1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/variance_c.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,458 @@
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 "variance.h"
1.16 +#include "filter.h"
1.17 +
1.18 +
1.19 +unsigned int vp8_get_mb_ss_c
1.20 +(
1.21 + const short *src_ptr
1.22 +)
1.23 +{
1.24 + unsigned int i = 0, sum = 0;
1.25 +
1.26 + do
1.27 + {
1.28 + sum += (src_ptr[i] * src_ptr[i]);
1.29 + i++;
1.30 + }
1.31 + while (i < 256);
1.32 +
1.33 + return sum;
1.34 +}
1.35 +
1.36 +
1.37 +static void variance(
1.38 + const unsigned char *src_ptr,
1.39 + int source_stride,
1.40 + const unsigned char *ref_ptr,
1.41 + int recon_stride,
1.42 + int w,
1.43 + int h,
1.44 + unsigned int *sse,
1.45 + int *sum)
1.46 +{
1.47 + int i, j;
1.48 + int diff;
1.49 +
1.50 + *sum = 0;
1.51 + *sse = 0;
1.52 +
1.53 + for (i = 0; i < h; i++)
1.54 + {
1.55 + for (j = 0; j < w; j++)
1.56 + {
1.57 + diff = src_ptr[j] - ref_ptr[j];
1.58 + *sum += diff;
1.59 + *sse += diff * diff;
1.60 + }
1.61 +
1.62 + src_ptr += source_stride;
1.63 + ref_ptr += recon_stride;
1.64 + }
1.65 +}
1.66 +
1.67 +
1.68 +unsigned int vp8_variance16x16_c(
1.69 + const unsigned char *src_ptr,
1.70 + int source_stride,
1.71 + const unsigned char *ref_ptr,
1.72 + int recon_stride,
1.73 + unsigned int *sse)
1.74 +{
1.75 + unsigned int var;
1.76 + int avg;
1.77 +
1.78 +
1.79 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
1.80 + *sse = var;
1.81 + return (var - (((unsigned int)avg * avg) >> 8));
1.82 +}
1.83 +
1.84 +unsigned int vp8_variance8x16_c(
1.85 + const unsigned char *src_ptr,
1.86 + int source_stride,
1.87 + const unsigned char *ref_ptr,
1.88 + int recon_stride,
1.89 + unsigned int *sse)
1.90 +{
1.91 + unsigned int var;
1.92 + int avg;
1.93 +
1.94 +
1.95 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16, &var, &avg);
1.96 + *sse = var;
1.97 + return (var - (((unsigned int)avg * avg) >> 7));
1.98 +}
1.99 +
1.100 +unsigned int vp8_variance16x8_c(
1.101 + const unsigned char *src_ptr,
1.102 + int source_stride,
1.103 + const unsigned char *ref_ptr,
1.104 + int recon_stride,
1.105 + unsigned int *sse)
1.106 +{
1.107 + unsigned int var;
1.108 + int avg;
1.109 +
1.110 +
1.111 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8, &var, &avg);
1.112 + *sse = var;
1.113 + return (var - (((unsigned int)avg * avg) >> 7));
1.114 +}
1.115 +
1.116 +
1.117 +unsigned int vp8_variance8x8_c(
1.118 + const unsigned char *src_ptr,
1.119 + int source_stride,
1.120 + const unsigned char *ref_ptr,
1.121 + int recon_stride,
1.122 + unsigned int *sse)
1.123 +{
1.124 + unsigned int var;
1.125 + int avg;
1.126 +
1.127 +
1.128 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, &var, &avg);
1.129 + *sse = var;
1.130 + return (var - (((unsigned int)avg * avg) >> 6));
1.131 +}
1.132 +
1.133 +unsigned int vp8_variance4x4_c(
1.134 + const unsigned char *src_ptr,
1.135 + int source_stride,
1.136 + const unsigned char *ref_ptr,
1.137 + int recon_stride,
1.138 + unsigned int *sse)
1.139 +{
1.140 + unsigned int var;
1.141 + int avg;
1.142 +
1.143 +
1.144 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 4, 4, &var, &avg);
1.145 + *sse = var;
1.146 + return (var - (((unsigned int)avg * avg) >> 4));
1.147 +}
1.148 +
1.149 +
1.150 +unsigned int vp8_mse16x16_c(
1.151 + const unsigned char *src_ptr,
1.152 + int source_stride,
1.153 + const unsigned char *ref_ptr,
1.154 + int recon_stride,
1.155 + unsigned int *sse)
1.156 +{
1.157 + unsigned int var;
1.158 + int avg;
1.159 +
1.160 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
1.161 + *sse = var;
1.162 + return var;
1.163 +}
1.164 +
1.165 +
1.166 +/****************************************************************************
1.167 + *
1.168 + * ROUTINE : filter_block2d_bil_first_pass
1.169 + *
1.170 + * INPUTS : UINT8 *src_ptr : Pointer to source block.
1.171 + * UINT32 src_pixels_per_line : Stride of input block.
1.172 + * UINT32 pixel_step : Offset between filter input samples (see notes).
1.173 + * UINT32 output_height : Input block height.
1.174 + * UINT32 output_width : Input block width.
1.175 + * INT32 *vp8_filter : Array of 2 bi-linear filter taps.
1.176 + *
1.177 + * OUTPUTS : INT32 *output_ptr : Pointer to filtered block.
1.178 + *
1.179 + * RETURNS : void
1.180 + *
1.181 + * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
1.182 + * either horizontal or vertical direction to produce the
1.183 + * filtered output block. Used to implement first-pass
1.184 + * of 2-D separable filter.
1.185 + *
1.186 + * SPECIAL NOTES : Produces INT32 output to retain precision for next pass.
1.187 + * Two filter taps should sum to VP8_FILTER_WEIGHT.
1.188 + * pixel_step defines whether the filter is applied
1.189 + * horizontally (pixel_step=1) or vertically (pixel_step=stride).
1.190 + * It defines the offset required to move from one input
1.191 + * to the next.
1.192 + *
1.193 + ****************************************************************************/
1.194 +static void var_filter_block2d_bil_first_pass
1.195 +(
1.196 + const unsigned char *src_ptr,
1.197 + unsigned short *output_ptr,
1.198 + unsigned int src_pixels_per_line,
1.199 + int pixel_step,
1.200 + unsigned int output_height,
1.201 + unsigned int output_width,
1.202 + const short *vp8_filter
1.203 +)
1.204 +{
1.205 + unsigned int i, j;
1.206 +
1.207 + for (i = 0; i < output_height; i++)
1.208 + {
1.209 + for (j = 0; j < output_width; j++)
1.210 + {
1.211 + /* Apply bilinear filter */
1.212 + output_ptr[j] = (((int)src_ptr[0] * vp8_filter[0]) +
1.213 + ((int)src_ptr[pixel_step] * vp8_filter[1]) +
1.214 + (VP8_FILTER_WEIGHT / 2)) >> VP8_FILTER_SHIFT;
1.215 + src_ptr++;
1.216 + }
1.217 +
1.218 + /* Next row... */
1.219 + src_ptr += src_pixels_per_line - output_width;
1.220 + output_ptr += output_width;
1.221 + }
1.222 +}
1.223 +
1.224 +/****************************************************************************
1.225 + *
1.226 + * ROUTINE : filter_block2d_bil_second_pass
1.227 + *
1.228 + * INPUTS : INT32 *src_ptr : Pointer to source block.
1.229 + * UINT32 src_pixels_per_line : Stride of input block.
1.230 + * UINT32 pixel_step : Offset between filter input samples (see notes).
1.231 + * UINT32 output_height : Input block height.
1.232 + * UINT32 output_width : Input block width.
1.233 + * INT32 *vp8_filter : Array of 2 bi-linear filter taps.
1.234 + *
1.235 + * OUTPUTS : UINT16 *output_ptr : Pointer to filtered block.
1.236 + *
1.237 + * RETURNS : void
1.238 + *
1.239 + * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
1.240 + * either horizontal or vertical direction to produce the
1.241 + * filtered output block. Used to implement second-pass
1.242 + * of 2-D separable filter.
1.243 + *
1.244 + * SPECIAL NOTES : Requires 32-bit input as produced by filter_block2d_bil_first_pass.
1.245 + * Two filter taps should sum to VP8_FILTER_WEIGHT.
1.246 + * pixel_step defines whether the filter is applied
1.247 + * horizontally (pixel_step=1) or vertically (pixel_step=stride).
1.248 + * It defines the offset required to move from one input
1.249 + * to the next.
1.250 + *
1.251 + ****************************************************************************/
1.252 +static void var_filter_block2d_bil_second_pass
1.253 +(
1.254 + const unsigned short *src_ptr,
1.255 + unsigned char *output_ptr,
1.256 + unsigned int src_pixels_per_line,
1.257 + unsigned int pixel_step,
1.258 + unsigned int output_height,
1.259 + unsigned int output_width,
1.260 + const short *vp8_filter
1.261 +)
1.262 +{
1.263 + unsigned int i, j;
1.264 + int Temp;
1.265 +
1.266 + for (i = 0; i < output_height; i++)
1.267 + {
1.268 + for (j = 0; j < output_width; j++)
1.269 + {
1.270 + /* Apply filter */
1.271 + Temp = ((int)src_ptr[0] * vp8_filter[0]) +
1.272 + ((int)src_ptr[pixel_step] * vp8_filter[1]) +
1.273 + (VP8_FILTER_WEIGHT / 2);
1.274 + output_ptr[j] = (unsigned int)(Temp >> VP8_FILTER_SHIFT);
1.275 + src_ptr++;
1.276 + }
1.277 +
1.278 + /* Next row... */
1.279 + src_ptr += src_pixels_per_line - output_width;
1.280 + output_ptr += output_width;
1.281 + }
1.282 +}
1.283 +
1.284 +
1.285 +unsigned int vp8_sub_pixel_variance4x4_c
1.286 +(
1.287 + const unsigned char *src_ptr,
1.288 + int src_pixels_per_line,
1.289 + int xoffset,
1.290 + int yoffset,
1.291 + const unsigned char *dst_ptr,
1.292 + int dst_pixels_per_line,
1.293 + unsigned int *sse
1.294 +)
1.295 +{
1.296 + unsigned char temp2[20*16];
1.297 + const short *HFilter, *VFilter;
1.298 + unsigned short FData3[5*4]; /* Temp data bufffer used in filtering */
1.299 +
1.300 + HFilter = vp8_bilinear_filters[xoffset];
1.301 + VFilter = vp8_bilinear_filters[yoffset];
1.302 +
1.303 + /* First filter 1d Horizontal */
1.304 + var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 5, 4, HFilter);
1.305 +
1.306 + /* Now filter Verticaly */
1.307 + var_filter_block2d_bil_second_pass(FData3, temp2, 4, 4, 4, 4, VFilter);
1.308 +
1.309 + return vp8_variance4x4_c(temp2, 4, dst_ptr, dst_pixels_per_line, sse);
1.310 +}
1.311 +
1.312 +
1.313 +unsigned int vp8_sub_pixel_variance8x8_c
1.314 +(
1.315 + const unsigned char *src_ptr,
1.316 + int src_pixels_per_line,
1.317 + int xoffset,
1.318 + int yoffset,
1.319 + const unsigned char *dst_ptr,
1.320 + int dst_pixels_per_line,
1.321 + unsigned int *sse
1.322 +)
1.323 +{
1.324 + unsigned short FData3[9*8]; /* Temp data bufffer used in filtering */
1.325 + unsigned char temp2[20*16];
1.326 + const short *HFilter, *VFilter;
1.327 +
1.328 + HFilter = vp8_bilinear_filters[xoffset];
1.329 + VFilter = vp8_bilinear_filters[yoffset];
1.330 +
1.331 + var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 9, 8, HFilter);
1.332 + var_filter_block2d_bil_second_pass(FData3, temp2, 8, 8, 8, 8, VFilter);
1.333 +
1.334 + return vp8_variance8x8_c(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
1.335 +}
1.336 +
1.337 +unsigned int vp8_sub_pixel_variance16x16_c
1.338 +(
1.339 + const unsigned char *src_ptr,
1.340 + int src_pixels_per_line,
1.341 + int xoffset,
1.342 + int yoffset,
1.343 + const unsigned char *dst_ptr,
1.344 + int dst_pixels_per_line,
1.345 + unsigned int *sse
1.346 +)
1.347 +{
1.348 + unsigned short FData3[17*16]; /* Temp data bufffer used in filtering */
1.349 + unsigned char temp2[20*16];
1.350 + const short *HFilter, *VFilter;
1.351 +
1.352 + HFilter = vp8_bilinear_filters[xoffset];
1.353 + VFilter = vp8_bilinear_filters[yoffset];
1.354 +
1.355 + var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 17, 16, HFilter);
1.356 + var_filter_block2d_bil_second_pass(FData3, temp2, 16, 16, 16, 16, VFilter);
1.357 +
1.358 + return vp8_variance16x16_c(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
1.359 +}
1.360 +
1.361 +
1.362 +unsigned int vp8_variance_halfpixvar16x16_h_c(
1.363 + const unsigned char *src_ptr,
1.364 + int source_stride,
1.365 + const unsigned char *ref_ptr,
1.366 + int recon_stride,
1.367 + unsigned int *sse)
1.368 +{
1.369 + return vp8_sub_pixel_variance16x16_c(src_ptr, source_stride, 4, 0,
1.370 + ref_ptr, recon_stride, sse);
1.371 +}
1.372 +
1.373 +
1.374 +unsigned int vp8_variance_halfpixvar16x16_v_c(
1.375 + const unsigned char *src_ptr,
1.376 + int source_stride,
1.377 + const unsigned char *ref_ptr,
1.378 + int recon_stride,
1.379 + unsigned int *sse)
1.380 +{
1.381 + return vp8_sub_pixel_variance16x16_c(src_ptr, source_stride, 0, 4,
1.382 + ref_ptr, recon_stride, sse);
1.383 +}
1.384 +
1.385 +
1.386 +unsigned int vp8_variance_halfpixvar16x16_hv_c(
1.387 + const unsigned char *src_ptr,
1.388 + int source_stride,
1.389 + const unsigned char *ref_ptr,
1.390 + int recon_stride,
1.391 + unsigned int *sse)
1.392 +{
1.393 + return vp8_sub_pixel_variance16x16_c(src_ptr, source_stride, 4, 4,
1.394 + ref_ptr, recon_stride, sse);
1.395 +}
1.396 +
1.397 +
1.398 +unsigned int vp8_sub_pixel_mse16x16_c
1.399 +(
1.400 + const unsigned char *src_ptr,
1.401 + int src_pixels_per_line,
1.402 + int xoffset,
1.403 + int yoffset,
1.404 + const unsigned char *dst_ptr,
1.405 + int dst_pixels_per_line,
1.406 + unsigned int *sse
1.407 +)
1.408 +{
1.409 + vp8_sub_pixel_variance16x16_c(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
1.410 + return *sse;
1.411 +}
1.412 +
1.413 +unsigned int vp8_sub_pixel_variance16x8_c
1.414 +(
1.415 + const unsigned char *src_ptr,
1.416 + int src_pixels_per_line,
1.417 + int xoffset,
1.418 + int yoffset,
1.419 + const unsigned char *dst_ptr,
1.420 + int dst_pixels_per_line,
1.421 + unsigned int *sse
1.422 +)
1.423 +{
1.424 + unsigned short FData3[16*9]; /* Temp data bufffer used in filtering */
1.425 + unsigned char temp2[20*16];
1.426 + const short *HFilter, *VFilter;
1.427 +
1.428 + HFilter = vp8_bilinear_filters[xoffset];
1.429 + VFilter = vp8_bilinear_filters[yoffset];
1.430 +
1.431 + var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 9, 16, HFilter);
1.432 + var_filter_block2d_bil_second_pass(FData3, temp2, 16, 16, 8, 16, VFilter);
1.433 +
1.434 + return vp8_variance16x8_c(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
1.435 +}
1.436 +
1.437 +unsigned int vp8_sub_pixel_variance8x16_c
1.438 +(
1.439 + const unsigned char *src_ptr,
1.440 + int src_pixels_per_line,
1.441 + int xoffset,
1.442 + int yoffset,
1.443 + const unsigned char *dst_ptr,
1.444 + int dst_pixels_per_line,
1.445 + unsigned int *sse
1.446 +)
1.447 +{
1.448 + unsigned short FData3[9*16]; /* Temp data bufffer used in filtering */
1.449 + unsigned char temp2[20*16];
1.450 + const short *HFilter, *VFilter;
1.451 +
1.452 +
1.453 + HFilter = vp8_bilinear_filters[xoffset];
1.454 + VFilter = vp8_bilinear_filters[yoffset];
1.455 +
1.456 +
1.457 + var_filter_block2d_bil_first_pass(src_ptr, FData3, src_pixels_per_line, 1, 17, 8, HFilter);
1.458 + var_filter_block2d_bil_second_pass(FData3, temp2, 8, 8, 16, 8, VFilter);
1.459 +
1.460 + return vp8_variance8x16_c(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
1.461 +}
