FFmpeg: libavcodec/x86/h264_qpel

00001 /*
00002  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include "dsputil_mmx.h"
00022 
00023 /***********************************/
00024 /* motion compensation */
00025 
00026 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
00027         "mov"#q" "#C", "#T"         \n\t"\
00028         "mov"#d" (%0), "#F"         \n\t"\
00029         "paddw "#D", "#T"           \n\t"\
00030         "psllw $2, "#T"             \n\t"\
00031         "psubw "#B", "#T"           \n\t"\
00032         "psubw "#E", "#T"           \n\t"\
00033         "punpcklbw "#Z", "#F"       \n\t"\
00034         "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
00035         "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
00036         "add %2, %0                 \n\t"\
00037         "paddw "#F", "#A"           \n\t"\
00038         "paddw "#A", "#T"           \n\t"\
00039         "psraw $5, "#T"             \n\t"\
00040         "packuswb "#T", "#T"        \n\t"\
00041         OP(T, (%1), A, d)\
00042         "add %3, %1                 \n\t"
00043 
00044 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
00045         "mov"#q" "#C", "#T"         \n\t"\
00046         "mov"#d" (%0), "#F"         \n\t"\
00047         "paddw "#D", "#T"           \n\t"\
00048         "psllw $2, "#T"             \n\t"\
00049         "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
00050         "psubw "#B", "#T"           \n\t"\
00051         "psubw "#E", "#T"           \n\t"\
00052         "punpcklbw "#Z", "#F"       \n\t"\
00053         "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
00054         "paddw "#F", "#A"           \n\t"\
00055         "add %2, %0                 \n\t"\
00056         "paddw "#A", "#T"           \n\t"\
00057         "mov"#q" "#T", "#OF"(%1)    \n\t"
00058 
00059 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
00060 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
00061 #define QPEL_H264V_XMM(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%xmm6,%%xmm7,q,dqa)
00062 #define QPEL_H264HV_XMM(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%xmm6,%%xmm7,q,dqa)
00063 
00064 
00065 #define QPEL_H264(OPNAME, OP, MMX)\
00066 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00067     int h=4;\
00068 \
00069     __asm__ volatile(\
00070         "pxor %%mm7, %%mm7          \n\t"\
00071         "movq "MANGLE(ff_pw_5) ", %%mm4\n\t"\
00072         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00073         "1:                         \n\t"\
00074         "movd  -1(%0), %%mm1        \n\t"\
00075         "movd    (%0), %%mm2        \n\t"\
00076         "movd   1(%0), %%mm3        \n\t"\
00077         "movd   2(%0), %%mm0        \n\t"\
00078         "punpcklbw %%mm7, %%mm1     \n\t"\
00079         "punpcklbw %%mm7, %%mm2     \n\t"\
00080         "punpcklbw %%mm7, %%mm3     \n\t"\
00081         "punpcklbw %%mm7, %%mm0     \n\t"\
00082         "paddw %%mm0, %%mm1         \n\t"\
00083         "paddw %%mm3, %%mm2         \n\t"\
00084         "movd  -2(%0), %%mm0        \n\t"\
00085         "movd   3(%0), %%mm3        \n\t"\
00086         "punpcklbw %%mm7, %%mm0     \n\t"\
00087         "punpcklbw %%mm7, %%mm3     \n\t"\
00088         "paddw %%mm3, %%mm0         \n\t"\
00089         "psllw $2, %%mm2            \n\t"\
00090         "psubw %%mm1, %%mm2         \n\t"\
00091         "pmullw %%mm4, %%mm2        \n\t"\
00092         "paddw %%mm5, %%mm0         \n\t"\
00093         "paddw %%mm2, %%mm0         \n\t"\
00094         "psraw $5, %%mm0            \n\t"\
00095         "packuswb %%mm0, %%mm0      \n\t"\
00096         OP(%%mm0, (%1),%%mm6, d)\
00097         "add %3, %0                 \n\t"\
00098         "add %4, %1                 \n\t"\
00099         "decl %2                    \n\t"\
00100         " jnz 1b                    \n\t"\
00101         : "+a"(src), "+c"(dst), "+g"(h)\
00102         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00103         : "memory"\
00104     );\
00105 }\
00106 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00107     int h=4;\
00108     __asm__ volatile(\
00109         "pxor %%mm7, %%mm7          \n\t"\
00110         "movq %0, %%mm4             \n\t"\
00111         "movq %1, %%mm5             \n\t"\
00112         :: "m"(ff_pw_5), "m"(ff_pw_16)\
00113     );\
00114     do{\
00115     __asm__ volatile(\
00116         "movd  -1(%0), %%mm1        \n\t"\
00117         "movd    (%0), %%mm2        \n\t"\
00118         "movd   1(%0), %%mm3        \n\t"\
00119         "movd   2(%0), %%mm0        \n\t"\
00120         "punpcklbw %%mm7, %%mm1     \n\t"\
00121         "punpcklbw %%mm7, %%mm2     \n\t"\
00122         "punpcklbw %%mm7, %%mm3     \n\t"\
00123         "punpcklbw %%mm7, %%mm0     \n\t"\
00124         "paddw %%mm0, %%mm1         \n\t"\
00125         "paddw %%mm3, %%mm2         \n\t"\
00126         "movd  -2(%0), %%mm0        \n\t"\
00127         "movd   3(%0), %%mm3        \n\t"\
00128         "punpcklbw %%mm7, %%mm0     \n\t"\
00129         "punpcklbw %%mm7, %%mm3     \n\t"\
00130         "paddw %%mm3, %%mm0         \n\t"\
00131         "psllw $2, %%mm2            \n\t"\
00132         "psubw %%mm1, %%mm2         \n\t"\
00133         "pmullw %%mm4, %%mm2        \n\t"\
00134         "paddw %%mm5, %%mm0         \n\t"\
00135         "paddw %%mm2, %%mm0         \n\t"\
00136         "movd   (%2), %%mm3         \n\t"\
00137         "psraw $5, %%mm0            \n\t"\
00138         "packuswb %%mm0, %%mm0      \n\t"\
00139         PAVGB" %%mm3, %%mm0         \n\t"\
00140         OP(%%mm0, (%1),%%mm6, d)\
00141         "add %4, %0                 \n\t"\
00142         "add %4, %1                 \n\t"\
00143         "add %3, %2                 \n\t"\
00144         : "+a"(src), "+c"(dst), "+d"(src2)\
00145         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00146         : "memory"\
00147     );\
00148     }while(--h);\
00149 }\
00150 static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00151     src -= 2*srcStride;\
00152     __asm__ volatile(\
00153         "pxor %%mm7, %%mm7          \n\t"\
00154         "movd (%0), %%mm0           \n\t"\
00155         "add %2, %0                 \n\t"\
00156         "movd (%0), %%mm1           \n\t"\
00157         "add %2, %0                 \n\t"\
00158         "movd (%0), %%mm2           \n\t"\
00159         "add %2, %0                 \n\t"\
00160         "movd (%0), %%mm3           \n\t"\
00161         "add %2, %0                 \n\t"\
00162         "movd (%0), %%mm4           \n\t"\
00163         "add %2, %0                 \n\t"\
00164         "punpcklbw %%mm7, %%mm0     \n\t"\
00165         "punpcklbw %%mm7, %%mm1     \n\t"\
00166         "punpcklbw %%mm7, %%mm2     \n\t"\
00167         "punpcklbw %%mm7, %%mm3     \n\t"\
00168         "punpcklbw %%mm7, %%mm4     \n\t"\
00169         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00170         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00171         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00172         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00173          \
00174         : "+a"(src), "+c"(dst)\
00175         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00176         : "memory"\
00177     );\
00178 }\
00179 static av_noinline void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00180     int h=4;\
00181     int w=3;\
00182     src -= 2*srcStride+2;\
00183     while(w--){\
00184         __asm__ volatile(\
00185             "pxor %%mm7, %%mm7      \n\t"\
00186             "movd (%0), %%mm0       \n\t"\
00187             "add %2, %0             \n\t"\
00188             "movd (%0), %%mm1       \n\t"\
00189             "add %2, %0             \n\t"\
00190             "movd (%0), %%mm2       \n\t"\
00191             "add %2, %0             \n\t"\
00192             "movd (%0), %%mm3       \n\t"\
00193             "add %2, %0             \n\t"\
00194             "movd (%0), %%mm4       \n\t"\
00195             "add %2, %0             \n\t"\
00196             "punpcklbw %%mm7, %%mm0 \n\t"\
00197             "punpcklbw %%mm7, %%mm1 \n\t"\
00198             "punpcklbw %%mm7, %%mm2 \n\t"\
00199             "punpcklbw %%mm7, %%mm3 \n\t"\
00200             "punpcklbw %%mm7, %%mm4 \n\t"\
00201             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
00202             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
00203             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
00204             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
00205              \
00206             : "+a"(src)\
00207             : "c"(tmp), "S"((x86_reg)srcStride)\
00208             : "memory"\
00209         );\
00210         tmp += 4;\
00211         src += 4 - 9*srcStride;\
00212     }\
00213     tmp -= 3*4;\
00214     __asm__ volatile(\
00215         "1:                         \n\t"\
00216         "movq     (%0), %%mm0       \n\t"\
00217         "paddw  10(%0), %%mm0       \n\t"\
00218         "movq    2(%0), %%mm1       \n\t"\
00219         "paddw   8(%0), %%mm1       \n\t"\
00220         "movq    4(%0), %%mm2       \n\t"\
00221         "paddw   6(%0), %%mm2       \n\t"\
00222         "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
00223         "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
00224         "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
00225         "paddsw %%mm2, %%mm0        \n\t"\
00226         "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
00227         "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
00228         "psraw $6, %%mm0            \n\t"\
00229         "packuswb %%mm0, %%mm0      \n\t"\
00230         OP(%%mm0, (%1),%%mm7, d)\
00231         "add $24, %0                \n\t"\
00232         "add %3, %1                 \n\t"\
00233         "decl %2                    \n\t"\
00234         " jnz 1b                    \n\t"\
00235         : "+a"(tmp), "+c"(dst), "+g"(h)\
00236         : "S"((x86_reg)dstStride)\
00237         : "memory"\
00238     );\
00239 }\
00240 \
00241 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00242     int h=8;\
00243     __asm__ volatile(\
00244         "pxor %%mm7, %%mm7          \n\t"\
00245         "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
00246         "1:                         \n\t"\
00247         "movq    (%0), %%mm0        \n\t"\
00248         "movq   1(%0), %%mm2        \n\t"\
00249         "movq %%mm0, %%mm1          \n\t"\
00250         "movq %%mm2, %%mm3          \n\t"\
00251         "punpcklbw %%mm7, %%mm0     \n\t"\
00252         "punpckhbw %%mm7, %%mm1     \n\t"\
00253         "punpcklbw %%mm7, %%mm2     \n\t"\
00254         "punpckhbw %%mm7, %%mm3     \n\t"\
00255         "paddw %%mm2, %%mm0         \n\t"\
00256         "paddw %%mm3, %%mm1         \n\t"\
00257         "psllw $2, %%mm0            \n\t"\
00258         "psllw $2, %%mm1            \n\t"\
00259         "movq   -1(%0), %%mm2       \n\t"\
00260         "movq    2(%0), %%mm4       \n\t"\
00261         "movq %%mm2, %%mm3          \n\t"\
00262         "movq %%mm4, %%mm5          \n\t"\
00263         "punpcklbw %%mm7, %%mm2     \n\t"\
00264         "punpckhbw %%mm7, %%mm3     \n\t"\
00265         "punpcklbw %%mm7, %%mm4     \n\t"\
00266         "punpckhbw %%mm7, %%mm5     \n\t"\
00267         "paddw %%mm4, %%mm2         \n\t"\
00268         "paddw %%mm3, %%mm5         \n\t"\
00269         "psubw %%mm2, %%mm0         \n\t"\
00270         "psubw %%mm5, %%mm1         \n\t"\
00271         "pmullw %%mm6, %%mm0        \n\t"\
00272         "pmullw %%mm6, %%mm1        \n\t"\
00273         "movd   -2(%0), %%mm2       \n\t"\
00274         "movd    7(%0), %%mm5       \n\t"\
00275         "punpcklbw %%mm7, %%mm2     \n\t"\
00276         "punpcklbw %%mm7, %%mm5     \n\t"\
00277         "paddw %%mm3, %%mm2         \n\t"\
00278         "paddw %%mm5, %%mm4         \n\t"\
00279         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00280         "paddw %%mm5, %%mm2         \n\t"\
00281         "paddw %%mm5, %%mm4         \n\t"\
00282         "paddw %%mm2, %%mm0         \n\t"\
00283         "paddw %%mm4, %%mm1         \n\t"\
00284         "psraw $5, %%mm0            \n\t"\
00285         "psraw $5, %%mm1            \n\t"\
00286         "packuswb %%mm1, %%mm0      \n\t"\
00287         OP(%%mm0, (%1),%%mm5, q)\
00288         "add %3, %0                 \n\t"\
00289         "add %4, %1                 \n\t"\
00290         "decl %2                    \n\t"\
00291         " jnz 1b                    \n\t"\
00292         : "+a"(src), "+c"(dst), "+g"(h)\
00293         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00294         : "memory"\
00295     );\
00296 }\
00297 \
00298 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00299     int h=8;\
00300     __asm__ volatile(\
00301         "pxor %%mm7, %%mm7          \n\t"\
00302         "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
00303         "1:                         \n\t"\
00304         "movq    (%0), %%mm0        \n\t"\
00305         "movq   1(%0), %%mm2        \n\t"\
00306         "movq %%mm0, %%mm1          \n\t"\
00307         "movq %%mm2, %%mm3          \n\t"\
00308         "punpcklbw %%mm7, %%mm0     \n\t"\
00309         "punpckhbw %%mm7, %%mm1     \n\t"\
00310         "punpcklbw %%mm7, %%mm2     \n\t"\
00311         "punpckhbw %%mm7, %%mm3     \n\t"\
00312         "paddw %%mm2, %%mm0         \n\t"\
00313         "paddw %%mm3, %%mm1         \n\t"\
00314         "psllw $2, %%mm0            \n\t"\
00315         "psllw $2, %%mm1            \n\t"\
00316         "movq   -1(%0), %%mm2       \n\t"\
00317         "movq    2(%0), %%mm4       \n\t"\
00318         "movq %%mm2, %%mm3          \n\t"\
00319         "movq %%mm4, %%mm5          \n\t"\
00320         "punpcklbw %%mm7, %%mm2     \n\t"\
00321         "punpckhbw %%mm7, %%mm3     \n\t"\
00322         "punpcklbw %%mm7, %%mm4     \n\t"\
00323         "punpckhbw %%mm7, %%mm5     \n\t"\
00324         "paddw %%mm4, %%mm2         \n\t"\
00325         "paddw %%mm3, %%mm5         \n\t"\
00326         "psubw %%mm2, %%mm0         \n\t"\
00327         "psubw %%mm5, %%mm1         \n\t"\
00328         "pmullw %%mm6, %%mm0        \n\t"\
00329         "pmullw %%mm6, %%mm1        \n\t"\
00330         "movd   -2(%0), %%mm2       \n\t"\
00331         "movd    7(%0), %%mm5       \n\t"\
00332         "punpcklbw %%mm7, %%mm2     \n\t"\
00333         "punpcklbw %%mm7, %%mm5     \n\t"\
00334         "paddw %%mm3, %%mm2         \n\t"\
00335         "paddw %%mm5, %%mm4         \n\t"\
00336         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00337         "paddw %%mm5, %%mm2         \n\t"\
00338         "paddw %%mm5, %%mm4         \n\t"\
00339         "paddw %%mm2, %%mm0         \n\t"\
00340         "paddw %%mm4, %%mm1         \n\t"\
00341         "psraw $5, %%mm0            \n\t"\
00342         "psraw $5, %%mm1            \n\t"\
00343         "movq (%2), %%mm4           \n\t"\
00344         "packuswb %%mm1, %%mm0      \n\t"\
00345         PAVGB" %%mm4, %%mm0         \n\t"\
00346         OP(%%mm0, (%1),%%mm5, q)\
00347         "add %5, %0                 \n\t"\
00348         "add %5, %1                 \n\t"\
00349         "add %4, %2                 \n\t"\
00350         "decl %3                    \n\t"\
00351         "jg 1b                      \n\t"\
00352         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00353         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00354         : "memory"\
00355     );\
00356 }\
00357 \
00358 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00359     int w= 2;\
00360     src -= 2*srcStride;\
00361     \
00362     while(w--){\
00363         __asm__ volatile(\
00364             "pxor %%mm7, %%mm7          \n\t"\
00365             "movd (%0), %%mm0           \n\t"\
00366             "add %2, %0                 \n\t"\
00367             "movd (%0), %%mm1           \n\t"\
00368             "add %2, %0                 \n\t"\
00369             "movd (%0), %%mm2           \n\t"\
00370             "add %2, %0                 \n\t"\
00371             "movd (%0), %%mm3           \n\t"\
00372             "add %2, %0                 \n\t"\
00373             "movd (%0), %%mm4           \n\t"\
00374             "add %2, %0                 \n\t"\
00375             "punpcklbw %%mm7, %%mm0     \n\t"\
00376             "punpcklbw %%mm7, %%mm1     \n\t"\
00377             "punpcklbw %%mm7, %%mm2     \n\t"\
00378             "punpcklbw %%mm7, %%mm3     \n\t"\
00379             "punpcklbw %%mm7, %%mm4     \n\t"\
00380             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00381             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00382             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00383             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00384             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00385             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00386             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00387             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00388             "cmpl $16, %4               \n\t"\
00389             "jne 2f                     \n\t"\
00390             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00391             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00392             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00393             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00394             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00395             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00396             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00397             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00398             "2:                         \n\t"\
00399             \
00400             : "+a"(src), "+c"(dst)\
00401             : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
00402             : "memory"\
00403         );\
00404         src += 4-(h+5)*srcStride;\
00405         dst += 4-h*dstStride;\
00406     }\
00407 }\
00408 static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
00409     int w = (size+8)>>2;\
00410     src -= 2*srcStride+2;\
00411     while(w--){\
00412         __asm__ volatile(\
00413             "pxor %%mm7, %%mm7      \n\t"\
00414             "movd (%0), %%mm0       \n\t"\
00415             "add %2, %0             \n\t"\
00416             "movd (%0), %%mm1       \n\t"\
00417             "add %2, %0             \n\t"\
00418             "movd (%0), %%mm2       \n\t"\
00419             "add %2, %0             \n\t"\
00420             "movd (%0), %%mm3       \n\t"\
00421             "add %2, %0             \n\t"\
00422             "movd (%0), %%mm4       \n\t"\
00423             "add %2, %0             \n\t"\
00424             "punpcklbw %%mm7, %%mm0 \n\t"\
00425             "punpcklbw %%mm7, %%mm1 \n\t"\
00426             "punpcklbw %%mm7, %%mm2 \n\t"\
00427             "punpcklbw %%mm7, %%mm3 \n\t"\
00428             "punpcklbw %%mm7, %%mm4 \n\t"\
00429             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
00430             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
00431             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
00432             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
00433             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
00434             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
00435             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
00436             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
00437             "cmpl $16, %3           \n\t"\
00438             "jne 2f                 \n\t"\
00439             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
00440             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
00441             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
00442             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
00443             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
00444             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
00445             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
00446             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
00447             "2:                     \n\t"\
00448             : "+a"(src)\
00449             : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)\
00450             : "memory"\
00451             );\
00452         tmp += 4;\
00453         src += 4 - (size+5)*srcStride;\
00454     }\
00455 }\
00456 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
00457     int w = size>>4;\
00458     do{\
00459     int h = size;\
00460     __asm__ volatile(\
00461         "1:                         \n\t"\
00462         "movq     (%0), %%mm0       \n\t"\
00463         "movq    8(%0), %%mm3       \n\t"\
00464         "movq    2(%0), %%mm1       \n\t"\
00465         "movq   10(%0), %%mm4       \n\t"\
00466         "paddw   %%mm4, %%mm0       \n\t"\
00467         "paddw   %%mm3, %%mm1       \n\t"\
00468         "paddw  18(%0), %%mm3       \n\t"\
00469         "paddw  16(%0), %%mm4       \n\t"\
00470         "movq    4(%0), %%mm2       \n\t"\
00471         "movq   12(%0), %%mm5       \n\t"\
00472         "paddw   6(%0), %%mm2       \n\t"\
00473         "paddw  14(%0), %%mm5       \n\t"\
00474         "psubw %%mm1, %%mm0         \n\t"\
00475         "psubw %%mm4, %%mm3         \n\t"\
00476         "psraw $2, %%mm0            \n\t"\
00477         "psraw $2, %%mm3            \n\t"\
00478         "psubw %%mm1, %%mm0         \n\t"\
00479         "psubw %%mm4, %%mm3         \n\t"\
00480         "paddsw %%mm2, %%mm0        \n\t"\
00481         "paddsw %%mm5, %%mm3        \n\t"\
00482         "psraw $2, %%mm0            \n\t"\
00483         "psraw $2, %%mm3            \n\t"\
00484         "paddw %%mm2, %%mm0         \n\t"\
00485         "paddw %%mm5, %%mm3         \n\t"\
00486         "psraw $6, %%mm0            \n\t"\
00487         "psraw $6, %%mm3            \n\t"\
00488         "packuswb %%mm3, %%mm0      \n\t"\
00489         OP(%%mm0, (%1),%%mm7, q)\
00490         "add $48, %0                \n\t"\
00491         "add %3, %1                 \n\t"\
00492         "decl %2                    \n\t"\
00493         " jnz 1b                    \n\t"\
00494         : "+a"(tmp), "+c"(dst), "+g"(h)\
00495         : "S"((x86_reg)dstStride)\
00496         : "memory"\
00497     );\
00498     tmp += 8 - size*24;\
00499     dst += 8 - size*dstStride;\
00500     }while(w--);\
00501 }\
00502 \
00503 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00504     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00505 }\
00506 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00507     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00508     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00509 }\
00510 \
00511 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00512     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00513     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00514     src += 8*srcStride;\
00515     dst += 8*dstStride;\
00516     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00517     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00518 }\
00519 \
00520 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00521     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00522     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00523     src += 8*dstStride;\
00524     dst += 8*dstStride;\
00525     src2 += 8*src2Stride;\
00526     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00527     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00528 }\
00529 \
00530 static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
00531           put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
00532     OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
00533 }\
00534 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00535     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
00536 }\
00537 \
00538 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00539     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
00540 }\
00541 \
00542 static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00543 {\
00544     __asm__ volatile(\
00545         "movq      (%1), %%mm0          \n\t"\
00546         "movq    24(%1), %%mm1          \n\t"\
00547         "psraw      $5,  %%mm0          \n\t"\
00548         "psraw      $5,  %%mm1          \n\t"\
00549         "packuswb %%mm0, %%mm0          \n\t"\
00550         "packuswb %%mm1, %%mm1          \n\t"\
00551         PAVGB"     (%0), %%mm0          \n\t"\
00552         PAVGB"  (%0,%3), %%mm1          \n\t"\
00553         OP(%%mm0, (%2),    %%mm4, d)\
00554         OP(%%mm1, (%2,%4), %%mm5, d)\
00555         "lea  (%0,%3,2), %0             \n\t"\
00556         "lea  (%2,%4,2), %2             \n\t"\
00557         "movq    48(%1), %%mm0          \n\t"\
00558         "movq    72(%1), %%mm1          \n\t"\
00559         "psraw      $5,  %%mm0          \n\t"\
00560         "psraw      $5,  %%mm1          \n\t"\
00561         "packuswb %%mm0, %%mm0          \n\t"\
00562         "packuswb %%mm1, %%mm1          \n\t"\
00563         PAVGB"     (%0), %%mm0          \n\t"\
00564         PAVGB"  (%0,%3), %%mm1          \n\t"\
00565         OP(%%mm0, (%2),    %%mm4, d)\
00566         OP(%%mm1, (%2,%4), %%mm5, d)\
00567         :"+a"(src8), "+c"(src16), "+d"(dst)\
00568         :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
00569         :"memory");\
00570 }\
00571 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00572 {\
00573     do{\
00574     __asm__ volatile(\
00575         "movq      (%1), %%mm0          \n\t"\
00576         "movq     8(%1), %%mm1          \n\t"\
00577         "movq    48(%1), %%mm2          \n\t"\
00578         "movq  8+48(%1), %%mm3          \n\t"\
00579         "psraw      $5,  %%mm0          \n\t"\
00580         "psraw      $5,  %%mm1          \n\t"\
00581         "psraw      $5,  %%mm2          \n\t"\
00582         "psraw      $5,  %%mm3          \n\t"\
00583         "packuswb %%mm1, %%mm0          \n\t"\
00584         "packuswb %%mm3, %%mm2          \n\t"\
00585         PAVGB"     (%0), %%mm0          \n\t"\
00586         PAVGB"  (%0,%3), %%mm2          \n\t"\
00587         OP(%%mm0, (%2), %%mm5, q)\
00588         OP(%%mm2, (%2,%4), %%mm5, q)\
00589         ::"a"(src8), "c"(src16), "d"(dst),\
00590           "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
00591         :"memory");\
00592         src8 += 2L*src8Stride;\
00593         src16 += 48;\
00594         dst += 2L*dstStride;\
00595     }while(h-=2);\
00596 }\
00597 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00598 {\
00599     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
00600     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
00601 }\
00602 
00603 
00604 #if ARCH_X86_64
00605 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00606 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00607     int h=16;\
00608     __asm__ volatile(\
00609         "pxor %%xmm15, %%xmm15      \n\t"\
00610         "movdqa %6, %%xmm14         \n\t"\
00611         "movdqa %7, %%xmm13         \n\t"\
00612         "1:                         \n\t"\
00613         "lddqu    6(%0), %%xmm1     \n\t"\
00614         "lddqu   -2(%0), %%xmm7     \n\t"\
00615         "movdqa  %%xmm1, %%xmm0     \n\t"\
00616         "punpckhbw %%xmm15, %%xmm1  \n\t"\
00617         "punpcklbw %%xmm15, %%xmm0  \n\t"\
00618         "punpcklbw %%xmm15, %%xmm7  \n\t"\
00619         "movdqa  %%xmm1, %%xmm2     \n\t"\
00620         "movdqa  %%xmm0, %%xmm6     \n\t"\
00621         "movdqa  %%xmm1, %%xmm3     \n\t"\
00622         "movdqa  %%xmm0, %%xmm8     \n\t"\
00623         "movdqa  %%xmm1, %%xmm4     \n\t"\
00624         "movdqa  %%xmm0, %%xmm9     \n\t"\
00625         "movdqa  %%xmm0, %%xmm12    \n\t"\
00626         "movdqa  %%xmm1, %%xmm11    \n\t"\
00627         "palignr $10,%%xmm0, %%xmm11\n\t"\
00628         "palignr $10,%%xmm7, %%xmm12\n\t"\
00629         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00630         "palignr $2, %%xmm7, %%xmm9 \n\t"\
00631         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00632         "palignr $4, %%xmm7, %%xmm8 \n\t"\
00633         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00634         "palignr $6, %%xmm7, %%xmm6 \n\t"\
00635         "paddw   %%xmm0 ,%%xmm11    \n\t"\
00636         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00637         "palignr $8, %%xmm7, %%xmm0 \n\t"\
00638         "paddw   %%xmm12,%%xmm7     \n\t"\
00639         "paddw   %%xmm3, %%xmm2     \n\t"\
00640         "paddw   %%xmm8, %%xmm6     \n\t"\
00641         "paddw   %%xmm4, %%xmm1     \n\t"\
00642         "paddw   %%xmm9, %%xmm0     \n\t"\
00643         "psllw   $2,     %%xmm2     \n\t"\
00644         "psllw   $2,     %%xmm6     \n\t"\
00645         "psubw   %%xmm1, %%xmm2     \n\t"\
00646         "psubw   %%xmm0, %%xmm6     \n\t"\
00647         "paddw   %%xmm13,%%xmm11    \n\t"\
00648         "paddw   %%xmm13,%%xmm7     \n\t"\
00649         "pmullw  %%xmm14,%%xmm2     \n\t"\
00650         "pmullw  %%xmm14,%%xmm6     \n\t"\
00651         "lddqu   (%2),   %%xmm3     \n\t"\
00652         "paddw   %%xmm11,%%xmm2     \n\t"\
00653         "paddw   %%xmm7, %%xmm6     \n\t"\
00654         "psraw   $5,     %%xmm2     \n\t"\
00655         "psraw   $5,     %%xmm6     \n\t"\
00656         "packuswb %%xmm2,%%xmm6     \n\t"\
00657         "pavgb   %%xmm3, %%xmm6     \n\t"\
00658         OP(%%xmm6, (%1), %%xmm4, dqa)\
00659         "add %5, %0                 \n\t"\
00660         "add %5, %1                 \n\t"\
00661         "add %4, %2                 \n\t"\
00662         "decl %3                    \n\t"\
00663         "jg 1b                      \n\t"\
00664         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00665         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
00666           "m"(ff_pw_5), "m"(ff_pw_16)\
00667         : XMM_CLOBBERS("%xmm0" , "%xmm1" , "%xmm2" , "%xmm3" , \
00668                        "%xmm4" , "%xmm5" , "%xmm6" , "%xmm7" , \
00669                        "%xmm8" , "%xmm9" , "%xmm10", "%xmm11", \
00670                        "%xmm12", "%xmm13", "%xmm14", "%xmm15",)\
00671           "memory"\
00672     );\
00673 }
00674 #else // ARCH_X86_64
00675 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00676 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00677     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00678     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00679     src += 8*dstStride;\
00680     dst += 8*dstStride;\
00681     src2 += 8*src2Stride;\
00682     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00683     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00684 }
00685 #endif // ARCH_X86_64
00686 
00687 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
00688 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00689     int h=8;\
00690     __asm__ volatile(\
00691         "pxor %%xmm7, %%xmm7        \n\t"\
00692         "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
00693         "1:                         \n\t"\
00694         "lddqu   -2(%0), %%xmm1     \n\t"\
00695         "movdqa  %%xmm1, %%xmm0     \n\t"\
00696         "punpckhbw %%xmm7, %%xmm1   \n\t"\
00697         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00698         "movdqa  %%xmm1, %%xmm2     \n\t"\
00699         "movdqa  %%xmm1, %%xmm3     \n\t"\
00700         "movdqa  %%xmm1, %%xmm4     \n\t"\
00701         "movdqa  %%xmm1, %%xmm5     \n\t"\
00702         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00703         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00704         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00705         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00706         "palignr $10,%%xmm0, %%xmm5 \n\t"\
00707         "paddw   %%xmm5, %%xmm0     \n\t"\
00708         "paddw   %%xmm3, %%xmm2     \n\t"\
00709         "paddw   %%xmm4, %%xmm1     \n\t"\
00710         "psllw   $2,     %%xmm2     \n\t"\
00711         "movq    (%2),   %%xmm3     \n\t"\
00712         "psubw   %%xmm1, %%xmm2     \n\t"\
00713         "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
00714         "pmullw  %%xmm6, %%xmm2     \n\t"\
00715         "paddw   %%xmm0, %%xmm2     \n\t"\
00716         "psraw   $5,     %%xmm2     \n\t"\
00717         "packuswb %%xmm2, %%xmm2    \n\t"\
00718         "pavgb   %%xmm3, %%xmm2     \n\t"\
00719         OP(%%xmm2, (%1), %%xmm4, q)\
00720         "add %5, %0                 \n\t"\
00721         "add %5, %1                 \n\t"\
00722         "add %4, %2                 \n\t"\
00723         "decl %3                    \n\t"\
00724         "jg 1b                      \n\t"\
00725         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00726         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00727         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00728                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00729           "memory"\
00730     );\
00731 }\
00732 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00733 \
00734 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00735     int h=8;\
00736     __asm__ volatile(\
00737         "pxor %%xmm7, %%xmm7        \n\t"\
00738         "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
00739         "1:                         \n\t"\
00740         "lddqu   -2(%0), %%xmm1     \n\t"\
00741         "movdqa  %%xmm1, %%xmm0     \n\t"\
00742         "punpckhbw %%xmm7, %%xmm1   \n\t"\
00743         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00744         "movdqa  %%xmm1, %%xmm2     \n\t"\
00745         "movdqa  %%xmm1, %%xmm3     \n\t"\
00746         "movdqa  %%xmm1, %%xmm4     \n\t"\
00747         "movdqa  %%xmm1, %%xmm5     \n\t"\
00748         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00749         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00750         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00751         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00752         "palignr $10,%%xmm0, %%xmm5 \n\t"\
00753         "paddw   %%xmm5, %%xmm0     \n\t"\
00754         "paddw   %%xmm3, %%xmm2     \n\t"\
00755         "paddw   %%xmm4, %%xmm1     \n\t"\
00756         "psllw   $2,     %%xmm2     \n\t"\
00757         "psubw   %%xmm1, %%xmm2     \n\t"\
00758         "paddw   "MANGLE(ff_pw_16)", %%xmm0\n\t"\
00759         "pmullw  %%xmm6, %%xmm2     \n\t"\
00760         "paddw   %%xmm0, %%xmm2     \n\t"\
00761         "psraw   $5,     %%xmm2     \n\t"\
00762         "packuswb %%xmm2, %%xmm2    \n\t"\
00763         OP(%%xmm2, (%1), %%xmm4, q)\
00764         "add %3, %0                 \n\t"\
00765         "add %4, %1                 \n\t"\
00766         "decl %2                    \n\t"\
00767         " jnz 1b                    \n\t"\
00768         : "+a"(src), "+c"(dst), "+g"(h)\
00769         : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00770         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00771                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00772           "memory"\
00773     );\
00774 }\
00775 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00776     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00777     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00778     src += 8*srcStride;\
00779     dst += 8*dstStride;\
00780     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00781     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00782 }\
00783 
00784 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
00785 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00786     src -= 2*srcStride;\
00787     \
00788     __asm__ volatile(\
00789         "pxor %%xmm7, %%xmm7        \n\t"\
00790         "movq (%0), %%xmm0          \n\t"\
00791         "add %2, %0                 \n\t"\
00792         "movq (%0), %%xmm1          \n\t"\
00793         "add %2, %0                 \n\t"\
00794         "movq (%0), %%xmm2          \n\t"\
00795         "add %2, %0                 \n\t"\
00796         "movq (%0), %%xmm3          \n\t"\
00797         "add %2, %0                 \n\t"\
00798         "movq (%0), %%xmm4          \n\t"\
00799         "add %2, %0                 \n\t"\
00800         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00801         "punpcklbw %%xmm7, %%xmm1   \n\t"\
00802         "punpcklbw %%xmm7, %%xmm2   \n\t"\
00803         "punpcklbw %%xmm7, %%xmm3   \n\t"\
00804         "punpcklbw %%xmm7, %%xmm4   \n\t"\
00805         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00806         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00807         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00808         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00809         QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
00810         QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
00811         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00812         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00813         "cmpl $16, %4               \n\t"\
00814         "jne 2f                     \n\t"\
00815         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00816         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00817         QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
00818         QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
00819         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00820         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00821         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00822         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00823         "2:                          \n\t"\
00824         \
00825         : "+a"(src), "+c"(dst)\
00826         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
00827         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00828                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00829           "memory"\
00830     );\
00831 }\
00832 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00833     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00834 }\
00835 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00836     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00837     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00838 }
00839 
00840 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
00841     int w = (size+8)>>3;
00842     src -= 2*srcStride+2;
00843     while(w--){
00844         __asm__ volatile(
00845             "pxor %%xmm7, %%xmm7        \n\t"
00846             "movq (%0), %%xmm0          \n\t"
00847             "add %2, %0                 \n\t"
00848             "movq (%0), %%xmm1          \n\t"
00849             "add %2, %0                 \n\t"
00850             "movq (%0), %%xmm2          \n\t"
00851             "add %2, %0                 \n\t"
00852             "movq (%0), %%xmm3          \n\t"
00853             "add %2, %0                 \n\t"
00854             "movq (%0), %%xmm4          \n\t"
00855             "add %2, %0                 \n\t"
00856             "punpcklbw %%xmm7, %%xmm0   \n\t"
00857             "punpcklbw %%xmm7, %%xmm1   \n\t"
00858             "punpcklbw %%xmm7, %%xmm2   \n\t"
00859             "punpcklbw %%xmm7, %%xmm3   \n\t"
00860             "punpcklbw %%xmm7, %%xmm4   \n\t"
00861             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
00862             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
00863             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
00864             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
00865             QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
00866             QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
00867             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
00868             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
00869             "cmpl $16, %3               \n\t"
00870             "jne 2f                     \n\t"
00871             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
00872             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
00873             QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
00874             QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
00875             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
00876             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
00877             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
00878             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
00879             "2:                         \n\t"
00880             : "+a"(src)
00881             : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)
00882             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3",
00883                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)
00884               "memory"
00885         );
00886         tmp += 8;
00887         src += 8 - (size+5)*srcStride;
00888     }
00889 }
00890 
00891 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
00892 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
00893     int h = size;\
00894     if(size == 16){\
00895         __asm__ volatile(\
00896             "1:                         \n\t"\
00897             "movdqa 32(%0), %%xmm4      \n\t"\
00898             "movdqa 16(%0), %%xmm5      \n\t"\
00899             "movdqa   (%0), %%xmm7      \n\t"\
00900             "movdqa %%xmm4, %%xmm3      \n\t"\
00901             "movdqa %%xmm4, %%xmm2      \n\t"\
00902             "movdqa %%xmm4, %%xmm1      \n\t"\
00903             "movdqa %%xmm4, %%xmm0      \n\t"\
00904             "palignr $10, %%xmm5, %%xmm0 \n\t"\
00905             "palignr  $8, %%xmm5, %%xmm1 \n\t"\
00906             "palignr  $6, %%xmm5, %%xmm2 \n\t"\
00907             "palignr  $4, %%xmm5, %%xmm3 \n\t"\
00908             "palignr  $2, %%xmm5, %%xmm4 \n\t"\
00909             "paddw  %%xmm5, %%xmm0      \n\t"\
00910             "paddw  %%xmm4, %%xmm1      \n\t"\
00911             "paddw  %%xmm3, %%xmm2      \n\t"\
00912             "movdqa %%xmm5, %%xmm6      \n\t"\
00913             "movdqa %%xmm5, %%xmm4      \n\t"\
00914             "movdqa %%xmm5, %%xmm3      \n\t"\
00915             "palignr  $8, %%xmm7, %%xmm4 \n\t"\
00916             "palignr  $2, %%xmm7, %%xmm6 \n\t"\
00917             "palignr $10, %%xmm7, %%xmm3 \n\t"\
00918             "paddw  %%xmm6, %%xmm4      \n\t"\
00919             "movdqa %%xmm5, %%xmm6      \n\t"\
00920             "palignr  $6, %%xmm7, %%xmm5 \n\t"\
00921             "palignr  $4, %%xmm7, %%xmm6 \n\t"\
00922             "paddw  %%xmm7, %%xmm3      \n\t"\
00923             "paddw  %%xmm6, %%xmm5      \n\t"\
00924             \
00925             "psubw  %%xmm1, %%xmm0      \n\t"\
00926             "psubw  %%xmm4, %%xmm3      \n\t"\
00927             "psraw      $2, %%xmm0      \n\t"\
00928             "psraw      $2, %%xmm3      \n\t"\
00929             "psubw  %%xmm1, %%xmm0      \n\t"\
00930             "psubw  %%xmm4, %%xmm3      \n\t"\
00931             "paddw  %%xmm2, %%xmm0      \n\t"\
00932             "paddw  %%xmm5, %%xmm3      \n\t"\
00933             "psraw      $2, %%xmm0      \n\t"\
00934             "psraw      $2, %%xmm3      \n\t"\
00935             "paddw  %%xmm2, %%xmm0      \n\t"\
00936             "paddw  %%xmm5, %%xmm3      \n\t"\
00937             "psraw      $6, %%xmm0      \n\t"\
00938             "psraw      $6, %%xmm3      \n\t"\
00939             "packuswb %%xmm0, %%xmm3    \n\t"\
00940             OP(%%xmm3, (%1), %%xmm7, dqa)\
00941             "add $48, %0                \n\t"\
00942             "add %3, %1                 \n\t"\
00943             "decl %2                    \n\t"\
00944             " jnz 1b                    \n\t"\
00945             : "+a"(tmp), "+c"(dst), "+g"(h)\
00946             : "S"((x86_reg)dstStride)\
00947             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00948                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00949               "memory"\
00950         );\
00951     }else{\
00952         __asm__ volatile(\
00953             "1:                         \n\t"\
00954             "movdqa 16(%0), %%xmm1      \n\t"\
00955             "movdqa   (%0), %%xmm0      \n\t"\
00956             "movdqa %%xmm1, %%xmm2      \n\t"\
00957             "movdqa %%xmm1, %%xmm3      \n\t"\
00958             "movdqa %%xmm1, %%xmm4      \n\t"\
00959             "movdqa %%xmm1, %%xmm5      \n\t"\
00960             "palignr $10, %%xmm0, %%xmm5 \n\t"\
00961             "palignr  $8, %%xmm0, %%xmm4 \n\t"\
00962             "palignr  $6, %%xmm0, %%xmm3 \n\t"\
00963             "palignr  $4, %%xmm0, %%xmm2 \n\t"\
00964             "palignr  $2, %%xmm0, %%xmm1 \n\t"\
00965             "paddw  %%xmm5, %%xmm0      \n\t"\
00966             "paddw  %%xmm4, %%xmm1      \n\t"\
00967             "paddw  %%xmm3, %%xmm2      \n\t"\
00968             "psubw  %%xmm1, %%xmm0      \n\t"\
00969             "psraw      $2, %%xmm0      \n\t"\
00970             "psubw  %%xmm1, %%xmm0      \n\t"\
00971             "paddw  %%xmm2, %%xmm0      \n\t"\
00972             "psraw      $2, %%xmm0      \n\t"\
00973             "paddw  %%xmm2, %%xmm0      \n\t"\
00974             "psraw      $6, %%xmm0      \n\t"\
00975             "packuswb %%xmm0, %%xmm0    \n\t"\
00976             OP(%%xmm0, (%1), %%xmm7, q)\
00977             "add $48, %0                \n\t"\
00978             "add %3, %1                 \n\t"\
00979             "decl %2                    \n\t"\
00980             " jnz 1b                    \n\t"\
00981             : "+a"(tmp), "+c"(dst), "+g"(h)\
00982             : "S"((x86_reg)dstStride)\
00983             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00984                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00985               "memory"\
00986         );\
00987     }\
00988 }
00989 
00990 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
00991 static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
00992           put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
00993     OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
00994 }\
00995 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00996     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
00997 }\
00998 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00999     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
01000 }\
01001 
01002 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
01003 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
01004 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
01005 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
01006 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
01007 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
01008 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
01009 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
01010 
01011 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
01012 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
01013 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
01014 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
01015 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
01016 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
01017 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
01018 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
01019 
01020 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
01021 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
01022 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
01023 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
01024 
01025 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
01026 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
01027 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
01028 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
01029 
01030 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
01031 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
01032 
01033 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
01034 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
01035 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
01036 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
01037 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
01038 
01039 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
01040     put_pixels16_sse2(dst, src, stride, 16);
01041 }
01042 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
01043     avg_pixels16_sse2(dst, src, stride, 16);
01044 }
01045 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
01046 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
01047 
01048 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
01049 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
01050     OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
01051 }\
01052 
01053 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
01054 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01055     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
01056 }\
01057 \
01058 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01059     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
01060 }\
01061 \
01062 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01063     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
01064 }\
01065 
01066 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
01067 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01068     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01069     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01070     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
01071 }\
01072 \
01073 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01074     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
01075 }\
01076 \
01077 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01078     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01079     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01080     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
01081 }\
01082 
01083 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
01084 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01085     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01086     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01087     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
01088 }\
01089 \
01090 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01091     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01092     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
01093     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
01094 }\
01095 \
01096 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01097     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01098     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01099     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
01100 }\
01101 \
01102 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01103     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01104     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
01105     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
01106 }\
01107 \
01108 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01109     DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
01110     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
01111 }\
01112 \
01113 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01114     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01115     uint8_t * const halfHV= temp;\
01116     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01117     assert(((int)temp & 7) == 0);\
01118     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01119     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
01120 }\
01121 \
01122 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01123     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01124     uint8_t * const halfHV= temp;\
01125     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01126     assert(((int)temp & 7) == 0);\
01127     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01128     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
01129 }\
01130 \
01131 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01132     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01133     uint8_t * const halfHV= temp;\
01134     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01135     assert(((int)temp & 7) == 0);\
01136     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01137     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
01138 }\
01139 \
01140 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01141     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01142     uint8_t * const halfHV= temp;\
01143     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01144     assert(((int)temp & 7) == 0);\
01145     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01146     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
01147 }\
01148 
01149 #define H264_MC_4816(MMX)\
01150 H264_MC(put_, 4, MMX, 8)\
01151 H264_MC(put_, 8, MMX, 8)\
01152 H264_MC(put_, 16,MMX, 8)\
01153 H264_MC(avg_, 4, MMX, 8)\
01154 H264_MC(avg_, 8, MMX, 8)\
01155 H264_MC(avg_, 16,MMX, 8)\
01156 
01157 #define H264_MC_816(QPEL, XMM)\
01158 QPEL(put_, 8, XMM, 16)\
01159 QPEL(put_, 16,XMM, 16)\
01160 QPEL(avg_, 8, XMM, 16)\
01161 QPEL(avg_, 16,XMM, 16)\
01162 
01163 
01164 #define AVG_3DNOW_OP(a,b,temp, size) \
01165 "mov" #size " " #b ", " #temp "   \n\t"\
01166 "pavgusb " #temp ", " #a "        \n\t"\
01167 "mov" #size " " #a ", " #b "      \n\t"
01168 #define AVG_MMX2_OP(a,b,temp, size) \
01169 "mov" #size " " #b ", " #temp "   \n\t"\
01170 "pavgb " #temp ", " #a "          \n\t"\
01171 "mov" #size " " #a ", " #b "      \n\t"
01172 
01173 #define PAVGB "pavgusb"
01174 QPEL_H264(put_,       PUT_OP, 3dnow)
01175 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
01176 #undef PAVGB
01177 #define PAVGB "pavgb"
01178 QPEL_H264(put_,       PUT_OP, mmx2)
01179 QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
01180 QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
01181 QPEL_H264_V_XMM(avg_,  AVG_MMX2_OP, sse2)
01182 QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
01183 QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, sse2)
01184 #if HAVE_SSSE3
01185 QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
01186 QPEL_H264_H_XMM(avg_,  AVG_MMX2_OP, ssse3)
01187 QPEL_H264_HV2_XMM(put_,       PUT_OP, ssse3)
01188 QPEL_H264_HV2_XMM(avg_,  AVG_MMX2_OP, ssse3)
01189 QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
01190 QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, ssse3)
01191 #endif
01192 #undef PAVGB
01193 
01194 H264_MC_4816(3dnow)
01195 H264_MC_4816(mmx2)
01196 H264_MC_816(H264_MC_V, sse2)
01197 H264_MC_816(H264_MC_HV, sse2)
01198 #if HAVE_SSSE3
01199 H264_MC_816(H264_MC_H, ssse3)
01200 H264_MC_816(H264_MC_HV, ssse3)
01201 #endif
libavcodec/x86/h264_qpel_mmx.c
