Author: post
Date: 2009-08-20 17:15:15 +0200 (Thu, 20 Aug 2009)
New Revision: 2626
Modified:
trunk/plugins/denoise/floatplanarimage-x86.cpp
Log:
Denoise: Refactored float yuv -> rgb shorts to avoid slow clamping, thanks to
akupenguin and Dark Shikari for assistance.
Modified: trunk/plugins/denoise/floatplanarimage-x86.cpp
===================================================================
--- trunk/plugins/denoise/floatplanarimage-x86.cpp 2009-08-19 17:53:17 UTC
(rev 2625)
+++ trunk/plugins/denoise/floatplanarimage-x86.cpp 2009-08-20 15:15:15 UTC
(rev 2626)
@@ -164,7 +164,8 @@
temp[i+12] = 1.772f; // Cb to b
temp[i+16] = (1.0f/redCorrection); // Red correction
temp[i+20] = (1.0f/blueCorrection); // Blue correction
- temp[i+24] = 65535.0f; // Saturation
+ *((gint*)&temp[i+24]) = 32768; // Subtract
+ *((guint*)&temp[i+28]) = 0x80008000; // xor sign shift
}
asm volatile
@@ -175,8 +176,9 @@
"movaps 48(%0), %%xmm13\n" // Cb to b
"movaps 64(%0), %%xmm14\n" // Red Correction
"movaps 80(%0), %%xmm15\n" // Blue Correction
- "movaps 96(%0), %%xmm9\n" // Saturation point
+ "movaps 96(%0), %%xmm9\n" // 0x00008000
"pxor %%xmm8, %%xmm8\n" // Zero
+ "movaps 112(%0), %%xmm7\n" // word 0x8000
: // no output registers
: "r" (temp)
: // %0
@@ -204,31 +206,28 @@
"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
- "minps %%xmm9, %%xmm1\n" // Saturate green
+ "cvtps2dq %%xmm1, %%xmm1\n" // Convert green to dwords
"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
+ "mulps %%xmm15, %%xmm3\n" // Multiply blue correction - maybe not
needed later
+ "mulps %%xmm14, %%xmm4\n" // Multiply red correction - maybe not
needed later
+ "psubd %%xmm9, %%xmm1\n" // g = g - 32768 ( to avoid saturation)
"cvtps2dq %%xmm3, %%xmm3\n" // Convert blue to dwords
+ "packssdw %%xmm1,%%xmm1\n" // g3g2 g1g0 g3g2 g1g0
"cvtps2dq %%xmm4, %%xmm4\n" // Convert red to dwords
- "movdqa %%xmm1, %%xmm0\n" // Copy green into xmm0
- "movdqa %%xmm3, %%xmm2\n" // Copy blue into xmm2
- "movdqa %%xmm4, %%xmm5\n" // Copy red into xmm5
- "pcmpgtd %%xmm8, %%xmm1\n" // if (xmm1 > 0) xmm1 = ones - green
- "pcmpgtd %%xmm8, %%xmm3\n" // same for blue
- "pcmpgtd %%xmm8, %%xmm4\n" // same for red
- "pand %%xmm0, %%xmm1\n" // Green in xmm1
- "pand %%xmm5, %%xmm4\n" // Red in xmm4
- "pslld $16, %%xmm1\n" // Shift up green
- "pand %%xmm2, %%xmm3\n" // Blue in xmm3
- "por %%xmm1, %%xmm4\n" // Interleave red & green
- "movdqa %%xmm4, %%xmm0\n" // Copy red &green into xmm0
+ "pxor %%xmm7, %%xmm1\n" // Shift sign
+ "psubd %%xmm9, %%xmm3\n" // b = b - 32768 ( to avoid saturation)
+ "psubd %%xmm9, %%xmm4\n" // r = r - 32768 ( to avoid saturation)
+ "packssdw %%xmm3,%%xmm3\n" // b3b2 b1b0 b3b2 b1b0
+ "packssdw %%xmm4,%%xmm4\n" // g3g2 g1g0 r3r2 r1r0
+ "pxor %%xmm7, %%xmm3\n" // Shift sign (b)
+ "pxor %%xmm7, %%xmm4\n" // Shift sign (r)
+ "punpcklwd %%xmm1, %%xmm4\n" // g3r3 g2r2 g1r1 g0r0
+ "punpcklwd %%xmm8, %%xmm3\n" // 00b3 00b2 00b1 00b0
+ "movdqa %%xmm4, %%xmm0\n" // Copy r&g
"punpckldq %%xmm3, %%xmm4\n" // Interleave lower blue into reg&green
in xmm4 Now 00b1 g1r1 00b0 g0r0
"punpckhdq %%xmm3, %%xmm0\n" // Interleave higher blue into reg&green
in xmm0 Now 00b3 g3r3 00b2 g2r2
- "movdqa %%xmm4, (%0)\n" // Store low pixels
+ "movdqa %%xmm4, (%0)\n" // Store low pixels
"movdqa %%xmm0, 16(%0)\n" // Store high pixels
"add $32, %0\n"
"add $16, %1\n"
@@ -255,7 +254,6 @@
temp[i+12] = 1.772f; // Cb to b
temp[i+16] = (1.0f/redCorrection); // Red correction
temp[i+20] = (1.0f/blueCorrection); // Blue correction
- temp[i+24] = 65535.0f; // Saturation
}
asm volatile
@@ -339,13 +337,15 @@
temp[i+12] = 1.772f; // Cb to b
temp[i+16] = (1.0f/redCorrection); // Red correction
temp[i+20] = (1.0f/blueCorrection); // Blue correction
- temp[i+24] = 65535.0f; // Saturation
+ *((gint*)&temp[i+24]) = 32768; // Subtract
+ *((guint*)&temp[i+28]) = 0x80008000; // xor sign shift
}
int* itemp = (int*)(&temp[28]);
asm volatile
(
- "movaps 96(%0), %%xmm7\n" // Saturation point
+ "movaps 96(%0), %%xmm7\n" // Subtract
+ "movaps 112(%0), %%xmm5\n" // Xor sign
"pxor %%xmm6, %%xmm6\n" // Zero
: // no output registers
: "r" (temp)
@@ -374,30 +374,28 @@
"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
- "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
+ "psubd %%xmm7, %%xmm1\n" // g = g - 32768 ( to avoid saturation)
"cvtps2dq %%xmm3, %%xmm3\n" // Convert blue to dwords
+ "packssdw %%xmm1,%%xmm1\n" // g3g2 g1g0 g3g2 g1g0
"cvtps2dq %%xmm4, %%xmm4\n" // Convert red to dwords
- "movdqa %%xmm1, %%xmm0\n" // Copy green into xmm0
- "movdqa %%xmm3, %%xmm2\n" // Copy blue into xmm2
- "movdqa %%xmm4, %%xmm5\n" // Copy red into xmm5
- "pcmpgtd %%xmm6, %%xmm1\n" // if (xmm1 > 0) xmm1 = ones - green
- "pcmpgtd %%xmm6, %%xmm3\n" // same for blue
- "pcmpgtd %%xmm6, %%xmm4\n" // same for red
- "pand %%xmm0, %%xmm1\n" // Green in xmm1
- "pand %%xmm5, %%xmm4\n" // Red in xmm4
- "pslld $16, %%xmm1\n" // Shift up green
- "pand %%xmm2, %%xmm3\n" // Blue in xmm3
- "por %%xmm1, %%xmm4\n" // Interleave red & green
- "movdqa %%xmm4, %%xmm0\n" // Copy red &green into xmm0
+ "pxor %%xmm5, %%xmm1\n" // Shift sign
+ "psubd %%xmm7, %%xmm3\n" // b = b - 32768 ( to avoid saturation)
+ "psubd %%xmm7, %%xmm4\n" // r = r - 32768 ( to avoid saturation)
+ "packssdw %%xmm3,%%xmm3\n" // b3b2 b1b0 b3b2 b1b0
+ "packssdw %%xmm4,%%xmm4\n" // g3g2 g1g0 r3r2 r1r0
+ "pxor %%xmm5, %%xmm3\n" // Shift sign (b)
+ "pxor %%xmm5, %%xmm4\n" // Shift sign (r)
+ "punpcklwd %%xmm1, %%xmm4\n" // g3r3 g2r2 g1r1 g0r0
+ "punpcklwd %%xmm8, %%xmm3\n" // 00b3 00b2 00b1 00b0
+ "movdqa %%xmm4, %%xmm0\n" // Copy r&g
"punpckldq %%xmm3, %%xmm4\n" // Interleave lower blue into reg&green
in xmm4 Now 00b1 g1r1 00b0 g0r0
"punpckhdq %%xmm3, %%xmm0\n" // Interleave higher blue into reg&green
in xmm0 Now 00b3 g3r3 00b2 g2r2
+
"movdqa %%xmm4, (%0)\n" // Store low pixels
"movdqa %%xmm0, 16(%0)\n" // Store high pixels
"add $32, %0\n"
_______________________________________________
Rawstudio-commit mailing list
[email protected]
http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-commit
