Mercurial > sdl-ios-xcode

diff src/video/SDL_RLEaccel.c @ 689:5bb080d35049

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Date: Tue, 19 Aug 2003 17:57:00 +0200 From: Stephane Marchesin Subject: Re: [SDL] [patch] MMX alpha blit patches with MMX detection I think everything is correct now. I've done as much testing as I could, but some real-world testing wouldn't hurt, I think. The patch is here : http://icps.u-strasbg.fr/~marchesin/sdl_mmxblit.patch If you do byte-by-byte comparison of the output between C and MMX functions, you'll notice that the results for 555 and 565 RGB alpha blits aren't exactly the same. This is because MMX functions for 555 and 565 RGB have an higher accuracy. If you want the exact same behaviour that's possible by masking the three lower alpha bits in the MMX functions. Just ask ! I removed one MMX function because after I fixed it to match its C equivalent, it revealed to be slower than the C version on a PIII (although a bit faster on an Athlon XP). I've also added MMX and PIII replacements for SDL_memcpy. Those provide some speed up in testvidinfo -benchmark (at least for me, under linux & X11).
author Sam Lantinga <slouken@libsdl.org>
date Fri, 22 Aug 2003 05:51:19 +0000
parents 4314a501d7be
children 22dbf364c017
line wrap: on
line diff
--- a/src/video/SDL_RLEaccel.c	Tue Aug 12 15:17:20 2003 +0000
+++ b/src/video/SDL_RLEaccel.c	Fri Aug 22 05:51:19 2003 +0000
@@ -102,6 +102,18 @@
 #include "SDL_memops.h"
 #include "SDL_RLEaccel_c.h"
 
+#if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
+#include "mmx.h"
+/* Function to check the CPU flags */
+#define MMX_CPU		0x800000
+#define CPU_Flags()	Hermes_X86_CPU()
+#define X86_ASSEMBLER
+#define HermesConverterInterface	void
+#define HermesClearInterface		void
+#define STACKCALL
+#include "HeadX86.h"
+#endif
+
 #ifndef MAX
 #define MAX(a, b) ((a) > (b) ? (a) : (b))
 #endif
@@ -125,6 +137,262 @@
 #define OPAQUE_BLIT(to, from, length, bpp, alpha)	\
     PIXEL_COPY(to, from, length, bpp)
 
+#if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
+
+#define ALPHA_BLIT32_888MMX(to, from, length, bpp, alpha)	\
+    do {							\
+	Uint32 *srcp = (Uint32 *)(from);			\
+	Uint32 *dstp = (Uint32 *)(to);				\
+        int i = 0x00FF00FF;					\
+        movd_m2r(*(&i), mm3);					\
+        punpckldq_r2r(mm3, mm3);				\
+        i = 0xFF000000;						\
+        movd_m2r(*(&i), mm7);					\
+        punpckldq_r2r(mm7, mm7);				\
+        i = alpha | alpha << 16;				\
+        movd_m2r(*(&i), mm4);					\
+        punpckldq_r2r(mm4, mm4);				\
+	pcmpeqd_r2r(mm5,mm5); /* set mm5 to "1" */		\
+	pxor_r2r(mm7, mm5); /* make clear alpha mask */		\
+        i = length;						\
+	if(i & 1) {						\
+          movd_m2r((*srcp), mm1); /* src -> mm1 */		\
+          punpcklbw_r2r(mm1, mm1);				\
+          pand_r2r(mm3, mm1);					\
+	  movd_m2r((*dstp), mm2); /* dst -> mm2 */		\
+          punpcklbw_r2r(mm2, mm2);				\
+          pand_r2r(mm3, mm2);					\
+	  psubw_r2r(mm2, mm1);					\
+	  pmullw_r2r(mm4, mm1);					\
+	  psrlw_i2r(8, mm1);					\
+	  paddw_r2r(mm1, mm2);					\
+	  pand_r2r(mm3, mm2);					\
+	  packuswb_r2r(mm2, mm2);				\
+	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
+	  movd_r2m(mm2, *dstp);					\
+	  ++srcp;						\
+	  ++dstp;						\
+	  i--;							\
+	}							\
+	for(; i > 0; --i) {					\
+          movq_m2r((*srcp), mm0);				\
+	  movq_r2r(mm0, mm1);					\
+          punpcklbw_r2r(mm0, mm0);				\
+	  movq_m2r((*dstp), mm2);				\
+	  punpckhbw_r2r(mm1, mm1);				\
+	  movq_r2r(mm2, mm6);					\
+          pand_r2r(mm3, mm0);					\
+          punpcklbw_r2r(mm2, mm2);				\
+	  pand_r2r(mm3, mm1);					\
+	  punpckhbw_r2r(mm6, mm6);				\
+          pand_r2r(mm3, mm2);					\
+	  psubw_r2r(mm2, mm0);					\
+	  pmullw_r2r(mm4, mm0);					\
+	  pand_r2r(mm3, mm6);					\
+	  psubw_r2r(mm6, mm1);					\
+	  pmullw_r2r(mm4, mm1);					\
+	  psrlw_i2r(8, mm0);					\
+	  paddw_r2r(mm0, mm2);					\
+	  psrlw_i2r(8, mm1);					\
+	  paddw_r2r(mm1, mm6);					\
+	  pand_r2r(mm3, mm2);					\
+	  pand_r2r(mm3, mm6);					\
+	  packuswb_r2r(mm2, mm2);				\
+	  packuswb_r2r(mm6, mm6);				\
+	  psrlq_i2r(32, mm2);					\
+	  psllq_i2r(32, mm6);					\
+	  por_r2r(mm6, mm2);					\
+	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
+         movq_r2m(mm2, *dstp);					\
+	  srcp += 2;						\
+	  dstp += 2;						\
+	  i--;							\
+	}							\
+	emms();							\
+    } while(0)
+
+#define ALPHA_BLIT16_565MMX(to, from, length, bpp, alpha)	\
+    do {						\
+        int i, n = 0;					\
+	Uint16 *srcp = (Uint16 *)(from);		\
+	Uint16 *dstp = (Uint16 *)(to);			\
+        Uint32 ALPHA = 0xF800;				\
+	movd_m2r(*(&ALPHA), mm1);			\
+        punpcklwd_r2r(mm1, mm1);			\
+        punpcklwd_r2r(mm1, mm1);			\
+	ALPHA = 0x07E0;					\
+	movd_m2r(*(&ALPHA), mm4);			\
+        punpcklwd_r2r(mm4, mm4);			\
+        punpcklwd_r2r(mm4, mm4);			\
+	ALPHA = 0x001F;					\
+	movd_m2r(*(&ALPHA), mm7);			\
+        punpcklwd_r2r(mm7, mm7);			\
+        punpcklwd_r2r(mm7, mm7);			\
+	alpha &= ~(1+2+4);				\
+        i = (Uint32)alpha | (Uint32)alpha << 16;	\
+        movd_m2r(*(&i), mm0);				\
+        punpckldq_r2r(mm0, mm0);			\
+        ALPHA = alpha >> 3;				\
+        i = ((int)(length) & 3);			\
+	for(; i > 0; --i) {				\
+	    Uint32 s = *srcp++;				\
+	    Uint32 d = *dstp;				\
+	    s = (s | s << 16) & 0x07e0f81f;		\
+	    d = (d | d << 16) & 0x07e0f81f;		\
+	    d += (s - d) * ALPHA >> 5;			\
+	    d &= 0x07e0f81f;				\
+	    *dstp++ = d | d >> 16;			\
+	    n++;					\
+	}						\
+	i = (int)(length) - n;				\
+	for(; i > 0; --i) {				\
+	  movq_m2r((*dstp), mm3);			\
+	  movq_m2r((*srcp), mm2);			\
+	  movq_r2r(mm2, mm5);				\
+	  pand_r2r(mm1 , mm5);				\
+	  psrlq_i2r(11, mm5);				\
+	  movq_r2r(mm3, mm6);				\
+	  pand_r2r(mm1 , mm6);				\
+	  psrlq_i2r(11, mm6);				\
+	  psubw_r2r(mm6, mm5);				\
+	  pmullw_r2r(mm0, mm5);				\
+	  psrlw_i2r(8, mm5);				\
+	  paddw_r2r(mm5, mm6);				\
+	  psllq_i2r(11, mm6);				\
+	  pand_r2r(mm1, mm6);				\
+	  movq_r2r(mm4, mm5);				\
+	  por_r2r(mm7, mm5);				\
+	  pand_r2r(mm5, mm3);				\
+	  por_r2r(mm6, mm3);				\
+	  movq_r2r(mm2, mm5);				\
+	  pand_r2r(mm4 , mm5);				\
+	  psrlq_i2r(5, mm5);				\
+	  movq_r2r(mm3, mm6);				\
+	  pand_r2r(mm4 , mm6);				\
+	  psrlq_i2r(5, mm6);				\
+	  psubw_r2r(mm6, mm5);				\
+	  pmullw_r2r(mm0, mm5);				\
+	  psrlw_i2r(8, mm5);				\
+	  paddw_r2r(mm5, mm6);				\
+	  psllq_i2r(5, mm6);				\
+	  pand_r2r(mm4, mm6);				\
+	  movq_r2r(mm1, mm5);				\
+	  por_r2r(mm7, mm5);				\
+	  pand_r2r(mm5, mm3);				\
+	  por_r2r(mm6, mm3);				\
+	  movq_r2r(mm2, mm5);				\
+	  pand_r2r(mm7 , mm5);				\
+          movq_r2r(mm3, mm6);				\
+	  pand_r2r(mm7 , mm6);				\
+	  psubw_r2r(mm6, mm5);				\
+	  pmullw_r2r(mm0, mm5);				\
+	  psrlw_i2r(8, mm5);				\
+	  paddw_r2r(mm5, mm6);				\
+	  pand_r2r(mm7, mm6);				\
+	  movq_r2r(mm1, mm5);				\
+	  por_r2r(mm4, mm5);				\
+	  pand_r2r(mm5, mm3);				\
+	  por_r2r(mm6, mm3);				\
+	  movq_r2m(mm3, *dstp);				\
+	  srcp += 4;					\
+	  dstp += 4;					\
+	  i -= 3;					\
+	}						\
+	emms();						\
+    } while(0)
+
+#define ALPHA_BLIT16_555MMX(to, from, length, bpp, alpha)	\
+    do {						\
+        int i, n = 0;					\
+	Uint16 *srcp = (Uint16 *)(from);		\
+	Uint16 *dstp = (Uint16 *)(to);			\
+        Uint32 ALPHA = 0x7C00;				\
+	movd_m2r(*(&ALPHA), mm1);			\
+        punpcklwd_r2r(mm1, mm1);			\
+        punpcklwd_r2r(mm1, mm1);			\
+	ALPHA = 0x03E0;					\
+        movd_m2r(*(&ALPHA), mm4);			\
+        punpcklwd_r2r(mm4, mm4);			\
+        punpcklwd_r2r(mm4, mm4);			\
+	ALPHA = 0x001F;					\
+	movd_m2r(*(&ALPHA), mm7);			\
+        punpcklwd_r2r(mm7, mm7);			\
+        punpcklwd_r2r(mm7, mm7);			\
+	alpha &= ~(1+2+4);				\
+        i = (Uint32)alpha | (Uint32)alpha << 16;	\
+        movd_m2r(*(&i), mm0);				\
+        punpckldq_r2r(mm0, mm0);			\
+        i = ((int)(length) & 3);				\
+        ALPHA = alpha >> 3;				\
+	for(; i > 0; --i) {				\
+	    Uint32 s = *srcp++;				\
+	    Uint32 d = *dstp;				\
+	    s = (s | s << 16) & 0x03e07c1f;		\
+	    d = (d | d << 16) & 0x03e07c1f;		\
+	    d += (s - d) * ALPHA >> 5;			\
+	    d &= 0x03e07c1f;				\
+	    *dstp++ = d | d >> 16;			\
+	    n++;					\
+	}						\
+	i = (int)(length) - n;				\
+	for(; i > 0; --i) {				\
+	  movq_m2r((*dstp), mm3);			\
+	  movq_m2r((*srcp), mm2);			\
+	  movq_r2r(mm2, mm5);				\
+	  pand_r2r(mm1 , mm5);				\
+	  psrlq_i2r(10, mm5);				\
+	  movq_r2r(mm3, mm6);				\
+	  pand_r2r(mm1 , mm6);				\
+	  psrlq_i2r(10, mm6);				\
+	  psubw_r2r(mm6, mm5);				\
+	  pmullw_r2r(mm0, mm5);				\
+	  psrlw_i2r(8, mm5);				\
+	  paddw_r2r(mm5, mm6);				\
+	  psllq_i2r(10, mm6);				\
+	  pand_r2r(mm1, mm6);				\
+	  movq_r2r(mm4, mm5);				\
+	  por_r2r(mm7, mm5);				\
+	  pand_r2r(mm5, mm3);				\
+	  por_r2r(mm6, mm3);				\
+	  movq_r2r(mm2, mm5);				\
+	  pand_r2r(mm4 , mm5);				\
+	  psrlq_i2r(5, mm5);				\
+	  movq_r2r(mm3, mm6);				\
+	  pand_r2r(mm4 , mm6);				\
+	  psrlq_i2r(5, mm6);				\
+	  psubw_r2r(mm6, mm5);				\
+	  pmullw_r2r(mm0, mm5);				\
+	  psrlw_i2r(8, mm5);				\
+	  paddw_r2r(mm5, mm6);				\
+	  psllq_i2r(5, mm6);				\
+	  pand_r2r(mm4, mm6);				\
+	  movq_r2r(mm1, mm5);				\
+	  por_r2r(mm7, mm5);				\
+	  pand_r2r(mm5, mm3);				\
+	  por_r2r(mm6, mm3);				\
+	  movq_r2r(mm2, mm5);				\
+	  pand_r2r(mm7 , mm5);				\
+          movq_r2r(mm3, mm6);				\
+	  pand_r2r(mm7 , mm6);				\
+	  psubw_r2r(mm6, mm5);				\
+	  pmullw_r2r(mm0, mm5);				\
+	  psrlw_i2r(8, mm5);				\
+	  paddw_r2r(mm5, mm6);				\
+	  pand_r2r(mm7, mm6);				\
+	  movq_r2r(mm1, mm5);				\
+	  por_r2r(mm4, mm5);				\
+	  pand_r2r(mm5, mm3);				\
+	  por_r2r(mm6, mm3);				\
+	  movq_r2m(mm3, *dstp);				\
+	  srcp += 4;					\
+	  dstp += 4;					\
+	  i -= 3;					\
+	}						\
+	emms();						\
+    } while(0)
+
+#endif
+
 /*
  * For 32bpp pixels on the form 0x00rrggbb:
  * If we treat the middle component separately, we can process the two
@@ -161,12 +429,13 @@
         int i;						\
 	Uint16 *src = (Uint16 *)(from);			\
 	Uint16 *dst = (Uint16 *)(to);			\
+	Uint32 ALPHA = alpha >> 3;			\
 	for(i = 0; i < (int)(length); i++) {		\
 	    Uint32 s = *src++;				\
 	    Uint32 d = *dst;				\
 	    s = (s | s << 16) & 0x07e0f81f;		\
 	    d = (d | d << 16) & 0x07e0f81f;		\
-	    d += (s - d) * alpha >> 5;			\
+	    d += (s - d) * ALPHA >> 5;			\
 	    d &= 0x07e0f81f;				\
 	    *dst++ = d | d >> 16;			\
 	}						\
@@ -177,12 +446,13 @@
         int i;						\
 	Uint16 *src = (Uint16 *)(from);			\
 	Uint16 *dst = (Uint16 *)(to);			\
+	Uint32 ALPHA = alpha >> 3;			\
 	for(i = 0; i < (int)(length); i++) {		\
 	    Uint32 s = *src++;				\
 	    Uint32 d = *dst;				\
 	    s = (s | s << 16) & 0x03e07c1f;		\
 	    d = (d | d << 16) & 0x03e07c1f;		\
-	    d += (s - d) * alpha >> 5;			\
+	    d += (s - d) * ALPHA >> 5;			\
 	    d &= 0x03e07c1f;				\
 	    *dst++ = d | d >> 16;			\
 	}						\
@@ -248,7 +518,48 @@
 	}								\
     } while(0)
 
+#if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
 
+#define ALPHA_BLIT32_888_50MMX(to, from, length, bpp, alpha)		\
+    do {								\
+	Uint32 *srcp = (Uint32 *)(from);				\
+	Uint32 *dstp = (Uint32 *)(to);					\
+        int i = 0x00fefefe;						\
+        movd_m2r(*(&i), mm4);						\
+        punpckldq_r2r(mm4, mm4);					\
+        i = 0x00010101;							\
+        movd_m2r(*(&i), mm3);						\
+        punpckldq_r2r(mm3, mm3);					\
+        i = (int)(length);						\
+        if( i & 1 ) {							\
+	  Uint32 s = *srcp++;						\
+	  Uint32 d = *dstp;						\
+	  *dstp++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
+		     + (s & d & 0x00010101);				\
+	  i--;								\
+	}								\
+	for(; i > 0; --i) {						\
+	    movq_m2r((*dstp), mm2); /* dst -> mm2 */			\
+	    movq_r2r(mm2, mm6);	/* dst -> mm6 */			\
+	    movq_m2r((*srcp), mm1); /* src -> mm1 */			\
+	    movq_r2r(mm1, mm5);	/* src -> mm5 */			\
+	    pand_r2r(mm4, mm6);	/* dst & 0x00fefefe -> mm6 */		\
+	    pand_r2r(mm4, mm5); /* src & 0x00fefefe -> mm5 */		\
+	    paddd_r2r(mm6, mm5); /* (dst & 0x00fefefe) + (dst & 0x00fefefe) -> mm5 */	\
+	    psrld_i2r(1, mm5);						\
+	    pand_r2r(mm1, mm2);	/* s & d -> mm2 */			\
+	    pand_r2r(mm3, mm2);	/* s & d & 0x00010101 -> mm2 */		\
+	    paddd_r2r(mm5, mm2);					\
+	    movq_r2m(mm2, (*dstp));					\
+	    dstp += 2;							\
+	    srcp += 2;							\
+	    i--;							\
+	}								\
+	emms();								\
+    } while(0)
+
+#endif
+    
 /*
  * Special case: 50% alpha (alpha=128)
  * This is treated specially because it can be optimized very well, and
@@ -320,6 +631,7 @@
 #define ALPHA_BLIT16_555_50(to, from, length, bpp, alpha)	\
     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xfbde)
 
+#if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
 
 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
     do {								\
@@ -345,7 +657,92 @@
 			if(alpha == 128)				\
 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
 			else {						\
-			    alpha >>= 3; /* use 5 bit alpha */		\
+			    if((CPU_Flags()&MMX_CPU)!=0)		\
+				blitter(2, Uint8, ALPHA_BLIT16_565MMX);	\
+			    else					\
+				blitter(2, Uint8, ALPHA_BLIT16_565);	\
+			}						\
+		    } else						\
+			goto general16;					\
+		    break;						\
+									\
+		case 0x7fff:						\
+		    if(fmt->Gmask == 0x03e0				\
+		       || fmt->Rmask == 0x03e0				\
+		       || fmt->Bmask == 0x03e0) {			\
+			if(alpha == 128)				\
+			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
+			else {						\
+			    if((CPU_Flags()&MMX_CPU)!=0)		\
+				blitter(2, Uint8, ALPHA_BLIT16_555MMX);	\
+			    else					\
+				blitter(2, Uint8, ALPHA_BLIT16_555);	\
+			}						\
+			break;						\
+		    }							\
+		    /* fallthrough */					\
+									\
+		default:						\
+		general16:						\
+		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
+		}							\
+		break;							\
+									\
+	    case 3:							\
+		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
+		break;							\
+									\
+	    case 4:							\
+		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
+		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
+		       || fmt->Bmask == 0xff00)) {			\
+		    if(alpha == 128)					\
+		    {							\
+			if((CPU_Flags()&MMX_CPU)!=0)			\
+				blitter(4, Uint16, ALPHA_BLIT32_888_50MMX);\
+			else						\
+				blitter(4, Uint16, ALPHA_BLIT32_888_50);\
+		    }							\
+		    else						\
+		    {							\
+			if((CPU_Flags()&MMX_CPU)!=0)			\
+				blitter(4, Uint16, ALPHA_BLIT32_888MMX);\
+			else						\
+				blitter(4, Uint16, ALPHA_BLIT32_888);	\
+		    }							\
+		} else							\
+		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
+		break;							\
+	    }								\
+	}								\
+    } while(0)
+
+#else
+	
+#define CHOOSE_BLIT(blitter, alpha, fmt)				\
+    do {								\
+        if(alpha == 255) {						\
+	    switch(fmt->BytesPerPixel) {				\
+	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
+	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
+	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
+	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
+	    }								\
+	} else {							\
+	    switch(fmt->BytesPerPixel) {				\
+	    case 1:							\
+		/* No 8bpp alpha blitting */				\
+		break;							\
+									\
+	    case 2:							\
+		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
+		case 0xffff:						\
+		    if(fmt->Gmask == 0x07e0				\
+		       || fmt->Rmask == 0x07e0				\
+		       || fmt->Bmask == 0x07e0) {			\
+			if(alpha == 128)				\
+			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
+			else {						\
 			    blitter(2, Uint8, ALPHA_BLIT16_565);	\
 			}						\
 		    } else						\
@@ -359,7 +756,6 @@
 			if(alpha == 128)				\
 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
 			else {						\
-			    alpha >>= 3; /* use 5 bit alpha */		\
 			    blitter(2, Uint8, ALPHA_BLIT16_555);	\
 			}						\
 			break;						\
@@ -391,6 +787,7 @@
 	}								\
     } while(0)
 
+#endif
 
 /*
  * This takes care of the case when the surface is clipped on the left and/or