ffmpeg/FFmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2024-09-21 05:46:55 +00:00
Code Issues Packages Projects Releases Wiki Activity
8c8f543b81
FFmpeg/libavcodec/arm/vp9dsp_init_arm.c
Martin Storsjö 6bec60a683 arm: vp9: Add NEON loop filters 
This work is sponsored by, and copyright, Google.

The implementation tries to have smart handling of cases
where no pixels need the full filtering for the 8/16 width
filters, skipping both calculation and writeback of the
unmodified pixels in those cases. The actual effect of this
is hard to test with checkasm though, since it tests the
full filtering, and the benefit depends on how many filtered
blocks use the shortcut.

Examples of relative speedup compared to the C version, from checkasm:
                          Cortex       A7     A8     A9    A53
vp9_loop_filter_h_4_8_neon:          2.72   2.68   1.78   3.15
vp9_loop_filter_h_8_8_neon:          2.36   2.38   1.70   2.91
vp9_loop_filter_h_16_8_neon:         1.80   1.89   1.45   2.01
vp9_loop_filter_h_16_16_neon:        2.81   2.78   2.18   3.16
vp9_loop_filter_mix2_h_44_16_neon:   2.65   2.67   1.93   3.05
vp9_loop_filter_mix2_h_48_16_neon:   2.46   2.38   1.81   2.85
vp9_loop_filter_mix2_h_84_16_neon:   2.50   2.41   1.73   2.85
vp9_loop_filter_mix2_h_88_16_neon:   2.77   2.66   1.96   3.23
vp9_loop_filter_mix2_v_44_16_neon:   4.28   4.46   3.22   5.70
vp9_loop_filter_mix2_v_48_16_neon:   3.92   4.00   3.03   5.19
vp9_loop_filter_mix2_v_84_16_neon:   3.97   4.31   2.98   5.33
vp9_loop_filter_mix2_v_88_16_neon:   3.91   4.19   3.06   5.18
vp9_loop_filter_v_4_8_neon:          4.53   4.47   3.31   6.05
vp9_loop_filter_v_8_8_neon:          3.58   3.99   2.92   5.17
vp9_loop_filter_v_16_8_neon:         3.40   3.50   2.81   4.68
vp9_loop_filter_v_16_16_neon:        4.66   4.41   3.74   6.02

The speedup vs C code is around 2-6x. The numbers are quite
inconclusive though, since the checkasm test runs multiple filterings
on top of each other, so later rounds might end up with different
codepaths (different decisions on which filter to apply, based
on input pixel differences). Disabling the early-exit in the asm
doesn't give a fair comparison either though, since the C code
only does the necessary calcuations for each row.

Based on START_TIMER/STOP_TIMER wrapping around a few individual
functions, the speedup vs C code is around 4-9x.

This is pretty similar in runtime to the corresponding routines
in libvpx. (This is comparing vpx_lpf_vertical_16_neon,
vpx_lpf_horizontal_edge_8_neon and vpx_lpf_horizontal_edge_16_neon
to vp9_loop_filter_h_16_8_neon, vp9_loop_filter_v_16_8_neon
and vp9_loop_filter_v_16_16_neon - note that the naming of horizonal
and vertical is flipped between the libraries.)

In order to have stable, comparable numbers, the early exits in both
asm versions were disabled, forcing the full filtering codepath.

                           Cortex           A7      A8      A9     A53
vp9_loop_filter_h_16_8_neon:             597.2   472.0   482.4   415.0
libvpx vpx_lpf_vertical_16_neon:         626.0   464.5   470.7   445.0
vp9_loop_filter_v_16_8_neon:             500.2   422.5   429.7   295.0
libvpx vpx_lpf_horizontal_edge_8_neon:   586.5   414.5   415.6   383.2
vp9_loop_filter_v_16_16_neon:            905.0   784.7   791.5   546.0
libvpx vpx_lpf_horizontal_edge_16_neon: 1060.2   751.7   743.5   685.2

Our version is consistently faster on on A7 and A53, marginally slower on
A8, and sometimes faster, sometimes slower on A9 (marginally slower in all
three tests in this particular test run).

This is an adapted cherry-pick from libav commit
dd299a2d6d.

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2016-11-15 15:10:03 -05:00

256 lines
9.8 KiB
C
Raw Blame History

/*
* Copyright (c) 2016 Google Inc.
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <stdint.h>
#include "libavutil/attributes.h"
#include "libavutil/arm/cpu.h"
#include "libavcodec/vp9dsp.h"
#define declare_fpel(type, sz) \
void ff_vp9_##type##sz##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
const uint8_t *src, ptrdiff_t src_stride, \
int h, int mx, int my)
#define declare_copy_avg(sz) \
declare_fpel(copy, sz); \
declare_fpel(avg , sz)
#define decl_mc_func(op, filter, dir, sz) \
void ff_vp9_##op##_##filter##sz##_##dir##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
const uint8_t *src, ptrdiff_t src_stride, \
int h, int mx, int my)
#define define_8tap_2d_fn(op, filter, sz) \
static void op##_##filter##sz##_hv_neon(uint8_t *dst, ptrdiff_t dst_stride, \
const uint8_t *src, ptrdiff_t src_stride, \
int h, int mx, int my) \
{ \
LOCAL_ALIGNED_16(uint8_t, temp, [((1 + (sz < 64)) * sz + 8) * sz]); \
/* We only need h + 7 lines, but the horizontal filter assumes an \
* even number of rows, so filter h + 8 lines here. */ \
ff_vp9_put_##filter##sz##_h_neon(temp, sz, \
src - 3 * src_stride, src_stride, \
h + 8, mx, 0); \
ff_vp9_##op##_##filter##sz##_v_neon(dst, dst_stride, \
temp + 3 * sz, sz, \
h, 0, my); \
}
#define decl_filter_funcs(op, dir, sz) \
decl_mc_func(op, regular, dir, sz); \
decl_mc_func(op, sharp, dir, sz); \
decl_mc_func(op, smooth, dir, sz)
#define decl_mc_funcs(sz) \
decl_filter_funcs(put, h, sz); \
decl_filter_funcs(avg, h, sz); \
decl_filter_funcs(put, v, sz); \
decl_filter_funcs(avg, v, sz); \
decl_filter_funcs(put, hv, sz); \
decl_filter_funcs(avg, hv, sz)
declare_copy_avg(64);
declare_copy_avg(32);
declare_copy_avg(16);
declare_copy_avg(8);
declare_copy_avg(4);
decl_mc_funcs(64);
decl_mc_funcs(32);
decl_mc_funcs(16);
decl_mc_funcs(8);
decl_mc_funcs(4);
#define define_8tap_2d_funcs(sz) \
define_8tap_2d_fn(put, regular, sz) \
define_8tap_2d_fn(put, sharp, sz) \
define_8tap_2d_fn(put, smooth, sz) \
define_8tap_2d_fn(avg, regular, sz) \
define_8tap_2d_fn(avg, sharp, sz) \
define_8tap_2d_fn(avg, smooth, sz)
define_8tap_2d_funcs(64)
define_8tap_2d_funcs(32)
define_8tap_2d_funcs(16)
define_8tap_2d_funcs(8)
define_8tap_2d_funcs(4)
static av_cold void vp9dsp_mc_init_arm(VP9DSPContext *dsp, int bpp)
{
int cpu_flags = av_get_cpu_flags();
if (bpp != 8)
return;
if (have_neon(cpu_flags)) {
#define init_fpel(idx1, idx2, sz, type) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_##type##sz##_neon
#define init_copy_avg(idx, sz) \
init_fpel(idx, 0, sz, copy); \
init_fpel(idx, 1, sz, avg)
#define init_mc_func(idx1, idx2, op, filter, fname, dir, mx, my, sz, pfx) \
dsp->mc[idx1][filter][idx2][mx][my] = pfx##op##_##fname##sz##_##dir##_neon
#define init_mc_funcs(idx, dir, mx, my, sz, pfx) \
init_mc_func(idx, 0, put, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \
init_mc_func(idx, 0, put, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \
init_mc_func(idx, 0, put, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx); \
init_mc_func(idx, 1, avg, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \
init_mc_func(idx, 1, avg, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \
init_mc_func(idx, 1, avg, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx)
#define init_mc_funcs_dirs(idx, sz) \
init_mc_funcs(idx, h, 1, 0, sz, ff_vp9_); \
init_mc_funcs(idx, v, 0, 1, sz, ff_vp9_); \
init_mc_funcs(idx, hv, 1, 1, sz,)
init_copy_avg(0, 64);
init_copy_avg(1, 32);
init_copy_avg(2, 16);
init_copy_avg(3, 8);
init_copy_avg(4, 4);
init_mc_funcs_dirs(0, 64);
init_mc_funcs_dirs(1, 32);
init_mc_funcs_dirs(2, 16);
init_mc_funcs_dirs(3, 8);
init_mc_funcs_dirs(4, 4);
}
}
#define define_itxfm(type_a, type_b, sz) \
void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_neon(uint8_t *_dst, \
ptrdiff_t stride, \
int16_t *_block, int eob)
#define define_itxfm_funcs(sz) \
define_itxfm(idct, idct, sz); \
define_itxfm(iadst, idct, sz); \
define_itxfm(idct, iadst, sz); \
define_itxfm(iadst, iadst, sz)
define_itxfm_funcs(4);
define_itxfm_funcs(8);
define_itxfm_funcs(16);
define_itxfm(idct, idct, 32);
define_itxfm(iwht, iwht, 4);
static av_cold void vp9dsp_itxfm_init_arm(VP9DSPContext *dsp, int bpp)
{
int cpu_flags = av_get_cpu_flags();
if (bpp != 8)
return;
if (have_neon(cpu_flags)) {
#define init_itxfm(tx, sz) \
dsp->itxfm_add[tx][DCT_DCT] = ff_vp9_idct_idct_##sz##_add_neon; \
dsp->itxfm_add[tx][DCT_ADST] = ff_vp9_iadst_idct_##sz##_add_neon; \
dsp->itxfm_add[tx][ADST_DCT] = ff_vp9_idct_iadst_##sz##_add_neon; \
dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_neon
#define init_idct(tx, nm) \
dsp->itxfm_add[tx][DCT_DCT] = \
dsp->itxfm_add[tx][ADST_DCT] = \
dsp->itxfm_add[tx][DCT_ADST] = \
dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_neon
init_itxfm(TX_4X4, 4x4);
init_itxfm(TX_8X8, 8x8);
init_itxfm(TX_16X16, 16x16);
init_idct(TX_32X32, idct_idct_32x32);
init_idct(4, iwht_iwht_4x4);
}
}
#define define_loop_filter(dir, wd, size) \
void ff_vp9_loop_filter_##dir##_##wd##_##size##_neon(uint8_t *dst, ptrdiff_t stride, int E, int I, int H)
#define define_loop_filters(wd, size) \
define_loop_filter(h, wd, size); \
define_loop_filter(v, wd, size)
define_loop_filters(4, 8);
define_loop_filters(8, 8);
define_loop_filters(16, 8);
define_loop_filters(16, 16);
#define lf_mix_fn(dir, wd1, wd2, stridea) \
static void loop_filter_##dir##_##wd1##wd2##_16_neon(uint8_t *dst, \
ptrdiff_t stride, \
int E, int I, int H) \
{ \
ff_vp9_loop_filter_##dir##_##wd1##_8_neon(dst, stride, E & 0xff, I & 0xff, H & 0xff); \
ff_vp9_loop_filter_##dir##_##wd2##_8_neon(dst + 8 * stridea, stride, E >> 8, I >> 8, H >> 8); \
}
#define lf_mix_fns(wd1, wd2) \
lf_mix_fn(h, wd1, wd2, stride) \
lf_mix_fn(v, wd1, wd2, sizeof(uint8_t))
lf_mix_fns(4, 4)
lf_mix_fns(4, 8)
lf_mix_fns(8, 4)
lf_mix_fns(8, 8)
static av_cold void vp9dsp_loopfilter_init_arm(VP9DSPContext *dsp, int bpp)
{
int cpu_flags = av_get_cpu_flags();
if (bpp != 8)
return;
if (have_neon(cpu_flags)) {
dsp->loop_filter_8[0][1] = ff_vp9_loop_filter_v_4_8_neon;
dsp->loop_filter_8[0][0] = ff_vp9_loop_filter_h_4_8_neon;
dsp->loop_filter_8[1][1] = ff_vp9_loop_filter_v_8_8_neon;
dsp->loop_filter_8[1][0] = ff_vp9_loop_filter_h_8_8_neon;
dsp->loop_filter_8[2][1] = ff_vp9_loop_filter_v_16_8_neon;
dsp->loop_filter_8[2][0] = ff_vp9_loop_filter_h_16_8_neon;
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_neon;
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_neon;
dsp->loop_filter_mix2[0][0][0] = loop_filter_h_44_16_neon;
dsp->loop_filter_mix2[0][0][1] = loop_filter_v_44_16_neon;
dsp->loop_filter_mix2[0][1][0] = loop_filter_h_48_16_neon;
dsp->loop_filter_mix2[0][1][1] = loop_filter_v_48_16_neon;
dsp->loop_filter_mix2[1][0][0] = loop_filter_h_84_16_neon;
dsp->loop_filter_mix2[1][0][1] = loop_filter_v_84_16_neon;
dsp->loop_filter_mix2[1][1][0] = loop_filter_h_88_16_neon;
dsp->loop_filter_mix2[1][1][1] = loop_filter_v_88_16_neon;
}
}
av_cold void ff_vp9dsp_init_arm(VP9DSPContext *dsp, int bpp)
{
vp9dsp_mc_init_arm(dsp, bpp);
vp9dsp_loopfilter_init_arm(dsp, bpp);
vp9dsp_itxfm_init_arm(dsp, bpp);
}
Reference in New Issue View Git Blame Copy Permalink