fabs(src[row * W + j] - tempave[row * W + j]) / len;
}
for (int col = 1; col <= radx; col++) {
temp[row * W + col] = (temp[row * W + col - 1] * len + fabs(src[row * W + col + radx] - tempave[row * W + col + radx])) / (len + 1);
len ++;
}
for (int col = radx + 1; col < W - radx; col++) {
temp[row * W + col] = temp[row * W + col - 1] + (fabs(src[row * W + col + radx] - tempave[row * W + col + radx]) - \
fabs(src[row * W + col - radx - 1] - tempave[row * W + col - radx - 1])) / len;
}
for (int col = W - radx; col < W; col++) {
temp[row * W + col] = (temp[row * W + col - 1] * len - fabs(src[row * W + col - radx - 1] - tempave[row * W + col - radx - 1])) / (len - 1);
len --;
}
}
}
if (rady == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++)
for (int col = 0; col < W; col++) {
dst[row * W + col] = temp[row * W + col];
}
} else {
//vertical blur
//OpenMP here
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int col = 0; col < W; col++) {
int len = rady + 1;
dst[0 * W + col] = temp[0 * W + col] / len;
for (int i = 1; i <= rady; i++) {
dst[0 * W + col] += temp[i * W + col] / len;
}
for (int row = 1; row <= rady; row++) {
dst[row * W + col] = (dst[(row - 1) * W + col] * len + temp[(row + rady) * W + col]) / (len + 1);
len ++;
}
for (int row = rady + 1; row < H - rady; row++) {
dst[row * W + col] = dst[(row - 1) * W + col] + (temp[(row + rady) * W + col] - temp[(row - rady - 1) * W + col]) / len;
}
for (int row = H - rady; row < H; row++) {
dst[row * W + col] = (dst[(row - 1) * W + col] * len - temp[(row - rady - 1) * W + col]) / (len - 1);
len --;
}
}
}
delete buffer1;
delete buffer2;
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
template<class T, class A> void boxsqblur (T* src, A* dst, 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)
AlignedBuffer<float>* buffer = new AlignedBuffer<float> (W * H);
float* temp = buffer->data;
if (radx == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++)
for (int col = 0; col < W; col++) {
temp[row * W + col] = SQR(src[row * W + col]);
}
} else {
//horizontal blur
//OpenMP here
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++) {
int len = radx + 1;
temp[row * W + 0] = SQR((float)src[row * W + 0]) / len;
for (int j = 1; j <= radx; j++) {
temp[row * W + 0] += SQR((float)src[row * W + j]) / len;
}
for (int col = 1; col <= radx; col++) {
temp[row * W + col] = (temp[row * W + col - 1] * len + SQR(src[row * W + col + radx])) / (len + 1);
len ++;
}
for (int col = radx + 1; col < W - radx; col++) {
temp[row * W + col] = temp[row * W + col - 1] + ((float)(SQR(src[row * W + col + radx]) - SQR(src[row * W + col - radx - 1]))) / len;
}
for (int col = W - radx; col < W; col++) {
temp[row * W + col] = (temp[row * W + col - 1] * len - SQR(src[row * W + col - radx - 1])) / (len - 1);
len --;
}
}
}
if (rady == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++)
for (int col = 0; col < W; col++) {
dst[row * W + col] = temp[row * W + col];
}
} else {
//vertical blur
//OpenMP here
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int col = 0; col < W; col++) {
int len = rady + 1;
dst[0 * W + col] = temp[0 * W + col] / len;
for (int i = 1; i <= rady; i++) {
dst[0 * W + col] += temp[i * W + col] / len;
}
for (int row = 1; row <= rady; row++) {
dst[row * W + col] = (dst[(row - 1) * W + col] * len + temp[(row + rady) * W + col]) / (len + 1);
len ++;
}
for (int row = rady + 1; row < H - rady; row++) {
dst[row * W + col] = dst[(row - 1) * W + col] + (temp[(row + rady) * W + col] - temp[(row - rady - 1) * W + col]) / len;
}
for (int row = H - rady; row < H; row++) {
dst[row * W + col] = (dst[(row - 1) * W + col] * len - temp[(row - rady - 1) * W + col]) / (len - 1);
len --;
}
}
}
delete buffer;
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
template<class T, class A> void boxcorrelate (T* src, A* dst, int dx, int dy, 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)
AlignedBuffer<float>* buffer = new AlignedBuffer<float> (W * H);
float* temp = buffer->data;
if (radx == 0) {
for (int row = 0; row < H; row++) {
int rr = min(H - 1, max(0, row + dy));
for (int col = 0; col < W; col++) {
int cc = min(W - 1, max(0, col + dx));
temp[row * W + col] = dy > 0 ? (src[row * W + col]) * (src[rr * W + cc]) : 0;
}
}
} else {
//horizontal blur
for (int row = 0; row < H; row++) {
int len = radx + 1;
int rr = min(H - 1, max(0, row + dy));
int cc = min(W - 1, max(0, 0 + dx));
temp[row * W + 0] = ((float)src[row * W + 0]) * (src[rr * W + cc]) / len;
for (int j = 1; j <= radx; j++) {
int cc = min(W - 1, max(0, j + dx));
temp[row * W + 0] += ((float)src[row * W + j]) * (src[rr * W + cc]) / len;
}
for (int col = 1; col <= radx; col++) {
int cc = min(W - 1, max(0, col + dx + radx));
temp[row * W + col] = (temp[row * W + col - 1] * len + (src[row * W + col + radx]) * (src[rr * W + cc])) / (len + 1);
len ++;
}
for (int col = radx + 1; col < W - radx; col++) {
int cc = min(W - 1, max(0, col + dx + radx));
int cc1 = min(W - 1, max(0, col + dx - radx - 1));
temp[row * W + col] = temp[row * W + col - 1] + ((float)((src[row * W + col + radx]) * (src[rr * W + cc]) -
(src[row * W + col - radx - 1]) * (src[rr * W + cc1]))) / len;
}
for (int col = W - radx; col < W; col++) {
int cc1 = min(W - 1, max(0, col + dx - radx - 1));
temp[row * W + col] = (temp[row * W + col - 1] * len - (src[row * W + col - radx - 1]) * (src[rr * W + cc1])) / (len - 1);
len --;
}
}
}
if (rady == 0) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; row++)
for (int col = 0; col < W; col++) {
dst[row * W + col] = temp[row * W + col];
}
} else {
//vertical blur
//OpenMP here
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int col = 0; col < W; col++) {
int len = rady + 1;
dst[0 * W + col] = temp[0 * W + col] / len;
for (int i = 1; i <= rady; i++) {
dst[0 * W + col] += temp[i * W + col] / len;
}
for (int row = 1; row <= rady; row++) {
dst[row * W + col] = (dst[(row - 1) * W + col] * len + temp[(row + rady) * W + col]) / (len + 1);
len ++;
}
for (int row = rady + 1; row < H - rady; row++) {
dst[row * W + col] = dst[(row - 1) * W + col] + (temp[(row + rady) * W + col] - temp[(row - rady - 1) * W + col]) / len;
}
for (int row = H - rady; row < H; row++) {
dst[row * W + col] = (dst[(row - 1) * W + col] * len - temp[(row - rady - 1) * W + col]) / (len - 1);
len --;
}
}
}
delete buffer;
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
template<class T, class A> SSEFUNCTION void boxabsblur (T* src, A* dst, int radx, int rady, int W, int H, float * temp)
{
@@ -1131,7 +639,5 @@ template<class T, class A> SSEFUNCTION void boxabsblur (T* src, A* dst, int radx
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
}
#endif /* _BOXBLUR_H_ */

1
rtengine/iimage.h
View File

@@ -28,6 +28,7 @@
#include "coord2d.h"
#include "procparams.h"
#include "color.h"
#include "../rtgui/threadutils.h"
#define TR_NONE 0
#define TR_R90 1