2010-04-30 21:30:27 +00:00
|
|
|
/*
|
|
|
|
|
* Copyright (c) 2010 Mans Rullgard <mans@mansr.com>
|
|
|
|
|
*
|
2011-03-18 17:35:10 +00:00
|
|
|
* This file is part of Libav.
|
2010-04-30 21:30:27 +00:00
|
|
|
*
|
2011-03-18 17:35:10 +00:00
|
|
|
* Libav is free software; you can redistribute it and/or
|
2010-04-30 21:30:27 +00:00
|
|
|
* 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.
|
|
|
|
|
*
|
2011-03-18 17:35:10 +00:00
|
|
|
* Libav is distributed in the hope that it will be useful,
|
2010-04-30 21:30:27 +00:00
|
|
|
* 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
|
2011-03-18 17:35:10 +00:00
|
|
|
* License along with Libav; if not, write to the Free Software
|
2010-04-30 21:30:27 +00:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include "asm.S"
|
|
|
|
|
|
|
|
|
|
.macro vp6_edge_filter
|
|
|
|
|
vdup.16 q3, r2 @ t
|
|
|
|
|
vmov.i16 q13, #1
|
|
|
|
|
vsubl.u8 q0, d20, d18 @ p[ 0] - p[-s]
|
|
|
|
|
vsubl.u8 q1, d16, d22 @ p[-2*s] - p[ s]
|
|
|
|
|
vsubl.u8 q14, d21, d19
|
|
|
|
|
vsubl.u8 q15, d17, d23
|
|
|
|
|
vadd.i16 q2, q0, q0 @ 2*(p[0]-p[-s])
|
|
|
|
|
vadd.i16 d29, d28, d28
|
|
|
|
|
vadd.i16 q0, q0, q1 @ p[0]-p[-s] + p[-2*s]-p[s]
|
|
|
|
|
vadd.i16 d28, d28, d30
|
|
|
|
|
vadd.i16 q0, q0, q2 @ 3*(p[0]-p[-s]) + p[-2*s]-p[s]
|
|
|
|
|
vadd.i16 d28, d28, d29
|
|
|
|
|
vrshr.s16 q0, q0, #3 @ v
|
|
|
|
|
vrshr.s16 d28, d28, #3
|
|
|
|
|
vsub.i16 q8, q3, q13 @ t-1
|
|
|
|
|
vabs.s16 q1, q0 @ V
|
|
|
|
|
vshr.s16 q2, q0, #15 @ s
|
|
|
|
|
vabs.s16 d30, d28
|
|
|
|
|
vshr.s16 d29, d28, #15
|
|
|
|
|
vsub.i16 q12, q1, q3 @ V-t
|
|
|
|
|
vsub.i16 d31, d30, d6
|
|
|
|
|
vsub.i16 q12, q12, q13 @ V-t-1
|
|
|
|
|
vsub.i16 d31, d31, d26
|
|
|
|
|
vcge.u16 q12, q12, q8 @ V-t-1 >= t-1
|
|
|
|
|
vcge.u16 d31, d31, d16
|
|
|
|
|
vadd.i16 q13, q3, q3 @ 2*t
|
|
|
|
|
vadd.i16 d16, d6, d6
|
|
|
|
|
vsub.i16 q13, q13, q1 @ 2*t - V
|
|
|
|
|
vsub.i16 d16, d16, d30
|
|
|
|
|
vadd.i16 q13, q13, q2 @ += s
|
|
|
|
|
vadd.i16 d16, d16, d29
|
|
|
|
|
veor q13, q13, q2 @ ^= s
|
|
|
|
|
veor d16, d16, d29
|
|
|
|
|
vbif q0, q13, q12
|
|
|
|
|
vbif d28, d16, d31
|
|
|
|
|
vmovl.u8 q1, d20
|
|
|
|
|
vmovl.u8 q15, d21
|
|
|
|
|
vaddw.u8 q2, q0, d18
|
|
|
|
|
vaddw.u8 q3, q14, d19
|
|
|
|
|
vsub.i16 q1, q1, q0
|
|
|
|
|
vsub.i16 d30, d30, d28
|
|
|
|
|
vqmovun.s16 d18, q2
|
|
|
|
|
vqmovun.s16 d19, q3
|
|
|
|
|
vqmovun.s16 d20, q1
|
|
|
|
|
vqmovun.s16 d21, q15
|
|
|
|
|
.endm
|
|
|
|
|
|
|
|
|
|
function ff_vp6_edge_filter_ver_neon, export=1
|
|
|
|
|
sub r0, r0, r1, lsl #1
|
|
|
|
|
vld1.8 {q8}, [r0], r1 @ p[-2*s]
|
|
|
|
|
vld1.8 {q9}, [r0], r1 @ p[-s]
|
|
|
|
|
vld1.8 {q10}, [r0], r1 @ p[0]
|
|
|
|
|
vld1.8 {q11}, [r0] @ p[s]
|
|
|
|
|
vp6_edge_filter
|
|
|
|
|
sub r0, r0, r1, lsl #1
|
|
|
|
|
sub r1, r1, #8
|
|
|
|
|
vst1.8 {d18}, [r0]!
|
|
|
|
|
vst1.32 {d19[0]}, [r0], r1
|
|
|
|
|
vst1.8 {d20}, [r0]!
|
|
|
|
|
vst1.32 {d21[0]}, [r0]
|
|
|
|
|
bx lr
|
|
|
|
|
endfunc
|
|
|
|
|
|
|
|
|
|
function ff_vp6_edge_filter_hor_neon, export=1
|
|
|
|
|
sub r3, r0, #1
|
|
|
|
|
sub r0, r0, #2
|
|
|
|
|
vld1.32 {d16[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d18[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d20[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d22[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d16[1]}, [r0], r1
|
|
|
|
|
vld1.32 {d18[1]}, [r0], r1
|
|
|
|
|
vld1.32 {d20[1]}, [r0], r1
|
|
|
|
|
vld1.32 {d22[1]}, [r0], r1
|
|
|
|
|
vld1.32 {d17[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d19[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d21[0]}, [r0], r1
|
|
|
|
|
vld1.32 {d23[0]}, [r0], r1
|
|
|
|
|
vtrn.8 q8, q9
|
|
|
|
|
vtrn.8 q10, q11
|
|
|
|
|
vtrn.16 q8, q10
|
|
|
|
|
vtrn.16 q9, q11
|
|
|
|
|
vp6_edge_filter
|
|
|
|
|
vtrn.8 q9, q10
|
|
|
|
|
vst1.16 {d18[0]}, [r3], r1
|
|
|
|
|
vst1.16 {d20[0]}, [r3], r1
|
|
|
|
|
vst1.16 {d18[1]}, [r3], r1
|
|
|
|
|
vst1.16 {d20[1]}, [r3], r1
|
|
|
|
|
vst1.16 {d18[2]}, [r3], r1
|
|
|
|
|
vst1.16 {d20[2]}, [r3], r1
|
|
|
|
|
vst1.16 {d18[3]}, [r3], r1
|
|
|
|
|
vst1.16 {d20[3]}, [r3], r1
|
|
|
|
|
vst1.16 {d19[0]}, [r3], r1
|
|
|
|
|
vst1.16 {d21[0]}, [r3], r1
|
|
|
|
|
vst1.16 {d19[1]}, [r3], r1
|
|
|
|
|
vst1.16 {d21[1]}, [r3], r1
|
|
|
|
|
bx lr
|
|
|
|
|
endfunc
|
