instead of using a stack array, use regular variables (marked with the register keyword to hint the optimizer) for the color data. At least over here this helps the optimizer *A LOT* - instead of keeping w[] on the stack, on the PowerPC all values stay in registers, for a nice speed boost. Shouldn't hurt x86 either

svn-id: r10522

1 files changed, 72 insertions, 71 deletions

diff --git a/common/scaler/hq3x.cpp b/common/scaler/hq3x.cpp
index 864b8c31cf..39be5e4f98 100644
--- a/common/scaler/hq3x.cpp
+++ b/common/scaler/hq3x.cpp
@@ -23,61 +23,61 @@
 #include "common/scaler/intern.h"
 
 
-#define PIXEL00_1M  *(q) = interpolate16_2<bitFormat,3,1>(w[5], w[1]);
-#define PIXEL00_1U  *(q) = interpolate16_2<bitFormat,3,1>(w[5], w[2]);
-#define PIXEL00_1L  *(q) = interpolate16_2<bitFormat,3,1>(w[5], w[4]);
-#define PIXEL00_2   *(q) = interpolate16_3<bitFormat,2,1,1>(w[5], w[4], w[2]);
-#define PIXEL00_4   *(q) = interpolate16_3<bitFormat,2,7,7>(w[5], w[4], w[2]);
-#define PIXEL00_5   *(q) = interpolate16_2<bitFormat,1,1>(w[4], w[2]);
-#define PIXEL00_C   *(q) = w[5];
+#define PIXEL00_1M  *(q) = interpolate16_2<bitFormat,3,1>(w5, w1);
+#define PIXEL00_1U  *(q) = interpolate16_2<bitFormat,3,1>(w5, w2);
+#define PIXEL00_1L  *(q) = interpolate16_2<bitFormat,3,1>(w5, w4);
+#define PIXEL00_2   *(q) = interpolate16_3<bitFormat,2,1,1>(w5, w4, w2);
+#define PIXEL00_4   *(q) = interpolate16_3<bitFormat,2,7,7>(w5, w4, w2);
+#define PIXEL00_5   *(q) = interpolate16_2<bitFormat,1,1>(w4, w2);
+#define PIXEL00_C   *(q) = w5;
 
-#define PIXEL01_1   *(q+1) = interpolate16_2<bitFormat,3,1>(w[5], w[2]);
-#define PIXEL01_3   *(q+1) = interpolate16_2<bitFormat,7,1>(w[5], w[2]);
-#define PIXEL01_6   *(q+1) = interpolate16_2<bitFormat,3,1>(w[2], w[5]);
-#define PIXEL01_C   *(q+1) = w[5];
+#define PIXEL01_1   *(q+1) = interpolate16_2<bitFormat,3,1>(w5, w2);
+#define PIXEL01_3   *(q+1) = interpolate16_2<bitFormat,7,1>(w5, w2);
+#define PIXEL01_6   *(q+1) = interpolate16_2<bitFormat,3,1>(w2, w5);
+#define PIXEL01_C   *(q+1) = w5;
 
-#define PIXEL02_1M  *(q+2) = interpolate16_2<bitFormat,3,1>(w[5], w[3]);
-#define PIXEL02_1U  *(q+2) = interpolate16_2<bitFormat,3,1>(w[5], w[2]);
-#define PIXEL02_1R  *(q+2) = interpolate16_2<bitFormat,3,1>(w[5], w[6]);
-#define PIXEL02_2   *(q+2) = interpolate16_3<bitFormat,2,1,1>(w[5], w[2], w[6]);
-#define PIXEL02_4   *(q+2) = interpolate16_3<bitFormat,2,7,7>(w[5], w[2], w[6]);
-#define PIXEL02_5   *(q+2) = interpolate16_2<bitFormat,1,1>(w[2], w[6]);
-#define PIXEL02_C   *(q+2) = w[5];
+#define PIXEL02_1M  *(q+2) = interpolate16_2<bitFormat,3,1>(w5, w3);
+#define PIXEL02_1U  *(q+2) = interpolate16_2<bitFormat,3,1>(w5, w2);
+#define PIXEL02_1R  *(q+2) = interpolate16_2<bitFormat,3,1>(w5, w6);
+#define PIXEL02_2   *(q+2) = interpolate16_3<bitFormat,2,1,1>(w5, w2, w6);
+#define PIXEL02_4   *(q+2) = interpolate16_3<bitFormat,2,7,7>(w5, w2, w6);
+#define PIXEL02_5   *(q+2) = interpolate16_2<bitFormat,1,1>(w2, w6);
+#define PIXEL02_C   *(q+2) = w5;
 
-#define PIXEL10_1   *(q+nextlineDst) = interpolate16_2<bitFormat,3,1>(w[5], w[4]);
-#define PIXEL10_3   *(q+nextlineDst) = interpolate16_2<bitFormat,7,1>(w[5], w[4]);
-#define PIXEL10_6   *(q+nextlineDst) = interpolate16_2<bitFormat,3,1>(w[4], w[5]);
-#define PIXEL10_C   *(q+nextlineDst) = w[5];
+#define PIXEL10_1   *(q+nextlineDst) = interpolate16_2<bitFormat,3,1>(w5, w4);
+#define PIXEL10_3   *(q+nextlineDst) = interpolate16_2<bitFormat,7,1>(w5, w4);
+#define PIXEL10_6   *(q+nextlineDst) = interpolate16_2<bitFormat,3,1>(w4, w5);
+#define PIXEL10_C   *(q+nextlineDst) = w5;
 
-#define PIXEL11     *(q+1+nextlineDst) = w[5];
+#define PIXEL11     *(q+1+nextlineDst) = w5;
 
-#define PIXEL12_1   *(q+2+nextlineDst) = interpolate16_2<bitFormat,3,1>(w[5], w[6]);
-#define PIXEL12_3   *(q+2+nextlineDst) = interpolate16_2<bitFormat,7,1>(w[5], w[6]);
-#define PIXEL12_6   *(q+2+nextlineDst) = interpolate16_2<bitFormat,3,1>(w[6], w[5]);
-#define PIXEL12_C   *(q+2+nextlineDst) = w[5];
+#define PIXEL12_1   *(q+2+nextlineDst) = interpolate16_2<bitFormat,3,1>(w5, w6);
+#define PIXEL12_3   *(q+2+nextlineDst) = interpolate16_2<bitFormat,7,1>(w5, w6);
+#define PIXEL12_6   *(q+2+nextlineDst) = interpolate16_2<bitFormat,3,1>(w6, w5);
+#define PIXEL12_C   *(q+2+nextlineDst) = w5;
 
-#define PIXEL20_1M  *(q+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[7]);
-#define PIXEL20_1D  *(q+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[8]);
-#define PIXEL20_1L  *(q+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[4]);
-#define PIXEL20_2   *(q+nextlineDst2) = interpolate16_3<bitFormat,2,1,1>(w[5], w[8], w[4]);
-#define PIXEL20_4   *(q+nextlineDst2) = interpolate16_3<bitFormat,2,7,7>(w[5], w[8], w[4]);
-#define PIXEL20_5   *(q+nextlineDst2) = interpolate16_2<bitFormat,1,1>(w[8], w[4]);
-#define PIXEL20_C   *(q+nextlineDst2) = w[5];
+#define PIXEL20_1M  *(q+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w7);
+#define PIXEL20_1D  *(q+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w8);
+#define PIXEL20_1L  *(q+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w4);
+#define PIXEL20_2   *(q+nextlineDst2) = interpolate16_3<bitFormat,2,1,1>(w5, w8, w4);
+#define PIXEL20_4   *(q+nextlineDst2) = interpolate16_3<bitFormat,2,7,7>(w5, w8, w4);
+#define PIXEL20_5   *(q+nextlineDst2) = interpolate16_2<bitFormat,1,1>(w8, w4);
+#define PIXEL20_C   *(q+nextlineDst2) = w5;
 
-#define PIXEL21_1   *(q+1+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[8]);
-#define PIXEL21_3   *(q+1+nextlineDst2) = interpolate16_2<bitFormat,7,1>(w[5], w[8]);
-#define PIXEL21_6   *(q+1+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[8], w[5]);
-#define PIXEL21_C   *(q+1+nextlineDst2) = w[5];
+#define PIXEL21_1   *(q+1+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w8);
+#define PIXEL21_3   *(q+1+nextlineDst2) = interpolate16_2<bitFormat,7,1>(w5, w8);
+#define PIXEL21_6   *(q+1+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w8, w5);
+#define PIXEL21_C   *(q+1+nextlineDst2) = w5;
 
-#define PIXEL22_1M  *(q+2+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[9]);
-#define PIXEL22_1D  *(q+2+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[8]);
-#define PIXEL22_1R  *(q+2+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w[5], w[6]);
-#define PIXEL22_2   *(q+2+nextlineDst2) = interpolate16_3<bitFormat,2,1,1>(w[5], w[6], w[8]);
-#define PIXEL22_4   *(q+2+nextlineDst2) = interpolate16_3<bitFormat,2,7,7>(w[5], w[6], w[8]);
-#define PIXEL22_5   *(q+2+nextlineDst2) = interpolate16_2<bitFormat,1,1>(w[6], w[8]);
-#define PIXEL22_C   *(q+2+nextlineDst2) = w[5];
+#define PIXEL22_1M  *(q+2+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w9);
+#define PIXEL22_1D  *(q+2+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w8);
+#define PIXEL22_1R  *(q+2+nextlineDst2) = interpolate16_2<bitFormat,3,1>(w5, w6);
+#define PIXEL22_2   *(q+2+nextlineDst2) = interpolate16_3<bitFormat,2,1,1>(w5, w6, w8);
+#define PIXEL22_4   *(q+2+nextlineDst2) = interpolate16_3<bitFormat,2,7,7>(w5, w6, w8);
+#define PIXEL22_5   *(q+2+nextlineDst2) = interpolate16_2<bitFormat,1,1>(w6, w8);
+#define PIXEL22_C   *(q+2+nextlineDst2) = w5;
 
-#define YUV(x)	RGBtoYUV[w[x]]
+#define YUV(x)	RGBtoYUV[w ## x]
 
 /**
  * The HQ3x high quality 3x graphics filter.
@@ -86,8 +86,9 @@
  */
 template<int bitFormat>
 void HQ3x(const uint8 *srcPtr, uint32 srcPitch, uint8 *dstPtr, uint32 dstPitch, int width, int height) {
-	int   w[10];
-  
+//	int   w[10];
+ 	register int  w1, w2, w3, w4, w5, w6, w7, w8, w9;
+ 
 	const uint32 nextlineSrc = srcPitch / sizeof(uint16);
 	const uint16 *p = (const uint16 *)srcPtr;
 
@@ -120,41 +121,41 @@ void HQ3x(const uint8 *srcPtr, uint32 srcPitch, uint8 *dstPtr, uint32 dstPitch,
 	//	 +----+----+----+
 
 	while (height--) {
-		w[2] = *(p - 1 - nextlineSrc);
-		w[5] = *(p - 1);
-		w[8] = *(p - 1 + nextlineSrc);
+		w2 = *(p - 1 - nextlineSrc);
+		w5 = *(p - 1);
+		w8 = *(p - 1 + nextlineSrc);
 
-		w[3] = *(p - nextlineSrc);
-		w[6] = *(p);
-		w[9] = *(p + nextlineSrc);
+		w3 = *(p - nextlineSrc);
+		w6 = *(p);
+		w9 = *(p + nextlineSrc);
 
 		int tmpWidth = width;
 		while (tmpWidth--) {
 			p++;
 
-			w[1] = w[2];
-			w[4] = w[5];
-			w[7] = w[8];
+			w1 = w2;
+			w4 = w5;
+			w7 = w8;
 
-			w[2] = w[3];
-			w[5] = w[6];
-			w[8] = w[9];
+			w2 = w3;
+			w5 = w6;
+			w8 = w9;
 
-			w[3] = *(p - nextlineSrc);
-			w[6] = *(p);
-			w[9] = *(p + nextlineSrc);
+			w3 = *(p - nextlineSrc);
+			w6 = *(p);
+			w9 = *(p + nextlineSrc);
 
 			int pattern = 0;
 			const int yuv5 = YUV(5);
 
-			if (w[5] != w[1] && diffYUV(yuv5, YUV(1))) pattern |= 0x0001;
-			if (w[5] != w[2] && diffYUV(yuv5, YUV(2))) pattern |= 0x0002;
-			if (w[5] != w[3] && diffYUV(yuv5, YUV(3))) pattern |= 0x0004;
-			if (w[5] != w[4] && diffYUV(yuv5, YUV(4))) pattern |= 0x0008;
-			if (w[5] != w[6] && diffYUV(yuv5, YUV(6))) pattern |= 0x0010;
-			if (w[5] != w[7] && diffYUV(yuv5, YUV(7))) pattern |= 0x0020;
-			if (w[5] != w[8] && diffYUV(yuv5, YUV(8))) pattern |= 0x0040;
-			if (w[5] != w[9] && diffYUV(yuv5, YUV(9))) pattern |= 0x0080;
+			if (w5 != w1 && diffYUV(yuv5, YUV(1))) pattern |= 0x0001;
+			if (w5 != w2 && diffYUV(yuv5, YUV(2))) pattern |= 0x0002;
+			if (w5 != w3 && diffYUV(yuv5, YUV(3))) pattern |= 0x0004;
+			if (w5 != w4 && diffYUV(yuv5, YUV(4))) pattern |= 0x0008;
+			if (w5 != w6 && diffYUV(yuv5, YUV(6))) pattern |= 0x0010;
+			if (w5 != w7 && diffYUV(yuv5, YUV(7))) pattern |= 0x0020;
+			if (w5 != w8 && diffYUV(yuv5, YUV(8))) pattern |= 0x0040;
+			if (w5 != w9 && diffYUV(yuv5, YUV(9))) pattern |= 0x0080;
 
 			switch (pattern) {
 			case 0: