1 /*
2 * Copyright (c) 2016 The WebM project authors. All Rights Reserved.
3 *
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.
9 */
10
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"
15
vpx_highbd_convolve8_neon(const uint16_t * src,ptrdiff_t src_stride,uint16_t * dst,ptrdiff_t dst_stride,const InterpKernel * filter,int x0_q4,int x_step_q4,int y0_q4,int y_step_q4,int w,int h,int bd)16 void vpx_highbd_convolve8_neon(const uint16_t *src, ptrdiff_t src_stride,
17 uint16_t *dst, ptrdiff_t dst_stride,
18 const InterpKernel *filter, int x0_q4,
19 int x_step_q4, int y0_q4, int y_step_q4, int w,
20 int h, int bd) {
21 // + 1 to make it divisible by 4
22 uint16_t temp[64 * 136];
23 const int intermediate_height =
24 (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
25
26 /* Filter starting 3 lines back. The neon implementation will ignore the given
27 * height and filter a multiple of 4 lines. Since this goes in to the temp
28 * buffer which has lots of extra room and is subsequently discarded this is
29 * safe if somewhat less than ideal. */
30 vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
31 filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w,
32 intermediate_height, bd);
33
34 /* Step into the temp buffer 3 lines to get the actual frame data */
35 vpx_highbd_convolve8_vert_neon(temp + w * 3, w, dst, dst_stride, filter,
36 x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd);
37 }
38
vpx_highbd_convolve8_avg_neon(const uint16_t * src,ptrdiff_t src_stride,uint16_t * dst,ptrdiff_t dst_stride,const InterpKernel * filter,int x0_q4,int x_step_q4,int y0_q4,int y_step_q4,int w,int h,int bd)39 void vpx_highbd_convolve8_avg_neon(const uint16_t *src, ptrdiff_t src_stride,
40 uint16_t *dst, ptrdiff_t dst_stride,
41 const InterpKernel *filter, int x0_q4,
42 int x_step_q4, int y0_q4, int y_step_q4,
43 int w, int h, int bd) {
44 // + 1 to make it divisible by 4
45 uint16_t temp[64 * 136];
46 const int intermediate_height =
47 (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
48
49 /* This implementation has the same issues as above. In addition, we only want
50 * to average the values after both passes.
51 */
52 vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
53 filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w,
54 intermediate_height, bd);
55 vpx_highbd_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter,
56 x0_q4, x_step_q4, y0_q4, y_step_q4, w, h,
57 bd);
58 }
59