libavcodec/x86/dsputil_h264_template_mmx.c

   1 /*
   2  * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>,
   3  *                    Loren Merritt
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * MMX optimized version of (put|avg)_h264_chroma_mc8.
  24  * H264_CHROMA_MC8_TMPL must be defined to the desired function name
  25  * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg
  26  * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function
  27  */
  28 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)
  29 {
  30     DECLARE_ALIGNED_8(uint64_t, AA);
  31     DECLARE_ALIGNED_8(uint64_t, DD);
  32     int i;
  33
  34     if(y==0 && x==0) {
  35         /* no filter needed */
  36         H264_CHROMA_MC8_MV0(dst, src, stride, h);
  37         return;
  38     }
  39
  40     assert(x<8 && y<8 && x>=0 && y>=0);
  41
  42     if(y==0 || x==0)
  43     {
  44         /* 1 dimensional filter only */
  45         const int dxy = x ? 1 : stride;
  46
  47         __asm__ volatile(
  48             "movd %0, %%mm5\n\t"
  49             "movq %1, %%mm4\n\t"
  50             "movq %2, %%mm6\n\t"         /* mm6 = rnd >> 3 */
  51             "punpcklwd %%mm5, %%mm5\n\t"
  52             "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
  53             "pxor %%mm7, %%mm7\n\t"
  54             "psubw %%mm5, %%mm4\n\t"     /* mm4 = A = 8-x */
  55             :: "rm"(x+y), "m"(ff_pw_8), "m"(*(rnd_reg+1)));
  56
  57         for(i=0; i<h; i++) {
  58             __asm__ volatile(
  59                 /* mm0 = src[0..7], mm1 = src[1..8] */
  60                 "movq %0, %%mm0\n\t"
  61                 "movq %1, %%mm2\n\t"
  62                 :: "m"(src[0]), "m"(src[dxy]));
  63
  64             __asm__ volatile(
  65                 /* [mm0,mm1] = A * src[0..7] */
  66                 /* [mm2,mm3] = B * src[1..8] */
  67                 "movq %%mm0, %%mm1\n\t"
  68                 "movq %%mm2, %%mm3\n\t"
  69                 "punpcklbw %%mm7, %%mm0\n\t"
  70                 "punpckhbw %%mm7, %%mm1\n\t"
  71                 "punpcklbw %%mm7, %%mm2\n\t"
  72                 "punpckhbw %%mm7, %%mm3\n\t"
  73                 "pmullw %%mm4, %%mm0\n\t"
  74                 "pmullw %%mm4, %%mm1\n\t"
  75                 "pmullw %%mm5, %%mm2\n\t"
  76                 "pmullw %%mm5, %%mm3\n\t"
  77
  78                 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3 */
  79                 "paddw %%mm6, %%mm0\n\t"
  80                 "paddw %%mm6, %%mm1\n\t"
  81                 "paddw %%mm2, %%mm0\n\t"
  82                 "paddw %%mm3, %%mm1\n\t"
  83                 "psrlw $3, %%mm0\n\t"
  84                 "psrlw $3, %%mm1\n\t"
  85                 "packuswb %%mm1, %%mm0\n\t"
  86                 H264_CHROMA_OP(%0, %%mm0)
  87                 "movq %%mm0, %0\n\t"
  88                 : "=m" (dst[0]));
  89
  90             src += stride;
  91             dst += stride;
  92         }
  93         return;
  94     }
  95
  96     /* general case, bilinear */
  97     __asm__ volatile("movd %2, %%mm4\n\t"
  98                  "movd %3, %%mm6\n\t"
  99                  "punpcklwd %%mm4, %%mm4\n\t"
 100                  "punpcklwd %%mm6, %%mm6\n\t"
 101                  "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
 102                  "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */
 103                  "movq %%mm4, %%mm5\n\t"
 104                  "pmullw %%mm6, %%mm4\n\t"    /* mm4 = x * y */
 105                  "psllw $3, %%mm5\n\t"
 106                  "psllw $3, %%mm6\n\t"
 107                  "movq %%mm5, %%mm7\n\t"
 108                  "paddw %%mm6, %%mm7\n\t"
 109                  "movq %%mm4, %1\n\t"         /* DD = x * y */
 110                  "psubw %%mm4, %%mm5\n\t"     /* mm5 = B = 8x - xy */
 111                  "psubw %%mm4, %%mm6\n\t"     /* mm6 = C = 8y - xy */
 112                  "paddw %4, %%mm4\n\t"
 113                  "psubw %%mm7, %%mm4\n\t"     /* mm4 = A = xy - (8x+8y) + 64 */
 114                  "pxor %%mm7, %%mm7\n\t"
 115                  "movq %%mm4, %0\n\t"
 116                  : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64));
 117
 118     __asm__ volatile(
 119         /* mm0 = src[0..7], mm1 = src[1..8] */
 120         "movq %0, %%mm0\n\t"
 121         "movq %1, %%mm1\n\t"
 122         : : "m" (src[0]), "m" (src[1]));
 123
 124     for(i=0; i<h; i++) {
 125         src += stride;
 126
 127         __asm__ volatile(
 128             /* mm2 = A * src[0..3] + B * src[1..4] */
 129             /* mm3 = A * src[4..7] + B * src[5..8] */
 130             "movq %%mm0, %%mm2\n\t"
 131             "movq %%mm1, %%mm3\n\t"
 132             "punpckhbw %%mm7, %%mm0\n\t"
 133             "punpcklbw %%mm7, %%mm1\n\t"
 134             "punpcklbw %%mm7, %%mm2\n\t"
 135             "punpckhbw %%mm7, %%mm3\n\t"
 136             "pmullw %0, %%mm0\n\t"
 137             "pmullw %0, %%mm2\n\t"
 138             "pmullw %%mm5, %%mm1\n\t"
 139             "pmullw %%mm5, %%mm3\n\t"
 140             "paddw %%mm1, %%mm2\n\t"
 141             "paddw %%mm0, %%mm3\n\t"
 142             : : "m" (AA));
 143
 144         __asm__ volatile(
 145             /* [mm2,mm3] += C * src[0..7] */
 146             "movq %0, %%mm0\n\t"
 147             "movq %%mm0, %%mm1\n\t"
 148             "punpcklbw %%mm7, %%mm0\n\t"
 149             "punpckhbw %%mm7, %%mm1\n\t"
 150             "pmullw %%mm6, %%mm0\n\t"
 151             "pmullw %%mm6, %%mm1\n\t"
 152             "paddw %%mm0, %%mm2\n\t"
 153             "paddw %%mm1, %%mm3\n\t"
 154             : : "m" (src[0]));
 155
 156         __asm__ volatile(
 157             /* [mm2,mm3] += D * src[1..8] */
 158             "movq %1, %%mm1\n\t"
 159             "movq %%mm1, %%mm0\n\t"
 160             "movq %%mm1, %%mm4\n\t"
 161             "punpcklbw %%mm7, %%mm0\n\t"
 162             "punpckhbw %%mm7, %%mm4\n\t"
 163             "pmullw %2, %%mm0\n\t"
 164             "pmullw %2, %%mm4\n\t"
 165             "paddw %%mm0, %%mm2\n\t"
 166             "paddw %%mm4, %%mm3\n\t"
 167             "movq %0, %%mm0\n\t"
 168             : : "m" (src[0]), "m" (src[1]), "m" (DD));
 169
 170         __asm__ volatile(
 171             /* dst[0..7] = ([mm2,mm3] + rnd) >> 6 */
 172             "paddw %1, %%mm2\n\t"
 173             "paddw %1, %%mm3\n\t"
 174             "psrlw $6, %%mm2\n\t"
 175             "psrlw $6, %%mm3\n\t"
 176             "packuswb %%mm3, %%mm2\n\t"
 177             H264_CHROMA_OP(%0, %%mm2)
 178             "movq %%mm2, %0\n\t"
 179             : "=m" (dst[0]) : "m" (*rnd_reg));
 180         dst+= stride;
 181     }
 182 }
 183
 184 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)
 185 {
 186     __asm__ volatile(
 187         "pxor   %%mm7, %%mm7        \n\t"
 188         "movd %5, %%mm2             \n\t"
 189         "movd %6, %%mm3             \n\t"
 190         "movq "MANGLE(ff_pw_8)", %%mm4\n\t"
 191         "movq "MANGLE(ff_pw_8)", %%mm5\n\t"
 192         "punpcklwd %%mm2, %%mm2     \n\t"
 193         "punpcklwd %%mm3, %%mm3     \n\t"
 194         "punpcklwd %%mm2, %%mm2     \n\t"
 195         "punpcklwd %%mm3, %%mm3     \n\t"
 196         "psubw %%mm2, %%mm4         \n\t"
 197         "psubw %%mm3, %%mm5         \n\t"
 198
 199         "movd  (%1), %%mm0          \n\t"
 200         "movd 1(%1), %%mm6          \n\t"
 201         "add %3, %1                 \n\t"
 202         "punpcklbw %%mm7, %%mm0     \n\t"
 203         "punpcklbw %%mm7, %%mm6     \n\t"
 204         "pmullw %%mm4, %%mm0        \n\t"
 205         "pmullw %%mm2, %%mm6        \n\t"
 206         "paddw %%mm0, %%mm6         \n\t"
 207
 208         "1:                         \n\t"
 209         "movd  (%1), %%mm0          \n\t"
 210         "movd 1(%1), %%mm1          \n\t"
 211         "add %3, %1                 \n\t"
 212         "punpcklbw %%mm7, %%mm0     \n\t"
 213         "punpcklbw %%mm7, %%mm1     \n\t"
 214         "pmullw %%mm4, %%mm0        \n\t"
 215         "pmullw %%mm2, %%mm1        \n\t"
 216         "paddw %%mm0, %%mm1         \n\t"
 217         "movq %%mm1, %%mm0          \n\t"
 218         "pmullw %%mm5, %%mm6        \n\t"
 219         "pmullw %%mm3, %%mm1        \n\t"
 220         "paddw %4, %%mm6            \n\t"
 221         "paddw %%mm6, %%mm1         \n\t"
 222         "psrlw $6, %%mm1            \n\t"
 223         "packuswb %%mm1, %%mm1      \n\t"
 224         H264_CHROMA_OP4((%0), %%mm1, %%mm6)
 225         "movd %%mm1, (%0)           \n\t"
 226         "add %3, %0                 \n\t"
 227         "movd  (%1), %%mm6          \n\t"
 228         "movd 1(%1), %%mm1          \n\t"
 229         "add %3, %1                 \n\t"
 230         "punpcklbw %%mm7, %%mm6     \n\t"
 231         "punpcklbw %%mm7, %%mm1     \n\t"
 232         "pmullw %%mm4, %%mm6        \n\t"
 233         "pmullw %%mm2, %%mm1        \n\t"
 234         "paddw %%mm6, %%mm1         \n\t"
 235         "movq %%mm1, %%mm6          \n\t"
 236         "pmullw %%mm5, %%mm0        \n\t"
 237         "pmullw %%mm3, %%mm1        \n\t"
 238         "paddw %4, %%mm0            \n\t"
 239         "paddw %%mm0, %%mm1         \n\t"
 240         "psrlw $6, %%mm1            \n\t"
 241         "packuswb %%mm1, %%mm1      \n\t"
 242         H264_CHROMA_OP4((%0), %%mm1, %%mm0)
 243         "movd %%mm1, (%0)           \n\t"
 244         "add %3, %0                 \n\t"
 245         "sub $2, %2                 \n\t"
 246         "jnz 1b                     \n\t"
 247         : "+r"(dst), "+r"(src), "+r"(h)
 248         : "r"((x86_reg)stride), "m"(*rnd_reg), "m"(x), "m"(y)
 249     );
 250 }
 251
 252 #ifdef H264_CHROMA_MC2_TMPL
 253 static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
 254 {
 255     int tmp = ((1<<16)-1)*x + 8;
 256     int CD= tmp*y;
 257     int AB= (tmp<<3) - CD;
 258     __asm__ volatile(
 259         /* mm5 = {A,B,A,B} */
 260         /* mm6 = {C,D,C,D} */
 261         "movd %0, %%mm5\n\t"
 262         "movd %1, %%mm6\n\t"
 263         "punpckldq %%mm5, %%mm5\n\t"
 264         "punpckldq %%mm6, %%mm6\n\t"
 265         "pxor %%mm7, %%mm7\n\t"
 266         /* mm0 = src[0,1,1,2] */
 267         "movd %2, %%mm2\n\t"
 268         "punpcklbw %%mm7, %%mm2\n\t"
 269         "pshufw $0x94, %%mm2, %%mm2\n\t"
 270         :: "r"(AB), "r"(CD), "m"(src[0]));
 271
 272
 273     __asm__ volatile(
 274         "1:\n\t"
 275         "add %4, %1\n\t"
 276         /* mm1 = A * src[0,1] + B * src[1,2] */
 277         "movq    %%mm2, %%mm1\n\t"
 278         "pmaddwd %%mm5, %%mm1\n\t"
 279         /* mm0 = src[0,1,1,2] */
 280         "movd (%1), %%mm0\n\t"
 281         "punpcklbw %%mm7, %%mm0\n\t"
 282         "pshufw $0x94, %%mm0, %%mm0\n\t"
 283         /* mm1 += C * src[0,1] + D * src[1,2] */
 284         "movq    %%mm0, %%mm2\n\t"
 285         "pmaddwd %%mm6, %%mm0\n\t"
 286         "paddw      %3, %%mm1\n\t"
 287         "paddw   %%mm0, %%mm1\n\t"
 288         /* dst[0,1] = pack((mm1 + 32) >> 6) */
 289         "psrlw $6, %%mm1\n\t"
 290         "packssdw %%mm7, %%mm1\n\t"
 291         "packuswb %%mm7, %%mm1\n\t"
 292         H264_CHROMA_OP4((%0), %%mm1, %%mm3)
 293         "movd %%mm1, %%esi\n\t"
 294         "movw %%si, (%0)\n\t"
 295         "add %4, %0\n\t"
 296         "sub $1, %2\n\t"
 297         "jnz 1b\n\t"
 298         : "+r" (dst), "+r"(src), "+r"(h)
 299         : "m" (ff_pw_32), "r"((x86_reg)stride)
 300         : "%esi");
 301
 302 }
 303 #endif
 304