FFmpeg: libavcodec/x86/h264_qpel.c Source File

00001 /*
00002  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
00003  * Copyright (c) 2011 Daniel Kang
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00022 #include "dsputil_mmx.h"
00023 
00024 #if HAVE_INLINE_ASM
00025 
00026 /***********************************/
00027 /* motion compensation */
00028 
00029 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
00030         "mov"#q" "#C", "#T"         \n\t"\
00031         "mov"#d" (%0), "#F"         \n\t"\
00032         "paddw "#D", "#T"           \n\t"\
00033         "psllw $2, "#T"             \n\t"\
00034         "psubw "#B", "#T"           \n\t"\
00035         "psubw "#E", "#T"           \n\t"\
00036         "punpcklbw "#Z", "#F"       \n\t"\
00037         "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
00038         "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
00039         "add %2, %0                 \n\t"\
00040         "paddw "#F", "#A"           \n\t"\
00041         "paddw "#A", "#T"           \n\t"\
00042         "psraw $5, "#T"             \n\t"\
00043         "packuswb "#T", "#T"        \n\t"\
00044         OP(T, (%1), A, d)\
00045         "add %3, %1                 \n\t"
00046 
00047 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
00048         "mov"#q" "#C", "#T"         \n\t"\
00049         "mov"#d" (%0), "#F"         \n\t"\
00050         "paddw "#D", "#T"           \n\t"\
00051         "psllw $2, "#T"             \n\t"\
00052         "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
00053         "psubw "#B", "#T"           \n\t"\
00054         "psubw "#E", "#T"           \n\t"\
00055         "punpcklbw "#Z", "#F"       \n\t"\
00056         "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
00057         "paddw "#F", "#A"           \n\t"\
00058         "add %2, %0                 \n\t"\
00059         "paddw "#A", "#T"           \n\t"\
00060         "mov"#q" "#T", "#OF"(%1)    \n\t"
00061 
00062 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
00063 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
00064 #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)
00065 #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)
00066 
00067 
00068 #define QPEL_H264(OPNAME, OP, MMX)\
00069 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00070     int h=4;\
00071 \
00072     __asm__ volatile(\
00073         "pxor %%mm7, %%mm7          \n\t"\
00074         "movq "MANGLE(ff_pw_5) ", %%mm4\n\t"\
00075         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00076         "1:                         \n\t"\
00077         "movd  -1(%0), %%mm1        \n\t"\
00078         "movd    (%0), %%mm2        \n\t"\
00079         "movd   1(%0), %%mm3        \n\t"\
00080         "movd   2(%0), %%mm0        \n\t"\
00081         "punpcklbw %%mm7, %%mm1     \n\t"\
00082         "punpcklbw %%mm7, %%mm2     \n\t"\
00083         "punpcklbw %%mm7, %%mm3     \n\t"\
00084         "punpcklbw %%mm7, %%mm0     \n\t"\
00085         "paddw %%mm0, %%mm1         \n\t"\
00086         "paddw %%mm3, %%mm2         \n\t"\
00087         "movd  -2(%0), %%mm0        \n\t"\
00088         "movd   3(%0), %%mm3        \n\t"\
00089         "punpcklbw %%mm7, %%mm0     \n\t"\
00090         "punpcklbw %%mm7, %%mm3     \n\t"\
00091         "paddw %%mm3, %%mm0         \n\t"\
00092         "psllw $2, %%mm2            \n\t"\
00093         "psubw %%mm1, %%mm2         \n\t"\
00094         "pmullw %%mm4, %%mm2        \n\t"\
00095         "paddw %%mm5, %%mm0         \n\t"\
00096         "paddw %%mm2, %%mm0         \n\t"\
00097         "psraw $5, %%mm0            \n\t"\
00098         "packuswb %%mm0, %%mm0      \n\t"\
00099         OP(%%mm0, (%1),%%mm6, d)\
00100         "add %3, %0                 \n\t"\
00101         "add %4, %1                 \n\t"\
00102         "decl %2                    \n\t"\
00103         " jnz 1b                    \n\t"\
00104         : "+a"(src), "+c"(dst), "+g"(h)\
00105         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00106         : "memory"\
00107     );\
00108 }\
00109 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00110     int h=4;\
00111     __asm__ volatile(\
00112         "pxor %%mm7, %%mm7          \n\t"\
00113         "movq %0, %%mm4             \n\t"\
00114         "movq %1, %%mm5             \n\t"\
00115         :: "m"(ff_pw_5), "m"(ff_pw_16)\
00116     );\
00117     do{\
00118     __asm__ volatile(\
00119         "movd  -1(%0), %%mm1        \n\t"\
00120         "movd    (%0), %%mm2        \n\t"\
00121         "movd   1(%0), %%mm3        \n\t"\
00122         "movd   2(%0), %%mm0        \n\t"\
00123         "punpcklbw %%mm7, %%mm1     \n\t"\
00124         "punpcklbw %%mm7, %%mm2     \n\t"\
00125         "punpcklbw %%mm7, %%mm3     \n\t"\
00126         "punpcklbw %%mm7, %%mm0     \n\t"\
00127         "paddw %%mm0, %%mm1         \n\t"\
00128         "paddw %%mm3, %%mm2         \n\t"\
00129         "movd  -2(%0), %%mm0        \n\t"\
00130         "movd   3(%0), %%mm3        \n\t"\
00131         "punpcklbw %%mm7, %%mm0     \n\t"\
00132         "punpcklbw %%mm7, %%mm3     \n\t"\
00133         "paddw %%mm3, %%mm0         \n\t"\
00134         "psllw $2, %%mm2            \n\t"\
00135         "psubw %%mm1, %%mm2         \n\t"\
00136         "pmullw %%mm4, %%mm2        \n\t"\
00137         "paddw %%mm5, %%mm0         \n\t"\
00138         "paddw %%mm2, %%mm0         \n\t"\
00139         "movd   (%2), %%mm3         \n\t"\
00140         "psraw $5, %%mm0            \n\t"\
00141         "packuswb %%mm0, %%mm0      \n\t"\
00142         PAVGB" %%mm3, %%mm0         \n\t"\
00143         OP(%%mm0, (%1),%%mm6, d)\
00144         "add %4, %0                 \n\t"\
00145         "add %4, %1                 \n\t"\
00146         "add %3, %2                 \n\t"\
00147         : "+a"(src), "+c"(dst), "+d"(src2)\
00148         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00149         : "memory"\
00150     );\
00151     }while(--h);\
00152 }\
00153 static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00154     src -= 2*srcStride;\
00155     __asm__ volatile(\
00156         "pxor %%mm7, %%mm7          \n\t"\
00157         "movd (%0), %%mm0           \n\t"\
00158         "add %2, %0                 \n\t"\
00159         "movd (%0), %%mm1           \n\t"\
00160         "add %2, %0                 \n\t"\
00161         "movd (%0), %%mm2           \n\t"\
00162         "add %2, %0                 \n\t"\
00163         "movd (%0), %%mm3           \n\t"\
00164         "add %2, %0                 \n\t"\
00165         "movd (%0), %%mm4           \n\t"\
00166         "add %2, %0                 \n\t"\
00167         "punpcklbw %%mm7, %%mm0     \n\t"\
00168         "punpcklbw %%mm7, %%mm1     \n\t"\
00169         "punpcklbw %%mm7, %%mm2     \n\t"\
00170         "punpcklbw %%mm7, %%mm3     \n\t"\
00171         "punpcklbw %%mm7, %%mm4     \n\t"\
00172         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00173         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00174         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00175         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00176          \
00177         : "+a"(src), "+c"(dst)\
00178         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00179         : "memory"\
00180     );\
00181 }\
00182 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){\
00183     int h=4;\
00184     int w=3;\
00185     src -= 2*srcStride+2;\
00186     while(w--){\
00187         __asm__ volatile(\
00188             "pxor %%mm7, %%mm7      \n\t"\
00189             "movd (%0), %%mm0       \n\t"\
00190             "add %2, %0             \n\t"\
00191             "movd (%0), %%mm1       \n\t"\
00192             "add %2, %0             \n\t"\
00193             "movd (%0), %%mm2       \n\t"\
00194             "add %2, %0             \n\t"\
00195             "movd (%0), %%mm3       \n\t"\
00196             "add %2, %0             \n\t"\
00197             "movd (%0), %%mm4       \n\t"\
00198             "add %2, %0             \n\t"\
00199             "punpcklbw %%mm7, %%mm0 \n\t"\
00200             "punpcklbw %%mm7, %%mm1 \n\t"\
00201             "punpcklbw %%mm7, %%mm2 \n\t"\
00202             "punpcklbw %%mm7, %%mm3 \n\t"\
00203             "punpcklbw %%mm7, %%mm4 \n\t"\
00204             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
00205             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
00206             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
00207             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
00208              \
00209             : "+a"(src)\
00210             : "c"(tmp), "S"((x86_reg)srcStride)\
00211             : "memory"\
00212         );\
00213         tmp += 4;\
00214         src += 4 - 9*srcStride;\
00215     }\
00216     tmp -= 3*4;\
00217     __asm__ volatile(\
00218         "1:                         \n\t"\
00219         "movq     (%0), %%mm0       \n\t"\
00220         "paddw  10(%0), %%mm0       \n\t"\
00221         "movq    2(%0), %%mm1       \n\t"\
00222         "paddw   8(%0), %%mm1       \n\t"\
00223         "movq    4(%0), %%mm2       \n\t"\
00224         "paddw   6(%0), %%mm2       \n\t"\
00225         "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
00226         "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
00227         "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
00228         "paddsw %%mm2, %%mm0        \n\t"\
00229         "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
00230         "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
00231         "psraw $6, %%mm0            \n\t"\
00232         "packuswb %%mm0, %%mm0      \n\t"\
00233         OP(%%mm0, (%1),%%mm7, d)\
00234         "add $24, %0                \n\t"\
00235         "add %3, %1                 \n\t"\
00236         "decl %2                    \n\t"\
00237         " jnz 1b                    \n\t"\
00238         : "+a"(tmp), "+c"(dst), "+g"(h)\
00239         : "S"((x86_reg)dstStride)\
00240         : "memory"\
00241     );\
00242 }\
00243 \
00244 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00245     int h=8;\
00246     __asm__ volatile(\
00247         "pxor %%mm7, %%mm7          \n\t"\
00248         "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
00249         "1:                         \n\t"\
00250         "movq    (%0), %%mm0        \n\t"\
00251         "movq   1(%0), %%mm2        \n\t"\
00252         "movq %%mm0, %%mm1          \n\t"\
00253         "movq %%mm2, %%mm3          \n\t"\
00254         "punpcklbw %%mm7, %%mm0     \n\t"\
00255         "punpckhbw %%mm7, %%mm1     \n\t"\
00256         "punpcklbw %%mm7, %%mm2     \n\t"\
00257         "punpckhbw %%mm7, %%mm3     \n\t"\
00258         "paddw %%mm2, %%mm0         \n\t"\
00259         "paddw %%mm3, %%mm1         \n\t"\
00260         "psllw $2, %%mm0            \n\t"\
00261         "psllw $2, %%mm1            \n\t"\
00262         "movq   -1(%0), %%mm2       \n\t"\
00263         "movq    2(%0), %%mm4       \n\t"\
00264         "movq %%mm2, %%mm3          \n\t"\
00265         "movq %%mm4, %%mm5          \n\t"\
00266         "punpcklbw %%mm7, %%mm2     \n\t"\
00267         "punpckhbw %%mm7, %%mm3     \n\t"\
00268         "punpcklbw %%mm7, %%mm4     \n\t"\
00269         "punpckhbw %%mm7, %%mm5     \n\t"\
00270         "paddw %%mm4, %%mm2         \n\t"\
00271         "paddw %%mm3, %%mm5         \n\t"\
00272         "psubw %%mm2, %%mm0         \n\t"\
00273         "psubw %%mm5, %%mm1         \n\t"\
00274         "pmullw %%mm6, %%mm0        \n\t"\
00275         "pmullw %%mm6, %%mm1        \n\t"\
00276         "movd   -2(%0), %%mm2       \n\t"\
00277         "movd    7(%0), %%mm5       \n\t"\
00278         "punpcklbw %%mm7, %%mm2     \n\t"\
00279         "punpcklbw %%mm7, %%mm5     \n\t"\
00280         "paddw %%mm3, %%mm2         \n\t"\
00281         "paddw %%mm5, %%mm4         \n\t"\
00282         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00283         "paddw %%mm5, %%mm2         \n\t"\
00284         "paddw %%mm5, %%mm4         \n\t"\
00285         "paddw %%mm2, %%mm0         \n\t"\
00286         "paddw %%mm4, %%mm1         \n\t"\
00287         "psraw $5, %%mm0            \n\t"\
00288         "psraw $5, %%mm1            \n\t"\
00289         "packuswb %%mm1, %%mm0      \n\t"\
00290         OP(%%mm0, (%1),%%mm5, q)\
00291         "add %3, %0                 \n\t"\
00292         "add %4, %1                 \n\t"\
00293         "decl %2                    \n\t"\
00294         " jnz 1b                    \n\t"\
00295         : "+a"(src), "+c"(dst), "+g"(h)\
00296         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00297         : "memory"\
00298     );\
00299 }\
00300 \
00301 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00302     int h=8;\
00303     __asm__ volatile(\
00304         "pxor %%mm7, %%mm7          \n\t"\
00305         "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
00306         "1:                         \n\t"\
00307         "movq    (%0), %%mm0        \n\t"\
00308         "movq   1(%0), %%mm2        \n\t"\
00309         "movq %%mm0, %%mm1          \n\t"\
00310         "movq %%mm2, %%mm3          \n\t"\
00311         "punpcklbw %%mm7, %%mm0     \n\t"\
00312         "punpckhbw %%mm7, %%mm1     \n\t"\
00313         "punpcklbw %%mm7, %%mm2     \n\t"\
00314         "punpckhbw %%mm7, %%mm3     \n\t"\
00315         "paddw %%mm2, %%mm0         \n\t"\
00316         "paddw %%mm3, %%mm1         \n\t"\
00317         "psllw $2, %%mm0            \n\t"\
00318         "psllw $2, %%mm1            \n\t"\
00319         "movq   -1(%0), %%mm2       \n\t"\
00320         "movq    2(%0), %%mm4       \n\t"\
00321         "movq %%mm2, %%mm3          \n\t"\
00322         "movq %%mm4, %%mm5          \n\t"\
00323         "punpcklbw %%mm7, %%mm2     \n\t"\
00324         "punpckhbw %%mm7, %%mm3     \n\t"\
00325         "punpcklbw %%mm7, %%mm4     \n\t"\
00326         "punpckhbw %%mm7, %%mm5     \n\t"\
00327         "paddw %%mm4, %%mm2         \n\t"\
00328         "paddw %%mm3, %%mm5         \n\t"\
00329         "psubw %%mm2, %%mm0         \n\t"\
00330         "psubw %%mm5, %%mm1         \n\t"\
00331         "pmullw %%mm6, %%mm0        \n\t"\
00332         "pmullw %%mm6, %%mm1        \n\t"\
00333         "movd   -2(%0), %%mm2       \n\t"\
00334         "movd    7(%0), %%mm5       \n\t"\
00335         "punpcklbw %%mm7, %%mm2     \n\t"\
00336         "punpcklbw %%mm7, %%mm5     \n\t"\
00337         "paddw %%mm3, %%mm2         \n\t"\
00338         "paddw %%mm5, %%mm4         \n\t"\
00339         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00340         "paddw %%mm5, %%mm2         \n\t"\
00341         "paddw %%mm5, %%mm4         \n\t"\
00342         "paddw %%mm2, %%mm0         \n\t"\
00343         "paddw %%mm4, %%mm1         \n\t"\
00344         "psraw $5, %%mm0            \n\t"\
00345         "psraw $5, %%mm1            \n\t"\
00346         "movq (%2), %%mm4           \n\t"\
00347         "packuswb %%mm1, %%mm0      \n\t"\
00348         PAVGB" %%mm4, %%mm0         \n\t"\
00349         OP(%%mm0, (%1),%%mm5, q)\
00350         "add %5, %0                 \n\t"\
00351         "add %5, %1                 \n\t"\
00352         "add %4, %2                 \n\t"\
00353         "decl %3                    \n\t"\
00354         "jg 1b                      \n\t"\
00355         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00356         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00357         : "memory"\
00358     );\
00359 }\
00360 \
00361 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00362     int w= 2;\
00363     src -= 2*srcStride;\
00364     \
00365     while(w--){\
00366         __asm__ volatile(\
00367             "pxor %%mm7, %%mm7          \n\t"\
00368             "movd (%0), %%mm0           \n\t"\
00369             "add %2, %0                 \n\t"\
00370             "movd (%0), %%mm1           \n\t"\
00371             "add %2, %0                 \n\t"\
00372             "movd (%0), %%mm2           \n\t"\
00373             "add %2, %0                 \n\t"\
00374             "movd (%0), %%mm3           \n\t"\
00375             "add %2, %0                 \n\t"\
00376             "movd (%0), %%mm4           \n\t"\
00377             "add %2, %0                 \n\t"\
00378             "punpcklbw %%mm7, %%mm0     \n\t"\
00379             "punpcklbw %%mm7, %%mm1     \n\t"\
00380             "punpcklbw %%mm7, %%mm2     \n\t"\
00381             "punpcklbw %%mm7, %%mm3     \n\t"\
00382             "punpcklbw %%mm7, %%mm4     \n\t"\
00383             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00384             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00385             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00386             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00387             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00388             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00389             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00390             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00391             "cmpl $16, %4               \n\t"\
00392             "jne 2f                     \n\t"\
00393             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00394             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00395             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00396             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00397             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00398             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00399             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00400             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00401             "2:                         \n\t"\
00402             \
00403             : "+a"(src), "+c"(dst)\
00404             : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
00405             : "memory"\
00406         );\
00407         src += 4-(h+5)*srcStride;\
00408         dst += 4-h*dstStride;\
00409     }\
00410 }\
00411 static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
00412     int w = (size+8)>>2;\
00413     src -= 2*srcStride+2;\
00414     while(w--){\
00415         __asm__ volatile(\
00416             "pxor %%mm7, %%mm7      \n\t"\
00417             "movd (%0), %%mm0       \n\t"\
00418             "add %2, %0             \n\t"\
00419             "movd (%0), %%mm1       \n\t"\
00420             "add %2, %0             \n\t"\
00421             "movd (%0), %%mm2       \n\t"\
00422             "add %2, %0             \n\t"\
00423             "movd (%0), %%mm3       \n\t"\
00424             "add %2, %0             \n\t"\
00425             "movd (%0), %%mm4       \n\t"\
00426             "add %2, %0             \n\t"\
00427             "punpcklbw %%mm7, %%mm0 \n\t"\
00428             "punpcklbw %%mm7, %%mm1 \n\t"\
00429             "punpcklbw %%mm7, %%mm2 \n\t"\
00430             "punpcklbw %%mm7, %%mm3 \n\t"\
00431             "punpcklbw %%mm7, %%mm4 \n\t"\
00432             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
00433             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
00434             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
00435             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
00436             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
00437             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
00438             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
00439             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
00440             "cmpl $16, %3           \n\t"\
00441             "jne 2f                 \n\t"\
00442             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
00443             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
00444             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
00445             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
00446             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
00447             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
00448             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
00449             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
00450             "2:                     \n\t"\
00451             : "+a"(src)\
00452             : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)\
00453             : "memory"\
00454             );\
00455         tmp += 4;\
00456         src += 4 - (size+5)*srcStride;\
00457     }\
00458 }\
00459 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
00460     int w = size>>4;\
00461     do{\
00462     int h = size;\
00463     __asm__ volatile(\
00464         "1:                         \n\t"\
00465         "movq     (%0), %%mm0       \n\t"\
00466         "movq    8(%0), %%mm3       \n\t"\
00467         "movq    2(%0), %%mm1       \n\t"\
00468         "movq   10(%0), %%mm4       \n\t"\
00469         "paddw   %%mm4, %%mm0       \n\t"\
00470         "paddw   %%mm3, %%mm1       \n\t"\
00471         "paddw  18(%0), %%mm3       \n\t"\
00472         "paddw  16(%0), %%mm4       \n\t"\
00473         "movq    4(%0), %%mm2       \n\t"\
00474         "movq   12(%0), %%mm5       \n\t"\
00475         "paddw   6(%0), %%mm2       \n\t"\
00476         "paddw  14(%0), %%mm5       \n\t"\
00477         "psubw %%mm1, %%mm0         \n\t"\
00478         "psubw %%mm4, %%mm3         \n\t"\
00479         "psraw $2, %%mm0            \n\t"\
00480         "psraw $2, %%mm3            \n\t"\
00481         "psubw %%mm1, %%mm0         \n\t"\
00482         "psubw %%mm4, %%mm3         \n\t"\
00483         "paddsw %%mm2, %%mm0        \n\t"\
00484         "paddsw %%mm5, %%mm3        \n\t"\
00485         "psraw $2, %%mm0            \n\t"\
00486         "psraw $2, %%mm3            \n\t"\
00487         "paddw %%mm2, %%mm0         \n\t"\
00488         "paddw %%mm5, %%mm3         \n\t"\
00489         "psraw $6, %%mm0            \n\t"\
00490         "psraw $6, %%mm3            \n\t"\
00491         "packuswb %%mm3, %%mm0      \n\t"\
00492         OP(%%mm0, (%1),%%mm7, q)\
00493         "add $48, %0                \n\t"\
00494         "add %3, %1                 \n\t"\
00495         "decl %2                    \n\t"\
00496         " jnz 1b                    \n\t"\
00497         : "+a"(tmp), "+c"(dst), "+g"(h)\
00498         : "S"((x86_reg)dstStride)\
00499         : "memory"\
00500     );\
00501     tmp += 8 - size*24;\
00502     dst += 8 - size*dstStride;\
00503     }while(w--);\
00504 }\
00505 \
00506 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00507     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00508 }\
00509 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00510     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00511     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00512 }\
00513 \
00514 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00515     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00516     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00517     src += 8*srcStride;\
00518     dst += 8*dstStride;\
00519     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00520     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00521 }\
00522 \
00523 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00524     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00525     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00526     src += 8*dstStride;\
00527     dst += 8*dstStride;\
00528     src2 += 8*src2Stride;\
00529     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00530     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00531 }\
00532 \
00533 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){\
00534           put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
00535     OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
00536 }\
00537 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00538     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
00539 }\
00540 \
00541 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00542     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
00543 }\
00544 \
00545 static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00546 {\
00547     __asm__ volatile(\
00548         "movq      (%1), %%mm0          \n\t"\
00549         "movq    24(%1), %%mm1          \n\t"\
00550         "psraw      $5,  %%mm0          \n\t"\
00551         "psraw      $5,  %%mm1          \n\t"\
00552         "packuswb %%mm0, %%mm0          \n\t"\
00553         "packuswb %%mm1, %%mm1          \n\t"\
00554         PAVGB"     (%0), %%mm0          \n\t"\
00555         PAVGB"  (%0,%3), %%mm1          \n\t"\
00556         OP(%%mm0, (%2),    %%mm4, d)\
00557         OP(%%mm1, (%2,%4), %%mm5, d)\
00558         "lea  (%0,%3,2), %0             \n\t"\
00559         "lea  (%2,%4,2), %2             \n\t"\
00560         "movq    48(%1), %%mm0          \n\t"\
00561         "movq    72(%1), %%mm1          \n\t"\
00562         "psraw      $5,  %%mm0          \n\t"\
00563         "psraw      $5,  %%mm1          \n\t"\
00564         "packuswb %%mm0, %%mm0          \n\t"\
00565         "packuswb %%mm1, %%mm1          \n\t"\
00566         PAVGB"     (%0), %%mm0          \n\t"\
00567         PAVGB"  (%0,%3), %%mm1          \n\t"\
00568         OP(%%mm0, (%2),    %%mm4, d)\
00569         OP(%%mm1, (%2,%4), %%mm5, d)\
00570         :"+a"(src8), "+c"(src16), "+d"(dst)\
00571         :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
00572         :"memory");\
00573 }\
00574 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00575 {\
00576     do{\
00577     __asm__ volatile(\
00578         "movq      (%1), %%mm0          \n\t"\
00579         "movq     8(%1), %%mm1          \n\t"\
00580         "movq    48(%1), %%mm2          \n\t"\
00581         "movq  8+48(%1), %%mm3          \n\t"\
00582         "psraw      $5,  %%mm0          \n\t"\
00583         "psraw      $5,  %%mm1          \n\t"\
00584         "psraw      $5,  %%mm2          \n\t"\
00585         "psraw      $5,  %%mm3          \n\t"\
00586         "packuswb %%mm1, %%mm0          \n\t"\
00587         "packuswb %%mm3, %%mm2          \n\t"\
00588         PAVGB"     (%0), %%mm0          \n\t"\
00589         PAVGB"  (%0,%3), %%mm2          \n\t"\
00590         OP(%%mm0, (%2), %%mm5, q)\
00591         OP(%%mm2, (%2,%4), %%mm5, q)\
00592         ::"a"(src8), "c"(src16), "d"(dst),\
00593           "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
00594         :"memory");\
00595         src8 += 2L*src8Stride;\
00596         src16 += 48;\
00597         dst += 2L*dstStride;\
00598     }while(h-=2);\
00599 }\
00600 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00601 {\
00602     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
00603     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
00604 }\
00605 
00606 
00607 #if ARCH_X86_64
00608 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00609 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00610     int h=16;\
00611     __asm__ volatile(\
00612         "pxor %%xmm15, %%xmm15      \n\t"\
00613         "movdqa %6, %%xmm14         \n\t"\
00614         "movdqa %7, %%xmm13         \n\t"\
00615         "1:                         \n\t"\
00616         "lddqu    6(%0), %%xmm1     \n\t"\
00617         "lddqu   -2(%0), %%xmm7     \n\t"\
00618         "movdqa  %%xmm1, %%xmm0     \n\t"\
00619         "punpckhbw %%xmm15, %%xmm1  \n\t"\
00620         "punpcklbw %%xmm15, %%xmm0  \n\t"\
00621         "punpcklbw %%xmm15, %%xmm7  \n\t"\
00622         "movdqa  %%xmm1, %%xmm2     \n\t"\
00623         "movdqa  %%xmm0, %%xmm6     \n\t"\
00624         "movdqa  %%xmm1, %%xmm3     \n\t"\
00625         "movdqa  %%xmm0, %%xmm8     \n\t"\
00626         "movdqa  %%xmm1, %%xmm4     \n\t"\
00627         "movdqa  %%xmm0, %%xmm9     \n\t"\
00628         "movdqa  %%xmm0, %%xmm12    \n\t"\
00629         "movdqa  %%xmm1, %%xmm11    \n\t"\
00630         "palignr $10,%%xmm0, %%xmm11\n\t"\
00631         "palignr $10,%%xmm7, %%xmm12\n\t"\
00632         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00633         "palignr $2, %%xmm7, %%xmm9 \n\t"\
00634         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00635         "palignr $4, %%xmm7, %%xmm8 \n\t"\
00636         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00637         "palignr $6, %%xmm7, %%xmm6 \n\t"\
00638         "paddw   %%xmm0 ,%%xmm11    \n\t"\
00639         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00640         "palignr $8, %%xmm7, %%xmm0 \n\t"\
00641         "paddw   %%xmm12,%%xmm7     \n\t"\
00642         "paddw   %%xmm3, %%xmm2     \n\t"\
00643         "paddw   %%xmm8, %%xmm6     \n\t"\
00644         "paddw   %%xmm4, %%xmm1     \n\t"\
00645         "paddw   %%xmm9, %%xmm0     \n\t"\
00646         "psllw   $2,     %%xmm2     \n\t"\
00647         "psllw   $2,     %%xmm6     \n\t"\
00648         "psubw   %%xmm1, %%xmm2     \n\t"\
00649         "psubw   %%xmm0, %%xmm6     \n\t"\
00650         "paddw   %%xmm13,%%xmm11    \n\t"\
00651         "paddw   %%xmm13,%%xmm7     \n\t"\
00652         "pmullw  %%xmm14,%%xmm2     \n\t"\
00653         "pmullw  %%xmm14,%%xmm6     \n\t"\
00654         "lddqu   (%2),   %%xmm3     \n\t"\
00655         "paddw   %%xmm11,%%xmm2     \n\t"\
00656         "paddw   %%xmm7, %%xmm6     \n\t"\
00657         "psraw   $5,     %%xmm2     \n\t"\
00658         "psraw   $5,     %%xmm6     \n\t"\
00659         "packuswb %%xmm2,%%xmm6     \n\t"\
00660         "pavgb   %%xmm3, %%xmm6     \n\t"\
00661         OP(%%xmm6, (%1), %%xmm4, dqa)\
00662         "add %5, %0                 \n\t"\
00663         "add %5, %1                 \n\t"\
00664         "add %4, %2                 \n\t"\
00665         "decl %3                    \n\t"\
00666         "jg 1b                      \n\t"\
00667         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00668         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
00669           "m"(ff_pw_5), "m"(ff_pw_16)\
00670         : XMM_CLOBBERS("%xmm0" , "%xmm1" , "%xmm2" , "%xmm3" , \
00671                        "%xmm4" , "%xmm5" , "%xmm6" , "%xmm7" , \
00672                        "%xmm8" , "%xmm9" , "%xmm10", "%xmm11", \
00673                        "%xmm12", "%xmm13", "%xmm14", "%xmm15",)\
00674           "memory"\
00675     );\
00676 }
00677 #else // ARCH_X86_64
00678 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00679 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00680     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00681     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00682     src += 8*dstStride;\
00683     dst += 8*dstStride;\
00684     src2 += 8*src2Stride;\
00685     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00686     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00687 }
00688 #endif // ARCH_X86_64
00689 
00690 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
00691 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00692     int h=8;\
00693     __asm__ volatile(\
00694         "pxor %%xmm7, %%xmm7        \n\t"\
00695         "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
00696         "1:                         \n\t"\
00697         "lddqu   -2(%0), %%xmm1     \n\t"\
00698         "movdqa  %%xmm1, %%xmm0     \n\t"\
00699         "punpckhbw %%xmm7, %%xmm1   \n\t"\
00700         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00701         "movdqa  %%xmm1, %%xmm2     \n\t"\
00702         "movdqa  %%xmm1, %%xmm3     \n\t"\
00703         "movdqa  %%xmm1, %%xmm4     \n\t"\
00704         "movdqa  %%xmm1, %%xmm5     \n\t"\
00705         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00706         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00707         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00708         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00709         "palignr $10,%%xmm0, %%xmm5 \n\t"\
00710         "paddw   %%xmm5, %%xmm0     \n\t"\
00711         "paddw   %%xmm3, %%xmm2     \n\t"\
00712         "paddw   %%xmm4, %%xmm1     \n\t"\
00713         "psllw   $2,     %%xmm2     \n\t"\
00714         "movq    (%2),   %%xmm3     \n\t"\
00715         "psubw   %%xmm1, %%xmm2     \n\t"\
00716         "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
00717         "pmullw  %%xmm6, %%xmm2     \n\t"\
00718         "paddw   %%xmm0, %%xmm2     \n\t"\
00719         "psraw   $5,     %%xmm2     \n\t"\
00720         "packuswb %%xmm2, %%xmm2    \n\t"\
00721         "pavgb   %%xmm3, %%xmm2     \n\t"\
00722         OP(%%xmm2, (%1), %%xmm4, q)\
00723         "add %5, %0                 \n\t"\
00724         "add %5, %1                 \n\t"\
00725         "add %4, %2                 \n\t"\
00726         "decl %3                    \n\t"\
00727         "jg 1b                      \n\t"\
00728         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00729         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00730         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00731                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00732           "memory"\
00733     );\
00734 }\
00735 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00736 \
00737 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00738     int h=8;\
00739     __asm__ volatile(\
00740         "pxor %%xmm7, %%xmm7        \n\t"\
00741         "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
00742         "1:                         \n\t"\
00743         "lddqu   -2(%0), %%xmm1     \n\t"\
00744         "movdqa  %%xmm1, %%xmm0     \n\t"\
00745         "punpckhbw %%xmm7, %%xmm1   \n\t"\
00746         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00747         "movdqa  %%xmm1, %%xmm2     \n\t"\
00748         "movdqa  %%xmm1, %%xmm3     \n\t"\
00749         "movdqa  %%xmm1, %%xmm4     \n\t"\
00750         "movdqa  %%xmm1, %%xmm5     \n\t"\
00751         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00752         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00753         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00754         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00755         "palignr $10,%%xmm0, %%xmm5 \n\t"\
00756         "paddw   %%xmm5, %%xmm0     \n\t"\
00757         "paddw   %%xmm3, %%xmm2     \n\t"\
00758         "paddw   %%xmm4, %%xmm1     \n\t"\
00759         "psllw   $2,     %%xmm2     \n\t"\
00760         "psubw   %%xmm1, %%xmm2     \n\t"\
00761         "paddw   "MANGLE(ff_pw_16)", %%xmm0\n\t"\
00762         "pmullw  %%xmm6, %%xmm2     \n\t"\
00763         "paddw   %%xmm0, %%xmm2     \n\t"\
00764         "psraw   $5,     %%xmm2     \n\t"\
00765         "packuswb %%xmm2, %%xmm2    \n\t"\
00766         OP(%%xmm2, (%1), %%xmm4, q)\
00767         "add %3, %0                 \n\t"\
00768         "add %4, %1                 \n\t"\
00769         "decl %2                    \n\t"\
00770         " jnz 1b                    \n\t"\
00771         : "+a"(src), "+c"(dst), "+g"(h)\
00772         : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00773         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00774                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00775           "memory"\
00776     );\
00777 }\
00778 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00779     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00780     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00781     src += 8*srcStride;\
00782     dst += 8*dstStride;\
00783     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00784     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00785 }\
00786 
00787 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
00788 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00789     src -= 2*srcStride;\
00790     \
00791     __asm__ volatile(\
00792         "pxor %%xmm7, %%xmm7        \n\t"\
00793         "movq (%0), %%xmm0          \n\t"\
00794         "add %2, %0                 \n\t"\
00795         "movq (%0), %%xmm1          \n\t"\
00796         "add %2, %0                 \n\t"\
00797         "movq (%0), %%xmm2          \n\t"\
00798         "add %2, %0                 \n\t"\
00799         "movq (%0), %%xmm3          \n\t"\
00800         "add %2, %0                 \n\t"\
00801         "movq (%0), %%xmm4          \n\t"\
00802         "add %2, %0                 \n\t"\
00803         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00804         "punpcklbw %%xmm7, %%xmm1   \n\t"\
00805         "punpcklbw %%xmm7, %%xmm2   \n\t"\
00806         "punpcklbw %%xmm7, %%xmm3   \n\t"\
00807         "punpcklbw %%xmm7, %%xmm4   \n\t"\
00808         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00809         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00810         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00811         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00812         QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
00813         QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
00814         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00815         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00816         "cmpl $16, %4               \n\t"\
00817         "jne 2f                     \n\t"\
00818         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00819         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00820         QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
00821         QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
00822         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00823         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00824         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00825         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00826         "2:                          \n\t"\
00827         \
00828         : "+a"(src), "+c"(dst)\
00829         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
00830         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00831                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00832           "memory"\
00833     );\
00834 }\
00835 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00836     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00837 }\
00838 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00839     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00840     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00841 }
00842 
00843 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
00844     int w = (size+8)>>3;
00845     src -= 2*srcStride+2;
00846     while(w--){
00847         __asm__ volatile(
00848             "pxor %%xmm7, %%xmm7        \n\t"
00849             "movq (%0), %%xmm0          \n\t"
00850             "add %2, %0                 \n\t"
00851             "movq (%0), %%xmm1          \n\t"
00852             "add %2, %0                 \n\t"
00853             "movq (%0), %%xmm2          \n\t"
00854             "add %2, %0                 \n\t"
00855             "movq (%0), %%xmm3          \n\t"
00856             "add %2, %0                 \n\t"
00857             "movq (%0), %%xmm4          \n\t"
00858             "add %2, %0                 \n\t"
00859             "punpcklbw %%xmm7, %%xmm0   \n\t"
00860             "punpcklbw %%xmm7, %%xmm1   \n\t"
00861             "punpcklbw %%xmm7, %%xmm2   \n\t"
00862             "punpcklbw %%xmm7, %%xmm3   \n\t"
00863             "punpcklbw %%xmm7, %%xmm4   \n\t"
00864             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
00865             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
00866             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
00867             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
00868             QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
00869             QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
00870             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
00871             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
00872             "cmpl $16, %3               \n\t"
00873             "jne 2f                     \n\t"
00874             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
00875             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
00876             QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
00877             QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
00878             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
00879             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
00880             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
00881             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
00882             "2:                         \n\t"
00883             : "+a"(src)
00884             : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)
00885             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3",
00886                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)
00887               "memory"
00888         );
00889         tmp += 8;
00890         src += 8 - (size+5)*srcStride;
00891     }
00892 }
00893 
00894 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
00895 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
00896     int h = size;\
00897     if(size == 16){\
00898         __asm__ volatile(\
00899             "1:                         \n\t"\
00900             "movdqa 32(%0), %%xmm4      \n\t"\
00901             "movdqa 16(%0), %%xmm5      \n\t"\
00902             "movdqa   (%0), %%xmm7      \n\t"\
00903             "movdqa %%xmm4, %%xmm3      \n\t"\
00904             "movdqa %%xmm4, %%xmm2      \n\t"\
00905             "movdqa %%xmm4, %%xmm1      \n\t"\
00906             "movdqa %%xmm4, %%xmm0      \n\t"\
00907             "palignr $10, %%xmm5, %%xmm0 \n\t"\
00908             "palignr  $8, %%xmm5, %%xmm1 \n\t"\
00909             "palignr  $6, %%xmm5, %%xmm2 \n\t"\
00910             "palignr  $4, %%xmm5, %%xmm3 \n\t"\
00911             "palignr  $2, %%xmm5, %%xmm4 \n\t"\
00912             "paddw  %%xmm5, %%xmm0      \n\t"\
00913             "paddw  %%xmm4, %%xmm1      \n\t"\
00914             "paddw  %%xmm3, %%xmm2      \n\t"\
00915             "movdqa %%xmm5, %%xmm6      \n\t"\
00916             "movdqa %%xmm5, %%xmm4      \n\t"\
00917             "movdqa %%xmm5, %%xmm3      \n\t"\
00918             "palignr  $8, %%xmm7, %%xmm4 \n\t"\
00919             "palignr  $2, %%xmm7, %%xmm6 \n\t"\
00920             "palignr $10, %%xmm7, %%xmm3 \n\t"\
00921             "paddw  %%xmm6, %%xmm4      \n\t"\
00922             "movdqa %%xmm5, %%xmm6      \n\t"\
00923             "palignr  $6, %%xmm7, %%xmm5 \n\t"\
00924             "palignr  $4, %%xmm7, %%xmm6 \n\t"\
00925             "paddw  %%xmm7, %%xmm3      \n\t"\
00926             "paddw  %%xmm6, %%xmm5      \n\t"\
00927             \
00928             "psubw  %%xmm1, %%xmm0      \n\t"\
00929             "psubw  %%xmm4, %%xmm3      \n\t"\
00930             "psraw      $2, %%xmm0      \n\t"\
00931             "psraw      $2, %%xmm3      \n\t"\
00932             "psubw  %%xmm1, %%xmm0      \n\t"\
00933             "psubw  %%xmm4, %%xmm3      \n\t"\
00934             "paddw  %%xmm2, %%xmm0      \n\t"\
00935             "paddw  %%xmm5, %%xmm3      \n\t"\
00936             "psraw      $2, %%xmm0      \n\t"\
00937             "psraw      $2, %%xmm3      \n\t"\
00938             "paddw  %%xmm2, %%xmm0      \n\t"\
00939             "paddw  %%xmm5, %%xmm3      \n\t"\
00940             "psraw      $6, %%xmm0      \n\t"\
00941             "psraw      $6, %%xmm3      \n\t"\
00942             "packuswb %%xmm0, %%xmm3    \n\t"\
00943             OP(%%xmm3, (%1), %%xmm7, dqa)\
00944             "add $48, %0                \n\t"\
00945             "add %3, %1                 \n\t"\
00946             "decl %2                    \n\t"\
00947             " jnz 1b                    \n\t"\
00948             : "+a"(tmp), "+c"(dst), "+g"(h)\
00949             : "S"((x86_reg)dstStride)\
00950             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00951                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00952               "memory"\
00953         );\
00954     }else{\
00955         __asm__ volatile(\
00956             "1:                         \n\t"\
00957             "movdqa 16(%0), %%xmm1      \n\t"\
00958             "movdqa   (%0), %%xmm0      \n\t"\
00959             "movdqa %%xmm1, %%xmm2      \n\t"\
00960             "movdqa %%xmm1, %%xmm3      \n\t"\
00961             "movdqa %%xmm1, %%xmm4      \n\t"\
00962             "movdqa %%xmm1, %%xmm5      \n\t"\
00963             "palignr $10, %%xmm0, %%xmm5 \n\t"\
00964             "palignr  $8, %%xmm0, %%xmm4 \n\t"\
00965             "palignr  $6, %%xmm0, %%xmm3 \n\t"\
00966             "palignr  $4, %%xmm0, %%xmm2 \n\t"\
00967             "palignr  $2, %%xmm0, %%xmm1 \n\t"\
00968             "paddw  %%xmm5, %%xmm0      \n\t"\
00969             "paddw  %%xmm4, %%xmm1      \n\t"\
00970             "paddw  %%xmm3, %%xmm2      \n\t"\
00971             "psubw  %%xmm1, %%xmm0      \n\t"\
00972             "psraw      $2, %%xmm0      \n\t"\
00973             "psubw  %%xmm1, %%xmm0      \n\t"\
00974             "paddw  %%xmm2, %%xmm0      \n\t"\
00975             "psraw      $2, %%xmm0      \n\t"\
00976             "paddw  %%xmm2, %%xmm0      \n\t"\
00977             "psraw      $6, %%xmm0      \n\t"\
00978             "packuswb %%xmm0, %%xmm0    \n\t"\
00979             OP(%%xmm0, (%1), %%xmm7, q)\
00980             "add $48, %0                \n\t"\
00981             "add %3, %1                 \n\t"\
00982             "decl %2                    \n\t"\
00983             " jnz 1b                    \n\t"\
00984             : "+a"(tmp), "+c"(dst), "+g"(h)\
00985             : "S"((x86_reg)dstStride)\
00986             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00987                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00988               "memory"\
00989         );\
00990     }\
00991 }
00992 
00993 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
00994 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){\
00995           put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
00996     OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
00997 }\
00998 static void OPNAME ## h264_qpel8_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, 8);\
01000 }\
01001 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
01002     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
01003 }\
01004 
01005 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
01006 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
01007 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
01008 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
01009 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
01010 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
01011 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
01012 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
01013 
01014 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
01015 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
01016 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
01017 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
01018 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
01019 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
01020 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
01021 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
01022 
01023 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
01024 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
01025 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
01026 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
01027 
01028 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
01029 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
01030 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
01031 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
01032 
01033 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
01034 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
01035 
01036 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
01037 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
01038 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
01039 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
01040 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
01041 
01042 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
01043     put_pixels16_sse2(dst, src, stride, 16);
01044 }
01045 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
01046     avg_pixels16_sse2(dst, src, stride, 16);
01047 }
01048 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
01049 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
01050 
01051 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
01052 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
01053     OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
01054 }\
01055 
01056 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
01057 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01058     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
01059 }\
01060 \
01061 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01062     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
01063 }\
01064 \
01065 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01066     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
01067 }\
01068 
01069 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
01070 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01071     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01072     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01073     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
01074 }\
01075 \
01076 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01077     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
01078 }\
01079 \
01080 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01081     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01082     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01083     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
01084 }\
01085 
01086 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
01087 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01088     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01089     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01090     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
01091 }\
01092 \
01093 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01094     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01095     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
01096     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
01097 }\
01098 \
01099 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01100     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01101     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01102     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
01103 }\
01104 \
01105 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01106     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01107     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
01108     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
01109 }\
01110 \
01111 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01112     DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
01113     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
01114 }\
01115 \
01116 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01117     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01118     uint8_t * const halfHV= temp;\
01119     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01120     assert(((int)temp & 7) == 0);\
01121     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01122     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
01123 }\
01124 \
01125 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01126     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01127     uint8_t * const halfHV= temp;\
01128     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01129     assert(((int)temp & 7) == 0);\
01130     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01131     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
01132 }\
01133 \
01134 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01135     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01136     uint8_t * const halfHV= temp;\
01137     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01138     assert(((int)temp & 7) == 0);\
01139     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01140     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
01141 }\
01142 \
01143 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01144     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01145     uint8_t * const halfHV= temp;\
01146     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01147     assert(((int)temp & 7) == 0);\
01148     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01149     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
01150 }\
01151 
01152 #define H264_MC_4816(MMX)\
01153 H264_MC(put_, 4, MMX, 8)\
01154 H264_MC(put_, 8, MMX, 8)\
01155 H264_MC(put_, 16,MMX, 8)\
01156 H264_MC(avg_, 4, MMX, 8)\
01157 H264_MC(avg_, 8, MMX, 8)\
01158 H264_MC(avg_, 16,MMX, 8)\
01159 
01160 #define H264_MC_816(QPEL, XMM)\
01161 QPEL(put_, 8, XMM, 16)\
01162 QPEL(put_, 16,XMM, 16)\
01163 QPEL(avg_, 8, XMM, 16)\
01164 QPEL(avg_, 16,XMM, 16)\
01165 
01166 #define PAVGB "pavgusb"
01167 QPEL_H264(put_,       PUT_OP, 3dnow)
01168 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
01169 #undef PAVGB
01170 #define PAVGB "pavgb"
01171 QPEL_H264(put_,       PUT_OP, mmx2)
01172 QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
01173 QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
01174 QPEL_H264_V_XMM(avg_,  AVG_MMX2_OP, sse2)
01175 QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
01176 QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, sse2)
01177 #if HAVE_SSSE3_INLINE
01178 QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
01179 QPEL_H264_H_XMM(avg_,  AVG_MMX2_OP, ssse3)
01180 QPEL_H264_HV2_XMM(put_,       PUT_OP, ssse3)
01181 QPEL_H264_HV2_XMM(avg_,  AVG_MMX2_OP, ssse3)
01182 QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
01183 QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, ssse3)
01184 #endif
01185 #undef PAVGB
01186 
01187 H264_MC_4816(3dnow)
01188 H264_MC_4816(mmx2)
01189 H264_MC_816(H264_MC_V, sse2)
01190 H264_MC_816(H264_MC_HV, sse2)
01191 #if HAVE_SSSE3_INLINE
01192 H264_MC_816(H264_MC_H, ssse3)
01193 H264_MC_816(H264_MC_HV, ssse3)
01194 #endif
01195 
01196 #endif /* HAVE_INLINE_ASM */
01197 
01198 //10bit
01199 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
01200 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
01201     (uint8_t *dst, uint8_t *src, int stride);
01202 
01203 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
01204     LUMA_MC_OP(put,  4, DEPTH, TYPE, OPT) \
01205     LUMA_MC_OP(avg,  4, DEPTH, TYPE, OPT) \
01206     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
01207     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
01208     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
01209     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
01210 
01211 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
01212     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
01213     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
01214     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
01215     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
01216 
01217 LUMA_MC_ALL(10, mc00, mmxext)
01218 LUMA_MC_ALL(10, mc10, mmxext)
01219 LUMA_MC_ALL(10, mc20, mmxext)
01220 LUMA_MC_ALL(10, mc30, mmxext)
01221 LUMA_MC_ALL(10, mc01, mmxext)
01222 LUMA_MC_ALL(10, mc11, mmxext)
01223 LUMA_MC_ALL(10, mc21, mmxext)
01224 LUMA_MC_ALL(10, mc31, mmxext)
01225 LUMA_MC_ALL(10, mc02, mmxext)
01226 LUMA_MC_ALL(10, mc12, mmxext)
01227 LUMA_MC_ALL(10, mc22, mmxext)
01228 LUMA_MC_ALL(10, mc32, mmxext)
01229 LUMA_MC_ALL(10, mc03, mmxext)
01230 LUMA_MC_ALL(10, mc13, mmxext)
01231 LUMA_MC_ALL(10, mc23, mmxext)
01232 LUMA_MC_ALL(10, mc33, mmxext)
01233 
01234 LUMA_MC_816(10, mc00, sse2)
01235 LUMA_MC_816(10, mc10, sse2)
01236 LUMA_MC_816(10, mc10, sse2_cache64)
01237 LUMA_MC_816(10, mc10, ssse3_cache64)
01238 LUMA_MC_816(10, mc20, sse2)
01239 LUMA_MC_816(10, mc20, sse2_cache64)
01240 LUMA_MC_816(10, mc20, ssse3_cache64)
01241 LUMA_MC_816(10, mc30, sse2)
01242 LUMA_MC_816(10, mc30, sse2_cache64)
01243 LUMA_MC_816(10, mc30, ssse3_cache64)
01244 LUMA_MC_816(10, mc01, sse2)
01245 LUMA_MC_816(10, mc11, sse2)
01246 LUMA_MC_816(10, mc21, sse2)
01247 LUMA_MC_816(10, mc31, sse2)
01248 LUMA_MC_816(10, mc02, sse2)
01249 LUMA_MC_816(10, mc12, sse2)
01250 LUMA_MC_816(10, mc22, sse2)
01251 LUMA_MC_816(10, mc32, sse2)
01252 LUMA_MC_816(10, mc03, sse2)
01253 LUMA_MC_816(10, mc13, sse2)
01254 LUMA_MC_816(10, mc23, sse2)
01255 LUMA_MC_816(10, mc33, sse2)
01256 
01257 #define QPEL16_OPMC(OP, MC, MMX)\
01258 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01259     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
01260     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
01261     src += 8*stride;\
01262     dst += 8*stride;\
01263     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
01264     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
01265 }
01266 
01267 #define QPEL16_OP(MC, MMX)\
01268 QPEL16_OPMC(put, MC, MMX)\
01269 QPEL16_OPMC(avg, MC, MMX)
01270 
01271 #define QPEL16(MMX)\
01272 QPEL16_OP(mc00, MMX)\
01273 QPEL16_OP(mc01, MMX)\
01274 QPEL16_OP(mc02, MMX)\
01275 QPEL16_OP(mc03, MMX)\
01276 QPEL16_OP(mc10, MMX)\
01277 QPEL16_OP(mc11, MMX)\
01278 QPEL16_OP(mc12, MMX)\
01279 QPEL16_OP(mc13, MMX)\
01280 QPEL16_OP(mc20, MMX)\
01281 QPEL16_OP(mc21, MMX)\
01282 QPEL16_OP(mc22, MMX)\
01283 QPEL16_OP(mc23, MMX)\
01284 QPEL16_OP(mc30, MMX)\
01285 QPEL16_OP(mc31, MMX)\
01286 QPEL16_OP(mc32, MMX)\
01287 QPEL16_OP(mc33, MMX)
01288 
01289 #if CONFIG_H264QPEL && ARCH_X86_32 && HAVE_YASM // ARCH_X86_64 implies sse2+
01290 QPEL16(mmxext)
01291 #endif