FFmpeg: libavcodec/x86/h264_qpel

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 /***********************************/
00025 /* motion compensation */
00026 
00027 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
00028         "mov"#q" "#C", "#T"         \n\t"\
00029         "mov"#d" (%0), "#F"         \n\t"\
00030         "paddw "#D", "#T"           \n\t"\
00031         "psllw $2, "#T"             \n\t"\
00032         "psubw "#B", "#T"           \n\t"\
00033         "psubw "#E", "#T"           \n\t"\
00034         "punpcklbw "#Z", "#F"       \n\t"\
00035         "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
00036         "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
00037         "add %2, %0                 \n\t"\
00038         "paddw "#F", "#A"           \n\t"\
00039         "paddw "#A", "#T"           \n\t"\
00040         "psraw $5, "#T"             \n\t"\
00041         "packuswb "#T", "#T"        \n\t"\
00042         OP(T, (%1), A, d)\
00043         "add %3, %1                 \n\t"
00044 
00045 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
00046         "mov"#q" "#C", "#T"         \n\t"\
00047         "mov"#d" (%0), "#F"         \n\t"\
00048         "paddw "#D", "#T"           \n\t"\
00049         "psllw $2, "#T"             \n\t"\
00050         "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
00051         "psubw "#B", "#T"           \n\t"\
00052         "psubw "#E", "#T"           \n\t"\
00053         "punpcklbw "#Z", "#F"       \n\t"\
00054         "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
00055         "paddw "#F", "#A"           \n\t"\
00056         "add %2, %0                 \n\t"\
00057         "paddw "#A", "#T"           \n\t"\
00058         "mov"#q" "#T", "#OF"(%1)    \n\t"
00059 
00060 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
00061 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
00062 #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)
00063 #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)
00064 
00065 
00066 #define QPEL_H264(OPNAME, OP, MMX)\
00067 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00068     int h=4;\
00069 \
00070     __asm__ volatile(\
00071         "pxor %%mm7, %%mm7          \n\t"\
00072         "movq "MANGLE(ff_pw_5) ", %%mm4\n\t"\
00073         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00074         "1:                         \n\t"\
00075         "movd  -1(%0), %%mm1        \n\t"\
00076         "movd    (%0), %%mm2        \n\t"\
00077         "movd   1(%0), %%mm3        \n\t"\
00078         "movd   2(%0), %%mm0        \n\t"\
00079         "punpcklbw %%mm7, %%mm1     \n\t"\
00080         "punpcklbw %%mm7, %%mm2     \n\t"\
00081         "punpcklbw %%mm7, %%mm3     \n\t"\
00082         "punpcklbw %%mm7, %%mm0     \n\t"\
00083         "paddw %%mm0, %%mm1         \n\t"\
00084         "paddw %%mm3, %%mm2         \n\t"\
00085         "movd  -2(%0), %%mm0        \n\t"\
00086         "movd   3(%0), %%mm3        \n\t"\
00087         "punpcklbw %%mm7, %%mm0     \n\t"\
00088         "punpcklbw %%mm7, %%mm3     \n\t"\
00089         "paddw %%mm3, %%mm0         \n\t"\
00090         "psllw $2, %%mm2            \n\t"\
00091         "psubw %%mm1, %%mm2         \n\t"\
00092         "pmullw %%mm4, %%mm2        \n\t"\
00093         "paddw %%mm5, %%mm0         \n\t"\
00094         "paddw %%mm2, %%mm0         \n\t"\
00095         "psraw $5, %%mm0            \n\t"\
00096         "packuswb %%mm0, %%mm0      \n\t"\
00097         OP(%%mm0, (%1),%%mm6, d)\
00098         "add %3, %0                 \n\t"\
00099         "add %4, %1                 \n\t"\
00100         "decl %2                    \n\t"\
00101         " jnz 1b                    \n\t"\
00102         : "+a"(src), "+c"(dst), "+g"(h)\
00103         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00104         : "memory"\
00105     );\
00106 }\
00107 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00108     int h=4;\
00109     __asm__ volatile(\
00110         "pxor %%mm7, %%mm7          \n\t"\
00111         "movq %0, %%mm4             \n\t"\
00112         "movq %1, %%mm5             \n\t"\
00113         :: "m"(ff_pw_5), "m"(ff_pw_16)\
00114     );\
00115     do{\
00116     __asm__ volatile(\
00117         "movd  -1(%0), %%mm1        \n\t"\
00118         "movd    (%0), %%mm2        \n\t"\
00119         "movd   1(%0), %%mm3        \n\t"\
00120         "movd   2(%0), %%mm0        \n\t"\
00121         "punpcklbw %%mm7, %%mm1     \n\t"\
00122         "punpcklbw %%mm7, %%mm2     \n\t"\
00123         "punpcklbw %%mm7, %%mm3     \n\t"\
00124         "punpcklbw %%mm7, %%mm0     \n\t"\
00125         "paddw %%mm0, %%mm1         \n\t"\
00126         "paddw %%mm3, %%mm2         \n\t"\
00127         "movd  -2(%0), %%mm0        \n\t"\
00128         "movd   3(%0), %%mm3        \n\t"\
00129         "punpcklbw %%mm7, %%mm0     \n\t"\
00130         "punpcklbw %%mm7, %%mm3     \n\t"\
00131         "paddw %%mm3, %%mm0         \n\t"\
00132         "psllw $2, %%mm2            \n\t"\
00133         "psubw %%mm1, %%mm2         \n\t"\
00134         "pmullw %%mm4, %%mm2        \n\t"\
00135         "paddw %%mm5, %%mm0         \n\t"\
00136         "paddw %%mm2, %%mm0         \n\t"\
00137         "movd   (%2), %%mm3         \n\t"\
00138         "psraw $5, %%mm0            \n\t"\
00139         "packuswb %%mm0, %%mm0      \n\t"\
00140         PAVGB" %%mm3, %%mm0         \n\t"\
00141         OP(%%mm0, (%1),%%mm6, d)\
00142         "add %4, %0                 \n\t"\
00143         "add %4, %1                 \n\t"\
00144         "add %3, %2                 \n\t"\
00145         : "+a"(src), "+c"(dst), "+d"(src2)\
00146         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00147         : "memory"\
00148     );\
00149     }while(--h);\
00150 }\
00151 static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00152     src -= 2*srcStride;\
00153     __asm__ volatile(\
00154         "pxor %%mm7, %%mm7          \n\t"\
00155         "movd (%0), %%mm0           \n\t"\
00156         "add %2, %0                 \n\t"\
00157         "movd (%0), %%mm1           \n\t"\
00158         "add %2, %0                 \n\t"\
00159         "movd (%0), %%mm2           \n\t"\
00160         "add %2, %0                 \n\t"\
00161         "movd (%0), %%mm3           \n\t"\
00162         "add %2, %0                 \n\t"\
00163         "movd (%0), %%mm4           \n\t"\
00164         "add %2, %0                 \n\t"\
00165         "punpcklbw %%mm7, %%mm0     \n\t"\
00166         "punpcklbw %%mm7, %%mm1     \n\t"\
00167         "punpcklbw %%mm7, %%mm2     \n\t"\
00168         "punpcklbw %%mm7, %%mm3     \n\t"\
00169         "punpcklbw %%mm7, %%mm4     \n\t"\
00170         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00171         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00172         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00173         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00174          \
00175         : "+a"(src), "+c"(dst)\
00176         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
00177         : "memory"\
00178     );\
00179 }\
00180 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){\
00181     int h=4;\
00182     int w=3;\
00183     src -= 2*srcStride+2;\
00184     while(w--){\
00185         __asm__ volatile(\
00186             "pxor %%mm7, %%mm7      \n\t"\
00187             "movd (%0), %%mm0       \n\t"\
00188             "add %2, %0             \n\t"\
00189             "movd (%0), %%mm1       \n\t"\
00190             "add %2, %0             \n\t"\
00191             "movd (%0), %%mm2       \n\t"\
00192             "add %2, %0             \n\t"\
00193             "movd (%0), %%mm3       \n\t"\
00194             "add %2, %0             \n\t"\
00195             "movd (%0), %%mm4       \n\t"\
00196             "add %2, %0             \n\t"\
00197             "punpcklbw %%mm7, %%mm0 \n\t"\
00198             "punpcklbw %%mm7, %%mm1 \n\t"\
00199             "punpcklbw %%mm7, %%mm2 \n\t"\
00200             "punpcklbw %%mm7, %%mm3 \n\t"\
00201             "punpcklbw %%mm7, %%mm4 \n\t"\
00202             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
00203             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
00204             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
00205             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
00206              \
00207             : "+a"(src)\
00208             : "c"(tmp), "S"((x86_reg)srcStride)\
00209             : "memory"\
00210         );\
00211         tmp += 4;\
00212         src += 4 - 9*srcStride;\
00213     }\
00214     tmp -= 3*4;\
00215     __asm__ volatile(\
00216         "1:                         \n\t"\
00217         "movq     (%0), %%mm0       \n\t"\
00218         "paddw  10(%0), %%mm0       \n\t"\
00219         "movq    2(%0), %%mm1       \n\t"\
00220         "paddw   8(%0), %%mm1       \n\t"\
00221         "movq    4(%0), %%mm2       \n\t"\
00222         "paddw   6(%0), %%mm2       \n\t"\
00223         "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
00224         "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
00225         "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
00226         "paddsw %%mm2, %%mm0        \n\t"\
00227         "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
00228         "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
00229         "psraw $6, %%mm0            \n\t"\
00230         "packuswb %%mm0, %%mm0      \n\t"\
00231         OP(%%mm0, (%1),%%mm7, d)\
00232         "add $24, %0                \n\t"\
00233         "add %3, %1                 \n\t"\
00234         "decl %2                    \n\t"\
00235         " jnz 1b                    \n\t"\
00236         : "+a"(tmp), "+c"(dst), "+g"(h)\
00237         : "S"((x86_reg)dstStride)\
00238         : "memory"\
00239     );\
00240 }\
00241 \
00242 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00243     int h=8;\
00244     __asm__ volatile(\
00245         "pxor %%mm7, %%mm7          \n\t"\
00246         "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
00247         "1:                         \n\t"\
00248         "movq    (%0), %%mm0        \n\t"\
00249         "movq   1(%0), %%mm2        \n\t"\
00250         "movq %%mm0, %%mm1          \n\t"\
00251         "movq %%mm2, %%mm3          \n\t"\
00252         "punpcklbw %%mm7, %%mm0     \n\t"\
00253         "punpckhbw %%mm7, %%mm1     \n\t"\
00254         "punpcklbw %%mm7, %%mm2     \n\t"\
00255         "punpckhbw %%mm7, %%mm3     \n\t"\
00256         "paddw %%mm2, %%mm0         \n\t"\
00257         "paddw %%mm3, %%mm1         \n\t"\
00258         "psllw $2, %%mm0            \n\t"\
00259         "psllw $2, %%mm1            \n\t"\
00260         "movq   -1(%0), %%mm2       \n\t"\
00261         "movq    2(%0), %%mm4       \n\t"\
00262         "movq %%mm2, %%mm3          \n\t"\
00263         "movq %%mm4, %%mm5          \n\t"\
00264         "punpcklbw %%mm7, %%mm2     \n\t"\
00265         "punpckhbw %%mm7, %%mm3     \n\t"\
00266         "punpcklbw %%mm7, %%mm4     \n\t"\
00267         "punpckhbw %%mm7, %%mm5     \n\t"\
00268         "paddw %%mm4, %%mm2         \n\t"\
00269         "paddw %%mm3, %%mm5         \n\t"\
00270         "psubw %%mm2, %%mm0         \n\t"\
00271         "psubw %%mm5, %%mm1         \n\t"\
00272         "pmullw %%mm6, %%mm0        \n\t"\
00273         "pmullw %%mm6, %%mm1        \n\t"\
00274         "movd   -2(%0), %%mm2       \n\t"\
00275         "movd    7(%0), %%mm5       \n\t"\
00276         "punpcklbw %%mm7, %%mm2     \n\t"\
00277         "punpcklbw %%mm7, %%mm5     \n\t"\
00278         "paddw %%mm3, %%mm2         \n\t"\
00279         "paddw %%mm5, %%mm4         \n\t"\
00280         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00281         "paddw %%mm5, %%mm2         \n\t"\
00282         "paddw %%mm5, %%mm4         \n\t"\
00283         "paddw %%mm2, %%mm0         \n\t"\
00284         "paddw %%mm4, %%mm1         \n\t"\
00285         "psraw $5, %%mm0            \n\t"\
00286         "psraw $5, %%mm1            \n\t"\
00287         "packuswb %%mm1, %%mm0      \n\t"\
00288         OP(%%mm0, (%1),%%mm5, q)\
00289         "add %3, %0                 \n\t"\
00290         "add %4, %1                 \n\t"\
00291         "decl %2                    \n\t"\
00292         " jnz 1b                    \n\t"\
00293         : "+a"(src), "+c"(dst), "+g"(h)\
00294         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00295         : "memory"\
00296     );\
00297 }\
00298 \
00299 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00300     int h=8;\
00301     __asm__ volatile(\
00302         "pxor %%mm7, %%mm7          \n\t"\
00303         "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
00304         "1:                         \n\t"\
00305         "movq    (%0), %%mm0        \n\t"\
00306         "movq   1(%0), %%mm2        \n\t"\
00307         "movq %%mm0, %%mm1          \n\t"\
00308         "movq %%mm2, %%mm3          \n\t"\
00309         "punpcklbw %%mm7, %%mm0     \n\t"\
00310         "punpckhbw %%mm7, %%mm1     \n\t"\
00311         "punpcklbw %%mm7, %%mm2     \n\t"\
00312         "punpckhbw %%mm7, %%mm3     \n\t"\
00313         "paddw %%mm2, %%mm0         \n\t"\
00314         "paddw %%mm3, %%mm1         \n\t"\
00315         "psllw $2, %%mm0            \n\t"\
00316         "psllw $2, %%mm1            \n\t"\
00317         "movq   -1(%0), %%mm2       \n\t"\
00318         "movq    2(%0), %%mm4       \n\t"\
00319         "movq %%mm2, %%mm3          \n\t"\
00320         "movq %%mm4, %%mm5          \n\t"\
00321         "punpcklbw %%mm7, %%mm2     \n\t"\
00322         "punpckhbw %%mm7, %%mm3     \n\t"\
00323         "punpcklbw %%mm7, %%mm4     \n\t"\
00324         "punpckhbw %%mm7, %%mm5     \n\t"\
00325         "paddw %%mm4, %%mm2         \n\t"\
00326         "paddw %%mm3, %%mm5         \n\t"\
00327         "psubw %%mm2, %%mm0         \n\t"\
00328         "psubw %%mm5, %%mm1         \n\t"\
00329         "pmullw %%mm6, %%mm0        \n\t"\
00330         "pmullw %%mm6, %%mm1        \n\t"\
00331         "movd   -2(%0), %%mm2       \n\t"\
00332         "movd    7(%0), %%mm5       \n\t"\
00333         "punpcklbw %%mm7, %%mm2     \n\t"\
00334         "punpcklbw %%mm7, %%mm5     \n\t"\
00335         "paddw %%mm3, %%mm2         \n\t"\
00336         "paddw %%mm5, %%mm4         \n\t"\
00337         "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
00338         "paddw %%mm5, %%mm2         \n\t"\
00339         "paddw %%mm5, %%mm4         \n\t"\
00340         "paddw %%mm2, %%mm0         \n\t"\
00341         "paddw %%mm4, %%mm1         \n\t"\
00342         "psraw $5, %%mm0            \n\t"\
00343         "psraw $5, %%mm1            \n\t"\
00344         "movq (%2), %%mm4           \n\t"\
00345         "packuswb %%mm1, %%mm0      \n\t"\
00346         PAVGB" %%mm4, %%mm0         \n\t"\
00347         OP(%%mm0, (%1),%%mm5, q)\
00348         "add %5, %0                 \n\t"\
00349         "add %5, %1                 \n\t"\
00350         "add %4, %2                 \n\t"\
00351         "decl %3                    \n\t"\
00352         "jg 1b                      \n\t"\
00353         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00354         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00355         : "memory"\
00356     );\
00357 }\
00358 \
00359 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00360     int w= 2;\
00361     src -= 2*srcStride;\
00362     \
00363     while(w--){\
00364         __asm__ volatile(\
00365             "pxor %%mm7, %%mm7          \n\t"\
00366             "movd (%0), %%mm0           \n\t"\
00367             "add %2, %0                 \n\t"\
00368             "movd (%0), %%mm1           \n\t"\
00369             "add %2, %0                 \n\t"\
00370             "movd (%0), %%mm2           \n\t"\
00371             "add %2, %0                 \n\t"\
00372             "movd (%0), %%mm3           \n\t"\
00373             "add %2, %0                 \n\t"\
00374             "movd (%0), %%mm4           \n\t"\
00375             "add %2, %0                 \n\t"\
00376             "punpcklbw %%mm7, %%mm0     \n\t"\
00377             "punpcklbw %%mm7, %%mm1     \n\t"\
00378             "punpcklbw %%mm7, %%mm2     \n\t"\
00379             "punpcklbw %%mm7, %%mm3     \n\t"\
00380             "punpcklbw %%mm7, %%mm4     \n\t"\
00381             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00382             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00383             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00384             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00385             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00386             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00387             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00388             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00389             "cmpl $16, %4               \n\t"\
00390             "jne 2f                     \n\t"\
00391             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00392             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00393             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00394             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00395             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00396             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00397             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00398             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00399             "2:                         \n\t"\
00400             \
00401             : "+a"(src), "+c"(dst)\
00402             : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
00403             : "memory"\
00404         );\
00405         src += 4-(h+5)*srcStride;\
00406         dst += 4-h*dstStride;\
00407     }\
00408 }\
00409 static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
00410     int w = (size+8)>>2;\
00411     src -= 2*srcStride+2;\
00412     while(w--){\
00413         __asm__ volatile(\
00414             "pxor %%mm7, %%mm7      \n\t"\
00415             "movd (%0), %%mm0       \n\t"\
00416             "add %2, %0             \n\t"\
00417             "movd (%0), %%mm1       \n\t"\
00418             "add %2, %0             \n\t"\
00419             "movd (%0), %%mm2       \n\t"\
00420             "add %2, %0             \n\t"\
00421             "movd (%0), %%mm3       \n\t"\
00422             "add %2, %0             \n\t"\
00423             "movd (%0), %%mm4       \n\t"\
00424             "add %2, %0             \n\t"\
00425             "punpcklbw %%mm7, %%mm0 \n\t"\
00426             "punpcklbw %%mm7, %%mm1 \n\t"\
00427             "punpcklbw %%mm7, %%mm2 \n\t"\
00428             "punpcklbw %%mm7, %%mm3 \n\t"\
00429             "punpcklbw %%mm7, %%mm4 \n\t"\
00430             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
00431             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
00432             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
00433             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
00434             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
00435             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
00436             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
00437             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
00438             "cmpl $16, %3           \n\t"\
00439             "jne 2f                 \n\t"\
00440             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
00441             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
00442             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
00443             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
00444             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
00445             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
00446             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
00447             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
00448             "2:                     \n\t"\
00449             : "+a"(src)\
00450             : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)\
00451             : "memory"\
00452             );\
00453         tmp += 4;\
00454         src += 4 - (size+5)*srcStride;\
00455     }\
00456 }\
00457 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
00458     int w = size>>4;\
00459     do{\
00460     int h = size;\
00461     __asm__ volatile(\
00462         "1:                         \n\t"\
00463         "movq     (%0), %%mm0       \n\t"\
00464         "movq    8(%0), %%mm3       \n\t"\
00465         "movq    2(%0), %%mm1       \n\t"\
00466         "movq   10(%0), %%mm4       \n\t"\
00467         "paddw   %%mm4, %%mm0       \n\t"\
00468         "paddw   %%mm3, %%mm1       \n\t"\
00469         "paddw  18(%0), %%mm3       \n\t"\
00470         "paddw  16(%0), %%mm4       \n\t"\
00471         "movq    4(%0), %%mm2       \n\t"\
00472         "movq   12(%0), %%mm5       \n\t"\
00473         "paddw   6(%0), %%mm2       \n\t"\
00474         "paddw  14(%0), %%mm5       \n\t"\
00475         "psubw %%mm1, %%mm0         \n\t"\
00476         "psubw %%mm4, %%mm3         \n\t"\
00477         "psraw $2, %%mm0            \n\t"\
00478         "psraw $2, %%mm3            \n\t"\
00479         "psubw %%mm1, %%mm0         \n\t"\
00480         "psubw %%mm4, %%mm3         \n\t"\
00481         "paddsw %%mm2, %%mm0        \n\t"\
00482         "paddsw %%mm5, %%mm3        \n\t"\
00483         "psraw $2, %%mm0            \n\t"\
00484         "psraw $2, %%mm3            \n\t"\
00485         "paddw %%mm2, %%mm0         \n\t"\
00486         "paddw %%mm5, %%mm3         \n\t"\
00487         "psraw $6, %%mm0            \n\t"\
00488         "psraw $6, %%mm3            \n\t"\
00489         "packuswb %%mm3, %%mm0      \n\t"\
00490         OP(%%mm0, (%1),%%mm7, q)\
00491         "add $48, %0                \n\t"\
00492         "add %3, %1                 \n\t"\
00493         "decl %2                    \n\t"\
00494         " jnz 1b                    \n\t"\
00495         : "+a"(tmp), "+c"(dst), "+g"(h)\
00496         : "S"((x86_reg)dstStride)\
00497         : "memory"\
00498     );\
00499     tmp += 8 - size*24;\
00500     dst += 8 - size*dstStride;\
00501     }while(w--);\
00502 }\
00503 \
00504 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00505     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00506 }\
00507 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00508     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00509     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00510 }\
00511 \
00512 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00513     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00514     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00515     src += 8*srcStride;\
00516     dst += 8*dstStride;\
00517     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00518     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00519 }\
00520 \
00521 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00522     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00523     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00524     src += 8*dstStride;\
00525     dst += 8*dstStride;\
00526     src2 += 8*src2Stride;\
00527     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00528     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00529 }\
00530 \
00531 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){\
00532           put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
00533     OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
00534 }\
00535 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00536     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
00537 }\
00538 \
00539 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00540     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
00541 }\
00542 \
00543 static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00544 {\
00545     __asm__ volatile(\
00546         "movq      (%1), %%mm0          \n\t"\
00547         "movq    24(%1), %%mm1          \n\t"\
00548         "psraw      $5,  %%mm0          \n\t"\
00549         "psraw      $5,  %%mm1          \n\t"\
00550         "packuswb %%mm0, %%mm0          \n\t"\
00551         "packuswb %%mm1, %%mm1          \n\t"\
00552         PAVGB"     (%0), %%mm0          \n\t"\
00553         PAVGB"  (%0,%3), %%mm1          \n\t"\
00554         OP(%%mm0, (%2),    %%mm4, d)\
00555         OP(%%mm1, (%2,%4), %%mm5, d)\
00556         "lea  (%0,%3,2), %0             \n\t"\
00557         "lea  (%2,%4,2), %2             \n\t"\
00558         "movq    48(%1), %%mm0          \n\t"\
00559         "movq    72(%1), %%mm1          \n\t"\
00560         "psraw      $5,  %%mm0          \n\t"\
00561         "psraw      $5,  %%mm1          \n\t"\
00562         "packuswb %%mm0, %%mm0          \n\t"\
00563         "packuswb %%mm1, %%mm1          \n\t"\
00564         PAVGB"     (%0), %%mm0          \n\t"\
00565         PAVGB"  (%0,%3), %%mm1          \n\t"\
00566         OP(%%mm0, (%2),    %%mm4, d)\
00567         OP(%%mm1, (%2,%4), %%mm5, d)\
00568         :"+a"(src8), "+c"(src16), "+d"(dst)\
00569         :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
00570         :"memory");\
00571 }\
00572 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00573 {\
00574     do{\
00575     __asm__ volatile(\
00576         "movq      (%1), %%mm0          \n\t"\
00577         "movq     8(%1), %%mm1          \n\t"\
00578         "movq    48(%1), %%mm2          \n\t"\
00579         "movq  8+48(%1), %%mm3          \n\t"\
00580         "psraw      $5,  %%mm0          \n\t"\
00581         "psraw      $5,  %%mm1          \n\t"\
00582         "psraw      $5,  %%mm2          \n\t"\
00583         "psraw      $5,  %%mm3          \n\t"\
00584         "packuswb %%mm1, %%mm0          \n\t"\
00585         "packuswb %%mm3, %%mm2          \n\t"\
00586         PAVGB"     (%0), %%mm0          \n\t"\
00587         PAVGB"  (%0,%3), %%mm2          \n\t"\
00588         OP(%%mm0, (%2), %%mm5, q)\
00589         OP(%%mm2, (%2,%4), %%mm5, q)\
00590         ::"a"(src8), "c"(src16), "d"(dst),\
00591           "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
00592         :"memory");\
00593         src8 += 2L*src8Stride;\
00594         src16 += 48;\
00595         dst += 2L*dstStride;\
00596     }while(h-=2);\
00597 }\
00598 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
00599 {\
00600     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
00601     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
00602 }\
00603 
00604 
00605 #if ARCH_X86_64
00606 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00607 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00608     int h=16;\
00609     __asm__ volatile(\
00610         "pxor %%xmm15, %%xmm15      \n\t"\
00611         "movdqa %6, %%xmm14         \n\t"\
00612         "movdqa %7, %%xmm13         \n\t"\
00613         "1:                         \n\t"\
00614         "lddqu    6(%0), %%xmm1     \n\t"\
00615         "lddqu   -2(%0), %%xmm7     \n\t"\
00616         "movdqa  %%xmm1, %%xmm0     \n\t"\
00617         "punpckhbw %%xmm15, %%xmm1  \n\t"\
00618         "punpcklbw %%xmm15, %%xmm0  \n\t"\
00619         "punpcklbw %%xmm15, %%xmm7  \n\t"\
00620         "movdqa  %%xmm1, %%xmm2     \n\t"\
00621         "movdqa  %%xmm0, %%xmm6     \n\t"\
00622         "movdqa  %%xmm1, %%xmm3     \n\t"\
00623         "movdqa  %%xmm0, %%xmm8     \n\t"\
00624         "movdqa  %%xmm1, %%xmm4     \n\t"\
00625         "movdqa  %%xmm0, %%xmm9     \n\t"\
00626         "movdqa  %%xmm0, %%xmm12    \n\t"\
00627         "movdqa  %%xmm1, %%xmm11    \n\t"\
00628         "palignr $10,%%xmm0, %%xmm11\n\t"\
00629         "palignr $10,%%xmm7, %%xmm12\n\t"\
00630         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00631         "palignr $2, %%xmm7, %%xmm9 \n\t"\
00632         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00633         "palignr $4, %%xmm7, %%xmm8 \n\t"\
00634         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00635         "palignr $6, %%xmm7, %%xmm6 \n\t"\
00636         "paddw   %%xmm0 ,%%xmm11    \n\t"\
00637         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00638         "palignr $8, %%xmm7, %%xmm0 \n\t"\
00639         "paddw   %%xmm12,%%xmm7     \n\t"\
00640         "paddw   %%xmm3, %%xmm2     \n\t"\
00641         "paddw   %%xmm8, %%xmm6     \n\t"\
00642         "paddw   %%xmm4, %%xmm1     \n\t"\
00643         "paddw   %%xmm9, %%xmm0     \n\t"\
00644         "psllw   $2,     %%xmm2     \n\t"\
00645         "psllw   $2,     %%xmm6     \n\t"\
00646         "psubw   %%xmm1, %%xmm2     \n\t"\
00647         "psubw   %%xmm0, %%xmm6     \n\t"\
00648         "paddw   %%xmm13,%%xmm11    \n\t"\
00649         "paddw   %%xmm13,%%xmm7     \n\t"\
00650         "pmullw  %%xmm14,%%xmm2     \n\t"\
00651         "pmullw  %%xmm14,%%xmm6     \n\t"\
00652         "lddqu   (%2),   %%xmm3     \n\t"\
00653         "paddw   %%xmm11,%%xmm2     \n\t"\
00654         "paddw   %%xmm7, %%xmm6     \n\t"\
00655         "psraw   $5,     %%xmm2     \n\t"\
00656         "psraw   $5,     %%xmm6     \n\t"\
00657         "packuswb %%xmm2,%%xmm6     \n\t"\
00658         "pavgb   %%xmm3, %%xmm6     \n\t"\
00659         OP(%%xmm6, (%1), %%xmm4, dqa)\
00660         "add %5, %0                 \n\t"\
00661         "add %5, %1                 \n\t"\
00662         "add %4, %2                 \n\t"\
00663         "decl %3                    \n\t"\
00664         "jg 1b                      \n\t"\
00665         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00666         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
00667           "m"(ff_pw_5), "m"(ff_pw_16)\
00668         : XMM_CLOBBERS("%xmm0" , "%xmm1" , "%xmm2" , "%xmm3" , \
00669                        "%xmm4" , "%xmm5" , "%xmm6" , "%xmm7" , \
00670                        "%xmm8" , "%xmm9" , "%xmm10", "%xmm11", \
00671                        "%xmm12", "%xmm13", "%xmm14", "%xmm15",)\
00672           "memory"\
00673     );\
00674 }
00675 #else // ARCH_X86_64
00676 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00677 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00678     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00679     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00680     src += 8*dstStride;\
00681     dst += 8*dstStride;\
00682     src2 += 8*src2Stride;\
00683     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
00684     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
00685 }
00686 #endif // ARCH_X86_64
00687 
00688 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
00689 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
00690     int h=8;\
00691     __asm__ volatile(\
00692         "pxor %%xmm7, %%xmm7        \n\t"\
00693         "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
00694         "1:                         \n\t"\
00695         "lddqu   -2(%0), %%xmm1     \n\t"\
00696         "movdqa  %%xmm1, %%xmm0     \n\t"\
00697         "punpckhbw %%xmm7, %%xmm1   \n\t"\
00698         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00699         "movdqa  %%xmm1, %%xmm2     \n\t"\
00700         "movdqa  %%xmm1, %%xmm3     \n\t"\
00701         "movdqa  %%xmm1, %%xmm4     \n\t"\
00702         "movdqa  %%xmm1, %%xmm5     \n\t"\
00703         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00704         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00705         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00706         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00707         "palignr $10,%%xmm0, %%xmm5 \n\t"\
00708         "paddw   %%xmm5, %%xmm0     \n\t"\
00709         "paddw   %%xmm3, %%xmm2     \n\t"\
00710         "paddw   %%xmm4, %%xmm1     \n\t"\
00711         "psllw   $2,     %%xmm2     \n\t"\
00712         "movq    (%2),   %%xmm3     \n\t"\
00713         "psubw   %%xmm1, %%xmm2     \n\t"\
00714         "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
00715         "pmullw  %%xmm6, %%xmm2     \n\t"\
00716         "paddw   %%xmm0, %%xmm2     \n\t"\
00717         "psraw   $5,     %%xmm2     \n\t"\
00718         "packuswb %%xmm2, %%xmm2    \n\t"\
00719         "pavgb   %%xmm3, %%xmm2     \n\t"\
00720         OP(%%xmm2, (%1), %%xmm4, q)\
00721         "add %5, %0                 \n\t"\
00722         "add %5, %1                 \n\t"\
00723         "add %4, %2                 \n\t"\
00724         "decl %3                    \n\t"\
00725         "jg 1b                      \n\t"\
00726         : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
00727         : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
00728         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00729                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00730           "memory"\
00731     );\
00732 }\
00733 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
00734 \
00735 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00736     int h=8;\
00737     __asm__ volatile(\
00738         "pxor %%xmm7, %%xmm7        \n\t"\
00739         "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
00740         "1:                         \n\t"\
00741         "lddqu   -2(%0), %%xmm1     \n\t"\
00742         "movdqa  %%xmm1, %%xmm0     \n\t"\
00743         "punpckhbw %%xmm7, %%xmm1   \n\t"\
00744         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00745         "movdqa  %%xmm1, %%xmm2     \n\t"\
00746         "movdqa  %%xmm1, %%xmm3     \n\t"\
00747         "movdqa  %%xmm1, %%xmm4     \n\t"\
00748         "movdqa  %%xmm1, %%xmm5     \n\t"\
00749         "palignr $2, %%xmm0, %%xmm4 \n\t"\
00750         "palignr $4, %%xmm0, %%xmm3 \n\t"\
00751         "palignr $6, %%xmm0, %%xmm2 \n\t"\
00752         "palignr $8, %%xmm0, %%xmm1 \n\t"\
00753         "palignr $10,%%xmm0, %%xmm5 \n\t"\
00754         "paddw   %%xmm5, %%xmm0     \n\t"\
00755         "paddw   %%xmm3, %%xmm2     \n\t"\
00756         "paddw   %%xmm4, %%xmm1     \n\t"\
00757         "psllw   $2,     %%xmm2     \n\t"\
00758         "psubw   %%xmm1, %%xmm2     \n\t"\
00759         "paddw   "MANGLE(ff_pw_16)", %%xmm0\n\t"\
00760         "pmullw  %%xmm6, %%xmm2     \n\t"\
00761         "paddw   %%xmm0, %%xmm2     \n\t"\
00762         "psraw   $5,     %%xmm2     \n\t"\
00763         "packuswb %%xmm2, %%xmm2    \n\t"\
00764         OP(%%xmm2, (%1), %%xmm4, q)\
00765         "add %3, %0                 \n\t"\
00766         "add %4, %1                 \n\t"\
00767         "decl %2                    \n\t"\
00768         " jnz 1b                    \n\t"\
00769         : "+a"(src), "+c"(dst), "+g"(h)\
00770         : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
00771         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00772                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00773           "memory"\
00774     );\
00775 }\
00776 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00777     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00778     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00779     src += 8*srcStride;\
00780     dst += 8*dstStride;\
00781     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
00782     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00783 }\
00784 
00785 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
00786 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00787     src -= 2*srcStride;\
00788     \
00789     __asm__ volatile(\
00790         "pxor %%xmm7, %%xmm7        \n\t"\
00791         "movq (%0), %%xmm0          \n\t"\
00792         "add %2, %0                 \n\t"\
00793         "movq (%0), %%xmm1          \n\t"\
00794         "add %2, %0                 \n\t"\
00795         "movq (%0), %%xmm2          \n\t"\
00796         "add %2, %0                 \n\t"\
00797         "movq (%0), %%xmm3          \n\t"\
00798         "add %2, %0                 \n\t"\
00799         "movq (%0), %%xmm4          \n\t"\
00800         "add %2, %0                 \n\t"\
00801         "punpcklbw %%xmm7, %%xmm0   \n\t"\
00802         "punpcklbw %%xmm7, %%xmm1   \n\t"\
00803         "punpcklbw %%xmm7, %%xmm2   \n\t"\
00804         "punpcklbw %%xmm7, %%xmm3   \n\t"\
00805         "punpcklbw %%xmm7, %%xmm4   \n\t"\
00806         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00807         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00808         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00809         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00810         QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
00811         QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
00812         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00813         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00814         "cmpl $16, %4               \n\t"\
00815         "jne 2f                     \n\t"\
00816         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00817         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00818         QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
00819         QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
00820         QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
00821         QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
00822         QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
00823         QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
00824         "2:                          \n\t"\
00825         \
00826         : "+a"(src), "+c"(dst)\
00827         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
00828         : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00829                        "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00830           "memory"\
00831     );\
00832 }\
00833 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00834     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00835 }\
00836 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00837     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00838     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00839 }
00840 
00841 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
00842     int w = (size+8)>>3;
00843     src -= 2*srcStride+2;
00844     while(w--){
00845         __asm__ volatile(
00846             "pxor %%xmm7, %%xmm7        \n\t"
00847             "movq (%0), %%xmm0          \n\t"
00848             "add %2, %0                 \n\t"
00849             "movq (%0), %%xmm1          \n\t"
00850             "add %2, %0                 \n\t"
00851             "movq (%0), %%xmm2          \n\t"
00852             "add %2, %0                 \n\t"
00853             "movq (%0), %%xmm3          \n\t"
00854             "add %2, %0                 \n\t"
00855             "movq (%0), %%xmm4          \n\t"
00856             "add %2, %0                 \n\t"
00857             "punpcklbw %%xmm7, %%xmm0   \n\t"
00858             "punpcklbw %%xmm7, %%xmm1   \n\t"
00859             "punpcklbw %%xmm7, %%xmm2   \n\t"
00860             "punpcklbw %%xmm7, %%xmm3   \n\t"
00861             "punpcklbw %%xmm7, %%xmm4   \n\t"
00862             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
00863             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
00864             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
00865             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
00866             QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
00867             QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
00868             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
00869             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
00870             "cmpl $16, %3               \n\t"
00871             "jne 2f                     \n\t"
00872             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
00873             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
00874             QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
00875             QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
00876             QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
00877             QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
00878             QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
00879             QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
00880             "2:                         \n\t"
00881             : "+a"(src)
00882             : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)
00883             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3",
00884                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)
00885               "memory"
00886         );
00887         tmp += 8;
00888         src += 8 - (size+5)*srcStride;
00889     }
00890 }
00891 
00892 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
00893 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
00894     int h = size;\
00895     if(size == 16){\
00896         __asm__ volatile(\
00897             "1:                         \n\t"\
00898             "movdqa 32(%0), %%xmm4      \n\t"\
00899             "movdqa 16(%0), %%xmm5      \n\t"\
00900             "movdqa   (%0), %%xmm7      \n\t"\
00901             "movdqa %%xmm4, %%xmm3      \n\t"\
00902             "movdqa %%xmm4, %%xmm2      \n\t"\
00903             "movdqa %%xmm4, %%xmm1      \n\t"\
00904             "movdqa %%xmm4, %%xmm0      \n\t"\
00905             "palignr $10, %%xmm5, %%xmm0 \n\t"\
00906             "palignr  $8, %%xmm5, %%xmm1 \n\t"\
00907             "palignr  $6, %%xmm5, %%xmm2 \n\t"\
00908             "palignr  $4, %%xmm5, %%xmm3 \n\t"\
00909             "palignr  $2, %%xmm5, %%xmm4 \n\t"\
00910             "paddw  %%xmm5, %%xmm0      \n\t"\
00911             "paddw  %%xmm4, %%xmm1      \n\t"\
00912             "paddw  %%xmm3, %%xmm2      \n\t"\
00913             "movdqa %%xmm5, %%xmm6      \n\t"\
00914             "movdqa %%xmm5, %%xmm4      \n\t"\
00915             "movdqa %%xmm5, %%xmm3      \n\t"\
00916             "palignr  $8, %%xmm7, %%xmm4 \n\t"\
00917             "palignr  $2, %%xmm7, %%xmm6 \n\t"\
00918             "palignr $10, %%xmm7, %%xmm3 \n\t"\
00919             "paddw  %%xmm6, %%xmm4      \n\t"\
00920             "movdqa %%xmm5, %%xmm6      \n\t"\
00921             "palignr  $6, %%xmm7, %%xmm5 \n\t"\
00922             "palignr  $4, %%xmm7, %%xmm6 \n\t"\
00923             "paddw  %%xmm7, %%xmm3      \n\t"\
00924             "paddw  %%xmm6, %%xmm5      \n\t"\
00925             \
00926             "psubw  %%xmm1, %%xmm0      \n\t"\
00927             "psubw  %%xmm4, %%xmm3      \n\t"\
00928             "psraw      $2, %%xmm0      \n\t"\
00929             "psraw      $2, %%xmm3      \n\t"\
00930             "psubw  %%xmm1, %%xmm0      \n\t"\
00931             "psubw  %%xmm4, %%xmm3      \n\t"\
00932             "paddw  %%xmm2, %%xmm0      \n\t"\
00933             "paddw  %%xmm5, %%xmm3      \n\t"\
00934             "psraw      $2, %%xmm0      \n\t"\
00935             "psraw      $2, %%xmm3      \n\t"\
00936             "paddw  %%xmm2, %%xmm0      \n\t"\
00937             "paddw  %%xmm5, %%xmm3      \n\t"\
00938             "psraw      $6, %%xmm0      \n\t"\
00939             "psraw      $6, %%xmm3      \n\t"\
00940             "packuswb %%xmm0, %%xmm3    \n\t"\
00941             OP(%%xmm3, (%1), %%xmm7, dqa)\
00942             "add $48, %0                \n\t"\
00943             "add %3, %1                 \n\t"\
00944             "decl %2                    \n\t"\
00945             " jnz 1b                    \n\t"\
00946             : "+a"(tmp), "+c"(dst), "+g"(h)\
00947             : "S"((x86_reg)dstStride)\
00948             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00949                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00950               "memory"\
00951         );\
00952     }else{\
00953         __asm__ volatile(\
00954             "1:                         \n\t"\
00955             "movdqa 16(%0), %%xmm1      \n\t"\
00956             "movdqa   (%0), %%xmm0      \n\t"\
00957             "movdqa %%xmm1, %%xmm2      \n\t"\
00958             "movdqa %%xmm1, %%xmm3      \n\t"\
00959             "movdqa %%xmm1, %%xmm4      \n\t"\
00960             "movdqa %%xmm1, %%xmm5      \n\t"\
00961             "palignr $10, %%xmm0, %%xmm5 \n\t"\
00962             "palignr  $8, %%xmm0, %%xmm4 \n\t"\
00963             "palignr  $6, %%xmm0, %%xmm3 \n\t"\
00964             "palignr  $4, %%xmm0, %%xmm2 \n\t"\
00965             "palignr  $2, %%xmm0, %%xmm1 \n\t"\
00966             "paddw  %%xmm5, %%xmm0      \n\t"\
00967             "paddw  %%xmm4, %%xmm1      \n\t"\
00968             "paddw  %%xmm3, %%xmm2      \n\t"\
00969             "psubw  %%xmm1, %%xmm0      \n\t"\
00970             "psraw      $2, %%xmm0      \n\t"\
00971             "psubw  %%xmm1, %%xmm0      \n\t"\
00972             "paddw  %%xmm2, %%xmm0      \n\t"\
00973             "psraw      $2, %%xmm0      \n\t"\
00974             "paddw  %%xmm2, %%xmm0      \n\t"\
00975             "psraw      $6, %%xmm0      \n\t"\
00976             "packuswb %%xmm0, %%xmm0    \n\t"\
00977             OP(%%xmm0, (%1), %%xmm7, q)\
00978             "add $48, %0                \n\t"\
00979             "add %3, %1                 \n\t"\
00980             "decl %2                    \n\t"\
00981             " jnz 1b                    \n\t"\
00982             : "+a"(tmp), "+c"(dst), "+g"(h)\
00983             : "S"((x86_reg)dstStride)\
00984             : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00985                            "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
00986               "memory"\
00987         );\
00988     }\
00989 }
00990 
00991 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
00992 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){\
00993           put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
00994     OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
00995 }\
00996 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
00997     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
00998 }\
00999 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
01000     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
01001 }\
01002 
01003 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
01004 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
01005 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
01006 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
01007 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
01008 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
01009 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
01010 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
01011 
01012 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
01013 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
01014 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
01015 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
01016 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
01017 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
01018 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
01019 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
01020 
01021 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
01022 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
01023 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
01024 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
01025 
01026 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
01027 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
01028 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
01029 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
01030 
01031 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
01032 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
01033 
01034 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
01035 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
01036 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
01037 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
01038 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
01039 
01040 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
01041     put_pixels16_sse2(dst, src, stride, 16);
01042 }
01043 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
01044     avg_pixels16_sse2(dst, src, stride, 16);
01045 }
01046 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
01047 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
01048 
01049 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
01050 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
01051     OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
01052 }\
01053 
01054 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
01055 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01056     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
01057 }\
01058 \
01059 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01060     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
01061 }\
01062 \
01063 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01064     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
01065 }\
01066 
01067 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
01068 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01069     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01070     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01071     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
01072 }\
01073 \
01074 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01075     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
01076 }\
01077 \
01078 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01079     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01080     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01081     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
01082 }\
01083 
01084 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
01085 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01086     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01087     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01088     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
01089 }\
01090 \
01091 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01092     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01093     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
01094     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
01095 }\
01096 \
01097 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01098     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01099     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
01100     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
01101 }\
01102 \
01103 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01104     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
01105     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
01106     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
01107 }\
01108 \
01109 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01110     DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
01111     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
01112 }\
01113 \
01114 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01115     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01116     uint8_t * const halfHV= temp;\
01117     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01118     assert(((int)temp & 7) == 0);\
01119     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01120     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
01121 }\
01122 \
01123 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01124     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01125     uint8_t * const halfHV= temp;\
01126     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01127     assert(((int)temp & 7) == 0);\
01128     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01129     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
01130 }\
01131 \
01132 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01133     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01134     uint8_t * const halfHV= temp;\
01135     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01136     assert(((int)temp & 7) == 0);\
01137     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01138     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
01139 }\
01140 \
01141 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01142     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
01143     uint8_t * const halfHV= temp;\
01144     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
01145     assert(((int)temp & 7) == 0);\
01146     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
01147     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
01148 }\
01149 
01150 #define H264_MC_4816(MMX)\
01151 H264_MC(put_, 4, MMX, 8)\
01152 H264_MC(put_, 8, MMX, 8)\
01153 H264_MC(put_, 16,MMX, 8)\
01154 H264_MC(avg_, 4, MMX, 8)\
01155 H264_MC(avg_, 8, MMX, 8)\
01156 H264_MC(avg_, 16,MMX, 8)\
01157 
01158 #define H264_MC_816(QPEL, XMM)\
01159 QPEL(put_, 8, XMM, 16)\
01160 QPEL(put_, 16,XMM, 16)\
01161 QPEL(avg_, 8, XMM, 16)\
01162 QPEL(avg_, 16,XMM, 16)\
01163 
01164 
01165 #define AVG_3DNOW_OP(a,b,temp, size) \
01166 "mov" #size " " #b ", " #temp "   \n\t"\
01167 "pavgusb " #temp ", " #a "        \n\t"\
01168 "mov" #size " " #a ", " #b "      \n\t"
01169 #define AVG_MMX2_OP(a,b,temp, size) \
01170 "mov" #size " " #b ", " #temp "   \n\t"\
01171 "pavgb " #temp ", " #a "          \n\t"\
01172 "mov" #size " " #a ", " #b "      \n\t"
01173 
01174 #define PAVGB "pavgusb"
01175 QPEL_H264(put_,       PUT_OP, 3dnow)
01176 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
01177 #undef PAVGB
01178 #define PAVGB "pavgb"
01179 QPEL_H264(put_,       PUT_OP, mmx2)
01180 QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
01181 QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
01182 QPEL_H264_V_XMM(avg_,  AVG_MMX2_OP, sse2)
01183 QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
01184 QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, sse2)
01185 #if HAVE_SSSE3
01186 QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
01187 QPEL_H264_H_XMM(avg_,  AVG_MMX2_OP, ssse3)
01188 QPEL_H264_HV2_XMM(put_,       PUT_OP, ssse3)
01189 QPEL_H264_HV2_XMM(avg_,  AVG_MMX2_OP, ssse3)
01190 QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
01191 QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, ssse3)
01192 #endif
01193 #undef PAVGB
01194 
01195 H264_MC_4816(3dnow)
01196 H264_MC_4816(mmx2)
01197 H264_MC_816(H264_MC_V, sse2)
01198 H264_MC_816(H264_MC_HV, sse2)
01199 #if HAVE_SSSE3
01200 H264_MC_816(H264_MC_H, ssse3)
01201 H264_MC_816(H264_MC_HV, ssse3)
01202 #endif
01203 
01204 
01205 
01206 //10bit
01207 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
01208 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
01209     (uint8_t *dst, uint8_t *src, int stride);
01210 
01211 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
01212     LUMA_MC_OP(put,  4, DEPTH, TYPE, OPT) \
01213     LUMA_MC_OP(avg,  4, DEPTH, TYPE, OPT) \
01214     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
01215     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
01216     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
01217     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
01218 
01219 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
01220     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
01221     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
01222     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
01223     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
01224 
01225 LUMA_MC_ALL(10, mc00, mmxext)
01226 LUMA_MC_ALL(10, mc10, mmxext)
01227 LUMA_MC_ALL(10, mc20, mmxext)
01228 LUMA_MC_ALL(10, mc30, mmxext)
01229 LUMA_MC_ALL(10, mc01, mmxext)
01230 LUMA_MC_ALL(10, mc11, mmxext)
01231 LUMA_MC_ALL(10, mc21, mmxext)
01232 LUMA_MC_ALL(10, mc31, mmxext)
01233 LUMA_MC_ALL(10, mc02, mmxext)
01234 LUMA_MC_ALL(10, mc12, mmxext)
01235 LUMA_MC_ALL(10, mc22, mmxext)
01236 LUMA_MC_ALL(10, mc32, mmxext)
01237 LUMA_MC_ALL(10, mc03, mmxext)
01238 LUMA_MC_ALL(10, mc13, mmxext)
01239 LUMA_MC_ALL(10, mc23, mmxext)
01240 LUMA_MC_ALL(10, mc33, mmxext)
01241 
01242 LUMA_MC_816(10, mc00, sse2)
01243 LUMA_MC_816(10, mc10, sse2)
01244 LUMA_MC_816(10, mc10, sse2_cache64)
01245 LUMA_MC_816(10, mc10, ssse3_cache64)
01246 LUMA_MC_816(10, mc20, sse2)
01247 LUMA_MC_816(10, mc20, sse2_cache64)
01248 LUMA_MC_816(10, mc20, ssse3_cache64)
01249 LUMA_MC_816(10, mc30, sse2)
01250 LUMA_MC_816(10, mc30, sse2_cache64)
01251 LUMA_MC_816(10, mc30, ssse3_cache64)
01252 LUMA_MC_816(10, mc01, sse2)
01253 LUMA_MC_816(10, mc11, sse2)
01254 LUMA_MC_816(10, mc21, sse2)
01255 LUMA_MC_816(10, mc31, sse2)
01256 LUMA_MC_816(10, mc02, sse2)
01257 LUMA_MC_816(10, mc12, sse2)
01258 LUMA_MC_816(10, mc22, sse2)
01259 LUMA_MC_816(10, mc32, sse2)
01260 LUMA_MC_816(10, mc03, sse2)
01261 LUMA_MC_816(10, mc13, sse2)
01262 LUMA_MC_816(10, mc23, sse2)
01263 LUMA_MC_816(10, mc33, sse2)
01264 
01265 #define QPEL16_OPMC(OP, MC, MMX)\
01266 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
01267     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
01268     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
01269     src += 8*stride;\
01270     dst += 8*stride;\
01271     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
01272     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
01273 }
01274 
01275 #define QPEL16_OP(MC, MMX)\
01276 QPEL16_OPMC(put, MC, MMX)\
01277 QPEL16_OPMC(avg, MC, MMX)
01278 
01279 #define QPEL16(MMX)\
01280 QPEL16_OP(mc00, MMX)\
01281 QPEL16_OP(mc01, MMX)\
01282 QPEL16_OP(mc02, MMX)\
01283 QPEL16_OP(mc03, MMX)\
01284 QPEL16_OP(mc10, MMX)\
01285 QPEL16_OP(mc11, MMX)\
01286 QPEL16_OP(mc12, MMX)\
01287 QPEL16_OP(mc13, MMX)\
01288 QPEL16_OP(mc20, MMX)\
01289 QPEL16_OP(mc21, MMX)\
01290 QPEL16_OP(mc22, MMX)\
01291 QPEL16_OP(mc23, MMX)\
01292 QPEL16_OP(mc30, MMX)\
01293 QPEL16_OP(mc31, MMX)\
01294 QPEL16_OP(mc32, MMX)\
01295 QPEL16_OP(mc33, MMX)
01296 
01297 #if ARCH_X86_32 && HAVE_YASM // ARCH_X86_64 implies sse2+
01298 QPEL16(mmxext)
01299 #endif
libavcodec/x86/h264_qpel_mmx.c
