libavcodec/x86/hpeldsp_init.c

   1 /*
   2  * SIMD-optimized halfpel functions
   3  * Copyright (c) 2000, 2001 Fabrice Bellard
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * This file is part of Libav.
   7  *
   8  * Libav is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * Libav is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with Libav; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  *
  22  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
  23  */
  24
  25 #include "libavutil/attributes.h"
  26 #include "libavutil/cpu.h"
  27 #include "libavutil/x86/cpu.h"
  28 #include "libavcodec/avcodec.h"
  29 #include "libavcodec/hpeldsp.h"
  30 #include "libavcodec/pixels.h"
  31 #include "fpel.h"
  32 #include "hpeldsp.h"
  33
  34 void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
  35                               ptrdiff_t line_size, int h);
  36 void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
  37                              ptrdiff_t line_size, int h);
  38 void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,
  39                                ptrdiff_t line_size, int h);
  40 void ff_put_pixels16_x2_3dnow(uint8_t *block, const uint8_t *pixels,
  41                               ptrdiff_t line_size, int h);
  42 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
  43                                      ptrdiff_t line_size, int h);
  44 void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
  45                                     ptrdiff_t line_size, int h);
  46 void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
  47                               ptrdiff_t line_size, int h);
  48 void ff_put_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
  49                              ptrdiff_t line_size, int h);
  50 void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
  51                                      ptrdiff_t line_size, int h);
  52 void ff_put_no_rnd_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
  53                                     ptrdiff_t line_size, int h);
  54 void ff_avg_pixels8_3dnow(uint8_t *block, const uint8_t *pixels,
  55                           ptrdiff_t line_size, int h);
  56 void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
  57                               ptrdiff_t line_size, int h);
  58 void ff_avg_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
  59                              ptrdiff_t line_size, int h);
  60 void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
  61                               ptrdiff_t line_size, int h);
  62 void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
  63                              ptrdiff_t line_size, int h);
  64 void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
  65                                ptrdiff_t line_size, int h);
  66 void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
  67                               ptrdiff_t line_size, int h);
  68
  69 #define avg_pixels8_mmx         ff_avg_pixels8_mmx
  70 #define avg_pixels8_x2_mmx      ff_avg_pixels8_x2_mmx
  71 #define avg_pixels16_mmx        ff_avg_pixels16_mmx
  72 #define avg_pixels8_xy2_mmx     ff_avg_pixels8_xy2_mmx
  73 #define avg_pixels16_xy2_mmx    ff_avg_pixels16_xy2_mmx
  74 #define put_pixels8_mmx         ff_put_pixels8_mmx
  75 #define put_pixels16_mmx        ff_put_pixels16_mmx
  76 #define put_pixels8_xy2_mmx     ff_put_pixels8_xy2_mmx
  77 #define put_pixels16_xy2_mmx    ff_put_pixels16_xy2_mmx
  78 #define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx
  79 #define put_no_rnd_pixels8_mmx  ff_put_pixels8_mmx
  80 #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx
  81
  82 #if HAVE_INLINE_ASM
  83
  84 /***********************************/
  85 /* MMX no rounding */
  86 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
  87 #define SET_RND  MOVQ_WONE
  88 #define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
  89 #define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
  90 #define STATIC static
  91
  92 #include "rnd_template.c"
  93 #include "hpeldsp_rnd_template.c"
  94
  95 #undef DEF
  96 #undef SET_RND
  97 #undef PAVGBP
  98 #undef PAVGB
  99 #undef STATIC
 100
 101 CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8)
 102 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8)
 103
 104 CALL_2X_PIXELS(avg_no_rnd_pixels16_xy2_mmx, avg_no_rnd_pixels8_xy2_mmx, 8)
 105 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8)
 106
 107 /***********************************/
 108 /* MMX rounding */
 109
 110 #define DEF(x, y) x ## _ ## y ## _mmx
 111 #define SET_RND  MOVQ_WTWO
 112 #define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
 113 #define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)
 114
 115 #include "hpeldsp_rnd_template.c"
 116
 117 #undef DEF
 118 #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx
 119 #define STATIC
 120
 121 #include "rnd_template.c"
 122
 123 #undef DEF
 124 #undef SET_RND
 125 #undef PAVGBP
 126 #undef PAVGB
 127
 128 CALL_2X_PIXELS(avg_pixels16_y2_mmx, avg_pixels8_y2_mmx, 8)
 129 CALL_2X_PIXELS(put_pixels16_y2_mmx, put_pixels8_y2_mmx, 8)
 130
 131 CALL_2X_PIXELS_EXPORT(ff_avg_pixels16_xy2_mmx, ff_avg_pixels8_xy2_mmx, 8)
 132 CALL_2X_PIXELS_EXPORT(ff_put_pixels16_xy2_mmx, ff_put_pixels8_xy2_mmx, 8)
 133
 134 #endif /* HAVE_INLINE_ASM */
 135
 136
 137 #if HAVE_X86ASM
 138
 139 #define HPELDSP_AVG_PIXELS16(CPUEXT)                      \
 140     CALL_2X_PIXELS(put_no_rnd_pixels16_x2 ## CPUEXT, ff_put_no_rnd_pixels8_x2 ## CPUEXT, 8) \
 141     CALL_2X_PIXELS(put_pixels16_y2        ## CPUEXT, ff_put_pixels8_y2        ## CPUEXT, 8) \
 142     CALL_2X_PIXELS(put_no_rnd_pixels16_y2 ## CPUEXT, ff_put_no_rnd_pixels8_y2 ## CPUEXT, 8) \
 143     CALL_2X_PIXELS(avg_pixels16           ## CPUEXT, ff_avg_pixels8           ## CPUEXT, 8) \
 144     CALL_2X_PIXELS(avg_pixels16_x2        ## CPUEXT, ff_avg_pixels8_x2        ## CPUEXT, 8) \
 145     CALL_2X_PIXELS(avg_pixels16_y2        ## CPUEXT, ff_avg_pixels8_y2        ## CPUEXT, 8) \
 146     CALL_2X_PIXELS(avg_pixels16_xy2       ## CPUEXT, ff_avg_pixels8_xy2       ## CPUEXT, 8)
 147
 148 HPELDSP_AVG_PIXELS16(_3dnow)
 149 HPELDSP_AVG_PIXELS16(_mmxext)
 150
 151 #endif /* HAVE_X86ASM */
 152
 153 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU)                                     \
 154     do {                                                                        \
 155         c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _     ## CPU; \
 156         c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_  ## CPU; \
 157         c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_  ## CPU; \
 158         c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
 159     } while (0)
 160
 161 static void hpeldsp_init_mmx(HpelDSPContext *c, int flags)
 162 {
 163 #if HAVE_MMX_INLINE
 164     SET_HPEL_FUNCS(put,        [0], 16, mmx);
 165     SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);
 166     SET_HPEL_FUNCS(avg,        [0], 16, mmx);
 167     SET_HPEL_FUNCS(avg_no_rnd,    , 16, mmx);
 168     SET_HPEL_FUNCS(put,        [1],  8, mmx);
 169     SET_HPEL_FUNCS(put_no_rnd, [1],  8, mmx);
 170     SET_HPEL_FUNCS(avg,        [1],  8, mmx);
 171 #endif /* HAVE_MMX_INLINE */
 172 }
 173
 174 static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags)
 175 {
 176 #if HAVE_MMXEXT_EXTERNAL
 177     c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;
 178     c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;
 179
 180     c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;
 181     c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
 182     c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
 183
 184     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
 185     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
 186
 187     c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
 188     c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
 189     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
 190
 191     if (!(flags & AV_CODEC_FLAG_BITEXACT)) {
 192         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;
 193         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmxext;
 194         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;
 195         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;
 196
 197         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmxext;
 198         c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
 199     }
 200 #endif /* HAVE_MMXEXT_EXTERNAL */
 201 }
 202
 203 static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags)
 204 {
 205 #if HAVE_AMD3DNOW_EXTERNAL
 206     c->put_pixels_tab[0][1] = ff_put_pixels16_x2_3dnow;
 207     c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
 208
 209     c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
 210     c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
 211     c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
 212
 213     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow;
 214     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow;
 215
 216     c->avg_pixels_tab[1][0] = ff_avg_pixels8_3dnow;
 217     c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_3dnow;
 218     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow;
 219
 220     if (!(flags & AV_CODEC_FLAG_BITEXACT)){
 221         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
 222         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
 223         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_3dnow;
 224         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_3dnow;
 225
 226         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
 227         c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow;
 228     }
 229 #endif /* HAVE_AMD3DNOW_EXTERNAL */
 230 }
 231
 232 static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags)
 233 {
 234 #if HAVE_SSE2_EXTERNAL
 235     c->put_pixels_tab[0][0]        = ff_put_pixels16_sse2;
 236     c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
 237     c->avg_pixels_tab[0][0]        = ff_avg_pixels16_sse2;
 238 #endif /* HAVE_SSE2_EXTERNAL */
 239 }
 240
 241 av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
 242 {
 243     int cpu_flags = av_get_cpu_flags();
 244
 245     if (INLINE_MMX(cpu_flags))
 246         hpeldsp_init_mmx(c, flags);
 247
 248     if (EXTERNAL_AMD3DNOW(cpu_flags))
 249         hpeldsp_init_3dnow(c, flags);
 250
 251     if (EXTERNAL_MMXEXT(cpu_flags))
 252         hpeldsp_init_mmxext(c, flags);
 253
 254     if (EXTERNAL_SSE2_FAST(cpu_flags))
 255         hpeldsp_init_sse2_fast(c, flags);
 256
 257     if (CONFIG_VP3_DECODER)
 258         ff_hpeldsp_vp3_init_x86(c, cpu_flags);
 259 }