Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/i386/h264dsp

00001 /*
00002  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
00003  *
00004  * This library is free software; you can redistribute it and/or
00005  * modify it under the terms of the GNU Lesser General Public
00006  * License as published by the Free Software Foundation; either
00007  * version 2 of the License, or (at your option) any later version.
00008  *
00009  * This library is distributed in the hope that it will be useful,
00010  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00011  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00012  * Lesser General Public License for more details.
00013  *
00014  * You should have received a copy of the GNU Lesser General Public
00015  * License along with this library; if not, write to the Free Software
00016  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
00017  */
00018 
00019 
00020 /***********************************/
00021 /* IDCT */
00022 
00023 /* in/out: mma=mma+mmb, mmb=mmb-mma */
00024 #define SUMSUB_BA( a, b ) \
00025     "paddw "#b", "#a" \n\t"\
00026     "paddw "#b", "#b" \n\t"\
00027     "psubw "#a", "#b" \n\t"
00028 
00029 #define SUMSUB_BADC( a, b, c, d ) \
00030     "paddw "#b", "#a" \n\t"\
00031     "paddw "#d", "#c" \n\t"\
00032     "paddw "#b", "#b" \n\t"\
00033     "paddw "#d", "#d" \n\t"\
00034     "psubw "#a", "#b" \n\t"\
00035     "psubw "#c", "#d" \n\t"
00036 
00037 #define SUMSUBD2_AB( a, b, t ) \
00038     "movq  "#b", "#t" \n\t"\
00039     "psraw  $1 , "#b" \n\t"\
00040     "paddw "#a", "#b" \n\t"\
00041     "psraw  $1 , "#a" \n\t"\
00042     "psubw "#t", "#a" \n\t"
00043 
00044 #define IDCT4_1D( s02, s13, d02, d13, t ) \
00045     SUMSUB_BA  ( s02, d02 )\
00046     SUMSUBD2_AB( s13, d13, t )\
00047     SUMSUB_BADC( d13, s02, s13, d02 )
00048 
00049 #define SBUTTERFLY(a,b,t,n)\
00050     "movq " #a ", " #t "                \n\t" /* abcd */\
00051     "punpckl" #n " " #b ", " #a "       \n\t" /* aebf */\
00052     "punpckh" #n " " #b ", " #t "       \n\t" /* cgdh */\
00053 
00054 #define TRANSPOSE4(a,b,c,d,t)\
00055     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
00056     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
00057     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
00058     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
00059 
00060 #define STORE_DIFF_4P( p, t, z ) \
00061     "psraw      $6,     "#p" \n\t"\
00062     "movd       (%0),   "#t" \n\t"\
00063     "punpcklbw "#z",    "#t" \n\t"\
00064     "paddsw    "#t",    "#p" \n\t"\
00065     "packuswb  "#z",    "#p" \n\t"\
00066     "movd      "#p",    (%0) \n\t"
00067 
00068 void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride)
00069 {
00070     /* Load dct coeffs */
00071     asm volatile(
00072         "movq   (%0), %%mm0 \n\t"
00073         "movq  8(%0), %%mm1 \n\t"
00074         "movq 16(%0), %%mm2 \n\t"
00075         "movq 24(%0), %%mm3 \n\t"
00076     :: "r"(block) );
00077 
00078     asm volatile(
00079         /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
00080         IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
00081 
00082         "movq      %0,    %%mm6 \n\t"
00083         /* in: 1,4,0,2  out: 1,2,3,0 */
00084         TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
00085 
00086         "paddw     %%mm6, %%mm3 \n\t"
00087 
00088         /* mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13 */
00089         IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
00090 
00091         "pxor %%mm7, %%mm7    \n\t"
00092     :: "m"(ff_pw_32));
00093 
00094     asm volatile(
00095     STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
00096         "add %1, %0             \n\t"
00097     STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
00098         "add %1, %0             \n\t"
00099     STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
00100         "add %1, %0             \n\t"
00101     STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
00102         : "+r"(dst)
00103         : "r" ((long)stride)
00104     );
00105 }
00106 
00107 
00108 /***********************************/
00109 /* deblocking */
00110 
00111 // out: o = |x-y|>a
00112 // clobbers: t
00113 #define DIFF_GT_MMX(x,y,a,o,t)\
00114     "movq     "#y", "#t"  \n\t"\
00115     "movq     "#x", "#o"  \n\t"\
00116     "psubusb  "#x", "#t"  \n\t"\
00117     "psubusb  "#y", "#o"  \n\t"\
00118     "por      "#t", "#o"  \n\t"\
00119     "psubusb  "#a", "#o"  \n\t"
00120 
00121 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
00122 // out: mm5=beta-1, mm7=mask
00123 // clobbers: mm4,mm6
00124 #define H264_DEBLOCK_MASK(alpha1, beta1) \
00125     "pshufw $0, "#alpha1", %%mm4 \n\t"\
00126     "pshufw $0, "#beta1 ", %%mm5 \n\t"\
00127     "packuswb  %%mm4, %%mm4      \n\t"\
00128     "packuswb  %%mm5, %%mm5      \n\t"\
00129     DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
00130     DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
00131     "por       %%mm4, %%mm7      \n\t"\
00132     DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
00133     "por       %%mm4, %%mm7      \n\t"\
00134     "pxor      %%mm6, %%mm6      \n\t"\
00135     "pcmpeqb   %%mm6, %%mm7      \n\t"
00136 
00137 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
00138 // out: mm1=p0' mm2=q0'
00139 // clobbers: mm0,3-6
00140 #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
00141         /* a = q0^p0^((p1-q1)>>2) */\
00142         "movq    %%mm0, %%mm4  \n\t"\
00143         "psubb   %%mm3, %%mm4  \n\t"\
00144         "psrlw   $2,    %%mm4  \n\t"\
00145         "pxor    %%mm1, %%mm4  \n\t"\
00146         "pxor    %%mm2, %%mm4  \n\t"\
00147         /* b = p0^(q1>>2) */\
00148         "psrlw   $2,    %%mm3  \n\t"\
00149         "pand "#pb_3f", %%mm3  \n\t"\
00150         "movq    %%mm1, %%mm5  \n\t"\
00151         "pxor    %%mm3, %%mm5  \n\t"\
00152         /* c = q0^(p1>>2) */\
00153         "psrlw   $2,    %%mm0  \n\t"\
00154         "pand "#pb_3f", %%mm0  \n\t"\
00155         "movq    %%mm2, %%mm6  \n\t"\
00156         "pxor    %%mm0, %%mm6  \n\t"\
00157         /* d = (c^b) & ~(b^a) & 1 */\
00158         "pxor    %%mm5, %%mm6  \n\t"\
00159         "pxor    %%mm4, %%mm5  \n\t"\
00160         "pandn   %%mm6, %%mm5  \n\t"\
00161         "pand "#pb_01", %%mm5  \n\t"\
00162         /* delta = (avg(q0, p1>>2) + (d&a))
00163          *       - (avg(p0, q1>>2) + (d&~a)) */\
00164         "pavgb   %%mm2, %%mm0  \n\t"\
00165         "movq    %%mm5, %%mm6  \n\t"\
00166         "pand    %%mm4, %%mm6  \n\t"\
00167         "paddusb %%mm6, %%mm0  \n\t"\
00168         "pavgb   %%mm1, %%mm3  \n\t"\
00169         "pandn   %%mm5, %%mm4  \n\t"\
00170         "paddusb %%mm4, %%mm3  \n\t"\
00171         /* p0 += clip(delta, -tc0, tc0)
00172          * q0 -= clip(delta, -tc0, tc0) */\
00173         "movq    %%mm0, %%mm4  \n\t"\
00174         "psubusb %%mm3, %%mm0  \n\t"\
00175         "psubusb %%mm4, %%mm3  \n\t"\
00176         "pminub  %%mm7, %%mm0  \n\t"\
00177         "pminub  %%mm7, %%mm3  \n\t"\
00178         "paddusb %%mm0, %%mm1  \n\t"\
00179         "paddusb %%mm3, %%mm2  \n\t"\
00180         "psubusb %%mm3, %%mm1  \n\t"\
00181         "psubusb %%mm0, %%mm2  \n\t"
00182 
00183 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
00184 // out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
00185 // clobbers: q2, tmp, tc0
00186 #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
00187         "movq     %%mm1,  "#tmp"   \n\t"\
00188         "pavgb    %%mm2,  "#tmp"   \n\t"\
00189         "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
00190         "pxor   "q2addr", "#tmp"   \n\t"\
00191         "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
00192         "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
00193         "movq     "#p1",  "#tmp"   \n\t"\
00194         "psubusb  "#tc0", "#tmp"   \n\t"\
00195         "paddusb  "#p1",  "#tc0"   \n\t"\
00196         "pmaxub   "#tmp", "#q2"    \n\t"\
00197         "pminub   "#tc0", "#q2"    \n\t"\
00198         "movq     "#q2",  "q1addr" \n\t"
00199 
00200 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
00201 {
00202     uint64_t tmp0;
00203     uint64_t tc = (uint8_t)tc0[1]*0x01010000 | (uint8_t)tc0[0]*0x0101;
00204     // with luma, tc0=0 doesn't mean no filtering, so we need a separate input mask
00205     uint32_t mask[2] = { (tc0[0]>=0)*0xffffffff, (tc0[1]>=0)*0xffffffff };
00206 
00207     asm volatile(
00208         "movq    (%1,%3), %%mm0    \n\t" //p1
00209         "movq    (%1,%3,2), %%mm1  \n\t" //p0
00210         "movq    (%2),    %%mm2    \n\t" //q0
00211         "movq    (%2,%3), %%mm3    \n\t" //q1
00212         H264_DEBLOCK_MASK(%6, %7)
00213         "pand     %5,     %%mm7    \n\t"
00214         "movq     %%mm7,  %0       \n\t"
00215 
00216         /* filter p1 */
00217         "movq     (%1),   %%mm3    \n\t" //p2
00218         DIFF_GT_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
00219         "pandn    %%mm7,  %%mm6    \n\t"
00220         "pcmpeqb  %%mm7,  %%mm6    \n\t"
00221         "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
00222         "pshufw  $80, %4, %%mm4    \n\t"
00223         "pand     %%mm7,  %%mm4    \n\t" // mask & tc0
00224         "movq     %8,     %%mm7    \n\t"
00225         "pand     %%mm6,  %%mm7    \n\t" // mask & |p2-p0|<beta & 1
00226         "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
00227         "paddb    %%mm4,  %%mm7    \n\t" // tc++
00228         H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
00229 
00230         /* filter q1 */
00231         "movq    (%2,%3,2), %%mm4  \n\t" //q2
00232         DIFF_GT_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
00233         "pandn    %0,     %%mm6    \n\t"
00234         "pcmpeqb  %0,     %%mm6    \n\t"
00235         "pand     %0,     %%mm6    \n\t"
00236         "pshufw  $80, %4, %%mm5    \n\t"
00237         "pand     %%mm6,  %%mm5    \n\t"
00238         "pand     %8,     %%mm6    \n\t"
00239         "paddb    %%mm6,  %%mm7    \n\t"
00240         "movq    (%2,%3), %%mm3    \n\t"
00241         H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
00242 
00243         /* filter p0, q0 */
00244         H264_DEBLOCK_P0_Q0(%8, %9)
00245         "movq      %%mm1, (%1,%3,2) \n\t"
00246         "movq      %%mm2, (%2)      \n\t"
00247 
00248         : "=m"(tmp0)
00249         : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
00250           "m"(tc), "m"(*(uint64_t*)mask), "m"(alpha1), "m"(beta1),
00251           "m"(mm_bone), "m"(ff_pb_3F)
00252     );
00253 }
00254 
00255 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
00256 {
00257     if((tc0[0] & tc0[1]) >= 0)
00258         h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
00259     if((tc0[2] & tc0[3]) >= 0)
00260         h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
00261 }
00262 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
00263 {
00264     //FIXME: could cut some load/stores by merging transpose with filter
00265     // also, it only needs to transpose 6x8
00266     uint8_t trans[8*8];
00267     int i;
00268     for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
00269         if((tc0[0] & tc0[1]) < 0)
00270             continue;
00271         transpose4x4(trans,       pix-4,          8, stride);
00272         transpose4x4(trans  +4*8, pix,            8, stride);
00273         transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
00274         transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
00275         h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
00276         transpose4x4(pix-2,          trans  +2*8, stride, 8);
00277         transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
00278     }
00279 }
00280 
00281 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
00282 {
00283     asm volatile(
00284         "movq    (%0),    %%mm0     \n\t" //p1
00285         "movq    (%0,%2), %%mm1     \n\t" //p0
00286         "movq    (%1),    %%mm2     \n\t" //q0
00287         "movq    (%1,%2), %%mm3     \n\t" //q1
00288         H264_DEBLOCK_MASK(%4, %5)
00289         "movd      %3,    %%mm6     \n\t"
00290         "punpcklbw %%mm6, %%mm6     \n\t"
00291         "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
00292         H264_DEBLOCK_P0_Q0(%6, %7)
00293         "movq      %%mm1, (%0,%2)   \n\t"
00294         "movq      %%mm2, (%1)      \n\t"
00295 
00296         :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
00297            "r"(*(uint32_t*)tc0),
00298            "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
00299     );
00300 }
00301 
00302 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
00303 {
00304     h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
00305 }
00306 
00307 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
00308 {
00309     //FIXME: could cut some load/stores by merging transpose with filter
00310     uint8_t trans[8*4];
00311     transpose4x4(trans, pix-2, 8, stride);
00312     transpose4x4(trans+4, pix-2+4*stride, 8, stride);
00313     h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
00314     transpose4x4(pix-2, trans, stride, 8);
00315     transpose4x4(pix-2+4*stride, trans+4, stride, 8);
00316 }
00317 
00318 // p0 = (p0 + q1 + 2*p1 + 2) >> 2
00319 #define H264_FILTER_CHROMA4(p0, p1, q1, one) \
00320     "movq    "#p0", %%mm4  \n\t"\
00321     "pxor    "#q1", %%mm4  \n\t"\
00322     "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
00323     "pavgb   "#q1", "#p0"  \n\t"\
00324     "psubusb %%mm4, "#p0"  \n\t"\
00325     "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
00326 
00327 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
00328 {
00329     asm volatile(
00330         "movq    (%0),    %%mm0     \n\t"
00331         "movq    (%0,%2), %%mm1     \n\t"
00332         "movq    (%1),    %%mm2     \n\t"
00333         "movq    (%1,%2), %%mm3     \n\t"
00334         H264_DEBLOCK_MASK(%3, %4)
00335         "movq    %%mm1,   %%mm5     \n\t"
00336         "movq    %%mm2,   %%mm6     \n\t"
00337         H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
00338         H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
00339         "psubb   %%mm5,   %%mm1     \n\t"
00340         "psubb   %%mm6,   %%mm2     \n\t"
00341         "pand    %%mm7,   %%mm1     \n\t"
00342         "pand    %%mm7,   %%mm2     \n\t"
00343         "paddb   %%mm5,   %%mm1     \n\t"
00344         "paddb   %%mm6,   %%mm2     \n\t"
00345         "movq    %%mm1,   (%0,%2)   \n\t"
00346         "movq    %%mm2,   (%1)      \n\t"
00347         :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
00348            "m"(alpha1), "m"(beta1), "m"(mm_bone)
00349     );
00350 }
00351 
00352 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
00353 {
00354     h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
00355 }
00356 
00357 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
00358 {
00359     //FIXME: could cut some load/stores by merging transpose with filter
00360     uint8_t trans[8*4];
00361     transpose4x4(trans, pix-2, 8, stride);
00362     transpose4x4(trans+4, pix-2+4*stride, 8, stride);
00363     h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
00364     transpose4x4(pix-2, trans, stride, 8);
00365     transpose4x4(pix-2+4*stride, trans+4, stride, 8);
00366 }
00367 
00368 
00369 /***********************************/
00370 /* motion compensation */
00371 
00372 #define QPEL_H264V(A,B,C,D,E,F,OP)\
00373         "movd (%0), "#F"                \n\t"\
00374         "movq "#C", %%mm6               \n\t"\
00375         "paddw "#D", %%mm6              \n\t"\
00376         "psllw $2, %%mm6                \n\t"\
00377         "psubw "#B", %%mm6              \n\t"\
00378         "psubw "#E", %%mm6              \n\t"\
00379         "pmullw %4, %%mm6               \n\t"\
00380         "add %2, %0                     \n\t"\
00381         "punpcklbw %%mm7, "#F"          \n\t"\
00382         "paddw %5, "#A"                 \n\t"\
00383         "paddw "#F", "#A"               \n\t"\
00384         "paddw "#A", %%mm6              \n\t"\
00385         "psraw $5, %%mm6                \n\t"\
00386         "packuswb %%mm6, %%mm6          \n\t"\
00387         OP(%%mm6, (%1), A, d)\
00388         "add %3, %1                     \n\t"     
00389 
00390 #define QPEL_H264HV(A,B,C,D,E,F,OF)\
00391         "movd (%0), "#F"                \n\t"\
00392         "movq "#C", %%mm6               \n\t"\
00393         "paddw "#D", %%mm6              \n\t"\
00394         "psllw $2, %%mm6                \n\t"\
00395         "psubw "#B", %%mm6              \n\t"\
00396         "psubw "#E", %%mm6              \n\t"\
00397         "pmullw %3, %%mm6               \n\t"\
00398         "add %2, %0                     \n\t"\
00399         "punpcklbw %%mm7, "#F"          \n\t"\
00400         "paddw "#F", "#A"               \n\t"\
00401         "paddw "#A", %%mm6              \n\t"\
00402         "movq %%mm6, "#OF"(%1)          \n\t"
00403         
00404 #define QPEL_H264(OPNAME, OP, MMX)\
00405 static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00406     int h=4;\
00407 \
00408     asm volatile(\
00409         "pxor %%mm7, %%mm7              \n\t"\
00410         "movq %5, %%mm4                 \n\t"\
00411         "movq %6, %%mm5                 \n\t"\
00412         "1:                             \n\t"\
00413         "movd  -1(%0), %%mm1            \n\t"\
00414         "movd    (%0), %%mm2            \n\t"\
00415         "movd   1(%0), %%mm3            \n\t"\
00416         "movd   2(%0), %%mm0            \n\t"\
00417         "punpcklbw %%mm7, %%mm1         \n\t"\
00418         "punpcklbw %%mm7, %%mm2         \n\t"\
00419         "punpcklbw %%mm7, %%mm3         \n\t"\
00420         "punpcklbw %%mm7, %%mm0         \n\t"\
00421         "paddw %%mm0, %%mm1             \n\t"\
00422         "paddw %%mm3, %%mm2             \n\t"\
00423         "movd  -2(%0), %%mm0            \n\t"\
00424         "movd   3(%0), %%mm3            \n\t"\
00425         "punpcklbw %%mm7, %%mm0         \n\t"\
00426         "punpcklbw %%mm7, %%mm3         \n\t"\
00427         "paddw %%mm3, %%mm0             \n\t"\
00428         "psllw $2, %%mm2                \n\t"\
00429         "psubw %%mm1, %%mm2             \n\t"\
00430         "pmullw %%mm4, %%mm2            \n\t"\
00431         "paddw %%mm5, %%mm0             \n\t"\
00432         "paddw %%mm2, %%mm0             \n\t"\
00433         "psraw $5, %%mm0                \n\t"\
00434         "packuswb %%mm0, %%mm0          \n\t"\
00435         OP(%%mm0, (%1),%%mm6, d)\
00436         "add %3, %0                     \n\t"\
00437         "add %4, %1                     \n\t"\
00438         "decl %2                        \n\t"\
00439         " jnz 1b                        \n\t"\
00440         : "+a"(src), "+c"(dst), "+m"(h)\
00441         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00442         : "memory"\
00443     );\
00444 }\
00445 static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00446     src -= 2*srcStride;\
00447     asm volatile(\
00448         "pxor %%mm7, %%mm7              \n\t"\
00449         "movd (%0), %%mm0               \n\t"\
00450         "add %2, %0                     \n\t"\
00451         "movd (%0), %%mm1               \n\t"\
00452         "add %2, %0                     \n\t"\
00453         "movd (%0), %%mm2               \n\t"\
00454         "add %2, %0                     \n\t"\
00455         "movd (%0), %%mm3               \n\t"\
00456         "add %2, %0                     \n\t"\
00457         "movd (%0), %%mm4               \n\t"\
00458         "add %2, %0                     \n\t"\
00459         "punpcklbw %%mm7, %%mm0         \n\t"\
00460         "punpcklbw %%mm7, %%mm1         \n\t"\
00461         "punpcklbw %%mm7, %%mm2         \n\t"\
00462         "punpcklbw %%mm7, %%mm3         \n\t"\
00463         "punpcklbw %%mm7, %%mm4         \n\t"\
00464         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00465         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00466         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00467         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00468          \
00469         : "+a"(src), "+c"(dst)\
00470         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00471         : "memory"\
00472     );\
00473 }\
00474 static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00475     int h=4;\
00476     int w=3;\
00477     src -= 2*srcStride+2;\
00478     while(w--){\
00479         asm volatile(\
00480             "pxor %%mm7, %%mm7                  \n\t"\
00481             "movd (%0), %%mm0                   \n\t"\
00482             "add %2, %0                         \n\t"\
00483             "movd (%0), %%mm1                   \n\t"\
00484             "add %2, %0                         \n\t"\
00485             "movd (%0), %%mm2                   \n\t"\
00486             "add %2, %0                         \n\t"\
00487             "movd (%0), %%mm3                   \n\t"\
00488             "add %2, %0                         \n\t"\
00489             "movd (%0), %%mm4                   \n\t"\
00490             "add %2, %0                         \n\t"\
00491             "punpcklbw %%mm7, %%mm0             \n\t"\
00492             "punpcklbw %%mm7, %%mm1             \n\t"\
00493             "punpcklbw %%mm7, %%mm2             \n\t"\
00494             "punpcklbw %%mm7, %%mm3             \n\t"\
00495             "punpcklbw %%mm7, %%mm4             \n\t"\
00496             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
00497             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
00498             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
00499             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
00500              \
00501             : "+a"(src)\
00502             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
00503             : "memory"\
00504         );\
00505         tmp += 4;\
00506         src += 4 - 9*srcStride;\
00507     }\
00508     tmp -= 3*4;\
00509     asm volatile(\
00510         "movq %4, %%mm6                 \n\t"\
00511         "1:                             \n\t"\
00512         "movq     (%0), %%mm0           \n\t"\
00513         "paddw  10(%0), %%mm0           \n\t"\
00514         "movq    2(%0), %%mm1           \n\t"\
00515         "paddw   8(%0), %%mm1           \n\t"\
00516         "movq    4(%0), %%mm2           \n\t"\
00517         "paddw   6(%0), %%mm2           \n\t"\
00518         "psubw %%mm1, %%mm0             \n\t"/*a-b   (abccba)*/\
00519         "psraw $2, %%mm0                \n\t"/*(a-b)/4 */\
00520         "psubw %%mm1, %%mm0             \n\t"/*(a-b)/4-b */\
00521         "paddsw %%mm2, %%mm0            \n\t"\
00522         "psraw $2, %%mm0                \n\t"/*((a-b)/4-b)/4 */\
00523         "paddw %%mm6, %%mm2             \n\t"\
00524         "paddw %%mm2, %%mm0             \n\t"\
00525         "psraw $6, %%mm0                \n\t"\
00526         "packuswb %%mm0, %%mm0          \n\t"\
00527         OP(%%mm0, (%1),%%mm7, d)\
00528         "add $24, %0                    \n\t"\
00529         "add %3, %1                     \n\t"\
00530         "decl %2                        \n\t"\
00531         " jnz 1b                        \n\t"\
00532         : "+a"(tmp), "+c"(dst), "+m"(h)\
00533         : "S"((long)dstStride), "m"(ff_pw_32)\
00534         : "memory"\
00535     );\
00536 }\
00537 \
00538 static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00539     int h=8;\
00540     asm volatile(\
00541         "pxor %%mm7, %%mm7              \n\t"\
00542         "movq %5, %%mm6                 \n\t"\
00543         "1:                             \n\t"\
00544         "movq    (%0), %%mm0            \n\t"\
00545         "movq   1(%0), %%mm2            \n\t"\
00546         "movq %%mm0, %%mm1              \n\t"\
00547         "movq %%mm2, %%mm3              \n\t"\
00548         "punpcklbw %%mm7, %%mm0         \n\t"\
00549         "punpckhbw %%mm7, %%mm1         \n\t"\
00550         "punpcklbw %%mm7, %%mm2         \n\t"\
00551         "punpckhbw %%mm7, %%mm3         \n\t"\
00552         "paddw %%mm2, %%mm0             \n\t"\
00553         "paddw %%mm3, %%mm1             \n\t"\
00554         "psllw $2, %%mm0                \n\t"\
00555         "psllw $2, %%mm1                \n\t"\
00556         "movq   -1(%0), %%mm2           \n\t"\
00557         "movq    2(%0), %%mm4           \n\t"\
00558         "movq %%mm2, %%mm3              \n\t"\
00559         "movq %%mm4, %%mm5              \n\t"\
00560         "punpcklbw %%mm7, %%mm2         \n\t"\
00561         "punpckhbw %%mm7, %%mm3         \n\t"\
00562         "punpcklbw %%mm7, %%mm4         \n\t"\
00563         "punpckhbw %%mm7, %%mm5         \n\t"\
00564         "paddw %%mm4, %%mm2             \n\t"\
00565         "paddw %%mm3, %%mm5             \n\t"\
00566         "psubw %%mm2, %%mm0             \n\t"\
00567         "psubw %%mm5, %%mm1             \n\t"\
00568         "pmullw %%mm6, %%mm0            \n\t"\
00569         "pmullw %%mm6, %%mm1            \n\t"\
00570         "movd   -2(%0), %%mm2           \n\t"\
00571         "movd    7(%0), %%mm5           \n\t"\
00572         "punpcklbw %%mm7, %%mm2         \n\t"\
00573         "punpcklbw %%mm7, %%mm5         \n\t"\
00574         "paddw %%mm3, %%mm2             \n\t"\
00575         "paddw %%mm5, %%mm4             \n\t"\
00576         "movq %6, %%mm5                 \n\t"\
00577         "paddw %%mm5, %%mm2             \n\t"\
00578         "paddw %%mm5, %%mm4             \n\t"\
00579         "paddw %%mm2, %%mm0             \n\t"\
00580         "paddw %%mm4, %%mm1             \n\t"\
00581         "psraw $5, %%mm0                \n\t"\
00582         "psraw $5, %%mm1                \n\t"\
00583         "packuswb %%mm1, %%mm0          \n\t"\
00584         OP(%%mm0, (%1),%%mm5, q)\
00585         "add %3, %0                     \n\t"\
00586         "add %4, %1                     \n\t"\
00587         "decl %2                        \n\t"\
00588         " jnz 1b                        \n\t"\
00589         : "+a"(src), "+c"(dst), "+m"(h)\
00590         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00591         : "memory"\
00592     );\
00593 }\
00594 \
00595 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00596     int h= 2;\
00597     src -= 2*srcStride;\
00598     \
00599     while(h--){\
00600       asm volatile(\
00601         "pxor %%mm7, %%mm7              \n\t"\
00602         "movd (%0), %%mm0               \n\t"\
00603         "add %2, %0                     \n\t"\
00604         "movd (%0), %%mm1               \n\t"\
00605         "add %2, %0                     \n\t"\
00606         "movd (%0), %%mm2               \n\t"\
00607         "add %2, %0                     \n\t"\
00608         "movd (%0), %%mm3               \n\t"\
00609         "add %2, %0                     \n\t"\
00610         "movd (%0), %%mm4               \n\t"\
00611         "add %2, %0                     \n\t"\
00612         "punpcklbw %%mm7, %%mm0         \n\t"\
00613         "punpcklbw %%mm7, %%mm1         \n\t"\
00614         "punpcklbw %%mm7, %%mm2         \n\t"\
00615         "punpcklbw %%mm7, %%mm3         \n\t"\
00616         "punpcklbw %%mm7, %%mm4         \n\t"\
00617         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00618         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00619         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00620         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00621         QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00622         QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00623         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00624         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00625          \
00626         : "+a"(src), "+c"(dst)\
00627         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00628         : "memory"\
00629      );\
00630      src += 4-13*srcStride;\
00631      dst +=  4-8*dstStride;\
00632    }\
00633 }\
00634 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00635     int h=8;\
00636     int w=4;\
00637     src -= 2*srcStride+2;\
00638     while(w--){\
00639         asm volatile(\
00640             "pxor %%mm7, %%mm7                  \n\t"\
00641             "movd (%0), %%mm0                   \n\t"\
00642             "add %2, %0                         \n\t"\
00643             "movd (%0), %%mm1                   \n\t"\
00644             "add %2, %0                         \n\t"\
00645             "movd (%0), %%mm2                   \n\t"\
00646             "add %2, %0                         \n\t"\
00647             "movd (%0), %%mm3                   \n\t"\
00648             "add %2, %0                         \n\t"\
00649             "movd (%0), %%mm4                   \n\t"\
00650             "add %2, %0                         \n\t"\
00651             "punpcklbw %%mm7, %%mm0             \n\t"\
00652             "punpcklbw %%mm7, %%mm1             \n\t"\
00653             "punpcklbw %%mm7, %%mm2             \n\t"\
00654             "punpcklbw %%mm7, %%mm3             \n\t"\
00655             "punpcklbw %%mm7, %%mm4             \n\t"\
00656             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*4)\
00657             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*4)\
00658             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*4)\
00659             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*4)\
00660             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*8*4)\
00661             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*8*4)\
00662             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*8*4)\
00663             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*8*4)\
00664              \
00665             : "+a"(src)\
00666             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
00667             : "memory"\
00668         );\
00669         tmp += 4;\
00670         src += 4 - 13*srcStride;\
00671     }\
00672     tmp -= 4*4;\
00673     asm volatile(\
00674         "movq %4, %%mm6                 \n\t"\
00675         "1:                             \n\t"\
00676         "movq     (%0), %%mm0           \n\t"\
00677         "movq    8(%0), %%mm3           \n\t"\
00678         "movq    2(%0), %%mm1           \n\t"\
00679         "movq   10(%0), %%mm4           \n\t"\
00680         "paddw   %%mm4, %%mm0           \n\t"\
00681         "paddw   %%mm3, %%mm1           \n\t"\
00682         "paddw  18(%0), %%mm3           \n\t"\
00683         "paddw  16(%0), %%mm4           \n\t"\
00684         "movq    4(%0), %%mm2           \n\t"\
00685         "movq   12(%0), %%mm5           \n\t"\
00686         "paddw   6(%0), %%mm2           \n\t"\
00687         "paddw  14(%0), %%mm5           \n\t"\
00688         "psubw %%mm1, %%mm0             \n\t"\
00689         "psubw %%mm4, %%mm3             \n\t"\
00690         "psraw $2, %%mm0                \n\t"\
00691         "psraw $2, %%mm3                \n\t"\
00692         "psubw %%mm1, %%mm0             \n\t"\
00693         "psubw %%mm4, %%mm3             \n\t"\
00694         "paddsw %%mm2, %%mm0            \n\t"\
00695         "paddsw %%mm5, %%mm3            \n\t"\
00696         "psraw $2, %%mm0                \n\t"\
00697         "psraw $2, %%mm3                \n\t"\
00698         "paddw %%mm6, %%mm2             \n\t"\
00699         "paddw %%mm6, %%mm5             \n\t"\
00700         "paddw %%mm2, %%mm0             \n\t"\
00701         "paddw %%mm5, %%mm3             \n\t"\
00702         "psraw $6, %%mm0                \n\t"\
00703         "psraw $6, %%mm3                \n\t"\
00704         "packuswb %%mm3, %%mm0          \n\t"\
00705         OP(%%mm0, (%1),%%mm7, q)\
00706         "add $32, %0                    \n\t"\
00707         "add %3, %1                     \n\t"\
00708         "decl %2                        \n\t"\
00709         " jnz 1b                        \n\t"\
00710         : "+a"(tmp), "+c"(dst), "+m"(h)\
00711         : "S"((long)dstStride), "m"(ff_pw_32)\
00712         : "memory"\
00713     );\
00714 }\
00715 static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00716     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00717     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00718     src += 8*srcStride;\
00719     dst += 8*dstStride;\
00720     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00721     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00722 }\
00723 \
00724 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00725     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00726     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00727     src += 8*srcStride;\
00728     dst += 8*dstStride;\
00729     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00730     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00731 }\
00732 \
00733 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00734     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride);\
00735     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp  , src+8, dstStride, tmpStride, srcStride);\
00736     src += 8*srcStride;\
00737     dst += 8*dstStride;\
00738     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride);\
00739     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp  , src+8, dstStride, tmpStride, srcStride);\
00740 }\
00741 
00742 #define H264_MC(OPNAME, SIZE, MMX) \
00743 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
00744     OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
00745 }\
00746 \
00747 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00748     uint64_t temp[SIZE*SIZE/8];\
00749     uint8_t * const half= (uint8_t*)temp;\
00750     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
00751     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
00752 }\
00753 \
00754 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00755     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
00756 }\
00757 \
00758 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00759     uint64_t temp[SIZE*SIZE/8];\
00760     uint8_t * const half= (uint8_t*)temp;\
00761     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
00762     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+1, half, stride, stride, SIZE);\
00763 }\
00764 \
00765 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00766     uint64_t temp[SIZE*SIZE/8];\
00767     uint8_t * const half= (uint8_t*)temp;\
00768     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
00769     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
00770 }\
00771 \
00772 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00773     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
00774 }\
00775 \
00776 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00777     uint64_t temp[SIZE*SIZE/8];\
00778     uint8_t * const half= (uint8_t*)temp;\
00779     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
00780     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
00781 }\
00782 \
00783 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00784     uint64_t temp[SIZE*SIZE/4];\
00785     uint8_t * const halfH= (uint8_t*)temp;\
00786     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
00787     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
00788     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
00789     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
00790 }\
00791 \
00792 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00793     uint64_t temp[SIZE*SIZE/4];\
00794     uint8_t * const halfH= (uint8_t*)temp;\
00795     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
00796     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
00797     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
00798     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
00799 }\
00800 \
00801 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00802     uint64_t temp[SIZE*SIZE/4];\
00803     uint8_t * const halfH= (uint8_t*)temp;\
00804     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
00805     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
00806     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
00807     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
00808 }\
00809 \
00810 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00811     uint64_t temp[SIZE*SIZE/4];\
00812     uint8_t * const halfH= (uint8_t*)temp;\
00813     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
00814     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
00815     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
00816     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
00817 }\
00818 \
00819 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00820     uint64_t temp[SIZE*(SIZE+8)/4];\
00821     int16_t * const tmp= (int16_t*)temp;\
00822     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
00823 }\
00824 \
00825 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00826     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
00827     uint8_t * const halfH= (uint8_t*)temp;\
00828     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
00829     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
00830     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
00831     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
00832     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
00833 }\
00834 \
00835 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00836     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
00837     uint8_t * const halfH= (uint8_t*)temp;\
00838     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
00839     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
00840     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
00841     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
00842     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
00843 }\
00844 \
00845 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00846     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
00847     uint8_t * const halfV= (uint8_t*)temp;\
00848     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
00849     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
00850     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
00851     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
00852     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
00853 }\
00854 \
00855 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00856     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
00857     uint8_t * const halfV= (uint8_t*)temp;\
00858     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
00859     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
00860     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
00861     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
00862     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
00863 }\
00864 
00865 
00866 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "     \n\t"
00867 #define AVG_3DNOW_OP(a,b,temp, size) \
00868 "mov" #size " " #b ", " #temp " \n\t"\
00869 "pavgusb " #temp ", " #a "      \n\t"\
00870 "mov" #size " " #a ", " #b "    \n\t"
00871 #define AVG_MMX2_OP(a,b,temp, size) \
00872 "mov" #size " " #b ", " #temp " \n\t"\
00873 "pavgb " #temp ", " #a "        \n\t"\
00874 "mov" #size " " #a ", " #b "    \n\t"
00875 
00876 QPEL_H264(put_,       PUT_OP, 3dnow)
00877 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
00878 QPEL_H264(put_,       PUT_OP, mmx2)
00879 QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
00880 
00881 H264_MC(put_, 4, 3dnow)
00882 H264_MC(put_, 8, 3dnow)
00883 H264_MC(put_, 16,3dnow)
00884 H264_MC(avg_, 4, 3dnow)
00885 H264_MC(avg_, 8, 3dnow)
00886 H264_MC(avg_, 16,3dnow)
00887 H264_MC(put_, 4, mmx2)
00888 H264_MC(put_, 8, mmx2)
00889 H264_MC(put_, 16,mmx2)
00890 H264_MC(avg_, 4, mmx2)
00891 H264_MC(avg_, 8, mmx2)
00892 H264_MC(avg_, 16,mmx2)
00893 
00894 
00895 #define H264_CHROMA_OP(S,D)
00896 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
00897 #include "dsputil_h264_template_mmx.c"
00898 #undef H264_CHROMA_OP
00899 #undef H264_CHROMA_MC8_TMPL
00900 
00901 #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
00902 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
00903 #include "dsputil_h264_template_mmx.c"
00904 #undef H264_CHROMA_OP
00905 #undef H264_CHROMA_MC8_TMPL
00906 
00907 #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
00908 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
00909 #include "dsputil_h264_template_mmx.c"
00910 #undef H264_CHROMA_OP
00911 #undef H264_CHROMA_MC8_TMPL
00912
hvirtual/quicktime/ffmpeg/libavcodec/i386/h264dsp_mmx.c
