mirror of https://git.ffmpeg.org/ffmpeg.git
363 lines
12 KiB
ArmAsm
363 lines
12 KiB
ArmAsm
/*
|
|
* ARM NEON IDCT
|
|
*
|
|
* Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
|
|
* Copyright (c) 2017 Matthieu Bouron <matthieu.bouron@gmail.com>
|
|
*
|
|
* Based on Simple IDCT
|
|
* Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
|
|
*
|
|
* 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 "libavutil/aarch64/asm.S"
|
|
|
|
#define Z1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
|
|
#define Z4c ((1<<(COL_SHIFT-1))/Z4)
|
|
#define ROW_SHIFT 11
|
|
#define COL_SHIFT 20
|
|
|
|
#define z1 v0.H[0]
|
|
#define z2 v0.H[1]
|
|
#define z3 v0.H[2]
|
|
#define z4 v0.H[3]
|
|
#define z5 v0.H[4]
|
|
#define z6 v0.H[5]
|
|
#define z7 v0.H[6]
|
|
#define z4c v0.H[7]
|
|
|
|
const idct_coeff_neon, align=4
|
|
.short Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z4c
|
|
endconst
|
|
|
|
.macro idct_start data
|
|
prfm pldl1keep, [\data]
|
|
mov x10, x30
|
|
movrel x3, idct_coeff_neon
|
|
ld1 {v0.2d}, [x3]
|
|
.endm
|
|
|
|
.macro idct_end
|
|
ret x10
|
|
.endm
|
|
|
|
.macro smull1 a, b, c
|
|
smull \a, \b, \c
|
|
.endm
|
|
|
|
.macro smlal1 a, b, c
|
|
smlal \a, \b, \c
|
|
.endm
|
|
|
|
.macro smlsl1 a, b, c
|
|
smlsl \a, \b, \c
|
|
.endm
|
|
|
|
.macro idct_col4_top y1, y2, y3, y4, i, l
|
|
smull\i v7.4s, \y3\l, z2
|
|
smull\i v16.4s, \y3\l, z6
|
|
smull\i v17.4s, \y2\l, z1
|
|
add v19.4s, v23.4s, v7.4s
|
|
smull\i v18.4s, \y2\l, z3
|
|
add v20.4s, v23.4s, v16.4s
|
|
smull\i v5.4s, \y2\l, z5
|
|
sub v21.4s, v23.4s, v16.4s
|
|
smull\i v6.4s, \y2\l, z7
|
|
sub v22.4s, v23.4s, v7.4s
|
|
|
|
smlal\i v17.4s, \y4\l, z3
|
|
smlsl\i v18.4s, \y4\l, z7
|
|
smlsl\i v5.4s, \y4\l, z1
|
|
smlsl\i v6.4s, \y4\l, z5
|
|
.endm
|
|
|
|
.macro idct_row4_neon y1, y2, y3, y4, pass
|
|
ld1 {\y1\().2d,\y2\().2d}, [x2], #32
|
|
movi v23.4s, #1<<2, lsl #8
|
|
orr v5.16b, \y1\().16b, \y2\().16b
|
|
ld1 {\y3\().2d,\y4\().2d}, [x2], #32
|
|
orr v6.16b, \y3\().16b, \y4\().16b
|
|
orr v5.16b, v5.16b, v6.16b
|
|
mov x3, v5.d[1]
|
|
smlal v23.4s, \y1\().4h, z4
|
|
|
|
idct_col4_top \y1, \y2, \y3, \y4, 1, .4h
|
|
|
|
cmp x3, #0
|
|
b.eq \pass\()f
|
|
|
|
smull2 v7.4s, \y1\().8h, z4
|
|
smlal2 v17.4s, \y2\().8h, z5
|
|
smlsl2 v18.4s, \y2\().8h, z1
|
|
smull2 v16.4s, \y3\().8h, z2
|
|
smlal2 v5.4s, \y2\().8h, z7
|
|
add v19.4s, v19.4s, v7.4s
|
|
sub v20.4s, v20.4s, v7.4s
|
|
sub v21.4s, v21.4s, v7.4s
|
|
add v22.4s, v22.4s, v7.4s
|
|
smlal2 v6.4s, \y2\().8h, z3
|
|
smull2 v7.4s, \y3\().8h, z6
|
|
smlal2 v17.4s, \y4\().8h, z7
|
|
smlsl2 v18.4s, \y4\().8h, z5
|
|
smlal2 v5.4s, \y4\().8h, z3
|
|
smlsl2 v6.4s, \y4\().8h, z1
|
|
add v19.4s, v19.4s, v7.4s
|
|
sub v20.4s, v20.4s, v16.4s
|
|
add v21.4s, v21.4s, v16.4s
|
|
sub v22.4s, v22.4s, v7.4s
|
|
|
|
\pass: add \y3\().4S, v19.4S, v17.4S
|
|
add \y4\().4s, v20.4s, v18.4s
|
|
shrn \y1\().4h, \y3\().4s, #ROW_SHIFT
|
|
shrn \y2\().4h, \y4\().4s, #ROW_SHIFT
|
|
add v7.4s, v21.4s, v5.4s
|
|
add v16.4s, v22.4s, v6.4s
|
|
shrn \y3\().4h, v7.4s, #ROW_SHIFT
|
|
shrn \y4\().4h, v16.4s, #ROW_SHIFT
|
|
sub v22.4s, v22.4s, v6.4s
|
|
sub v19.4s, v19.4s, v17.4s
|
|
sub v21.4s, v21.4s, v5.4s
|
|
shrn2 \y1\().8h, v22.4s, #ROW_SHIFT
|
|
sub v20.4s, v20.4s, v18.4s
|
|
shrn2 \y2\().8h, v21.4s, #ROW_SHIFT
|
|
shrn2 \y3\().8h, v20.4s, #ROW_SHIFT
|
|
shrn2 \y4\().8h, v19.4s, #ROW_SHIFT
|
|
|
|
trn1 v16.8h, \y1\().8h, \y2\().8h
|
|
trn2 v17.8h, \y1\().8h, \y2\().8h
|
|
trn1 v18.8h, \y3\().8h, \y4\().8h
|
|
trn2 v19.8h, \y3\().8h, \y4\().8h
|
|
trn1 \y1\().4s, v16.4s, v18.4s
|
|
trn1 \y2\().4s, v17.4s, v19.4s
|
|
trn2 \y3\().4s, v16.4s, v18.4s
|
|
trn2 \y4\().4s, v17.4s, v19.4s
|
|
.endm
|
|
|
|
.macro declare_idct_col4_neon i, l
|
|
function idct_col4_neon\i
|
|
dup v23.4h, z4c
|
|
.if \i == 1
|
|
add v23.4h, v23.4h, v24.4h
|
|
.else
|
|
mov v5.d[0], v24.d[1]
|
|
add v23.4h, v23.4h, v5.4h
|
|
.endif
|
|
smull v23.4s, v23.4h, z4
|
|
|
|
idct_col4_top v24, v25, v26, v27, \i, \l
|
|
|
|
mov x4, v28.d[\i - 1]
|
|
mov x5, v29.d[\i - 1]
|
|
cmp x4, #0
|
|
b.eq 1f
|
|
|
|
smull\i v7.4s, v28\l, z4
|
|
add v19.4s, v19.4s, v7.4s
|
|
sub v20.4s, v20.4s, v7.4s
|
|
sub v21.4s, v21.4s, v7.4s
|
|
add v22.4s, v22.4s, v7.4s
|
|
|
|
1: mov x4, v30.d[\i - 1]
|
|
cmp x5, #0
|
|
b.eq 2f
|
|
|
|
smlal\i v17.4s, v29\l, z5
|
|
smlsl\i v18.4s, v29\l, z1
|
|
smlal\i v5.4s, v29\l, z7
|
|
smlal\i v6.4s, v29\l, z3
|
|
|
|
2: mov x5, v31.d[\i - 1]
|
|
cmp x4, #0
|
|
b.eq 3f
|
|
|
|
smull\i v7.4s, v30\l, z6
|
|
smull\i v16.4s, v30\l, z2
|
|
add v19.4s, v19.4s, v7.4s
|
|
sub v22.4s, v22.4s, v7.4s
|
|
sub v20.4s, v20.4s, v16.4s
|
|
add v21.4s, v21.4s, v16.4s
|
|
|
|
3: cmp x5, #0
|
|
b.eq 4f
|
|
|
|
smlal\i v17.4s, v31\l, z7
|
|
smlsl\i v18.4s, v31\l, z5
|
|
smlal\i v5.4s, v31\l, z3
|
|
smlsl\i v6.4s, v31\l, z1
|
|
|
|
4: addhn v7.4h, v19.4s, v17.4s
|
|
addhn2 v7.8h, v20.4s, v18.4s
|
|
subhn v18.4h, v20.4s, v18.4s
|
|
subhn2 v18.8h, v19.4s, v17.4s
|
|
|
|
addhn v16.4h, v21.4s, v5.4s
|
|
addhn2 v16.8h, v22.4s, v6.4s
|
|
subhn v17.4h, v22.4s, v6.4s
|
|
subhn2 v17.8h, v21.4s, v5.4s
|
|
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
declare_idct_col4_neon 1, .4H
|
|
declare_idct_col4_neon 2, .8H
|
|
|
|
function ff_simple_idct_put_neon, export=1
|
|
idct_start x2
|
|
|
|
idct_row4_neon v24, v25, v26, v27, 1
|
|
idct_row4_neon v28, v29, v30, v31, 2
|
|
bl idct_col4_neon1
|
|
|
|
sqshrun v1.8b, v7.8h, #COL_SHIFT-16
|
|
sqshrun2 v1.16b, v16.8h, #COL_SHIFT-16
|
|
sqshrun v3.8b, v17.8h, #COL_SHIFT-16
|
|
sqshrun2 v3.16b, v18.8h, #COL_SHIFT-16
|
|
|
|
bl idct_col4_neon2
|
|
|
|
sqshrun v2.8b, v7.8h, #COL_SHIFT-16
|
|
sqshrun2 v2.16b, v16.8h, #COL_SHIFT-16
|
|
sqshrun v4.8b, v17.8h, #COL_SHIFT-16
|
|
sqshrun2 v4.16b, v18.8h, #COL_SHIFT-16
|
|
|
|
zip1 v16.4s, v1.4s, v2.4s
|
|
zip2 v17.4s, v1.4s, v2.4s
|
|
|
|
st1 {v16.d}[0], [x0], x1
|
|
st1 {v16.d}[1], [x0], x1
|
|
|
|
zip1 v18.4s, v3.4s, v4.4s
|
|
zip2 v19.4s, v3.4s, v4.4s
|
|
|
|
st1 {v17.d}[0], [x0], x1
|
|
st1 {v17.d}[1], [x0], x1
|
|
st1 {v18.d}[0], [x0], x1
|
|
st1 {v18.d}[1], [x0], x1
|
|
st1 {v19.d}[0], [x0], x1
|
|
st1 {v19.d}[1], [x0], x1
|
|
|
|
idct_end
|
|
endfunc
|
|
|
|
function ff_simple_idct_add_neon, export=1
|
|
idct_start x2
|
|
|
|
idct_row4_neon v24, v25, v26, v27, 1
|
|
idct_row4_neon v28, v29, v30, v31, 2
|
|
bl idct_col4_neon1
|
|
|
|
sshr v1.8h, v7.8h, #COL_SHIFT-16
|
|
sshr v2.8h, v16.8h, #COL_SHIFT-16
|
|
sshr v3.8h, v17.8h, #COL_SHIFT-16
|
|
sshr v4.8h, v18.8h, #COL_SHIFT-16
|
|
|
|
bl idct_col4_neon2
|
|
|
|
sshr v7.8h, v7.8h, #COL_SHIFT-16
|
|
sshr v16.8h, v16.8h, #COL_SHIFT-16
|
|
sshr v17.8h, v17.8h, #COL_SHIFT-16
|
|
sshr v18.8h, v18.8h, #COL_SHIFT-16
|
|
|
|
mov x9, x0
|
|
ld1 {v19.d}[0], [x0], x1
|
|
zip1 v23.2d, v1.2d, v7.2d
|
|
zip2 v24.2d, v1.2d, v7.2d
|
|
ld1 {v19.d}[1], [x0], x1
|
|
zip1 v25.2d, v2.2d, v16.2d
|
|
zip2 v26.2d, v2.2d, v16.2d
|
|
ld1 {v20.d}[0], [x0], x1
|
|
zip1 v27.2d, v3.2d, v17.2d
|
|
zip2 v28.2d, v3.2d, v17.2d
|
|
ld1 {v20.d}[1], [x0], x1
|
|
zip1 v29.2d, v4.2d, v18.2d
|
|
zip2 v30.2d, v4.2d, v18.2d
|
|
ld1 {v21.d}[0], [x0], x1
|
|
uaddw v23.8h, v23.8h, v19.8b
|
|
uaddw2 v24.8h, v24.8h, v19.16b
|
|
ld1 {v21.d}[1], [x0], x1
|
|
sqxtun v23.8b, v23.8h
|
|
sqxtun2 v23.16b, v24.8h
|
|
ld1 {v22.d}[0], [x0], x1
|
|
uaddw v24.8h, v25.8h, v20.8b
|
|
uaddw2 v25.8h, v26.8h, v20.16b
|
|
ld1 {v22.d}[1], [x0], x1
|
|
sqxtun v24.8b, v24.8h
|
|
sqxtun2 v24.16b, v25.8h
|
|
st1 {v23.d}[0], [x9], x1
|
|
uaddw v25.8h, v27.8h, v21.8b
|
|
uaddw2 v26.8h, v28.8h, v21.16b
|
|
st1 {v23.d}[1], [x9], x1
|
|
sqxtun v25.8b, v25.8h
|
|
sqxtun2 v25.16b, v26.8h
|
|
st1 {v24.d}[0], [x9], x1
|
|
uaddw v26.8h, v29.8h, v22.8b
|
|
uaddw2 v27.8h, v30.8h, v22.16b
|
|
st1 {v24.d}[1], [x9], x1
|
|
sqxtun v26.8b, v26.8h
|
|
sqxtun2 v26.16b, v27.8h
|
|
st1 {v25.d}[0], [x9], x1
|
|
st1 {v25.d}[1], [x9], x1
|
|
st1 {v26.d}[0], [x9], x1
|
|
st1 {v26.d}[1], [x9], x1
|
|
|
|
idct_end
|
|
endfunc
|
|
|
|
function ff_simple_idct_neon, export=1
|
|
idct_start x0
|
|
|
|
mov x2, x0
|
|
idct_row4_neon v24, v25, v26, v27, 1
|
|
idct_row4_neon v28, v29, v30, v31, 2
|
|
sub x2, x2, #128
|
|
bl idct_col4_neon1
|
|
|
|
sshr v1.8h, v7.8h, #COL_SHIFT-16
|
|
sshr v2.8h, v16.8h, #COL_SHIFT-16
|
|
sshr v3.8h, v17.8h, #COL_SHIFT-16
|
|
sshr v4.8h, v18.8h, #COL_SHIFT-16
|
|
|
|
bl idct_col4_neon2
|
|
|
|
sshr v7.8h, v7.8h, #COL_SHIFT-16
|
|
sshr v16.8h, v16.8h, #COL_SHIFT-16
|
|
sshr v17.8h, v17.8h, #COL_SHIFT-16
|
|
sshr v18.8h, v18.8h, #COL_SHIFT-16
|
|
|
|
zip1 v23.2d, v1.2d, v7.2d
|
|
zip2 v24.2d, v1.2d, v7.2d
|
|
st1 {v23.2d,v24.2d}, [x2], #32
|
|
zip1 v25.2d, v2.2d, v16.2d
|
|
zip2 v26.2d, v2.2d, v16.2d
|
|
st1 {v25.2d,v26.2d}, [x2], #32
|
|
zip1 v27.2d, v3.2d, v17.2d
|
|
zip2 v28.2d, v3.2d, v17.2d
|
|
st1 {v27.2d,v28.2d}, [x2], #32
|
|
zip1 v29.2d, v4.2d, v18.2d
|
|
zip2 v30.2d, v4.2d, v18.2d
|
|
st1 {v29.2d,v30.2d}, [x2], #32
|
|
|
|
idct_end
|
|
endfunc
|