1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/sad_c.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,302 @@
1.4 +/*
1.5 + * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +
1.15 +#include <limits.h>
1.16 +#include <stdlib.h>
1.17 +#include "vpx_config.h"
1.18 +#include "vpx/vpx_integer.h"
1.19 +
1.20 +static unsigned int sad_mx_n_c(const unsigned char *src_ptr, int src_stride,
1.21 + const unsigned char *ref_ptr, int ref_stride,
1.22 + unsigned int max_sad, int m, int n)
1.23 +{
1.24 + int r, c;
1.25 + unsigned int sad = 0;
1.26 +
1.27 + for (r = 0; r < n; r++)
1.28 + {
1.29 + for (c = 0; c < m; c++)
1.30 + {
1.31 + sad += abs(src_ptr[c] - ref_ptr[c]);
1.32 + }
1.33 +
1.34 + if (sad > max_sad)
1.35 + break;
1.36 +
1.37 + src_ptr += src_stride;
1.38 + ref_ptr += ref_stride;
1.39 + }
1.40 +
1.41 + return sad;
1.42 +}
1.43 +
1.44 +/* max_sad is provided as an optional optimization point. Alternative
1.45 + * implementations of these functions are not required to check it.
1.46 + */
1.47 +
1.48 +unsigned int vp8_sad16x16_c(const unsigned char *src_ptr, int src_stride,
1.49 + const unsigned char *ref_ptr, int ref_stride,
1.50 + unsigned int max_sad)
1.51 +{
1.52 + return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, max_sad, 16, 16);
1.53 +}
1.54 +
1.55 +unsigned int vp8_sad8x8_c(const unsigned char *src_ptr, int src_stride,
1.56 + const unsigned char *ref_ptr, int ref_stride,
1.57 + unsigned int max_sad)
1.58 +{
1.59 + return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, max_sad, 8, 8);
1.60 +}
1.61 +
1.62 +unsigned int vp8_sad16x8_c(const unsigned char *src_ptr, int src_stride,
1.63 + const unsigned char *ref_ptr, int ref_stride,
1.64 + unsigned int max_sad)
1.65 +{
1.66 + return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, max_sad, 16, 8);
1.67 +
1.68 +}
1.69 +
1.70 +unsigned int vp8_sad8x16_c(const unsigned char *src_ptr, int src_stride,
1.71 + const unsigned char *ref_ptr, int ref_stride,
1.72 + unsigned int max_sad)
1.73 +{
1.74 + return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, max_sad, 8, 16);
1.75 +}
1.76 +
1.77 +unsigned int vp8_sad4x4_c(const unsigned char *src_ptr, int src_stride,
1.78 + const unsigned char *ref_ptr, int ref_stride,
1.79 + unsigned int max_sad)
1.80 +{
1.81 + return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, max_sad, 4, 4);
1.82 +}
1.83 +
1.84 +void vp8_sad16x16x3_c(const unsigned char *src_ptr, int src_stride,
1.85 + const unsigned char *ref_ptr, int ref_stride,
1.86 + unsigned int *sad_array)
1.87 +{
1.88 + sad_array[0] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.89 + sad_array[1] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.90 + sad_array[2] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.91 +}
1.92 +
1.93 +void vp8_sad16x16x8_c(const unsigned char *src_ptr, int src_stride,
1.94 + const unsigned char *ref_ptr, int ref_stride,
1.95 + unsigned short *sad_array)
1.96 +{
1.97 + sad_array[0] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.98 + sad_array[1] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.99 + sad_array[2] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.100 + sad_array[3] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 3, ref_stride, UINT_MAX);
1.101 + sad_array[4] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 4, ref_stride, UINT_MAX);
1.102 + sad_array[5] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 5, ref_stride, UINT_MAX);
1.103 + sad_array[6] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 6, ref_stride, UINT_MAX);
1.104 + sad_array[7] = (unsigned short)vp8_sad16x16_c(src_ptr, src_stride, ref_ptr + 7, ref_stride, UINT_MAX);
1.105 +}
1.106 +
1.107 +void vp8_sad16x8x3_c(const unsigned char *src_ptr, int src_stride,
1.108 + const unsigned char *ref_ptr, int ref_stride,
1.109 + unsigned int *sad_array)
1.110 +{
1.111 + sad_array[0] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.112 + sad_array[1] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.113 + sad_array[2] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.114 +}
1.115 +
1.116 +void vp8_sad16x8x8_c(const unsigned char *src_ptr, int src_stride,
1.117 + const unsigned char *ref_ptr, int ref_stride,
1.118 + unsigned short *sad_array)
1.119 +{
1.120 + sad_array[0] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.121 + sad_array[1] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.122 + sad_array[2] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.123 + sad_array[3] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 3, ref_stride, UINT_MAX);
1.124 + sad_array[4] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 4, ref_stride, UINT_MAX);
1.125 + sad_array[5] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 5, ref_stride, UINT_MAX);
1.126 + sad_array[6] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 6, ref_stride, UINT_MAX);
1.127 + sad_array[7] = (unsigned short)vp8_sad16x8_c(src_ptr, src_stride, ref_ptr + 7, ref_stride, UINT_MAX);
1.128 +}
1.129 +
1.130 +void vp8_sad8x8x3_c(const unsigned char *src_ptr, int src_stride,
1.131 + const unsigned char *ref_ptr, int ref_stride,
1.132 + unsigned int *sad_array)
1.133 +{
1.134 + sad_array[0] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.135 + sad_array[1] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.136 + sad_array[2] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.137 +}
1.138 +
1.139 +void vp8_sad8x8x8_c(const unsigned char *src_ptr, int src_stride,
1.140 + const unsigned char *ref_ptr, int ref_stride,
1.141 + unsigned short *sad_array)
1.142 +{
1.143 + sad_array[0] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.144 + sad_array[1] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.145 + sad_array[2] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.146 + sad_array[3] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 3, ref_stride, UINT_MAX);
1.147 + sad_array[4] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 4, ref_stride, UINT_MAX);
1.148 + sad_array[5] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 5, ref_stride, UINT_MAX);
1.149 + sad_array[6] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 6, ref_stride, UINT_MAX);
1.150 + sad_array[7] = (unsigned short)vp8_sad8x8_c(src_ptr, src_stride, ref_ptr + 7, ref_stride, UINT_MAX);
1.151 +}
1.152 +
1.153 +void vp8_sad8x16x3_c(const unsigned char *src_ptr, int src_stride,
1.154 + const unsigned char *ref_ptr, int ref_stride,
1.155 + unsigned int *sad_array)
1.156 +{
1.157 + sad_array[0] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.158 + sad_array[1] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.159 + sad_array[2] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.160 +}
1.161 +
1.162 +void vp8_sad8x16x8_c(const unsigned char *src_ptr, int src_stride,
1.163 + const unsigned char *ref_ptr, int ref_stride,
1.164 + unsigned short *sad_array)
1.165 +{
1.166 + sad_array[0] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.167 + sad_array[1] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.168 + sad_array[2] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.169 + sad_array[3] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 3, ref_stride, UINT_MAX);
1.170 + sad_array[4] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 4, ref_stride, UINT_MAX);
1.171 + sad_array[5] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 5, ref_stride, UINT_MAX);
1.172 + sad_array[6] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 6, ref_stride, UINT_MAX);
1.173 + sad_array[7] = (unsigned short)vp8_sad8x16_c(src_ptr, src_stride, ref_ptr + 7, ref_stride, UINT_MAX);
1.174 +}
1.175 +
1.176 +void vp8_sad4x4x3_c(const unsigned char *src_ptr, int src_stride,
1.177 + const unsigned char *ref_ptr, int ref_stride,
1.178 + unsigned int *sad_array)
1.179 +{
1.180 + sad_array[0] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.181 + sad_array[1] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.182 + sad_array[2] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.183 +}
1.184 +
1.185 +void vp8_sad4x4x8_c(const unsigned char *src_ptr, int src_stride,
1.186 + const unsigned char *ref_ptr, int ref_stride,
1.187 + unsigned short *sad_array)
1.188 +{
1.189 + sad_array[0] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 0, ref_stride, UINT_MAX);
1.190 + sad_array[1] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 1, ref_stride, UINT_MAX);
1.191 + sad_array[2] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 2, ref_stride, UINT_MAX);
1.192 + sad_array[3] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 3, ref_stride, UINT_MAX);
1.193 + sad_array[4] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 4, ref_stride, UINT_MAX);
1.194 + sad_array[5] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 5, ref_stride, UINT_MAX);
1.195 + sad_array[6] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 6, ref_stride, UINT_MAX);
1.196 + sad_array[7] = (unsigned short)vp8_sad4x4_c(src_ptr, src_stride, ref_ptr + 7, ref_stride, UINT_MAX);
1.197 +}
1.198 +
1.199 +void vp8_sad16x16x4d_c(const unsigned char *src_ptr, int src_stride,
1.200 + const unsigned char * const ref_ptr[], int ref_stride,
1.201 + unsigned int *sad_array)
1.202 +{
1.203 + sad_array[0] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr[0], ref_stride, UINT_MAX);
1.204 + sad_array[1] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr[1], ref_stride, UINT_MAX);
1.205 + sad_array[2] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr[2], ref_stride, UINT_MAX);
1.206 + sad_array[3] = vp8_sad16x16_c(src_ptr, src_stride, ref_ptr[3], ref_stride, UINT_MAX);
1.207 +}
1.208 +
1.209 +void vp8_sad16x8x4d_c(const unsigned char *src_ptr, int src_stride,
1.210 + const unsigned char * const ref_ptr[], int ref_stride,
1.211 + unsigned int *sad_array)
1.212 +{
1.213 + sad_array[0] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr[0], ref_stride, UINT_MAX);
1.214 + sad_array[1] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr[1], ref_stride, UINT_MAX);
1.215 + sad_array[2] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr[2], ref_stride, UINT_MAX);
1.216 + sad_array[3] = vp8_sad16x8_c(src_ptr, src_stride, ref_ptr[3], ref_stride, UINT_MAX);
1.217 +}
1.218 +
1.219 +void vp8_sad8x8x4d_c(const unsigned char *src_ptr, int src_stride,
1.220 + const unsigned char * const ref_ptr[], int ref_stride,
1.221 + unsigned int *sad_array)
1.222 +{
1.223 + sad_array[0] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr[0], ref_stride, UINT_MAX);
1.224 + sad_array[1] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr[1], ref_stride, UINT_MAX);
1.225 + sad_array[2] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr[2], ref_stride, UINT_MAX);
1.226 + sad_array[3] = vp8_sad8x8_c(src_ptr, src_stride, ref_ptr[3], ref_stride, UINT_MAX);
1.227 +}
1.228 +
1.229 +void vp8_sad8x16x4d_c(const unsigned char *src_ptr, int src_stride,
1.230 + const unsigned char * const ref_ptr[], int ref_stride,
1.231 + unsigned int *sad_array)
1.232 +{
1.233 + sad_array[0] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr[0], ref_stride, UINT_MAX);
1.234 + sad_array[1] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr[1], ref_stride, UINT_MAX);
1.235 + sad_array[2] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr[2], ref_stride, UINT_MAX);
1.236 + sad_array[3] = vp8_sad8x16_c(src_ptr, src_stride, ref_ptr[3], ref_stride, UINT_MAX);
1.237 +}
1.238 +
1.239 +void vp8_sad4x4x4d_c(const unsigned char *src_ptr, int src_stride,
1.240 + const unsigned char * const ref_ptr[], int ref_stride,
1.241 + unsigned int *sad_array)
1.242 +{
1.243 + sad_array[0] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr[0], ref_stride, UINT_MAX);
1.244 + sad_array[1] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr[1], ref_stride, UINT_MAX);
1.245 + sad_array[2] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr[2], ref_stride, UINT_MAX);
1.246 + sad_array[3] = vp8_sad4x4_c(src_ptr, src_stride, ref_ptr[3], ref_stride, UINT_MAX);
1.247 +}
1.248 +
1.249 +/* Copy 2 macroblocks to a buffer */
1.250 +void vp8_copy32xn_c(unsigned char *src_ptr, int src_stride,
1.251 + unsigned char *dst_ptr, int dst_stride,
1.252 + int height)
1.253 +{
1.254 + int r;
1.255 +
1.256 + for (r = 0; r < height; r++)
1.257 + {
1.258 +#if !(CONFIG_FAST_UNALIGNED)
1.259 + dst_ptr[0] = src_ptr[0];
1.260 + dst_ptr[1] = src_ptr[1];
1.261 + dst_ptr[2] = src_ptr[2];
1.262 + dst_ptr[3] = src_ptr[3];
1.263 + dst_ptr[4] = src_ptr[4];
1.264 + dst_ptr[5] = src_ptr[5];
1.265 + dst_ptr[6] = src_ptr[6];
1.266 + dst_ptr[7] = src_ptr[7];
1.267 + dst_ptr[8] = src_ptr[8];
1.268 + dst_ptr[9] = src_ptr[9];
1.269 + dst_ptr[10] = src_ptr[10];
1.270 + dst_ptr[11] = src_ptr[11];
1.271 + dst_ptr[12] = src_ptr[12];
1.272 + dst_ptr[13] = src_ptr[13];
1.273 + dst_ptr[14] = src_ptr[14];
1.274 + dst_ptr[15] = src_ptr[15];
1.275 + dst_ptr[16] = src_ptr[16];
1.276 + dst_ptr[17] = src_ptr[17];
1.277 + dst_ptr[18] = src_ptr[18];
1.278 + dst_ptr[19] = src_ptr[19];
1.279 + dst_ptr[20] = src_ptr[20];
1.280 + dst_ptr[21] = src_ptr[21];
1.281 + dst_ptr[22] = src_ptr[22];
1.282 + dst_ptr[23] = src_ptr[23];
1.283 + dst_ptr[24] = src_ptr[24];
1.284 + dst_ptr[25] = src_ptr[25];
1.285 + dst_ptr[26] = src_ptr[26];
1.286 + dst_ptr[27] = src_ptr[27];
1.287 + dst_ptr[28] = src_ptr[28];
1.288 + dst_ptr[29] = src_ptr[29];
1.289 + dst_ptr[30] = src_ptr[30];
1.290 + dst_ptr[31] = src_ptr[31];
1.291 +#else
1.292 + ((uint32_t *)dst_ptr)[0] = ((uint32_t *)src_ptr)[0] ;
1.293 + ((uint32_t *)dst_ptr)[1] = ((uint32_t *)src_ptr)[1] ;
1.294 + ((uint32_t *)dst_ptr)[2] = ((uint32_t *)src_ptr)[2] ;
1.295 + ((uint32_t *)dst_ptr)[3] = ((uint32_t *)src_ptr)[3] ;
1.296 + ((uint32_t *)dst_ptr)[4] = ((uint32_t *)src_ptr)[4] ;
1.297 + ((uint32_t *)dst_ptr)[5] = ((uint32_t *)src_ptr)[5] ;
1.298 + ((uint32_t *)dst_ptr)[6] = ((uint32_t *)src_ptr)[6] ;
1.299 + ((uint32_t *)dst_ptr)[7] = ((uint32_t *)src_ptr)[7] ;
1.300 +#endif
1.301 + src_ptr += src_stride;
1.302 + dst_ptr += dst_stride;
1.303 +
1.304 + }
1.305 +}
