libavcodec/i386/dsputil_mmx_qns.h

   1 /*
   2  * DSP utils : QNS functions are compiled 3 times for mmx/3dnow/ssse3
   3  * Copyright (c) 2004 Michael Niedermayer
   4  *
   5  * MMX optimization by Michael Niedermayer <michaelni@gmx.at>
   6  * 3DNow! and SSSE3 optimization by Zuxy Meng <zuxy.meng@gmail.com>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 /* This header intentionally has no multiple inclusion guards. It is meant to
  26  * be included multiple times and generates different code depending on the
  27  * value of certain #defines. */
  28
  29 #define MAX_ABS (512 >> (SCALE_OFFSET>0 ? SCALE_OFFSET : 0))
  30
  31 static int DEF(try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale)
  32 {
  33     x86_reg i=0;
  34
  35     assert(FFABS(scale) < MAX_ABS);
  36     scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT;
  37
  38     SET_RND(mm6);
  39     asm volatile(
  40         "pxor %%mm7, %%mm7              \n\t"
  41         "movd  %4, %%mm5                \n\t"
  42         "punpcklwd %%mm5, %%mm5         \n\t"
  43         "punpcklwd %%mm5, %%mm5         \n\t"
  44         ASMALIGN(4)
  45         "1:                             \n\t"
  46         "movq  (%1, %0), %%mm0          \n\t"
  47         "movq  8(%1, %0), %%mm1         \n\t"
  48         PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6)
  49         "paddw (%2, %0), %%mm0          \n\t"
  50         "paddw 8(%2, %0), %%mm1         \n\t"
  51         "psraw $6, %%mm0                \n\t"
  52         "psraw $6, %%mm1                \n\t"
  53         "pmullw (%3, %0), %%mm0         \n\t"
  54         "pmullw 8(%3, %0), %%mm1        \n\t"
  55         "pmaddwd %%mm0, %%mm0           \n\t"
  56         "pmaddwd %%mm1, %%mm1           \n\t"
  57         "paddd %%mm1, %%mm0             \n\t"
  58         "psrld $4, %%mm0                \n\t"
  59         "paddd %%mm0, %%mm7             \n\t"
  60         "add $16, %0                    \n\t"
  61         "cmp $128, %0                   \n\t" //FIXME optimize & bench
  62         " jb 1b                         \n\t"
  63         PHADDD(%%mm7, %%mm6)
  64         "psrld $2, %%mm7                \n\t"
  65         "movd %%mm7, %0                 \n\t"
  66
  67         : "+r" (i)
  68         : "r"(basis), "r"(rem), "r"(weight), "g"(scale)
  69     );
  70     return i;
  71 }
  72
  73 static void DEF(add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale)
  74 {
  75     x86_reg i=0;
  76
  77     if(FFABS(scale) < MAX_ABS){
  78         scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT;
  79         SET_RND(mm6);
  80         asm volatile(
  81                 "movd  %3, %%mm5        \n\t"
  82                 "punpcklwd %%mm5, %%mm5 \n\t"
  83                 "punpcklwd %%mm5, %%mm5 \n\t"
  84                 ASMALIGN(4)
  85                 "1:                     \n\t"
  86                 "movq  (%1, %0), %%mm0  \n\t"
  87                 "movq  8(%1, %0), %%mm1 \n\t"
  88                 PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6)
  89                 "paddw (%2, %0), %%mm0  \n\t"
  90                 "paddw 8(%2, %0), %%mm1 \n\t"
  91                 "movq %%mm0, (%2, %0)   \n\t"
  92                 "movq %%mm1, 8(%2, %0)  \n\t"
  93                 "add $16, %0            \n\t"
  94                 "cmp $128, %0           \n\t" // FIXME optimize & bench
  95                 " jb 1b                 \n\t"
  96
  97                 : "+r" (i)
  98                 : "r"(basis), "r"(rem), "g"(scale)
  99         );
 100     }else{
 101         for(i=0; i<8*8; i++){
 102             rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);
 103         }
 104     }
 105 }