1/* 2 * Copyright (C) 2009 David Conrad 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#include "libavutil/attributes.h" 23#include "libavutil/common.h" 24#include "diracdsp.h" 25 26#define FILTER(src, stride) \ 27 ((21*((src)[ 0*stride] + (src)[1*stride]) \ 28 -7*((src)[-1*stride] + (src)[2*stride]) \ 29 +3*((src)[-2*stride] + (src)[3*stride]) \ 30 -1*((src)[-3*stride] + (src)[4*stride]) + 16) >> 5) 31 32static void dirac_hpel_filter(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, const uint8_t *src, 33 int stride, int width, int height) 34{ 35 int x, y; 36 37 for (y = 0; y < height; y++) { 38 for (x = -3; x < width+5; x++) 39 dstv[x] = av_clip_uint8(FILTER(src+x, stride)); 40 41 for (x = 0; x < width; x++) 42 dstc[x] = av_clip_uint8(FILTER(dstv+x, 1)); 43 44 for (x = 0; x < width; x++) 45 dsth[x] = av_clip_uint8(FILTER(src+x, 1)); 46 47 src += stride; 48 dsth += stride; 49 dstv += stride; 50 dstc += stride; 51 } 52} 53 54#define PIXOP_BILINEAR(PFX, OP, WIDTH) \ 55 static void ff_ ## PFX ## _dirac_pixels ## WIDTH ## _bilinear_c(uint8_t *dst, const uint8_t *src[5], int stride, int h) \ 56 { \ 57 int x; \ 58 const uint8_t *s0 = src[0]; \ 59 const uint8_t *s1 = src[1]; \ 60 const uint8_t *s2 = src[2]; \ 61 const uint8_t *s3 = src[3]; \ 62 const uint8_t *w = src[4]; \ 63 \ 64 while (h--) { \ 65 for (x = 0; x < WIDTH; x++) { \ 66 OP(dst[x], (s0[x]*w[0] + s1[x]*w[1] + s2[x]*w[2] + s3[x]*w[3] + 8) >> 4); \ 67 } \ 68 \ 69 dst += stride; \ 70 s0 += stride; \ 71 s1 += stride; \ 72 s2 += stride; \ 73 s3 += stride; \ 74 } \ 75 } 76 77#define OP_PUT(dst, val) (dst) = (val) 78#define OP_AVG(dst, val) (dst) = (((dst) + (val) + 1)>>1) 79 80PIXOP_BILINEAR(put, OP_PUT, 8) 81PIXOP_BILINEAR(put, OP_PUT, 16) 82PIXOP_BILINEAR(put, OP_PUT, 32) 83PIXOP_BILINEAR(avg, OP_AVG, 8) 84PIXOP_BILINEAR(avg, OP_AVG, 16) 85PIXOP_BILINEAR(avg, OP_AVG, 32) 86 87#define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + (1<<(log2_denom-1))) >> log2_denom) 88#define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + (1<<(log2_denom-1))) >> log2_denom) 89 90#define DIRAC_WEIGHT(W) \ 91 static void weight_dirac_pixels ## W ## _c(uint8_t *block, int stride, int log2_denom, \ 92 int weight, int h) { \ 93 int x; \ 94 while (h--) { \ 95 for (x = 0; x < W; x++) { \ 96 op_scale1(x); \ 97 op_scale1(x+1); \ 98 } \ 99 block += stride; \ 100 } \ 101 } \ 102 static void biweight_dirac_pixels ## W ## _c(uint8_t *dst, const uint8_t *src, int stride, int log2_denom, \ 103 int weightd, int weights, int h) { \ 104 int x; \ 105 while (h--) { \ 106 for (x = 0; x < W; x++) { \ 107 op_scale2(x); \ 108 op_scale2(x+1); \ 109 } \ 110 dst += stride; \ 111 src += stride; \ 112 } \ 113 } 114 115DIRAC_WEIGHT(8) 116DIRAC_WEIGHT(16) 117DIRAC_WEIGHT(32) 118 119#define ADD_OBMC(xblen) \ 120 static void add_obmc ## xblen ## _c(uint16_t *dst, const uint8_t *src, int stride, \ 121 const uint8_t *obmc_weight, int yblen) \ 122 { \ 123 int x; \ 124 while (yblen--) { \ 125 for (x = 0; x < xblen; x += 2) { \ 126 dst[x ] += src[x ] * obmc_weight[x ]; \ 127 dst[x+1] += src[x+1] * obmc_weight[x+1]; \ 128 } \ 129 dst += stride; \ 130 src += stride; \ 131 obmc_weight += 32; \ 132 } \ 133 } 134 135ADD_OBMC(8) 136ADD_OBMC(16) 137ADD_OBMC(32) 138 139static void put_signed_rect_clamped_8bit_c(uint8_t *dst, int dst_stride, const uint8_t *_src, int src_stride, int width, int height) 140{ 141 int x, y; 142 int16_t *src = (int16_t *)_src; 143 for (y = 0; y < height; y++) { 144 for (x = 0; x < width; x+=4) { 145 dst[x ] = av_clip_uint8(src[x ] + 128); 146 dst[x+1] = av_clip_uint8(src[x+1] + 128); 147 dst[x+2] = av_clip_uint8(src[x+2] + 128); 148 dst[x+3] = av_clip_uint8(src[x+3] + 128); 149 } 150 dst += dst_stride; 151 src += src_stride >> 1; 152 } 153} 154 155#define PUT_SIGNED_RECT_CLAMPED(PX) \ 156static void put_signed_rect_clamped_ ## PX ## bit_c(uint8_t *_dst, int dst_stride, const uint8_t *_src, \ 157 int src_stride, int width, int height) \ 158{ \ 159 int x, y; \ 160 uint16_t *dst = (uint16_t *)_dst; \ 161 int32_t *src = (int32_t *)_src; \ 162 for (y = 0; y < height; y++) { \ 163 for (x = 0; x < width; x+=4) { \ 164 dst[x ] = av_clip_uintp2(src[x ] + (1U << (PX - 1)), PX); \ 165 dst[x+1] = av_clip_uintp2(src[x+1] + (1U << (PX - 1)), PX); \ 166 dst[x+2] = av_clip_uintp2(src[x+2] + (1U << (PX - 1)), PX); \ 167 dst[x+3] = av_clip_uintp2(src[x+3] + (1U << (PX - 1)), PX); \ 168 } \ 169 dst += dst_stride >> 1; \ 170 src += src_stride >> 2; \ 171 } \ 172} 173 174PUT_SIGNED_RECT_CLAMPED(10) 175PUT_SIGNED_RECT_CLAMPED(12) 176 177static void add_rect_clamped_c(uint8_t *dst, const uint16_t *src, int stride, 178 const int16_t *idwt, int idwt_stride, 179 int width, int height) 180{ 181 int x, y; 182 183 for (y = 0; y < height; y++) { 184 for (x = 0; x < width; x+=2) { 185 dst[x ] = av_clip_uint8(((src[x ]+32)>>6) + idwt[x ]); 186 dst[x+1] = av_clip_uint8(((src[x+1]+32)>>6) + idwt[x+1]); 187 } 188 dst += stride; 189 src += stride; 190 idwt += idwt_stride; 191 } 192} 193 194#define DEQUANT_SUBBAND(PX) \ 195static void dequant_subband_ ## PX ## _c(uint8_t *src, uint8_t *dst, ptrdiff_t stride, \ 196 const int qf, const int qs, int tot_v, int tot_h) \ 197{ \ 198 int i, y; \ 199 for (y = 0; y < tot_v; y++) { \ 200 PX c, *src_r = (PX *)src, *dst_r = (PX *)dst; \ 201 for (i = 0; i < tot_h; i++) { \ 202 c = *src_r++; \ 203 if (c < 0) c = -((-(unsigned)c*qf + qs) >> 2); \ 204 else if(c > 0) c = (( (unsigned)c*qf + qs) >> 2); \ 205 *dst_r++ = c; \ 206 } \ 207 src += tot_h << (sizeof(PX) >> 1); \ 208 dst += stride; \ 209 } \ 210} 211 212DEQUANT_SUBBAND(int16_t) 213DEQUANT_SUBBAND(int32_t) 214 215#define PIXFUNC(PFX, WIDTH) \ 216 c->PFX ## _dirac_pixels_tab[WIDTH>>4][0] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _c; \ 217 c->PFX ## _dirac_pixels_tab[WIDTH>>4][1] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _l2_c; \ 218 c->PFX ## _dirac_pixels_tab[WIDTH>>4][2] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _l4_c; \ 219 c->PFX ## _dirac_pixels_tab[WIDTH>>4][3] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _bilinear_c 220 221av_cold void ff_diracdsp_init(DiracDSPContext *c) 222{ 223 c->dirac_hpel_filter = dirac_hpel_filter; 224 c->add_rect_clamped = add_rect_clamped_c; 225 c->put_signed_rect_clamped[0] = put_signed_rect_clamped_8bit_c; 226 c->put_signed_rect_clamped[1] = put_signed_rect_clamped_10bit_c; 227 c->put_signed_rect_clamped[2] = put_signed_rect_clamped_12bit_c; 228 229 c->add_dirac_obmc[0] = add_obmc8_c; 230 c->add_dirac_obmc[1] = add_obmc16_c; 231 c->add_dirac_obmc[2] = add_obmc32_c; 232 233 c->weight_dirac_pixels_tab[0] = weight_dirac_pixels8_c; 234 c->weight_dirac_pixels_tab[1] = weight_dirac_pixels16_c; 235 c->weight_dirac_pixels_tab[2] = weight_dirac_pixels32_c; 236 c->biweight_dirac_pixels_tab[0] = biweight_dirac_pixels8_c; 237 c->biweight_dirac_pixels_tab[1] = biweight_dirac_pixels16_c; 238 c->biweight_dirac_pixels_tab[2] = biweight_dirac_pixels32_c; 239 240 c->dequant_subband[0] = c->dequant_subband[2] = dequant_subband_int16_t_c; 241 c->dequant_subband[1] = c->dequant_subband[3] = dequant_subband_int32_t_c; 242 243 PIXFUNC(put, 8); 244 PIXFUNC(put, 16); 245 PIXFUNC(put, 32); 246 PIXFUNC(avg, 8); 247 PIXFUNC(avg, 16); 248 PIXFUNC(avg, 32); 249 250#if ARCH_X86 251 ff_diracdsp_init_x86(c); 252#endif 253} 254