mirror of
https://github.com/FFmpeg/FFmpeg.git
synced 2024-12-12 19:18:44 +02:00
a67ae67083
This work is sponsored by, and copyright, Google.
For the transforms up to 8x8, we can fit all the data (including
temporaries) in registers and just do a straightforward transform
of all the data. For 16x16, we do a transform of 4x16 pixels in
4 slices, using a temporary buffer. For 32x32, we transform 4x32
pixels at a time, in two steps of 4x16 pixels each.
Examples of relative speedup compared to the C version, from checkasm:
Cortex A7 A8 A9 A53
vp9_inv_adst_adst_4x4_add_neon: 3.39 5.83 4.17 4.01
vp9_inv_adst_adst_8x8_add_neon: 3.79 4.86 4.23 3.98
vp9_inv_adst_adst_16x16_add_neon: 3.33 4.36 4.11 4.16
vp9_inv_dct_dct_4x4_add_neon: 4.06 6.16 4.59 4.46
vp9_inv_dct_dct_8x8_add_neon: 4.61 6.01 4.98 4.86
vp9_inv_dct_dct_16x16_add_neon: 3.35 3.44 3.36 3.79
vp9_inv_dct_dct_32x32_add_neon: 3.89 3.50 3.79 4.42
vp9_inv_wht_wht_4x4_add_neon: 3.22 5.13 3.53 3.77
Thus, the speedup vs C code is around 3-6x.
This is mostly marginally faster than the corresponding routines
in libvpx on most cores, tested with their 32x32 idct (compared to
vpx_idct32x32_1024_add_neon). These numbers are slightly in libvpx's
favour since their version doesn't clear the input buffer like ours
do (although the effect of that on the total runtime probably is
negligible.)
Cortex A7 A8 A9 A53
vp9_inv_dct_dct_32x32_add_neon: 18436.8 16874.1 14235.1 11988.9
libvpx vpx_idct32x32_1024_add_neon 20789.0 13344.3 15049.9 13030.5
Only on the Cortex A8, the libvpx function is faster. On the other cores,
ours is slightly faster even though ours has got source block clearing
integrated.
Signed-off-by: Martin Storsjö <martin@martin.st>
190 lines
7.1 KiB
C
190 lines
7.1 KiB
C
/*
|
|
* Copyright (c) 2016 Google Inc.
|
|
*
|
|
* This file is part of Libav.
|
|
*
|
|
* Libav 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.
|
|
*
|
|
* Libav 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 Libav; 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/vp9.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 cpu_flags = av_get_cpu_flags();
|
|
|
|
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 cpu_flags = av_get_cpu_flags();
|
|
|
|
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);
|
|
}
|
|
}
|
|
|
|
av_cold void ff_vp9dsp_init_arm(VP9DSPContext *dsp)
|
|
{
|
|
vp9dsp_mc_init_arm(dsp);
|
|
vp9dsp_itxfm_init_arm(dsp);
|
|
}
|