2013-07-16 21:13:06 +04:00
|
|
|
/*
|
2016-09-02 00:32:49 +03:00
|
|
|
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
|
2013-07-16 21:13:06 +04:00
|
|
|
*
|
2016-09-02 00:32:49 +03:00
|
|
|
* This source code is subject to the terms of the BSD 2 Clause License and
|
|
|
|
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
|
|
|
|
* was not distributed with this source code in the LICENSE file, you can
|
|
|
|
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
|
|
|
|
* Media Patent License 1.0 was not distributed with this source code in the
|
|
|
|
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
|
2013-07-16 21:13:06 +04:00
|
|
|
*/
|
|
|
|
|
2015-08-13 02:14:53 +03:00
|
|
|
#include <assert.h>
|
|
|
|
|
2016-08-31 00:01:10 +03:00
|
|
|
#include "./aom_dsp_rtcd.h"
|
|
|
|
#include "aom_dsp/aom_dsp_common.h"
|
2016-08-23 02:08:15 +03:00
|
|
|
#include "aom_ports/mem.h"
|
2013-07-16 21:13:06 +04:00
|
|
|
|
2016-08-31 00:01:10 +03:00
|
|
|
void aom_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
|
2016-08-09 08:59:08 +03:00
|
|
|
ptrdiff_t dst_stride, const int16_t *filter_x,
|
|
|
|
int x_step_q4, const int16_t *filter_y, int y_step_q4,
|
2013-07-16 21:13:06 +04:00
|
|
|
int w, int h) {
|
|
|
|
/* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the
|
|
|
|
* maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4).
|
|
|
|
*/
|
2015-05-02 23:24:16 +03:00
|
|
|
DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
|
2013-07-16 21:13:06 +04:00
|
|
|
|
|
|
|
// Account for the vertical phase needing 3 lines prior and 4 lines post
|
|
|
|
int intermediate_height = h + 7;
|
|
|
|
|
2015-08-13 02:14:53 +03:00
|
|
|
assert(y_step_q4 == 16);
|
|
|
|
assert(x_step_q4 == 16);
|
2013-07-16 21:13:06 +04:00
|
|
|
|
|
|
|
/* Filter starting 3 lines back. The neon implementation will ignore the
|
|
|
|
* given height and filter a multiple of 4 lines. Since this goes in to
|
|
|
|
* the temp buffer which has lots of extra room and is subsequently discarded
|
|
|
|
* this is safe if somewhat less than ideal.
|
|
|
|
*/
|
2016-08-31 00:01:10 +03:00
|
|
|
aom_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, 64, filter_x,
|
2016-08-09 08:59:08 +03:00
|
|
|
x_step_q4, filter_y, y_step_q4, w,
|
|
|
|
intermediate_height);
|
2013-07-16 21:13:06 +04:00
|
|
|
|
|
|
|
/* Step into the temp buffer 3 lines to get the actual frame data */
|
2016-08-31 00:01:10 +03:00
|
|
|
aom_convolve8_vert_neon(temp + 64 * 3, 64, dst, dst_stride, filter_x,
|
2016-08-09 08:59:08 +03:00
|
|
|
x_step_q4, filter_y, y_step_q4, w, h);
|
2013-07-16 21:13:06 +04:00
|
|
|
}
|
|
|
|
|
2016-08-31 00:01:10 +03:00
|
|
|
void aom_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
|
2013-07-16 21:13:06 +04:00
|
|
|
uint8_t *dst, ptrdiff_t dst_stride,
|
|
|
|
const int16_t *filter_x, int x_step_q4,
|
2016-08-09 08:59:08 +03:00
|
|
|
const int16_t *filter_y, int y_step_q4, int w,
|
|
|
|
int h) {
|
2015-05-02 23:24:16 +03:00
|
|
|
DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
|
2013-07-16 21:13:06 +04:00
|
|
|
int intermediate_height = h + 7;
|
|
|
|
|
2015-08-13 02:14:53 +03:00
|
|
|
assert(y_step_q4 == 16);
|
|
|
|
assert(x_step_q4 == 16);
|
2013-07-16 21:13:06 +04:00
|
|
|
|
|
|
|
/* This implementation has the same issues as above. In addition, we only want
|
|
|
|
* to average the values after both passes.
|
|
|
|
*/
|
2016-08-31 00:01:10 +03:00
|
|
|
aom_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, 64, filter_x,
|
2016-08-09 08:59:08 +03:00
|
|
|
x_step_q4, filter_y, y_step_q4, w,
|
|
|
|
intermediate_height);
|
2016-08-31 00:01:10 +03:00
|
|
|
aom_convolve8_avg_vert_neon(temp + 64 * 3, 64, dst, dst_stride, filter_x,
|
2016-08-09 08:59:08 +03:00
|
|
|
x_step_q4, filter_y, y_step_q4, w, h);
|
2013-07-16 21:13:06 +04:00
|
|
|
}
|