1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/encoder/vp9_variance_c.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,1094 @@
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 +#include "./vp9_rtcd.h"
1.15 +
1.16 +#include "vpx_ports/mem.h"
1.17 +#include "vpx/vpx_integer.h"
1.18 +
1.19 +#include "vp9/common/vp9_common.h"
1.20 +#include "vp9/common/vp9_filter.h"
1.21 +
1.22 +#include "vp9/encoder/vp9_variance.h"
1.23 +
1.24 +void variance(const uint8_t *src_ptr,
1.25 + int source_stride,
1.26 + const uint8_t *ref_ptr,
1.27 + int recon_stride,
1.28 + int w,
1.29 + int h,
1.30 + unsigned int *sse,
1.31 + int *sum) {
1.32 + int i, j;
1.33 + int diff;
1.34 +
1.35 + *sum = 0;
1.36 + *sse = 0;
1.37 +
1.38 + for (i = 0; i < h; i++) {
1.39 + for (j = 0; j < w; j++) {
1.40 + diff = src_ptr[j] - ref_ptr[j];
1.41 + *sum += diff;
1.42 + *sse += diff * diff;
1.43 + }
1.44 +
1.45 + src_ptr += source_stride;
1.46 + ref_ptr += recon_stride;
1.47 + }
1.48 +}
1.49 +
1.50 +/****************************************************************************
1.51 + *
1.52 + * ROUTINE : filter_block2d_bil_first_pass
1.53 + *
1.54 + * INPUTS : uint8_t *src_ptr : Pointer to source block.
1.55 + * uint32_t src_pixels_per_line : Stride of input block.
1.56 + * uint32_t pixel_step : Offset between filter input
1.57 + * samples (see notes).
1.58 + * uint32_t output_height : Input block height.
1.59 + * uint32_t output_width : Input block width.
1.60 + * int32_t *vp9_filter : Array of 2 bi-linear filter
1.61 + * taps.
1.62 + *
1.63 + * OUTPUTS : int32_t *output_ptr : Pointer to filtered block.
1.64 + *
1.65 + * RETURNS : void
1.66 + *
1.67 + * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
1.68 + * either horizontal or vertical direction to produce the
1.69 + * filtered output block. Used to implement first-pass
1.70 + * of 2-D separable filter.
1.71 + *
1.72 + * SPECIAL NOTES : Produces int32_t output to retain precision for next pass.
1.73 + * Two filter taps should sum to VP9_FILTER_WEIGHT.
1.74 + * pixel_step defines whether the filter is applied
1.75 + * horizontally (pixel_step=1) or vertically (pixel_step=
1.76 + * stride).
1.77 + * It defines the offset required to move from one input
1.78 + * to the next.
1.79 + *
1.80 + ****************************************************************************/
1.81 +static void var_filter_block2d_bil_first_pass(const uint8_t *src_ptr,
1.82 + uint16_t *output_ptr,
1.83 + unsigned int src_pixels_per_line,
1.84 + int pixel_step,
1.85 + unsigned int output_height,
1.86 + unsigned int output_width,
1.87 + const int16_t *vp9_filter) {
1.88 + unsigned int i, j;
1.89 +
1.90 + for (i = 0; i < output_height; i++) {
1.91 + for (j = 0; j < output_width; j++) {
1.92 + output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
1.93 + (int)src_ptr[pixel_step] * vp9_filter[1],
1.94 + FILTER_BITS);
1.95 +
1.96 + src_ptr++;
1.97 + }
1.98 +
1.99 + // Next row...
1.100 + src_ptr += src_pixels_per_line - output_width;
1.101 + output_ptr += output_width;
1.102 + }
1.103 +}
1.104 +
1.105 +/****************************************************************************
1.106 + *
1.107 + * ROUTINE : filter_block2d_bil_second_pass
1.108 + *
1.109 + * INPUTS : int32_t *src_ptr : Pointer to source block.
1.110 + * uint32_t src_pixels_per_line : Stride of input block.
1.111 + * uint32_t pixel_step : Offset between filter input
1.112 + * samples (see notes).
1.113 + * uint32_t output_height : Input block height.
1.114 + * uint32_t output_width : Input block width.
1.115 + * int32_t *vp9_filter : Array of 2 bi-linear filter
1.116 + * taps.
1.117 + *
1.118 + * OUTPUTS : uint16_t *output_ptr : Pointer to filtered block.
1.119 + *
1.120 + * RETURNS : void
1.121 + *
1.122 + * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
1.123 + * either horizontal or vertical direction to produce the
1.124 + * filtered output block. Used to implement second-pass
1.125 + * of 2-D separable filter.
1.126 + *
1.127 + * SPECIAL NOTES : Requires 32-bit input as produced by
1.128 + * filter_block2d_bil_first_pass.
1.129 + * Two filter taps should sum to VP9_FILTER_WEIGHT.
1.130 + * pixel_step defines whether the filter is applied
1.131 + * horizontally (pixel_step=1) or vertically (pixel_step=
1.132 + * stride).
1.133 + * It defines the offset required to move from one input
1.134 + * to the next.
1.135 + *
1.136 + ****************************************************************************/
1.137 +static void var_filter_block2d_bil_second_pass(const uint16_t *src_ptr,
1.138 + uint8_t *output_ptr,
1.139 + unsigned int src_pixels_per_line,
1.140 + unsigned int pixel_step,
1.141 + unsigned int output_height,
1.142 + unsigned int output_width,
1.143 + const int16_t *vp9_filter) {
1.144 + unsigned int i, j;
1.145 +
1.146 + for (i = 0; i < output_height; i++) {
1.147 + for (j = 0; j < output_width; j++) {
1.148 + output_ptr[j] = ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
1.149 + (int)src_ptr[pixel_step] * vp9_filter[1],
1.150 + FILTER_BITS);
1.151 + src_ptr++;
1.152 + }
1.153 +
1.154 + src_ptr += src_pixels_per_line - output_width;
1.155 + output_ptr += output_width;
1.156 + }
1.157 +}
1.158 +
1.159 +unsigned int vp9_get_mb_ss_c(const int16_t *src_ptr) {
1.160 + unsigned int i, sum = 0;
1.161 +
1.162 + for (i = 0; i < 256; i++) {
1.163 + sum += (src_ptr[i] * src_ptr[i]);
1.164 + }
1.165 +
1.166 + return sum;
1.167 +}
1.168 +
1.169 +unsigned int vp9_variance64x32_c(const uint8_t *src_ptr,
1.170 + int source_stride,
1.171 + const uint8_t *ref_ptr,
1.172 + int recon_stride,
1.173 + unsigned int *sse) {
1.174 + unsigned int var;
1.175 + int avg;
1.176 +
1.177 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 64, 32, &var, &avg);
1.178 + *sse = var;
1.179 + return (var - (((int64_t)avg * avg) >> 11));
1.180 +}
1.181 +
1.182 +unsigned int vp9_sub_pixel_variance64x32_c(const uint8_t *src_ptr,
1.183 + int src_pixels_per_line,
1.184 + int xoffset,
1.185 + int yoffset,
1.186 + const uint8_t *dst_ptr,
1.187 + int dst_pixels_per_line,
1.188 + unsigned int *sse) {
1.189 + uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
1.190 + uint8_t temp2[68 * 64];
1.191 + const int16_t *hfilter, *vfilter;
1.192 +
1.193 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.194 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.195 +
1.196 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.197 + 1, 33, 64, hfilter);
1.198 + var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 32, 64, vfilter);
1.199 +
1.200 + return vp9_variance64x32(temp2, 64, dst_ptr, dst_pixels_per_line, sse);
1.201 +}
1.202 +
1.203 +unsigned int vp9_sub_pixel_avg_variance64x32_c(const uint8_t *src_ptr,
1.204 + int src_pixels_per_line,
1.205 + int xoffset,
1.206 + int yoffset,
1.207 + const uint8_t *dst_ptr,
1.208 + int dst_pixels_per_line,
1.209 + unsigned int *sse,
1.210 + const uint8_t *second_pred) {
1.211 + uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
1.212 + uint8_t temp2[68 * 64];
1.213 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64); // compound pred buffer
1.214 + const int16_t *hfilter, *vfilter;
1.215 +
1.216 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.217 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.218 +
1.219 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.220 + 1, 33, 64, hfilter);
1.221 + var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 32, 64, vfilter);
1.222 + comp_avg_pred(temp3, second_pred, 64, 32, temp2, 64);
1.223 + return vp9_variance64x32(temp3, 64, dst_ptr, dst_pixels_per_line, sse);
1.224 +}
1.225 +
1.226 +unsigned int vp9_variance32x64_c(const uint8_t *src_ptr,
1.227 + int source_stride,
1.228 + const uint8_t *ref_ptr,
1.229 + int recon_stride,
1.230 + unsigned int *sse) {
1.231 + unsigned int var;
1.232 + int avg;
1.233 +
1.234 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 32, 64, &var, &avg);
1.235 + *sse = var;
1.236 + return (var - (((int64_t)avg * avg) >> 11));
1.237 +}
1.238 +
1.239 +unsigned int vp9_sub_pixel_variance32x64_c(const uint8_t *src_ptr,
1.240 + int src_pixels_per_line,
1.241 + int xoffset,
1.242 + int yoffset,
1.243 + const uint8_t *dst_ptr,
1.244 + int dst_pixels_per_line,
1.245 + unsigned int *sse) {
1.246 + uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
1.247 + uint8_t temp2[68 * 64];
1.248 + const int16_t *hfilter, *vfilter;
1.249 +
1.250 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.251 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.252 +
1.253 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.254 + 1, 65, 32, hfilter);
1.255 + var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 64, 32, vfilter);
1.256 +
1.257 + return vp9_variance32x64(temp2, 32, dst_ptr, dst_pixels_per_line, sse);
1.258 +}
1.259 +
1.260 +unsigned int vp9_sub_pixel_avg_variance32x64_c(const uint8_t *src_ptr,
1.261 + int src_pixels_per_line,
1.262 + int xoffset,
1.263 + int yoffset,
1.264 + const uint8_t *dst_ptr,
1.265 + int dst_pixels_per_line,
1.266 + unsigned int *sse,
1.267 + const uint8_t *second_pred) {
1.268 + uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
1.269 + uint8_t temp2[68 * 64];
1.270 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 64); // compound pred buffer
1.271 + const int16_t *hfilter, *vfilter;
1.272 +
1.273 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.274 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.275 +
1.276 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.277 + 1, 65, 32, hfilter);
1.278 + var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 64, 32, vfilter);
1.279 + comp_avg_pred(temp3, second_pred, 32, 64, temp2, 32);
1.280 + return vp9_variance32x64(temp3, 32, dst_ptr, dst_pixels_per_line, sse);
1.281 +}
1.282 +
1.283 +unsigned int vp9_variance32x16_c(const uint8_t *src_ptr,
1.284 + int source_stride,
1.285 + const uint8_t *ref_ptr,
1.286 + int recon_stride,
1.287 + unsigned int *sse) {
1.288 + unsigned int var;
1.289 + int avg;
1.290 +
1.291 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 32, 16, &var, &avg);
1.292 + *sse = var;
1.293 + return (var - (((int64_t)avg * avg) >> 9));
1.294 +}
1.295 +
1.296 +unsigned int vp9_sub_pixel_variance32x16_c(const uint8_t *src_ptr,
1.297 + int src_pixels_per_line,
1.298 + int xoffset,
1.299 + int yoffset,
1.300 + const uint8_t *dst_ptr,
1.301 + int dst_pixels_per_line,
1.302 + unsigned int *sse) {
1.303 + uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
1.304 + uint8_t temp2[36 * 32];
1.305 + const int16_t *hfilter, *vfilter;
1.306 +
1.307 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.308 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.309 +
1.310 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.311 + 1, 17, 32, hfilter);
1.312 + var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 16, 32, vfilter);
1.313 +
1.314 + return vp9_variance32x16(temp2, 32, dst_ptr, dst_pixels_per_line, sse);
1.315 +}
1.316 +
1.317 +unsigned int vp9_sub_pixel_avg_variance32x16_c(const uint8_t *src_ptr,
1.318 + int src_pixels_per_line,
1.319 + int xoffset,
1.320 + int yoffset,
1.321 + const uint8_t *dst_ptr,
1.322 + int dst_pixels_per_line,
1.323 + unsigned int *sse,
1.324 + const uint8_t *second_pred) {
1.325 + uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
1.326 + uint8_t temp2[36 * 32];
1.327 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 16); // compound pred buffer
1.328 + const int16_t *hfilter, *vfilter;
1.329 +
1.330 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.331 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.332 +
1.333 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.334 + 1, 17, 32, hfilter);
1.335 + var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 16, 32, vfilter);
1.336 + comp_avg_pred(temp3, second_pred, 32, 16, temp2, 32);
1.337 + return vp9_variance32x16(temp3, 32, dst_ptr, dst_pixels_per_line, sse);
1.338 +}
1.339 +
1.340 +unsigned int vp9_variance16x32_c(const uint8_t *src_ptr,
1.341 + int source_stride,
1.342 + const uint8_t *ref_ptr,
1.343 + int recon_stride,
1.344 + unsigned int *sse) {
1.345 + unsigned int var;
1.346 + int avg;
1.347 +
1.348 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 32, &var, &avg);
1.349 + *sse = var;
1.350 + return (var - (((int64_t)avg * avg) >> 9));
1.351 +}
1.352 +
1.353 +unsigned int vp9_sub_pixel_variance16x32_c(const uint8_t *src_ptr,
1.354 + int src_pixels_per_line,
1.355 + int xoffset,
1.356 + int yoffset,
1.357 + const uint8_t *dst_ptr,
1.358 + int dst_pixels_per_line,
1.359 + unsigned int *sse) {
1.360 + uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
1.361 + uint8_t temp2[36 * 32];
1.362 + const int16_t *hfilter, *vfilter;
1.363 +
1.364 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.365 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.366 +
1.367 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.368 + 1, 33, 16, hfilter);
1.369 + var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 32, 16, vfilter);
1.370 +
1.371 + return vp9_variance16x32(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
1.372 +}
1.373 +
1.374 +unsigned int vp9_sub_pixel_avg_variance16x32_c(const uint8_t *src_ptr,
1.375 + int src_pixels_per_line,
1.376 + int xoffset,
1.377 + int yoffset,
1.378 + const uint8_t *dst_ptr,
1.379 + int dst_pixels_per_line,
1.380 + unsigned int *sse,
1.381 + const uint8_t *second_pred) {
1.382 + uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
1.383 + uint8_t temp2[36 * 32];
1.384 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 32); // compound pred buffer
1.385 + const int16_t *hfilter, *vfilter;
1.386 +
1.387 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.388 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.389 +
1.390 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.391 + 1, 33, 16, hfilter);
1.392 + var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 32, 16, vfilter);
1.393 + comp_avg_pred(temp3, second_pred, 16, 32, temp2, 16);
1.394 + return vp9_variance16x32(temp3, 16, dst_ptr, dst_pixels_per_line, sse);
1.395 +}
1.396 +
1.397 +unsigned int vp9_variance64x64_c(const uint8_t *src_ptr,
1.398 + int source_stride,
1.399 + const uint8_t *ref_ptr,
1.400 + int recon_stride,
1.401 + unsigned int *sse) {
1.402 + unsigned int var;
1.403 + int avg;
1.404 +
1.405 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 64, 64, &var, &avg);
1.406 + *sse = var;
1.407 + return (var - (((int64_t)avg * avg) >> 12));
1.408 +}
1.409 +
1.410 +unsigned int vp9_variance32x32_c(const uint8_t *src_ptr,
1.411 + int source_stride,
1.412 + const uint8_t *ref_ptr,
1.413 + int recon_stride,
1.414 + unsigned int *sse) {
1.415 + unsigned int var;
1.416 + int avg;
1.417 +
1.418 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 32, 32, &var, &avg);
1.419 + *sse = var;
1.420 + return (var - (((int64_t)avg * avg) >> 10));
1.421 +}
1.422 +
1.423 +unsigned int vp9_variance16x16_c(const uint8_t *src_ptr,
1.424 + int source_stride,
1.425 + const uint8_t *ref_ptr,
1.426 + int recon_stride,
1.427 + unsigned int *sse) {
1.428 + unsigned int var;
1.429 + int avg;
1.430 +
1.431 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
1.432 + *sse = var;
1.433 + return (var - (((unsigned int)avg * avg) >> 8));
1.434 +}
1.435 +
1.436 +unsigned int vp9_variance8x16_c(const uint8_t *src_ptr,
1.437 + int source_stride,
1.438 + const uint8_t *ref_ptr,
1.439 + int recon_stride,
1.440 + unsigned int *sse) {
1.441 + unsigned int var;
1.442 + int avg;
1.443 +
1.444 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16, &var, &avg);
1.445 + *sse = var;
1.446 + return (var - (((unsigned int)avg * avg) >> 7));
1.447 +}
1.448 +
1.449 +unsigned int vp9_variance16x8_c(const uint8_t *src_ptr,
1.450 + int source_stride,
1.451 + const uint8_t *ref_ptr,
1.452 + int recon_stride,
1.453 + unsigned int *sse) {
1.454 + unsigned int var;
1.455 + int avg;
1.456 +
1.457 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8, &var, &avg);
1.458 + *sse = var;
1.459 + return (var - (((unsigned int)avg * avg) >> 7));
1.460 +}
1.461 +
1.462 +void vp9_get_sse_sum_8x8_c(const uint8_t *src_ptr, int source_stride,
1.463 + const uint8_t *ref_ptr, int ref_stride,
1.464 + unsigned int *sse, int *sum) {
1.465 + variance(src_ptr, source_stride, ref_ptr, ref_stride, 8, 8, sse, sum);
1.466 +}
1.467 +
1.468 +unsigned int vp9_variance8x8_c(const uint8_t *src_ptr,
1.469 + int source_stride,
1.470 + const uint8_t *ref_ptr,
1.471 + int recon_stride,
1.472 + unsigned int *sse) {
1.473 + unsigned int var;
1.474 + int avg;
1.475 +
1.476 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, &var, &avg);
1.477 + *sse = var;
1.478 + return (var - (((unsigned int)avg * avg) >> 6));
1.479 +}
1.480 +
1.481 +unsigned int vp9_variance8x4_c(const uint8_t *src_ptr,
1.482 + int source_stride,
1.483 + const uint8_t *ref_ptr,
1.484 + int recon_stride,
1.485 + unsigned int *sse) {
1.486 + unsigned int var;
1.487 + int avg;
1.488 +
1.489 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 4, &var, &avg);
1.490 + *sse = var;
1.491 + return (var - (((unsigned int)avg * avg) >> 5));
1.492 +}
1.493 +
1.494 +unsigned int vp9_variance4x8_c(const uint8_t *src_ptr,
1.495 + int source_stride,
1.496 + const uint8_t *ref_ptr,
1.497 + int recon_stride,
1.498 + unsigned int *sse) {
1.499 + unsigned int var;
1.500 + int avg;
1.501 +
1.502 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 4, 8, &var, &avg);
1.503 + *sse = var;
1.504 + return (var - (((unsigned int)avg * avg) >> 5));
1.505 +}
1.506 +
1.507 +unsigned int vp9_variance4x4_c(const uint8_t *src_ptr,
1.508 + int source_stride,
1.509 + const uint8_t *ref_ptr,
1.510 + int recon_stride,
1.511 + unsigned int *sse) {
1.512 + unsigned int var;
1.513 + int avg;
1.514 +
1.515 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 4, 4, &var, &avg);
1.516 + *sse = var;
1.517 + return (var - (((unsigned int)avg * avg) >> 4));
1.518 +}
1.519 +
1.520 +
1.521 +unsigned int vp9_mse16x16_c(const uint8_t *src_ptr,
1.522 + int source_stride,
1.523 + const uint8_t *ref_ptr,
1.524 + int recon_stride,
1.525 + unsigned int *sse) {
1.526 + unsigned int var;
1.527 + int avg;
1.528 +
1.529 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
1.530 + *sse = var;
1.531 + return var;
1.532 +}
1.533 +
1.534 +unsigned int vp9_mse16x8_c(const uint8_t *src_ptr,
1.535 + int source_stride,
1.536 + const uint8_t *ref_ptr,
1.537 + int recon_stride,
1.538 + unsigned int *sse) {
1.539 + unsigned int var;
1.540 + int avg;
1.541 +
1.542 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8, &var, &avg);
1.543 + *sse = var;
1.544 + return var;
1.545 +}
1.546 +
1.547 +unsigned int vp9_mse8x16_c(const uint8_t *src_ptr,
1.548 + int source_stride,
1.549 + const uint8_t *ref_ptr,
1.550 + int recon_stride,
1.551 + unsigned int *sse) {
1.552 + unsigned int var;
1.553 + int avg;
1.554 +
1.555 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16, &var, &avg);
1.556 + *sse = var;
1.557 + return var;
1.558 +}
1.559 +
1.560 +unsigned int vp9_mse8x8_c(const uint8_t *src_ptr,
1.561 + int source_stride,
1.562 + const uint8_t *ref_ptr,
1.563 + int recon_stride,
1.564 + unsigned int *sse) {
1.565 + unsigned int var;
1.566 + int avg;
1.567 +
1.568 + variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, &var, &avg);
1.569 + *sse = var;
1.570 + return var;
1.571 +}
1.572 +
1.573 +
1.574 +unsigned int vp9_sub_pixel_variance4x4_c(const uint8_t *src_ptr,
1.575 + int src_pixels_per_line,
1.576 + int xoffset,
1.577 + int yoffset,
1.578 + const uint8_t *dst_ptr,
1.579 + int dst_pixels_per_line,
1.580 + unsigned int *sse) {
1.581 + uint8_t temp2[20 * 16];
1.582 + const int16_t *hfilter, *vfilter;
1.583 + uint16_t fdata3[5 * 4]; // Temp data buffer used in filtering
1.584 +
1.585 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.586 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.587 +
1.588 + // First filter 1d Horizontal
1.589 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.590 + 1, 5, 4, hfilter);
1.591 +
1.592 + // Now filter Verticaly
1.593 + var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, vfilter);
1.594 +
1.595 + return vp9_variance4x4(temp2, 4, dst_ptr, dst_pixels_per_line, sse);
1.596 +}
1.597 +
1.598 +unsigned int vp9_sub_pixel_avg_variance4x4_c(const uint8_t *src_ptr,
1.599 + int src_pixels_per_line,
1.600 + int xoffset,
1.601 + int yoffset,
1.602 + const uint8_t *dst_ptr,
1.603 + int dst_pixels_per_line,
1.604 + unsigned int *sse,
1.605 + const uint8_t *second_pred) {
1.606 + uint8_t temp2[20 * 16];
1.607 + const int16_t *hfilter, *vfilter;
1.608 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 4); // compound pred buffer
1.609 + uint16_t fdata3[5 * 4]; // Temp data buffer used in filtering
1.610 +
1.611 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.612 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.613 +
1.614 + // First filter 1d Horizontal
1.615 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.616 + 1, 5, 4, hfilter);
1.617 +
1.618 + // Now filter Verticaly
1.619 + var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, vfilter);
1.620 + comp_avg_pred(temp3, second_pred, 4, 4, temp2, 4);
1.621 + return vp9_variance4x4(temp3, 4, dst_ptr, dst_pixels_per_line, sse);
1.622 +}
1.623 +
1.624 +unsigned int vp9_sub_pixel_variance8x8_c(const uint8_t *src_ptr,
1.625 + int src_pixels_per_line,
1.626 + int xoffset,
1.627 + int yoffset,
1.628 + const uint8_t *dst_ptr,
1.629 + int dst_pixels_per_line,
1.630 + unsigned int *sse) {
1.631 + uint16_t fdata3[9 * 8]; // Temp data buffer used in filtering
1.632 + uint8_t temp2[20 * 16];
1.633 + const int16_t *hfilter, *vfilter;
1.634 +
1.635 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.636 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.637 +
1.638 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.639 + 1, 9, 8, hfilter);
1.640 + var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 8, 8, vfilter);
1.641 +
1.642 + return vp9_variance8x8(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
1.643 +}
1.644 +
1.645 +unsigned int vp9_sub_pixel_avg_variance8x8_c(const uint8_t *src_ptr,
1.646 + int src_pixels_per_line,
1.647 + int xoffset,
1.648 + int yoffset,
1.649 + const uint8_t *dst_ptr,
1.650 + int dst_pixels_per_line,
1.651 + unsigned int *sse,
1.652 + const uint8_t *second_pred) {
1.653 + uint16_t fdata3[9 * 8]; // Temp data buffer used in filtering
1.654 + uint8_t temp2[20 * 16];
1.655 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 8); // compound pred buffer
1.656 + const int16_t *hfilter, *vfilter;
1.657 +
1.658 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.659 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.660 +
1.661 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.662 + 1, 9, 8, hfilter);
1.663 + var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 8, 8, vfilter);
1.664 + comp_avg_pred(temp3, second_pred, 8, 8, temp2, 8);
1.665 + return vp9_variance8x8(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
1.666 +}
1.667 +
1.668 +unsigned int vp9_sub_pixel_variance16x16_c(const uint8_t *src_ptr,
1.669 + int src_pixels_per_line,
1.670 + int xoffset,
1.671 + int yoffset,
1.672 + const uint8_t *dst_ptr,
1.673 + int dst_pixels_per_line,
1.674 + unsigned int *sse) {
1.675 + uint16_t fdata3[17 * 16]; // Temp data buffer used in filtering
1.676 + uint8_t temp2[20 * 16];
1.677 + const int16_t *hfilter, *vfilter;
1.678 +
1.679 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.680 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.681 +
1.682 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.683 + 1, 17, 16, hfilter);
1.684 + var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 16, 16, vfilter);
1.685 +
1.686 + return vp9_variance16x16(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
1.687 +}
1.688 +
1.689 +unsigned int vp9_sub_pixel_avg_variance16x16_c(const uint8_t *src_ptr,
1.690 + int src_pixels_per_line,
1.691 + int xoffset,
1.692 + int yoffset,
1.693 + const uint8_t *dst_ptr,
1.694 + int dst_pixels_per_line,
1.695 + unsigned int *sse,
1.696 + const uint8_t *second_pred) {
1.697 + uint16_t fdata3[17 * 16];
1.698 + uint8_t temp2[20 * 16];
1.699 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 16); // compound pred buffer
1.700 + const int16_t *hfilter, *vfilter;
1.701 +
1.702 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.703 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.704 +
1.705 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.706 + 1, 17, 16, hfilter);
1.707 + var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 16, 16, vfilter);
1.708 +
1.709 + comp_avg_pred(temp3, second_pred, 16, 16, temp2, 16);
1.710 + return vp9_variance16x16(temp3, 16, dst_ptr, dst_pixels_per_line, sse);
1.711 +}
1.712 +
1.713 +unsigned int vp9_sub_pixel_variance64x64_c(const uint8_t *src_ptr,
1.714 + int src_pixels_per_line,
1.715 + int xoffset,
1.716 + int yoffset,
1.717 + const uint8_t *dst_ptr,
1.718 + int dst_pixels_per_line,
1.719 + unsigned int *sse) {
1.720 + uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
1.721 + uint8_t temp2[68 * 64];
1.722 + const int16_t *hfilter, *vfilter;
1.723 +
1.724 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.725 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.726 +
1.727 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.728 + 1, 65, 64, hfilter);
1.729 + var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 64, 64, vfilter);
1.730 +
1.731 + return vp9_variance64x64(temp2, 64, dst_ptr, dst_pixels_per_line, sse);
1.732 +}
1.733 +
1.734 +unsigned int vp9_sub_pixel_avg_variance64x64_c(const uint8_t *src_ptr,
1.735 + int src_pixels_per_line,
1.736 + int xoffset,
1.737 + int yoffset,
1.738 + const uint8_t *dst_ptr,
1.739 + int dst_pixels_per_line,
1.740 + unsigned int *sse,
1.741 + const uint8_t *second_pred) {
1.742 + uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
1.743 + uint8_t temp2[68 * 64];
1.744 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64); // compound pred buffer
1.745 + const int16_t *hfilter, *vfilter;
1.746 +
1.747 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.748 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.749 +
1.750 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.751 + 1, 65, 64, hfilter);
1.752 + var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 64, 64, vfilter);
1.753 + comp_avg_pred(temp3, second_pred, 64, 64, temp2, 64);
1.754 + return vp9_variance64x64(temp3, 64, dst_ptr, dst_pixels_per_line, sse);
1.755 +}
1.756 +
1.757 +unsigned int vp9_sub_pixel_variance32x32_c(const uint8_t *src_ptr,
1.758 + int src_pixels_per_line,
1.759 + int xoffset,
1.760 + int yoffset,
1.761 + const uint8_t *dst_ptr,
1.762 + int dst_pixels_per_line,
1.763 + unsigned int *sse) {
1.764 + uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
1.765 + uint8_t temp2[36 * 32];
1.766 + const int16_t *hfilter, *vfilter;
1.767 +
1.768 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.769 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.770 +
1.771 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.772 + 1, 33, 32, hfilter);
1.773 + var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 32, 32, vfilter);
1.774 +
1.775 + return vp9_variance32x32(temp2, 32, dst_ptr, dst_pixels_per_line, sse);
1.776 +}
1.777 +
1.778 +unsigned int vp9_sub_pixel_avg_variance32x32_c(const uint8_t *src_ptr,
1.779 + int src_pixels_per_line,
1.780 + int xoffset,
1.781 + int yoffset,
1.782 + const uint8_t *dst_ptr,
1.783 + int dst_pixels_per_line,
1.784 + unsigned int *sse,
1.785 + const uint8_t *second_pred) {
1.786 + uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
1.787 + uint8_t temp2[36 * 32];
1.788 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 32); // compound pred buffer
1.789 + const int16_t *hfilter, *vfilter;
1.790 +
1.791 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.792 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.793 +
1.794 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.795 + 1, 33, 32, hfilter);
1.796 + var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 32, 32, vfilter);
1.797 + comp_avg_pred(temp3, second_pred, 32, 32, temp2, 32);
1.798 + return vp9_variance32x32(temp3, 32, dst_ptr, dst_pixels_per_line, sse);
1.799 +}
1.800 +
1.801 +unsigned int vp9_variance_halfpixvar16x16_h_c(const uint8_t *src_ptr,
1.802 + int source_stride,
1.803 + const uint8_t *ref_ptr,
1.804 + int recon_stride,
1.805 + unsigned int *sse) {
1.806 + return vp9_sub_pixel_variance16x16_c(src_ptr, source_stride, 8, 0,
1.807 + ref_ptr, recon_stride, sse);
1.808 +}
1.809 +
1.810 +unsigned int vp9_variance_halfpixvar32x32_h_c(const uint8_t *src_ptr,
1.811 + int source_stride,
1.812 + const uint8_t *ref_ptr,
1.813 + int recon_stride,
1.814 + unsigned int *sse) {
1.815 + return vp9_sub_pixel_variance32x32_c(src_ptr, source_stride, 8, 0,
1.816 + ref_ptr, recon_stride, sse);
1.817 +}
1.818 +
1.819 +unsigned int vp9_variance_halfpixvar64x64_h_c(const uint8_t *src_ptr,
1.820 + int source_stride,
1.821 + const uint8_t *ref_ptr,
1.822 + int recon_stride,
1.823 + unsigned int *sse) {
1.824 + return vp9_sub_pixel_variance64x64_c(src_ptr, source_stride, 8, 0,
1.825 + ref_ptr, recon_stride, sse);
1.826 +}
1.827 +
1.828 +unsigned int vp9_variance_halfpixvar16x16_v_c(const uint8_t *src_ptr,
1.829 + int source_stride,
1.830 + const uint8_t *ref_ptr,
1.831 + int recon_stride,
1.832 + unsigned int *sse) {
1.833 + return vp9_sub_pixel_variance16x16_c(src_ptr, source_stride, 0, 8,
1.834 + ref_ptr, recon_stride, sse);
1.835 +}
1.836 +
1.837 +unsigned int vp9_variance_halfpixvar32x32_v_c(const uint8_t *src_ptr,
1.838 + int source_stride,
1.839 + const uint8_t *ref_ptr,
1.840 + int recon_stride,
1.841 + unsigned int *sse) {
1.842 + return vp9_sub_pixel_variance32x32_c(src_ptr, source_stride, 0, 8,
1.843 + ref_ptr, recon_stride, sse);
1.844 +}
1.845 +
1.846 +unsigned int vp9_variance_halfpixvar64x64_v_c(const uint8_t *src_ptr,
1.847 + int source_stride,
1.848 + const uint8_t *ref_ptr,
1.849 + int recon_stride,
1.850 + unsigned int *sse) {
1.851 + return vp9_sub_pixel_variance64x64_c(src_ptr, source_stride, 0, 8,
1.852 + ref_ptr, recon_stride, sse);
1.853 +}
1.854 +
1.855 +unsigned int vp9_variance_halfpixvar16x16_hv_c(const uint8_t *src_ptr,
1.856 + int source_stride,
1.857 + const uint8_t *ref_ptr,
1.858 + int recon_stride,
1.859 + unsigned int *sse) {
1.860 + return vp9_sub_pixel_variance16x16_c(src_ptr, source_stride, 8, 8,
1.861 + ref_ptr, recon_stride, sse);
1.862 +}
1.863 +
1.864 +unsigned int vp9_variance_halfpixvar32x32_hv_c(const uint8_t *src_ptr,
1.865 + int source_stride,
1.866 + const uint8_t *ref_ptr,
1.867 + int recon_stride,
1.868 + unsigned int *sse) {
1.869 + return vp9_sub_pixel_variance32x32_c(src_ptr, source_stride, 8, 8,
1.870 + ref_ptr, recon_stride, sse);
1.871 +}
1.872 +
1.873 +unsigned int vp9_variance_halfpixvar64x64_hv_c(const uint8_t *src_ptr,
1.874 + int source_stride,
1.875 + const uint8_t *ref_ptr,
1.876 + int recon_stride,
1.877 + unsigned int *sse) {
1.878 + return vp9_sub_pixel_variance64x64_c(src_ptr, source_stride, 8, 8,
1.879 + ref_ptr, recon_stride, sse);
1.880 +}
1.881 +
1.882 +unsigned int vp9_sub_pixel_mse16x16_c(const uint8_t *src_ptr,
1.883 + int src_pixels_per_line,
1.884 + int xoffset,
1.885 + int yoffset,
1.886 + const uint8_t *dst_ptr,
1.887 + int dst_pixels_per_line,
1.888 + unsigned int *sse) {
1.889 + vp9_sub_pixel_variance16x16_c(src_ptr, src_pixels_per_line,
1.890 + xoffset, yoffset, dst_ptr,
1.891 + dst_pixels_per_line, sse);
1.892 + return *sse;
1.893 +}
1.894 +
1.895 +unsigned int vp9_sub_pixel_mse32x32_c(const uint8_t *src_ptr,
1.896 + int src_pixels_per_line,
1.897 + int xoffset,
1.898 + int yoffset,
1.899 + const uint8_t *dst_ptr,
1.900 + int dst_pixels_per_line,
1.901 + unsigned int *sse) {
1.902 + vp9_sub_pixel_variance32x32_c(src_ptr, src_pixels_per_line,
1.903 + xoffset, yoffset, dst_ptr,
1.904 + dst_pixels_per_line, sse);
1.905 + return *sse;
1.906 +}
1.907 +
1.908 +unsigned int vp9_sub_pixel_mse64x64_c(const uint8_t *src_ptr,
1.909 + int src_pixels_per_line,
1.910 + int xoffset,
1.911 + int yoffset,
1.912 + const uint8_t *dst_ptr,
1.913 + int dst_pixels_per_line,
1.914 + unsigned int *sse) {
1.915 + vp9_sub_pixel_variance64x64_c(src_ptr, src_pixels_per_line,
1.916 + xoffset, yoffset, dst_ptr,
1.917 + dst_pixels_per_line, sse);
1.918 + return *sse;
1.919 +}
1.920 +
1.921 +unsigned int vp9_sub_pixel_variance16x8_c(const uint8_t *src_ptr,
1.922 + int src_pixels_per_line,
1.923 + int xoffset,
1.924 + int yoffset,
1.925 + const uint8_t *dst_ptr,
1.926 + int dst_pixels_per_line,
1.927 + unsigned int *sse) {
1.928 + uint16_t fdata3[16 * 9]; // Temp data buffer used in filtering
1.929 + uint8_t temp2[20 * 16];
1.930 + const int16_t *hfilter, *vfilter;
1.931 +
1.932 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.933 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.934 +
1.935 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.936 + 1, 9, 16, hfilter);
1.937 + var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 8, 16, vfilter);
1.938 +
1.939 + return vp9_variance16x8(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
1.940 +}
1.941 +
1.942 +unsigned int vp9_sub_pixel_avg_variance16x8_c(const uint8_t *src_ptr,
1.943 + int src_pixels_per_line,
1.944 + int xoffset,
1.945 + int yoffset,
1.946 + const uint8_t *dst_ptr,
1.947 + int dst_pixels_per_line,
1.948 + unsigned int *sse,
1.949 + const uint8_t *second_pred) {
1.950 + uint16_t fdata3[16 * 9]; // Temp data buffer used in filtering
1.951 + uint8_t temp2[20 * 16];
1.952 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 8); // compound pred buffer
1.953 + const int16_t *hfilter, *vfilter;
1.954 +
1.955 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.956 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.957 +
1.958 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.959 + 1, 9, 16, hfilter);
1.960 + var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 8, 16, vfilter);
1.961 + comp_avg_pred(temp3, second_pred, 16, 8, temp2, 16);
1.962 + return vp9_variance16x8(temp3, 16, dst_ptr, dst_pixels_per_line, sse);
1.963 +}
1.964 +
1.965 +unsigned int vp9_sub_pixel_variance8x16_c(const uint8_t *src_ptr,
1.966 + int src_pixels_per_line,
1.967 + int xoffset,
1.968 + int yoffset,
1.969 + const uint8_t *dst_ptr,
1.970 + int dst_pixels_per_line,
1.971 + unsigned int *sse) {
1.972 + uint16_t fdata3[9 * 16]; // Temp data buffer used in filtering
1.973 + uint8_t temp2[20 * 16];
1.974 + const int16_t *hfilter, *vfilter;
1.975 +
1.976 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.977 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.978 +
1.979 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.980 + 1, 17, 8, hfilter);
1.981 + var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 16, 8, vfilter);
1.982 +
1.983 + return vp9_variance8x16(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
1.984 +}
1.985 +
1.986 +unsigned int vp9_sub_pixel_avg_variance8x16_c(const uint8_t *src_ptr,
1.987 + int src_pixels_per_line,
1.988 + int xoffset,
1.989 + int yoffset,
1.990 + const uint8_t *dst_ptr,
1.991 + int dst_pixels_per_line,
1.992 + unsigned int *sse,
1.993 + const uint8_t *second_pred) {
1.994 + uint16_t fdata3[9 * 16]; // Temp data buffer used in filtering
1.995 + uint8_t temp2[20 * 16];
1.996 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 16); // compound pred buffer
1.997 + const int16_t *hfilter, *vfilter;
1.998 +
1.999 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.1000 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.1001 +
1.1002 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.1003 + 1, 17, 8, hfilter);
1.1004 + var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 16, 8, vfilter);
1.1005 + comp_avg_pred(temp3, second_pred, 8, 16, temp2, 8);
1.1006 + return vp9_variance8x16(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
1.1007 +}
1.1008 +
1.1009 +unsigned int vp9_sub_pixel_variance8x4_c(const uint8_t *src_ptr,
1.1010 + int src_pixels_per_line,
1.1011 + int xoffset,
1.1012 + int yoffset,
1.1013 + const uint8_t *dst_ptr,
1.1014 + int dst_pixels_per_line,
1.1015 + unsigned int *sse) {
1.1016 + uint16_t fdata3[8 * 5]; // Temp data buffer used in filtering
1.1017 + uint8_t temp2[20 * 16];
1.1018 + const int16_t *hfilter, *vfilter;
1.1019 +
1.1020 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.1021 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.1022 +
1.1023 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.1024 + 1, 5, 8, hfilter);
1.1025 + var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 4, 8, vfilter);
1.1026 +
1.1027 + return vp9_variance8x4(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
1.1028 +}
1.1029 +
1.1030 +unsigned int vp9_sub_pixel_avg_variance8x4_c(const uint8_t *src_ptr,
1.1031 + int src_pixels_per_line,
1.1032 + int xoffset,
1.1033 + int yoffset,
1.1034 + const uint8_t *dst_ptr,
1.1035 + int dst_pixels_per_line,
1.1036 + unsigned int *sse,
1.1037 + const uint8_t *second_pred) {
1.1038 + uint16_t fdata3[8 * 5]; // Temp data buffer used in filtering
1.1039 + uint8_t temp2[20 * 16];
1.1040 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 4); // compound pred buffer
1.1041 + const int16_t *hfilter, *vfilter;
1.1042 +
1.1043 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.1044 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.1045 +
1.1046 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.1047 + 1, 5, 8, hfilter);
1.1048 + var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 4, 8, vfilter);
1.1049 + comp_avg_pred(temp3, second_pred, 8, 4, temp2, 8);
1.1050 + return vp9_variance8x4(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
1.1051 +}
1.1052 +
1.1053 +unsigned int vp9_sub_pixel_variance4x8_c(const uint8_t *src_ptr,
1.1054 + int src_pixels_per_line,
1.1055 + int xoffset,
1.1056 + int yoffset,
1.1057 + const uint8_t *dst_ptr,
1.1058 + int dst_pixels_per_line,
1.1059 + unsigned int *sse) {
1.1060 + uint16_t fdata3[5 * 8]; // Temp data buffer used in filtering
1.1061 + // FIXME(jingning,rbultje): this temp2 buffer probably doesn't need to be
1.1062 + // of this big? same issue appears in all other block size settings.
1.1063 + uint8_t temp2[20 * 16];
1.1064 + const int16_t *hfilter, *vfilter;
1.1065 +
1.1066 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.1067 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.1068 +
1.1069 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.1070 + 1, 9, 4, hfilter);
1.1071 + var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 8, 4, vfilter);
1.1072 +
1.1073 + return vp9_variance4x8(temp2, 4, dst_ptr, dst_pixels_per_line, sse);
1.1074 +}
1.1075 +
1.1076 +unsigned int vp9_sub_pixel_avg_variance4x8_c(const uint8_t *src_ptr,
1.1077 + int src_pixels_per_line,
1.1078 + int xoffset,
1.1079 + int yoffset,
1.1080 + const uint8_t *dst_ptr,
1.1081 + int dst_pixels_per_line,
1.1082 + unsigned int *sse,
1.1083 + const uint8_t *second_pred) {
1.1084 + uint16_t fdata3[5 * 8]; // Temp data buffer used in filtering
1.1085 + uint8_t temp2[20 * 16];
1.1086 + DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 8); // compound pred buffer
1.1087 + const int16_t *hfilter, *vfilter;
1.1088 +
1.1089 + hfilter = BILINEAR_FILTERS_2TAP(xoffset);
1.1090 + vfilter = BILINEAR_FILTERS_2TAP(yoffset);
1.1091 +
1.1092 + var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1.1093 + 1, 9, 4, hfilter);
1.1094 + var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 8, 4, vfilter);
1.1095 + comp_avg_pred(temp3, second_pred, 4, 8, temp2, 4);
1.1096 + return vp9_variance4x8(temp3, 4, dst_ptr, dst_pixels_per_line, sse);
1.1097 +}
