|
|
|
/*
|
|
|
|
* Copyright (c) 2012 Mans Rullgard
|
|
|
|
*
|
|
|
|
* 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/arm/asm.S"
|
|
|
|
|
|
|
|
function ff_ps_add_squares_neon, export=1
|
|
|
|
mov r3, r0
|
|
|
|
sub r2, r2, #4
|
|
|
|
vld1.32 {q0}, [r1,:128]!
|
|
|
|
vmul.f32 q0, q0, q0
|
|
|
|
vld1.32 {q2}, [r1,:128]!
|
|
|
|
vmul.f32 q2, q2, q2
|
|
|
|
vld1.32 {q1}, [r0,:128]!
|
|
|
|
1:
|
|
|
|
vpadd.f32 d6, d0, d1
|
|
|
|
vld1.32 {q0}, [r1,:128]!
|
|
|
|
vpadd.f32 d7, d4, d5
|
|
|
|
vmul.f32 q0, q0, q0
|
|
|
|
vld1.32 {q2}, [r1,:128]!
|
|
|
|
vadd.f32 q3, q1, q3
|
|
|
|
vld1.32 {q1}, [r0,:128]!
|
|
|
|
vmul.f32 q2, q2, q2
|
|
|
|
vst1.32 {q3}, [r3,:128]!
|
|
|
|
subs r2, r2, #4
|
|
|
|
bgt 1b
|
|
|
|
vpadd.f32 d6, d0, d1
|
|
|
|
vpadd.f32 d7, d4, d5
|
|
|
|
vadd.f32 q1, q1, q3
|
|
|
|
vst1.32 {q1}, [r3,:128]!
|
|
|
|
bx lr
|
|
|
|
endfunc
|
|
|
|
|
|
|
|
function ff_ps_mul_pair_single_neon, export=1
|
|
|
|
sub r3, r3, #4
|
|
|
|
tst r1, #8
|
|
|
|
bne 2f
|
|
|
|
vld1.32 {q0}, [r1,:128]!
|
|
|
|
1:
|
|
|
|
vld1.32 {q3}, [r2,:128]!
|
|
|
|
vmul.f32 d4, d0, d6[0]
|
|
|
|
vmul.f32 d5, d1, d6[1]
|
|
|
|
vld1.32 {q1}, [r1,:128]!
|
|
|
|
vmul.f32 d6, d2, d7[0]
|
|
|
|
vmul.f32 d7, d3, d7[1]
|
|
|
|
vld1.32 {q0}, [r1,:128]!
|
|
|
|
vst1.32 {q2,q3}, [r0,:128]!
|
|
|
|
subs r3, r3, #4
|
|
|
|
bgt 1b
|
|
|
|
vld1.32 {q3}, [r2,:128]!
|
|
|
|
vmul.f32 d4, d0, d6[0]
|
|
|
|
vmul.f32 d5, d1, d6[1]
|
|
|
|
vld1.32 {q1}, [r1,:128]!
|
|
|
|
vmul.f32 d6, d2, d7[0]
|
|
|
|
vmul.f32 d7, d3, d7[1]
|
|
|
|
vst1.32 {q2,q3}, [r0,:128]!
|
|
|
|
bx lr
|
|
|
|
2:
|
|
|
|
vld1.32 {d0}, [r1,:64]!
|
|
|
|
vld1.32 {d1,d2}, [r1,:128]!
|
|
|
|
1:
|
|
|
|
vld1.32 {q3}, [r2,:128]!
|
|
|
|
vmul.f32 d4, d0, d6[0]
|
|
|
|
vmul.f32 d5, d1, d6[1]
|
|
|
|
vld1.32 {d0,d1}, [r1,:128]!
|
|
|
|
vmul.f32 d6, d2, d7[0]
|
|
|
|
vmul.f32 d7, d0, d7[1]
|
|
|
|
vmov d0, d1
|
|
|
|
vld1.32 {d1,d2}, [r1,:128]!
|
|
|
|
vst1.32 {q2,q3}, [r0,:128]!
|
|
|
|
subs r3, r3, #4
|
|
|
|
bgt 1b
|
|
|
|
vld1.32 {q3}, [r2,:128]!
|
|
|
|
vmul.f32 d4, d0, d6[0]
|
|
|
|
vmul.f32 d5, d1, d6[1]
|
|
|
|
vld1.32 {d0}, [r1,:64]!
|
|
|
|
vmul.f32 d6, d2, d7[0]
|
|
|
|
vmul.f32 d7, d0, d7[1]
|
|
|
|
vst1.32 {q2,q3}, [r0,:128]!
|
|
|
|
bx lr
|
|
|
|
endfunc
|
|
|
|
|
|
|
|
function ff_ps_hybrid_synthesis_deint_neon, export=1
|
|
|
|
push {r4-r8,lr}
|
|
|
|
add r0, r0, r2, lsl #2
|
|
|
|
add r1, r1, r2, lsl #5+1+2
|
|
|
|
rsb r2, r2, #64
|
|
|
|
mov r5, #64*4
|
|
|
|
mov lr, r0
|
|
|
|
add r4, r0, #38*64*4
|
|
|
|
mov r12, r3
|
|
|
|
2:
|
|
|
|
vld1.32 {d0,d1}, [r1,:128]!
|
|
|
|
vst1.32 {d0[0]}, [lr,:32], r5
|
|
|
|
vst1.32 {d0[1]}, [r4,:32], r5
|
|
|
|
vst1.32 {d1[0]}, [lr,:32], r5
|
|
|
|
vst1.32 {d1[1]}, [r4,:32], r5
|
|
|
|
subs r12, r12, #2
|
|
|
|
bgt 2b
|
|
|
|
add r0, r0, #4
|
|
|
|
sub r2, r2, #1
|
|
|
|
tst r2, #2
|
|
|
|
bne 6f
|
|
|
|
1:
|
|
|
|
mov lr, r0
|
|
|
|
add r4, r0, #38*64*4
|
|
|
|
add r6, r1, # 32*2*4
|
|
|
|
add r7, r1, #2*32*2*4
|
|
|
|
add r8, r1, #3*32*2*4
|
|
|
|
mov r12, r3
|
|
|
|
2:
|
|
|
|
vld1.32 {d0,d1}, [r1,:128]!
|
|
|
|
vld1.32 {d2,d3}, [r6,:128]!
|
|
|
|
vld1.32 {d4,d5}, [r7,:128]!
|
|
|
|
vld1.32 {d6,d7}, [r8,:128]!
|
|
|
|
vst4.32 {d0[0],d2[0],d4[0],d6[0]}, [lr,:128], r5
|
|
|
|
vst4.32 {d0[1],d2[1],d4[1],d6[1]}, [r4,:128], r5
|
|
|
|
vst4.32 {d1[0],d3[0],d5[0],d7[0]}, [lr,:128], r5
|
|
|
|
vst4.32 {d1[1],d3[1],d5[1],d7[1]}, [r4,:128], r5
|
|
|
|
subs r12, r12, #2
|
|
|
|
bgt 2b
|
|
|
|
add r0, r0, #16
|
|
|
|
add r1, r1, #3*32*2*4
|
|
|
|
subs r2, r2, #4
|
|
|
|
bgt 1b
|
|
|
|
pop {r4-r8,pc}
|
|
|
|
6:
|
|
|
|
mov lr, r0
|
|
|
|
add r4, r0, #38*64*4
|
|
|
|
add r6, r1, #32*2*4
|
|
|
|
mov r12, r3
|
|
|
|
2:
|
|
|
|
vld1.32 {d0,d1}, [r1,:128]!
|
|
|
|
vld1.32 {d2,d3}, [r6,:128]!
|
|
|
|
vst2.32 {d0[0],d2[0]}, [lr,:64], r5
|
|
|
|
vst2.32 {d0[1],d2[1]}, [r4,:64], r5
|
|
|
|
vst2.32 {d1[0],d3[0]}, [lr,:64], r5
|
|
|
|
vst2.32 {d1[1],d3[1]}, [r4,:64], r5
|
|
|
|
subs r12, r12, #2
|
|
|
|
bgt 2b
|
|
|
|
add r0, r0, #8
|
|
|
|
add r1, r1, #32*2*4
|
|
|
|
sub r2, r2, #2
|
|
|
|
b 1b
|
|
|
|
endfunc
|
|
|
|
|
|
|
|
function ff_ps_hybrid_analysis_neon, export=1
|
|
|
|
vldm r1, {d19-d31}
|
|
|
|
ldr r12, [sp]
|
|
|
|
lsl r3, r3, #3
|
|
|
|
vadd.f32 d16, d19, d31
|
|
|
|
vadd.f32 d17, d20, d30
|
|
|
|
vsub.f32 d18, d19, d31
|
|
|
|
vsub.f32 d19, d20, d30
|
|
|
|
vsub.f32 d0, d21, d29
|
|
|
|
vsub.f32 d1, d22, d28
|
|
|
|
vadd.f32 d2, d21, d29
|
|
|
|
vadd.f32 d3, d22, d28
|
|
|
|
vadd.f32 d20, d23, d27
|
|
|
|
vadd.f32 d21, d24, d26
|
|
|
|
vsub.f32 d22, d23, d27
|
|
|
|
vsub.f32 d23, d24, d26
|
|
|
|
vmov.i32 d6, #1<<31
|
|
|
|
vmov.i32 d7, #0
|
|
|
|
vmov.f32 q14, #0.0
|
|
|
|
vmov.f32 q15, #0.0
|
|
|
|
vtrn.32 d6, d7
|
|
|
|
vrev64.32 q9, q9
|
|
|
|
vrev64.32 q0, q0
|
|
|
|
vrev64.32 q11, q11
|
|
|
|
veor q9, q9, q3
|
|
|
|
veor q0, q0, q3
|
|
|
|
veor q11, q11, q3
|
|
|
|
vld1.32 {q13}, [r2,:128]!
|
|
|
|
vtrn.32 q8, q9
|
|
|
|
vtrn.32 q1, q0
|
|
|
|
vtrn.32 q10, q11
|
|
|
|
sub r12, r12, #1
|
|
|
|
vmla.f32 q14, q8, q13
|
|
|
|
vld1.32 {q2}, [r2,:128]!
|
|
|
|
vmla.f32 q15, q9, q13
|
|
|
|
1:
|
|
|
|
vmla.f32 q14, q1, q2
|
|
|
|
vld1.32 {q13}, [r2,:128]!
|
|
|
|
vmla.f32 q15, q0, q2
|
|
|
|
vmla.f32 q14, q10, q13
|
|
|
|
vld1.32 {q2}, [r2,:128]!
|
|
|
|
vmla.f32 q15, q11, q13
|
|
|
|
vld1.32 {q13}, [r2,:128]!
|
|
|
|
vadd.f32 d6, d28, d29
|
|
|
|
vadd.f32 d7, d30, d31
|
|
|
|
vmov.f32 q14, #0.0
|
|
|
|
vmov.f32 q15, #0.0
|
|
|
|
vmla.f32 q14, q8, q13
|
|
|
|
vpadd.f32 d6, d6, d7
|
|
|
|
vmla.f32 q15, q9, q13
|
|
|
|
vmla.f32 d6, d25, d4[0]
|
|
|
|
vld1.32 {q2}, [r2,:128]!
|
|
|
|
vst1.32 {d6}, [r0,:64], r3
|
|
|
|
subs r12, r12, #1
|
|
|
|
bgt 1b
|
|
|
|
vmla.f32 q14, q1, q2
|
|
|
|
vld1.32 {q13}, [r2,:128]!
|
|
|
|
vmla.f32 q15, q0, q2
|
|
|
|
vmla.f32 q14, q10, q13
|
|
|
|
vld1.32 {q2}, [r2,:128]!
|
|
|
|
vmla.f32 q15, q11, q13
|
|
|
|
vadd.f32 d6, d28, d29
|
|
|
|
vadd.f32 d7, d30, d31
|
|
|
|
vpadd.f32 d6, d6, d7
|
|
|
|
vmla.f32 d6, d25, d4[0]
|
|
|
|
vst1.32 {d6}, [r0,:64], r3
|
|
|
|
bx lr
|
|
|
|
endfunc
|
|
|
|
|
|
|
|
function ff_ps_stereo_interpolate_neon, export=1
|
|
|
|
vld1.32 {q0}, [r2]
|
|
|
|
vld1.32 {q14}, [r3]
|
|
|
|
mov r2, r0
|
|
|
|
mov r3, r1
|
|
|
|
ldr r12, [sp]
|
|
|
|
vadd.f32 q1, q0, q14
|
lavc/arm: fix lack of precision in ff_ps_stereo_interpolate_neon
The code originally pre-multiply by 2 the steps, causing the running sum
of the h factors to drift away due to the lack of precision. It quickly
causes an inaccuracy > 0.01.
I tried diverse approaches such as multiply by 2.0 (instead of adding
the value itself) without success.
I'm unable to bench the impact of this change, feel free to compare.
This commit fixes the incoming aacpsdsp tests.
Following is an alternative simplified function (matching the incoming
AArch64 code) that may be used:
function ff_ps_stereo_interpolate_neon, export=1
vld1.32 {q0}, [r2]
vld1.32 {q1}, [r3]
ldr r12, [sp]
vmov.f32 q8, q0
vmov.f32 q9, q1
vzip.32 q8, q0
vzip.32 q9, q1
1:
vld1.32 {d4}, [r0,:64]
vld1.32 {d6}, [r1,:64]
vadd.f32 q8, q8, q9
vadd.f32 q0, q0, q1
vmov.f32 d5, d4
vmov.f32 d7, d6
vmul.f32 q2, q2, q8
vmla.f32 q2, q3, q0
vst1.32 {d4}, [r0,:64]!
vst1.32 {d5}, [r1,:64]!
subs r12, r12, #1
bgt 1b
bx lr
endfunc
8 years ago
|
|
|
vadd.f32 q0, q1, q14
|
|
|
|
vld1.32 {q2}, [r0,:64]!
|
|
|
|
vld1.32 {q3}, [r1,:64]!
|
|
|
|
subs r12, r12, #1
|
|
|
|
beq 2f
|
|
|
|
1:
|
|
|
|
vmul.f32 d16, d4, d2[0]
|
|
|
|
vmul.f32 d17, d5, d0[0]
|
|
|
|
vmul.f32 d18, d4, d2[1]
|
|
|
|
vmul.f32 d19, d5, d0[1]
|
|
|
|
vmla.f32 d16, d6, d3[0]
|
|
|
|
vmla.f32 d17, d7, d1[0]
|
|
|
|
vmla.f32 d18, d6, d3[1]
|
|
|
|
vmla.f32 d19, d7, d1[1]
|
lavc/arm: fix lack of precision in ff_ps_stereo_interpolate_neon
The code originally pre-multiply by 2 the steps, causing the running sum
of the h factors to drift away due to the lack of precision. It quickly
causes an inaccuracy > 0.01.
I tried diverse approaches such as multiply by 2.0 (instead of adding
the value itself) without success.
I'm unable to bench the impact of this change, feel free to compare.
This commit fixes the incoming aacpsdsp tests.
Following is an alternative simplified function (matching the incoming
AArch64 code) that may be used:
function ff_ps_stereo_interpolate_neon, export=1
vld1.32 {q0}, [r2]
vld1.32 {q1}, [r3]
ldr r12, [sp]
vmov.f32 q8, q0
vmov.f32 q9, q1
vzip.32 q8, q0
vzip.32 q9, q1
1:
vld1.32 {d4}, [r0,:64]
vld1.32 {d6}, [r1,:64]
vadd.f32 q8, q8, q9
vadd.f32 q0, q0, q1
vmov.f32 d5, d4
vmov.f32 d7, d6
vmul.f32 q2, q2, q8
vmla.f32 q2, q3, q0
vst1.32 {d4}, [r0,:64]!
vst1.32 {d5}, [r1,:64]!
subs r12, r12, #1
bgt 1b
bx lr
endfunc
8 years ago
|
|
|
vadd.f32 q1, q1, q14
|
|
|
|
vadd.f32 q0, q0, q14
|
|
|
|
vadd.f32 q1, q1, q14
|
|
|
|
vadd.f32 q0, q0, q14
|
|
|
|
vld1.32 {q2}, [r0,:64]!
|
|
|
|
vld1.32 {q3}, [r1,:64]!
|
|
|
|
vst1.32 {q8}, [r2,:64]!
|
|
|
|
vst1.32 {q9}, [r3,:64]!
|
|
|
|
subs r12, r12, #2
|
|
|
|
bgt 1b
|
|
|
|
it lt
|
|
|
|
bxlt lr
|
|
|
|
2:
|
|
|
|
vmul.f32 d16, d4, d2[0]
|
|
|
|
vmul.f32 d18, d4, d2[1]
|
|
|
|
vmla.f32 d16, d6, d3[0]
|
|
|
|
vmla.f32 d18, d6, d3[1]
|
|
|
|
vst1.32 {d16}, [r2,:64]!
|
|
|
|
vst1.32 {d18}, [r3,:64]!
|
|
|
|
bx lr
|
|
|
|
endfunc
|