1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/common/x86/vp9_asm_stubs.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,468 @@
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 <assert.h>
1.15 +
1.16 +#include "./vpx_config.h"
1.17 +#include "./vp9_rtcd.h"
1.18 +#include "vpx_ports/mem.h"
1.19 +///////////////////////////////////////////////////////////////////////////
1.20 +// the mmx function that does the bilinear filtering and var calculation //
1.21 +// int one pass //
1.22 +///////////////////////////////////////////////////////////////////////////
1.23 +DECLARE_ALIGNED(16, const short, vp9_bilinear_filters_mmx[16][8]) = {
1.24 + { 128, 128, 128, 128, 0, 0, 0, 0 },
1.25 + { 120, 120, 120, 120, 8, 8, 8, 8 },
1.26 + { 112, 112, 112, 112, 16, 16, 16, 16 },
1.27 + { 104, 104, 104, 104, 24, 24, 24, 24 },
1.28 + { 96, 96, 96, 96, 32, 32, 32, 32 },
1.29 + { 88, 88, 88, 88, 40, 40, 40, 40 },
1.30 + { 80, 80, 80, 80, 48, 48, 48, 48 },
1.31 + { 72, 72, 72, 72, 56, 56, 56, 56 },
1.32 + { 64, 64, 64, 64, 64, 64, 64, 64 },
1.33 + { 56, 56, 56, 56, 72, 72, 72, 72 },
1.34 + { 48, 48, 48, 48, 80, 80, 80, 80 },
1.35 + { 40, 40, 40, 40, 88, 88, 88, 88 },
1.36 + { 32, 32, 32, 32, 96, 96, 96, 96 },
1.37 + { 24, 24, 24, 24, 104, 104, 104, 104 },
1.38 + { 16, 16, 16, 16, 112, 112, 112, 112 },
1.39 + { 8, 8, 8, 8, 120, 120, 120, 120 }
1.40 +};
1.41 +
1.42 +typedef void filter8_1dfunction (
1.43 + const unsigned char *src_ptr,
1.44 + const unsigned int src_pitch,
1.45 + unsigned char *output_ptr,
1.46 + unsigned int out_pitch,
1.47 + unsigned int output_height,
1.48 + const short *filter
1.49 +);
1.50 +
1.51 +#if HAVE_SSSE3
1.52 +filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
1.53 +filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
1.54 +filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
1.55 +filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
1.56 +filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
1.57 +filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
1.58 +filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
1.59 +filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
1.60 +filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
1.61 +filter8_1dfunction vp9_filter_block1d8_h8_avg_ssse3;
1.62 +filter8_1dfunction vp9_filter_block1d4_v8_avg_ssse3;
1.63 +filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
1.64 +
1.65 +void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
1.66 + uint8_t *dst, ptrdiff_t dst_stride,
1.67 + const int16_t *filter_x, int x_step_q4,
1.68 + const int16_t *filter_y, int y_step_q4,
1.69 + int w, int h) {
1.70 + /* Ensure the filter can be compressed to int16_t. */
1.71 + if (x_step_q4 == 16 && filter_x[3] != 128) {
1.72 + while (w >= 16) {
1.73 + vp9_filter_block1d16_h8_ssse3(src, src_stride,
1.74 + dst, dst_stride,
1.75 + h, filter_x);
1.76 + src += 16;
1.77 + dst += 16;
1.78 + w -= 16;
1.79 + }
1.80 + while (w >= 8) {
1.81 + vp9_filter_block1d8_h8_ssse3(src, src_stride,
1.82 + dst, dst_stride,
1.83 + h, filter_x);
1.84 + src += 8;
1.85 + dst += 8;
1.86 + w -= 8;
1.87 + }
1.88 + while (w >= 4) {
1.89 + vp9_filter_block1d4_h8_ssse3(src, src_stride,
1.90 + dst, dst_stride,
1.91 + h, filter_x);
1.92 + src += 4;
1.93 + dst += 4;
1.94 + w -= 4;
1.95 + }
1.96 + }
1.97 + if (w) {
1.98 + vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
1.99 + filter_x, x_step_q4, filter_y, y_step_q4,
1.100 + w, h);
1.101 + }
1.102 +}
1.103 +
1.104 +void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
1.105 + uint8_t *dst, ptrdiff_t dst_stride,
1.106 + const int16_t *filter_x, int x_step_q4,
1.107 + const int16_t *filter_y, int y_step_q4,
1.108 + int w, int h) {
1.109 + if (y_step_q4 == 16 && filter_y[3] != 128) {
1.110 + while (w >= 16) {
1.111 + vp9_filter_block1d16_v8_ssse3(src - src_stride * 3, src_stride,
1.112 + dst, dst_stride,
1.113 + h, filter_y);
1.114 + src += 16;
1.115 + dst += 16;
1.116 + w -= 16;
1.117 + }
1.118 + while (w >= 8) {
1.119 + vp9_filter_block1d8_v8_ssse3(src - src_stride * 3, src_stride,
1.120 + dst, dst_stride,
1.121 + h, filter_y);
1.122 + src += 8;
1.123 + dst += 8;
1.124 + w -= 8;
1.125 + }
1.126 + while (w >= 4) {
1.127 + vp9_filter_block1d4_v8_ssse3(src - src_stride * 3, src_stride,
1.128 + dst, dst_stride,
1.129 + h, filter_y);
1.130 + src += 4;
1.131 + dst += 4;
1.132 + w -= 4;
1.133 + }
1.134 + }
1.135 + if (w) {
1.136 + vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
1.137 + filter_x, x_step_q4, filter_y, y_step_q4,
1.138 + w, h);
1.139 + }
1.140 +}
1.141 +
1.142 +void vp9_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
1.143 + uint8_t *dst, ptrdiff_t dst_stride,
1.144 + const int16_t *filter_x, int x_step_q4,
1.145 + const int16_t *filter_y, int y_step_q4,
1.146 + int w, int h) {
1.147 + if (x_step_q4 == 16 && filter_x[3] != 128) {
1.148 + while (w >= 16) {
1.149 + vp9_filter_block1d16_h8_avg_ssse3(src, src_stride,
1.150 + dst, dst_stride,
1.151 + h, filter_x);
1.152 + src += 16;
1.153 + dst += 16;
1.154 + w -= 16;
1.155 + }
1.156 + while (w >= 8) {
1.157 + vp9_filter_block1d8_h8_avg_ssse3(src, src_stride,
1.158 + dst, dst_stride,
1.159 + h, filter_x);
1.160 + src += 8;
1.161 + dst += 8;
1.162 + w -= 8;
1.163 + }
1.164 + while (w >= 4) {
1.165 + vp9_filter_block1d4_h8_avg_ssse3(src, src_stride,
1.166 + dst, dst_stride,
1.167 + h, filter_x);
1.168 + src += 4;
1.169 + dst += 4;
1.170 + w -= 4;
1.171 + }
1.172 + }
1.173 + if (w) {
1.174 + vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
1.175 + filter_x, x_step_q4, filter_y, y_step_q4,
1.176 + w, h);
1.177 + }
1.178 +}
1.179 +
1.180 +void vp9_convolve8_avg_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
1.181 + uint8_t *dst, ptrdiff_t dst_stride,
1.182 + const int16_t *filter_x, int x_step_q4,
1.183 + const int16_t *filter_y, int y_step_q4,
1.184 + int w, int h) {
1.185 + if (y_step_q4 == 16 && filter_y[3] != 128) {
1.186 + while (w >= 16) {
1.187 + vp9_filter_block1d16_v8_avg_ssse3(src - src_stride * 3, src_stride,
1.188 + dst, dst_stride,
1.189 + h, filter_y);
1.190 + src += 16;
1.191 + dst += 16;
1.192 + w -= 16;
1.193 + }
1.194 + while (w >= 8) {
1.195 + vp9_filter_block1d8_v8_avg_ssse3(src - src_stride * 3, src_stride,
1.196 + dst, dst_stride,
1.197 + h, filter_y);
1.198 + src += 8;
1.199 + dst += 8;
1.200 + w -= 8;
1.201 + }
1.202 + while (w >= 4) {
1.203 + vp9_filter_block1d4_v8_avg_ssse3(src - src_stride * 3, src_stride,
1.204 + dst, dst_stride,
1.205 + h, filter_y);
1.206 + src += 4;
1.207 + dst += 4;
1.208 + w -= 4;
1.209 + }
1.210 + }
1.211 + if (w) {
1.212 + vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
1.213 + filter_x, x_step_q4, filter_y, y_step_q4,
1.214 + w, h);
1.215 + }
1.216 +}
1.217 +
1.218 +void vp9_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
1.219 + uint8_t *dst, ptrdiff_t dst_stride,
1.220 + const int16_t *filter_x, int x_step_q4,
1.221 + const int16_t *filter_y, int y_step_q4,
1.222 + int w, int h) {
1.223 + DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 71);
1.224 +
1.225 + assert(w <= 64);
1.226 + assert(h <= 64);
1.227 + if (x_step_q4 == 16 && y_step_q4 == 16) {
1.228 + vp9_convolve8_horiz_ssse3(src - 3 * src_stride, src_stride, fdata2, 64,
1.229 + filter_x, x_step_q4, filter_y, y_step_q4,
1.230 + w, h + 7);
1.231 + vp9_convolve8_vert_ssse3(fdata2 + 3 * 64, 64, dst, dst_stride,
1.232 + filter_x, x_step_q4, filter_y, y_step_q4, w, h);
1.233 + } else {
1.234 + vp9_convolve8_c(src, src_stride, dst, dst_stride,
1.235 + filter_x, x_step_q4, filter_y, y_step_q4, w, h);
1.236 + }
1.237 +}
1.238 +
1.239 +void vp9_convolve8_avg_ssse3(const uint8_t *src, ptrdiff_t src_stride,
1.240 + uint8_t *dst, ptrdiff_t dst_stride,
1.241 + const int16_t *filter_x, int x_step_q4,
1.242 + const int16_t *filter_y, int y_step_q4,
1.243 + int w, int h) {
1.244 + DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 71);
1.245 +
1.246 + assert(w <= 64);
1.247 + assert(h <= 64);
1.248 + if (x_step_q4 == 16 && y_step_q4 == 16) {
1.249 + vp9_convolve8_horiz_ssse3(src - 3 * src_stride, src_stride, fdata2, 64,
1.250 + filter_x, x_step_q4, filter_y, y_step_q4,
1.251 + w, h + 7);
1.252 + vp9_convolve8_avg_vert_ssse3(fdata2 + 3 * 64, 64, dst, dst_stride,
1.253 + filter_x, x_step_q4, filter_y, y_step_q4,
1.254 + w, h);
1.255 + } else {
1.256 + vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
1.257 + filter_x, x_step_q4, filter_y, y_step_q4, w, h);
1.258 + }
1.259 +}
1.260 +#endif
1.261 +
1.262 +#if HAVE_SSE2
1.263 +filter8_1dfunction vp9_filter_block1d16_v8_sse2;
1.264 +filter8_1dfunction vp9_filter_block1d16_h8_sse2;
1.265 +filter8_1dfunction vp9_filter_block1d8_v8_sse2;
1.266 +filter8_1dfunction vp9_filter_block1d8_h8_sse2;
1.267 +filter8_1dfunction vp9_filter_block1d4_v8_sse2;
1.268 +filter8_1dfunction vp9_filter_block1d4_h8_sse2;
1.269 +filter8_1dfunction vp9_filter_block1d16_v8_avg_sse2;
1.270 +filter8_1dfunction vp9_filter_block1d16_h8_avg_sse2;
1.271 +filter8_1dfunction vp9_filter_block1d8_v8_avg_sse2;
1.272 +filter8_1dfunction vp9_filter_block1d8_h8_avg_sse2;
1.273 +filter8_1dfunction vp9_filter_block1d4_v8_avg_sse2;
1.274 +filter8_1dfunction vp9_filter_block1d4_h8_avg_sse2;
1.275 +
1.276 +void vp9_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
1.277 + uint8_t *dst, ptrdiff_t dst_stride,
1.278 + const int16_t *filter_x, int x_step_q4,
1.279 + const int16_t *filter_y, int y_step_q4,
1.280 + int w, int h) {
1.281 + /* Ensure the filter can be compressed to int16_t. */
1.282 + if (x_step_q4 == 16 && filter_x[3] != 128) {
1.283 + while (w >= 16) {
1.284 + vp9_filter_block1d16_h8_sse2(src, src_stride,
1.285 + dst, dst_stride,
1.286 + h, filter_x);
1.287 + src += 16;
1.288 + dst += 16;
1.289 + w -= 16;
1.290 + }
1.291 + while (w >= 8) {
1.292 + vp9_filter_block1d8_h8_sse2(src, src_stride,
1.293 + dst, dst_stride,
1.294 + h, filter_x);
1.295 + src += 8;
1.296 + dst += 8;
1.297 + w -= 8;
1.298 + }
1.299 + while (w >= 4) {
1.300 + vp9_filter_block1d4_h8_sse2(src, src_stride,
1.301 + dst, dst_stride,
1.302 + h, filter_x);
1.303 + src += 4;
1.304 + dst += 4;
1.305 + w -= 4;
1.306 + }
1.307 + }
1.308 + if (w) {
1.309 + vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
1.310 + filter_x, x_step_q4, filter_y, y_step_q4,
1.311 + w, h);
1.312 + }
1.313 +}
1.314 +
1.315 +void vp9_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
1.316 + uint8_t *dst, ptrdiff_t dst_stride,
1.317 + const int16_t *filter_x, int x_step_q4,
1.318 + const int16_t *filter_y, int y_step_q4,
1.319 + int w, int h) {
1.320 + if (y_step_q4 == 16 && filter_y[3] != 128) {
1.321 + while (w >= 16) {
1.322 + vp9_filter_block1d16_v8_sse2(src - src_stride * 3, src_stride,
1.323 + dst, dst_stride,
1.324 + h, filter_y);
1.325 + src += 16;
1.326 + dst += 16;
1.327 + w -= 16;
1.328 + }
1.329 + while (w >= 8) {
1.330 + vp9_filter_block1d8_v8_sse2(src - src_stride * 3, src_stride,
1.331 + dst, dst_stride,
1.332 + h, filter_y);
1.333 + src += 8;
1.334 + dst += 8;
1.335 + w -= 8;
1.336 + }
1.337 + while (w >= 4) {
1.338 + vp9_filter_block1d4_v8_sse2(src - src_stride * 3, src_stride,
1.339 + dst, dst_stride,
1.340 + h, filter_y);
1.341 + src += 4;
1.342 + dst += 4;
1.343 + w -= 4;
1.344 + }
1.345 + }
1.346 + if (w) {
1.347 + vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
1.348 + filter_x, x_step_q4, filter_y, y_step_q4,
1.349 + w, h);
1.350 + }
1.351 +}
1.352 +
1.353 +void vp9_convolve8_avg_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
1.354 + uint8_t *dst, ptrdiff_t dst_stride,
1.355 + const int16_t *filter_x, int x_step_q4,
1.356 + const int16_t *filter_y, int y_step_q4,
1.357 + int w, int h) {
1.358 + if (x_step_q4 == 16 && filter_x[3] != 128) {
1.359 + while (w >= 16) {
1.360 + vp9_filter_block1d16_h8_avg_sse2(src, src_stride,
1.361 + dst, dst_stride,
1.362 + h, filter_x);
1.363 + src += 16;
1.364 + dst += 16;
1.365 + w -= 16;
1.366 + }
1.367 + while (w >= 8) {
1.368 + vp9_filter_block1d8_h8_avg_sse2(src, src_stride,
1.369 + dst, dst_stride,
1.370 + h, filter_x);
1.371 + src += 8;
1.372 + dst += 8;
1.373 + w -= 8;
1.374 + }
1.375 + while (w >= 4) {
1.376 + vp9_filter_block1d4_h8_avg_sse2(src, src_stride,
1.377 + dst, dst_stride,
1.378 + h, filter_x);
1.379 + src += 4;
1.380 + dst += 4;
1.381 + w -= 4;
1.382 + }
1.383 + }
1.384 + if (w) {
1.385 + vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
1.386 + filter_x, x_step_q4, filter_y, y_step_q4,
1.387 + w, h);
1.388 + }
1.389 +}
1.390 +
1.391 +void vp9_convolve8_avg_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
1.392 + uint8_t *dst, ptrdiff_t dst_stride,
1.393 + const int16_t *filter_x, int x_step_q4,
1.394 + const int16_t *filter_y, int y_step_q4,
1.395 + int w, int h) {
1.396 + if (y_step_q4 == 16 && filter_y[3] != 128) {
1.397 + while (w >= 16) {
1.398 + vp9_filter_block1d16_v8_avg_sse2(src - src_stride * 3, src_stride,
1.399 + dst, dst_stride,
1.400 + h, filter_y);
1.401 + src += 16;
1.402 + dst += 16;
1.403 + w -= 16;
1.404 + }
1.405 + while (w >= 8) {
1.406 + vp9_filter_block1d8_v8_avg_sse2(src - src_stride * 3, src_stride,
1.407 + dst, dst_stride,
1.408 + h, filter_y);
1.409 + src += 8;
1.410 + dst += 8;
1.411 + w -= 8;
1.412 + }
1.413 + while (w >= 4) {
1.414 + vp9_filter_block1d4_v8_avg_sse2(src - src_stride * 3, src_stride,
1.415 + dst, dst_stride,
1.416 + h, filter_y);
1.417 + src += 4;
1.418 + dst += 4;
1.419 + w -= 4;
1.420 + }
1.421 + }
1.422 + if (w) {
1.423 + vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
1.424 + filter_x, x_step_q4, filter_y, y_step_q4,
1.425 + w, h);
1.426 + }
1.427 +}
1.428 +
1.429 +void vp9_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
1.430 + uint8_t *dst, ptrdiff_t dst_stride,
1.431 + const int16_t *filter_x, int x_step_q4,
1.432 + const int16_t *filter_y, int y_step_q4,
1.433 + int w, int h) {
1.434 + DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 71);
1.435 +
1.436 + assert(w <= 64);
1.437 + assert(h <= 64);
1.438 + if (x_step_q4 == 16 && y_step_q4 == 16) {
1.439 + vp9_convolve8_horiz_sse2(src - 3 * src_stride, src_stride, fdata2, 64,
1.440 + filter_x, x_step_q4, filter_y, y_step_q4,
1.441 + w, h + 7);
1.442 + vp9_convolve8_vert_sse2(fdata2 + 3 * 64, 64, dst, dst_stride,
1.443 + filter_x, x_step_q4, filter_y, y_step_q4, w, h);
1.444 + } else {
1.445 + vp9_convolve8_c(src, src_stride, dst, dst_stride,
1.446 + filter_x, x_step_q4, filter_y, y_step_q4, w, h);
1.447 + }
1.448 +}
1.449 +
1.450 +void vp9_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
1.451 + uint8_t *dst, ptrdiff_t dst_stride,
1.452 + const int16_t *filter_x, int x_step_q4,
1.453 + const int16_t *filter_y, int y_step_q4,
1.454 + int w, int h) {
1.455 + DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 71);
1.456 +
1.457 + assert(w <= 64);
1.458 + assert(h <= 64);
1.459 + if (x_step_q4 == 16 && y_step_q4 == 16) {
1.460 + vp9_convolve8_horiz_sse2(src - 3 * src_stride, src_stride, fdata2, 64,
1.461 + filter_x, x_step_q4, filter_y, y_step_q4,
1.462 + w, h + 7);
1.463 + vp9_convolve8_avg_vert_sse2(fdata2 + 3 * 64, 64, dst, dst_stride,
1.464 + filter_x, x_step_q4, filter_y, y_step_q4,
1.465 + w, h);
1.466 + } else {
1.467 + vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
1.468 + filter_x, x_step_q4, filter_y, y_step_q4, w, h);
1.469 + }
1.470 +}
1.471 +#endif
