2 * Copyright (c) 2016 The WebM project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
11 #include "./vpx_dsp_rtcd.h"
12 #include "vpx_dsp/vpx_dsp_common.h"
13 #include "vpx_dsp/vpx_filter.h"
14 #include "vpx_ports/mem.h"
16 void vpx_highbd_convolve8_neon(const uint16_t *src, ptrdiff_t src_stride,
17 uint16_t *dst, ptrdiff_t dst_stride,
18 const int16_t *filter_x, int x_step_q4,
19 const int16_t *filter_y, int y_step_q4, int w,
21 const int y0_q4 = get_filter_offset(filter_y, get_filter_base(filter_y));
22 // + 1 to make it divisible by 4
23 DECLARE_ALIGNED(16, uint16_t, temp[64 * 136]);
24 const int intermediate_height =
25 (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
27 /* Filter starting 3 lines back. The neon implementation will ignore the given
28 * height and filter a multiple of 4 lines. Since this goes in to the temp
29 * buffer which has lots of extra room and is subsequently discarded this is
30 * safe if somewhat less than ideal. */
31 vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
32 filter_x, x_step_q4, filter_y, y_step_q4, w,
33 intermediate_height, bd);
35 /* Step into the temp buffer 3 lines to get the actual frame data */
36 vpx_highbd_convolve8_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x,
37 x_step_q4, filter_y, y_step_q4, w, h, bd);
40 void vpx_highbd_convolve8_avg_neon(const uint16_t *src, ptrdiff_t src_stride,
41 uint16_t *dst, ptrdiff_t dst_stride,
42 const int16_t *filter_x, int x_step_q4,
43 const int16_t *filter_y, int y_step_q4,
44 int w, int h, int bd) {
45 const int y0_q4 = get_filter_offset(filter_y, get_filter_base(filter_y));
46 // + 1 to make it divisible by 4
47 DECLARE_ALIGNED(16, uint16_t, temp[64 * 136]);
48 const int intermediate_height =
49 (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
51 /* This implementation has the same issues as above. In addition, we only want
52 * to average the values after both passes.
54 vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
55 filter_x, x_step_q4, filter_y, y_step_q4, w,
56 intermediate_height, bd);
57 vpx_highbd_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x,
58 x_step_q4, filter_y, y_step_q4, w, h, bd);