sl@0: /*
sl@0:  * Copyright © 2004 Red Hat, Inc.
sl@0:  * Copyright © 2004 Nicholas Miell
sl@0:  * Copyright © 2005 Trolltech AS
sl@0:  *
sl@0:  * Permission to use, copy, modify, distribute, and sell this software and its
sl@0:  * documentation for any purpose is hereby granted without fee, provided that
sl@0:  * the above copyright notice appear in all copies and that both that
sl@0:  * copyright notice and this permission notice appear in supporting
sl@0:  * documentation, and that the name of Red Hat not be used in advertising or
sl@0:  * publicity pertaining to distribution of the software without specific,
sl@0:  * written prior permission.  Red Hat makes no representations about the
sl@0:  * suitability of this software for any purpose.  It is provided "as is"
sl@0:  * without express or implied warranty.
sl@0:  *
sl@0:  * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
sl@0:  * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
sl@0:  * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
sl@0:  * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
sl@0:  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
sl@0:  * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
sl@0:  * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
sl@0:  * SOFTWARE.
sl@0:  *
sl@0:  * Author:  Søren Sandmann (sandmann@redhat.com)
sl@0:  * Minor Improvements: Nicholas Miell (nmiell@gmail.com)
sl@0:  * MMX code paths for fbcompose.c by Lars Knoll (lars@trolltech.com) 
sl@0:  *
sl@0:  * Based on work by Owen Taylor
sl@0:  */
sl@0: //Portions Copyright (c)  2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved. 
sl@0: 
sl@0: #ifdef HAVE_CONFIG_H
sl@0: #include "config.h"
sl@0: #endif
sl@0: 
sl@0: #include <liboil/liboil.h>
sl@0: #include <liboil/liboilfunction.h>
sl@0:  
sl@0: #include <xmmintrin.h> /* for _mm_shuffle_pi16 and _MM_SHUFFLE */
sl@0: 
sl@0: typedef uint32_t CARD32;
sl@0: typedef uint16_t CARD16;
sl@0: typedef int16_t INT16;
sl@0: typedef uint8_t CARD8;
sl@0: typedef uint64_t ullong;
sl@0: typedef CARD32* PicturePtr;
sl@0: typedef CARD32* FbBits;
sl@0: typedef int FbStride;
sl@0: 
sl@0: 
sl@0: #include "fbmmx.h"
sl@0: #include "fbpict.h"
sl@0: 
sl@0: #define CHECKPOINT()
sl@0: 
sl@0: OIL_DECLARE_CLASS (composite_in_argb);
sl@0: OIL_DECLARE_CLASS (composite_in_argb_const_src);
sl@0: OIL_DECLARE_CLASS (composite_in_argb_const_mask);
sl@0: OIL_DECLARE_CLASS (composite_over_argb);
sl@0: OIL_DECLARE_CLASS (composite_over_argb_const_src);
sl@0: OIL_DECLARE_CLASS (composite_add_argb);
sl@0: OIL_DECLARE_CLASS (composite_add_argb_const_src);
sl@0: OIL_DECLARE_CLASS (composite_in_over_argb);
sl@0: OIL_DECLARE_CLASS (composite_in_over_argb_const_src);
sl@0: OIL_DECLARE_CLASS (composite_in_over_argb_const_mask);
sl@0: OIL_DECLARE_CLASS (composite_over_u8);
sl@0: OIL_DECLARE_CLASS (composite_add_u8);
sl@0: 
sl@0: 
sl@0: /* --------------- MMX code patch for fbcompose.c --------------------- */
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: mmxCombineMaskU (uint32_t *dest, const uint32_t *src, const uint8_t *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     
sl@0:     const uint32_t *end = mask + width;
sl@0:     while (mask < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         a = MmxAlpha(a);
sl@0:         MmxMul(s, a);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef ENABLE_BROKEN_IMPLS
sl@0: static void
sl@0: mmxCombineOverU (uint32_t *dest, const uint32_t *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const uint32_t *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 x, y, a;
sl@0:         x = MmxTo(*src);
sl@0:         y = MmxTo(*dest);
sl@0:         a = MmxAlpha(x);
sl@0:         a = MmxNegate(a);
sl@0:         MmxMulAdd(y, a, x);
sl@0:         *dest = MmxFrom(y);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL(mmxCombineOverU, composite_over_argb, OIL_IMPL_FLAG_MMX);
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static FASTCALL void
sl@0: mmxCombineOverReverseU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 x, y, a;
sl@0:         x = MmxTo(*dest);
sl@0:         y = MmxTo(*src);
sl@0:         a = MmxAlpha(x);
sl@0:         a = MmxNegate(a);
sl@0:         MmxMulAdd(y, a, x);
sl@0:         *dest = MmxFrom(y);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: mmxCombineInU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 x, a;
sl@0:         x = MmxTo(*src);
sl@0:         a = MmxTo(*dest);
sl@0:         a = MmxAlpha(a);
sl@0:         MmxMul(x, a);
sl@0:         *dest = MmxFrom(x);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static FASTCALL void
sl@0: mmxCombineInReverseU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 x, a;
sl@0:         x = MmxTo(*dest);
sl@0:         a = MmxTo(*src);
sl@0:         a = MmxAlpha(a);
sl@0:         MmxMul(x, a);
sl@0:         *dest = MmxFrom(x);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static FASTCALL void
sl@0: mmxCombineOutU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 x, a;
sl@0:         x = MmxTo(*src);
sl@0:         a = MmxTo(*dest);
sl@0:         a = MmxAlpha(a);
sl@0:         a = MmxNegate(a);
sl@0:         MmxMul(x, a);
sl@0:         *dest = MmxFrom(x);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static FASTCALL void
sl@0: mmxCombineOutReverseU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 x, a;
sl@0:         x = MmxTo(*dest);
sl@0:         a = MmxTo(*src);
sl@0:         a = MmxAlpha(a);
sl@0:         a = MmxNegate(a);
sl@0:         MmxMul(x, a);
sl@0:         *dest = MmxFrom(x);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineAtopU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 s, da, d, sia;
sl@0:         s = MmxTo(*src);
sl@0:         d = MmxTo(*dest);
sl@0:         sia = MmxAlpha(s);
sl@0:         sia = MmxNegate(sia);
sl@0:         da = MmxAlpha(d);
sl@0:         MmxAddMul(s, da, d, sia);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineAtopReverseU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const CARD32 *end;
sl@0: 
sl@0:     end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 s, dia, d, sa;
sl@0:         s = MmxTo(*src);
sl@0:         d = MmxTo(*dest);
sl@0:         sa = MmxAlpha(s);
sl@0:         dia = MmxAlpha(d);
sl@0:         dia = MmxNegate(dia);
sl@0:         MmxAddMul(s, dia, d, sa);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineXorU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0: 
sl@0:     while (dest < end) {
sl@0:         __m64 s, dia, d, sia;
sl@0:         s = MmxTo(*src);
sl@0:         d = MmxTo(*dest);
sl@0:         sia = MmxAlpha(s);
sl@0:         dia = MmxAlpha(d);
sl@0:         sia = MmxNegate(sia);
sl@0:         dia = MmxNegate(dia);
sl@0:         MmxAddMul(s, dia, d, sia);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: static void
sl@0: mmxCombineAddU (uint32_t *dest, const uint32_t *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0: 
sl@0:     const uint32_t *end = dest + width;
sl@0:     while (dest < end) {
sl@0:         __m64 s, d;
sl@0:         s = MmxTo(*src);
sl@0:         d = MmxTo(*dest);
sl@0:         s = MmxAdd(s, d);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++dest;
sl@0:         ++src;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL(mmxCombineAddU, composite_add_argb, OIL_IMPL_FLAG_MMX | OIL_IMPL_FLAG_SSE);
sl@0: 
sl@0: #if 0
sl@0: static FASTCALL void
sl@0: mmxCombineSaturateU (CARD32 *dest, const CARD32 *src, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0: 
sl@0:     const CARD32 *end = dest + width;
sl@0:     while (dest < end) {
sl@0:         CARD32 s = *src;
sl@0:         CARD32 d = *dest;
sl@0:         __m64 ms = MmxTo(s);
sl@0:         __m64 md = MmxTo(d);
sl@0:         CARD32 sa = s >> 24;
sl@0:         CARD32 da = ~d >> 24;
sl@0: 
sl@0:         if (sa > da) {
sl@0:             __m64 msa = MmxTo(FbIntDiv(da, sa));
sl@0:             msa = MmxAlpha(msa);
sl@0:             MmxMul(ms, msa);
sl@0:         }
sl@0:         MmxAdd(md, ms);
sl@0:         *dest = MmxFrom(md);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineSrcC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0: 
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         MmxMul(s, a);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++src;
sl@0:         ++mask;
sl@0:         ++dest;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineOverC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 sa = MmxAlpha(s);
sl@0:         MmxMul(s, a);
sl@0:         MmxMul(a, sa);
sl@0:         a = MmxNegate(a);
sl@0:         MmxMulAdd(d, a, s);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineOverReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 da = MmxAlpha(d);
sl@0:         da = MmxNegate(da);
sl@0:         MmxMul(s, a);
sl@0:         MmxMulAdd(s, da, d);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineInC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 da = MmxAlpha(d);
sl@0:         MmxMul(s, a);
sl@0:         MmxMul(s, da);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineInReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 sa = MmxAlpha(s);
sl@0:         MmxMul(a, sa);
sl@0:         MmxMul(d, a);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineOutC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 da = MmxAlpha(d);
sl@0:         da = MmxNegate(da);
sl@0:         MmxMul(s, a);
sl@0:         MmxMul(s, da);
sl@0:         *dest = MmxFrom(s);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineOutReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 sa = MmxAlpha(s);
sl@0:         MmxMul(a, sa);
sl@0:         a = MmxNegate(a);
sl@0:         MmxMul(d, a);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineAtopC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 da = MmxAlpha(d);
sl@0:         __m64 sa = MmxAlpha(s); 
sl@0:         MmxMul(s, a);
sl@0:         MmxMul(a, sa);
sl@0:         a = MmxNegate(a);
sl@0:         MmxAddMul(d, a, s, da);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineAtopReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 da = MmxAlpha(d);
sl@0:         __m64 sa = MmxAlpha(s)
sl@0:         MmxMul(s, a);
sl@0:         MmxMul(a, sa);
sl@0:         da = MmxNegate(da);
sl@0:         MmxAddMul(d, a, s, da);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineXorC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     const __m64 mmx_4x00ff = (__m64) 0x00ff00ff00ff00ffULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         __m64 da = MmxAlpha(d);
sl@0:         __m64 sa = MmxAlpha(s);
sl@0:         MmxMul(s, a);
sl@0:         MmxMul(a, sa);
sl@0:         da = MmxNegate(da);
sl@0:         a = MmxNegate(a);
sl@0:         MmxAddMul(d, a, s, da);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: static FASTCALL void
sl@0: mmxCombineAddC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
sl@0: {
sl@0:     const __m64 mmx_0 = _mm_setzero_si64();
sl@0:     const __m64 mmx_4x0080 = (__m64) 0x0080008000800080ULL;
sl@0:     
sl@0:     const CARD32 *end = src + width;
sl@0:     while (src < end) {
sl@0:         __m64 a = MmxTo(*mask);
sl@0:         __m64 s = MmxTo(*src);
sl@0:         __m64 d = MmxTo(*dest);
sl@0:         MmxMul(s, a);
sl@0:         d = MmxAdd(s, d);
sl@0:         *dest = MmxFrom(d);
sl@0:         ++src;
sl@0:         ++dest;
sl@0:         ++mask;
sl@0:     }
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: extern FbComposeFunctions composeFunctions;
sl@0: 
sl@0: void fbComposeSetupMMX(void)
sl@0: {
sl@0:     /* check if we have MMX support and initialize accordingly */
sl@0:     if (fbHaveMMX()) {
sl@0:         composeFunctions.combineU[PictOpOver] = mmxCombineOverU;
sl@0:         composeFunctions.combineU[PictOpOverReverse] = mmxCombineOverReverseU;
sl@0:         composeFunctions.combineU[PictOpIn] = mmxCombineInU;
sl@0:         composeFunctions.combineU[PictOpInReverse] = mmxCombineInReverseU;
sl@0:         composeFunctions.combineU[PictOpOut] = mmxCombineOutU;
sl@0:         composeFunctions.combineU[PictOpOutReverse] = mmxCombineOutReverseU;
sl@0:         composeFunctions.combineU[PictOpAtop] = mmxCombineAtopU;
sl@0:         composeFunctions.combineU[PictOpAtopReverse] = mmxCombineAtopReverseU;
sl@0:         composeFunctions.combineU[PictOpXor] = mmxCombineXorU;
sl@0:         composeFunctions.combineU[PictOpAdd] = mmxCombineAddU;
sl@0:         composeFunctions.combineU[PictOpSaturate] = mmxCombineSaturateU;
sl@0: 
sl@0:         composeFunctions.combineC[PictOpSrc] = mmxCombineSrcC;
sl@0:         composeFunctions.combineC[PictOpOver] = mmxCombineOverC;
sl@0:         composeFunctions.combineC[PictOpOverReverse] = mmxCombineOverReverseC;
sl@0:         composeFunctions.combineC[PictOpIn] = mmxCombineInC;
sl@0:         composeFunctions.combineC[PictOpInReverse] = mmxCombineInReverseC;
sl@0:         composeFunctions.combineC[PictOpOut] = mmxCombineOutC;
sl@0:         composeFunctions.combineC[PictOpOutReverse] = mmxCombineOutReverseC;
sl@0:         composeFunctions.combineC[PictOpAtop] = mmxCombineAtopC;
sl@0:         composeFunctions.combineC[PictOpAtopReverse] = mmxCombineAtopReverseC;
sl@0:         composeFunctions.combineC[PictOpXor] = mmxCombineXorC;
sl@0:         composeFunctions.combineC[PictOpAdd] = mmxCombineAddC;
sl@0: 
sl@0:         composeFunctions.combineMaskU = mmxCombineMaskU;
sl@0:     } 
sl@0: }
sl@0: #endif
sl@0: 
sl@0: 
sl@0: /* ------------------ MMX code paths called from fbpict.c ----------------------- */
sl@0: 
sl@0: typedef union {
sl@0:   __m64 m64;
sl@0:   uint64_t ull;
sl@0: } m64_ull;
sl@0: 
sl@0: typedef struct
sl@0: {
sl@0:     m64_ull mmx_4x00ff;
sl@0:     m64_ull mmx_4x0080;
sl@0:     m64_ull mmx_565_rgb;
sl@0:     m64_ull mmx_565_unpack_multiplier;
sl@0:     m64_ull mmx_565_r;
sl@0:     m64_ull mmx_565_g;
sl@0:     m64_ull mmx_565_b;
sl@0:     m64_ull mmx_mask_0;
sl@0:     m64_ull mmx_mask_1;
sl@0:     m64_ull mmx_mask_2;
sl@0:     m64_ull mmx_mask_3;
sl@0:     m64_ull mmx_full_alpha;
sl@0:     m64_ull mmx_ffff0000ffff0000;
sl@0:     m64_ull mmx_0000ffff00000000;
sl@0:     m64_ull mmx_000000000000ffff;
sl@0: } MMXData;
sl@0: 
sl@0: static const MMXData c =
sl@0: {
sl@0:     .mmx_4x00ff.ull =			0x00ff00ff00ff00ffULL,
sl@0:     .mmx_4x0080.ull =			0x0080008000800080ULL,
sl@0:     .mmx_565_rgb.ull =			0x000001f0003f001fULL,
sl@0:     .mmx_565_r.ull =			0x000000f800000000ULL,
sl@0:     .mmx_565_g.ull =			0x0000000000fc0000ULL,
sl@0:     .mmx_565_b.ull =			0x00000000000000f8ULL,
sl@0:     .mmx_mask_0.ull =			0xffffffffffff0000ULL,
sl@0:     .mmx_mask_1.ull =			0xffffffff0000ffffULL,
sl@0:     .mmx_mask_2.ull =			0xffff0000ffffffffULL,
sl@0:     .mmx_mask_3.ull =			0x0000ffffffffffffULL,
sl@0:     .mmx_full_alpha.ull =			0x00ff000000000000ULL,
sl@0:     .mmx_565_unpack_multiplier.ull =	0x0000008404100840ULL,
sl@0:     .mmx_ffff0000ffff0000.ull =		0xffff0000ffff0000ULL,
sl@0:     .mmx_0000ffff00000000.ull =		0x0000ffff00000000ULL,
sl@0:     .mmx_000000000000ffff.ull =		0x000000000000ffffULL,
sl@0: };
sl@0: 
sl@0: #define MC(x) ((__m64) c.mmx_##x.m64)
sl@0: 
sl@0: static __inline__ __m64
sl@0: shift (__m64 v, int s)
sl@0: {
sl@0:     if (s > 0)
sl@0: 	return _mm_slli_si64 (v, s);
sl@0:     else if (s < 0)
sl@0: 	return _mm_srli_si64 (v, -s);
sl@0:     else
sl@0: 	return v;
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: negate (__m64 mask)
sl@0: {
sl@0:     return _mm_xor_si64 (mask, MC(4x00ff));
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: pix_multiply (__m64 a, __m64 b)
sl@0: {
sl@0:     __m64 res;
sl@0:     
sl@0:     res = _mm_mullo_pi16 (a, b);
sl@0:     res = _mm_adds_pu16 (res, MC(4x0080));
sl@0:     res = _mm_adds_pu16 (res, _mm_srli_pi16 (res, 8));
sl@0:     res = _mm_srli_pi16 (res, 8);
sl@0:     
sl@0:     return res;
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: expand_alpha (__m64 pixel)
sl@0: {
sl@0:     return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE(3, 3, 3, 3));
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: expand_alpha_rev (__m64 pixel)
sl@0: {
sl@0:     return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE(0, 0, 0, 0));
sl@0: }    
sl@0: 
sl@0: static __inline__ __m64
sl@0: invert_colors (__m64 pixel)
sl@0: {
sl@0:     return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE(3, 0, 1, 2));
sl@0: }
sl@0: 
sl@0: /* Notes about writing mmx code
sl@0:  *
sl@0:  * give memory operands as the second operand. If you give it as the
sl@0:  * first, gcc will first load it into a register, then use that
sl@0:  * register
sl@0:  *
sl@0:  *   ie. use
sl@0:  *
sl@0:  *         _mm_mullo_pi16 (x, mmx_constant);
sl@0:  *
sl@0:  *   not
sl@0:  *
sl@0:  *         _mm_mullo_pi16 (mmx_constant, x);
sl@0:  *
sl@0:  * Also try to minimize dependencies. i.e. when you need a value, try
sl@0:  * to calculate it from a value that was calculated as early as
sl@0:  * possible.
sl@0:  */
sl@0: 
sl@0: static __inline__ __m64
sl@0: over (__m64 src, __m64 srca, __m64 dest)
sl@0: {
sl@0:     return  _mm_adds_pu8 (src, pix_multiply(dest, negate(srca)));
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: over_rev_non_pre (__m64 src, __m64 dest)
sl@0: {
sl@0:     __m64 srca = expand_alpha (src);
sl@0:     __m64 srcfaaa = _mm_or_si64 (srca, MC(full_alpha));
sl@0:     
sl@0:     return over(pix_multiply(invert_colors(src), srcfaaa), srca, dest);
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: in (__m64 src,
sl@0:     __m64 mask)
sl@0: {
sl@0:     return pix_multiply (src, mask);
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: in_over (__m64 src,
sl@0: 	 __m64 srca,
sl@0: 	 __m64 mask,
sl@0: 	 __m64 dest)
sl@0: {
sl@0:     return over(in(src, mask), pix_multiply(srca, mask), dest);
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: load8888 (CARD32 v)
sl@0: {
sl@0:     return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (v), _mm_setzero_si64());
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: pack8888 (__m64 lo, __m64 hi)
sl@0: {
sl@0:     __m64 r;
sl@0:     r = _mm_packs_pu16 (lo, hi);
sl@0:     return r;
sl@0: }
sl@0: 
sl@0: static __inline__ CARD32
sl@0: store8888 (__m64 v)
sl@0: {
sl@0:     return _mm_cvtsi64_si32(pack8888(v, _mm_setzero_si64()));
sl@0: }
sl@0: 
sl@0: /* Expand 16 bits positioned at @pos (0-3) of a mmx register into
sl@0:  *
sl@0:  *    00RR00GG00BB
sl@0:  * 
sl@0:  * --- Expanding 565 in the low word ---
sl@0:  * 
sl@0:  * m = (m << (32 - 3)) | (m << (16 - 5)) | m;
sl@0:  * m = m & (01f0003f001f);
sl@0:  * m = m * (008404100840);
sl@0:  * m = m >> 8;
sl@0:  * 
sl@0:  * Note the trick here - the top word is shifted by another nibble to
sl@0:  * avoid it bumping into the middle word
sl@0:  */
sl@0: static __inline__ __m64
sl@0: expand565 (__m64 pixel, int pos)
sl@0: {
sl@0:     __m64 p = pixel;
sl@0:     __m64 t1, t2;
sl@0:     
sl@0:     /* move pixel to low 16 bit and zero the rest */
sl@0:     p = shift (shift (p, (3 - pos) * 16), -48); 
sl@0:     
sl@0:     t1 = shift (p, 36 - 11);
sl@0:     t2 = shift (p, 16 - 5);
sl@0:     
sl@0:     p = _mm_or_si64 (t1, p);
sl@0:     p = _mm_or_si64 (t2, p);
sl@0:     p = _mm_and_si64 (p, MC(565_rgb));
sl@0:     
sl@0:     pixel = _mm_mullo_pi16 (p, MC(565_unpack_multiplier));
sl@0:     return _mm_srli_pi16 (pixel, 8);
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: expand8888 (__m64 in, int pos)
sl@0: {
sl@0:     if (pos == 0)
sl@0: 	return _mm_unpacklo_pi8 (in, _mm_setzero_si64());
sl@0:     else
sl@0: 	return _mm_unpackhi_pi8 (in, _mm_setzero_si64());
sl@0: }
sl@0: 
sl@0: static __inline__ __m64
sl@0: pack565 (__m64 pixel, __m64 target, int pos)
sl@0: {
sl@0:     __m64 p = pixel;
sl@0:     __m64 t = target;
sl@0:     __m64 r, g, b;
sl@0:     
sl@0:     r = _mm_and_si64 (p, MC(565_r));
sl@0:     g = _mm_and_si64 (p, MC(565_g));
sl@0:     b = _mm_and_si64 (p, MC(565_b));
sl@0:     
sl@0:     r = shift (r, - (32 - 8) + pos * 16);
sl@0:     g = shift (g, - (16 - 3) + pos * 16);
sl@0:     b = shift (b, - (0  + 3) + pos * 16);
sl@0:     
sl@0:     if (pos == 0)
sl@0: 	t = _mm_and_si64 (t, MC(mask_0));
sl@0:     else if (pos == 1)
sl@0: 	t = _mm_and_si64 (t, MC(mask_1));
sl@0:     else if (pos == 2)
sl@0: 	t = _mm_and_si64 (t, MC(mask_2));
sl@0:     else if (pos == 3)
sl@0: 	t = _mm_and_si64 (t, MC(mask_3));
sl@0:     
sl@0:     p = _mm_or_si64 (r, t);
sl@0:     p = _mm_or_si64 (g, p);
sl@0:     
sl@0:     return _mm_or_si64 (b, p);
sl@0: }
sl@0: 
sl@0: #ifdef ENABLE_BROKEN_IMPLS
sl@0: /* broken.  See Debian bug #340932 */
sl@0: static void
sl@0: fbCompositeSolid_nx8888mmx (uint32_t *dst, uint32_t *src, int w)
sl@0: {
sl@0:     __m64	vsrc, vsrca;
sl@0: 
sl@0:     vsrc = load8888 (*src);
sl@0:     vsrca = expand_alpha (vsrc);
sl@0: 
sl@0:     while (w && (unsigned long)dst & 7)
sl@0:     {
sl@0:         *dst = store8888(over(vsrc, vsrca, load8888(*dst)));
sl@0:         
sl@0:         w--;
sl@0:         dst++;
sl@0:     }
sl@0:     
sl@0:     while (w >= 2)
sl@0:     {
sl@0:         __m64 vdest;
sl@0:         __m64 dest0, dest1;
sl@0:         
sl@0:         vdest = *(__m64 *)dst;
sl@0:         
sl@0:         dest0 = over(vsrc, vsrca, expand8888(vdest, 0));
sl@0:         dest1 = over(vsrc, vsrca, expand8888(vdest, 1));
sl@0:         
sl@0:         *(__m64 *)dst = pack8888(dest0, dest1);
sl@0:         
sl@0:         dst += 2;
sl@0:         w -= 2;
sl@0:     }
sl@0:     
sl@0:     while (w)
sl@0:     {
sl@0:         *dst = store8888(over(vsrc, vsrca, load8888(*dst)));
sl@0:         
sl@0:         w--;
sl@0:         dst++;
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL(fbCompositeSolid_nx8888mmx, composite_over_argb_const_src,
sl@0:     OIL_IMPL_FLAG_MMX| OIL_IMPL_FLAG_MMXEXT);
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: void
sl@0: fbCompositeSolid_nx0565mmx (CARD8	op,
sl@0: 			    PicturePtr pSrc,
sl@0: 			    PicturePtr pMask,
sl@0: 			    PicturePtr pDst,
sl@0: 			    INT16	xSrc,
sl@0: 			    INT16	ySrc,
sl@0: 			    INT16	xMask,
sl@0: 			    INT16	yMask,
sl@0: 			    INT16	xDst,
sl@0: 			    INT16	yDst,
sl@0: 			    CARD16	width,
sl@0: 			    CARD16	height)
sl@0: {
sl@0:     CARD32	src;
sl@0:     CARD16	*dstLine, *dst;
sl@0:     CARD16	w;
sl@0:     FbStride	dstStride;
sl@0:     __m64	vsrc, vsrca;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetSolid(pSrc, src, pDst->format);
sl@0:     
sl@0:     if (src >> 24 == 0)
sl@0: 	return;
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD16, dstStride, dstLine, 1);
sl@0:     
sl@0:     vsrc = load8888 (src);
sl@0:     vsrca = expand_alpha (vsrc);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	w = width;
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    ullong d = *dst;
sl@0: 	    __m64 vdest = expand565 ((__m64)d, 0);
sl@0: 	    vdest = pack565(over(vsrc, vsrca, vdest), vdest, 0);
sl@0: 	    *dst = (ullong)vdest;
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	}
sl@0: 	
sl@0: 	while (w >= 4)
sl@0: 	{
sl@0: 	    __m64 vdest;
sl@0: 	    
sl@0: 	    vdest = *(__m64 *)dst;
sl@0: 	    
sl@0: 	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 0)), vdest, 0);
sl@0: 	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 1)), vdest, 1);
sl@0: 	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 2)), vdest, 2);
sl@0: 	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 3)), vdest, 3);
sl@0: 	    
sl@0: 	    *(__m64 *)dst = vdest;
sl@0: 	    
sl@0: 	    dst += 4;
sl@0: 	    w -= 4;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    ullong d = *dst;
sl@0: 	    __m64 vdest = expand565 ((__m64)d, 0);
sl@0: 	    vdest = pack565(over(vsrc, vsrca, vdest), vdest, 0);
sl@0: 	    *dst = (ullong)vdest;
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: fbCompositeSolidMask_nx8888x8888Cmmx (uint32_t *dst, uint32_t *src, uint8_t *mask, int w)
sl@0: {
sl@0:     CARD32	src, srca;
sl@0:     CARD32	*dstLine;
sl@0:     CARD32	*maskLine;
sl@0:     FbStride	dstStride, maskStride;
sl@0:     __m64	vsrc, vsrca;
sl@0:     
sl@0:     
sl@0:     while (twidth && (unsigned long)q & 7)
sl@0:     {
sl@0:         CARD32 m = *(CARD32 *)p;
sl@0:         
sl@0:         if (m)
sl@0:         {
sl@0:             __m64 vdest = load8888(*q);
sl@0:             vdest = in_over(vsrc, vsrca, load8888(m), vdest);
sl@0:             *q = (ullong)pack8888(vdest, _mm_setzero_si64());
sl@0:         }
sl@0:         
sl@0:         twidth--;
sl@0:         p++;
sl@0:         q++;
sl@0:     }
sl@0:     
sl@0:     while (twidth >= 2)
sl@0:     {
sl@0:         CARD32 m0, m1;
sl@0:         m0 = *p;
sl@0:         m1 = *(p + 1);
sl@0:         
sl@0:         if (m0 | m1)
sl@0:         {
sl@0:             __m64 dest0, dest1;
sl@0:             __m64 vdest = *(__m64 *)q;
sl@0:             
sl@0:             dest0 = in_over(vsrc, vsrca, load8888(m0),
sl@0:                             expand8888 (vdest, 0));
sl@0:             dest1 = in_over(vsrc, vsrca, load8888(m1),
sl@0:                             expand8888 (vdest, 1));
sl@0:             
sl@0:             *(__m64 *)q = pack8888(dest0, dest1);
sl@0:         }
sl@0:         
sl@0:         p += 2;
sl@0:         q += 2;
sl@0:         twidth -= 2;
sl@0:     }
sl@0:     
sl@0:     while (twidth)
sl@0:     {
sl@0:         CARD32 m = *(CARD32 *)p;
sl@0:         
sl@0:         if (m)
sl@0:         {
sl@0:             __m64 vdest = load8888(*q);
sl@0:             vdest = in_over(vsrc, vsrca, load8888(m), vdest);
sl@0:             *q = (ullong)pack8888(vdest, _mm_setzero_si64());
sl@0:         }
sl@0:         
sl@0:         twidth--;
sl@0:         p++;
sl@0:         q++;
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: fbCompositeSrc_8888x8x8888mmx (uint32_t *dest, uint32_t *src, uint8_t *mask,
sl@0:     int width)
sl@0: {
sl@0: 
sl@0:     mask = *maskLine << 24 | *maskLine << 16 | *maskLine << 8 | *maskLine;
sl@0:     vmask = load8888 (mask);
sl@0:     srca = MC(4x00ff);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	src = srcLine;
sl@0: 	srcLine += srcStride;
sl@0: 	w = width;
sl@0: 
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    __m64 s = load8888 (*src);
sl@0: 	    __m64 d = load8888 (*dst);
sl@0: 	    
sl@0: 	    *dst = (ullong)pack8888 (in_over (s, srca, vmask, d), (__m64)_mm_setzero_si64());
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0: 
sl@0: 	while (w >= 16)
sl@0: 	{
sl@0: 	    __m64 vd0 = *(__m64 *)(dst + 0);
sl@0: 	    __m64 vd1 = *(__m64 *)(dst + 2);
sl@0: 	    __m64 vd2 = *(__m64 *)(dst + 4);
sl@0: 	    __m64 vd3 = *(__m64 *)(dst + 6);
sl@0: 	    __m64 vd4 = *(__m64 *)(dst + 8);
sl@0: 	    __m64 vd5 = *(__m64 *)(dst + 10);
sl@0: 	    __m64 vd6 = *(__m64 *)(dst + 12);
sl@0: 	    __m64 vd7 = *(__m64 *)(dst + 14);
sl@0: 
sl@0: 	    __m64 vs0 = *(__m64 *)(src + 0);
sl@0: 	    __m64 vs1 = *(__m64 *)(src + 2);
sl@0: 	    __m64 vs2 = *(__m64 *)(src + 4);
sl@0: 	    __m64 vs3 = *(__m64 *)(src + 6);
sl@0: 	    __m64 vs4 = *(__m64 *)(src + 8);
sl@0: 	    __m64 vs5 = *(__m64 *)(src + 10);
sl@0: 	    __m64 vs6 = *(__m64 *)(src + 12);
sl@0: 	    __m64 vs7 = *(__m64 *)(src + 14);
sl@0: 
sl@0: 	    vd0 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs0, 0), srca, vmask, expand8888 (vd0, 0)),
sl@0: 		in_over (expand8888 (vs0, 1), srca, vmask, expand8888 (vd0, 1)));
sl@0: 	
sl@0: 	    vd1 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs1, 0), srca, vmask, expand8888 (vd1, 0)),
sl@0: 		in_over (expand8888 (vs1, 1), srca, vmask, expand8888 (vd1, 1)));
sl@0: 	
sl@0: 	    vd2 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs2, 0), srca, vmask, expand8888 (vd2, 0)),
sl@0: 		in_over (expand8888 (vs2, 1), srca, vmask, expand8888 (vd2, 1)));
sl@0: 	
sl@0: 	    vd3 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs3, 0), srca, vmask, expand8888 (vd3, 0)),
sl@0: 		in_over (expand8888 (vs3, 1), srca, vmask, expand8888 (vd3, 1)));
sl@0: 	
sl@0: 	    vd4 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs4, 0), srca, vmask, expand8888 (vd4, 0)),
sl@0: 		in_over (expand8888 (vs4, 1), srca, vmask, expand8888 (vd4, 1)));
sl@0: 	
sl@0: 	    vd5 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs5, 0), srca, vmask, expand8888 (vd5, 0)),
sl@0: 		in_over (expand8888 (vs5, 1), srca, vmask, expand8888 (vd5, 1)));
sl@0: 	
sl@0: 	    vd6 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs6, 0), srca, vmask, expand8888 (vd6, 0)),
sl@0: 		in_over (expand8888 (vs6, 1), srca, vmask, expand8888 (vd6, 1)));
sl@0: 	
sl@0: 	    vd7 = (__m64)pack8888 (
sl@0: 		in_over (expand8888 (vs7, 0), srca, vmask, expand8888 (vd7, 0)),
sl@0: 		in_over (expand8888 (vs7, 1), srca, vmask, expand8888 (vd7, 1)));
sl@0: 
sl@0:     	    *(__m64 *)(dst + 0) = vd0;
sl@0: 	    *(__m64 *)(dst + 2) = vd1;
sl@0: 	    *(__m64 *)(dst + 4) = vd2;
sl@0: 	    *(__m64 *)(dst + 6) = vd3;
sl@0: 	    *(__m64 *)(dst + 8) = vd4;
sl@0: 	    *(__m64 *)(dst + 10) = vd5;
sl@0: 	    *(__m64 *)(dst + 12) = vd6;
sl@0: 	    *(__m64 *)(dst + 14) = vd7;
sl@0: 	
sl@0: 	    w -= 16;
sl@0: 	    dst += 16;
sl@0: 	    src += 16;
sl@0: 	}
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    __m64 s = load8888 (*src);
sl@0: 	    __m64 d = load8888 (*dst);
sl@0: 	    
sl@0: 	    *dst = (ullong)pack8888 (in_over (s, srca, vmask, d), (__m64)_mm_setzero_si64());
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0:     }
sl@0: 
sl@0:     _mm_empty(); 
sl@0: }
sl@0: 
sl@0: void
sl@0: fbCompositeSrc_8888x8888mmx (CARD8	op,
sl@0: 			     PicturePtr pSrc,
sl@0: 			     PicturePtr pMask,
sl@0: 			     PicturePtr pDst,
sl@0: 			     INT16	xSrc,
sl@0: 			     INT16	ySrc,
sl@0: 			     INT16      xMask,
sl@0: 			     INT16      yMask,
sl@0: 			     INT16      xDst,
sl@0: 			     INT16      yDst,
sl@0: 			     CARD16     width,
sl@0: 			     CARD16     height)
sl@0: {
sl@0:     CARD32	*dstLine, *dst;
sl@0:     CARD32	*srcLine, *src;
sl@0:     FbStride	dstStride, srcStride;
sl@0:     CARD16	w;
sl@0:     __m64  srca;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
sl@0:     fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
sl@0: 
sl@0:     srca = MC (4x00ff);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	src = srcLine;
sl@0: 	srcLine += srcStride;
sl@0: 	w = width;
sl@0: 
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    __m64 s = load8888 (*src);
sl@0: 	    __m64 d = load8888 (*dst);
sl@0: 	    
sl@0: 	    *dst = (ullong)pack8888 (over (s, expand_alpha (s), d), (__m64)_mm_setzero_si64());
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0: 
sl@0: 	while (w >= 2)
sl@0: 	{
sl@0: 	    __m64 vd = *(__m64 *)(dst + 0);
sl@0: 	    __m64 vs = *(__m64 *)(src + 0);
sl@0: 	    __m64 vs0 = expand8888 (vs, 0);
sl@0: 	    __m64 vs1 = expand8888 (vs, 1);
sl@0: 
sl@0: 	    *(__m64 *)dst = (__m64)pack8888 (
sl@0: 		over (vs0, expand_alpha (vs0), expand8888 (vd, 0)),
sl@0: 		over (vs1, expand_alpha (vs1), expand8888 (vd, 1)));
sl@0: 	    
sl@0: 	    w -= 2;
sl@0: 	    dst += 2;
sl@0: 	    src += 2;
sl@0: 	}
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    __m64 s = load8888 (*src);
sl@0: 	    __m64 d = load8888 (*dst);
sl@0: 	    
sl@0: 	    *dst = (ullong)pack8888 (over (s, expand_alpha (s), d),
sl@0: 				     (__m64)_mm_setzero_si64());
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0:     }
sl@0: 
sl@0:     _mm_empty(); 
sl@0: }
sl@0: 
sl@0: void
sl@0: fbCompositeSolidMask_nx8x8888mmx (CARD8      op,
sl@0: 				  PicturePtr pSrc,
sl@0: 				  PicturePtr pMask,
sl@0: 				  PicturePtr pDst,
sl@0: 				  INT16      xSrc,
sl@0: 				  INT16      ySrc,
sl@0: 				  INT16      xMask,
sl@0: 				  INT16      yMask,
sl@0: 				  INT16      xDst,
sl@0: 				  INT16      yDst,
sl@0: 				  CARD16     width,
sl@0: 				  CARD16     height)
sl@0: {
sl@0:     CARD32	src, srca;
sl@0:     CARD32	*dstLine, *dst;
sl@0:     CARD8	*maskLine, *mask;
sl@0:     FbStride	dstStride, maskStride;
sl@0:     CARD16	w;
sl@0:     __m64	vsrc, vsrca;
sl@0:     ullong	srcsrc;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetSolid(pSrc, src, pDst->format);
sl@0:     
sl@0:     srca = src >> 24;
sl@0:     if (srca == 0)
sl@0: 	return;
sl@0:     
sl@0:     srcsrc = (unsigned long long)src << 32 | src;
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
sl@0:     fbComposeGetStart (pMask, xMask, yMask, CARD8, maskStride, maskLine, 1);
sl@0:     
sl@0:     vsrc = load8888 (src);
sl@0:     vsrca = expand_alpha (vsrc);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	mask = maskLine;
sl@0: 	maskLine += maskStride;
sl@0: 	w = width;
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    ullong m = *mask;
sl@0: 	    
sl@0: 	    if (m)
sl@0: 	    {
sl@0: 		__m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), load8888(*dst));
sl@0: 		*dst = (ullong)pack8888(vdest, _mm_setzero_si64());
sl@0: 	    }
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    mask++;
sl@0: 	    dst++;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w >= 2)
sl@0: 	{
sl@0: 	    ullong m0, m1;
sl@0: 	    m0 = *mask;
sl@0: 	    m1 = *(mask + 1);
sl@0: 	    
sl@0: 	    if (srca == 0xff && (m0 & m1) == 0xff)
sl@0: 	    {
sl@0: 		*(unsigned long long *)dst = srcsrc;
sl@0: 	    }
sl@0: 	    else if (m0 | m1)
sl@0: 	    {
sl@0: 		__m64 vdest;
sl@0: 		__m64 dest0, dest1;
sl@0: 		
sl@0: 		vdest = *(__m64 *)dst;
sl@0: 		
sl@0: 		dest0 = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m0), expand8888(vdest, 0));
sl@0: 		dest1 = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m1), expand8888(vdest, 1));
sl@0: 		
sl@0: 		*(__m64 *)dst = pack8888(dest0, dest1);
sl@0: 	    }
sl@0: 	    
sl@0: 	    mask += 2;
sl@0: 	    dst += 2;
sl@0: 	    w -= 2;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    ullong m = *mask;
sl@0: 	    
sl@0: 	    if (m)
sl@0: 	    {
sl@0: 		__m64 vdest = load8888(*dst);
sl@0: 		vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), vdest);
sl@0: 		*dst = (ullong)pack8888(vdest, _mm_setzero_si64());
sl@0: 	    }
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    mask++;
sl@0: 	    dst++;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: 
sl@0: void
sl@0: fbCompositeSolidMask_nx8x0565mmx (CARD8      op,
sl@0: 				  PicturePtr pSrc,
sl@0: 				  PicturePtr pMask,
sl@0: 				  PicturePtr pDst,
sl@0: 				  INT16      xSrc,
sl@0: 				  INT16      ySrc,
sl@0: 				  INT16      xMask,
sl@0: 				  INT16      yMask,
sl@0: 				  INT16      xDst,
sl@0: 				  INT16      yDst,
sl@0: 				  CARD16     width,
sl@0: 				  CARD16     height)
sl@0: {
sl@0:     CARD32	src, srca;
sl@0:     CARD16	*dstLine, *dst;
sl@0:     CARD8	*maskLine, *mask;
sl@0:     FbStride	dstStride, maskStride;
sl@0:     CARD16	w;
sl@0:     __m64	vsrc, vsrca;
sl@0:     unsigned long long srcsrcsrcsrc, src16;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetSolid(pSrc, src, pDst->format);
sl@0:     
sl@0:     srca = src >> 24;
sl@0:     if (srca == 0)
sl@0: 	return;
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD16, dstStride, dstLine, 1);
sl@0:     fbComposeGetStart (pMask, xMask, yMask, CARD8, maskStride, maskLine, 1);
sl@0:     
sl@0:     vsrc = load8888 (src);
sl@0:     vsrca = expand_alpha (vsrc);
sl@0:     
sl@0:     src16 = (ullong)pack565(vsrc, _mm_setzero_si64(), 0);
sl@0:     
sl@0:     srcsrcsrcsrc = (ullong)src16 << 48 | (ullong)src16 << 32 |
sl@0: 	(ullong)src16 << 16 | (ullong)src16;
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	mask = maskLine;
sl@0: 	maskLine += maskStride;
sl@0: 	w = width;
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    ullong m = *mask;
sl@0: 	    
sl@0: 	    if (m)
sl@0: 	    {
sl@0: 		ullong d = *dst;
sl@0: 		__m64 vd = (__m64)d;
sl@0: 		__m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), expand565(vd, 0));
sl@0: 		*dst = (ullong)pack565(vdest, _mm_setzero_si64(), 0);
sl@0: 	    }
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    mask++;
sl@0: 	    dst++;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w >= 4)
sl@0: 	{
sl@0: 	    ullong m0, m1, m2, m3;
sl@0: 	    m0 = *mask;
sl@0: 	    m1 = *(mask + 1);
sl@0: 	    m2 = *(mask + 2);
sl@0: 	    m3 = *(mask + 3);
sl@0: 	    
sl@0: 	    if (srca == 0xff && (m0 & m1 & m2 & m3) == 0xff)
sl@0: 	    {
sl@0: 		*(unsigned long long *)dst = srcsrcsrcsrc;
sl@0: 	    }
sl@0: 	    else if (m0 | m1 | m2 | m3)
sl@0: 	    {
sl@0: 		__m64 vdest;
sl@0: 		__m64 vm0, vm1, vm2, vm3;
sl@0: 		
sl@0: 		vdest = *(__m64 *)dst;
sl@0: 		
sl@0: 		vm0 = (__m64)m0;
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm0), expand565(vdest, 0)), vdest, 0);
sl@0: 		vm1 = (__m64)m1;
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm1), expand565(vdest, 1)), vdest, 1);
sl@0: 		vm2 = (__m64)m2;
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm2), expand565(vdest, 2)), vdest, 2);
sl@0: 		vm3 = (__m64)m3;
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm3), expand565(vdest, 3)), vdest, 3);
sl@0: 		
sl@0: 		*(__m64 *)dst = vdest;
sl@0: 	    }
sl@0: 	    
sl@0: 	    w -= 4;
sl@0: 	    mask += 4;
sl@0: 	    dst += 4;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    ullong m = *mask;
sl@0: 	    
sl@0: 	    if (m)
sl@0: 	    {
sl@0: 		ullong d = *dst;
sl@0: 		__m64 vd = (__m64)d;
sl@0: 		__m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), expand565(vd, 0));
sl@0: 		*dst = (ullong)pack565(vdest, _mm_setzero_si64(), 0);
sl@0: 	    }
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    mask++;
sl@0: 	    dst++;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: void
sl@0: fbCompositeSrc_8888RevNPx0565mmx (CARD8      op,
sl@0: 				  PicturePtr pSrc,
sl@0: 				  PicturePtr pMask,
sl@0: 				  PicturePtr pDst,
sl@0: 				  INT16      xSrc,
sl@0: 				  INT16      ySrc,
sl@0: 				  INT16      xMask,
sl@0: 				  INT16      yMask,
sl@0: 				  INT16      xDst,
sl@0: 				  INT16      yDst,
sl@0: 				  CARD16     width,
sl@0: 				  CARD16     height)
sl@0: {
sl@0:     CARD16	*dstLine, *dst;
sl@0:     CARD32	*srcLine, *src;
sl@0:     FbStride	dstStride, srcStride;
sl@0:     CARD16	w;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD16, dstStride, dstLine, 1);
sl@0:     fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
sl@0:     
sl@0:     assert (pSrc->pDrawable == pMask->pDrawable);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	src = srcLine;
sl@0: 	srcLine += srcStride;
sl@0: 	w = width;
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    __m64 vsrc = load8888 (*src);
sl@0: 	    ullong d = *dst;
sl@0: 	    __m64 vdest = expand565 ((__m64)d, 0);
sl@0: 	    
sl@0: 	    vdest = pack565(over_rev_non_pre(vsrc, vdest), vdest, 0);
sl@0: 	    
sl@0: 	    *dst = (ullong)vdest;
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w >= 4)
sl@0: 	{
sl@0: 	    CARD32 s0, s1, s2, s3;
sl@0: 	    unsigned char a0, a1, a2, a3;
sl@0: 	    
sl@0: 	    s0 = *src;
sl@0: 	    s1 = *(src + 1);
sl@0: 	    s2 = *(src + 2);
sl@0: 	    s3 = *(src + 3);
sl@0: 	    
sl@0: 	    a0 = (s0 >> 24);
sl@0: 	    a1 = (s1 >> 24);
sl@0: 	    a2 = (s2 >> 24);
sl@0: 	    a3 = (s3 >> 24);
sl@0: 	    
sl@0: 	    if ((a0 & a1 & a2 & a3) == 0xFF)
sl@0: 	    {
sl@0: 		__m64 vdest;
sl@0: 		vdest = pack565(invert_colors(load8888(s0)), _mm_setzero_si64(), 0);
sl@0: 		vdest = pack565(invert_colors(load8888(s1)), vdest, 1);
sl@0: 		vdest = pack565(invert_colors(load8888(s2)), vdest, 2);
sl@0: 		vdest = pack565(invert_colors(load8888(s3)), vdest, 3);
sl@0: 		
sl@0: 		*(__m64 *)dst = vdest;
sl@0: 	    }
sl@0: 	    else if (a0 | a1 | a2 | a3)
sl@0: 	    {
sl@0: 		__m64 vdest = *(__m64 *)dst;
sl@0: 		
sl@0: 		vdest = pack565(over_rev_non_pre(load8888(s0), expand565(vdest, 0)), vdest, 0);
sl@0: 	        vdest = pack565(over_rev_non_pre(load8888(s1), expand565(vdest, 1)), vdest, 1);
sl@0: 		vdest = pack565(over_rev_non_pre(load8888(s2), expand565(vdest, 2)), vdest, 2);
sl@0: 		vdest = pack565(over_rev_non_pre(load8888(s3), expand565(vdest, 3)), vdest, 3);
sl@0: 		
sl@0: 		*(__m64 *)dst = vdest;
sl@0: 	    }
sl@0: 	    
sl@0: 	    w -= 4;
sl@0: 	    dst += 4;
sl@0: 	    src += 4;
sl@0: 	}
sl@0: 	
sl@0: 	CHECKPOINT();
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    __m64 vsrc = load8888 (*src);
sl@0: 	    ullong d = *dst;
sl@0: 	    __m64 vdest = expand565 ((__m64)d, 0);
sl@0: 	    
sl@0: 	    vdest = pack565(over_rev_non_pre(vsrc, vdest), vdest, 0);
sl@0: 	    
sl@0: 	    *dst = (ullong)vdest;
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: /* "8888RevNP" is GdkPixbuf's format: ABGR, non premultiplied */
sl@0: 
sl@0: void
sl@0: fbCompositeSrc_8888RevNPx8888mmx (CARD8      op,
sl@0: 				  PicturePtr pSrc,
sl@0: 				  PicturePtr pMask,
sl@0: 				  PicturePtr pDst,
sl@0: 				  INT16      xSrc,
sl@0: 				  INT16      ySrc,
sl@0: 				  INT16      xMask,
sl@0: 				  INT16      yMask,
sl@0: 				  INT16      xDst,
sl@0: 				  INT16      yDst,
sl@0: 				  CARD16     width,
sl@0: 				  CARD16     height)
sl@0: {
sl@0:     CARD32	*dstLine, *dst;
sl@0:     CARD32	*srcLine, *src;
sl@0:     FbStride	dstStride, srcStride;
sl@0:     CARD16	w;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
sl@0:     fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
sl@0:     
sl@0:     assert (pSrc->pDrawable == pMask->pDrawable);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	dst = dstLine;
sl@0: 	dstLine += dstStride;
sl@0: 	src = srcLine;
sl@0: 	srcLine += srcStride;
sl@0: 	w = width;
sl@0: 	
sl@0: 	while (w && (unsigned long)dst & 7)
sl@0: 	{
sl@0: 	    __m64 s = load8888 (*src);
sl@0: 	    __m64 d = load8888 (*dst);
sl@0: 	    
sl@0: 	    *dst = (ullong)pack8888 (over_rev_non_pre (s, d), _mm_setzero_si64());
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0: 	
sl@0: 	while (w >= 2)
sl@0: 	{
sl@0: 	    ullong s0, s1;
sl@0: 	    unsigned char a0, a1;
sl@0: 	    __m64 d0, d1;
sl@0: 	    
sl@0: 	    s0 = *src;
sl@0: 	    s1 = *(src + 1);
sl@0: 	    
sl@0: 	    a0 = (s0 >> 24);
sl@0: 	    a1 = (s1 >> 24);
sl@0: 	    
sl@0: 	    if ((a0 & a1) == 0xFF)
sl@0: 	    {
sl@0: 		d0 = invert_colors(load8888(s0));
sl@0: 		d1 = invert_colors(load8888(s1));
sl@0: 		
sl@0: 		*(__m64 *)dst = pack8888 (d0, d1);
sl@0: 	    }
sl@0: 	    else if (a0 | a1)
sl@0: 	    {
sl@0: 		__m64 vdest = *(__m64 *)dst;
sl@0: 		
sl@0: 		d0 = over_rev_non_pre (load8888(s0), expand8888 (vdest, 0));
sl@0: 		d1 = over_rev_non_pre (load8888(s1), expand8888 (vdest, 1));
sl@0: 		
sl@0: 		*(__m64 *)dst = pack8888 (d0, d1);
sl@0: 	    }
sl@0: 	    
sl@0: 	    w -= 2;
sl@0: 	    dst += 2;
sl@0: 	    src += 2;
sl@0: 	}
sl@0: 	
sl@0: 	while (w)
sl@0: 	{
sl@0: 	    __m64 s = load8888 (*src);
sl@0: 	    __m64 d = load8888 (*dst);
sl@0: 	    
sl@0: 	    *dst = (ullong)pack8888 (over_rev_non_pre (s, d), _mm_setzero_si64());
sl@0: 	    
sl@0: 	    w--;
sl@0: 	    dst++;
sl@0: 	    src++;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: 
sl@0: void
sl@0: fbCompositeSolidMask_nx8888x0565Cmmx (CARD8      op,
sl@0: 				      PicturePtr pSrc,
sl@0: 				      PicturePtr pMask,
sl@0: 				      PicturePtr pDst,
sl@0: 				      INT16      xSrc,
sl@0: 				      INT16      ySrc,
sl@0: 				      INT16      xMask,
sl@0: 				      INT16      yMask,
sl@0: 				      INT16      xDst,
sl@0: 				      INT16      yDst,
sl@0: 				      CARD16     width,
sl@0: 				      CARD16     height)
sl@0: {
sl@0:     CARD32	src, srca;
sl@0:     CARD16	*dstLine;
sl@0:     CARD32	*maskLine;
sl@0:     FbStride	dstStride, maskStride;
sl@0:     __m64  vsrc, vsrca;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbComposeGetSolid(pSrc, src, pDst->format);
sl@0:     
sl@0:     srca = src >> 24;
sl@0:     if (srca == 0)
sl@0: 	return;
sl@0:     
sl@0:     fbComposeGetStart (pDst, xDst, yDst, CARD16, dstStride, dstLine, 1);
sl@0:     fbComposeGetStart (pMask, xMask, yMask, CARD32, maskStride, maskLine, 1);
sl@0:     
sl@0:     vsrc = load8888 (src);
sl@0:     vsrca = expand_alpha (vsrc);
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	int twidth = width;
sl@0: 	CARD32 *p = (CARD32 *)maskLine;
sl@0: 	CARD16 *q = (CARD16 *)dstLine;
sl@0: 	
sl@0: 	while (twidth && ((unsigned long)q & 7))
sl@0: 	{
sl@0: 	    CARD32 m = *(CARD32 *)p;
sl@0: 	    
sl@0: 	    if (m)
sl@0: 	    {
sl@0: 		ullong d = *q;
sl@0: 		__m64 vdest = expand565 ((__m64)d, 0);
sl@0: 		vdest = pack565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
sl@0: 		*q = (ullong)vdest;
sl@0: 	    }
sl@0: 	    
sl@0: 	    twidth--;
sl@0: 	    p++;
sl@0: 	    q++;
sl@0: 	}
sl@0: 	
sl@0: 	while (twidth >= 4)
sl@0: 	{
sl@0: 	    CARD32 m0, m1, m2, m3;
sl@0: 	    
sl@0: 	    m0 = *p;
sl@0: 	    m1 = *(p + 1);
sl@0: 	    m2 = *(p + 2);
sl@0: 	    m3 = *(p + 3);
sl@0: 	    
sl@0: 	    if ((m0 | m1 | m2 | m3))
sl@0: 	    {
sl@0: 		__m64 vdest = *(__m64 *)q;
sl@0: 		
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, load8888(m0), expand565(vdest, 0)), vdest, 0);
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, load8888(m1), expand565(vdest, 1)), vdest, 1);
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, load8888(m2), expand565(vdest, 2)), vdest, 2);
sl@0: 		vdest = pack565(in_over(vsrc, vsrca, load8888(m3), expand565(vdest, 3)), vdest, 3);
sl@0: 		
sl@0: 		*(__m64 *)q = vdest;
sl@0: 	    }
sl@0: 	    twidth -= 4;
sl@0: 	    p += 4;
sl@0: 	    q += 4;
sl@0: 	}
sl@0: 	
sl@0: 	while (twidth)
sl@0: 	{
sl@0: 	    CARD32 m;
sl@0: 	    
sl@0: 	    m = *(CARD32 *)p;
sl@0: 	    if (m)
sl@0: 	    {
sl@0: 		ullong d = *q;
sl@0: 		__m64 vdest = expand565((__m64)d, 0);
sl@0: 		vdest = pack565 (in_over(vsrc, vsrca, load8888(m), vdest), vdest, 0);
sl@0: 		*q = (ullong)vdest;
sl@0: 	    }
sl@0: 	    
sl@0: 	    twidth--;
sl@0: 	    p++;
sl@0: 	    q++;
sl@0: 	}
sl@0: 	
sl@0: 	maskLine += maskStride;
sl@0: 	dstLine += dstStride;
sl@0:     }
sl@0:     
sl@0:     _mm_empty ();
sl@0: }
sl@0: #endif
sl@0: 
sl@0: static void
sl@0: fbCompositeSrcAdd_8000x8000mmx (uint8_t *dst, uint8_t *src, int w)
sl@0: {
sl@0:     int s;
sl@0:     int d;
sl@0:     int t;
sl@0: 
sl@0:     while (w && (unsigned long)dst & 7)
sl@0:     {
sl@0:         s = *src;
sl@0:         d = *dst;
sl@0:         t = d + s;
sl@0:         s = t | (0 - (t >> 8));
sl@0:         *dst = s;
sl@0:         
sl@0:         dst++;
sl@0:         src++;
sl@0:         w--;
sl@0:     }
sl@0:     
sl@0:     while (w >= 8)
sl@0:     {
sl@0:         *(__m64*)dst = _mm_adds_pu8(*(__m64*)src, *(__m64*)dst);
sl@0:         dst += 8;
sl@0:         src += 8;
sl@0:         w -= 8;
sl@0:     }
sl@0:     
sl@0:     while (w)
sl@0:     {
sl@0:         s = *src;
sl@0:         d = *dst;
sl@0:         t = d + s;
sl@0:         s = t | (0 - (t >> 8));
sl@0:         *dst = s;
sl@0:         
sl@0:         dst++;
sl@0:         src++;
sl@0:         w--;
sl@0:     }
sl@0: 
sl@0:     _mm_empty();
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (fbCompositeSrcAdd_8000x8000mmx, composite_add_u8, OIL_IMPL_FLAG_MMX);
sl@0: 
sl@0: static void
sl@0: fbCompositeSrcAdd_8888x8888mmx (uint32_t *dst, uint32_t *src, int w)
sl@0: {
sl@0:     while (w && (unsigned long)dst & 7)
sl@0:     {
sl@0:         *dst = _mm_cvtsi64_si32(_mm_adds_pu8(_mm_cvtsi32_si64(*src),
sl@0:                                              _mm_cvtsi32_si64(*dst)));
sl@0:         dst++;
sl@0:         src++;
sl@0:         w--;
sl@0:     }
sl@0:     
sl@0:     while (w >= 2)
sl@0:     {
sl@0:         *(__m64 *)dst = _mm_adds_pu8(*(__m64*)src, *(__m64*)dst);
sl@0:         dst += 2;
sl@0:         src += 2;
sl@0:         w -= 2;
sl@0:     }
sl@0:     
sl@0:     if (w)
sl@0:     {
sl@0:         *dst = _mm_cvtsi64_si32(_mm_adds_pu8(_mm_cvtsi32_si64(*src),
sl@0:                                              _mm_cvtsi32_si64(*dst)));
sl@0:         
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (fbCompositeSrcAdd_8888x8888mmx, composite_add_argb, OIL_IMPL_FLAG_MMX | OIL_IMPL_FLAG_SSE);
sl@0: 
sl@0: #if 0
sl@0: #define GetStart(drw,x,y,type,stride,line,bpp) {\
sl@0:     FbBits	*__bits__;									\
sl@0:     FbStride	__stride__;									\
sl@0:     int		__xoff__,__yoff__;								\
sl@0: 												\
sl@0:     fbGetDrawable((drw),__bits__,__stride__,bpp,__xoff__,__yoff__);				\
sl@0:     (stride) = __stride__ * sizeof (FbBits) / sizeof (type);					\
sl@0:     (line) = ((type *) __bits__) + (stride) * ((y) - __yoff__) + ((x) - __xoff__);		\
sl@0: }
sl@0: 
sl@0: Bool
sl@0: fbSolidFillmmx (DrawablePtr	pDraw,
sl@0: 		int		x,
sl@0: 		int		y,
sl@0: 		int		width,
sl@0: 		int		height,
sl@0: 		FbBits		xor)
sl@0: { 
sl@0:     FbStride	stride;
sl@0:     int		bpp;
sl@0:     ullong	fill;
sl@0:     __m64	vfill;
sl@0:     CARD32	byte_width;
sl@0:     CARD8	*byte_line;
sl@0:     FbBits      *bits;
sl@0:     int		xoff, yoff;
sl@0:     
sl@0:     CHECKPOINT();
sl@0:     
sl@0:     fbGetDrawable(pDraw, bits, stride, bpp, xoff, yoff);
sl@0:     
sl@0:     if (bpp == 16 && (xor >> 16 != (xor & 0xffff)))
sl@0: 	return FALSE;
sl@0:     
sl@0:     if (bpp != 16 && bpp != 32)
sl@0: 	return FALSE;
sl@0:     
sl@0:     if (bpp == 16)
sl@0:     {
sl@0: 	stride = stride * sizeof (FbBits) / 2;
sl@0: 	byte_line = (CARD8 *)(((CARD16 *)bits) + stride * (y - yoff) + (x - xoff));
sl@0: 	byte_width = 2 * width;
sl@0: 	stride *= 2;
sl@0:     }
sl@0:     else
sl@0:     {
sl@0: 	stride = stride * sizeof (FbBits) / 4;
sl@0: 	byte_line = (CARD8 *)(((CARD32 *)bits) + stride * (y - yoff) + (x - xoff));
sl@0: 	byte_width = 4 * width;
sl@0: 	stride *= 4;
sl@0:     }
sl@0:     
sl@0:     fill = ((ullong)xor << 32) | xor;
sl@0:     vfill = (__m64)fill;
sl@0:     
sl@0:     while (height--)
sl@0:     {
sl@0: 	int w;
sl@0: 	CARD8 *d = byte_line;
sl@0: 	byte_line += stride;
sl@0: 	w = byte_width;
sl@0: 	
sl@0: 	while (w >= 2 && ((unsigned long)d & 3))
sl@0: 	{
sl@0: 	    *(CARD16 *)d = xor;
sl@0: 	    w -= 2;
sl@0: 	    d += 2;
sl@0: 	}
sl@0: 	
sl@0: 	while (w >= 4 && ((unsigned long)d & 7))
sl@0: 	{
sl@0: 	    *(CARD32 *)d = xor;
sl@0: 	    
sl@0: 	    w -= 4;
sl@0: 	    d += 4;
sl@0: 	}
sl@0: 	
sl@0: 	while (w >= 64)
sl@0: 	{
sl@0: 	    *(__m64*) (d +  0) = vfill;
sl@0: 	    *(__m64*) (d +  8) = vfill;
sl@0: 	    *(__m64*) (d + 16) = vfill;
sl@0: 	    *(__m64*) (d + 24) = vfill;
sl@0: 	    *(__m64*) (d + 32) = vfill;
sl@0: 	    *(__m64*) (d + 40) = vfill;
sl@0: 	    *(__m64*) (d + 48) = vfill;
sl@0: 	    *(__m64*) (d + 56) = vfill;
sl@0: 	    
sl@0: 	    w -= 64;
sl@0: 	    d += 64;
sl@0: 	}
sl@0: 	while (w >= 4)
sl@0: 	{
sl@0: 	    *(CARD32 *)d = xor;
sl@0: 	    
sl@0: 	    w -= 4;
sl@0: 	    d += 4;
sl@0: 	}
sl@0: 	if (w >= 2)
sl@0: 	{
sl@0: 	    *(CARD16 *)d = xor;
sl@0: 	    w -= 2;
sl@0: 	    d += 2;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0:     return TRUE;
sl@0: }
sl@0: 
sl@0: Bool
sl@0: fbCopyAreammx (DrawablePtr	pSrc,
sl@0: 	       DrawablePtr	pDst,
sl@0: 	       int		src_x,
sl@0: 	       int		src_y,
sl@0: 	       int		dst_x,
sl@0: 	       int		dst_y,
sl@0: 	       int		width,
sl@0: 	       int		height)
sl@0: {
sl@0:     FbBits *	src_bits;
sl@0:     FbStride	src_stride;
sl@0:     int		src_bpp;
sl@0:     int		src_xoff;
sl@0:     int		src_yoff;
sl@0: 
sl@0:     FbBits *	dst_bits;
sl@0:     FbStride	dst_stride;
sl@0:     int		dst_bpp;
sl@0:     int		dst_xoff;
sl@0:     int		dst_yoff;
sl@0: 
sl@0:     CARD8 *	src_bytes;
sl@0:     CARD8 *	dst_bytes;
sl@0:     int		byte_width;
sl@0:     
sl@0:     fbGetDrawable(pSrc, src_bits, src_stride, src_bpp, src_xoff, src_yoff);
sl@0:     fbGetDrawable(pDst, dst_bits, dst_stride, dst_bpp, dst_xoff, dst_yoff);
sl@0: 
sl@0:     if (src_bpp != 16 && src_bpp != 32)
sl@0: 	return FALSE;
sl@0: 
sl@0:     if (dst_bpp != 16 && dst_bpp != 32)
sl@0: 	return FALSE;
sl@0: 
sl@0:     if (src_bpp != dst_bpp)
sl@0:     {
sl@0: 	return FALSE;
sl@0:     }
sl@0:     
sl@0:     if (src_bpp == 16)
sl@0:     {
sl@0: 	src_stride = src_stride * sizeof (FbBits) / 2;
sl@0: 	dst_stride = dst_stride * sizeof (FbBits) / 2;
sl@0: 	src_bytes = (CARD8 *)(((CARD16 *)src_bits) + src_stride * (src_y - src_yoff) + (src_x - src_xoff));
sl@0: 	dst_bytes = (CARD8 *)(((CARD16 *)dst_bits) + dst_stride * (dst_y - dst_yoff) + (dst_x - dst_xoff));
sl@0: 	byte_width = 2 * width;
sl@0: 	src_stride *= 2;
sl@0: 	dst_stride *= 2;
sl@0:     }
sl@0:     else
sl@0:     {
sl@0: 	src_stride = src_stride * sizeof (FbBits) / 4;
sl@0: 	dst_stride = dst_stride * sizeof (FbBits) / 4;
sl@0: 	src_bytes = (CARD8 *)(((CARD32 *)src_bits) + src_stride * (src_y - src_yoff) + (src_x - src_xoff));
sl@0: 	dst_bytes = (CARD8 *)(((CARD32 *)dst_bits) + dst_stride * (dst_y - dst_yoff) + (dst_x - dst_xoff));
sl@0: 	byte_width = 4 * width;
sl@0: 	src_stride *= 4;
sl@0: 	dst_stride *= 4;
sl@0:     }
sl@0: 
sl@0:     while (height--)
sl@0:     {
sl@0: 	int w;
sl@0: 	CARD8 *s = src_bytes;
sl@0: 	CARD8 *d = dst_bytes;
sl@0: 	src_bytes += src_stride;
sl@0: 	dst_bytes += dst_stride;
sl@0: 	w = byte_width;
sl@0: 	
sl@0: 	while (w >= 2 && ((unsigned long)d & 3))
sl@0: 	{
sl@0: 	    *(CARD16 *)d = *(CARD16 *)s;
sl@0: 	    w -= 2;
sl@0: 	    s += 2;
sl@0: 	    d += 2;
sl@0: 	}
sl@0: 	
sl@0: 	while (w >= 4 && ((unsigned long)d & 7))
sl@0: 	{
sl@0: 	    *(CARD32 *)d = *(CARD32 *)s;
sl@0: 	    
sl@0: 	    w -= 4;
sl@0: 	    s += 4;
sl@0: 	    d += 4;
sl@0: 	}
sl@0: 	
sl@0: 	while (w >= 64)
sl@0: 	{
sl@0: 	    *(__m64 *)(d + 0)  = *(__m64 *)(s + 0);
sl@0: 	    *(__m64 *)(d + 8)  = *(__m64 *)(s + 8);
sl@0: 	    *(__m64 *)(d + 16) = *(__m64 *)(s + 16);
sl@0: 	    *(__m64 *)(d + 24) = *(__m64 *)(s + 24);
sl@0: 	    *(__m64 *)(d + 32) = *(__m64 *)(s + 32);
sl@0: 	    *(__m64 *)(d + 40) = *(__m64 *)(s + 40);
sl@0: 	    *(__m64 *)(d + 48) = *(__m64 *)(s + 48);
sl@0: 	    *(__m64 *)(d + 56) = *(__m64 *)(s + 56);
sl@0: 	    w -= 64;
sl@0: 	    s += 64;
sl@0: 	    d += 64;
sl@0: 	}
sl@0: 	while (w >= 4)
sl@0: 	{
sl@0: 	    *(CARD32 *)d = *(CARD32 *)s;
sl@0: 
sl@0: 	    w -= 4;
sl@0: 	    s += 4;
sl@0: 	    d += 4;
sl@0: 	}
sl@0: 	if (w >= 2)
sl@0: 	{
sl@0: 	    *(CARD16 *)d = *(CARD16 *)s;
sl@0: 	    w -= 2;
sl@0: 	    s += 2;
sl@0: 	    d += 2;
sl@0: 	}
sl@0:     }
sl@0:     
sl@0:     _mm_empty();
sl@0:     return TRUE;
sl@0: }
sl@0: 
sl@0: void
sl@0: fbCompositeCopyAreammx (CARD8		op,
sl@0: 			PicturePtr	pSrc,
sl@0: 			PicturePtr	pMask,
sl@0: 			PicturePtr	pDst,
sl@0: 			INT16		xSrc,
sl@0: 			INT16		ySrc,
sl@0: 			INT16		xMask,
sl@0: 			INT16		yMask,
sl@0: 			INT16		xDst,
sl@0: 			INT16		yDst,
sl@0: 			CARD16		width,
sl@0: 			CARD16		height)
sl@0: {
sl@0:     fbCopyAreammx (pSrc->pDrawable,
sl@0: 		   pDst->pDrawable,
sl@0: 		   xSrc, ySrc,
sl@0: 		   xDst, yDst,
sl@0: 		   width, height);
sl@0: }
sl@0: 
sl@0: #if !defined(__amd64__) && !defined(__x86_64__)
sl@0: 
sl@0: enum CPUFeatures {
sl@0:     NoFeatures = 0,
sl@0:     MMX = 0x1,
sl@0:     MMX_Extensions = 0x2, 
sl@0:     SSE = 0x6,
sl@0:     SSE2 = 0x8,
sl@0:     CMOV = 0x10
sl@0: };
sl@0: 
sl@0: static unsigned int detectCPUFeatures(void) {
sl@0:     unsigned int result;
sl@0:     char vendor[13];
sl@0:     vendor[0] = 0;
sl@0:     vendor[12] = 0;
sl@0:     /* see p. 118 of amd64 instruction set manual Vol3 */
sl@0:     __asm__ ("push %%ebx\n"
sl@0:              "pushf\n"
sl@0:              "pop %%eax\n"
sl@0:              "mov %%eax, %%ebx\n"
sl@0:              "xor $0x00200000, %%eax\n"
sl@0:              "push %%eax\n"
sl@0:              "popf\n"
sl@0:              "pushf\n"
sl@0:              "pop %%eax\n"
sl@0:              "mov $0x0, %%edx\n"
sl@0:              "xor %%ebx, %%eax\n"
sl@0:              "jz skip\n"
sl@0: 
sl@0:              "mov $0x00000000, %%eax\n"
sl@0:              "cpuid\n"
sl@0:              "mov %%ebx, %1\n"
sl@0:              "mov %%edx, %2\n"
sl@0:              "mov %%ecx, %3\n"
sl@0:              "mov $0x00000001, %%eax\n"
sl@0:              "cpuid\n"
sl@0:              "skip:\n"
sl@0:              "pop %%ebx\n"
sl@0:              "mov %%edx, %0\n"
sl@0:              : "=r" (result), 
sl@0:                "=m" (vendor[0]), 
sl@0:                "=m" (vendor[4]), 
sl@0:                "=m" (vendor[8])
sl@0:              :
sl@0:              : "%eax", "%ecx", "%edx"
sl@0:         );
sl@0: 
sl@0:     unsigned int features = 0;
sl@0:     if (result) {
sl@0:         /* result now contains the standard feature bits */
sl@0:         if (result & (1 << 15))
sl@0:             features |= CMOV;
sl@0:         if (result & (1 << 23))
sl@0:             features |= MMX;
sl@0:         if (result & (1 << 25))
sl@0:             features |= SSE;
sl@0:         if (result & (1 << 26))
sl@0:             features |= SSE2;
sl@0:         if ((result & MMX) && !(result & SSE) && (strcmp(vendor, "AuthenticAMD") == 0)) {
sl@0:             /* check for AMD MMX extensions */
sl@0: 
sl@0:             unsigned int result;            
sl@0:             __asm__("push %%ebx\n"
sl@0:                     "mov $0x80000000, %%eax\n"
sl@0:                     "cpuid\n"
sl@0:                     "xor %%edx, %%edx\n"
sl@0:                     "cmp $0x1, %%eax\n"
sl@0:                     "jge skip2\n"
sl@0:                     "mov $0x80000001, %%eax\n"
sl@0:                     "cpuid\n"
sl@0:                     "skip2:\n"
sl@0:                     "mov %%edx, %0\n"
sl@0:                     "pop %%ebx\n"
sl@0:                     : "=r" (result)
sl@0:                     :
sl@0:                     : "%eax", "%ecx", "%edx"
sl@0:                 );
sl@0:             if (result & (1<<22))
sl@0:                 features |= MMX_Extensions;
sl@0:         }
sl@0:     }
sl@0:     return features;
sl@0: }
sl@0: 
sl@0: Bool
sl@0: fbHaveMMX (void)
sl@0: {
sl@0:     static Bool initialized = FALSE;
sl@0:     static Bool mmx_present;
sl@0:     
sl@0:     if (!initialized)
sl@0:     {
sl@0:         unsigned int features = detectCPUFeatures();
sl@0: 	mmx_present = (features & (MMX|MMX_Extensions)) == (MMX|MMX_Extensions);
sl@0:         initialized = TRUE;
sl@0:     }
sl@0:     
sl@0:     return mmx_present;
sl@0: }
sl@0: #endif /* __amd64__ */
sl@0: 
sl@0: 
sl@0: #endif
sl@0: 
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_mmxCombineOverU, composite_over_argb() {
sl@0: 		return &_oil_function_impl_mmxCombineOverU, composite_over_argb;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_mmxCombineAddU, composite_add_argb() {
sl@0: 		return &_oil_function_impl_mmxCombineAddU, composite_add_argb;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_fbCompositeSolid_nx8888mmx, composite_over_argb_const_src() {
sl@0: 		return &_oil_function_impl_fbCompositeSolid_nx8888mmx, composite_over_argb_const_src;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_fbCompositeSrcAdd_8000x8000mmx, composite_add_u8() {
sl@0: 		return &_oil_function_impl_fbCompositeSrcAdd_8000x8000mmx, composite_add_u8;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_fbCompositeSrcAdd_8888x8888mmx, composite_add_argb() {
sl@0: 		return &_oil_function_impl_fbCompositeSrcAdd_8888x8888mmx, composite_add_argb;
sl@0: }
sl@0: #endif
sl@0: