2006-09-26 23:26:22 +03:00
|
|
|
/*
|
|
|
|
* Simple IDCT
|
|
|
|
*
|
|
|
|
* Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
|
|
|
|
* Copyright (c) 2006 Mans Rullgard <mru@inprovide.com>
|
|
|
|
*
|
2006-10-07 18:30:46 +03:00
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2006-09-26 23:26:22 +03:00
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
2006-10-07 18:30:46 +03:00
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
2006-09-26 23:26:22 +03:00
|
|
|
*
|
2006-10-07 18:30:46 +03:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2006-09-26 23:26:22 +03:00
|
|
|
* 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
|
2006-10-07 18:30:46 +03:00
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
2006-09-26 23:26:22 +03:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define W1 22725 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define W2 21407 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define W3 19266 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define W4 16383 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define W5 12873 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define W6 8867 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define W7 4520 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
|
|
|
|
#define ROW_SHIFT 11
|
|
|
|
#define COL_SHIFT 20
|
|
|
|
|
|
|
|
#define W13 (W1 | (W3 << 16))
|
|
|
|
#define W26 (W2 | (W6 << 16))
|
|
|
|
#define W57 (W5 | (W7 << 16))
|
|
|
|
|
|
|
|
.text
|
|
|
|
.align
|
|
|
|
w13: .long W13
|
|
|
|
w26: .long W26
|
|
|
|
w57: .long W57
|
|
|
|
|
|
|
|
.align
|
|
|
|
.func idct_row_armv5te
|
|
|
|
idct_row_armv5te:
|
|
|
|
str lr, [sp, #-4]!
|
|
|
|
|
|
|
|
ldrd v1, [a1, #8]
|
|
|
|
ldrd a3, [a1] /* a3 = row[1:0], a4 = row[3:2] */
|
|
|
|
orrs v1, v1, v2
|
|
|
|
cmpeq v1, a4
|
|
|
|
cmpeq v1, a3, lsr #16
|
|
|
|
beq row_dc_only
|
|
|
|
|
|
|
|
mov v1, #(1<<(ROW_SHIFT-1))
|
|
|
|
mov ip, #16384
|
|
|
|
sub ip, ip, #1 /* ip = W4 */
|
|
|
|
smlabb v1, ip, a3, v1 /* v1 = W4*row[0]+(1<<(RS-1)) */
|
|
|
|
ldr ip, [pc, #(w26-.-8)] /* ip = W2 | (W6 << 16) */
|
|
|
|
smultb a2, ip, a4
|
|
|
|
smulbb lr, ip, a4
|
|
|
|
add v2, v1, a2
|
|
|
|
sub v3, v1, a2
|
|
|
|
sub v4, v1, lr
|
|
|
|
add v1, v1, lr
|
|
|
|
|
|
|
|
ldr ip, [pc, #(w13-.-8)] /* ip = W1 | (W3 << 16) */
|
|
|
|
ldr lr, [pc, #(w57-.-8)] /* lr = W5 | (W7 << 16) */
|
|
|
|
smulbt v5, ip, a3
|
|
|
|
smultt v6, lr, a4
|
|
|
|
smlatt v5, ip, a4, v5
|
|
|
|
smultt a2, ip, a3
|
|
|
|
smulbt v7, lr, a3
|
|
|
|
sub v6, v6, a2
|
|
|
|
smulbt a2, ip, a4
|
|
|
|
smultt fp, lr, a3
|
|
|
|
sub v7, v7, a2
|
|
|
|
smulbt a2, lr, a4
|
|
|
|
ldrd a3, [a1, #8] /* a3=row[5:4] a4=row[7:6] */
|
|
|
|
sub fp, fp, a2
|
|
|
|
|
|
|
|
orrs a2, a3, a4
|
|
|
|
beq 1f
|
|
|
|
|
|
|
|
smlabt v5, lr, a3, v5
|
|
|
|
smlabt v6, ip, a3, v6
|
|
|
|
smlatt v5, lr, a4, v5
|
|
|
|
smlabt v6, lr, a4, v6
|
|
|
|
smlatt v7, lr, a3, v7
|
|
|
|
smlatt fp, ip, a3, fp
|
|
|
|
smulbt a2, ip, a4
|
|
|
|
smlatt v7, ip, a4, v7
|
|
|
|
sub fp, fp, a2
|
|
|
|
|
|
|
|
ldr ip, [pc, #(w26-.-8)] /* ip = W2 | (W6 << 16) */
|
|
|
|
mov a2, #16384
|
|
|
|
sub a2, a2, #1 /* a2 = W4 */
|
|
|
|
smulbb a2, a2, a3 /* a2 = W4*row[4] */
|
|
|
|
smultb lr, ip, a4 /* lr = W6*row[6] */
|
|
|
|
add v1, v1, a2 /* v1 += W4*row[4] */
|
|
|
|
add v1, v1, lr /* v1 += W6*row[6] */
|
|
|
|
add v4, v4, a2 /* v4 += W4*row[4] */
|
|
|
|
sub v4, v4, lr /* v4 -= W6*row[6] */
|
|
|
|
smulbb lr, ip, a4 /* lr = W2*row[6] */
|
|
|
|
sub v2, v2, a2 /* v2 -= W4*row[4] */
|
|
|
|
sub v2, v2, lr /* v2 -= W2*row[6] */
|
|
|
|
sub v3, v3, a2 /* v3 -= W4*row[4] */
|
|
|
|
add v3, v3, lr /* v3 += W2*row[6] */
|
|
|
|
|
|
|
|
1: add a2, v1, v5
|
|
|
|
mov a3, a2, lsr #11
|
|
|
|
bic a3, a3, #0x1f0000
|
|
|
|
sub a2, v2, v6
|
|
|
|
mov a2, a2, lsr #11
|
|
|
|
add a3, a3, a2, lsl #16
|
|
|
|
add a2, v3, v7
|
|
|
|
mov a4, a2, lsr #11
|
|
|
|
bic a4, a4, #0x1f0000
|
|
|
|
add a2, v4, fp
|
|
|
|
mov a2, a2, lsr #11
|
|
|
|
add a4, a4, a2, lsl #16
|
|
|
|
strd a3, [a1]
|
|
|
|
|
|
|
|
sub a2, v4, fp
|
|
|
|
mov a3, a2, lsr #11
|
|
|
|
bic a3, a3, #0x1f0000
|
|
|
|
sub a2, v3, v7
|
|
|
|
mov a2, a2, lsr #11
|
|
|
|
add a3, a3, a2, lsl #16
|
|
|
|
add a2, v2, v6
|
|
|
|
mov a4, a2, lsr #11
|
|
|
|
bic a4, a4, #0x1f0000
|
|
|
|
sub a2, v1, v5
|
|
|
|
mov a2, a2, lsr #11
|
|
|
|
add a4, a4, a2, lsl #16
|
|
|
|
strd a3, [a1, #8]
|
|
|
|
|
|
|
|
ldr pc, [sp], #4
|
|
|
|
|
|
|
|
row_dc_only:
|
|
|
|
orr a3, a3, a3, lsl #16
|
|
|
|
bic a3, a3, #0xe000
|
|
|
|
mov a3, a3, lsl #3
|
|
|
|
mov a4, a3
|
|
|
|
strd a3, [a1]
|
|
|
|
strd a3, [a1, #8]
|
|
|
|
|
|
|
|
ldr pc, [sp], #4
|
|
|
|
.endfunc
|
|
|
|
|
|
|
|
.macro idct_col
|
|
|
|
ldr a4, [a1] /* a4 = col[1:0] */
|
|
|
|
mov ip, #16384
|
|
|
|
sub ip, ip, #1 /* ip = W4 */
|
|
|
|
#if 0
|
|
|
|
mov v1, #(1<<(COL_SHIFT-1))
|
|
|
|
smlabt v2, ip, a4, v1 /* v2 = W4*col[1] + (1<<(COL_SHIFT-1)) */
|
|
|
|
smlabb v1, ip, a4, v1 /* v1 = W4*col[0] + (1<<(COL_SHIFT-1)) */
|
|
|
|
ldr a4, [a1, #(16*4)]
|
|
|
|
#else
|
|
|
|
mov v1, #((1<<(COL_SHIFT-1))/W4) /* this matches the C version */
|
|
|
|
add v2, v1, a4, asr #16
|
|
|
|
rsb v2, v2, v2, lsl #14
|
|
|
|
mov a4, a4, lsl #16
|
|
|
|
add v1, v1, a4, asr #16
|
|
|
|
ldr a4, [a1, #(16*4)]
|
|
|
|
rsb v1, v1, v1, lsl #14
|
|
|
|
#endif
|
|
|
|
|
|
|
|
smulbb lr, ip, a4
|
|
|
|
smulbt a3, ip, a4
|
|
|
|
sub v3, v1, lr
|
|
|
|
sub v5, v1, lr
|
|
|
|
add v7, v1, lr
|
|
|
|
add v1, v1, lr
|
|
|
|
sub v4, v2, a3
|
|
|
|
sub v6, v2, a3
|
|
|
|
add fp, v2, a3
|
|
|
|
ldr ip, [pc, #(w26-.-8)]
|
|
|
|
ldr a4, [a1, #(16*2)]
|
|
|
|
add v2, v2, a3
|
|
|
|
|
|
|
|
smulbb lr, ip, a4
|
|
|
|
smultb a3, ip, a4
|
|
|
|
add v1, v1, lr
|
|
|
|
sub v7, v7, lr
|
|
|
|
add v3, v3, a3
|
|
|
|
sub v5, v5, a3
|
|
|
|
smulbt lr, ip, a4
|
|
|
|
smultt a3, ip, a4
|
|
|
|
add v2, v2, lr
|
|
|
|
sub fp, fp, lr
|
|
|
|
add v4, v4, a3
|
|
|
|
ldr a4, [a1, #(16*6)]
|
|
|
|
sub v6, v6, a3
|
|
|
|
|
|
|
|
smultb lr, ip, a4
|
|
|
|
smulbb a3, ip, a4
|
|
|
|
add v1, v1, lr
|
|
|
|
sub v7, v7, lr
|
|
|
|
sub v3, v3, a3
|
|
|
|
add v5, v5, a3
|
|
|
|
smultt lr, ip, a4
|
|
|
|
smulbt a3, ip, a4
|
|
|
|
add v2, v2, lr
|
|
|
|
sub fp, fp, lr
|
|
|
|
sub v4, v4, a3
|
|
|
|
add v6, v6, a3
|
|
|
|
|
|
|
|
stmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp}
|
|
|
|
|
|
|
|
ldr ip, [pc, #(w13-.-8)]
|
|
|
|
ldr a4, [a1, #(16*1)]
|
|
|
|
ldr lr, [pc, #(w57-.-8)]
|
|
|
|
smulbb v1, ip, a4
|
|
|
|
smultb v3, ip, a4
|
|
|
|
smulbb v5, lr, a4
|
|
|
|
smultb v7, lr, a4
|
|
|
|
smulbt v2, ip, a4
|
|
|
|
smultt v4, ip, a4
|
|
|
|
smulbt v6, lr, a4
|
|
|
|
smultt fp, lr, a4
|
|
|
|
rsb v4, v4, #0
|
|
|
|
ldr a4, [a1, #(16*3)]
|
|
|
|
rsb v3, v3, #0
|
|
|
|
|
|
|
|
smlatb v1, ip, a4, v1
|
|
|
|
smlatb v3, lr, a4, v3
|
|
|
|
smulbb a3, ip, a4
|
|
|
|
smulbb a2, lr, a4
|
|
|
|
sub v5, v5, a3
|
|
|
|
sub v7, v7, a2
|
|
|
|
smlatt v2, ip, a4, v2
|
|
|
|
smlatt v4, lr, a4, v4
|
|
|
|
smulbt a3, ip, a4
|
|
|
|
smulbt a2, lr, a4
|
|
|
|
sub v6, v6, a3
|
|
|
|
ldr a4, [a1, #(16*5)]
|
|
|
|
sub fp, fp, a2
|
|
|
|
|
|
|
|
smlabb v1, lr, a4, v1
|
|
|
|
smlabb v3, ip, a4, v3
|
|
|
|
smlatb v5, lr, a4, v5
|
|
|
|
smlatb v7, ip, a4, v7
|
|
|
|
smlabt v2, lr, a4, v2
|
|
|
|
smlabt v4, ip, a4, v4
|
|
|
|
smlatt v6, lr, a4, v6
|
|
|
|
ldr a3, [a1, #(16*7)]
|
|
|
|
smlatt fp, ip, a4, fp
|
|
|
|
|
|
|
|
smlatb v1, lr, a3, v1
|
|
|
|
smlabb v3, lr, a3, v3
|
|
|
|
smlatb v5, ip, a3, v5
|
|
|
|
smulbb a4, ip, a3
|
|
|
|
smlatt v2, lr, a3, v2
|
|
|
|
sub v7, v7, a4
|
|
|
|
smlabt v4, lr, a3, v4
|
|
|
|
smulbt a4, ip, a3
|
|
|
|
smlatt v6, ip, a3, v6
|
|
|
|
sub fp, fp, a4
|
|
|
|
.endm
|
|
|
|
|
|
|
|
.align
|
|
|
|
.func idct_col_armv5te
|
|
|
|
idct_col_armv5te:
|
|
|
|
str lr, [sp, #-4]!
|
|
|
|
|
|
|
|
idct_col
|
|
|
|
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
adds a2, a3, v1
|
|
|
|
mov a2, a2, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
add ip, a4, v2
|
|
|
|
mov ip, ip, asr #20
|
|
|
|
orr a2, a2, ip, lsl #16
|
|
|
|
str a2, [a1]
|
|
|
|
subs a3, a3, v1
|
|
|
|
mov a2, a3, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
sub a4, a4, v2
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
orr a2, a2, a4, lsl #16
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
str a2, [a1, #(16*7)]
|
|
|
|
|
|
|
|
subs a2, a3, v3
|
|
|
|
mov a2, a2, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
sub ip, a4, v4
|
|
|
|
mov ip, ip, asr #20
|
|
|
|
orr a2, a2, ip, lsl #16
|
|
|
|
str a2, [a1, #(16*1)]
|
|
|
|
adds a3, a3, v3
|
|
|
|
mov a2, a3, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
add a4, a4, v4
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
orr a2, a2, a4, lsl #16
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
str a2, [a1, #(16*6)]
|
|
|
|
|
|
|
|
adds a2, a3, v5
|
|
|
|
mov a2, a2, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
add ip, a4, v6
|
|
|
|
mov ip, ip, asr #20
|
|
|
|
orr a2, a2, ip, lsl #16
|
|
|
|
str a2, [a1, #(16*2)]
|
|
|
|
subs a3, a3, v5
|
|
|
|
mov a2, a3, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
sub a4, a4, v6
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
orr a2, a2, a4, lsl #16
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
str a2, [a1, #(16*5)]
|
|
|
|
|
|
|
|
adds a2, a3, v7
|
|
|
|
mov a2, a2, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
add ip, a4, fp
|
|
|
|
mov ip, ip, asr #20
|
|
|
|
orr a2, a2, ip, lsl #16
|
|
|
|
str a2, [a1, #(16*3)]
|
|
|
|
subs a3, a3, v7
|
|
|
|
mov a2, a3, lsr #20
|
|
|
|
orrmi a2, a2, #0xf000
|
|
|
|
sub a4, a4, fp
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
orr a2, a2, a4, lsl #16
|
|
|
|
str a2, [a1, #(16*4)]
|
|
|
|
|
|
|
|
ldr pc, [sp], #4
|
|
|
|
.endfunc
|
|
|
|
|
|
|
|
.align
|
|
|
|
.func idct_col_put_armv5te
|
|
|
|
idct_col_put_armv5te:
|
|
|
|
str lr, [sp, #-4]!
|
|
|
|
|
|
|
|
idct_col
|
|
|
|
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
ldr lr, [sp, #32]
|
|
|
|
add a2, a3, v1
|
|
|
|
movs a2, a2, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add ip, a4, v2
|
|
|
|
movs ip, ip, asr #20
|
|
|
|
movmi ip, #0
|
|
|
|
cmp ip, #255
|
|
|
|
movgt ip, #255
|
|
|
|
orr a2, a2, ip, lsl #8
|
|
|
|
sub a3, a3, v1
|
|
|
|
movs a3, a3, asr #20
|
|
|
|
movmi a3, #0
|
|
|
|
cmp a3, #255
|
|
|
|
movgt a3, #255
|
|
|
|
sub a4, a4, v2
|
|
|
|
movs a4, a4, asr #20
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
ldr v1, [sp, #28]
|
|
|
|
movgt a4, #255
|
|
|
|
strh a2, [v1]
|
|
|
|
add a2, v1, #2
|
|
|
|
str a2, [sp, #28]
|
|
|
|
orr a2, a3, a4, lsl #8
|
|
|
|
rsb v2, lr, lr, lsl #3
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
strh a2, [v2, v1]!
|
|
|
|
|
|
|
|
sub a2, a3, v3
|
|
|
|
movs a2, a2, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
sub ip, a4, v4
|
|
|
|
movs ip, ip, asr #20
|
|
|
|
movmi ip, #0
|
|
|
|
cmp ip, #255
|
|
|
|
movgt ip, #255
|
|
|
|
orr a2, a2, ip, lsl #8
|
|
|
|
strh a2, [v1, lr]!
|
|
|
|
add a3, a3, v3
|
|
|
|
movs a2, a3, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add a4, a4, v4
|
|
|
|
movs a4, a4, asr #20
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
orr a2, a2, a4, lsl #8
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
strh a2, [v2, -lr]!
|
|
|
|
|
|
|
|
add a2, a3, v5
|
|
|
|
movs a2, a2, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add ip, a4, v6
|
|
|
|
movs ip, ip, asr #20
|
|
|
|
movmi ip, #0
|
|
|
|
cmp ip, #255
|
|
|
|
movgt ip, #255
|
|
|
|
orr a2, a2, ip, lsl #8
|
|
|
|
strh a2, [v1, lr]!
|
|
|
|
sub a3, a3, v5
|
|
|
|
movs a2, a3, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
sub a4, a4, v6
|
|
|
|
movs a4, a4, asr #20
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
orr a2, a2, a4, lsl #8
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
strh a2, [v2, -lr]!
|
|
|
|
|
|
|
|
add a2, a3, v7
|
|
|
|
movs a2, a2, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add ip, a4, fp
|
|
|
|
movs ip, ip, asr #20
|
|
|
|
movmi ip, #0
|
|
|
|
cmp ip, #255
|
|
|
|
movgt ip, #255
|
|
|
|
orr a2, a2, ip, lsl #8
|
|
|
|
strh a2, [v1, lr]
|
|
|
|
sub a3, a3, v7
|
|
|
|
movs a2, a3, asr #20
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
sub a4, a4, fp
|
|
|
|
movs a4, a4, asr #20
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
orr a2, a2, a4, lsl #8
|
|
|
|
strh a2, [v2, -lr]
|
|
|
|
|
|
|
|
ldr pc, [sp], #4
|
|
|
|
.endfunc
|
|
|
|
|
|
|
|
.align
|
|
|
|
.func idct_col_add_armv5te
|
|
|
|
idct_col_add_armv5te:
|
|
|
|
str lr, [sp, #-4]!
|
|
|
|
|
|
|
|
idct_col
|
|
|
|
|
|
|
|
ldr lr, [sp, #36]
|
|
|
|
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
ldrh ip, [lr]
|
|
|
|
add a2, a3, v1
|
|
|
|
mov a2, a2, asr #20
|
|
|
|
sub a3, a3, v1
|
|
|
|
and v1, ip, #255
|
|
|
|
adds a2, a2, v1
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add v1, a4, v2
|
|
|
|
mov v1, v1, asr #20
|
|
|
|
adds v1, v1, ip, lsr #8
|
|
|
|
movmi v1, #0
|
|
|
|
cmp v1, #255
|
|
|
|
movgt v1, #255
|
|
|
|
orr a2, a2, v1, lsl #8
|
|
|
|
ldr v1, [sp, #32]
|
|
|
|
sub a4, a4, v2
|
|
|
|
rsb v2, v1, v1, lsl #3
|
|
|
|
ldrh ip, [v2, lr]!
|
|
|
|
strh a2, [lr]
|
|
|
|
mov a3, a3, asr #20
|
|
|
|
and a2, ip, #255
|
|
|
|
adds a3, a3, a2
|
|
|
|
movmi a3, #0
|
|
|
|
cmp a3, #255
|
|
|
|
movgt a3, #255
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
adds a4, a4, ip, lsr #8
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
add a2, lr, #2
|
|
|
|
str a2, [sp, #28]
|
|
|
|
orr a2, a3, a4, lsl #8
|
|
|
|
strh a2, [v2]
|
|
|
|
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
ldrh ip, [lr, v1]!
|
|
|
|
sub a2, a3, v3
|
|
|
|
mov a2, a2, asr #20
|
|
|
|
add a3, a3, v3
|
|
|
|
and v3, ip, #255
|
|
|
|
adds a2, a2, v3
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
sub v3, a4, v4
|
|
|
|
mov v3, v3, asr #20
|
|
|
|
adds v3, v3, ip, lsr #8
|
|
|
|
movmi v3, #0
|
|
|
|
cmp v3, #255
|
|
|
|
movgt v3, #255
|
|
|
|
orr a2, a2, v3, lsl #8
|
|
|
|
add a4, a4, v4
|
|
|
|
ldrh ip, [v2, -v1]!
|
|
|
|
strh a2, [lr]
|
|
|
|
mov a3, a3, asr #20
|
|
|
|
and a2, ip, #255
|
|
|
|
adds a3, a3, a2
|
|
|
|
movmi a3, #0
|
|
|
|
cmp a3, #255
|
|
|
|
movgt a3, #255
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
adds a4, a4, ip, lsr #8
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
orr a2, a3, a4, lsl #8
|
|
|
|
strh a2, [v2]
|
|
|
|
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
ldrh ip, [lr, v1]!
|
|
|
|
add a2, a3, v5
|
|
|
|
mov a2, a2, asr #20
|
|
|
|
sub a3, a3, v5
|
|
|
|
and v3, ip, #255
|
|
|
|
adds a2, a2, v3
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add v3, a4, v6
|
|
|
|
mov v3, v3, asr #20
|
|
|
|
adds v3, v3, ip, lsr #8
|
|
|
|
movmi v3, #0
|
|
|
|
cmp v3, #255
|
|
|
|
movgt v3, #255
|
|
|
|
orr a2, a2, v3, lsl #8
|
|
|
|
sub a4, a4, v6
|
|
|
|
ldrh ip, [v2, -v1]!
|
|
|
|
strh a2, [lr]
|
|
|
|
mov a3, a3, asr #20
|
|
|
|
and a2, ip, #255
|
|
|
|
adds a3, a3, a2
|
|
|
|
movmi a3, #0
|
|
|
|
cmp a3, #255
|
|
|
|
movgt a3, #255
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
adds a4, a4, ip, lsr #8
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
orr a2, a3, a4, lsl #8
|
|
|
|
strh a2, [v2]
|
|
|
|
|
|
|
|
ldmfd sp!, {a3, a4}
|
|
|
|
ldrh ip, [lr, v1]!
|
|
|
|
add a2, a3, v7
|
|
|
|
mov a2, a2, asr #20
|
|
|
|
sub a3, a3, v7
|
|
|
|
and v3, ip, #255
|
|
|
|
adds a2, a2, v3
|
|
|
|
movmi a2, #0
|
|
|
|
cmp a2, #255
|
|
|
|
movgt a2, #255
|
|
|
|
add v3, a4, fp
|
|
|
|
mov v3, v3, asr #20
|
|
|
|
adds v3, v3, ip, lsr #8
|
|
|
|
movmi v3, #0
|
|
|
|
cmp v3, #255
|
|
|
|
movgt v3, #255
|
|
|
|
orr a2, a2, v3, lsl #8
|
|
|
|
sub a4, a4, fp
|
|
|
|
ldrh ip, [v2, -v1]!
|
|
|
|
strh a2, [lr]
|
|
|
|
mov a3, a3, asr #20
|
|
|
|
and a2, ip, #255
|
|
|
|
adds a3, a3, a2
|
|
|
|
movmi a3, #0
|
|
|
|
cmp a3, #255
|
|
|
|
movgt a3, #255
|
|
|
|
mov a4, a4, asr #20
|
|
|
|
adds a4, a4, ip, lsr #8
|
|
|
|
movmi a4, #0
|
|
|
|
cmp a4, #255
|
|
|
|
movgt a4, #255
|
|
|
|
orr a2, a3, a4, lsl #8
|
|
|
|
strh a2, [v2]
|
|
|
|
|
|
|
|
ldr pc, [sp], #4
|
|
|
|
.endfunc
|
|
|
|
|
|
|
|
.align
|
|
|
|
.global simple_idct_armv5te
|
|
|
|
.func simple_idct_armv5te
|
|
|
|
simple_idct_armv5te:
|
|
|
|
stmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, lr}
|
|
|
|
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
|
|
|
|
sub a1, a1, #(16*7)
|
|
|
|
|
|
|
|
bl idct_col_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_armv5te
|
|
|
|
|
|
|
|
ldmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, pc}
|
|
|
|
.endfunc
|
|
|
|
|
|
|
|
.align
|
|
|
|
.global simple_idct_add_armv5te
|
|
|
|
.func simple_idct_add_armv5te
|
|
|
|
simple_idct_add_armv5te:
|
|
|
|
stmfd sp!, {a1, a2, v1, v2, v3, v4, v5, v6, v7, fp, lr}
|
|
|
|
|
|
|
|
mov a1, a3
|
|
|
|
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
|
|
|
|
sub a1, a1, #(16*7)
|
|
|
|
|
|
|
|
bl idct_col_add_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_add_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_add_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_add_armv5te
|
|
|
|
|
|
|
|
add sp, sp, #8
|
|
|
|
ldmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, pc}
|
|
|
|
.endfunc
|
|
|
|
|
|
|
|
.align
|
|
|
|
.global simple_idct_put_armv5te
|
|
|
|
.func simple_idct_put_armv5te
|
|
|
|
simple_idct_put_armv5te:
|
|
|
|
stmfd sp!, {a1, a2, v1, v2, v3, v4, v5, v6, v7, fp, lr}
|
|
|
|
|
|
|
|
mov a1, a3
|
|
|
|
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
add a1, a1, #16
|
|
|
|
bl idct_row_armv5te
|
|
|
|
|
|
|
|
sub a1, a1, #(16*7)
|
|
|
|
|
|
|
|
bl idct_col_put_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_put_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_put_armv5te
|
|
|
|
add a1, a1, #4
|
|
|
|
bl idct_col_put_armv5te
|
|
|
|
|
|
|
|
add sp, sp, #8
|
|
|
|
ldmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, pc}
|
|
|
|
.endfunc
|