1/* 2 * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.com> 3 * 4 * This file is part of FFmpeg. 5 * 6 * FFmpeg is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License as published by 8 * the Free Software Foundation; either version 2 of the License, or 9 * (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 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License along 17 * with FFmpeg; if not, write to the Free Software Foundation, Inc., 18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. 19 */ 20 21#include <string.h> 22 23#include "libavutil/intreadwrite.h" 24#include "libavutil/mem_internal.h" 25 26#include "libavcodec/avcodec.h" 27 28#include "libavcodec/hevcdsp.h" 29 30#include "checkasm.h" 31 32static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff }; 33static const uint32_t sao_size[5] = {8, 16, 32, 48, 64}; 34 35#define SIZEOF_PIXEL ((bit_depth + 7) / 8) 36#define PIXEL_STRIDE (2*MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE) //same with sao_edge src_stride 37#define BUF_SIZE (PIXEL_STRIDE * (64+2) * 2) //+2 for top and bottom row, *2 for high bit depth 38#define OFFSET_THRESH (1 << (bit_depth - 5)) 39#define OFFSET_LENGTH 5 40 41#define randomize_buffers(buf0, buf1, size) \ 42 do { \ 43 uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \ 44 int k; \ 45 for (k = 0; k < size; k += 4) { \ 46 uint32_t r = rnd() & mask; \ 47 AV_WN32A(buf0 + k, r); \ 48 AV_WN32A(buf1 + k, r); \ 49 } \ 50 } while (0) 51 52#define randomize_buffers2(buf, size) \ 53 do { \ 54 uint32_t max_offset = OFFSET_THRESH; \ 55 int k; \ 56 if (bit_depth == 8) { \ 57 for (k = 0; k < size; k++) { \ 58 uint8_t r = rnd() % max_offset; \ 59 buf[k] = r; \ 60 } \ 61 } else { \ 62 for (k = 0; k < size; k++) { \ 63 uint16_t r = rnd() % max_offset; \ 64 buf[k] = r; \ 65 } \ 66 } \ 67 } while (0) 68 69static void check_sao_band(HEVCDSPContext h, int bit_depth) 70{ 71 int i; 72 LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]); 73 LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]); 74 LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]); 75 LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]); 76 int16_t offset_val[OFFSET_LENGTH]; 77 int left_class = rnd()%32; 78 79 for (i = 0; i <= 4; i++) { 80 int block_size = sao_size[i]; 81 int prev_size = i > 0 ? sao_size[i - 1] : 0; 82 ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL; 83 declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride, 84 int16_t *sao_offset_val, int sao_left_class, int width, int height); 85 86 if (check_func(h.sao_band_filter[i], "hevc_sao_band_%d_%d", block_size, bit_depth)) { 87 88 for (int w = prev_size + 4; w <= block_size; w += 4) { 89 randomize_buffers(src0, src1, BUF_SIZE); 90 randomize_buffers2(offset_val, OFFSET_LENGTH); 91 memset(dst0, 0, BUF_SIZE); 92 memset(dst1, 0, BUF_SIZE); 93 94 call_ref(dst0, src0, stride, stride, offset_val, left_class, w, block_size); 95 call_new(dst1, src1, stride, stride, offset_val, left_class, w, block_size); 96 for (int j = 0; j < block_size; j++) { 97 if (memcmp(dst0 + j*stride, dst1 + j*stride, w*SIZEOF_PIXEL)) 98 fail(); 99 } 100 } 101 bench_new(dst1, src1, stride, stride, offset_val, left_class, block_size, block_size); 102 } 103 } 104} 105 106static void check_sao_edge(HEVCDSPContext h, int bit_depth) 107{ 108 int i; 109 LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]); 110 LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]); 111 LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]); 112 LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]); 113 int16_t offset_val[OFFSET_LENGTH]; 114 int eo = rnd()%4; 115 116 for (i = 0; i <= 4; i++) { 117 int block_size = sao_size[i]; 118 int prev_size = i > 0 ? sao_size[i - 1] : 0; 119 ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL; 120 int offset = (AV_INPUT_BUFFER_PADDING_SIZE + PIXEL_STRIDE)*SIZEOF_PIXEL; 121 declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst, 122 int16_t *sao_offset_val, int eo, int width, int height); 123 124 for (int w = prev_size + 4; w <= block_size; w += 4) { 125 randomize_buffers(src0, src1, BUF_SIZE); 126 randomize_buffers2(offset_val, OFFSET_LENGTH); 127 memset(dst0, 0, BUF_SIZE); 128 memset(dst1, 0, BUF_SIZE); 129 130 if (check_func(h.sao_edge_filter[i], "hevc_sao_edge_%d_%d", block_size, bit_depth)) { 131 call_ref(dst0, src0 + offset, stride, offset_val, eo, w, block_size); 132 call_new(dst1, src1 + offset, stride, offset_val, eo, w, block_size); 133 for (int j = 0; j < block_size; j++) { 134 if (memcmp(dst0 + j*stride, dst1 + j*stride, w*SIZEOF_PIXEL)) 135 fail(); 136 } 137 } 138 bench_new(dst1, src1 + offset, stride, offset_val, eo, block_size, block_size); 139 } 140 } 141} 142 143void checkasm_check_hevc_sao(void) 144{ 145 int bit_depth; 146 147 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) { 148 HEVCDSPContext h; 149 150 ff_hevc_dsp_init(&h, bit_depth); 151 check_sao_band(h, bit_depth); 152 } 153 report("sao_band"); 154 155 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) { 156 HEVCDSPContext h; 157 158 ff_hevc_dsp_init(&h, bit_depth); 159 check_sao_edge(h, bit_depth); 160 } 161 report("sao_edge"); 162} 163