FFmpeg: libavcodec/x86/dsputil_h264_template

00001 /*
00002  * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>,
00003  *                    Loren Merritt
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 
00028 static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, const uint64_t *rnd_reg)
00029 {
00030     DECLARE_ALIGNED_8(uint64_t, AA);
00031     DECLARE_ALIGNED_8(uint64_t, DD);
00032     int i;
00033 
00034     if(y==0 && x==0) {
00035         /* no filter needed */
00036         H264_CHROMA_MC8_MV0(dst, src, stride, h);
00037         return;
00038     }
00039 
00040     assert(x<8 && y<8 && x>=0 && y>=0);
00041 
00042     if(y==0 || x==0)
00043     {
00044         /* 1 dimensional filter only */
00045         const int dxy = x ? 1 : stride;
00046 
00047         __asm__ volatile(
00048             "movd %0, %%mm5\n\t"
00049             "movq %1, %%mm4\n\t"
00050             "movq %2, %%mm6\n\t"         /* mm6 = rnd >> 3 */
00051             "punpcklwd %%mm5, %%mm5\n\t"
00052             "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
00053             "pxor %%mm7, %%mm7\n\t"
00054             "psubw %%mm5, %%mm4\n\t"     /* mm4 = A = 8-x */
00055             :: "rm"(x+y), "m"(ff_pw_8), "m"(*(rnd_reg+1)));
00056 
00057         for(i=0; i<h; i++) {
00058             __asm__ volatile(
00059                 /* mm0 = src[0..7], mm1 = src[1..8] */
00060                 "movq %0, %%mm0\n\t"
00061                 "movq %1, %%mm2\n\t"
00062                 :: "m"(src[0]), "m"(src[dxy]));
00063 
00064             __asm__ volatile(
00065                 /* [mm0,mm1] = A * src[0..7] */
00066                 /* [mm2,mm3] = B * src[1..8] */
00067                 "movq %%mm0, %%mm1\n\t"
00068                 "movq %%mm2, %%mm3\n\t"
00069                 "punpcklbw %%mm7, %%mm0\n\t"
00070                 "punpckhbw %%mm7, %%mm1\n\t"
00071                 "punpcklbw %%mm7, %%mm2\n\t"
00072                 "punpckhbw %%mm7, %%mm3\n\t"
00073                 "pmullw %%mm4, %%mm0\n\t"
00074                 "pmullw %%mm4, %%mm1\n\t"
00075                 "pmullw %%mm5, %%mm2\n\t"
00076                 "pmullw %%mm5, %%mm3\n\t"
00077 
00078                 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3 */
00079                 "paddw %%mm6, %%mm0\n\t"
00080                 "paddw %%mm6, %%mm1\n\t"
00081                 "paddw %%mm2, %%mm0\n\t"
00082                 "paddw %%mm3, %%mm1\n\t"
00083                 "psrlw $3, %%mm0\n\t"
00084                 "psrlw $3, %%mm1\n\t"
00085                 "packuswb %%mm1, %%mm0\n\t"
00086                 H264_CHROMA_OP(%0, %%mm0)
00087                 "movq %%mm0, %0\n\t"
00088                 : "=m" (dst[0]));
00089 
00090             src += stride;
00091             dst += stride;
00092         }
00093         return;
00094     }
00095 
00096     /* general case, bilinear */
00097     __asm__ volatile("movd %2, %%mm4\n\t"
00098                  "movd %3, %%mm6\n\t"
00099                  "punpcklwd %%mm4, %%mm4\n\t"
00100                  "punpcklwd %%mm6, %%mm6\n\t"
00101                  "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
00102                  "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */
00103                  "movq %%mm4, %%mm5\n\t"
00104                  "pmullw %%mm6, %%mm4\n\t"    /* mm4 = x * y */
00105                  "psllw $3, %%mm5\n\t"
00106                  "psllw $3, %%mm6\n\t"
00107                  "movq %%mm5, %%mm7\n\t"
00108                  "paddw %%mm6, %%mm7\n\t"
00109                  "movq %%mm4, %1\n\t"         /* DD = x * y */
00110                  "psubw %%mm4, %%mm5\n\t"     /* mm5 = B = 8x - xy */
00111                  "psubw %%mm4, %%mm6\n\t"     /* mm6 = C = 8y - xy */
00112                  "paddw %4, %%mm4\n\t"
00113                  "psubw %%mm7, %%mm4\n\t"     /* mm4 = A = xy - (8x+8y) + 64 */
00114                  "pxor %%mm7, %%mm7\n\t"
00115                  "movq %%mm4, %0\n\t"
00116                  : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64));
00117 
00118     __asm__ volatile(
00119         /* mm0 = src[0..7], mm1 = src[1..8] */
00120         "movq %0, %%mm0\n\t"
00121         "movq %1, %%mm1\n\t"
00122         : : "m" (src[0]), "m" (src[1]));
00123 
00124     for(i=0; i<h; i++) {
00125         src += stride;
00126 
00127         __asm__ volatile(
00128             /* mm2 = A * src[0..3] + B * src[1..4] */
00129             /* mm3 = A * src[4..7] + B * src[5..8] */
00130             "movq %%mm0, %%mm2\n\t"
00131             "movq %%mm1, %%mm3\n\t"
00132             "punpckhbw %%mm7, %%mm0\n\t"
00133             "punpcklbw %%mm7, %%mm1\n\t"
00134             "punpcklbw %%mm7, %%mm2\n\t"
00135             "punpckhbw %%mm7, %%mm3\n\t"
00136             "pmullw %0, %%mm0\n\t"
00137             "pmullw %0, %%mm2\n\t"
00138             "pmullw %%mm5, %%mm1\n\t"
00139             "pmullw %%mm5, %%mm3\n\t"
00140             "paddw %%mm1, %%mm2\n\t"
00141             "paddw %%mm0, %%mm3\n\t"
00142             : : "m" (AA));
00143 
00144         __asm__ volatile(
00145             /* [mm2,mm3] += C * src[0..7] */
00146             "movq %0, %%mm0\n\t"
00147             "movq %%mm0, %%mm1\n\t"
00148             "punpcklbw %%mm7, %%mm0\n\t"
00149             "punpckhbw %%mm7, %%mm1\n\t"
00150             "pmullw %%mm6, %%mm0\n\t"
00151             "pmullw %%mm6, %%mm1\n\t"
00152             "paddw %%mm0, %%mm2\n\t"
00153             "paddw %%mm1, %%mm3\n\t"
00154             : : "m" (src[0]));
00155 
00156         __asm__ volatile(
00157             /* [mm2,mm3] += D * src[1..8] */
00158             "movq %1, %%mm1\n\t"
00159             "movq %%mm1, %%mm0\n\t"
00160             "movq %%mm1, %%mm4\n\t"
00161             "punpcklbw %%mm7, %%mm0\n\t"
00162             "punpckhbw %%mm7, %%mm4\n\t"
00163             "pmullw %2, %%mm0\n\t"
00164             "pmullw %2, %%mm4\n\t"
00165             "paddw %%mm0, %%mm2\n\t"
00166             "paddw %%mm4, %%mm3\n\t"
00167             "movq %0, %%mm0\n\t"
00168             : : "m" (src[0]), "m" (src[1]), "m" (DD));
00169 
00170         __asm__ volatile(
00171             /* dst[0..7] = ([mm2,mm3] + rnd) >> 6 */
00172             "paddw %1, %%mm2\n\t"
00173             "paddw %1, %%mm3\n\t"
00174             "psrlw $6, %%mm2\n\t"
00175             "psrlw $6, %%mm3\n\t"
00176             "packuswb %%mm3, %%mm2\n\t"
00177             H264_CHROMA_OP(%0, %%mm2)
00178             "movq %%mm2, %0\n\t"
00179             : "=m" (dst[0]) : "m" (*rnd_reg));
00180         dst+= stride;
00181     }
00182 }
00183 
00184 static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, const uint64_t *rnd_reg)
00185 {
00186     __asm__ volatile(
00187         "pxor   %%mm7, %%mm7        \n\t"
00188         "movd %5, %%mm2             \n\t"
00189         "movd %6, %%mm3             \n\t"
00190         "movq "MANGLE(ff_pw_8)", %%mm4\n\t"
00191         "movq "MANGLE(ff_pw_8)", %%mm5\n\t"
00192         "punpcklwd %%mm2, %%mm2     \n\t"
00193         "punpcklwd %%mm3, %%mm3     \n\t"
00194         "punpcklwd %%mm2, %%mm2     \n\t"
00195         "punpcklwd %%mm3, %%mm3     \n\t"
00196         "psubw %%mm2, %%mm4         \n\t"
00197         "psubw %%mm3, %%mm5         \n\t"
00198 
00199         "movd  (%1), %%mm0          \n\t"
00200         "movd 1(%1), %%mm6          \n\t"
00201         "add %3, %1                 \n\t"
00202         "punpcklbw %%mm7, %%mm0     \n\t"
00203         "punpcklbw %%mm7, %%mm6     \n\t"
00204         "pmullw %%mm4, %%mm0        \n\t"
00205         "pmullw %%mm2, %%mm6        \n\t"
00206         "paddw %%mm0, %%mm6         \n\t"
00207 
00208         "1:                         \n\t"
00209         "movd  (%1), %%mm0          \n\t"
00210         "movd 1(%1), %%mm1          \n\t"
00211         "add %3, %1                 \n\t"
00212         "punpcklbw %%mm7, %%mm0     \n\t"
00213         "punpcklbw %%mm7, %%mm1     \n\t"
00214         "pmullw %%mm4, %%mm0        \n\t"
00215         "pmullw %%mm2, %%mm1        \n\t"
00216         "paddw %%mm0, %%mm1         \n\t"
00217         "movq %%mm1, %%mm0          \n\t"
00218         "pmullw %%mm5, %%mm6        \n\t"
00219         "pmullw %%mm3, %%mm1        \n\t"
00220         "paddw %4, %%mm6            \n\t"
00221         "paddw %%mm6, %%mm1         \n\t"
00222         "psrlw $6, %%mm1            \n\t"
00223         "packuswb %%mm1, %%mm1      \n\t"
00224         H264_CHROMA_OP4((%0), %%mm1, %%mm6)
00225         "movd %%mm1, (%0)           \n\t"
00226         "add %3, %0                 \n\t"
00227         "movd  (%1), %%mm6          \n\t"
00228         "movd 1(%1), %%mm1          \n\t"
00229         "add %3, %1                 \n\t"
00230         "punpcklbw %%mm7, %%mm6     \n\t"
00231         "punpcklbw %%mm7, %%mm1     \n\t"
00232         "pmullw %%mm4, %%mm6        \n\t"
00233         "pmullw %%mm2, %%mm1        \n\t"
00234         "paddw %%mm6, %%mm1         \n\t"
00235         "movq %%mm1, %%mm6          \n\t"
00236         "pmullw %%mm5, %%mm0        \n\t"
00237         "pmullw %%mm3, %%mm1        \n\t"
00238         "paddw %4, %%mm0            \n\t"
00239         "paddw %%mm0, %%mm1         \n\t"
00240         "psrlw $6, %%mm1            \n\t"
00241         "packuswb %%mm1, %%mm1      \n\t"
00242         H264_CHROMA_OP4((%0), %%mm1, %%mm0)
00243         "movd %%mm1, (%0)           \n\t"
00244         "add %3, %0                 \n\t"
00245         "sub $2, %2                 \n\t"
00246         "jnz 1b                     \n\t"
00247         : "+r"(dst), "+r"(src), "+r"(h)
00248         : "r"((x86_reg)stride), "m"(*rnd_reg), "m"(x), "m"(y)
00249     );
00250 }
00251 
00252 #ifdef H264_CHROMA_MC2_TMPL
00253 static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
00254 {
00255     int tmp = ((1<<16)-1)*x + 8;
00256     int CD= tmp*y;
00257     int AB= (tmp<<3) - CD;
00258     __asm__ volatile(
00259         /* mm5 = {A,B,A,B} */
00260         /* mm6 = {C,D,C,D} */
00261         "movd %0, %%mm5\n\t"
00262         "movd %1, %%mm6\n\t"
00263         "punpckldq %%mm5, %%mm5\n\t"
00264         "punpckldq %%mm6, %%mm6\n\t"
00265         "pxor %%mm7, %%mm7\n\t"
00266         /* mm0 = src[0,1,1,2] */
00267         "movd %2, %%mm2\n\t"
00268         "punpcklbw %%mm7, %%mm2\n\t"
00269         "pshufw $0x94, %%mm2, %%mm2\n\t"
00270         :: "r"(AB), "r"(CD), "m"(src[0]));
00271 
00272 
00273     __asm__ volatile(
00274         "1:\n\t"
00275         "add %4, %1\n\t"
00276         /* mm1 = A * src[0,1] + B * src[1,2] */
00277         "movq    %%mm2, %%mm1\n\t"
00278         "pmaddwd %%mm5, %%mm1\n\t"
00279         /* mm0 = src[0,1,1,2] */
00280         "movd (%1), %%mm0\n\t"
00281         "punpcklbw %%mm7, %%mm0\n\t"
00282         "pshufw $0x94, %%mm0, %%mm0\n\t"
00283         /* mm1 += C * src[0,1] + D * src[1,2] */
00284         "movq    %%mm0, %%mm2\n\t"
00285         "pmaddwd %%mm6, %%mm0\n\t"
00286         "paddw      %3, %%mm1\n\t"
00287         "paddw   %%mm0, %%mm1\n\t"
00288         /* dst[0,1] = pack((mm1 + 32) >> 6) */
00289         "psrlw $6, %%mm1\n\t"
00290         "packssdw %%mm7, %%mm1\n\t"
00291         "packuswb %%mm7, %%mm1\n\t"
00292         H264_CHROMA_OP4((%0), %%mm1, %%mm3)
00293         "movd %%mm1, %%esi\n\t"
00294         "movw %%si, (%0)\n\t"
00295         "add %4, %0\n\t"
00296         "sub $1, %2\n\t"
00297         "jnz 1b\n\t"
00298         : "+r" (dst), "+r"(src), "+r"(h)
00299         : "m" (ff_pw_32), "r"((x86_reg)stride)
00300         : "%esi");
00301 
00302 }
00303 #endif
00304
libavcodec/x86/dsputil_h264_template_mmx.c
