Author: post
Date: 2009-08-17 23:16:07 +0200 (Mon, 17 Aug 2009)
New Revision: 2622
Modified:
trunk/plugins/denoise/Makefile.am
trunk/plugins/denoise/floatplanarimage-x86.cpp
trunk/plugins/denoise/floatplanarimage.cpp
trunk/plugins/denoise/floatplanarimage.h
Log:
Denoise: WB correction was being applied at the wrong stage, resulting in color
cast when denoising/sharpening. One assembler function disabled until fixed.
Modified: trunk/plugins/denoise/Makefile.am
===================================================================
--- trunk/plugins/denoise/Makefile.am 2009-08-10 19:58:00 UTC (rev 2621)
+++ trunk/plugins/denoise/Makefile.am 2009-08-17 21:16:07 UTC (rev 2622)
@@ -10,7 +10,7 @@
INCLUDES = \
-DPACKAGE_DATA_DIR=\""$(datadir)"\" \
-DPACKAGE_LOCALE_DIR=\""$(prefix)/$(DATADIRNAME)/locale"\" \
- @PACKAGE_CFLAGS@ \
+ @PACKAGE_CFLAGS@ -ffast-math \
-I../../librawstudio/
lib_LTLIBRARIES = denoise.la
Modified: trunk/plugins/denoise/floatplanarimage-x86.cpp
===================================================================
--- trunk/plugins/denoise/floatplanarimage-x86.cpp 2009-08-10 19:58:00 UTC
(rev 2621)
+++ trunk/plugins/denoise/floatplanarimage-x86.cpp 2009-08-17 21:16:07 UTC
(rev 2622)
@@ -22,6 +22,7 @@
#if defined (__x86_64__)
// Only 64 bits, and only if pixelsize is 4
+// FIXME: Apply R/B correction prior to square root
void FloatPlanarImage::unpackInterleavedYUV_SSE( const ImgConvertJob* j )
{
RS_IMAGE16* image = j->rs;
@@ -170,12 +171,12 @@
"addps %%xmm0, %%xmm3\n" // Add Y to blue
"addps %%xmm0, %%xmm4\n" // Add Y to red - xmm 0 free
"mulps %%xmm1, %%xmm1\n" // Square green
- "mulps %%xmm15, %%xmm3\n" // Multiply blue correction - maybe not
needed later
- "mulps %%xmm14, %%xmm4\n" // Multiply red correction - maybe not
needed later
"minps %%xmm9, %%xmm1\n" // Saturate green
"mulps %%xmm3, %%xmm3\n" // Square blue
"mulps %%xmm4, %%xmm4\n" // Square red
"cvtps2dq %%xmm1, %%xmm1\n" // Convert green to dwords
+ "mulps %%xmm15, %%xmm3\n" // Multiply blue correction - maybe not
needed later
+ "mulps %%xmm14, %%xmm4\n" // Multiply red correction - maybe not
needed later
"minps %%xmm9, %%xmm3\n" // Saturate blue
"minps %%xmm9, %%xmm4\n" // Saturate red
"cvtps2dq %%xmm3, %%xmm3\n" // Convert blue to dwords
@@ -259,11 +260,11 @@
"addps %%xmm0, %%xmm3\n" // Add Y to blue
"addps %%xmm0, %%xmm4\n" // Add Y to red - xmm 0 free
"mulps %%xmm1, %%xmm1\n" // Square green
- "mulps %%xmm15, %%xmm3\n" // Multiply blue correction - maybe not
needed later
- "mulps %%xmm14, %%xmm4\n" // Multiply red correction - maybe not
needed later
"mulps %%xmm3, %%xmm3\n" // Square blue
"mulps %%xmm4, %%xmm4\n" // Square red
"cvtps2dq %%xmm1, %%xmm1\n" // Convert green to dwords
+ "mulps %%xmm15, %%xmm3\n" // Multiply blue correction - maybe not
needed later
+ "mulps %%xmm14, %%xmm4\n" // Multiply red correction - maybe not
needed later
"cvtps2dq %%xmm4, %%xmm4\n" // Convert red to dwords
"cvtps2dq %%xmm3, %%xmm3\n" // Convert blue to dwords
"packusdw %%xmm1, %%xmm1\n" // green g3g2 g1g0 g3g2 g1g0
@@ -340,12 +341,12 @@
"addps %%xmm0, %%xmm3\n" // Add Y to blue
"addps %%xmm0, %%xmm4\n" // Add Y to red - xmm 0 free
"mulps %%xmm1, %%xmm1\n" // Square green
- "mulps 80(%4), %%xmm3\n" // Multiply blue correction - maybe not
needed later
- "mulps 64(%4), %%xmm4\n" // Multiply red correction - maybe not
needed later
"minps %%xmm7, %%xmm1\n" // Saturate green
"mulps %%xmm3, %%xmm3\n" // Square blue
"mulps %%xmm4, %%xmm4\n" // Square red
"cvtps2dq %%xmm1, %%xmm1\n" // Convert green to dwords
+ "mulps 80(%4), %%xmm3\n" // Multiply blue correction - maybe not
needed later
+ "mulps 64(%4), %%xmm4\n" // Multiply red correction - maybe not
needed later
"minps %%xmm7, %%xmm3\n" // Saturate blue
"minps %%xmm7, %%xmm4\n" // Saturate red
"cvtps2dq %%xmm3, %%xmm3\n" // Convert blue to dwords
Modified: trunk/plugins/denoise/floatplanarimage.cpp
===================================================================
--- trunk/plugins/denoise/floatplanarimage.cpp 2009-08-10 19:58:00 UTC (rev
2621)
+++ trunk/plugins/denoise/floatplanarimage.cpp 2009-08-17 21:16:07 UTC (rev
2622)
@@ -19,7 +19,7 @@
#include "complexfilter.h"
#include <math.h>
-float FloatPlanarImage::shortToFloat[65536] = {0};
+float FloatPlanarImage::shortToFloat[65536*4] = {0};
FloatPlanarImage::FloatPlanarImage(void) {
p = 0;
@@ -161,29 +161,29 @@
RS_IMAGE16* image = j->rs;
#if defined (__x86_64__)
- if (image->pixelsize == 4)
- return unpackInterleavedYUV_SSE(j);
+// if (image->pixelsize == 4)
+// return unpackInterleavedYUV_SSE(j);
#endif
- gfloat r1 = 0.299 * redCorrection;
- gfloat r2 = -0.169 * redCorrection;
- gfloat r3 = 0.499 * redCorrection;
- gfloat b1 = 0.114 * blueCorrection;
- gfloat b2 = 0.499 * blueCorrection;
- gfloat b3 = -0.0813 * blueCorrection;
+ // We cannot look up more than 65535*4
+ redCorrection = MIN( 4.0f, redCorrection);
+ blueCorrection = MIN( 4.0f, blueCorrection);
+ gint redc = (gint)(16384 * redCorrection + 0.5);
+ gint bluec = (gint)(16384 * blueCorrection + 0.5);
+
for (int y = j->start_y; y < j->end_y; y++ ) {
const gushort* pix = GET_PIXEL(image,0,y);
gfloat *Y = p[0]->getAt(ox, y+oy);
gfloat *Cb = p[1]->getAt(ox, y+oy);
gfloat *Cr = p[2]->getAt(ox, y+oy);
for (int x=0; x<image->w; x++) {
- float r = shortToFloat[(*pix)];
+ float r = shortToFloat[((*pix)*redc)>>14];
float g = shortToFloat[(*(pix+1))];
- float b = shortToFloat[(*(pix+2))];
- *Y++ = r * r1 + g * 0.587 + b * b1 ;
- *Cb++ = r * r2 + g * -0.331 + b * b2;
- *Cr++ = r * r3 + g * -0.418 + b * b3;
+ float b = shortToFloat[((*(pix+2))*bluec)>>14];
+ *Y++ = r * 0.299 + g * 0.587 + b * 0.114 ;
+ *Cb++ = r * -0.169 + g * -0.331 + b * 0.499;
+ *Cr++ = r * 0.499 + g * -0.418 + b * -0.0813;
pix += image->pixelsize;
}
}
@@ -237,12 +237,12 @@
gfloat *Cr = p[2]->getAt(ox, y+oy);
gushort* out = GET_PIXEL(image,0,y);
for (int x=0; x<image->w; x++) {
- float fr = (Y[x] + 1.402 * Cr[x]) * r_factor;
+ float fr = (Y[x] + 1.402 * Cr[x]);
float fg = Y[x] - 0.344 * Cb[x] - 0.714 * Cr[x];
- float fb = (Y[x] + 1.772 * Cb[x]) * b_factor;
- int r = (int)(fr*fr);
+ float fb = (Y[x] + 1.772 * Cb[x]) ;
+ int r = (int)(fr*fr* r_factor);
int g = (int)(fg*fg);
- int b = (int)(fb*fb);
+ int b = (int)(fb*fb* b_factor);
out[0] = clampbits(r,16);
out[1] = clampbits(g,16);
out[2] = clampbits(b,16);
@@ -268,7 +268,7 @@
}
void FloatPlanarImage::initConvTable() {
- for (int i = 0; i < 65536; i++) {
+ for (int i = 0; i < 65536*4; i++) {
shortToFloat[i] = sqrt((float)i);
}
}
Modified: trunk/plugins/denoise/floatplanarimage.h
===================================================================
--- trunk/plugins/denoise/floatplanarimage.h 2009-08-10 19:58:00 UTC (rev
2621)
+++ trunk/plugins/denoise/floatplanarimage.h 2009-08-17 21:16:07 UTC (rev
2622)
@@ -65,7 +65,7 @@
float blueCorrection;
static void initConvTable();
- static float shortToFloat[65536];
+ static float shortToFloat[65536*4];
};
#endif // floatplanarimage_h__
_______________________________________________
Rawstudio-commit mailing list
[email protected]
http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-commit
