summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--src/mesa/drivers/dri/common/spantmp2.h564
-rw-r--r--src/mesa/drivers/dri/r128/r128_span.c105
-rw-r--r--src/mesa/drivers/dri/r200/r200_span.c78
-rw-r--r--src/mesa/drivers/dri/unichrome/via_span.c53
-rw-r--r--src/mesa/sources1
-rw-r--r--src/mesa/x86/read_rgba_span_x86.S453
-rw-r--r--src/mesa/x86/read_rgba_span_x86.h53
7 files changed, 1114 insertions, 193 deletions
diff --git a/src/mesa/drivers/dri/common/spantmp2.h b/src/mesa/drivers/dri/common/spantmp2.h
new file mode 100644
index 00000000000..ce0a66de702
--- /dev/null
+++ b/src/mesa/drivers/dri/common/spantmp2.h
@@ -0,0 +1,564 @@
+/*
+ * Copyright 2000-2001 VA Linux Systems, Inc.
+ * (C) Copyright IBM Corporation 2004
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * VA LINUX SYSTEM, IBM AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file spantmp2.h
+ *
+ * Template file of span read / write functions.
+ *
+ * \author Keith Whitwell <[email protected]>
+ * \author Gareth Hughes <[email protected]>
+ * \author Ian Romanick <[email protected]>
+ */
+
+#include "colormac.h"
+
+#ifndef DBG
+#define DBG 0
+#endif
+
+#ifndef HW_WRITE_LOCK
+#define HW_WRITE_LOCK() HW_LOCK()
+#endif
+
+#ifndef HW_WRITE_UNLOCK
+#define HW_WRITE_UNLOCK() HW_UNLOCK()
+#endif
+
+#ifndef HW_READ_LOCK
+#define HW_READ_LOCK() HW_LOCK()
+#endif
+
+#ifndef HW_READ_UNLOCK
+#define HW_READ_UNLOCK() HW_UNLOCK()
+#endif
+
+#ifndef HW_READ_CLIPLOOP
+#define HW_READ_CLIPLOOP() HW_CLIPLOOP()
+#endif
+
+#ifndef HW_WRITE_CLIPLOOP
+#define HW_WRITE_CLIPLOOP() HW_CLIPLOOP()
+#endif
+
+#if (SPANTMP_PIXEL_FMT == GL_RGB) && (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)
+
+#define INIT_MONO_PIXEL(p, color) \
+ p = PACK_COLOR_565( color[0], color[1], color[2] )
+
+#define WRITE_RGBA( _x, _y, r, g, b, a ) \
+ do { \
+ GLshort * _p = (GLshort *) GET_DST_PTR(_x, _y); \
+ _p[0] = ((((int)r & 0xf8) << 8) | (((int)g & 0xfc) << 3) | \
+ (((int)b & 0xf8) >> 3)); \
+ } while(0)
+
+#define WRITE_PIXEL( _x, _y, p ) \
+ do { \
+ GLushort * _p = (GLushort *) GET_DST_PTR(_x, _y); \
+ _p[0] = p; \
+ } while(0)
+
+#define READ_RGBA( rgba, _x, _y ) \
+ do { \
+ GLushort p = *(volatile GLshort *) GET_SRC_PTR(_x, _y); \
+ rgba[0] = ((p >> 8) & 0xf8) * 255 / 0xf8; \
+ rgba[1] = ((p >> 3) & 0xfc) * 255 / 0xfc; \
+ rgba[2] = ((p << 3) & 0xf8) * 255 / 0xf8; \
+ rgba[3] = 0xff; \
+ } while (0)
+
+#elif (SPANTMP_PIXEL_FMT == GL_BGRA) && (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
+
+# define INIT_MONO_PIXEL(p, color) \
+ p = PACK_COLOR_8888(color[3], color[0], color[1], color[2])
+
+# define WRITE_RGBA(_x, _y, r, g, b, a) \
+ do { \
+ GLuint * _p = (GLuint *) GET_DST_PTR(_x, _y); \
+ _p[0] = ((r << 16) | (g << 8) | (b << 0) | (a << 24)); \
+ } while(0)
+
+#define WRITE_PIXEL(_x, _y, p) \
+ do { \
+ GLuint * _p = (GLuint *) GET_DST_PTR(_x, _y); \
+ _p[0] = p; \
+ } while(0)
+
+# if defined( USE_X86_ASM )
+# define READ_RGBA(rgba, _x, _y) \
+ do { \
+ GLuint p = *(volatile GLuint *) GET_SRC_PTR(_x, _y); \
+ __asm__ __volatile__( "bswap %0; rorl $8, %0" \
+ : "=r" (p) : "r" (p) ); \
+ ((GLuint *)rgba)[0] = p; \
+ } while (0)
+# else
+# define READ_RGBA( rgba, _x, _y ) \
+ do { \
+ GLuint p = *(volatile GLuint *) GET_SRC_PTR(_x, _y); \
+ rgba[0] = (p >> 16) & 0xff; \
+ rgba[1] = (p >> 8) & 0xff; \
+ rgba[2] = (p >> 0) & 0xff; \
+ rgba[3] = (p >> 24) & 0xff; \
+ } while (0)
+# endif
+
+#else
+#error SPANTMP_PIXEL_FMT must be set to a valid value!
+#endif
+
+#if defined( USE_MMX_ASM ) || defined( USE_SSE_ASM )
+#include "x86/read_rgba_span_x86.h"
+#include "x86/common_x86_asm.h"
+#endif
+
+static void TAG(WriteRGBASpan)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ const GLubyte rgba[][4],
+ const GLubyte mask[] )
+{
+ HW_WRITE_LOCK()
+ {
+ GLint x1;
+ GLint n1;
+ LOCAL_VARS;
+
+ y = Y_FLIP(y);
+
+ HW_WRITE_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+
+ if (DBG) fprintf(stderr, "WriteRGBASpan %d..%d (x1 %d)\n",
+ (int)i, (int)n1, (int)x1);
+
+ if (mask)
+ {
+ for (;n1>0;i++,x1++,n1--)
+ if (mask[i])
+ WRITE_RGBA( x1, y,
+ rgba[i][0], rgba[i][1],
+ rgba[i][2], rgba[i][3] );
+ }
+ else
+ {
+ for (;n1>0;i++,x1++,n1--)
+ WRITE_RGBA( x1, y,
+ rgba[i][0], rgba[i][1],
+ rgba[i][2], rgba[i][3] );
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_WRITE_UNLOCK();
+}
+
+static void TAG(WriteRGBSpan)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ const GLubyte rgb[][3],
+ const GLubyte mask[] )
+{
+ HW_WRITE_LOCK()
+ {
+ GLint x1;
+ GLint n1;
+ LOCAL_VARS;
+
+ y = Y_FLIP(y);
+
+ HW_WRITE_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+
+ if (DBG) fprintf(stderr, "WriteRGBSpan %d..%d (x1 %d)\n",
+ (int)i, (int)n1, (int)x1);
+
+ if (mask)
+ {
+ for (;n1>0;i++,x1++,n1--)
+ if (mask[i])
+ WRITE_RGBA( x1, y, rgb[i][0], rgb[i][1], rgb[i][2], 255 );
+ }
+ else
+ {
+ for (;n1>0;i++,x1++,n1--)
+ WRITE_RGBA( x1, y, rgb[i][0], rgb[i][1], rgb[i][2], 255 );
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_WRITE_UNLOCK();
+}
+
+static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
+ GLuint n,
+ const GLint x[],
+ const GLint y[],
+ const GLubyte rgba[][4],
+ const GLubyte mask[] )
+{
+ HW_WRITE_LOCK()
+ {
+ GLint i;
+ LOCAL_VARS;
+
+ if (DBG) fprintf(stderr, "WriteRGBAPixels\n");
+
+ HW_WRITE_CLIPLOOP()
+ {
+ if (mask)
+ {
+ for (i=0;i<n;i++)
+ {
+ if (mask[i]) {
+ const int fy = Y_FLIP(y[i]);
+ if (CLIPPIXEL(x[i],fy))
+ WRITE_RGBA( x[i], fy,
+ rgba[i][0], rgba[i][1],
+ rgba[i][2], rgba[i][3] );
+ }
+ }
+ }
+ else
+ {
+ for (i=0;i<n;i++)
+ {
+ const int fy = Y_FLIP(y[i]);
+ if (CLIPPIXEL(x[i],fy))
+ WRITE_RGBA( x[i], fy,
+ rgba[i][0], rgba[i][1],
+ rgba[i][2], rgba[i][3] );
+ }
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_WRITE_UNLOCK();
+}
+
+
+static void TAG(WriteMonoRGBASpan)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ const GLchan color[4],
+ const GLubyte mask[] )
+{
+ HW_WRITE_LOCK()
+ {
+ GLint x1;
+ GLint n1;
+ LOCAL_VARS;
+ INIT_MONO_PIXEL(p, color);
+
+ y = Y_FLIP( y );
+
+ if (DBG) fprintf(stderr, "WriteMonoRGBASpan\n");
+
+ HW_WRITE_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+ if (mask)
+ {
+ for (;n1>0;i++,x1++,n1--)
+ if (mask[i])
+ WRITE_PIXEL( x1, y, p );
+ }
+ else
+ {
+ for (;n1>0;i++,x1++,n1--)
+ WRITE_PIXEL( x1, y, p );
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_WRITE_UNLOCK();
+}
+
+
+static void TAG(WriteMonoRGBAPixels)( const GLcontext *ctx,
+ GLuint n,
+ const GLint x[], const GLint y[],
+ const GLchan color[],
+ const GLubyte mask[] )
+{
+ HW_WRITE_LOCK()
+ {
+ GLint i;
+ LOCAL_VARS;
+ INIT_MONO_PIXEL(p, color);
+
+ if (DBG) fprintf(stderr, "WriteMonoRGBAPixels\n");
+
+ HW_WRITE_CLIPLOOP()
+ {
+ if (mask)
+ {
+ for (i=0;i<n;i++)
+ if (mask[i]) {
+ int fy = Y_FLIP(y[i]);
+ if (CLIPPIXEL( x[i], fy ))
+ WRITE_PIXEL( x[i], fy, p );
+ }
+ }
+ else
+ {
+ for (i=0;i<n;i++) {
+ int fy = Y_FLIP(y[i]);
+ if (CLIPPIXEL( x[i], fy ))
+ WRITE_PIXEL( x[i], fy, p );
+ }
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_WRITE_UNLOCK();
+}
+
+
+static void TAG(ReadRGBASpan)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ GLubyte rgba[][4])
+{
+ HW_READ_LOCK()
+ {
+ GLint x1,n1;
+ LOCAL_VARS;
+
+ y = Y_FLIP(y);
+
+ if (DBG) fprintf(stderr, "ReadRGBASpan\n");
+
+ HW_READ_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+ for (;n1>0;i++,x1++,n1--)
+ READ_RGBA( rgba[i], x1, y );
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_READ_UNLOCK();
+}
+
+
+#if defined(USE_MMX_ASM)
+static void TAG2(ReadRGBASpan,_MMX)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ GLubyte rgba[][4])
+{
+#ifndef USE_INNER_EMMS
+ /* The EMMS instruction is directly in-lined here because using GCC's
+ * built-in _mm_empty function was found to utterly destroy performance.
+ */
+ __asm__ __volatile__( "emms" );
+#endif
+
+ HW_LOCK()
+ {
+ GLint x1,n1;
+ LOCAL_VARS;
+
+ y = Y_FLIP(y);
+
+ if (DBG) fprintf(stderr, "ReadRGBASpan\n");
+
+ HW_READ_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+
+ {
+ const char * src = (read_buf + x1*4 + y*pitch);
+ _generic_read_RGBA_span_BGRA8888_REV_MMX( src, rgba[i], n1 );
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_UNLOCK();
+#ifndef USE_INNER_EMMS
+ __asm__ __volatile__( "emms" );
+#endif
+}
+#endif
+
+
+#if defined(USE_SSE_ASM)
+static void TAG2(ReadRGBASpan,_SSE2)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ GLubyte rgba[][4])
+{
+ HW_LOCK()
+ {
+ GLint x1,n1;
+ LOCAL_VARS;
+
+ y = Y_FLIP(y);
+
+ if (DBG) fprintf(stderr, "ReadRGBASpan\n");
+
+ HW_READ_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+
+ {
+ const char * src = (read_buf + x1*4 + y*pitch);
+ _generic_read_RGBA_span_BGRA8888_REV_SSE2( src, rgba[i], n1 );
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_UNLOCK();
+}
+#endif
+
+#if defined(USE_SSE_ASM)
+static void TAG2(ReadRGBASpan,_SSE)( const GLcontext *ctx,
+ GLuint n, GLint x, GLint y,
+ GLubyte rgba[][4])
+{
+#ifndef USE_INNER_EMMS
+ /* The EMMS instruction is directly in-lined here because using GCC's
+ * built-in _mm_empty function was found to utterly destroy performance.
+ */
+ __asm__ __volatile__( "emms" );
+#endif
+
+ HW_LOCK()
+ {
+ GLint x1,n1;
+ LOCAL_VARS;
+
+ y = Y_FLIP(y);
+
+ if (DBG) fprintf(stderr, "ReadRGBASpan\n");
+
+ HW_READ_CLIPLOOP()
+ {
+ GLint i = 0;
+ CLIPSPAN(x,y,n,x1,n1,i);
+
+ {
+ const char * src = (read_buf + x1*4 + y*pitch);
+ _generic_read_RGBA_span_BGRA8888_REV_SSE( src, rgba[i], n1 );
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_UNLOCK();
+#ifndef USE_INNER_EMMS
+ __asm__ __volatile__( "emms" );
+#endif
+}
+#endif
+
+
+static void TAG(ReadRGBAPixels)( const GLcontext *ctx,
+ GLuint n, const GLint x[], const GLint y[],
+ GLubyte rgba[][4], const GLubyte mask[] )
+{
+ HW_READ_LOCK()
+ {
+ GLint i;
+ LOCAL_VARS;
+
+ if (DBG) fprintf(stderr, "ReadRGBAPixels\n");
+
+ HW_READ_CLIPLOOP()
+ {
+ if (mask)
+ {
+ for (i=0;i<n;i++)
+ if (mask[i]) {
+ int fy = Y_FLIP( y[i] );
+ if (CLIPPIXEL( x[i], fy ))
+ READ_RGBA( rgba[i], x[i], fy );
+ }
+ }
+ else
+ {
+ for (i=0;i<n;i++) {
+ int fy = Y_FLIP( y[i] );
+ if (CLIPPIXEL( x[i], fy ))
+ READ_RGBA( rgba[i], x[i], fy );
+ }
+ }
+ }
+ HW_ENDCLIPLOOP();
+ }
+ HW_READ_UNLOCK();
+}
+
+static void TAG(InitPointers)(struct swrast_device_driver *swdd)
+{
+ swdd->WriteRGBASpan = TAG(WriteRGBASpan);
+ swdd->WriteRGBSpan = TAG(WriteRGBSpan);
+ swdd->WriteMonoRGBASpan = TAG(WriteMonoRGBASpan);
+ swdd->WriteRGBAPixels = TAG(WriteRGBAPixels);
+ swdd->WriteMonoRGBAPixels = TAG(WriteMonoRGBAPixels);
+ swdd->ReadRGBAPixels = TAG(ReadRGBAPixels);
+
+#if defined(USE_SSE_ASM)
+ if ( cpu_has_xmm2 ) {
+ if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "SSE2" );
+ swdd->ReadRGBASpan = TAG2(ReadRGBASpan, _SSE2);
+ }
+ else
+#endif
+#if defined(USE_SSE_ASM)
+ if ( cpu_has_xmm ) {
+ if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "SSE" );
+ swdd->ReadRGBASpan = TAG2(ReadRGBASpan, _SSE);
+ }
+ else
+#endif
+#if defined(USE_MMX_ASM)
+ if ( cpu_has_mmx ) {
+ if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "MMX" );
+ swdd->ReadRGBASpan = TAG2(ReadRGBASpan, _MMX);
+ }
+ else
+#endif
+ {
+ if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "C" );
+ swdd->ReadRGBASpan = TAG(ReadRGBASpan);
+ }
+
+}
+
+
+#undef INIT_MONO_PIXEL
+#undef WRITE_PIXEL
+#undef WRITE_RGBA
+#undef READ_RGBA
+#undef TAG
+#undef TAG2
+#undef GET_SRC_PTR
+#undef GET_DST_PTR
+#undef SPANTMP_PIXEL_FMT
+#undef SPANTMP_PIXEL_TYPE
diff --git a/src/mesa/drivers/dri/r128/r128_span.c b/src/mesa/drivers/dri/r128/r128_span.c
index db2ec44fbf2..b169dc7509e 100644
--- a/src/mesa/drivers/dri/r128/r128_span.c
+++ b/src/mesa/drivers/dri/r128/r128_span.c
@@ -125,85 +125,27 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
/* 16 bit, RGB565 color spanline and pixel functions
*/
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p, color) \
- p = R128PACKCOLOR565( color[0], color[1], color[2] )
+#define GET_SRC_PTR(_x, _y) (read_buf + _x * 2 + _y * pitch)
+#define GET_DST_PTR(_x, _y) ( buf + _x * 2 + _y * pitch)
+#define SPANTMP_PIXEL_FMT GL_RGB
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-#define WRITE_RGBA( _x, _y, r, g, b, a ) \
- *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) | \
- (((int)g & 0xfc) << 3) | \
- (((int)b & 0xf8) >> 3))
+#define TAG(x) r128##x##_RGB565
+#define TAG2(x,y) r128##x##_RGB565##y
+#include "spantmp2.h"
-#define WRITE_PIXEL( _x, _y, p ) \
- *(GLushort *)(buf + _x*2 + _y*pitch) = p
-
-#define READ_RGBA( rgba, _x, _y ) \
- do { \
- GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch); \
- rgba[0] = (p >> 8) & 0xf8; \
- rgba[1] = (p >> 3) & 0xfc; \
- rgba[2] = (p << 3) & 0xf8; \
- rgba[3] = 0xff; \
- if ( rgba[0] & 0x08 ) rgba[0] |= 0x07; \
- if ( rgba[1] & 0x04 ) rgba[1] |= 0x03; \
- if ( rgba[2] & 0x08 ) rgba[2] |= 0x07; \
- } while (0)
-
-#define TAG(x) r128##x##_RGB565
-#include "spantmp.h"
-
-#define READ_DEPTH(d, _x, _y) \
- d = *(GLushort *)(buf + _x*2 + _y*pitch)
/* 32 bit, ARGB8888 color spanline and pixel functions
*/
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p, color) \
- p = R128PACKCOLOR8888( color[0], color[1], color[2], color[3] )
-
-#define WRITE_RGBA( _x, _y, r, g, b, a ) \
- *(GLuint *)(buf + _x*4 + _y*pitch) = ((b << 0) | \
- (g << 8) | \
- (r << 16) | \
- (a << 24) )
+#define GET_SRC_PTR(_x, _y) (read_buf + _x * 4 + _y * pitch)
+#define GET_DST_PTR(_x, _y) ( buf + _x * 4 + _y * pitch)
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-#define WRITE_PIXEL( _x, _y, p ) \
- *(GLuint *)(buf + _x*4 + _y*pitch) = p
+#define TAG(x) r128##x##_ARGB8888
+#define TAG2(x,y) r128##x##_ARGB8888##y
+#include "spantmp2.h"
-#define READ_RGBA( rgba, _x, _y ) \
-do { \
- GLuint p = *(GLuint *)(read_buf + _x*4 + _y*pitch); \
- rgba[0] = (p >> 16) & 0xff; \
- rgba[1] = (p >> 8) & 0xff; \
- rgba[2] = (p >> 0) & 0xff; \
- rgba[3] = 0xff;/*(p >> 24) & 0xff;*/ \
-} while (0)
-
-#define TAG(x) r128##x##_ARGB8888
-#include "spantmp.h"
-
-
-/* 24 bit, RGB888 color spanline and pixel functions */
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p, color) \
- p = R128PACKCOLOR888( color[0], color[1], color[2] )
-
-#define WRITE_RGBA(_x, _y, r, g, b, a) \
- *(GLuint *)(buf + _x*3 + _y*pitch) = ((r << 16) | \
- (g << 8) | \
- (b << 0))
-
-#define WRITE_PIXEL(_x, _y, p) \
- *(GLuint *)(buf + _x*3 + _y*pitch) = p
-
-#define READ_RGBA(rgba, _x, _y) \
- do { \
- GLuint p = *(GLuint *)(read_buf + _x*3 + _y*pitch); \
- rgba[0] = (p >> 16) & 0xff; \
- rgba[1] = (p >> 8) & 0xff; \
- rgba[2] = (p >> 0) & 0xff; \
- rgba[3] = 0xff; \
- } while (0)
/* ================================================================
* Depth buffer
@@ -211,6 +153,9 @@ do { \
/* 16-bit depth buffer functions
*/
+#define READ_DEPTH(d, _x, _y) \
+ d = *(GLushort *)(buf + _x*2 + _y*pitch)
+
#define WRITE_DEPTH_SPAN() \
r128WriteDepthSpanLocked( rmesa, n, \
x + dPriv->x, \
@@ -423,23 +368,11 @@ void r128DDInitSpanFuncs( GLcontext *ctx )
switch ( rmesa->r128Screen->cpp ) {
case 2:
- swdd->WriteRGBASpan = r128WriteRGBASpan_RGB565;
- swdd->WriteRGBSpan = r128WriteRGBSpan_RGB565;
- swdd->WriteMonoRGBASpan = r128WriteMonoRGBASpan_RGB565;
- swdd->WriteRGBAPixels = r128WriteRGBAPixels_RGB565;
- swdd->WriteMonoRGBAPixels = r128WriteMonoRGBAPixels_RGB565;
- swdd->ReadRGBASpan = r128ReadRGBASpan_RGB565;
- swdd->ReadRGBAPixels = r128ReadRGBAPixels_RGB565;
+ r128InitPointers_RGB565( swdd );
break;
case 4:
- swdd->WriteRGBASpan = r128WriteRGBASpan_ARGB8888;
- swdd->WriteRGBSpan = r128WriteRGBSpan_ARGB8888;
- swdd->WriteMonoRGBASpan = r128WriteMonoRGBASpan_ARGB8888;
- swdd->WriteRGBAPixels = r128WriteRGBAPixels_ARGB8888;
- swdd->WriteMonoRGBAPixels = r128WriteMonoRGBAPixels_ARGB8888;
- swdd->ReadRGBASpan = r128ReadRGBASpan_ARGB8888;
- swdd->ReadRGBAPixels = r128ReadRGBAPixels_ARGB8888;
+ r128InitPointers_ARGB8888( swdd );
break;
default:
diff --git a/src/mesa/drivers/dri/r200/r200_span.c b/src/mesa/drivers/dri/r200/r200_span.c
index e1ad976ce3b..efb5e6ea7c5 100644
--- a/src/mesa/drivers/dri/r200/r200_span.c
+++ b/src/mesa/drivers/dri/r200/r200_span.c
@@ -122,61 +122,27 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
/* 16 bit, RGB565 color spanline and pixel functions
*/
-#define INIT_MONO_PIXEL(p, color) \
- p = PACK_COLOR_565( color[0], color[1], color[2] )
-#define WRITE_RGBA( _x, _y, r, g, b, a ) \
- *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) | \
- (((int)g & 0xfc) << 3) | \
- (((int)b & 0xf8) >> 3))
+#define GET_SRC_PTR(_x, _y) (read_buf + _x * 2 + _y * pitch)
+#define GET_DST_PTR(_x, _y) ( buf + _x * 2 + _y * pitch)
+#define SPANTMP_PIXEL_FMT GL_RGB
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-#define WRITE_PIXEL( _x, _y, p ) \
- *(GLushort *)(buf + _x*2 + _y*pitch) = p
-
-#define READ_RGBA( rgba, _x, _y ) \
- do { \
- GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch); \
- rgba[0] = ((p >> 8) & 0xf8) * 255 / 0xf8; \
- rgba[1] = ((p >> 3) & 0xfc) * 255 / 0xfc; \
- rgba[2] = ((p << 3) & 0xf8) * 255 / 0xf8; \
- rgba[3] = 0xff; \
- } while (0)
-
-#define TAG(x) r200##x##_RGB565
-#include "spantmp.h"
+#define TAG(x) r200##x##_RGB565
+#define TAG2(x,y) r200##x##_RGB565##y
+#include "spantmp2.h"
/* 32 bit, ARGB8888 color spanline and pixel functions
*/
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p, color) \
- p = PACK_COLOR_8888( color[3], color[0], color[1], color[2] )
-
-#define WRITE_RGBA( _x, _y, r, g, b, a ) \
-do { \
- *(GLuint *)(buf + _x*4 + _y*pitch) = ((b << 0) | \
- (g << 8) | \
- (r << 16) | \
- (a << 24) ); \
-} while (0)
-
-#define WRITE_PIXEL( _x, _y, p ) \
-do { \
- *(GLuint *)(buf + _x*4 + _y*pitch) = p; \
-} while (0)
-
-#define READ_RGBA( rgba, _x, _y ) \
-do { \
- volatile GLuint *ptr = (volatile GLuint *)(read_buf + _x*4 + _y*pitch); \
- GLuint p = *ptr; \
- rgba[0] = (p >> 16) & 0xff; \
- rgba[1] = (p >> 8) & 0xff; \
- rgba[2] = (p >> 0) & 0xff; \
- rgba[3] = (p >> 24) & 0xff; \
-} while (0)
-#define TAG(x) r200##x##_ARGB8888
-#include "spantmp.h"
+#define GET_SRC_PTR(_x, _y) (read_buf + _x * 4 + _y * pitch)
+#define GET_DST_PTR(_x, _y) ( buf + _x * 4 + _y * pitch)
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
+#define TAG(x) r200##x##_ARGB8888
+#define TAG2(x,y) r200##x##_ARGB8888##y
+#include "spantmp2.h"
/* ================================================================
@@ -380,23 +346,11 @@ void r200InitSpanFuncs( GLcontext *ctx )
switch ( rmesa->r200Screen->cpp ) {
case 2:
- swdd->WriteRGBASpan = r200WriteRGBASpan_RGB565;
- swdd->WriteRGBSpan = r200WriteRGBSpan_RGB565;
- swdd->WriteMonoRGBASpan = r200WriteMonoRGBASpan_RGB565;
- swdd->WriteRGBAPixels = r200WriteRGBAPixels_RGB565;
- swdd->WriteMonoRGBAPixels = r200WriteMonoRGBAPixels_RGB565;
- swdd->ReadRGBASpan = r200ReadRGBASpan_RGB565;
- swdd->ReadRGBAPixels = r200ReadRGBAPixels_RGB565;
+ r200InitPointers_RGB565( swdd );
break;
case 4:
- swdd->WriteRGBASpan = r200WriteRGBASpan_ARGB8888;
- swdd->WriteRGBSpan = r200WriteRGBSpan_ARGB8888;
- swdd->WriteMonoRGBASpan = r200WriteMonoRGBASpan_ARGB8888;
- swdd->WriteRGBAPixels = r200WriteRGBAPixels_ARGB8888;
- swdd->WriteMonoRGBAPixels = r200WriteMonoRGBAPixels_ARGB8888;
- swdd->ReadRGBASpan = r200ReadRGBASpan_ARGB8888;
- swdd->ReadRGBAPixels = r200ReadRGBAPixels_ARGB8888;
+ r200InitPointers_ARGB8888( swdd );
break;
default:
diff --git a/src/mesa/drivers/dri/unichrome/via_span.c b/src/mesa/drivers/dri/unichrome/via_span.c
index 693b6de1428..3a747a3d999 100644
--- a/src/mesa/drivers/dri/unichrome/via_span.c
+++ b/src/mesa/drivers/dri/unichrome/via_span.c
@@ -204,19 +204,6 @@
#undef LOCAL_VARS
#undef LOCAL_DEPTH_VARS
-/*=* [DBG] csmash : fix options worng position *=*/
-/*#define LOCAL_VARS \
- __DRIdrawablePrivate *dPriv = vmesa->driDrawable; \
- GLuint pitch = vmesa->drawPitch; \
- GLuint height = dPriv->h; \
- GLuint p; \
- char *buf = (char *)(vmesa->drawMap + \
- dPriv->x * 4 + \
- dPriv->y * pitch); \
- char *read_buf = (char *)(vmesa->readMap + \
- dPriv->x * 4 + \
- dPriv->y * pitch); \
- (void)read_buf; (void)buf; (void)p*/
#define LOCAL_VARS \
__DRIdrawablePrivate *dPriv = vmesa->driDrawable; \
GLuint pitch = vmesa->drawPitch; \
@@ -237,33 +224,15 @@
dPriv->y * pitch); \
}
+#define GET_SRC_PTR(_x, _y) (read_buf + _x * 4 + _y * pitch)
+#define GET_DST_PTR(_x, _y) ( buf + _x * 4 + _y * pitch)
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p, color) \
- p = PACK_COLOR_8888(color[3], color[0], color[1], color[2])
-
-#define WRITE_RGBA(_x, _y, r, g, b, a) \
- *(GLuint *)(buf + _x * 4 + _y * pitch) = ((r << 16) | \
- (g << 8) | \
- (b << 0) | \
- (a << 24));
-
-
-#define WRITE_PIXEL(_x, _y, p) \
- *(GLuint *)(buf + _x * 4 + _y * pitch) = p
-
-#define READ_RGBA(rgba, _x, _y) \
- do { \
- GLuint p = *(GLuint *)(read_buf + _x * 4 + _y * pitch); \
- rgba[0] = (p >> 16) & 0xff; \
- rgba[1] = (p >> 8) & 0xff; \
- rgba[2] = (p >> 0) & 0xff; \
- rgba[3] = 255; \
- } while (0)
+#define TAG(x) via##x##_8888
+#define TAG2(x,y) via##x##_8888##y
+#include "spantmp2.h"
-#define TAG(x) via##x##_8888
-#include "spantmp.h"
-/*#include "via_spantmp.h"*/
/* 16 bit depthbuffer functions.
*/
@@ -367,13 +336,7 @@ void viaInitSpanFuncs(GLcontext *ctx)
swdd->ReadRGBAPixels = viaReadRGBAPixels_565;
}
else if (vmesa->viaScreen->bitsPerPixel == 0x20) {
- swdd->WriteRGBASpan = viaWriteRGBASpan_8888;
- swdd->WriteRGBSpan = viaWriteRGBSpan_8888;
- swdd->WriteMonoRGBASpan = viaWriteMonoRGBASpan_8888;
- swdd->WriteRGBAPixels = viaWriteRGBAPixels_8888;
- swdd->WriteMonoRGBAPixels = viaWriteMonoRGBAPixels_8888;
- swdd->ReadRGBASpan = viaReadRGBASpan_8888;
- swdd->ReadRGBAPixels = viaReadRGBAPixels_8888;
+ viaInitPointers_8888( swdd );
}
else
ASSERT(0);
diff --git a/src/mesa/sources b/src/mesa/sources
index 7013182f6f2..f82d344bf58 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -170,6 +170,7 @@ X86_SOURCES = \
x86/sse_xform3.S \
x86/sse_xform4.S \
x86/sse_normal.S \
+ x86/read_rgba_span_x86.S \
tnl/t_vtx_x86_gcc.S
X86_API = \
diff --git a/src/mesa/x86/read_rgba_span_x86.S b/src/mesa/x86/read_rgba_span_x86.S
new file mode 100644
index 00000000000..e637f22da37
--- /dev/null
+++ b/src/mesa/x86/read_rgba_span_x86.S
@@ -0,0 +1,453 @@
+/*
+ * (C) Copyright IBM Corporation 2004
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * IBM AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file read_rgba_span_x86.S
+ * Optimized routines to transfer pixel data from the framebuffer to a
+ * buffer in main memory.
+ *
+ * \author Ian Romanick <[email protected]>
+ */
+
+ .file "read_rgba_span_x86.S"
+ .section .rodata
+ .align 16
+ .type mask, @object
+ .size mask, 32
+mask:
+ .long 0xff00ff00
+ .long 0xff00ff00
+ .long 0xff00ff00
+ .long 0xff00ff00
+ .long 0x00ff0000
+ .long 0x00ff0000
+ .long 0x00ff0000
+ .long 0x00ff0000
+
+
+/* I implemented these as macros because the appear in quite a few places,
+ * and I've tweaked them a number of times. I got tired of changing every
+ * place they appear. :)
+ */
+
+#define DO_ONE_PIXEL() \
+ movl (%ebx), %eax ; \
+ addl $4, %ebx ; \
+ bswap %eax /* ARGB -> BGRA */ ; \
+ rorl $8, %eax /* BGRA -> ABGR */ ; \
+ movl %eax, (%ecx) /* ABGR -> R, G, B, A */ ; \
+ addl $4, %ecx
+
+#define DO_ONE_LAST_PIXEL() \
+ movl (%ebx), %eax ; \
+ bswap %eax /* ARGB -> BGRA */ ; \
+ rorl $8, %eax /* BGRA -> ABGR */ ; \
+ movl %eax, (%ecx) /* ABGR -> R, G, B, A */ ; \
+
+
+/**
+ * MMX optimized version of the BGRA8888_REV to RGBA copy routine.
+ *
+ * \warning
+ * This function assumes that the caller will issue the EMMS instruction
+ * at the correct places.
+ */
+
+.globl _generic_read_RGBA_span_BGRA8888_REV_MMX
+ .type _generic_read_RGBA_span_BGRA8888_REV_MMX, @function
+_generic_read_RGBA_span_BGRA8888_REV_MMX:
+ pushl %ebx
+
+#ifdef USE_INNER_EMMS
+ emms
+#endif
+ movq mask, %mm1
+ movq mask+16, %mm2
+
+ movl 8(%esp), %ebx /* source pointer */
+ movl 16(%esp), %edx /* number of pixels to copy */
+ movl 12(%esp), %ecx /* destination pointer */
+
+ testl %edx, %edx
+ je .L20 /* Bail if there's nothing to do. */
+
+ movl %ebx, %eax
+
+ negl %eax
+ sarl $2, %eax
+ andl $1, %eax
+ je .L17
+
+ subl %eax, %edx
+ DO_ONE_PIXEL()
+.L17:
+
+ /* Would it be faster to unroll this loop once and process 4 pixels
+ * per pass, instead of just two?
+ */
+
+ movl %edx, %eax
+ shrl %eax
+ jmp .L18
+.L19:
+ movq (%ebx), %mm0
+ addl $8, %ebx
+
+ /* These 9 instructions do what PSHUFB (if there were such an
+ * instruction) could do in 1. :(
+ */
+
+ movq %mm0, %mm3
+ movq %mm0, %mm4
+
+ pand %mm2, %mm3
+ psllq $16, %mm4
+ psrlq $16, %mm3
+ pand %mm2, %mm4
+
+ pand %mm1, %mm0
+ por %mm4, %mm3
+ por %mm3, %mm0
+
+ movq %mm0, (%ecx)
+ addl $8, %ecx
+ subl $1, %eax
+.L18:
+ jne .L19
+
+#ifdef USE_INNER_EMMS
+ emms
+#endif
+
+ /* At this point there are either 1 or 0 pixels remaining to be
+ * converted. Convert the last pixel, if needed.
+ */
+
+ testl $1, %edx
+ je .L20
+
+ DO_ONE_LAST_PIXEL()
+
+.L20:
+ popl %ebx
+ ret
+ .size _generic_read_RGBA_span_BGRA8888_REV_MMX, .-_generic_read_RGBA_span_BGRA8888_REV_MMX
+
+
+/**
+ * SSE optimized version of the BGRA8888_REV to RGBA copy routine. SSE
+ * instructions are only actually used to read data from the framebuffer.
+ * In practice, the speed-up is pretty small.
+ *
+ * \todo
+ * Do some more testing and determine if there's any reason to have this
+ * function in addition to the MMX version.
+ *
+ * \warning
+ * This function assumes that the caller will issue the EMMS instruction
+ * at the correct places.
+ */
+
+.globl _generic_read_RGBA_span_BGRA8888_REV_SSE
+ .type _generic_read_RGBA_span_BGRA8888_REV_SSE, @function
+_generic_read_RGBA_span_BGRA8888_REV_SSE:
+ pushl %esi
+ pushl %ebx
+ pushl %ebp
+
+#ifdef USE_INNER_EMMS
+ emms
+#endif
+ movq mask, %mm1
+ movq mask+16, %mm2
+
+ movl 16(%esp), %ebx /* source pointer */
+ movl 24(%esp), %edx /* number of pixels to copy */
+ movl 20(%esp), %ecx /* destination pointer */
+
+ movl %esp, %ebp
+ subl $16, %esp
+ andl $0xfffffff0, %esp
+
+ movl %ebx, %eax
+ movl %edx, %esi
+
+ negl %eax
+ andl $15, %eax
+ sarl $2, %eax
+ cmpl %edx, %eax
+ cmovle %eax, %esi
+
+ subl %esi, %edx
+
+ testl $1, %esi
+ je .L32
+
+ DO_ONE_PIXEL()
+.L32:
+
+ testl $2, %esi
+ je .L31
+
+ movq (%ebx), %mm0
+ addl $8, %ebx
+
+ movq %mm0, %mm3
+ movq %mm0, %mm4
+
+ pand %mm2, %mm3
+ psllq $16, %mm4
+ psrlq $16, %mm3
+ pand %mm2, %mm4
+
+ pand %mm1, %mm0
+ por %mm4, %mm3
+ por %mm3, %mm0
+
+ movq %mm0, (%ecx)
+ addl $8, %ecx
+.L31:
+
+ movl %edx, %eax
+ shrl $2, %eax
+ jmp .L33
+.L34:
+ movaps (%ebx), %xmm0
+ addl $16, %ebx
+
+ /* This would be so much better if we could just move directly from
+ * an SSE register to an MMX register. Unfortunately, that
+ * functionality wasn't introduced until SSE2 with the MOVDQ2Q
+ * instruction.
+ */
+
+ movaps %xmm0, (%esp)
+ movq (%esp), %mm0
+ movq 8(%esp), %mm5
+
+ movq %mm0, %mm3
+ movq %mm0, %mm4
+ movq %mm5, %mm6
+ movq %mm5, %mm7
+
+ pand %mm2, %mm3
+ pand %mm2, %mm6
+
+ psllq $16, %mm4
+ psllq $16, %mm7
+
+ psrlq $16, %mm3
+ psrlq $16, %mm6
+
+ pand %mm2, %mm4
+ pand %mm2, %mm7
+
+ pand %mm1, %mm0
+ pand %mm1, %mm5
+
+ por %mm4, %mm3
+ por %mm7, %mm6
+
+ por %mm3, %mm0
+ por %mm6, %mm5
+
+ movq %mm0, (%ecx)
+ movq %mm5, 8(%ecx)
+ addl $16, %ecx
+
+ subl $1, %eax
+.L33:
+ jne .L34
+
+#ifdef USE_INNER_EMMS
+ emms
+#endif
+ movl %ebp, %esp
+
+ /* At this point there are either [0, 3] pixels remaining to be
+ * converted.
+ */
+
+ testl $2, %edx
+ je .L36
+
+ movq (%ebx), %mm0
+ addl $8, %ebx
+
+ movq %mm0, %mm3
+ movq %mm0, %mm4
+
+ pand %mm2, %mm3
+ psllq $16, %mm4
+ psrlq $16, %mm3
+ pand %mm2, %mm4
+
+ pand %mm1, %mm0
+ por %mm4, %mm3
+ por %mm3, %mm0
+
+ movq %mm0, (%ecx)
+ addl $8, %ecx
+.L36:
+
+ testl $1, %edx
+ je .L35
+
+ DO_ONE_LAST_PIXEL()
+.L35:
+ popl %ebp
+ popl %ebx
+ popl %esi
+ ret
+ .size _generic_read_RGBA_span_BGRA8888_REV_SSE, .-_generic_read_RGBA_span_BGRA8888_REV_SSE
+
+
+/**
+ * SSE2 optimized version of the BGRA8888_REV to RGBA copy routine.
+ */
+
+ .text
+.globl _generic_read_RGBA_span_BGRA8888_REV_SSE2
+ .type _generic_read_RGBA_span_BGRA8888_REV_SSE2, @function
+_generic_read_RGBA_span_BGRA8888_REV_SSE2:
+ pushl %esi
+ pushl %ebx
+
+ movdqa mask, %xmm1
+ movdqa mask+16, %xmm2
+
+ movl 12(%esp), %ebx /* source pointer */
+ movl 20(%esp), %edx /* number of pixels to copy */
+ movl 16(%esp), %ecx /* destination pointer */
+
+ movl %ebx, %eax
+ movl %edx, %esi
+
+ /* If the source pointer isn't a multiple of 16 we have to process
+ * a few pixels the "slow" way to get the address aligned for
+ * the SSE fetch intsructions.
+ */
+
+ negl %eax
+ andl $15, %eax
+ sarl $2, %eax
+
+ cmpl %edx, %eax
+ cmovbe %eax, %esi
+ subl %esi, %edx
+
+ testl $1, %esi
+ je .L41
+
+ DO_ONE_PIXEL()
+.L41:
+ testl $2, %esi
+ je .L40
+
+ movq (%ebx), %xmm0
+ addl $8, %ebx
+
+ movdqa %xmm0, %xmm3
+ movdqa %xmm0, %xmm4
+ andps %xmm1, %xmm0
+
+ andps %xmm2, %xmm3
+ pslldq $2, %xmm4
+ psrldq $2, %xmm3
+ andps %xmm2, %xmm4
+
+ orps %xmm4, %xmm3
+ orps %xmm3, %xmm0
+
+ movq %xmm0, (%ecx)
+ addl $8, %ecx
+.L40:
+
+ /* Would it be worth having a specialized version of this loop for
+ * the case where the destination is 16-byte aligned? That version
+ * would be identical except that it could use movedqa instead of
+ * movdqu.
+ */
+
+ movl %edx, %eax
+ shrl $2, %eax
+ jmp .L42
+.L43:
+ movdqa (%ebx), %xmm0
+ addl $16, %ebx
+
+ movdqa %xmm0, %xmm3
+ movdqa %xmm0, %xmm4
+ andps %xmm1, %xmm0
+
+ andps %xmm2, %xmm3
+ pslldq $2, %xmm4
+ psrldq $2, %xmm3
+ andps %xmm2, %xmm4
+
+ orps %xmm4, %xmm3
+ orps %xmm3, %xmm0
+
+ movdqu %xmm0, (%ecx)
+ addl $16, %ecx
+ subl $1, %eax
+.L42:
+ jne .L43
+
+
+ /* There may be upto 3 pixels remaining to be copied. Take care
+ * of them now. We do the 2 pixel case first because the data
+ * will be aligned.
+ */
+
+ testl $2, %edx
+ je .L47
+
+ movq (%ebx), %xmm0
+
+ movdqa %xmm0, %xmm3
+ movdqa %xmm0, %xmm4
+ andps %xmm1, %xmm0
+
+ andps %xmm2, %xmm3
+ pslldq $2, %xmm4
+ psrldq $2, %xmm3
+ andps %xmm2, %xmm4
+
+ orps %xmm4, %xmm3
+ orps %xmm3, %xmm0
+
+ movq %xmm0, (%ecx)
+.L47:
+
+ testl $1, %edx
+ je .L46
+
+ DO_ONE_LAST_PIXEL()
+.L46:
+
+ popl %ebx
+ popl %esi
+ ret
+ .size _generic_read_RGBA_span_BGRA8888_REV_SSE2, .-_generic_read_RGBA_span_BGRA8888_REV_SSE2
diff --git a/src/mesa/x86/read_rgba_span_x86.h b/src/mesa/x86/read_rgba_span_x86.h
new file mode 100644
index 00000000000..99dd0e365d0
--- /dev/null
+++ b/src/mesa/x86/read_rgba_span_x86.h
@@ -0,0 +1,53 @@
+/*
+ * (C) Copyright IBM Corporation 2004
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * IBM AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file read_rgba_span_x86.h
+ *
+ * \author Ian Romanick <[email protected]>
+ */
+
+#ifndef READ_RGBA_SPAN_X86_H
+#define READ_RGBA_SPAN_X86_H
+
+#if defined(USE_SSE_ASM) || defined(USE_MMX_ASM)
+#include "x86/common_x86_asm.h"
+#endif
+
+#if defined(USE_SSE_ASM)
+extern void _generic_read_RGBA_span_BGRA8888_REV_SSE2( const unsigned char *,
+ unsigned char *, unsigned );
+#endif
+
+#if defined(USE_SSE_ASM)
+extern void _generic_read_RGBA_span_BGRA8888_REV_SSE( const unsigned char *,
+ unsigned char *, unsigned );
+#endif
+
+#if defined(USE_MMX_ASM)
+extern void _generic_read_RGBA_span_BGRA8888_REV_MMX( const unsigned char *,
+ unsigned char *, unsigned );
+#endif
+
+#endif /* READ_RGBA_SPAN_X86_H */