Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 1 | /* |
Yaowu Xu | 9c01aa1 | 2016-09-01 14:32:49 -0700 | [diff] [blame] | 2 | * Copyright (c) 2016, Alliance for Open Media. All rights reserved |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 3 | * |
Yaowu Xu | 9c01aa1 | 2016-09-01 14:32:49 -0700 | [diff] [blame] | 4 | * This source code is subject to the terms of the BSD 2 Clause License and |
| 5 | * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License |
| 6 | * was not distributed with this source code in the LICENSE file, you can |
| 7 | * obtain it at www.aomedia.org/license/software. If the Alliance for Open |
| 8 | * Media Patent License 1.0 was not distributed with this source code in the |
| 9 | * PATENTS file, you can obtain it at www.aomedia.org/license/patent. |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 10 | */ |
| 11 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 12 | #include "./aom_config.h" |
| 13 | #include "./aom_dsp_rtcd.h" |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 14 | #include "aom_dsp/x86/convolve.h" |
| 15 | |
| 16 | #if HAVE_SSE2 |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 17 | filter8_1dfunction aom_filter_block1d16_v8_sse2; |
| 18 | filter8_1dfunction aom_filter_block1d16_h8_sse2; |
| 19 | filter8_1dfunction aom_filter_block1d8_v8_sse2; |
| 20 | filter8_1dfunction aom_filter_block1d8_h8_sse2; |
| 21 | filter8_1dfunction aom_filter_block1d4_v8_sse2; |
| 22 | filter8_1dfunction aom_filter_block1d4_h8_sse2; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 23 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 24 | filter8_1dfunction aom_filter_block1d16_v2_sse2; |
| 25 | filter8_1dfunction aom_filter_block1d16_h2_sse2; |
| 26 | filter8_1dfunction aom_filter_block1d8_v2_sse2; |
| 27 | filter8_1dfunction aom_filter_block1d8_h2_sse2; |
| 28 | filter8_1dfunction aom_filter_block1d4_v2_sse2; |
| 29 | filter8_1dfunction aom_filter_block1d4_h2_sse2; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 30 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 31 | // void aom_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 32 | // uint8_t *dst, ptrdiff_t dst_stride, |
| 33 | // const int16_t *filter_x, int x_step_q4, |
| 34 | // const int16_t *filter_y, int y_step_q4, |
| 35 | // int w, int h); |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 36 | // void aom_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 37 | // uint8_t *dst, ptrdiff_t dst_stride, |
| 38 | // const int16_t *filter_x, int x_step_q4, |
| 39 | // const int16_t *filter_y, int y_step_q4, |
| 40 | // int w, int h); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 41 | FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2); |
| 42 | FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 43 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 44 | // void aom_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 45 | // uint8_t *dst, ptrdiff_t dst_stride, |
| 46 | // const int16_t *filter_x, int x_step_q4, |
| 47 | // const int16_t *filter_y, int y_step_q4, |
| 48 | // int w, int h); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 49 | FUN_CONV_2D(, sse2); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 50 | |
Yaowu Xu | d3e7c68 | 2017-12-21 14:08:25 -0800 | [diff] [blame] | 51 | #if ARCH_X86_64 |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 52 | highbd_filter8_1dfunction aom_highbd_filter_block1d16_v8_sse2; |
| 53 | highbd_filter8_1dfunction aom_highbd_filter_block1d16_h8_sse2; |
| 54 | highbd_filter8_1dfunction aom_highbd_filter_block1d8_v8_sse2; |
| 55 | highbd_filter8_1dfunction aom_highbd_filter_block1d8_h8_sse2; |
| 56 | highbd_filter8_1dfunction aom_highbd_filter_block1d4_v8_sse2; |
| 57 | highbd_filter8_1dfunction aom_highbd_filter_block1d4_h8_sse2; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 58 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 59 | highbd_filter8_1dfunction aom_highbd_filter_block1d16_v2_sse2; |
| 60 | highbd_filter8_1dfunction aom_highbd_filter_block1d16_h2_sse2; |
| 61 | highbd_filter8_1dfunction aom_highbd_filter_block1d8_v2_sse2; |
| 62 | highbd_filter8_1dfunction aom_highbd_filter_block1d8_h2_sse2; |
| 63 | highbd_filter8_1dfunction aom_highbd_filter_block1d4_v2_sse2; |
| 64 | highbd_filter8_1dfunction aom_highbd_filter_block1d4_h2_sse2; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 65 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 66 | // void aom_highbd_convolve8_horiz_sse2(const uint8_t *src, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 67 | // ptrdiff_t src_stride, |
| 68 | // uint8_t *dst, |
| 69 | // ptrdiff_t dst_stride, |
| 70 | // const int16_t *filter_x, |
| 71 | // int x_step_q4, |
| 72 | // const int16_t *filter_y, |
| 73 | // int y_step_q4, |
| 74 | // int w, int h, int bd); |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 75 | // void aom_highbd_convolve8_vert_sse2(const uint8_t *src, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 76 | // ptrdiff_t src_stride, |
| 77 | // uint8_t *dst, |
| 78 | // ptrdiff_t dst_stride, |
| 79 | // const int16_t *filter_x, |
| 80 | // int x_step_q4, |
| 81 | // const int16_t *filter_y, |
| 82 | // int y_step_q4, |
| 83 | // int w, int h, int bd); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 84 | HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2); |
| 85 | HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 86 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 87 | // void aom_highbd_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 88 | // uint8_t *dst, ptrdiff_t dst_stride, |
| 89 | // const int16_t *filter_x, int x_step_q4, |
| 90 | // const int16_t *filter_y, int y_step_q4, |
| 91 | // int w, int h, int bd); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 92 | HIGH_FUN_CONV_2D(, sse2); |
David Barker | be6cc07 | 2016-12-15 15:39:10 +0000 | [diff] [blame] | 93 | |
Yaowu Xu | d3e7c68 | 2017-12-21 14:08:25 -0800 | [diff] [blame] | 94 | #endif // ARCH_X86_64 |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 95 | #endif // HAVE_SSE2 |