diff options
author | Matt A. Tobin <mattatobin@localhost.localdomain> | 2018-02-02 04:16:08 -0500 |
---|---|---|
committer | Matt A. Tobin <mattatobin@localhost.localdomain> | 2018-02-02 04:16:08 -0500 |
commit | 5f8de423f190bbb79a62f804151bc24824fa32d8 (patch) | |
tree | 10027f336435511475e392454359edea8e25895d /media/libvpx/vp9/common/arm/neon/vp9_convolve_neon.c | |
parent | 49ee0794b5d912db1f95dce6eb52d781dc210db5 (diff) | |
download | UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.gz UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.lz UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.xz UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.zip |
Add m-esr52 at 52.6.0
Diffstat (limited to 'media/libvpx/vp9/common/arm/neon/vp9_convolve_neon.c')
-rw-r--r-- | media/libvpx/vp9/common/arm/neon/vp9_convolve_neon.c | 82 |
1 files changed, 82 insertions, 0 deletions
diff --git a/media/libvpx/vp9/common/arm/neon/vp9_convolve_neon.c b/media/libvpx/vp9/common/arm/neon/vp9_convolve_neon.c new file mode 100644 index 000000000..2e28cb20e --- /dev/null +++ b/media/libvpx/vp9/common/arm/neon/vp9_convolve_neon.c @@ -0,0 +1,82 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_common.h" +#include "vpx_ports/mem.h" + +void vp9_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + 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). + */ + DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); + + // Account for the vertical phase needing 3 lines prior and 4 lines post + int intermediate_height = h + 7; + + if (x_step_q4 != 16 || y_step_q4 != 16) { + vp9_convolve8_c(src, src_stride, + dst, dst_stride, + filter_x, x_step_q4, + filter_y, y_step_q4, + w, h); + return; + } + + /* 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. + */ + vp9_convolve8_horiz_neon(src - src_stride * 3, src_stride, + temp, 64, + filter_x, x_step_q4, filter_y, y_step_q4, + w, intermediate_height); + + /* Step into the temp buffer 3 lines to get the actual frame data */ + vp9_convolve8_vert_neon(temp + 64 * 3, 64, + dst, dst_stride, + filter_x, x_step_q4, filter_y, y_step_q4, + w, h); +} + +void vp9_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride, + uint8_t *dst, ptrdiff_t dst_stride, + const int16_t *filter_x, int x_step_q4, + const int16_t *filter_y, int y_step_q4, + int w, int h) { + DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); + int intermediate_height = h + 7; + + if (x_step_q4 != 16 || y_step_q4 != 16) { + vp9_convolve8_avg_c(src, src_stride, + dst, dst_stride, + filter_x, x_step_q4, + filter_y, y_step_q4, + w, h); + return; + } + + /* This implementation has the same issues as above. In addition, we only want + * to average the values after both passes. + */ + vp9_convolve8_horiz_neon(src - src_stride * 3, src_stride, + temp, 64, + filter_x, x_step_q4, filter_y, y_step_q4, + w, intermediate_height); + vp9_convolve8_avg_vert_neon(temp + 64 * 3, + 64, dst, dst_stride, + filter_x, x_step_q4, filter_y, y_step_q4, + w, h); +} |