Magellan Linux

  • Back to homepage
    • /[pkg-src]/trunk/vnc/patches/9914_all_6.8.2-mmx-gcc4.patch

    Annotation of /trunk/vnc/patches/9914_all_6.8.2-mmx-gcc4.patch

    Parent Directory Parent Directory | Revision Log Revision Log

    Revision 153 - (hide annotations) (download)
    Tue May 8 20:52:56 2007 UTC (17 years, 1 month ago) by niro
    File size: 52896 byte(s) 
    -import
    1 niro 153 diff -ur xc-orig/programs/Xserver/fb/Imakefile xc/programs/Xserver/fb/Imakefile
    2     --- xc-orig/programs/Xserver/fb/Imakefile 2005-02-11 04:00:50.004092510 -0500
    3     +++ xc/programs/Xserver/fb/Imakefile 2005-02-11 04:01:32.059345739 -0500
    4     @@ -3,13 +3,22 @@
    5     XCOMM
    6     XCOMM Id: Imakefile,v 1.1 1999/11/02 03:54:44 keithp Exp $
    7    
    8     -#if defined(i386Architecture) && defined(HasGcc34) && HasGcc34
    9     +#if defined(HasGcc34) && HasGcc34
    10     MMXOPTIONS= -mmmx -Winline --param inline-unit-growth=10000 \
    11     - --param large-function-growth=10000 -DUSE_GCC34_MMX
    12     + --param large-function-growth=10000 -DUSE_MMX
    13     +SSEOPTIONS= $(MMXOPTIONS) -msse -DUSE_SSE
    14    
    15     +#if defined(i386Architecture)
    16     SpecialCObjectRule(fbmmx,fbmmx.c,$(MMXOPTIONS))
    17     +#elif defined(AMD64Architecture)
    18     +SpecialCObjectRule(fbmmx,fbmmx.c,$(SSEOPTIONS))
    19     +#endif
    20     +
    21     +#if defined(i386Architecture) || defined(AMD64Architecture)
    22     SpecialCObjectRule(fbpict,fbpict.c,$(MMXOPTIONS))
    23     SpecialCObjectRule(fbfill,fbfill.c,$(MMXOPTIONS))
    24     +SpecialCObjectRule(fbcopy,fbcopy.c,$(MMXOPTIONS))
    25     +#endif
    26    
    27     #endif
    28    
    29     diff -ur xc-orig/programs/Xserver/fb/fbcompose.c xc/programs/Xserver/fb/fbcompose.c
    30     --- xc-orig/programs/Xserver/fb/fbcompose.c 2005-02-11 04:00:50.009092659 -0500
    31     +++ xc/programs/Xserver/fb/fbcompose.c 2005-02-11 04:01:32.067345977 -0500
    32     @@ -1,8 +1,8 @@
    33     /*
    34     - * $XdotOrg: xc/programs/Xserver/fb/fbcompose.c,v 1.3 2004/05/12 01:49:46 anholt Exp $
    35     + * $XdotOrg: xc/programs/Xserver/fb/fbcompose.c,v 1.5 2005/01/13 20:49:21 sandmann Exp $
    36     * $XFree86: xc/programs/Xserver/fb/fbcompose.c,v 1.17tsi Exp $
    37     *
    38     - * Copyright © 2000 Keith Packard, member of The XFree86 Project, Inc.
    39     + * Copyright © 2000 Keith Packard, member of The XFree86 Project, Inc.
    40     *
    41     * Permission to use, copy, modify, distribute, and sell this software and its
    42     * documentation for any purpose is hereby granted without fee, provided that
    43     @@ -2693,7 +2693,6 @@
    44     op->u.transform.y = y - op->u.transform.top_y;
    45     }
    46    
    47     -
    48     Bool
    49     fbBuildCompositeOperand (PicturePtr pPict,
    50     FbCompositeOperand op[4],
    51     @@ -2710,7 +2709,6 @@
    52    
    53     op->u.transform.top_y = pPict->pDrawable->y;
    54     op->u.transform.left_x = pPict->pDrawable->x;
    55     -
    56     op->u.transform.start_x = x - op->u.transform.left_x;
    57     op->u.transform.x = op->u.transform.start_x;
    58     op->u.transform.y = y - op->u.transform.top_y;
    59     @@ -2822,6 +2820,21 @@
    60     FbCombineFunc f;
    61     int w;
    62    
    63     +#if 0
    64     + ErrorF ("op: %d\n"
    65     + "src format: %lx\n"
    66     + "msk format %lx\n"
    67     + "dst format %lx\n"
    68     + "width: %d\n"
    69     + "height %d\n",
    70     + op,
    71     + pSrc? pSrc->format : 0,
    72     + pMask? pMask->format : 0,
    73     + pDst? pDst->format : 0,
    74     + width, height);
    75     + ErrorF ("PICT_x8r8g8b8: %lx\n", PICT_x8r8g8b8);
    76     +#endif
    77     +
    78     if (!fbBuildCompositeOperand (pSrc, src, xSrc, ySrc, TRUE, TRUE))
    79     return;
    80     if (!fbBuildCompositeOperand (pDst, dst, xDst, yDst, FALSE, TRUE))
    81     diff -ur xc-orig/programs/Xserver/fb/fbcopy.c xc/programs/Xserver/fb/fbcopy.c
    82     --- xc-orig/programs/Xserver/fb/fbcopy.c 2005-02-11 04:00:50.004092510 -0500
    83     +++ xc/programs/Xserver/fb/fbcopy.c 2005-02-11 04:01:32.068346007 -0500
    84     @@ -1,7 +1,7 @@
    85     /*
    86     * Id: fbcopy.c,v 1.1 1999/11/02 03:54:45 keithp Exp $
    87     *
    88     - * Copyright © 1998 Keith Packard
    89     + * Copyright © 1998 Keith Packard
    90     *
    91     * Permission to use, copy, modify, distribute, and sell this software and its
    92     * documentation for any purpose is hereby granted without fee, provided that
    93     @@ -27,6 +27,7 @@
    94     #ifdef IN_MODULE
    95     #include "xf86_ansic.h"
    96     #endif
    97     +#include "fbmmx.h"
    98    
    99     void
    100     fbCopyNtoN (DrawablePtr pSrcDrawable,
    101     @@ -54,28 +55,51 @@
    102    
    103     fbGetDrawable (pSrcDrawable, src, srcStride, srcBpp, srcXoff, srcYoff);
    104     fbGetDrawable (pDstDrawable, dst, dstStride, dstBpp, dstXoff, dstYoff);
    105     -
    106     +
    107     while (nbox--)
    108     {
    109     +#ifdef USE_MMX
    110     + if (!reverse && !upsidedown && fbHaveMMX())
    111     + {
    112     + if (!fbCopyAreammx (pSrcDrawable,
    113     + pDstDrawable,
    114     +
    115     + (pbox->x1 + dx + srcXoff),
    116     + (pbox->y1 + dy + srcYoff),
    117     +
    118     + (pbox->x1 + dstXoff),
    119     + (pbox->y1 + dstYoff),
    120     +
    121     + (pbox->x2 - pbox->x1),
    122     + (pbox->y2 - pbox->y1)))
    123     + goto fallback;
    124     + else
    125     + goto next;
    126     + }
    127     + fallback:
    128     +#endif
    129     fbBlt (src + (pbox->y1 + dy + srcYoff) * srcStride,
    130     srcStride,
    131     (pbox->x1 + dx + srcXoff) * srcBpp,
    132     -
    133     +
    134     dst + (pbox->y1 + dstYoff) * dstStride,
    135     dstStride,
    136     (pbox->x1 + dstXoff) * dstBpp,
    137     -
    138     +
    139     (pbox->x2 - pbox->x1) * dstBpp,
    140     (pbox->y2 - pbox->y1),
    141     -
    142     +
    143     alu,
    144     pm,
    145     dstBpp,
    146     -
    147     +
    148     reverse,
    149     upsidedown);
    150     +#ifdef USE_MMX
    151     + next:
    152     +#endif
    153     pbox++;
    154     - }
    155     + }
    156     }
    157    
    158     void
    159     @@ -594,7 +618,7 @@
    160     int yOut)
    161     {
    162     fbCopyProc copy;
    163     -
    164     +
    165     #ifdef FB_24_32BIT
    166     if (pSrcDrawable->bitsPerPixel != pDstDrawable->bitsPerPixel)
    167     copy = fb24_32CopyMtoN;
    168     diff -ur xc-orig/programs/Xserver/fb/fbfill.c xc/programs/Xserver/fb/fbfill.c
    169     --- xc-orig/programs/Xserver/fb/fbfill.c 2005-02-11 04:00:50.006092570 -0500
    170     +++ xc/programs/Xserver/fb/fbfill.c 2005-02-11 04:01:32.069346037 -0500
    171     @@ -1,7 +1,7 @@
    172     /*
    173     * Id: fbfill.c,v 1.1 1999/11/02 03:54:45 keithp Exp $
    174     *
    175     - * Copyright © 1998 Keith Packard
    176     + * Copyright © 1998 Keith Packard
    177     *
    178     * Permission to use, copy, modify, distribute, and sell this software and its
    179     * documentation for any purpose is hereby granted without fee, provided that
    180     @@ -44,7 +44,7 @@
    181    
    182     switch (pGC->fillStyle) {
    183     case FillSolid:
    184     -#ifdef USE_GCC34_MMX
    185     +#ifdef USE_MMX
    186     if (!pPriv->and && fbHaveMMX())
    187     if (fbSolidFillmmx (pDrawable, x, y, width, height, pPriv->xor))
    188     return;
    189    
    190     diff -ur xc-orig/programs/Xserver/fb/fbmmx.c xc/programs/Xserver/fb/fbmmx.c
    191     --- xc-orig/programs/Xserver/fb/fbmmx.c 2005-02-11 04:00:50.006092570 -0500
    192     +++ xc/programs/Xserver/fb/fbmmx.c 2005-02-11 04:01:32.072346126 -0500
    193     @@ -1,5 +1,6 @@
    194     /*
    195     - * Copyright © 2004 Red Hat, Inc.
    196     + * Copyright © 2004 Red Hat, Inc.
    197     + * Copyright © 2004 Nicholas Miell
    198     *
    199     * Permission to use, copy, modify, distribute, and sell this software and its
    200     * documentation for any purpose is hereby granted without fee, provided that
    201     @@ -18,14 +19,23 @@
    202     * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
    203     * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
    204     *
    205     - * Author: Søren Sandmann (sandmann@redhat.com)
    206     - *
    207     + * Author: Søren Sandmann (sandmann@redhat.com)
    208     + * Minor Improvements: Nicholas Miell (nmiell@gmail.com)
    209     + *
    210     * Based on work by Owen Taylor
    211     */
    212    
    213     +
    214     +#ifdef USE_MMX
    215     +
    216     #include "fb.h"
    217     +#include "fbmmx.h"
    218     +
    219     +#include <mmintrin.h>
    220    
    221     -#ifdef USE_GCC34_MMX
    222     +#ifdef USE_SSE
    223     +#include <xmmintrin.h> /* for _mm_shuffle_pi16 and _MM_SHUFFLE */
    224     +#endif
    225    
    226     #ifdef RENDER
    227    
    228     @@ -33,11 +43,6 @@
    229     #include "mipict.h"
    230     #include "fbpict.h"
    231    
    232     -typedef int Vector1x64 __attribute__ ((mode(DI)));
    233     -typedef int Vector2x32 __attribute__ ((mode(V2SI)));
    234     -typedef int Vector4x16 __attribute__ ((mode(V4HI)));
    235     -typedef int Vector8x8 __attribute__ ((mode(V8QI)));
    236     -
    237     typedef unsigned long long ullong;
    238    
    239     #define noVERBOSE
    240     @@ -50,7 +55,6 @@
    241    
    242     typedef struct
    243     {
    244     - ullong mmx_zero;
    245     ullong mmx_4x00ff;
    246     ullong mmx_4x0080;
    247     ullong mmx_565_rgb;
    248     @@ -70,7 +74,6 @@
    249    
    250     static const MMXData c =
    251     {
    252     - .mmx_zero = 0x0000000000000000ULL,
    253     .mmx_4x00ff = 0x00ff00ff00ff00ffULL,
    254     .mmx_4x0080 = 0x0080008000800080ULL,
    255     .mmx_565_rgb = 0x000001f0003f001fULL,
    256     @@ -88,121 +91,112 @@
    257     .mmx_000000000000ffff = 0x000000000000ffffULL,
    258     };
    259    
    260     -static __inline__ Vector1x64
    261     -shift (Vector1x64 v, int s)
    262     +#define MC(x) ((__m64) c.mmx_##x)
    263     +
    264     +static __inline__ __m64
    265     +shift (__m64 v, int s)
    266     {
    267     if (s > 0)
    268     - return __builtin_ia32_psllq (v, s);
    269     + return _mm_slli_si64 (v, s);
    270     else if (s < 0)
    271     - return __builtin_ia32_psrlq (v, -s);
    272     + return _mm_srli_si64 (v, -s);
    273     else
    274     return v;
    275     }
    276    
    277     -static __inline__ Vector4x16
    278     -negate (Vector4x16 mask)
    279     +static __inline__ __m64
    280     +negate (__m64 mask)
    281     {
    282     - return (Vector4x16)__builtin_ia32_pxor (
    283     - (Vector1x64)mask,
    284     - (Vector1x64)c.mmx_4x00ff);
    285     + return _mm_xor_si64 (mask, MC(4x00ff));
    286     }
    287    
    288     -static __inline__ Vector4x16
    289     -pix_multiply (Vector4x16 a, Vector4x16 b)
    290     +static __inline__ __m64
    291     +pix_multiply (__m64 a, __m64 b)
    292     {
    293     - Vector4x16 res;
    294     + __m64 res;
    295    
    296     - res = __builtin_ia32_pmullw (a, b);
    297     - res = __builtin_ia32_paddw (res, (Vector4x16)c.mmx_4x0080);
    298     - res = __builtin_ia32_psrlw (res, 8);
    299     + res = _mm_mullo_pi16 (a, b);
    300     + res = _mm_add_pi16 (res, MC(4x0080));
    301     + res = _mm_srli_pi16 (res, 8);
    302    
    303     return res;
    304     }
    305    
    306     -#if 0
    307     +#ifdef USE_SSE
    308     #define HAVE_PSHUFW
    309     #endif
    310    
    311     #ifdef HAVE_PSHUFW
    312    
    313     -static __inline__ Vector4x16
    314     -expand_alpha (Vector4x16 pixel)
    315     +static __inline__ __m64
    316     +expand_alpha (__m64 pixel)
    317     {
    318     - Vector4x16 result;
    319     - __asm__ ("pshufw $0xFF, %1, %0\n\t" : "=y" (result) : "y" (pixel));
    320     - return result;
    321     + return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE(3, 3, 3, 3));
    322     }
    323    
    324     -static __inline__ Vector4x16
    325     -expand_alpha_rev (Vector4x16 pixel)
    326     +static __inline__ __m64
    327     +expand_alpha_rev (__m64 pixel)
    328     {
    329     - Vector4x16 result;
    330     - __asm__ ("pshufw $0x00, %1, %0\n\t" : "=y" (result) : "y" (pixel));
    331     - return result;
    332     + return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE(0, 0, 0, 0));
    333     }
    334    
    335     -static __inline__ Vector4x16
    336     -invert_colors (Vector4x16 pixel)
    337     +static __inline__ __m64
    338     +invert_colors (__m64 pixel)
    339     {
    340     - Vector4x16 result;
    341     -
    342     - /* 0xC6 = 11000110 */
    343     - /* 3 0 1 2 */
    344     -
    345     - __asm__ ("pshufw $0xC6, %1, %0\n\t" : "=y" (result) : "y" (pixel));
    346     -
    347     - return result;
    348     + return _mm_shuffle_pi16 (pixel, _MM_SHUFFLE(3, 0, 1, 2));
    349     }
    350    
    351     #else
    352    
    353     -static __inline__ Vector4x16
    354     -expand_alpha (Vector4x16 pixel)
    355     +static __inline__ __m64
    356     +expand_alpha (__m64 pixel)
    357     {
    358     - Vector1x64 t1, t2;
    359     -
    360     - t1 = shift ((Vector1x64)pixel, -48);
    361     + __m64 t1, t2;
    362     +
    363     + t1 = shift (pixel, -48);
    364     t2 = shift (t1, 16);
    365     - t1 = __builtin_ia32_por (t1, t2);
    366     + t1 = _mm_or_si64 (t1, t2);
    367     t2 = shift (t1, 32);
    368     - t1 = __builtin_ia32_por (t1, t2);
    369     -
    370     - return (Vector4x16)t1;
    371     + t1 = _mm_or_si64 (t1, t2);
    372     +
    373     + return t1;
    374     }
    375    
    376     -static __inline__ Vector4x16
    377     -expand_alpha_rev (Vector4x16 pixel)
    378     +static __inline__ __m64
    379     +expand_alpha_rev (__m64 pixel)
    380     {
    381     - Vector1x64 t1, t2;
    382     -
    383     - t1 = shift ((Vector1x64)pixel, 48);
    384     + __m64 t1, t2;
    385     +
    386     + /* move alpha to low 16 bits and zero the rest */
    387     + t1 = shift (pixel, 48);
    388     t1 = shift (t1, -48);
    389     +
    390     t2 = shift (t1, 16);
    391     - t1 = __builtin_ia32_por (t1, t2);
    392     + t1 = _mm_or_si64 (t1, t2);
    393     t2 = shift (t1, 32);
    394     - t1 = __builtin_ia32_por (t1, t2);
    395     -
    396     - return (Vector4x16)t1;
    397     + t1 = _mm_or_si64 (t1, t2);
    398     +
    399     + return t1;
    400     }
    401    
    402     -static __inline__ Vector4x16
    403     -invert_colors (Vector4x16 pixel)
    404     +static __inline__ __m64
    405     +invert_colors (__m64 pixel)
    406     {
    407     - Vector1x64 x, y, z;
    408     -
    409     - x = y = z = (Vector1x64)pixel;
    410     -
    411     - x = __builtin_ia32_pand (x, (Vector1x64)c.mmx_ffff0000ffff0000);
    412     - y = __builtin_ia32_pand (y, (Vector1x64)c.mmx_000000000000ffff);
    413     - z = __builtin_ia32_pand (z, (Vector1x64)c.mmx_0000ffff00000000);
    414     -
    415     + __m64 x, y, z;
    416     +
    417     + x = y = z = pixel;
    418     +
    419     + x = _mm_and_si64 (x, MC(ffff0000ffff0000));
    420     + y = _mm_and_si64 (y, MC(000000000000ffff));
    421     + z = _mm_and_si64 (z, MC(0000ffff00000000));
    422     +
    423     y = shift (y, 32);
    424     z = shift (z, -32);
    425     -
    426     - x = __builtin_ia32_por (x, y);
    427     - x = __builtin_ia32_por (x, z);
    428     -
    429     - return (Vector4x16)x;
    430     +
    431     + x = _mm_or_si64 (x, y);
    432     + x = _mm_or_si64 (x, z);
    433     +
    434     + return x;
    435     }
    436    
    437     #endif
    438     @@ -210,147 +204,138 @@
    439     /* Notes about writing mmx code
    440     *
    441     * give memory operands as the second operand. If you give it as the
    442     - * first, gcc will first load it into a register, then use that register
    443     + * first, gcc will first load it into a register, then use that
    444     + * register
    445     *
    446     * ie. use
    447     *
    448     - * __builtin_pmullw (x, mmx_constant[8]);
    449     + * _mm_mullo_pi16 (x, mmx_constant);
    450     *
    451     * not
    452     *
    453     - * __builtin_pmullw (mmx_constant[8], x);
    454     + * _mm_mullo_pi16 (mmx_constant, x);
    455     *
    456     - * Also try to minimize dependencies. Ie. when you need a value, try to calculate
    457     - * it from a value that was calculated as early as possible.
    458     + * Also try to minimize dependencies. i.e. when you need a value, try
    459     + * to calculate it from a value that was calculated as early as
    460     + * possible.
    461     */
    462    
    463     -static __inline__ Vector4x16
    464     -over (Vector4x16 src, Vector4x16 srca, Vector4x16 dest)
    465     +static __inline__ __m64
    466     +over (__m64 src, __m64 srca, __m64 dest)
    467     {
    468     - return (Vector4x16)__builtin_ia32_paddusb ((Vector8x8)src, (Vector8x8)pix_multiply(dest, negate(srca)));
    469     + return _mm_adds_pu8 (src, pix_multiply(dest, negate(srca)));
    470     }
    471    
    472     -static __inline__ Vector4x16
    473     -over_rev_non_pre (Vector4x16 src, Vector4x16 dest)
    474     +static __inline__ __m64
    475     +over_rev_non_pre (__m64 src, __m64 dest)
    476     {
    477     - Vector4x16 srca = expand_alpha (src);
    478     - Vector4x16 srcfaaa = (Vector4x16)__builtin_ia32_por((Vector1x64)srca, (Vector1x64)c.mmx_full_alpha);
    479     -
    480     + __m64 srca = expand_alpha (src);
    481     + __m64 srcfaaa = _mm_or_si64 (srca, MC(full_alpha));
    482     +
    483     return over(pix_multiply(invert_colors(src), srcfaaa), srca, dest);
    484     }
    485    
    486     -static __inline__ Vector4x16
    487     -in (Vector4x16 src,
    488     - Vector4x16 mask)
    489     +static __inline__ __m64
    490     +in (__m64 src,
    491     + __m64 mask)
    492     {
    493     return pix_multiply (src, mask);
    494     }
    495    
    496     -static __inline__ Vector4x16
    497     -in_over (Vector4x16 src,
    498     - Vector4x16 srca,
    499     - Vector4x16 mask,
    500     - Vector4x16 dest)
    501     +static __inline__ __m64
    502     +in_over (__m64 src,
    503     + __m64 srca,
    504     + __m64 mask,
    505     + __m64 dest)
    506     {
    507     return over(in(src, mask), pix_multiply(srca, mask), dest);
    508     }
    509    
    510     -static __inline__ Vector8x8
    511     -cvt32to64 (CARD32 v)
    512     -{
    513     - ullong r = v;
    514     - return (Vector8x8)r;
    515     -}
    516     -
    517     -static __inline__ Vector4x16
    518     +static __inline__ __m64
    519     load8888 (CARD32 v)
    520     {
    521     - return (Vector4x16)__builtin_ia32_punpcklbw (cvt32to64 (v),
    522     - (Vector8x8)c.mmx_zero);
    523     + return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (v), _mm_setzero_si64());
    524     }
    525    
    526     -static __inline__ Vector8x8
    527     -pack8888 (Vector4x16 lo, Vector4x16 hi)
    528     +static __inline__ __m64
    529     +pack8888 (__m64 lo, __m64 hi)
    530     {
    531     - Vector8x8 r;
    532     - r = __builtin_ia32_packuswb ((Vector4x16)lo, (Vector4x16)hi);
    533     + __m64 r;
    534     + r = _mm_packs_pu16 (lo, hi);
    535     return r;
    536     }
    537    
    538     -/* Expand 16 bits positioned at @pos (0-3) of a mmx register into 00RR00GG00BB
    539     -
    540     ---- Expanding 565 in the low word ---
    541     -
    542     -m = (m << (32 - 3)) | (m << (16 - 5)) | m;
    543     -m = m & (01f0003f001f);
    544     -m = m * (008404100840);
    545     -m = m >> 8;
    546     -
    547     -Note the trick here - the top word is shifted by another nibble to avoid
    548     -it bumping into the middle word
    549     -*/
    550     -static __inline__ Vector4x16
    551     -expand565 (Vector4x16 pixel, int pos)
    552     +/* Expand 16 bits positioned at @pos (0-3) of a mmx register into
    553     + *
    554     + * 00RR00GG00BB
    555     + *
    556     + * --- Expanding 565 in the low word ---
    557     + *
    558     + * m = (m << (32 - 3)) | (m << (16 - 5)) | m;
    559     + * m = m & (01f0003f001f);
    560     + * m = m * (008404100840);
    561     + * m = m >> 8;
    562     + *
    563     + * Note the trick here - the top word is shifted by another nibble to
    564     + * avoid it bumping into the middle word
    565     + */
    566     +static __inline__ __m64
    567     +expand565 (__m64 pixel, int pos)
    568     {
    569     - Vector1x64 p = (Vector1x64)pixel;
    570     + __m64 p = pixel;
    571     + __m64 t1, t2;
    572    
    573     /* move pixel to low 16 bit and zero the rest */
    574     p = shift (shift (p, (3 - pos) * 16), -48);
    575    
    576     - Vector1x64 t1 = shift (p, 36 - 11);
    577     - Vector1x64 t2 = shift (p, 16 - 5);
    578     + t1 = shift (p, 36 - 11);
    579     + t2 = shift (p, 16 - 5);
    580    
    581     - p = __builtin_ia32_por (t1, p);
    582     - p = __builtin_ia32_por (t2, p);
    583     - p = __builtin_ia32_pand (p, (Vector1x64)c.mmx_565_rgb);
    584     + p = _mm_or_si64 (t1, p);
    585     + p = _mm_or_si64 (t2, p);
    586     + p = _mm_and_si64 (p, MC(565_rgb));
    587    
    588     - pixel = __builtin_ia32_pmullw ((Vector4x16)p, (Vector4x16)c.mmx_565_unpack_multiplier);
    589     - return __builtin_ia32_psrlw (pixel, 8);
    590     + pixel = _mm_mullo_pi16 (p, MC(565_unpack_multiplier));
    591     + return _mm_srli_pi16 (pixel, 8);
    592     }
    593    
    594     -static __inline__ Vector4x16
    595     -expand8888 (Vector4x16 in, int pos)
    596     +static __inline__ __m64
    597     +expand8888 (__m64 in, int pos)
    598     {
    599     if (pos == 0)
    600     - return (Vector4x16)__builtin_ia32_punpcklbw ((Vector8x8)in, (Vector8x8)c.mmx_zero);
    601     + return _mm_unpacklo_pi8 (in, _mm_setzero_si64());
    602     else
    603     - return (Vector4x16)__builtin_ia32_punpckhbw ((Vector8x8)in, (Vector8x8)c.mmx_zero);
    604     + return _mm_unpackhi_pi8 (in, _mm_setzero_si64());
    605     }
    606    
    607     -static __inline__ Vector4x16
    608     -pack565 (Vector4x16 pixel, Vector4x16 target, int pos)
    609     +static __inline__ __m64
    610     +pack565 (__m64 pixel, __m64 target, int pos)
    611     {
    612     - Vector1x64 p = (Vector1x64)pixel;
    613     - Vector1x64 t = (Vector1x64)target;
    614     - Vector1x64 r, g, b;
    615     + __m64 p = pixel;
    616     + __m64 t = target;
    617     + __m64 r, g, b;
    618    
    619     - r = __builtin_ia32_pand (p, (Vector1x64)c.mmx_565_r);
    620     - g = __builtin_ia32_pand (p, (Vector1x64)c.mmx_565_g);
    621     - b = __builtin_ia32_pand (p, (Vector1x64)c.mmx_565_b);
    622     + r = _mm_and_si64 (p, MC(565_r));
    623     + g = _mm_and_si64 (p, MC(565_g));
    624     + b = _mm_and_si64 (p, MC(565_b));
    625    
    626     r = shift (r, - (32 - 8) + pos * 16);
    627     g = shift (g, - (16 - 3) + pos * 16);
    628     b = shift (b, - (0 + 3) + pos * 16);
    629     -
    630     +
    631     if (pos == 0)
    632     - t = __builtin_ia32_pand (t, (Vector1x64)c.mmx_mask_0);
    633     + t = _mm_and_si64 (t, MC(mask_0));
    634     else if (pos == 1)
    635     - t = __builtin_ia32_pand (t, (Vector1x64)c.mmx_mask_1);
    636     + t = _mm_and_si64 (t, MC(mask_1));
    637     else if (pos == 2)
    638     - t = __builtin_ia32_pand (t, (Vector1x64)c.mmx_mask_2);
    639     + t = _mm_and_si64 (t, MC(mask_2));
    640     else if (pos == 3)
    641     - t = __builtin_ia32_pand (t, (Vector1x64)c.mmx_mask_3);
    642     + t = _mm_and_si64 (t, MC(mask_3));
    643    
    644     - p = __builtin_ia32_por (r, t);
    645     - p = __builtin_ia32_por (g, p);
    646     + p = _mm_or_si64 (r, t);
    647     + p = _mm_or_si64 (g, p);
    648    
    649     - return (Vector4x16)__builtin_ia32_por (b, p);
    650     -}
    651     -
    652     -static __inline__ void
    653     -emms (void)
    654     -{
    655     - __asm__ __volatile__ ("emms");
    656     + return _mm_or_si64 (b, p);
    657     }
    658    
    659     void
    660     @@ -371,8 +356,8 @@
    661     CARD32 *dstLine, *dst;
    662     CARD16 w;
    663     FbStride dstStride;
    664     - Vector4x16 vsrc, vsrca;
    665     -
    666     + __m64 vsrc, vsrca;
    667     +
    668     CHECKPOINT();
    669    
    670     fbComposeGetSolid(pSrc, src, pDst->format);
    671     @@ -384,51 +369,52 @@
    672    
    673     vsrc = load8888 (src);
    674     vsrca = expand_alpha (vsrc);
    675     -
    676     +
    677     while (height--)
    678     {
    679     dst = dstLine;
    680     dstLine += dstStride;
    681     w = width;
    682     -
    683     +
    684     CHECKPOINT();
    685    
    686     while (w && (unsigned long)dst & 7)
    687     {
    688     - *dst = (ullong) pack8888(over(vsrc, vsrca, load8888(*dst)), (Vector4x16)c.mmx_zero);
    689     + *dst = (ullong) pack8888(over(vsrc, vsrca, load8888(*dst)),
    690     + _mm_setzero_si64());
    691    
    692     w--;
    693     dst++;
    694     }
    695     -
    696     +
    697     while (w >= 2)
    698     {
    699     - Vector4x16 vdest;
    700     - Vector4x16 dest0, dest1;
    701     -
    702     - vdest = *(Vector4x16 *)dst;
    703     + __m64 vdest;
    704     + __m64 dest0, dest1;
    705     +
    706     + vdest = *(__m64 *)dst;
    707    
    708     dest0 = over(vsrc, vsrca, expand8888(vdest, 0));
    709     dest1 = over(vsrc, vsrca, expand8888(vdest, 1));
    710    
    711     - *(Vector8x8 *)dst = (Vector8x8)pack8888(dest0, dest1);
    712     + *(__m64 *)dst = pack8888(dest0, dest1);
    713    
    714     dst += 2;
    715     w -= 2;
    716     }
    717     -
    718     +
    719     CHECKPOINT();
    720    
    721     while (w)
    722     {
    723     - *dst = (ullong) pack8888(over(vsrc, vsrca, load8888(*dst)), (Vector4x16)c.mmx_zero);
    724     + *dst = (ullong) pack8888(over(vsrc, vsrca, load8888(*dst)), _mm_setzero_si64());
    725    
    726     w--;
    727     dst++;
    728     }
    729     }
    730    
    731     - emms();
    732     + _mm_empty();
    733     }
    734    
    735     void
    736     @@ -449,8 +435,8 @@
    737     CARD16 *dstLine, *dst;
    738     CARD16 w;
    739     FbStride dstStride;
    740     - Vector4x16 vsrc, vsrca;
    741     -
    742     + __m64 vsrc, vsrca;
    743     +
    744     CHECKPOINT();
    745    
    746     fbComposeGetSolid(pSrc, src, pDst->format);
    747     @@ -462,49 +448,49 @@
    748    
    749     vsrc = load8888 (src);
    750     vsrca = expand_alpha (vsrc);
    751     -
    752     +
    753     while (height--)
    754     {
    755     dst = dstLine;
    756     dstLine += dstStride;
    757     w = width;
    758     -
    759     +
    760     CHECKPOINT();
    761    
    762     while (w && (unsigned long)dst & 7)
    763     {
    764     ullong d = *dst;
    765     - Vector4x16 vdest = expand565 ((Vector4x16)d, 0);
    766     + __m64 vdest = expand565 ((__m64)d, 0);
    767     vdest = pack565(over(vsrc, vsrca, vdest), vdest, 0);
    768     *dst = (ullong)vdest;
    769    
    770     w--;
    771     dst++;
    772     }
    773     -
    774     +
    775     while (w >= 4)
    776     {
    777     - Vector4x16 vdest;
    778     -
    779     - vdest = *(Vector4x16 *)dst;
    780     + __m64 vdest;
    781     +
    782     + vdest = *(__m64 *)dst;
    783    
    784     vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 0)), vdest, 0);
    785     vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 1)), vdest, 1);
    786     vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 2)), vdest, 2);
    787     vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 3)), vdest, 3);
    788    
    789     - *(Vector8x8 *)dst = (Vector8x8)vdest;
    790     + *(__m64 *)dst = vdest;
    791    
    792     dst += 4;
    793     w -= 4;
    794     }
    795     -
    796     +
    797     CHECKPOINT();
    798    
    799     while (w)
    800     {
    801     ullong d = *dst;
    802     - Vector4x16 vdest = expand565 ((Vector4x16)d, 0);
    803     + __m64 vdest = expand565 ((__m64)d, 0);
    804     vdest = pack565(over(vsrc, vsrca, vdest), vdest, 0);
    805     *dst = (ullong)vdest;
    806    
    807     @@ -513,7 +499,7 @@
    808     }
    809     }
    810    
    811     - emms();
    812     + _mm_empty();
    813     }
    814    
    815     void
    816     @@ -534,8 +520,8 @@
    817     CARD32 *dstLine;
    818     CARD32 *maskLine;
    819     FbStride dstStride, maskStride;
    820     - Vector4x16 vsrc, vsrca;
    821     -
    822     + __m64 vsrc, vsrca;
    823     +
    824     CHECKPOINT();
    825    
    826     fbComposeGetSolid(pSrc, src, pDst->format);
    827     @@ -562,9 +548,9 @@
    828    
    829     if (m)
    830     {
    831     - Vector4x16 vdest = load8888(*q);
    832     + __m64 vdest = load8888(*q);
    833     vdest = in_over(vsrc, vsrca, load8888(m), vdest);
    834     - *q = (ullong)pack8888(vdest, (Vector4x16)c.mmx_zero);
    835     + *q = (ullong)pack8888(vdest, _mm_setzero_si64());
    836     }
    837    
    838     twidth--;
    839     @@ -580,15 +566,15 @@
    840    
    841     if (m0 | m1)
    842     {
    843     - Vector4x16 dest0, dest1;
    844     - Vector4x16 vdest = *(Vector4x16 *)q;
    845     + __m64 dest0, dest1;
    846     + __m64 vdest = *(__m64 *)q;
    847    
    848     dest0 = in_over(vsrc, vsrca, load8888(m0),
    849     expand8888 (vdest, 0));
    850     dest1 = in_over(vsrc, vsrca, load8888(m1),
    851     expand8888 (vdest, 1));
    852    
    853     - *(Vector8x8 *)q = (Vector8x8)pack8888(dest0, dest1);
    854     + *(__m64 *)q = pack8888(dest0, dest1);
    855     }
    856    
    857     p += 2;
    858     @@ -602,9 +588,9 @@
    859    
    860     if (m)
    861     {
    862     - Vector4x16 vdest = load8888(*q);
    863     + __m64 vdest = load8888(*q);
    864     vdest = in_over(vsrc, vsrca, load8888(m), vdest);
    865     - *q = (ullong)pack8888(vdest, (Vector4x16)c.mmx_zero);
    866     + *q = (ullong)pack8888(vdest, _mm_setzero_si64());
    867     }
    868    
    869     twidth--;
    870     @@ -616,7 +602,133 @@
    871     maskLine += maskStride;
    872     }
    873    
    874     - emms();
    875     + _mm_empty();
    876     +}
    877     +
    878     +void
    879     +fbCompositeSrc_8888x8x8888mmx (CARD8 op,
    880     + PicturePtr pSrc,
    881     + PicturePtr pMask,
    882     + PicturePtr pDst,
    883     + INT16 xSrc,
    884     + INT16 ySrc,
    885     + INT16 xMask,
    886     + INT16 yMask,
    887     + INT16 xDst,
    888     + INT16 yDst,
    889     + CARD16 width,
    890     + CARD16 height)
    891     +{
    892     + CARD32 *dstLine, *dst;
    893     + CARD32 *srcLine, *src;
    894     + CARD8 *maskLine;
    895     + CARD32 mask;
    896     + __m64 vmask;
    897     + FbStride dstStride, srcStride, maskStride;
    898     + CARD16 w;
    899     + __m64 srca;
    900     +
    901     + CHECKPOINT();
    902     +
    903     + fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
    904     + fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
    905     + fbComposeGetStart (pMask, xMask, yMask, CARD8, maskStride, maskLine, 1);
    906     +
    907     + mask = *maskLine << 24 | *maskLine << 16 | *maskLine << 8 | *maskLine;
    908     + vmask = load8888 (mask);
    909     + srca = MC(4x00ff);
    910     +
    911     + while (height--)
    912     + {
    913     + dst = dstLine;
    914     + dstLine += dstStride;
    915     + src = srcLine;
    916     + srcLine += srcStride;
    917     + w = width;
    918     +
    919     + while (w && (unsigned long)dst & 7)
    920     + {
    921     + __m64 s = load8888 (*src);
    922     + __m64 d = load8888 (*dst);
    923     +
    924     + *dst = (ullong)pack8888 (in_over (s, srca, vmask, d), (__m64)_mm_setzero_si64());
    925     +
    926     + w--;
    927     + dst++;
    928     + src++;
    929     + }
    930     +
    931     + while (w >= 16)
    932     + {
    933     + __m64 vd0 = *(__m64 *)(dst + 0);
    934     + __m64 vd1 = *(__m64 *)(dst + 2);
    935     + __m64 vd2 = *(__m64 *)(dst + 4);
    936     + __m64 vd3 = *(__m64 *)(dst + 6);
    937     + __m64 vd4 = *(__m64 *)(dst + 8);
    938     + __m64 vd5 = *(__m64 *)(dst + 10);
    939     + __m64 vd6 = *(__m64 *)(dst + 12);
    940     + __m64 vd7 = *(__m64 *)(dst + 14);
    941     +
    942     + __m64 vs0 = *(__m64 *)(src + 0);
    943     + __m64 vs1 = *(__m64 *)(src + 2);
    944     + __m64 vs2 = *(__m64 *)(src + 4);
    945     + __m64 vs3 = *(__m64 *)(src + 6);
    946     + __m64 vs4 = *(__m64 *)(src + 8);
    947     + __m64 vs5 = *(__m64 *)(src + 10);
    948     + __m64 vs6 = *(__m64 *)(src + 12);
    949     + __m64 vs7 = *(__m64 *)(dst + 14);
    950     +
    951     + vd0 = (__m64)pack8888 (
    952     + in_over (expand8888 (vs0, 0), srca, vmask, expand8888 (vd0, 0)),
    953     + in_over (expand8888 (vs0, 1), srca, vmask, expand8888 (vd0, 1)));
    954     +
    955     + vd1 = (__m64)pack8888 (
    956     + in_over (expand8888 (vs1, 0), srca, vmask, expand8888 (vd1, 0)),
    957     + in_over (expand8888 (vs1, 1), srca, vmask, expand8888 (vd1, 1)));
    958     +
    959     + vd2 = (__m64)pack8888 (
    960     + in_over (expand8888 (vs2, 0), srca, vmask, expand8888 (vd2, 0)),
    961     + in_over (expand8888 (vs2, 1), srca, vmask, expand8888 (vd2, 1)));
    962     +
    963     + vd3 = (__m64)pack8888 (
    964     + in_over (expand8888 (vs3, 0), srca, vmask, expand8888 (vd3, 0)),
    965     + in_over (expand8888 (vs3, 1), srca, vmask, expand8888 (vd3, 1)));
    966     +
    967     + vd4 = (__m64)pack8888 (
    968     + in_over (expand8888 (vs4, 0), srca, vmask, expand8888 (vd4, 0)),
    969     + in_over (expand8888 (vs4, 1), srca, vmask, expand8888 (vd4, 1)));
    970     +
    971     + vd5 = (__m64)pack8888 (
    972     + in_over (expand8888 (vs5, 0), srca, vmask, expand8888 (vd5, 0)),
    973     + in_over (expand8888 (vs5, 1), srca, vmask, expand8888 (vd5, 1)));
    974     +
    975     + vd6 = (__m64)pack8888 (
    976     + in_over (expand8888 (vs6, 0), srca, vmask, expand8888 (vd6, 0)),
    977     + in_over (expand8888 (vs6, 1), srca, vmask, expand8888 (vd6, 1)));
    978     +
    979     + vd7 = (__m64)pack8888 (
    980     + in_over (expand8888 (vs7, 0), srca, vmask, expand8888 (vd7, 0)),
    981     + in_over (expand8888 (vs7, 1), srca, vmask, expand8888 (vd7, 1)));
    982     +
    983     + w -= 16;
    984     + dst += 16;
    985     + src += 16;
    986     + }
    987     +
    988     + while (w)
    989     + {
    990     + __m64 s = load8888 (*src);
    991     + __m64 d = load8888 (*dst);
    992     +
    993     + *dst = (ullong)pack8888 (in_over (s, srca, vmask, d), (__m64)_mm_setzero_si64());
    994     +
    995     + w--;
    996     + dst++;
    997     + src++;
    998     + }
    999     + }
    1000     +
    1001     + _mm_empty();
    1002     }
    1003    
    1004     void
    1005     @@ -638,7 +750,7 @@
    1006     CARD8 *maskLine, *mask;
    1007     FbStride dstStride, maskStride;
    1008     CARD16 w;
    1009     - Vector4x16 vsrc, vsrca;
    1010     + __m64 vsrc, vsrca;
    1011     ullong srcsrc;
    1012    
    1013     CHECKPOINT();
    1014     @@ -648,7 +760,7 @@
    1015     srca = src >> 24;
    1016     if (srca == 0)
    1017     return;
    1018     -
    1019     +
    1020     srcsrc = (unsigned long long)src << 32 | src;
    1021    
    1022     fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
    1023     @@ -664,7 +776,7 @@
    1024     mask = maskLine;
    1025     maskLine += maskStride;
    1026     w = width;
    1027     -
    1028     +
    1029     CHECKPOINT();
    1030    
    1031     while (w && (unsigned long)dst & 7)
    1032     @@ -673,15 +785,15 @@
    1033    
    1034     if (m)
    1035     {
    1036     - Vector4x16 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((Vector4x16)m), load8888(*dst));
    1037     - *dst = (ullong)pack8888(vdest, (Vector4x16)c.mmx_zero);
    1038     + __m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), load8888(*dst));
    1039     + *dst = (ullong)pack8888(vdest, _mm_setzero_si64());
    1040     }
    1041    
    1042     w--;
    1043     mask++;
    1044     dst++;
    1045     }
    1046     -
    1047     +
    1048     CHECKPOINT();
    1049    
    1050     while (w >= 2)
    1051     @@ -689,29 +801,29 @@
    1052     ullong m0, m1;
    1053     m0 = *mask;
    1054     m1 = *(mask + 1);
    1055     -
    1056     +
    1057     if (srca == 0xff && (m0 & m1) == 0xff)
    1058     {
    1059     *(unsigned long long *)dst = srcsrc;
    1060     }
    1061     else if (m0 | m1)
    1062     {
    1063     - Vector4x16 vdest;
    1064     - Vector4x16 dest0, dest1;
    1065     -
    1066     - vdest = *(Vector4x16 *)dst;
    1067     + __m64 vdest;
    1068     + __m64 dest0, dest1;
    1069     +
    1070     + vdest = *(__m64 *)dst;
    1071    
    1072     - dest0 = in_over(vsrc, vsrca, expand_alpha_rev ((Vector4x16)m0), expand8888(vdest, 0));
    1073     - dest1 = in_over(vsrc, vsrca, expand_alpha_rev ((Vector4x16)m1), expand8888(vdest, 1));
    1074     + dest0 = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m0), expand8888(vdest, 0));
    1075     + dest1 = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m1), expand8888(vdest, 1));
    1076    
    1077     - *(Vector8x8 *)dst = (Vector8x8)pack8888(dest0, dest1);
    1078     + *(__m64 *)dst = pack8888(dest0, dest1);
    1079     }
    1080    
    1081     mask += 2;
    1082     dst += 2;
    1083     w -= 2;
    1084     }
    1085     -
    1086     +
    1087     CHECKPOINT();
    1088    
    1089     while (w)
    1090     @@ -720,9 +832,9 @@
    1091    
    1092     if (m)
    1093     {
    1094     - Vector4x16 vdest = load8888(*dst);
    1095     - vdest = in_over(vsrc, vsrca, expand_alpha_rev ((Vector4x16)m), vdest);
    1096     - *dst = (ullong)pack8888(vdest, (Vector4x16)c.mmx_zero);
    1097     + __m64 vdest = load8888(*dst);
    1098     + vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), vdest);
    1099     + *dst = (ullong)pack8888(vdest, _mm_setzero_si64());
    1100     }
    1101    
    1102     w--;
    1103     @@ -731,7 +843,7 @@
    1104     }
    1105     }
    1106    
    1107     - emms();
    1108     + _mm_empty();
    1109     }
    1110    
    1111    
    1112     @@ -754,7 +866,7 @@
    1113     CARD8 *maskLine, *mask;
    1114     FbStride dstStride, maskStride;
    1115     CARD16 w;
    1116     - Vector4x16 vsrc, vsrca;
    1117     + __m64 vsrc, vsrca;
    1118     unsigned long long srcsrcsrcsrc, src16;
    1119    
    1120     CHECKPOINT();
    1121     @@ -770,9 +882,9 @@
    1122    
    1123     vsrc = load8888 (src);
    1124     vsrca = expand_alpha (vsrc);
    1125     -
    1126     - src16 = (ullong)pack565(vsrc, (Vector4x16)c.mmx_zero, 0);
    1127     -
    1128     +
    1129     + src16 = (ullong)pack565(vsrc, _mm_setzero_si64(), 0);
    1130     +
    1131     srcsrcsrcsrc = (ullong)src16 << 48 | (ullong)src16 << 32 |
    1132     (ullong)src16 << 16 | (ullong)src16;
    1133    
    1134     @@ -783,7 +895,7 @@
    1135     mask = maskLine;
    1136     maskLine += maskStride;
    1137     w = width;
    1138     -
    1139     +
    1140     CHECKPOINT();
    1141    
    1142     while (w && (unsigned long)dst & 7)
    1143     @@ -793,16 +905,16 @@
    1144     if (m)
    1145     {
    1146     ullong d = *dst;
    1147     - Vector4x16 vd = (Vector4x16)d;
    1148     - Vector4x16 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((Vector4x16)m), expand565(vd, 0));
    1149     - *dst = (ullong)pack565(vdest, (Vector4x16)c.mmx_zero, 0);
    1150     + __m64 vd = (__m64)d;
    1151     + __m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), expand565(vd, 0));
    1152     + *dst = (ullong)pack565(vdest, _mm_setzero_si64(), 0);
    1153     }
    1154    
    1155     w--;
    1156     mask++;
    1157     dst++;
    1158     }
    1159     -
    1160     +
    1161     CHECKPOINT();
    1162    
    1163     while (w >= 4)
    1164     @@ -812,35 +924,35 @@
    1165     m1 = *(mask + 1);
    1166     m2 = *(mask + 2);
    1167     m3 = *(mask + 3);
    1168     -
    1169     +
    1170     if (srca == 0xff && (m0 & m1 & m2 & m3) == 0xff)
    1171     {
    1172     *(unsigned long long *)dst = srcsrcsrcsrc;
    1173     }
    1174     else if (m0 | m1 | m2 | m3)
    1175     {
    1176     - Vector4x16 vdest;
    1177     - Vector4x16 vm0, vm1, vm2, vm3;
    1178     -
    1179     - vdest = *(Vector4x16 *)dst;
    1180     -
    1181     - vm0 = (Vector4x16)m0;
    1182     + __m64 vdest;
    1183     + __m64 vm0, vm1, vm2, vm3;
    1184     +
    1185     + vdest = *(__m64 *)dst;
    1186     +
    1187     + vm0 = (__m64)m0;
    1188     vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm0), expand565(vdest, 0)), vdest, 0);
    1189     - vm1 = (Vector4x16)m1;
    1190     + vm1 = (__m64)m1;
    1191     vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm1), expand565(vdest, 1)), vdest, 1);
    1192     - vm2 = (Vector4x16)m2;
    1193     + vm2 = (__m64)m2;
    1194     vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm2), expand565(vdest, 2)), vdest, 2);
    1195     - vm3 = (Vector4x16)m3;
    1196     + vm3 = (__m64)m3;
    1197     vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm3), expand565(vdest, 3)), vdest, 3);
    1198    
    1199     - *(Vector4x16 *)dst = vdest;
    1200     + *(__m64 *)dst = vdest;
    1201     }
    1202    
    1203     w -= 4;
    1204     mask += 4;
    1205     dst += 4;
    1206     }
    1207     -
    1208     +
    1209     CHECKPOINT();
    1210    
    1211     while (w)
    1212     @@ -850,9 +962,9 @@
    1213     if (m)
    1214     {
    1215     ullong d = *dst;
    1216     - Vector4x16 vd = (Vector4x16)d;
    1217     - Vector4x16 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((Vector4x16)m), expand565(vd, 0));
    1218     - *dst = (ullong)pack565(vdest, (Vector4x16)c.mmx_zero, 0);
    1219     + __m64 vd = (__m64)d;
    1220     + __m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev ((__m64)m), expand565(vd, 0));
    1221     + *dst = (ullong)pack565(vdest, _mm_setzero_si64(), 0);
    1222     }
    1223    
    1224     w--;
    1225     @@ -861,7 +973,7 @@
    1226     }
    1227     }
    1228    
    1229     - emms();
    1230     + _mm_empty();
    1231     }
    1232    
    1233     void
    1234     @@ -887,9 +999,9 @@
    1235    
    1236     fbComposeGetStart (pDst, xDst, yDst, CARD16, dstStride, dstLine, 1);
    1237     fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
    1238     -
    1239     +
    1240     assert (pSrc->pDrawable == pMask->pDrawable);
    1241     -
    1242     +
    1243     while (height--)
    1244     {
    1245     dst = dstLine;
    1246     @@ -897,14 +1009,14 @@
    1247     src = srcLine;
    1248     srcLine += srcStride;
    1249     w = width;
    1250     -
    1251     +
    1252     CHECKPOINT();
    1253    
    1254     while (w && (unsigned long)dst & 7)
    1255     {
    1256     - Vector4x16 vsrc = load8888 (*src);
    1257     + __m64 vsrc = load8888 (*src);
    1258     ullong d = *dst;
    1259     - Vector4x16 vdest = expand565 ((Vector4x16)d, 0);
    1260     + __m64 vdest = expand565 ((__m64)d, 0);
    1261    
    1262     vdest = pack565(over_rev_non_pre(vsrc, vdest), vdest, 0);
    1263    
    1264     @@ -914,19 +1026,19 @@
    1265     dst++;
    1266     src++;
    1267     }
    1268     -
    1269     +
    1270     CHECKPOINT();
    1271    
    1272     while (w >= 4)
    1273     {
    1274     CARD32 s0, s1, s2, s3;
    1275     unsigned char a0, a1, a2, a3;
    1276     -
    1277     +
    1278     s0 = *src;
    1279     s1 = *(src + 1);
    1280     s2 = *(src + 2);
    1281     s3 = *(src + 3);
    1282     -
    1283     +
    1284     a0 = (s0 >> 24);
    1285     a1 = (s1 >> 24);
    1286     a2 = (s2 >> 24);
    1287     @@ -934,38 +1046,38 @@
    1288    
    1289     if ((a0 & a1 & a2 & a3) == 0xFF)
    1290     {
    1291     - Vector4x16 vdest;
    1292     - vdest = pack565(invert_colors(load8888(s0)), (Vector4x16)c.mmx_zero, 0);
    1293     + __m64 vdest;
    1294     + vdest = pack565(invert_colors(load8888(s0)), _mm_setzero_si64(), 0);
    1295     vdest = pack565(invert_colors(load8888(s1)), vdest, 1);
    1296     vdest = pack565(invert_colors(load8888(s2)), vdest, 2);
    1297     vdest = pack565(invert_colors(load8888(s3)), vdest, 3);
    1298     -
    1299     - *(Vector4x16 *)dst = vdest;
    1300     +
    1301     + *(__m64 *)dst = vdest;
    1302     }
    1303     else if (a0 | a1 | a2 | a3)
    1304     {
    1305     - Vector4x16 vdest = *(Vector4x16 *)dst;
    1306     -
    1307     + __m64 vdest = *(__m64 *)dst;
    1308     +
    1309     vdest = pack565(over_rev_non_pre(load8888(s0), expand565(vdest, 0)), vdest, 0);
    1310     vdest = pack565(over_rev_non_pre(load8888(s1), expand565(vdest, 1)), vdest, 1);
    1311     vdest = pack565(over_rev_non_pre(load8888(s2), expand565(vdest, 2)), vdest, 2);
    1312     vdest = pack565(over_rev_non_pre(load8888(s3), expand565(vdest, 3)), vdest, 3);
    1313     -
    1314     - *(Vector4x16 *)dst = vdest;
    1315     +
    1316     + *(__m64 *)dst = vdest;
    1317     }
    1318    
    1319     w -= 4;
    1320     dst += 4;
    1321     src += 4;
    1322     }
    1323     -
    1324     +
    1325     CHECKPOINT();
    1326    
    1327     while (w)
    1328     {
    1329     - Vector4x16 vsrc = load8888 (*src);
    1330     + __m64 vsrc = load8888 (*src);
    1331     ullong d = *dst;
    1332     - Vector4x16 vdest = expand565 ((Vector4x16)d, 0);
    1333     + __m64 vdest = expand565 ((__m64)d, 0);
    1334    
    1335     vdest = pack565(over_rev_non_pre(vsrc, vdest), vdest, 0);
    1336    
    1337     @@ -976,11 +1088,11 @@
    1338     src++;
    1339     }
    1340     }
    1341     -
    1342     - emms();
    1343     +
    1344     + _mm_empty();
    1345     }
    1346    
    1347     -/* "888RevNP" is GdkPixbuf's format: ABGR, non premultiplied */
    1348     +/* "8888RevNP" is GdkPixbuf's format: ABGR, non premultiplied */
    1349    
    1350     void
    1351     fbCompositeSrc_8888RevNPx8888mmx (CARD8 op,
    1352     @@ -1005,9 +1117,9 @@
    1353    
    1354     fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
    1355     fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
    1356     -
    1357     +
    1358     assert (pSrc->pDrawable == pMask->pDrawable);
    1359     -
    1360     +
    1361     while (height--)
    1362     {
    1363     dst = dstLine;
    1364     @@ -1015,28 +1127,28 @@
    1365     src = srcLine;
    1366     srcLine += srcStride;
    1367     w = width;
    1368     -
    1369     +
    1370     while (w && (unsigned long)dst & 7)
    1371     {
    1372     - Vector4x16 s = load8888 (*src);
    1373     - Vector4x16 d = load8888 (*dst);
    1374     + __m64 s = load8888 (*src);
    1375     + __m64 d = load8888 (*dst);
    1376    
    1377     - *dst = (ullong)pack8888 (over_rev_non_pre (s, d), (Vector4x16)c.mmx_zero);
    1378     + *dst = (ullong)pack8888 (over_rev_non_pre (s, d), _mm_setzero_si64());
    1379    
    1380     w--;
    1381     dst++;
    1382     src++;
    1383     }
    1384     -
    1385     +
    1386     while (w >= 2)
    1387     {
    1388     ullong s0, s1;
    1389     unsigned char a0, a1;
    1390     - Vector4x16 d0, d1;
    1391     -
    1392     + __m64 d0, d1;
    1393     +
    1394     s0 = *src;
    1395     s1 = *(src + 1);
    1396     -
    1397     +
    1398     a0 = (s0 >> 24);
    1399     a1 = (s1 >> 24);
    1400    
    1401     @@ -1044,17 +1156,17 @@
    1402     {
    1403     d0 = invert_colors(load8888(s0));
    1404     d1 = invert_colors(load8888(s1));
    1405     -
    1406     - *(Vector8x8 *)dst = pack8888 (d0, d1);
    1407     +
    1408     + *(__m64 *)dst = pack8888 (d0, d1);
    1409     }
    1410     else if (a0 | a1)
    1411     {
    1412     - Vector4x16 vdest = *(Vector4x16 *)dst;
    1413     -
    1414     + __m64 vdest = *(__m64 *)dst;
    1415     +
    1416     d0 = over_rev_non_pre (load8888(s0), expand8888 (vdest, 0));
    1417     d1 = over_rev_non_pre (load8888(s1), expand8888 (vdest, 1));
    1418     -
    1419     - *(Vector8x8 *)dst = pack8888 (d0, d1);
    1420     +
    1421     + *(__m64 *)dst = pack8888 (d0, d1);
    1422     }
    1423    
    1424     w -= 2;
    1425     @@ -1064,18 +1176,18 @@
    1426    
    1427     while (w)
    1428     {
    1429     - Vector4x16 s = load8888 (*src);
    1430     - Vector4x16 d = load8888 (*dst);
    1431     + __m64 s = load8888 (*src);
    1432     + __m64 d = load8888 (*dst);
    1433    
    1434     - *dst = (ullong)pack8888 (over_rev_non_pre (s, d), (Vector4x16)c.mmx_zero);
    1435     + *dst = (ullong)pack8888 (over_rev_non_pre (s, d), _mm_setzero_si64());
    1436    
    1437     w--;
    1438     dst++;
    1439     src++;
    1440     }
    1441     }
    1442     -
    1443     - emms();
    1444     +
    1445     + _mm_empty();
    1446     }
    1447    
    1448     void
    1449     @@ -1096,7 +1208,7 @@
    1450     CARD16 *dstLine;
    1451     CARD32 *maskLine;
    1452     FbStride dstStride, maskStride;
    1453     - Vector4x16 vsrc, vsrca;
    1454     + __m64 vsrc, vsrca;
    1455    
    1456     CHECKPOINT();
    1457    
    1458     @@ -1125,7 +1237,7 @@
    1459     if (m)
    1460     {
    1461     ullong d = *q;
    1462     - Vector4x16 vdest = expand565 ((Vector4x16)d, 0);
    1463     + __m64 vdest = expand565 ((__m64)d, 0);
    1464     vdest = pack565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
    1465     *q = (ullong)vdest;
    1466     }
    1467     @@ -1146,14 +1258,14 @@
    1468    
    1469     if ((m0 | m1 | m2 | m3))
    1470     {
    1471     - Vector4x16 vdest = *(Vector4x16 *)q;
    1472     + __m64 vdest = *(__m64 *)q;
    1473    
    1474     vdest = pack565(in_over(vsrc, vsrca, load8888(m0), expand565(vdest, 0)), vdest, 0);
    1475     vdest = pack565(in_over(vsrc, vsrca, load8888(m1), expand565(vdest, 1)), vdest, 1);
    1476     vdest = pack565(in_over(vsrc, vsrca, load8888(m2), expand565(vdest, 2)), vdest, 2);
    1477     vdest = pack565(in_over(vsrc, vsrca, load8888(m3), expand565(vdest, 3)), vdest, 3);
    1478    
    1479     - *(Vector4x16 *)q = vdest;
    1480     + *(__m64 *)q = vdest;
    1481     }
    1482     twidth -= 4;
    1483     p += 4;
    1484     @@ -1168,7 +1280,7 @@
    1485     if (m)
    1486     {
    1487     ullong d = *q;
    1488     - Vector4x16 vdest = expand565((Vector4x16)d, 0);
    1489     + __m64 vdest = expand565((__m64)d, 0);
    1490     vdest = pack565 (in_over(vsrc, vsrca, load8888(m), vdest), vdest, 0);
    1491     *q = (ullong)vdest;
    1492     }
    1493     @@ -1182,7 +1294,7 @@
    1494     dstLine += dstStride;
    1495     }
    1496    
    1497     - emms ();
    1498     + _mm_empty ();
    1499     }
    1500    
    1501     void
    1502     @@ -1210,7 +1322,7 @@
    1503    
    1504     fbComposeGetStart (pSrc, xSrc, ySrc, CARD8, srcStride, srcLine, 1);
    1505     fbComposeGetStart (pDst, xDst, yDst, CARD8, dstStride, dstLine, 1);
    1506     -
    1507     +
    1508     while (height--)
    1509     {
    1510     dst = dstLine;
    1511     @@ -1218,7 +1330,7 @@
    1512     src = srcLine;
    1513     srcLine += srcStride;
    1514     w = width;
    1515     -
    1516     +
    1517     while (w && (unsigned long)dst & 7)
    1518     {
    1519     s = *src;
    1520     @@ -1234,13 +1346,7 @@
    1521    
    1522     while (w >= 8)
    1523     {
    1524     - __asm__ __volatile__ (
    1525     - "movq (%0), %%mm2\n\t"
    1526     - "movq (%1), %%mm3\n\t"
    1527     - "paddusb %%mm2, %%mm3\n\t"
    1528     - "movq %%mm3, (%1)\n\t"
    1529     - : /* no output */ : "r" (src), "r" (dst));
    1530     -
    1531     + *(__m64*)dst = _mm_adds_pu8(*(__m64*)src, *(__m64*)dst);
    1532     dst += 8;
    1533     src += 8;
    1534     w -= 8;
    1535     @@ -1259,8 +1365,8 @@
    1536     w--;
    1537     }
    1538     }
    1539     -
    1540     - emms();
    1541     +
    1542     + _mm_empty();
    1543     }
    1544    
    1545     void
    1546     @@ -1297,13 +1403,8 @@
    1547    
    1548     while (w && (unsigned long)dst & 7)
    1549     {
    1550     - __asm__ __volatile__ (
    1551     - "movd %0, %%mm2\n\t"
    1552     - "movd %1, %%mm3\n\t"
    1553     - "paddusb %%mm2, %%mm3\n\t"
    1554     - "movd %%mm3, %1\n\t"
    1555     - : /* no output */ : "m" (*src), "m" (*dst));
    1556     -
    1557     + *dst = _mm_cvtsi64_si32(_mm_adds_pu8(_mm_cvtsi32_si64(*src),
    1558     + _mm_cvtsi32_si64(*dst)));
    1559     dst++;
    1560     src++;
    1561     w--;
    1562     @@ -1311,13 +1412,7 @@
    1563    
    1564     while (w >= 2)
    1565     {
    1566     - __asm__ __volatile__ (
    1567     - "movq (%0), %%mm2\n\t"
    1568     - "movq (%1), %%mm3\n\t"
    1569     - "paddusb %%mm2, %%mm3\n\t"
    1570     - "movq %%mm3, (%1)\n\t"
    1571     - : /* no output */ : "r" (src), "r" (dst));
    1572     -
    1573     + *(ullong*)dst = (ullong) _mm_adds_pu8(*(__m64*)src, *(__m64*)dst);
    1574     dst += 2;
    1575     src += 2;
    1576     w -= 2;
    1577     @@ -1325,16 +1420,13 @@
    1578    
    1579     if (w)
    1580     {
    1581     - __asm__ __volatile__ (
    1582     - "movd %0, %%mm2\n\t"
    1583     - "movd %1, %%mm3\n\t"
    1584     - "paddusb %%mm2, %%mm3\n\t"
    1585     - "movd %%mm3, %1\n\t"
    1586     - : /* no output */ : "m" (*src), "m" (*dst));
    1587     + *dst = _mm_cvtsi64_si32(_mm_adds_pu8(_mm_cvtsi32_si64(*src),
    1588     + _mm_cvtsi32_si64(*dst)));
    1589     +
    1590     }
    1591     }
    1592     -
    1593     - emms();
    1594     +
    1595     + _mm_empty();
    1596     }
    1597    
    1598     #define GetStart(drw,x,y,type,stride,line,bpp) {\
    1599     @@ -1358,19 +1450,19 @@
    1600     FbStride stride;
    1601     int bpp;
    1602     ullong fill;
    1603     - Vector8x8 vfill;
    1604     + __m64 vfill;
    1605     CARD32 byte_width;
    1606     CARD8 *byte_line;
    1607     FbBits *bits;
    1608     int xoff, yoff;
    1609    
    1610     CHECKPOINT();
    1611     -
    1612     +
    1613     fbGetDrawable(pDraw, bits, stride, bpp, xoff, yoff);
    1614     -
    1615     +
    1616     if (bpp == 16 && (xor >> 16 != (xor & 0xffff)))
    1617     return FALSE;
    1618     -
    1619     +
    1620     if (bpp != 16 && bpp != 32)
    1621     return FALSE;
    1622    
    1623     @@ -1388,9 +1480,9 @@
    1624     byte_width = 4 * width;
    1625     stride *= 4;
    1626     }
    1627     -
    1628     +
    1629     fill = ((ullong)xor << 32) | xor;
    1630     - vfill = (Vector8x8)fill;
    1631     + vfill = (__m64)fill;
    1632    
    1633     while (height--)
    1634     {
    1635     @@ -1398,7 +1490,7 @@
    1636     CARD8 *d = byte_line;
    1637     byte_line += stride;
    1638     w = byte_width;
    1639     -
    1640     +
    1641     while (w >= 2 && ((unsigned long)d & 3))
    1642     {
    1643     *(CARD16 *)d = xor;
    1644     @@ -1406,35 +1498,32 @@
    1645     d += 2;
    1646     }
    1647    
    1648     - while (w >= 4 && ((unsigned int)d & 7))
    1649     + while (w >= 4 && ((unsigned long)d & 7))
    1650     {
    1651     *(CARD32 *)d = xor;
    1652     -
    1653     +
    1654     w -= 4;
    1655     d += 4;
    1656     }
    1657    
    1658     while (w >= 64)
    1659     {
    1660     - __asm__ __volatile (
    1661     - "movq %0, (%1)\n\t"
    1662     - "movq %0, 8(%1)\n\t"
    1663     - "movq %0, 16(%1)\n\t"
    1664     - "movq %0, 24(%1)\n\t"
    1665     - "movq %0, 32(%1)\n\t"
    1666     - "movq %0, 40(%1)\n\t"
    1667     - "movq %0, 48(%1)\n\t"
    1668     - "movq %0, 56(%1)\n\t"
    1669     - : /* no output */
    1670     - : "y" (vfill), "r" (d)
    1671     - : "memory");
    1672     + *(__m64*) (d + 0) = vfill;
    1673     + *(__m64*) (d + 8) = vfill;
    1674     + *(__m64*) (d + 16) = vfill;
    1675     + *(__m64*) (d + 24) = vfill;
    1676     + *(__m64*) (d + 32) = vfill;
    1677     + *(__m64*) (d + 40) = vfill;
    1678     + *(__m64*) (d + 48) = vfill;
    1679     + *(__m64*) (d + 56) = vfill;
    1680     +
    1681     w -= 64;
    1682     d += 64;
    1683     }
    1684     while (w >= 4)
    1685     {
    1686     *(CARD32 *)d = xor;
    1687     -
    1688     +
    1689     w -= 4;
    1690     d += 4;
    1691     }
    1692     @@ -1446,16 +1535,160 @@
    1693     }
    1694     }
    1695    
    1696     - emms();
    1697     + _mm_empty();
    1698     + return TRUE;
    1699     +}
    1700     +
    1701     +Bool
    1702     +fbCopyAreammx (DrawablePtr pSrc,
    1703     + DrawablePtr pDst,
    1704     + int src_x,
    1705     + int src_y,
    1706     + int dst_x,
    1707     + int dst_y,
    1708     + int width,
    1709     + int height)
    1710     +{
    1711     + FbBits * src_bits;
    1712     + FbStride src_stride;
    1713     + int src_bpp;
    1714     + int src_xoff;
    1715     + int src_yoff;
    1716     +
    1717     + FbBits * dst_bits;
    1718     + FbStride dst_stride;
    1719     + int dst_bpp;
    1720     + int dst_xoff;
    1721     + int dst_yoff;
    1722     +
    1723     + CARD8 * src_bytes;
    1724     + CARD8 * dst_bytes;
    1725     + int byte_width;
    1726     +
    1727     + fbGetDrawable(pSrc, src_bits, src_stride, src_bpp, src_xoff, src_yoff);
    1728     + fbGetDrawable(pDst, dst_bits, dst_stride, dst_bpp, dst_xoff, dst_yoff);
    1729     +
    1730     + if (src_bpp != 16 && src_bpp != 32)
    1731     + return FALSE;
    1732     +
    1733     + if (dst_bpp != 16 && dst_bpp != 32)
    1734     + return FALSE;
    1735     +
    1736     + if (src_bpp != dst_bpp)
    1737     + {
    1738     + return FALSE;
    1739     + }
    1740     +
    1741     + if (src_bpp == 16)
    1742     + {
    1743     + src_stride = src_stride * sizeof (FbBits) / 2;
    1744     + dst_stride = dst_stride * sizeof (FbBits) / 2;
    1745     + src_bytes = (CARD8 *)(((CARD16 *)src_bits) + src_stride * (src_y - src_yoff) + (src_x - src_xoff));
    1746     + dst_bytes = (CARD8 *)(((CARD16 *)dst_bits) + dst_stride * (dst_y - dst_yoff) + (dst_x - dst_xoff));
    1747     + byte_width = 2 * width;
    1748     + src_stride *= 2;
    1749     + dst_stride *= 2;
    1750     + }
    1751     + else
    1752     + {
    1753     + src_stride = src_stride * sizeof (FbBits) / 4;
    1754     + dst_stride = dst_stride * sizeof (FbBits) / 4;
    1755     + src_bytes = (CARD8 *)(((CARD32 *)src_bits) + src_stride * (src_y - src_yoff) + (src_x - src_xoff));
    1756     + dst_bytes = (CARD8 *)(((CARD32 *)dst_bits) + dst_stride * (dst_y - dst_yoff) + (dst_x - dst_xoff));
    1757     + byte_width = 4 * width;
    1758     + src_stride *= 4;
    1759     + dst_stride *= 4;
    1760     + }
    1761     +
    1762     + while (height--)
    1763     + {
    1764     + int w;
    1765     + CARD8 *s = src_bytes;
    1766     + CARD8 *d = dst_bytes;
    1767     + src_bytes += src_stride;
    1768     + dst_bytes += dst_stride;
    1769     + w = byte_width;
    1770     +
    1771     + while (w >= 2 && ((unsigned long)d & 3))
    1772     + {
    1773     + *(CARD16 *)d = *(CARD16 *)s;
    1774     + w -= 2;
    1775     + s += 2;
    1776     + d += 2;
    1777     + }
    1778     +
    1779     + while (w >= 4 && ((unsigned int)d & 7))
    1780     + {
    1781     + *(CARD32 *)d = *(CARD32 *)s;
    1782     +
    1783     + w -= 4;
    1784     + s += 4;
    1785     + d += 4;
    1786     + }
    1787     +
    1788     + while (w >= 64)
    1789     + {
    1790     + *(__m64 *)(d + 0) = *(__m64 *)(s + 0);
    1791     + *(__m64 *)(d + 8) = *(__m64 *)(s + 8);
    1792     + *(__m64 *)(d + 16) = *(__m64 *)(s + 16);
    1793     + *(__m64 *)(d + 24) = *(__m64 *)(s + 24);
    1794     + *(__m64 *)(d + 32) = *(__m64 *)(s + 32);
    1795     + *(__m64 *)(d + 40) = *(__m64 *)(s + 40);
    1796     + *(__m64 *)(d + 48) = *(__m64 *)(s + 48);
    1797     + *(__m64 *)(d + 56) = *(__m64 *)(s + 56);
    1798     + w -= 64;
    1799     + s += 64;
    1800     + d += 64;
    1801     + }
    1802     + while (w >= 4)
    1803     + {
    1804     + *(CARD32 *)d = *(CARD32 *)s;
    1805     +
    1806     + w -= 4;
    1807     + s += 4;
    1808     + d += 4;
    1809     + }
    1810     + if (w >= 2)
    1811     + {
    1812     + *(CARD16 *)d = *(CARD16 *)s;
    1813     + w -= 2;
    1814     + s += 2;
    1815     + d += 2;
    1816     + }
    1817     + }
    1818     +
    1819     + _mm_empty();
    1820     return TRUE;
    1821     }
    1822    
    1823     +void
    1824     +fbCompositeCopyAreammx (CARD8 op,
    1825     + PicturePtr pSrc,
    1826     + PicturePtr pMask,
    1827     + PicturePtr pDst,
    1828     + INT16 xSrc,
    1829     + INT16 ySrc,
    1830     + INT16 xMask,
    1831     + INT16 yMask,
    1832     + INT16 xDst,
    1833     + INT16 yDst,
    1834     + CARD16 width,
    1835     + CARD16 height)
    1836     +{
    1837     + fbCopyAreammx (pSrc->pDrawable,
    1838     + pDst->pDrawable,
    1839     + xSrc, ySrc,
    1840     + xDst, yDst,
    1841     + width, height);
    1842     +}
    1843     +
    1844     +#ifndef __amd64__
    1845     Bool
    1846     fbHaveMMX (void)
    1847     {
    1848     static Bool initialized = FALSE;
    1849     static Bool mmx_present;
    1850     -
    1851     +
    1852     if (!initialized)
    1853     {
    1854     int tmp; /* static variables are accessed through %ebx,
    1855     @@ -1466,7 +1699,7 @@
    1856    
    1857     __asm__ __volatile__ (
    1858     /* Check if bit 21 in flags word is writeable */
    1859     -
    1860     +
    1861     "pusha \n\t"
    1862     "pushfl \n\t"
    1863     "popl %%eax \n\t"
    1864     @@ -1502,13 +1735,14 @@
    1865     : /* no input */);
    1866    
    1867     initialized = TRUE;
    1868     -
    1869     +
    1870     mmx_present = tmp;
    1871     }
    1872    
    1873     return mmx_present;
    1874     }
    1875     +#endif /* __amd64__ */
    1876    
    1877    
    1878     #endif /* RENDER */
    1879     -#endif /* USE_GCC34_MMX */
    1880     +#endif /* USE_MMX */
    1881     diff -ur xc-orig/programs/Xserver/fb/fbmmx.h xc/programs/Xserver/fb/fbmmx.h
    1882     --- xc-orig/programs/Xserver/fb/fbmmx.h 2005-02-11 04:00:50.006092570 -0500
    1883     +++ xc/programs/Xserver/fb/fbmmx.h 2005-02-11 04:01:32.072346126 -0500
    1884     @@ -1,5 +1,5 @@
    1885     /*
    1886     - * Copyright © 2004 Red Hat, Inc.
    1887     + * Copyright © 2004 Red Hat, Inc.
    1888     *
    1889     * Permission to use, copy, modify, distribute, and sell this software and its
    1890     * documentation for any purpose is hereby granted without fee, provided that
    1891     @@ -18,17 +18,23 @@
    1892     * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
    1893     * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
    1894     *
    1895     - * Author: Søren Sandmann (sandmann@redhat.com)
    1896     + * Author: Søren Sandmann (sandmann@redhat.com)
    1897     *
    1898     * Based on work by Owen Taylor
    1899     */
    1900     -#ifdef USE_GCC34_MMX
    1901     +#ifdef USE_MMX
    1902     +
    1903     +#ifndef __amd64__
    1904     Bool fbHaveMMX(void);
    1905     #else
    1906     -#define fbHaveMMX FALSE
    1907     +#define fbHaveMMX() TRUE
    1908     +#endif
    1909     +
    1910     +#else
    1911     +#define fbHaveMMX() FALSE
    1912     #endif
    1913    
    1914     -#ifdef USE_GCC34_MMX
    1915     +#ifdef USE_MMX
    1916    
    1917     void fbCompositeSolidMask_nx8888x0565Cmmx (CARD8 op,
    1918     PicturePtr pSrc,
    1919     @@ -150,6 +156,38 @@
    1920     INT16 yDst,
    1921     CARD16 width,
    1922     CARD16 height);
    1923     +void fbCompositeSrc_8888x8x8888mmx (CARD8 op,
    1924     + PicturePtr pSrc,
    1925     + PicturePtr pMask,
    1926     + PicturePtr pDst,
    1927     + INT16 xSrc,
    1928     + INT16 ySrc,
    1929     + INT16 xMask,
    1930     + INT16 yMask,
    1931     + INT16 xDst,
    1932     + INT16 yDst,
    1933     + CARD16 width,
    1934     + CARD16 height);
    1935     +Bool fbCopyAreammx (DrawablePtr pSrc,
    1936     + DrawablePtr pDst,
    1937     + int src_x,
    1938     + int src_y,
    1939     + int dst_x,
    1940     + int dst_y,
    1941     + int width,
    1942     + int height);
    1943     +void fbCompositeCopyAreammx (CARD8 op,
    1944     + PicturePtr pSrc,
    1945     + PicturePtr pMask,
    1946     + PicturePtr pDst,
    1947     + INT16 xSrc,
    1948     + INT16 ySrc,
    1949     + INT16 xMask,
    1950     + INT16 yMask,
    1951     + INT16 xDst,
    1952     + INT16 yDst,
    1953     + CARD16 width,
    1954     + CARD16 height);
    1955     Bool fbSolidFillmmx (DrawablePtr pDraw,
    1956     int x,
    1957     int y,
    1958     @@ -157,4 +195,4 @@
    1959     int height,
    1960     FbBits xor);
    1961    
    1962     -#endif /* USE_GCC34_MMX */
    1963     +#endif /* USE_MMX */
    1964    
    1965     diff -ur xc-orig/programs/Xserver/fb/fbpict.c xc/programs/Xserver/fb/fbpict.c
    1966     --- xc-orig/programs/Xserver/fb/fbpict.c 2005-02-11 04:00:50.007092600 -0500
    1967     +++ xc/programs/Xserver/fb/fbpict.c 2005-02-11 04:01:32.075346216 -0500
    1968     @@ -1,7 +1,7 @@
    1969     /*
    1970     * $XFree86: xc/programs/Xserver/fb/fbpict.c,v 1.15 2002/09/26 02:56:48 keithp Exp $
    1971     *
    1972     - * Copyright © 2000 SuSE, Inc.
    1973     + * Copyright © 2000 SuSE, Inc.
    1974     *
    1975     * Permission to use, copy, modify, distribute, and sell this software and its
    1976     * documentation for any purpose is hereby granted without fee, provided that
    1977     @@ -863,6 +863,15 @@
    1978     if (!pSrc->transform && !(pMask && pMask->transform))
    1979     if (!maskAlphaMap && !srcAlphaMap && !dstAlphaMap)
    1980     switch (op) {
    1981     + case PictOpSrc:
    1982     +#ifdef USE_MMX
    1983     + if (!pMask && pSrc->format == pDst->format &&
    1984     + pSrc->pDrawable != pDst->pDrawable)
    1985     + {
    1986     + func = fbCompositeCopyAreammx;
    1987     + }
    1988     +#endif
    1989     + break;
    1990     case PictOpOver:
    1991     if (pMask)
    1992     {
    1993     @@ -877,7 +886,7 @@
    1994     switch (pDst->format) {
    1995     case PICT_r5g6b5:
    1996     case PICT_b5g6r5:
    1997     -#ifdef USE_GCC34_MMX
    1998     +#ifdef USE_MMX
    1999     if (fbHaveMMX())
    2000     func = fbCompositeSolidMask_nx8x0565mmx;
    2001     else
    2002     @@ -892,7 +901,7 @@
    2003     case PICT_x8r8g8b8:
    2004     case PICT_a8b8g8r8:
    2005     case PICT_x8b8g8r8:
    2006     -#ifdef USE_GCC34_MMX
    2007     +#ifdef USE_MMX
    2008     if (fbHaveMMX())
    2009     func = fbCompositeSolidMask_nx8x8888mmx;
    2010     else
    2011     @@ -906,7 +915,7 @@
    2012     switch (pDst->format) {
    2013     case PICT_a8r8g8b8:
    2014     case PICT_x8r8g8b8:
    2015     -#ifdef USE_GCC34_MMX
    2016     +#ifdef USE_MMX
    2017     if (fbHaveMMX())
    2018     func = fbCompositeSolidMask_nx8888x8888Cmmx;
    2019     else
    2020     @@ -914,7 +923,7 @@
    2021     func = fbCompositeSolidMask_nx8888x8888C;
    2022     break;
    2023     case PICT_r5g6b5:
    2024     -#ifdef USE_GCC34_MMX
    2025     +#ifdef USE_MMX
    2026     if (fbHaveMMX())
    2027     func = fbCompositeSolidMask_nx8888x0565Cmmx;
    2028     else
    2029     @@ -929,7 +938,7 @@
    2030     switch (pDst->format) {
    2031     case PICT_a8b8g8r8:
    2032     case PICT_x8b8g8r8:
    2033     -#ifdef USE_GCC34_MMX
    2034     +#ifdef USE_MMX
    2035     if (fbHaveMMX())
    2036     func = fbCompositeSolidMask_nx8888x8888Cmmx;
    2037     else
    2038     @@ -937,7 +946,7 @@
    2039     func = fbCompositeSolidMask_nx8888x8888C;
    2040     break;
    2041     case PICT_b5g6r5:
    2042     -#ifdef USE_GCC34_MMX
    2043     +#ifdef USE_MMX
    2044     if (fbHaveMMX())
    2045     func = fbCompositeSolidMask_nx8888x0565Cmmx;
    2046     else
    2047     @@ -970,6 +979,7 @@
    2048     xSrc == xMask && ySrc == yMask &&
    2049     !pMask->componentAlpha)
    2050     {
    2051     + /* source == mask: non-premultiplied data */
    2052     switch (pSrc->format) {
    2053     case PICT_x8b8g8r8:
    2054     switch (pMask->format) {
    2055     @@ -978,13 +988,13 @@
    2056     switch (pDst->format) {
    2057     case PICT_a8r8g8b8:
    2058     case PICT_x8r8g8b8:
    2059     -#ifdef USE_GCC34_MMX
    2060     +#ifdef USE_MMX
    2061     if (fbHaveMMX())
    2062     func = fbCompositeSrc_8888RevNPx8888mmx;
    2063     #endif
    2064     break;
    2065     case PICT_r5g6b5:
    2066     -#ifdef USE_GCC34_MMX
    2067     +#ifdef USE_MMX
    2068     if (fbHaveMMX())
    2069     func = fbCompositeSrc_8888RevNPx0565mmx;
    2070     #endif
    2071     @@ -1000,13 +1010,13 @@
    2072     switch (pDst->format) {
    2073     case PICT_a8b8g8r8:
    2074     case PICT_x8b8g8r8:
    2075     -#ifdef USE_GCC34_MMX
    2076     +#ifdef USE_MMX
    2077     if (fbHaveMMX())
    2078     func = fbCompositeSrc_8888RevNPx8888mmx;
    2079     #endif
    2080     break;
    2081     case PICT_r5g6b5:
    2082     -#ifdef USE_GCC34_MMX
    2083     +#ifdef USE_MMX
    2084     if (fbHaveMMX())
    2085     func = fbCompositeSrc_8888RevNPx0565mmx;
    2086     #endif
    2087     @@ -1018,9 +1028,27 @@
    2088     }
    2089     break;
    2090     }
    2091     + else
    2092     + {
    2093     + /* non-repeating source, repeating mask => translucent window */
    2094     + if (maskRepeat &&
    2095     + pMask->pDrawable->width == 1 &&
    2096     + pMask->pDrawable->height == 1)
    2097     + {
    2098     + if (pSrc->format == PICT_x8r8g8b8 &&
    2099     + pDst->format == PICT_x8r8g8b8 &&
    2100     + pMask->format == PICT_a8)
    2101     + {
    2102     +#ifdef USE_MMX
    2103     + if (fbHaveMMX())
    2104     + func = fbCompositeSrc_8888x8x8888mmx;
    2105     +#endif
    2106     + }
    2107     + }
    2108     + }
    2109     }
    2110     }
    2111     - else
    2112     + else /* no mask */
    2113     {
    2114     if (srcRepeat &&
    2115     pSrc->pDrawable->width == 1 &&
    2116     @@ -1032,7 +1060,7 @@
    2117     switch (pDst->format) {
    2118     case PICT_a8r8g8b8:
    2119     case PICT_x8r8g8b8:
    2120     -#ifdef USE_GCC34_MMX
    2121     +#ifdef USE_MMX
    2122     if (fbHaveMMX())
    2123     {
    2124     srcRepeat = FALSE;
    2125     @@ -1041,7 +1069,7 @@
    2126     #endif
    2127     break;
    2128     case PICT_r5g6b5:
    2129     -#ifdef USE_GCC34_MMX
    2130     +#ifdef USE_MMX
    2131     if (fbHaveMMX())
    2132     {
    2133     srcRepeat = FALSE;
    2134     @@ -1070,6 +1098,27 @@
    2135     break;
    2136     }
    2137     break;
    2138     + case PICT_x8r8g8b8:
    2139     + switch (pDst->format) {
    2140     + case PICT_a8r8g8b8:
    2141     + case PICT_x8r8g8b8:
    2142     +#ifdef USE_MMX
    2143     + if (fbHaveMMX())
    2144     + func = fbCompositeCopyAreammx;
    2145     +#endif
    2146     + break;
    2147     + }
    2148     + case PICT_x8b8g8r8:
    2149     + switch (pDst->format) {
    2150     + case PICT_a8b8g8r8:
    2151     + case PICT_x8b8g8r8:
    2152     +#ifdef USE_MMX
    2153     + if (fbHaveMMX())
    2154     + func = fbCompositeCopyAreammx;
    2155     +#endif
    2156     + break;
    2157     + }
    2158     + break;
    2159     case PICT_a8b8g8r8:
    2160     switch (pDst->format) {
    2161     case PICT_a8b8g8r8:
    2162     @@ -1109,7 +1158,7 @@
    2163     case PICT_a8r8g8b8:
    2164     switch (pDst->format) {
    2165     case PICT_a8r8g8b8:
    2166     -#ifdef USE_GCC34_MMX
    2167     +#ifdef USE_MMX
    2168     if (fbHaveMMX())
    2169     func = fbCompositeSrcAdd_8888x8888mmx;
    2170     else
    2171     @@ -1121,7 +1170,7 @@
    2172     case PICT_a8b8g8r8:
    2173     switch (pDst->format) {
    2174     case PICT_a8b8g8r8:
    2175     -#ifdef USE_GCC34_MMX
    2176     +#ifdef USE_MMX
    2177     if (fbHaveMMX())
    2178     func = fbCompositeSrcAdd_8888x8888mmx;
    2179     else
    2180     @@ -1133,7 +1182,7 @@
    2181     case PICT_a8:
    2182     switch (pDst->format) {
    2183     case PICT_a8:
    2184     -#ifdef USE_GCC34_MMX
    2185     +#ifdef USE_MMX
    2186     if (fbHaveMMX())
    2187     func = fbCompositeSrcAdd_8000x8000mmx;
    2188     else