marfrit
5c8b09349c
Last unmeasured H.264 kernel. mc20 picked as representative (horizontal half-pel, 6-tap filter; canonical for the H.264 luma qpel family). M1 PASS 10000/10000 first try, M3 = 131.477 Mblock/s on a single core (7.6 ns/block), 135x the 1080p30 floor. Per the cycles 6+7 lightweight-kernel rationale, Phase 4 deferred: QPU dispatch floor (~250 ns/block) is 33x above the NEON per-block cost; R9 ≈ 0.03 deep RED. No realistic QPU offload value. Generalization: all H.264 luma MC variants (mc02, mc11, mc22, etc.) will share this verdict. No need to measure each variant individually. H.264 NEON is dramatically faster than VP9 NEON across the board: - IDCT 4x4: 175 vs N/A (no VP9 analog) - IDCT 8x8: 151 vs 8.2 Mblock/s (18x faster) - MC 6/8-tap: 131 vs 7.0 (19x faster) - Deblock: 92 vs 48 Medge/s (2x faster) H.264 deployment recipe: all CPU NEON except deblock (opportunistic QPU). On a Pi 5 running H.264-only, the QPU is mostly idle. Cycles 1-9 complete. Public API exposes all 9. Next: daedalus-v4l2 sibling repo per locked Phase 8 architecture (B + γ + sibling), then README polish. - external/ffmpeg-snapshot/libavcodec/aarch64/h264qpel_neon.S vendored (1467 lines, all qpel variants) - tests/h264_qpel8_mc20_ref.c: 40-line C ref (clip255 of 6-tap convolution) - tests/bench_neon_h264qpel_mc20.c: M1 + M3 bench - docs/k9_h264qpel_mc20.md: cycle 9 closure with comparison matrix Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
1468 lines
52 KiB
ArmAsm
1468 lines
52 KiB
ArmAsm
/*
|
|
* Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
|
|
* Copyright (c) 2013 Janne Grunau <janne-libav@jannau.net>
|
|
*
|
|
* 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"
|
|
#include "neon.S"
|
|
|
|
/* H.264 qpel MC */
|
|
|
|
.macro lowpass_const r
|
|
movz \r, #20, lsl #16
|
|
movk \r, #5
|
|
mov v6.s[0], \r
|
|
.endm
|
|
|
|
//trashes v0-v5
|
|
.macro lowpass_8 r0, r1, r2, r3, d0, d1, narrow=1
|
|
ext v2.8b, \r0\().8b, \r1\().8b, #2
|
|
ext v3.8b, \r0\().8b, \r1\().8b, #3
|
|
uaddl v2.8h, v2.8b, v3.8b
|
|
ext v4.8b, \r0\().8b, \r1\().8b, #1
|
|
ext v5.8b, \r0\().8b, \r1\().8b, #4
|
|
uaddl v4.8h, v4.8b, v5.8b
|
|
ext v1.8b, \r0\().8b, \r1\().8b, #5
|
|
uaddl \d0\().8h, \r0\().8b, v1.8b
|
|
ext v0.8b, \r2\().8b, \r3\().8b, #2
|
|
mla \d0\().8h, v2.8h, v6.h[1]
|
|
ext v1.8b, \r2\().8b, \r3\().8b, #3
|
|
uaddl v0.8h, v0.8b, v1.8b
|
|
ext v1.8b, \r2\().8b, \r3\().8b, #1
|
|
mls \d0\().8h, v4.8h, v6.h[0]
|
|
ext v3.8b, \r2\().8b, \r3\().8b, #4
|
|
uaddl v1.8h, v1.8b, v3.8b
|
|
ext v2.8b, \r2\().8b, \r3\().8b, #5
|
|
uaddl \d1\().8h, \r2\().8b, v2.8b
|
|
mla \d1\().8h, v0.8h, v6.h[1]
|
|
mls \d1\().8h, v1.8h, v6.h[0]
|
|
.if \narrow
|
|
sqrshrun \d0\().8b, \d0\().8h, #5
|
|
sqrshrun \d1\().8b, \d1\().8h, #5
|
|
.endif
|
|
.endm
|
|
|
|
//trashes v0-v4
|
|
.macro lowpass_8_v r0, r1, r2, r3, r4, r5, r6, d0, d1, narrow=1
|
|
uaddl v2.8h, \r2\().8b, \r3\().8b
|
|
uaddl v0.8h, \r3\().8b, \r4\().8b
|
|
uaddl v4.8h, \r1\().8b, \r4\().8b
|
|
uaddl v1.8h, \r2\().8b, \r5\().8b
|
|
uaddl \d0\().8h, \r0\().8b, \r5\().8b
|
|
uaddl \d1\().8h, \r1\().8b, \r6\().8b
|
|
mla \d0\().8h, v2.8h, v6.h[1]
|
|
mls \d0\().8h, v4.8h, v6.h[0]
|
|
mla \d1\().8h, v0.8h, v6.h[1]
|
|
mls \d1\().8h, v1.8h, v6.h[0]
|
|
.if \narrow
|
|
sqrshrun \d0\().8b, \d0\().8h, #5
|
|
sqrshrun \d1\().8b, \d1\().8h, #5
|
|
.endif
|
|
.endm
|
|
|
|
//trashes v0-v5, v7, v30-v31
|
|
.macro lowpass_8H r0, r1
|
|
ext v0.16b, \r0\().16b, \r0\().16b, #2
|
|
ext v1.16b, \r0\().16b, \r0\().16b, #3
|
|
uaddl v0.8h, v0.8b, v1.8b
|
|
ext v2.16b, \r0\().16b, \r0\().16b, #1
|
|
ext v3.16b, \r0\().16b, \r0\().16b, #4
|
|
uaddl v2.8h, v2.8b, v3.8b
|
|
ext v30.16b, \r0\().16b, \r0\().16b, #5
|
|
uaddl \r0\().8h, \r0\().8b, v30.8b
|
|
ext v4.16b, \r1\().16b, \r1\().16b, #2
|
|
mla \r0\().8h, v0.8h, v6.h[1]
|
|
ext v5.16b, \r1\().16b, \r1\().16b, #3
|
|
uaddl v4.8h, v4.8b, v5.8b
|
|
ext v7.16b, \r1\().16b, \r1\().16b, #1
|
|
mls \r0\().8h, v2.8h, v6.h[0]
|
|
ext v0.16b, \r1\().16b, \r1\().16b, #4
|
|
uaddl v7.8h, v7.8b, v0.8b
|
|
ext v31.16b, \r1\().16b, \r1\().16b, #5
|
|
uaddl \r1\().8h, \r1\().8b, v31.8b
|
|
mla \r1\().8h, v4.8h, v6.h[1]
|
|
mls \r1\().8h, v7.8h, v6.h[0]
|
|
.endm
|
|
|
|
// trashes v2-v5, v30
|
|
.macro lowpass_8_1 r0, r1, d0, narrow=1
|
|
ext v2.8b, \r0\().8b, \r1\().8b, #2
|
|
ext v3.8b, \r0\().8b, \r1\().8b, #3
|
|
uaddl v2.8h, v2.8b, v3.8b
|
|
ext v4.8b, \r0\().8b, \r1\().8b, #1
|
|
ext v5.8b, \r0\().8b, \r1\().8b, #4
|
|
uaddl v4.8h, v4.8b, v5.8b
|
|
ext v30.8b, \r0\().8b, \r1\().8b, #5
|
|
uaddl \d0\().8h, \r0\().8b, v30.8b
|
|
mla \d0\().8h, v2.8h, v6.h[1]
|
|
mls \d0\().8h, v4.8h, v6.h[0]
|
|
.if \narrow
|
|
sqrshrun \d0\().8b, \d0\().8h, #5
|
|
.endif
|
|
.endm
|
|
|
|
// trashed v0-v7
|
|
.macro lowpass_8.16 r0, r1, r2, r3, r4, r5
|
|
saddl v5.4s, \r2\().4h, \r3\().4h
|
|
saddl2 v1.4s, \r2\().8h, \r3\().8h
|
|
saddl v6.4s, \r1\().4h, \r4\().4h
|
|
saddl2 v2.4s, \r1\().8h, \r4\().8h
|
|
saddl v0.4s, \r0\().4h, \r5\().4h
|
|
saddl2 v4.4s, \r0\().8h, \r5\().8h
|
|
|
|
shl v3.4s, v5.4s, #4
|
|
shl v5.4s, v5.4s, #2
|
|
shl v7.4s, v6.4s, #2
|
|
add v5.4s, v5.4s, v3.4s
|
|
add v6.4s, v6.4s, v7.4s
|
|
|
|
shl v3.4s, v1.4s, #4
|
|
shl v1.4s, v1.4s, #2
|
|
shl v7.4s, v2.4s, #2
|
|
add v1.4s, v1.4s, v3.4s
|
|
add v2.4s, v2.4s, v7.4s
|
|
|
|
add v5.4s, v5.4s, v0.4s
|
|
sub v5.4s, v5.4s, v6.4s
|
|
|
|
add v1.4s, v1.4s, v4.4s
|
|
sub v1.4s, v1.4s, v2.4s
|
|
|
|
rshrn v5.4h, v5.4s, #10
|
|
rshrn2 v5.8h, v1.4s, #10
|
|
|
|
sqxtun \r0\().8b, v5.8h
|
|
.endm
|
|
|
|
function put_h264_qpel16_h_lowpass_neon_packed
|
|
mov x4, x30
|
|
mov x12, #16
|
|
mov x3, #8
|
|
bl put_h264_qpel8_h_lowpass_neon
|
|
sub x1, x1, x2, lsl #4
|
|
add x1, x1, #8
|
|
mov x12, #16
|
|
mov x30, x4
|
|
b put_h264_qpel8_h_lowpass_neon
|
|
endfunc
|
|
|
|
.macro h264_qpel_h_lowpass type
|
|
function \type\()_h264_qpel16_h_lowpass_neon
|
|
mov x13, x30
|
|
mov x12, #16
|
|
bl \type\()_h264_qpel8_h_lowpass_neon
|
|
sub x0, x0, x3, lsl #4
|
|
sub x1, x1, x2, lsl #4
|
|
add x0, x0, #8
|
|
add x1, x1, #8
|
|
mov x12, #16
|
|
mov x30, x13
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_h_lowpass_neon
|
|
1: ld1 {v28.8b, v29.8b}, [x1], x2
|
|
ld1 {v16.8b, v17.8b}, [x1], x2
|
|
subs x12, x12, #2
|
|
lowpass_8 v28, v29, v16, v17, v28, v16
|
|
.ifc \type,avg
|
|
ld1 {v2.8b}, [x0], x3
|
|
ld1 {v3.8b}, [x0]
|
|
urhadd v28.8b, v28.8b, v2.8b
|
|
urhadd v16.8b, v16.8b, v3.8b
|
|
sub x0, x0, x3
|
|
.endif
|
|
st1 {v28.8b}, [x0], x3
|
|
st1 {v16.8b}, [x0], x3
|
|
b.ne 1b
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_h_lowpass put
|
|
h264_qpel_h_lowpass avg
|
|
|
|
.macro h264_qpel_h_lowpass_l2 type
|
|
function \type\()_h264_qpel16_h_lowpass_l2_neon
|
|
mov x13, x30
|
|
mov x12, #16
|
|
bl \type\()_h264_qpel8_h_lowpass_l2_neon
|
|
sub x0, x0, x2, lsl #4
|
|
sub x1, x1, x2, lsl #4
|
|
sub x3, x3, x2, lsl #4
|
|
add x0, x0, #8
|
|
add x1, x1, #8
|
|
add x3, x3, #8
|
|
mov x12, #16
|
|
mov x30, x13
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_h_lowpass_l2_neon
|
|
1: ld1 {v26.8b, v27.8b}, [x1], x2
|
|
ld1 {v16.8b, v17.8b}, [x1], x2
|
|
ld1 {v28.8b}, [x3], x2
|
|
ld1 {v29.8b}, [x3], x2
|
|
subs x12, x12, #2
|
|
lowpass_8 v26, v27, v16, v17, v26, v27
|
|
urhadd v26.8b, v26.8b, v28.8b
|
|
urhadd v27.8b, v27.8b, v29.8b
|
|
.ifc \type,avg
|
|
ld1 {v2.8b}, [x0], x2
|
|
ld1 {v3.8b}, [x0]
|
|
urhadd v26.8b, v26.8b, v2.8b
|
|
urhadd v27.8b, v27.8b, v3.8b
|
|
sub x0, x0, x2
|
|
.endif
|
|
st1 {v26.8b}, [x0], x2
|
|
st1 {v27.8b}, [x0], x2
|
|
b.ne 1b
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_h_lowpass_l2 put
|
|
h264_qpel_h_lowpass_l2 avg
|
|
|
|
function put_h264_qpel16_v_lowpass_neon_packed
|
|
mov x4, x30
|
|
mov x2, #8
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #8
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x4
|
|
b put_h264_qpel8_v_lowpass_neon
|
|
endfunc
|
|
|
|
.macro h264_qpel_v_lowpass type
|
|
function \type\()_h264_qpel16_v_lowpass_neon
|
|
mov x4, x30
|
|
bl \type\()_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
bl \type\()_h264_qpel8_v_lowpass_neon
|
|
sub x0, x0, x2, lsl #4
|
|
add x0, x0, #8
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #8
|
|
bl \type\()_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x4
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_v_lowpass_neon
|
|
ld1 {v16.8b}, [x1], x3
|
|
ld1 {v17.8b}, [x1], x3
|
|
ld1 {v18.8b}, [x1], x3
|
|
ld1 {v19.8b}, [x1], x3
|
|
ld1 {v20.8b}, [x1], x3
|
|
ld1 {v21.8b}, [x1], x3
|
|
ld1 {v22.8b}, [x1], x3
|
|
ld1 {v23.8b}, [x1], x3
|
|
ld1 {v24.8b}, [x1], x3
|
|
ld1 {v25.8b}, [x1], x3
|
|
ld1 {v26.8b}, [x1], x3
|
|
ld1 {v27.8b}, [x1], x3
|
|
ld1 {v28.8b}, [x1]
|
|
|
|
lowpass_8_v v16, v17, v18, v19, v20, v21, v22, v16, v17
|
|
lowpass_8_v v18, v19, v20, v21, v22, v23, v24, v18, v19
|
|
lowpass_8_v v20, v21, v22, v23, v24, v25, v26, v20, v21
|
|
lowpass_8_v v22, v23, v24, v25, v26, v27, v28, v22, v23
|
|
.ifc \type,avg
|
|
ld1 {v24.8b}, [x0], x2
|
|
ld1 {v25.8b}, [x0], x2
|
|
ld1 {v26.8b}, [x0], x2
|
|
urhadd v16.8b, v16.8b, v24.8b
|
|
ld1 {v27.8b}, [x0], x2
|
|
urhadd v17.8b, v17.8b, v25.8b
|
|
ld1 {v28.8b}, [x0], x2
|
|
urhadd v18.8b, v18.8b, v26.8b
|
|
ld1 {v29.8b}, [x0], x2
|
|
urhadd v19.8b, v19.8b, v27.8b
|
|
ld1 {v30.8b}, [x0], x2
|
|
urhadd v20.8b, v20.8b, v28.8b
|
|
ld1 {v31.8b}, [x0], x2
|
|
urhadd v21.8b, v21.8b, v29.8b
|
|
urhadd v22.8b, v22.8b, v30.8b
|
|
urhadd v23.8b, v23.8b, v31.8b
|
|
sub x0, x0, x2, lsl #3
|
|
.endif
|
|
|
|
st1 {v16.8b}, [x0], x2
|
|
st1 {v17.8b}, [x0], x2
|
|
st1 {v18.8b}, [x0], x2
|
|
st1 {v19.8b}, [x0], x2
|
|
st1 {v20.8b}, [x0], x2
|
|
st1 {v21.8b}, [x0], x2
|
|
st1 {v22.8b}, [x0], x2
|
|
st1 {v23.8b}, [x0], x2
|
|
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_v_lowpass put
|
|
h264_qpel_v_lowpass avg
|
|
|
|
.macro h264_qpel_v_lowpass_l2 type
|
|
function \type\()_h264_qpel16_v_lowpass_l2_neon
|
|
mov x4, x30
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon
|
|
sub x1, x1, x3, lsl #2
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon
|
|
sub x0, x0, x3, lsl #4
|
|
sub x12, x12, x2, lsl #4
|
|
add x0, x0, #8
|
|
add x12, x12, #8
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #8
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x4
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_v_lowpass_l2_neon
|
|
ld1 {v16.8b}, [x1], x3
|
|
ld1 {v17.8b}, [x1], x3
|
|
ld1 {v18.8b}, [x1], x3
|
|
ld1 {v19.8b}, [x1], x3
|
|
ld1 {v20.8b}, [x1], x3
|
|
ld1 {v21.8b}, [x1], x3
|
|
ld1 {v22.8b}, [x1], x3
|
|
ld1 {v23.8b}, [x1], x3
|
|
ld1 {v24.8b}, [x1], x3
|
|
ld1 {v25.8b}, [x1], x3
|
|
ld1 {v26.8b}, [x1], x3
|
|
ld1 {v27.8b}, [x1], x3
|
|
ld1 {v28.8b}, [x1]
|
|
|
|
lowpass_8_v v16, v17, v18, v19, v20, v21, v22, v16, v17
|
|
lowpass_8_v v18, v19, v20, v21, v22, v23, v24, v18, v19
|
|
lowpass_8_v v20, v21, v22, v23, v24, v25, v26, v20, v21
|
|
lowpass_8_v v22, v23, v24, v25, v26, v27, v28, v22, v23
|
|
|
|
ld1 {v24.8b}, [x12], x2
|
|
ld1 {v25.8b}, [x12], x2
|
|
ld1 {v26.8b}, [x12], x2
|
|
ld1 {v27.8b}, [x12], x2
|
|
ld1 {v28.8b}, [x12], x2
|
|
urhadd v16.8b, v24.8b, v16.8b
|
|
urhadd v17.8b, v25.8b, v17.8b
|
|
ld1 {v29.8b}, [x12], x2
|
|
urhadd v18.8b, v26.8b, v18.8b
|
|
urhadd v19.8b, v27.8b, v19.8b
|
|
ld1 {v30.8b}, [x12], x2
|
|
urhadd v20.8b, v28.8b, v20.8b
|
|
urhadd v21.8b, v29.8b, v21.8b
|
|
ld1 {v31.8b}, [x12], x2
|
|
urhadd v22.8b, v30.8b, v22.8b
|
|
urhadd v23.8b, v31.8b, v23.8b
|
|
|
|
.ifc \type,avg
|
|
ld1 {v24.8b}, [x0], x3
|
|
ld1 {v25.8b}, [x0], x3
|
|
ld1 {v26.8b}, [x0], x3
|
|
urhadd v16.8b, v16.8b, v24.8b
|
|
ld1 {v27.8b}, [x0], x3
|
|
urhadd v17.8b, v17.8b, v25.8b
|
|
ld1 {v28.8b}, [x0], x3
|
|
urhadd v18.8b, v18.8b, v26.8b
|
|
ld1 {v29.8b}, [x0], x3
|
|
urhadd v19.8b, v19.8b, v27.8b
|
|
ld1 {v30.8b}, [x0], x3
|
|
urhadd v20.8b, v20.8b, v28.8b
|
|
ld1 {v31.8b}, [x0], x3
|
|
urhadd v21.8b, v21.8b, v29.8b
|
|
urhadd v22.8b, v22.8b, v30.8b
|
|
urhadd v23.8b, v23.8b, v31.8b
|
|
sub x0, x0, x3, lsl #3
|
|
.endif
|
|
|
|
st1 {v16.8b}, [x0], x3
|
|
st1 {v17.8b}, [x0], x3
|
|
st1 {v18.8b}, [x0], x3
|
|
st1 {v19.8b}, [x0], x3
|
|
st1 {v20.8b}, [x0], x3
|
|
st1 {v21.8b}, [x0], x3
|
|
st1 {v22.8b}, [x0], x3
|
|
st1 {v23.8b}, [x0], x3
|
|
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_v_lowpass_l2 put
|
|
h264_qpel_v_lowpass_l2 avg
|
|
|
|
function put_h264_qpel8_hv_lowpass_neon_top
|
|
lowpass_const w12
|
|
ld1 {v16.8h}, [x1], x3
|
|
ld1 {v17.8h}, [x1], x3
|
|
ld1 {v18.8h}, [x1], x3
|
|
ld1 {v19.8h}, [x1], x3
|
|
ld1 {v20.8h}, [x1], x3
|
|
ld1 {v21.8h}, [x1], x3
|
|
ld1 {v22.8h}, [x1], x3
|
|
ld1 {v23.8h}, [x1], x3
|
|
ld1 {v24.8h}, [x1], x3
|
|
ld1 {v25.8h}, [x1], x3
|
|
ld1 {v26.8h}, [x1], x3
|
|
ld1 {v27.8h}, [x1], x3
|
|
ld1 {v28.8h}, [x1]
|
|
lowpass_8H v16, v17
|
|
lowpass_8H v18, v19
|
|
lowpass_8H v20, v21
|
|
lowpass_8H v22, v23
|
|
lowpass_8H v24, v25
|
|
lowpass_8H v26, v27
|
|
lowpass_8H v28, v29
|
|
|
|
lowpass_8.16 v16, v17, v18, v19, v20, v21
|
|
lowpass_8.16 v17, v18, v19, v20, v21, v22
|
|
|
|
lowpass_8.16 v18, v19, v20, v21, v22, v23
|
|
lowpass_8.16 v19, v20, v21, v22, v23, v24
|
|
|
|
lowpass_8.16 v20, v21, v22, v23, v24, v25
|
|
lowpass_8.16 v21, v22, v23, v24, v25, v26
|
|
|
|
lowpass_8.16 v22, v23, v24, v25, v26, v27
|
|
lowpass_8.16 v23, v24, v25, v26, v27, v28
|
|
|
|
ret
|
|
endfunc
|
|
|
|
.macro h264_qpel8_hv_lowpass type
|
|
function \type\()_h264_qpel8_hv_lowpass_neon
|
|
mov x10, x30
|
|
bl put_h264_qpel8_hv_lowpass_neon_top
|
|
.ifc \type,avg
|
|
ld1 {v0.8b}, [x0], x2
|
|
ld1 {v1.8b}, [x0], x2
|
|
ld1 {v2.8b}, [x0], x2
|
|
urhadd v16.8b, v16.8b, v0.8b
|
|
ld1 {v3.8b}, [x0], x2
|
|
urhadd v17.8b, v17.8b, v1.8b
|
|
ld1 {v4.8b}, [x0], x2
|
|
urhadd v18.8b, v18.8b, v2.8b
|
|
ld1 {v5.8b}, [x0], x2
|
|
urhadd v19.8b, v19.8b, v3.8b
|
|
ld1 {v6.8b}, [x0], x2
|
|
urhadd v20.8b, v20.8b, v4.8b
|
|
ld1 {v7.8b}, [x0], x2
|
|
urhadd v21.8b, v21.8b, v5.8b
|
|
urhadd v22.8b, v22.8b, v6.8b
|
|
urhadd v23.8b, v23.8b, v7.8b
|
|
sub x0, x0, x2, lsl #3
|
|
.endif
|
|
|
|
st1 {v16.8b}, [x0], x2
|
|
st1 {v17.8b}, [x0], x2
|
|
st1 {v18.8b}, [x0], x2
|
|
st1 {v19.8b}, [x0], x2
|
|
st1 {v20.8b}, [x0], x2
|
|
st1 {v21.8b}, [x0], x2
|
|
st1 {v22.8b}, [x0], x2
|
|
st1 {v23.8b}, [x0], x2
|
|
|
|
ret x10
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel8_hv_lowpass put
|
|
h264_qpel8_hv_lowpass avg
|
|
|
|
.macro h264_qpel8_hv_lowpass_l2 type
|
|
function \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
mov x10, x30
|
|
bl put_h264_qpel8_hv_lowpass_neon_top
|
|
|
|
ld1 {v0.8b, v1.8b}, [x2], #16
|
|
ld1 {v2.8b, v3.8b}, [x2], #16
|
|
urhadd v0.8b, v0.8b, v16.8b
|
|
urhadd v1.8b, v1.8b, v17.8b
|
|
ld1 {v4.8b, v5.8b}, [x2], #16
|
|
urhadd v2.8b, v2.8b, v18.8b
|
|
urhadd v3.8b, v3.8b, v19.8b
|
|
ld1 {v6.8b, v7.8b}, [x2], #16
|
|
urhadd v4.8b, v4.8b, v20.8b
|
|
urhadd v5.8b, v5.8b, v21.8b
|
|
urhadd v6.8b, v6.8b, v22.8b
|
|
urhadd v7.8b, v7.8b, v23.8b
|
|
.ifc \type,avg
|
|
ld1 {v16.8b}, [x0], x3
|
|
ld1 {v17.8b}, [x0], x3
|
|
ld1 {v18.8b}, [x0], x3
|
|
urhadd v0.8b, v0.8b, v16.8b
|
|
ld1 {v19.8b}, [x0], x3
|
|
urhadd v1.8b, v1.8b, v17.8b
|
|
ld1 {v20.8b}, [x0], x3
|
|
urhadd v2.8b, v2.8b, v18.8b
|
|
ld1 {v21.8b}, [x0], x3
|
|
urhadd v3.8b, v3.8b, v19.8b
|
|
ld1 {v22.8b}, [x0], x3
|
|
urhadd v4.8b, v4.8b, v20.8b
|
|
ld1 {v23.8b}, [x0], x3
|
|
urhadd v5.8b, v5.8b, v21.8b
|
|
urhadd v6.8b, v6.8b, v22.8b
|
|
urhadd v7.8b, v7.8b, v23.8b
|
|
sub x0, x0, x3, lsl #3
|
|
.endif
|
|
st1 {v0.8b}, [x0], x3
|
|
st1 {v1.8b}, [x0], x3
|
|
st1 {v2.8b}, [x0], x3
|
|
st1 {v3.8b}, [x0], x3
|
|
st1 {v4.8b}, [x0], x3
|
|
st1 {v5.8b}, [x0], x3
|
|
st1 {v6.8b}, [x0], x3
|
|
st1 {v7.8b}, [x0], x3
|
|
|
|
ret x10
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel8_hv_lowpass_l2 put
|
|
h264_qpel8_hv_lowpass_l2 avg
|
|
|
|
.macro h264_qpel16_hv type
|
|
function \type\()_h264_qpel16_hv_lowpass_neon
|
|
mov x13, x30
|
|
bl \type\()_h264_qpel8_hv_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
bl \type\()_h264_qpel8_hv_lowpass_neon
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #8
|
|
sub x0, x0, x2, lsl #4
|
|
add x0, x0, #8
|
|
bl \type\()_h264_qpel8_hv_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x13
|
|
b \type\()_h264_qpel8_hv_lowpass_neon
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel16_hv_lowpass_l2_neon
|
|
mov x13, x30
|
|
sub x2, x4, #256
|
|
bl \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
sub x1, x1, x3, lsl #2
|
|
bl \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #8
|
|
sub x0, x0, x3, lsl #4
|
|
add x0, x0, #8
|
|
bl \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x13
|
|
b \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel16_hv put
|
|
h264_qpel16_hv avg
|
|
|
|
.macro h264_qpel8 type
|
|
function ff_\type\()_h264_qpel8_mc10_neon, export=1
|
|
lowpass_const w3
|
|
mov x3, x1
|
|
sub x1, x1, #2
|
|
mov x12, #8
|
|
b \type\()_h264_qpel8_h_lowpass_l2_neon
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc20_neon, export=1
|
|
lowpass_const w3
|
|
sub x1, x1, #2
|
|
mov x3, x2
|
|
mov x12, #8
|
|
b \type\()_h264_qpel8_h_lowpass_neon
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc30_neon, export=1
|
|
lowpass_const w3
|
|
add x3, x1, #1
|
|
sub x1, x1, #2
|
|
mov x12, #8
|
|
b \type\()_h264_qpel8_h_lowpass_l2_neon
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc01_neon, export=1
|
|
mov x14, x30
|
|
mov x12, x1
|
|
\type\()_h264_qpel8_mc01:
|
|
lowpass_const w3
|
|
mov x3, x2
|
|
sub x1, x1, x2, lsl #1
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc11_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel8_mc11:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #64
|
|
mov x0, sp
|
|
sub x1, x1, #2
|
|
mov x3, #8
|
|
mov x12, #8
|
|
bl put_h264_qpel8_h_lowpass_neon
|
|
mov x0, x8
|
|
mov x3, x2
|
|
mov x12, sp
|
|
sub x1, x9, x2, lsl #1
|
|
mov x2, #8
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc21_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel8_mc21:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #(8*8+16*12)
|
|
sub x1, x1, #2
|
|
mov x3, #8
|
|
mov x0, sp
|
|
mov x12, #8
|
|
bl put_h264_qpel8_h_lowpass_neon
|
|
mov x4, x0
|
|
mov x0, x8
|
|
sub x1, x9, x2, lsl #1
|
|
sub x1, x1, #2
|
|
mov x3, x2
|
|
sub x2, x4, #64
|
|
bl \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc31_neon, export=1
|
|
add x1, x1, #1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
sub x1, x1, #1
|
|
b \type\()_h264_qpel8_mc11
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc02_neon, export=1
|
|
mov x14, x30
|
|
lowpass_const w3
|
|
sub x1, x1, x2, lsl #1
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel8_v_lowpass_neon
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc12_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel8_mc12:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #(8*8+16*12)
|
|
sub x1, x1, x2, lsl #1
|
|
mov x3, x2
|
|
mov x2, #8
|
|
mov x0, sp
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
mov x4, x0
|
|
mov x0, x8
|
|
sub x1, x9, x3, lsl #1
|
|
sub x1, x1, #2
|
|
sub x2, x4, #64
|
|
bl \type\()_h264_qpel8_hv_lowpass_l2_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc22_neon, export=1
|
|
mov x14, x30
|
|
mov x11, sp
|
|
sub x1, x1, x2, lsl #1
|
|
sub x1, x1, #2
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel8_hv_lowpass_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc32_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, #1
|
|
b \type\()_h264_qpel8_mc12
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc03_neon, export=1
|
|
mov x14, x30
|
|
add x12, x1, x2
|
|
b \type\()_h264_qpel8_mc01
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc13_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
b \type\()_h264_qpel8_mc11
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc23_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
b \type\()_h264_qpel8_mc21
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc33_neon, export=1
|
|
add x1, x1, #1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
sub x1, x1, #1
|
|
b \type\()_h264_qpel8_mc11
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel8 put
|
|
h264_qpel8 avg
|
|
|
|
.macro h264_qpel16 type
|
|
function ff_\type\()_h264_qpel16_mc10_neon, export=1
|
|
lowpass_const w3
|
|
mov x3, x1
|
|
sub x1, x1, #2
|
|
b \type\()_h264_qpel16_h_lowpass_l2_neon
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc20_neon, export=1
|
|
lowpass_const w3
|
|
sub x1, x1, #2
|
|
mov x3, x2
|
|
b \type\()_h264_qpel16_h_lowpass_neon
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc30_neon, export=1
|
|
lowpass_const w3
|
|
add x3, x1, #1
|
|
sub x1, x1, #2
|
|
b \type\()_h264_qpel16_h_lowpass_l2_neon
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc01_neon, export=1
|
|
mov x14, x30
|
|
mov x12, x1
|
|
\type\()_h264_qpel16_mc01:
|
|
lowpass_const w3
|
|
mov x3, x2
|
|
sub x1, x1, x2, lsl #1
|
|
bl \type\()_h264_qpel16_v_lowpass_l2_neon
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc11_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel16_mc11:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #256
|
|
mov x0, sp
|
|
sub x1, x1, #2
|
|
mov x3, #16
|
|
bl put_h264_qpel16_h_lowpass_neon
|
|
mov x0, x8
|
|
mov x3, x2
|
|
mov x12, sp
|
|
sub x1, x9, x2, lsl #1
|
|
mov x2, #16
|
|
bl \type\()_h264_qpel16_v_lowpass_l2_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc21_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel16_mc21:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #(16*16+16*12)
|
|
sub x1, x1, #2
|
|
mov x0, sp
|
|
bl put_h264_qpel16_h_lowpass_neon_packed
|
|
mov x4, x0
|
|
mov x0, x8
|
|
sub x1, x9, x2, lsl #1
|
|
sub x1, x1, #2
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel16_hv_lowpass_l2_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc31_neon, export=1
|
|
add x1, x1, #1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
sub x1, x1, #1
|
|
b \type\()_h264_qpel16_mc11
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc02_neon, export=1
|
|
mov x14, x30
|
|
lowpass_const w3
|
|
sub x1, x1, x2, lsl #1
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel16_v_lowpass_neon
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc12_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel16_mc12:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #(16*16+16*12)
|
|
sub x1, x1, x2, lsl #1
|
|
mov x0, sp
|
|
mov x3, x2
|
|
bl put_h264_qpel16_v_lowpass_neon_packed
|
|
mov x4, x0
|
|
mov x0, x8
|
|
sub x1, x9, x3, lsl #1
|
|
sub x1, x1, #2
|
|
mov x2, x3
|
|
bl \type\()_h264_qpel16_hv_lowpass_l2_neon
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc22_neon, export=1
|
|
mov x14, x30
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub x1, x1, x2, lsl #1
|
|
sub x1, x1, #2
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel16_hv_lowpass_neon
|
|
mov sp, x11 // restore stack
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc32_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, #1
|
|
b \type\()_h264_qpel16_mc12
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc03_neon, export=1
|
|
mov x14, x30
|
|
add x12, x1, x2
|
|
b \type\()_h264_qpel16_mc01
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc13_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
b \type\()_h264_qpel16_mc11
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc23_neon, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
b \type\()_h264_qpel16_mc21
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc33_neon, export=1
|
|
add x1, x1, #1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
sub x1, x1, #1
|
|
b \type\()_h264_qpel16_mc11
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel16 put
|
|
h264_qpel16 avg
|
|
|
|
//trashes v0-v5
|
|
.macro lowpass_8_10 r0, r1, r2, r3, d0, d1
|
|
ext v2.16b, \r0\().16b, \r1\().16b, #4
|
|
ext v3.16b, \r0\().16b, \r1\().16b, #6
|
|
add v2.8h, v2.8h, v3.8h
|
|
ext v4.16b, \r0\().16b, \r1\().16b, #2
|
|
ext v5.16b, \r0\().16b, \r1\().16b, #8
|
|
add v4.8h, v4.8h, v5.8h
|
|
ext v1.16b, \r0\().16b, \r1\().16b, #10
|
|
|
|
add \d0\().8h, \r0\().8h, v1.8h
|
|
ext v0.16b, \r2\().16b, \r3\().16b, #4
|
|
mla \d0\().8h, v2.8h, v6.h[1]
|
|
ext v1.16b, \r2\().16b, \r3\().16b, #6
|
|
add v0.8h, v0.8h, v1.8h
|
|
ext v1.16b, \r2\().16b, \r3\().16b, #2
|
|
mul v5.8h, v4.8h, v6.h[0]
|
|
uqsub \d0\().8h, \d0\().8h, v5.8h
|
|
urshr \d0\().8h, \d0\().8h, #5
|
|
|
|
ext v3.16b, \r2\().16b, \r3\().16b, #8
|
|
add v1.8h, v1.8h, v3.8h
|
|
ext v2.16b, \r2\().16b, \r3\().16b, #10
|
|
|
|
add \d1\().8h, \r2\().8h, v2.8h
|
|
mla \d1\().8h, v0.8h, v6.h[1]
|
|
mul v5.8h, v1.8h, v6.h[0]
|
|
uqsub \d1\().8h, \d1\().8h, v5.8h
|
|
mvni v5.8h, #0xFC, lsl #8 // 1023 for clipping
|
|
urshr \d1\().8h, \d1\().8h, #5
|
|
|
|
umin \d0\().8h, \d0\().8h, v5.8h
|
|
umin \d1\().8h, \d1\().8h, v5.8h
|
|
.endm
|
|
|
|
//trashes v0-v4
|
|
.macro lowpass_8_10_v r0, r1, r2, r3, r4, r5, r6, d0, d1
|
|
add v2.8h, \r2\().8h, \r3\().8h
|
|
add v0.8h, \r3\().8h, \r4\().8h
|
|
add v4.8h, \r1\().8h, \r4\().8h
|
|
add v1.8h, \r2\().8h, \r5\().8h
|
|
|
|
add \d0\().8h, \r0\().8h, \r5\().8h
|
|
add \d1\().8h, \r1\().8h, \r6\().8h
|
|
mla \d0\().8h, v2.8h, v6.h[1]
|
|
mla \d1\().8h, v0.8h, v6.h[1]
|
|
mul v2.8h, v4.8h, v6.h[0]
|
|
mul v0.8h, v1.8h, v6.h[0]
|
|
uqsub \d0\().8h, \d0\().8h, v2.8h
|
|
uqsub \d1\().8h, \d1\().8h, v0.8h
|
|
|
|
mvni v0.8h, #0xFC, lsl #8 // 1023 for clipping
|
|
|
|
urshr \d0\().8h, \d0\().8h, #5
|
|
urshr \d1\().8h, \d1\().8h, #5
|
|
|
|
umin \d0\().8h, \d0\().8h, v0.8h
|
|
umin \d1\().8h, \d1\().8h, v0.8h
|
|
.endm
|
|
|
|
function put_h264_qpel16_h_lowpass_neon_packed_10
|
|
mov x4, x30
|
|
mov x12, #32
|
|
mov x3, #16
|
|
bl put_h264_qpel8_h_lowpass_neon_10
|
|
sub x1, x1, x2, lsl #4
|
|
add x1, x1, #16
|
|
mov x12, #32
|
|
mov x30, x4
|
|
b put_h264_qpel8_h_lowpass_neon_10
|
|
endfunc
|
|
|
|
.macro h264_qpel_h_lowpass_10 type
|
|
function \type\()_h264_qpel16_h_lowpass_neon_10
|
|
mov x13, x30
|
|
mov x12, #32
|
|
bl \type\()_h264_qpel8_h_lowpass_neon_10
|
|
sub x0, x0, x3, lsl #4
|
|
sub x1, x1, x2, lsl #4
|
|
add x0, x0, #16
|
|
add x1, x1, #16
|
|
mov x12, #32
|
|
mov x30, x13
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_h_lowpass_neon_10
|
|
1: ld1 {v28.8h, v29.8h}, [x1], x2
|
|
ld1 {v16.8h, v17.8h}, [x1], x2
|
|
subs x12, x12, #4
|
|
lowpass_8_10 v28, v29, v16, v17, v28, v20
|
|
.ifc \type,avg
|
|
ld1 {v2.8h}, [x0], x3
|
|
ld1 {v3.8h}, [x0]
|
|
urhadd v28.8h, v28.8h, v2.8h
|
|
urhadd v20.8h, v20.8h, v3.8h
|
|
sub x0, x0, x3
|
|
.endif
|
|
st1 {v28.8h}, [x0], x3
|
|
st1 {v20.8h}, [x0], x3
|
|
b.ne 1b
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_h_lowpass_10 put
|
|
h264_qpel_h_lowpass_10 avg
|
|
|
|
.macro h264_qpel_h_lowpass_l2_10 type
|
|
function \type\()_h264_qpel16_h_lowpass_l2_neon_10
|
|
mov x13, x30
|
|
mov x12, #32
|
|
bl \type\()_h264_qpel8_h_lowpass_l2_neon_10
|
|
sub x0, x0, x2, lsl #4
|
|
sub x1, x1, x2, lsl #4
|
|
sub x3, x3, x2, lsl #4
|
|
add x0, x0, #16
|
|
add x1, x1, #16
|
|
add x3, x3, #16
|
|
mov x12, #32
|
|
mov x30, x13
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_h_lowpass_l2_neon_10
|
|
1: ld1 {v26.8h, v27.8h}, [x1], x2
|
|
ld1 {v16.8h, v17.8h}, [x1], x2
|
|
ld1 {v28.8h}, [x3], x2
|
|
ld1 {v29.8h}, [x3], x2
|
|
subs x12, x12, #4
|
|
lowpass_8_10 v26, v27, v16, v17, v26, v27
|
|
urhadd v26.8h, v26.8h, v28.8h
|
|
urhadd v27.8h, v27.8h, v29.8h
|
|
.ifc \type,avg
|
|
ld1 {v2.8h}, [x0], x2
|
|
ld1 {v3.8h}, [x0]
|
|
urhadd v26.8h, v26.8h, v2.8h
|
|
urhadd v27.8h, v27.8h, v3.8h
|
|
sub x0, x0, x2
|
|
.endif
|
|
st1 {v26.8h}, [x0], x2
|
|
st1 {v27.8h}, [x0], x2
|
|
b.ne 1b
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_h_lowpass_l2_10 put
|
|
h264_qpel_h_lowpass_l2_10 avg
|
|
|
|
function put_h264_qpel16_v_lowpass_neon_packed_10
|
|
mov x4, x30
|
|
mov x2, #8
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #8
|
|
bl put_h264_qpel8_v_lowpass_neon
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x4
|
|
b put_h264_qpel8_v_lowpass_neon
|
|
endfunc
|
|
|
|
.macro h264_qpel_v_lowpass_10 type
|
|
function \type\()_h264_qpel16_v_lowpass_neon_10
|
|
mov x4, x30
|
|
bl \type\()_h264_qpel8_v_lowpass_neon_10
|
|
sub x1, x1, x3, lsl #2
|
|
bl \type\()_h264_qpel8_v_lowpass_neon_10
|
|
sub x0, x0, x2, lsl #4
|
|
add x0, x0, #16
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #16
|
|
bl \type\()_h264_qpel8_v_lowpass_neon_10
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x4
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_v_lowpass_neon_10
|
|
ld1 {v16.8h}, [x1], x3
|
|
ld1 {v17.8h}, [x1], x3
|
|
ld1 {v18.8h}, [x1], x3
|
|
ld1 {v19.8h}, [x1], x3
|
|
ld1 {v20.8h}, [x1], x3
|
|
ld1 {v21.8h}, [x1], x3
|
|
ld1 {v22.8h}, [x1], x3
|
|
ld1 {v23.8h}, [x1], x3
|
|
ld1 {v24.8h}, [x1], x3
|
|
ld1 {v25.8h}, [x1], x3
|
|
ld1 {v26.8h}, [x1], x3
|
|
ld1 {v27.8h}, [x1], x3
|
|
ld1 {v28.8h}, [x1]
|
|
|
|
lowpass_8_10_v v16, v17, v18, v19, v20, v21, v22, v16, v17
|
|
lowpass_8_10_v v18, v19, v20, v21, v22, v23, v24, v18, v19
|
|
lowpass_8_10_v v20, v21, v22, v23, v24, v25, v26, v20, v21
|
|
lowpass_8_10_v v22, v23, v24, v25, v26, v27, v28, v22, v23
|
|
|
|
.ifc \type,avg
|
|
ld1 {v24.8h}, [x0], x2
|
|
ld1 {v25.8h}, [x0], x2
|
|
ld1 {v26.8h}, [x0], x2
|
|
urhadd v16.8h, v16.8h, v24.8h
|
|
ld1 {v27.8h}, [x0], x2
|
|
urhadd v17.8h, v17.8h, v25.8h
|
|
ld1 {v28.8h}, [x0], x2
|
|
urhadd v18.8h, v18.8h, v26.8h
|
|
ld1 {v29.8h}, [x0], x2
|
|
urhadd v19.8h, v19.8h, v27.8h
|
|
ld1 {v30.8h}, [x0], x2
|
|
urhadd v20.8h, v20.8h, v28.8h
|
|
ld1 {v31.8h}, [x0], x2
|
|
urhadd v21.8h, v21.8h, v29.8h
|
|
urhadd v22.8h, v22.8h, v30.8h
|
|
urhadd v23.8h, v23.8h, v31.8h
|
|
sub x0, x0, x2, lsl #3
|
|
.endif
|
|
|
|
st1 {v16.8h}, [x0], x2
|
|
st1 {v17.8h}, [x0], x2
|
|
st1 {v18.8h}, [x0], x2
|
|
st1 {v19.8h}, [x0], x2
|
|
st1 {v20.8h}, [x0], x2
|
|
st1 {v21.8h}, [x0], x2
|
|
st1 {v22.8h}, [x0], x2
|
|
st1 {v23.8h}, [x0], x2
|
|
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_v_lowpass_10 put
|
|
h264_qpel_v_lowpass_10 avg
|
|
|
|
.macro h264_qpel_v_lowpass_l2_10 type
|
|
function \type\()_h264_qpel16_v_lowpass_l2_neon_10
|
|
mov x4, x30
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon_10
|
|
sub x1, x1, x3, lsl #2
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon_10
|
|
sub x0, x0, x3, lsl #4
|
|
sub x12, x12, x2, lsl #4
|
|
add x0, x0, #16
|
|
add x12, x12, #16
|
|
sub x1, x1, x3, lsl #4
|
|
sub x1, x1, x3, lsl #2
|
|
add x1, x1, #16
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon_10
|
|
sub x1, x1, x3, lsl #2
|
|
mov x30, x4
|
|
endfunc
|
|
|
|
function \type\()_h264_qpel8_v_lowpass_l2_neon_10
|
|
ld1 {v16.8h}, [x1], x3
|
|
ld1 {v17.8h}, [x1], x3
|
|
ld1 {v18.8h}, [x1], x3
|
|
ld1 {v19.8h}, [x1], x3
|
|
ld1 {v20.8h}, [x1], x3
|
|
ld1 {v21.8h}, [x1], x3
|
|
ld1 {v22.8h}, [x1], x3
|
|
ld1 {v23.8h}, [x1], x3
|
|
ld1 {v24.8h}, [x1], x3
|
|
ld1 {v25.8h}, [x1], x3
|
|
ld1 {v26.8h}, [x1], x3
|
|
ld1 {v27.8h}, [x1], x3
|
|
ld1 {v28.8h}, [x1]
|
|
|
|
lowpass_8_10_v v16, v17, v18, v19, v20, v21, v22, v16, v17
|
|
lowpass_8_10_v v18, v19, v20, v21, v22, v23, v24, v18, v19
|
|
lowpass_8_10_v v20, v21, v22, v23, v24, v25, v26, v20, v21
|
|
lowpass_8_10_v v22, v23, v24, v25, v26, v27, v28, v22, v23
|
|
|
|
ld1 {v24.8h}, [x12], x2
|
|
ld1 {v25.8h}, [x12], x2
|
|
ld1 {v26.8h}, [x12], x2
|
|
ld1 {v27.8h}, [x12], x2
|
|
ld1 {v28.8h}, [x12], x2
|
|
urhadd v16.8h, v24.8h, v16.8h
|
|
urhadd v17.8h, v25.8h, v17.8h
|
|
ld1 {v29.8h}, [x12], x2
|
|
urhadd v18.8h, v26.8h, v18.8h
|
|
urhadd v19.8h, v27.8h, v19.8h
|
|
ld1 {v30.8h}, [x12], x2
|
|
urhadd v20.8h, v28.8h, v20.8h
|
|
urhadd v21.8h, v29.8h, v21.8h
|
|
ld1 {v31.8h}, [x12], x2
|
|
urhadd v22.8h, v30.8h, v22.8h
|
|
urhadd v23.8h, v31.8h, v23.8h
|
|
|
|
.ifc \type,avg
|
|
ld1 {v24.8h}, [x0], x3
|
|
ld1 {v25.8h}, [x0], x3
|
|
ld1 {v26.8h}, [x0], x3
|
|
urhadd v16.8h, v16.8h, v24.8h
|
|
ld1 {v27.8h}, [x0], x3
|
|
urhadd v17.8h, v17.8h, v25.8h
|
|
ld1 {v28.8h}, [x0], x3
|
|
urhadd v18.8h, v18.8h, v26.8h
|
|
ld1 {v29.8h}, [x0], x3
|
|
urhadd v19.8h, v19.8h, v27.8h
|
|
ld1 {v30.8h}, [x0], x3
|
|
urhadd v20.8h, v20.8h, v28.8h
|
|
ld1 {v31.8h}, [x0], x3
|
|
urhadd v21.8h, v21.8h, v29.8h
|
|
urhadd v22.8h, v22.8h, v30.8h
|
|
urhadd v23.8h, v23.8h, v31.8h
|
|
sub x0, x0, x3, lsl #3
|
|
.endif
|
|
|
|
st1 {v16.8h}, [x0], x3
|
|
st1 {v17.8h}, [x0], x3
|
|
st1 {v18.8h}, [x0], x3
|
|
st1 {v19.8h}, [x0], x3
|
|
st1 {v20.8h}, [x0], x3
|
|
st1 {v21.8h}, [x0], x3
|
|
st1 {v22.8h}, [x0], x3
|
|
st1 {v23.8h}, [x0], x3
|
|
|
|
ret
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel_v_lowpass_l2_10 put
|
|
h264_qpel_v_lowpass_l2_10 avg
|
|
|
|
.macro h264_qpel8_10 type
|
|
function ff_\type\()_h264_qpel8_mc10_neon_10, export=1
|
|
lowpass_const w3
|
|
mov x3, x1
|
|
sub x1, x1, #4
|
|
mov x12, #16
|
|
b \type\()_h264_qpel8_h_lowpass_l2_neon_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc20_neon_10, export=1
|
|
lowpass_const w3
|
|
sub x1, x1, #4
|
|
mov x3, x2
|
|
mov x12, #16
|
|
b \type\()_h264_qpel8_h_lowpass_neon_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc30_neon_10, export=1
|
|
lowpass_const w3
|
|
add x3, x1, #2
|
|
sub x1, x1, #4
|
|
mov x12, #16
|
|
b \type\()_h264_qpel8_h_lowpass_l2_neon_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc01_neon_10, export=1
|
|
mov x14, x30
|
|
mov x12, x1
|
|
\type\()_h264_qpel8_mc01_10:
|
|
lowpass_const w3
|
|
mov x3, x2
|
|
sub x1, x1, x2, lsl #1
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon_10
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc11_neon_10, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel8_mc11_10:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #128
|
|
mov x0, sp
|
|
sub x1, x1, #4
|
|
mov x3, #16
|
|
mov x12, #16
|
|
bl put_h264_qpel8_h_lowpass_neon_10
|
|
mov x0, x8
|
|
mov x3, x2
|
|
mov x12, sp
|
|
sub x1, x9, x2, lsl #1
|
|
mov x2, #16
|
|
bl \type\()_h264_qpel8_v_lowpass_l2_neon_10
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc31_neon_10, export=1
|
|
add x1, x1, #2
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
sub x1, x1, #2
|
|
b \type\()_h264_qpel8_mc11_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc02_neon_10, export=1
|
|
mov x14, x30
|
|
lowpass_const w3
|
|
sub x1, x1, x2, lsl #1
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel8_v_lowpass_neon_10
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc03_neon_10, export=1
|
|
mov x14, x30
|
|
add x12, x1, x2
|
|
b \type\()_h264_qpel8_mc01_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc13_neon_10, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
b \type\()_h264_qpel8_mc11_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel8_mc33_neon_10, export=1
|
|
add x1, x1, #2
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
sub x1, x1, #2
|
|
b \type\()_h264_qpel8_mc11_10
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel8_10 put
|
|
h264_qpel8_10 avg
|
|
|
|
.macro h264_qpel16_10 type
|
|
function ff_\type\()_h264_qpel16_mc10_neon_10, export=1
|
|
lowpass_const w3
|
|
mov x3, x1
|
|
sub x1, x1, #4
|
|
b \type\()_h264_qpel16_h_lowpass_l2_neon_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc20_neon_10, export=1
|
|
lowpass_const w3
|
|
sub x1, x1, #4
|
|
mov x3, x2
|
|
b \type\()_h264_qpel16_h_lowpass_neon_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc30_neon_10, export=1
|
|
lowpass_const w3
|
|
add x3, x1, #2
|
|
sub x1, x1, #4
|
|
b \type\()_h264_qpel16_h_lowpass_l2_neon_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc01_neon_10, export=1
|
|
mov x14, x30
|
|
mov x12, x1
|
|
\type\()_h264_qpel16_mc01_10:
|
|
lowpass_const w3
|
|
mov x3, x2
|
|
sub x1, x1, x2, lsl #1
|
|
bl \type\()_h264_qpel16_v_lowpass_l2_neon_10
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc11_neon_10, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
\type\()_h264_qpel16_mc11_10:
|
|
lowpass_const w3
|
|
mov x11, sp
|
|
sub sp, sp, #512
|
|
mov x0, sp
|
|
sub x1, x1, #4
|
|
mov x3, #32
|
|
bl put_h264_qpel16_h_lowpass_neon_10
|
|
mov x0, x8
|
|
mov x3, x2
|
|
mov x12, sp
|
|
sub x1, x9, x2, lsl #1
|
|
mov x2, #32
|
|
bl \type\()_h264_qpel16_v_lowpass_l2_neon_10
|
|
mov sp, x11
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc31_neon_10, export=1
|
|
add x1, x1, #2
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
sub x1, x1, #2
|
|
b \type\()_h264_qpel16_mc11_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc02_neon_10, export=1
|
|
mov x14, x30
|
|
lowpass_const w3
|
|
sub x1, x1, x2, lsl #1
|
|
mov x3, x2
|
|
bl \type\()_h264_qpel16_v_lowpass_neon_10
|
|
ret x14
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc03_neon_10, export=1
|
|
mov x14, x30
|
|
add x12, x1, x2
|
|
b \type\()_h264_qpel16_mc01_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc13_neon_10, export=1
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
b \type\()_h264_qpel16_mc11_10
|
|
endfunc
|
|
|
|
function ff_\type\()_h264_qpel16_mc33_neon_10, export=1
|
|
add x1, x1, #2
|
|
mov x14, x30
|
|
mov x8, x0
|
|
mov x9, x1
|
|
add x1, x1, x2
|
|
sub x1, x1, #2
|
|
b \type\()_h264_qpel16_mc11_10
|
|
endfunc
|
|
.endm
|
|
|
|
h264_qpel16_10 put
|
|
h264_qpel16_10 avg
|