libavcodec/riscv/vp8dsp_init.c

   1 /*
   2  * Copyright (c) 2024 Institue of Software Chinese Academy of Sciences (ISCAS).
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg 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  * FFmpeg 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 FFmpeg; 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
  23 #include "libavutil/attributes.h"
  24 #include "libavutil/cpu.h"
  25 #include "libavutil/riscv/cpu.h"
  26 #include "libavcodec/vp8dsp.h"
  27 #include "vp8dsp.h"
  28
  29 void ff_vp8_luma_dc_wht_rvv(int16_t block[4][4][16], int16_t dc[16]);
  30 void ff_vp8_idct_dc_add_rvv(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
  31 void ff_vp8_idct_dc_add4y_rvv(uint8_t *dst, int16_t block[4][16], ptrdiff_t stride);
  32 void ff_vp8_idct_dc_add4uv_rvv(uint8_t *dst, int16_t block[4][16], ptrdiff_t stride);
  33
  34 VP8_EPEL(16, rvi);
  35 VP8_EPEL(8,  rvi);
  36 VP8_EPEL(4,  rvi);
  37 VP8_EPEL(16, rvv);
  38 VP8_EPEL(8,  rvv);
  39 VP8_EPEL(4,  rvv);
  40
  41 VP8_BILIN(16, rvv);
  42 VP8_BILIN(8,  rvv);
  43 VP8_BILIN(4,  rvv);
  44
  45 av_cold void ff_vp78dsp_init_riscv(VP8DSPContext *c)
  46 {
  47 #if HAVE_RV
  48     int flags = av_get_cpu_flags();
  49     if (flags & AV_CPU_FLAG_RV_MISALIGNED) {
  50 #if __riscv_xlen >= 64
  51         c->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_rvi;
  52         c->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_rvi;
  53         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_rvi;
  54         c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_rvi;
  55 #endif
  56         c->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_rvi;
  57         c->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_rvi;
  58     }
  59 #if HAVE_RVV
  60     if (flags & AV_CPU_FLAG_RVV_I32 && ff_rv_vlen_least(128)) {
  61         c->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilin16_h_rvv;
  62         c->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_rvv;
  63         c->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_rvv;
  64         c->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_rvv;
  65         c->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_rvv;
  66         c->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_rvv;
  67
  68         c->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilin16_v_rvv;
  69         c->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_rvv;
  70         c->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_rvv;
  71         c->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_rvv;
  72         c->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_rvv;
  73         c->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_rvv;
  74
  75         c->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilin16_hv_rvv;
  76         c->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilin16_hv_rvv;
  77         c->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilin16_hv_rvv;
  78         c->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_rvv;
  79         c->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_rvv;
  80         c->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_rvv;
  81         c->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_rvv;
  82         c->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_rvv;
  83         c->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_rvv;
  84         c->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_rvv;
  85         c->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_rvv;
  86         c->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_rvv;
  87
  88         if (flags & AV_CPU_FLAG_RVB_ADDR) {
  89             c->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_rvv;
  90             c->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_rvv;
  91             c->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_rvv;
  92             c->put_vp8_epel_pixels_tab[0][0][1] = ff_put_vp8_epel16_h4_rvv;
  93             c->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_rvv;
  94             c->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_rvv;
  95
  96             c->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_rvv;
  97             c->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_rvv;
  98             c->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_rvv;
  99             c->put_vp8_epel_pixels_tab[0][1][0] = ff_put_vp8_epel16_v4_rvv;
 100             c->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_rvv;
 101             c->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_rvv;
 102 #if __riscv_xlen <= 64
 103             c->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_rvv;
 104             c->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_rvv;
 105             c->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_rvv;
 106             c->put_vp8_epel_pixels_tab[0][2][1] = ff_put_vp8_epel16_h4v6_rvv;
 107             c->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_rvv;
 108             c->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_rvv;
 109             c->put_vp8_epel_pixels_tab[0][1][1] = ff_put_vp8_epel16_h4v4_rvv;
 110             c->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_rvv;
 111             c->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_rvv;
 112             c->put_vp8_epel_pixels_tab[0][1][2] = ff_put_vp8_epel16_h6v4_rvv;
 113             c->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_rvv;
 114             c->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_rvv;
 115 #endif
 116         }
 117     }
 118 #endif
 119 #endif
 120 }
 121
 122 av_cold void ff_vp8dsp_init_riscv(VP8DSPContext *c)
 123 {
 124 #if HAVE_RVV
 125     int flags = av_get_cpu_flags();
 126
 127     if (flags & AV_CPU_FLAG_RVV_I32 && ff_rv_vlen_least(128)) {
 128 #if __riscv_xlen >= 64
 129         if (flags & AV_CPU_FLAG_RVV_I64)
 130             c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_rvv;
 131 #endif
 132         c->vp8_idct_dc_add = ff_vp8_idct_dc_add_rvv;
 133         c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_rvv;
 134         if (flags & AV_CPU_FLAG_RVB_ADDR) {
 135             c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_rvv;
 136         }
 137     }
 138 #endif
 139 }