libavcodec/arm/dsputil_arm.c

   1 /*
   2  * ARM optimized DSP utils
   3  * Copyright (c) 2001 Lionel Ulmer
   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 #include "libavcodec/dsputil.h"
  23 #if HAVE_IPP
  24 #include <ipp.h>
  25 #endif
  26
  27 void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx);
  28 void ff_float_init_arm_vfp(DSPContext* c, AVCodecContext *avctx);
  29 void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx);
  30
  31 void j_rev_dct_ARM(DCTELEM *data);
  32 void simple_idct_ARM(DCTELEM *data);
  33
  34 void simple_idct_armv5te(DCTELEM *data);
  35 void simple_idct_put_armv5te(uint8_t *dest, int line_size, DCTELEM *data);
  36 void simple_idct_add_armv5te(uint8_t *dest, int line_size, DCTELEM *data);
  37
  38 void ff_simple_idct_armv6(DCTELEM *data);
  39 void ff_simple_idct_put_armv6(uint8_t *dest, int line_size, DCTELEM *data);
  40 void ff_simple_idct_add_armv6(uint8_t *dest, int line_size, DCTELEM *data);
  41
  42 void ff_simple_idct_neon(DCTELEM *data);
  43 void ff_simple_idct_put_neon(uint8_t *dest, int line_size, DCTELEM *data);
  44 void ff_simple_idct_add_neon(uint8_t *dest, int line_size, DCTELEM *data);
  45
  46 void ff_vp3_idct_neon(DCTELEM *data);
  47 void ff_vp3_idct_put_neon(uint8_t *dest, int line_size, DCTELEM *data);
  48 void ff_vp3_idct_add_neon(uint8_t *dest, int line_size, DCTELEM *data);
  49
  50 /* XXX: local hack */
  51 static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
  52 static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
  53
  54 void put_pixels8_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  55 void put_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  56 void put_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  57 void put_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  58
  59 void put_no_rnd_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  60 void put_no_rnd_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  61 void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  62
  63 void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  64
  65 void ff_prefetch_arm(void *mem, int stride, int h);
  66
  67 CALL_2X_PIXELS(put_pixels16_x2_arm , put_pixels8_x2_arm , 8)
  68 CALL_2X_PIXELS(put_pixels16_y2_arm , put_pixels8_y2_arm , 8)
  69 CALL_2X_PIXELS(put_pixels16_xy2_arm, put_pixels8_xy2_arm, 8)
  70 CALL_2X_PIXELS(put_no_rnd_pixels16_x2_arm , put_no_rnd_pixels8_x2_arm , 8)
  71 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_arm , put_no_rnd_pixels8_y2_arm , 8)
  72 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_arm, put_no_rnd_pixels8_xy2_arm, 8)
  73
  74 void ff_add_pixels_clamped_ARM(short *block, unsigned char *dest,
  75                                       int line_size);
  76
  77 /* XXX: those functions should be suppressed ASAP when all IDCTs are
  78    converted */
  79 static void j_rev_dct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
  80 {
  81     j_rev_dct_ARM (block);
  82     ff_put_pixels_clamped(block, dest, line_size);
  83 }
  84 static void j_rev_dct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
  85 {
  86     j_rev_dct_ARM (block);
  87     ff_add_pixels_clamped(block, dest, line_size);
  88 }
  89 static void simple_idct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
  90 {
  91     simple_idct_ARM (block);
  92     ff_put_pixels_clamped(block, dest, line_size);
  93 }
  94 static void simple_idct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
  95 {
  96     simple_idct_ARM (block);
  97     ff_add_pixels_clamped(block, dest, line_size);
  98 }
  99
 100 #if HAVE_IPP
 101 static void simple_idct_ipp(DCTELEM *block)
 102 {
 103     ippiDCT8x8Inv_Video_16s_C1I(block);
 104 }
 105 static void simple_idct_ipp_put(uint8_t *dest, int line_size, DCTELEM *block)
 106 {
 107     ippiDCT8x8Inv_Video_16s8u_C1R(block, dest, line_size);
 108 }
 109
 110 void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size);
 111
 112 static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
 113 {
 114     ippiDCT8x8Inv_Video_16s_C1I(block);
 115 #if HAVE_IWMMXT
 116     add_pixels_clamped_iwmmxt(block, dest, line_size);
 117 #else
 118     ff_add_pixels_clamped_ARM(block, dest, line_size);
 119 #endif
 120 }
 121 #endif
 122
 123 int mm_support(void)
 124 {
 125     return HAVE_IWMMXT * FF_MM_IWMMXT;
 126 }
 127
 128 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx)
 129 {
 130     int idct_algo= avctx->idct_algo;
 131
 132     ff_put_pixels_clamped = c->put_pixels_clamped;
 133     ff_add_pixels_clamped = c->add_pixels_clamped;
 134
 135     if (avctx->lowres == 0) {
 136         if(idct_algo == FF_IDCT_AUTO){
 137 #if   HAVE_IPP
 138             idct_algo = FF_IDCT_IPP;
 139 #elif HAVE_NEON
 140             idct_algo = FF_IDCT_SIMPLENEON;
 141 #elif HAVE_ARMV6
 142             idct_algo = FF_IDCT_SIMPLEARMV6;
 143 #elif HAVE_ARMV5TE
 144             idct_algo = FF_IDCT_SIMPLEARMV5TE;
 145 #else
 146             idct_algo = FF_IDCT_ARM;
 147 #endif
 148         }
 149
 150         if(idct_algo==FF_IDCT_ARM){
 151             c->idct_put= j_rev_dct_ARM_put;
 152             c->idct_add= j_rev_dct_ARM_add;
 153             c->idct    = j_rev_dct_ARM;
 154             c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
 155         } else if (idct_algo==FF_IDCT_SIMPLEARM){
 156             c->idct_put= simple_idct_ARM_put;
 157             c->idct_add= simple_idct_ARM_add;
 158             c->idct    = simple_idct_ARM;
 159             c->idct_permutation_type= FF_NO_IDCT_PERM;
 160 #if HAVE_ARMV6
 161         } else if (idct_algo==FF_IDCT_SIMPLEARMV6){
 162             c->idct_put= ff_simple_idct_put_armv6;
 163             c->idct_add= ff_simple_idct_add_armv6;
 164             c->idct    = ff_simple_idct_armv6;
 165             c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
 166 #endif
 167 #if HAVE_ARMV5TE
 168         } else if (idct_algo==FF_IDCT_SIMPLEARMV5TE){
 169             c->idct_put= simple_idct_put_armv5te;
 170             c->idct_add= simple_idct_add_armv5te;
 171             c->idct    = simple_idct_armv5te;
 172             c->idct_permutation_type = FF_NO_IDCT_PERM;
 173 #endif
 174 #if HAVE_IPP
 175         } else if (idct_algo==FF_IDCT_IPP){
 176             c->idct_put= simple_idct_ipp_put;
 177             c->idct_add= simple_idct_ipp_add;
 178             c->idct    = simple_idct_ipp;
 179             c->idct_permutation_type= FF_NO_IDCT_PERM;
 180 #endif
 181 #if HAVE_NEON
 182         } else if (idct_algo==FF_IDCT_SIMPLENEON){
 183             c->idct_put= ff_simple_idct_put_neon;
 184             c->idct_add= ff_simple_idct_add_neon;
 185             c->idct    = ff_simple_idct_neon;
 186             c->idct_permutation_type = FF_PARTTRANS_IDCT_PERM;
 187         } else if ((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&
 188                    idct_algo==FF_IDCT_VP3){
 189             c->idct_put= ff_vp3_idct_put_neon;
 190             c->idct_add= ff_vp3_idct_add_neon;
 191             c->idct    = ff_vp3_idct_neon;
 192             c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
 193 #endif
 194         }
 195     }
 196
 197     c->put_pixels_tab[0][0] = put_pixels16_arm;
 198     c->put_pixels_tab[0][1] = put_pixels16_x2_arm;
 199     c->put_pixels_tab[0][2] = put_pixels16_y2_arm;
 200     c->put_pixels_tab[0][3] = put_pixels16_xy2_arm;
 201     c->put_no_rnd_pixels_tab[0][0] = put_pixels16_arm;
 202     c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_arm;
 203     c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_arm;
 204     c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_arm;
 205     c->put_pixels_tab[1][0] = put_pixels8_arm;
 206     c->put_pixels_tab[1][1] = put_pixels8_x2_arm;
 207     c->put_pixels_tab[1][2] = put_pixels8_y2_arm;
 208     c->put_pixels_tab[1][3] = put_pixels8_xy2_arm;
 209     c->put_no_rnd_pixels_tab[1][0] = put_pixels8_arm;
 210     c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_arm;
 211     c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_arm;
 212     c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;
 213
 214 #if HAVE_ARMV5TE
 215     c->prefetch = ff_prefetch_arm;
 216 #endif
 217
 218 #if HAVE_IWMMXT
 219     dsputil_init_iwmmxt(c, avctx);
 220 #endif
 221 #if HAVE_ARMVFP
 222     ff_float_init_arm_vfp(c, avctx);
 223 #endif
 224 #if HAVE_NEON
 225     ff_dsputil_init_neon(c, avctx);
 226 #endif
 227 }