Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * Loongson SIMD optimized pixblockdsp
  3.  *
  4.  * Copyright (c) 2015 Loongson Technology Corporation Limited
  5.  * Copyright (c) 2015 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
  6.  *
  7.  * This file is part of FFmpeg.
  8.  *
  9.  * FFmpeg is free software; you can redistribute it and/or
  10.  * modify it under the terms of the GNU Lesser General Public
  11.  * License as published by the Free Software Foundation; either
  12.  * version 2.1 of the License, or (at your option) any later version.
  13.  *
  14.  * FFmpeg is distributed in the hope that it will be useful,
  15.  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17.  * Lesser General Public License for more details.
  18.  *
  19.  * You should have received a copy of the GNU Lesser General Public
  20.  * License along with FFmpeg; if not, write to the Free Software
  21.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22.  */
  23.  
  24. #include "pixblockdsp_mips.h"
  25.  
  26. void ff_get_pixels_8_mmi(int16_t *av_restrict block, const uint8_t *pixels,
  27.         ptrdiff_t line_size)
  28. {
  29.     __asm__ volatile (
  30.         "move $8, $0                    \n\t"
  31.         "xor $f0, $f0, $f0              \n\t"
  32.         "1:                             \n\t"
  33.         "gsldlc1 $f2, 7(%1)             \n\t"
  34.         "gsldrc1 $f2, 0(%1)             \n\t"
  35.         "punpcklbh $f4, $f2, $f0        \n\t"
  36.         "punpckhbh $f6, $f2, $f0        \n\t"
  37.         "gssdxc1 $f4, 0(%0, $8)         \n\t"
  38.         "gssdxc1 $f6, 8(%0, $8)         \n\t"
  39.         "daddiu $8, $8, 16              \n\t"
  40.         "daddu %1, %1, %2               \n\t"
  41.         "daddi %3, %3, -1               \n\t"
  42.         "bnez %3, 1b                    \n\t"
  43.         ::"r"((uint8_t *)block),"r"(pixels),"r"(line_size),"r"(8)
  44.         : "$8","memory"
  45.     );
  46. }
  47.  
  48. void ff_diff_pixels_mmi(int16_t *av_restrict block, const uint8_t *src1,
  49.         const uint8_t *src2, int stride)
  50. {
  51.     __asm__ volatile (
  52.         "dli $2, 8                     \n\t"
  53.         "xor $f14, $f14, $f14          \n\t"
  54.         "1:                            \n\t"
  55.         "gsldlc1 $f0, 7(%1)            \n\t"
  56.         "gsldrc1 $f0, 0(%1)            \n\t"
  57.         "or $f2, $f0, $f0              \n\t"
  58.         "gsldlc1 $f4, 7(%2)            \n\t"
  59.         "gsldrc1 $f4, 0(%2)            \n\t"
  60.         "or $f6, $f4, $f4              \n\t"
  61.         "punpcklbh $f0, $f0, $f14      \n\t"
  62.         "punpckhbh $f2, $f2, $f14      \n\t"
  63.         "punpcklbh $f4, $f4, $f14      \n\t"
  64.         "punpckhbh $f6, $f6, $f14      \n\t"
  65.         "psubh $f0, $f0, $f4           \n\t"
  66.         "psubh $f2, $f2, $f6           \n\t"
  67.         "gssdlc1 $f0, 7(%0)            \n\t"
  68.         "gssdrc1 $f0, 0(%0)            \n\t"
  69.         "gssdlc1 $f2, 15(%0)           \n\t"
  70.         "gssdrc1 $f2, 8(%0)            \n\t"
  71.         "daddi %0, %0, 16              \n\t"
  72.         "daddu %1, %1, %3              \n\t"
  73.         "daddu %2, %2, %3              \n\t"
  74.         "daddi $2, $2, -1              \n\t"
  75.         "bgtz $2, 1b                   \n\t"
  76.         ::"r"(block),"r"(src1),"r"(src2),"r"(stride)
  77.         : "$2","memory"
  78.     );
  79. }
  80.