libavcodec/x86/huffyuvdsp_init.c

   1 /*
   2  * Copyright (c) 2009 Loren Merritt <lorenm@u.washington.edu>
   3  *
   4  * This file is part of Libav.
   5  *
   6  * Libav is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * Libav is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with Libav; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "config.h"
  22 #include "libavutil/attributes.h"
  23 #include "libavutil/cpu.h"
  24 #include "libavutil/x86/asm.h"
  25 #include "libavutil/x86/cpu.h"
  26 #include "libavcodec/huffyuvdsp.h"
  27
  28 void ff_add_hfyu_median_pred_mmxext(uint8_t *dst, const uint8_t *top,
  29                                     const uint8_t *diff, int w,
  30                                     int *left, int *left_top);
  31
  32 int  ff_add_hfyu_left_pred_ssse3(uint8_t *dst, const uint8_t *src,
  33                                  int w, int left);
  34 int  ff_add_hfyu_left_pred_unaligned_ssse3(uint8_t *dst, const uint8_t *src,
  35                                            int w, int left);
  36
  37 #if HAVE_INLINE_ASM
  38
  39 #if HAVE_7REGS
  40 static void add_hfyu_median_pred_cmov(uint8_t *dst, const uint8_t *top,
  41                                       const uint8_t *diff, int w,
  42                                       int *left, int *left_top)
  43 {
  44     x86_reg w2 = -w;
  45     x86_reg x;
  46     int l  = *left     & 0xff;
  47     int tl = *left_top & 0xff;
  48     int t;
  49     __asm__ volatile (
  50         "mov          %7, %3            \n"
  51         "1:                             \n"
  52         "movzbl (%3, %4), %2            \n"
  53         "mov          %2, %k3           \n"
  54         "sub         %b1, %b3           \n"
  55         "add         %b0, %b3           \n"
  56         "mov          %2, %1            \n"
  57         "cmp          %0, %2            \n"
  58         "cmovg        %0, %2            \n"
  59         "cmovg        %1, %0            \n"
  60         "cmp         %k3, %0            \n"
  61         "cmovg       %k3, %0            \n"
  62         "mov          %7, %3            \n"
  63         "cmp          %2, %0            \n"
  64         "cmovl        %2, %0            \n"
  65         "add    (%6, %4), %b0           \n"
  66         "mov         %b0, (%5, %4)      \n"
  67         "inc          %4                \n"
  68         "jl           1b                \n"
  69         : "+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
  70         : "r"(dst + w), "r"(diff + w), "rm"(top + w)
  71     );
  72     *left     = l;
  73     *left_top = tl;
  74 }
  75 #endif /* HAVE_7REGS */
  76
  77 static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w)
  78 {
  79     x86_reg i = 0;
  80
  81     __asm__ volatile (
  82         "jmp          2f                \n\t"
  83         "1:                             \n\t"
  84         "movq   (%1, %0), %%mm0         \n\t"
  85         "movq   (%2, %0), %%mm1         \n\t"
  86         "paddb     %%mm0, %%mm1         \n\t"
  87         "movq      %%mm1, (%2, %0)      \n\t"
  88         "movq  8(%1, %0), %%mm0         \n\t"
  89         "movq  8(%2, %0), %%mm1         \n\t"
  90         "paddb     %%mm0, %%mm1         \n\t"
  91         "movq      %%mm1, 8(%2, %0)     \n\t"
  92         "add         $16, %0            \n\t"
  93         "2:                             \n\t"
  94         "cmp          %3, %0            \n\t"
  95         "js           1b                \n\t"
  96         : "+r" (i)
  97         : "r" (src), "r" (dst), "r" ((x86_reg) w - 15));
  98
  99     for (; i < w; i++)
 100         dst[i + 0] += src[i + 0];
 101 }
 102
 103 #endif /* HAVE_INLINE_ASM */
 104
 105 av_cold void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c)
 106 {
 107     int cpu_flags = av_get_cpu_flags();
 108
 109 #if HAVE_INLINE_ASM
 110 #if HAVE_7REGS
 111     if (cpu_flags & AV_CPU_FLAG_CMOV)
 112         c->add_hfyu_median_pred = add_hfyu_median_pred_cmov;
 113 #endif /* HAVE_7REGS */
 114
 115     if (INLINE_MMX(cpu_flags))
 116         c->add_bytes = add_bytes_mmx;
 117 #endif /* HAVE_INLINE_ASM */
 118
 119     if (EXTERNAL_MMXEXT(cpu_flags)) {
 120         /* slower than cmov version on AMD */
 121         if (!(cpu_flags & AV_CPU_FLAG_3DNOW))
 122             c->add_hfyu_median_pred = ff_add_hfyu_median_pred_mmxext;
 123     }
 124
 125     if (EXTERNAL_SSSE3(cpu_flags)) {
 126         c->add_hfyu_left_pred = ff_add_hfyu_left_pred_ssse3;
 127     }
 128
 129     if (EXTERNAL_SSSE3_FAST(cpu_flags)) {
 130         c->add_hfyu_left_pred = ff_add_hfyu_left_pred_unaligned_ssse3;
 131     }
 132 }