Nyaaori/rockbox
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
rockbox/apps/plugins/mpegplayer/motion_comp_coldfire_s.S
Jens Arnold e1b4bf7a4d Mpegplayer: Assembler optimised motion compensation for coldfire (just the variants that are assemblerised for ARM) for a nice speedup. 
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@15168 a1c6a512-1295-4272-9138-f99709370657
2007-10-17 15:46:09 +00:00

434 lines
9.7 KiB
ArmAsm
Raw Blame History

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2007 Jens Arnold
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
.macro LEFT8_PW dW1, dW2 | needs %d0 == 24, clobbers %d2
lsl.l #8, \dW1 | changes dW1, keeps dW2
move.l \dW2, %d2
lsr.l %d0, %d2
or.l %d2, \dW1
.endm
.macro LEFT24_PW dW1, dW2 | needs %d0 == 24, clobbers %d2
lsl.l %d0, \dW1 | changes dW1, keeps dW2
move.l \dW2, %d2
lsr.l #8, %d2
or.l %d2, \dW1
.endm
/*****************************************************************************/
.align 2
.global MC_put_o_8
.type MC_put_o_8, @function
MC_put_o_8:
movem.l (4,%sp), %a0-%a1 | dest, source
move.l %a1, %d0
and.l #3, %d0
sub.l %d0, %a1 | align source
jmp.l (2, %pc, %d0.l*4)
bra.w .po8_0
bra.w .po8_1
bra.w .po8_2
| last table entry coincides with target
.po8_3:
lea.l (-5*4,%sp), %sp
movem.l %d2-%d5/%a2, (%sp) | save some registers
move.l (5*4+12,%sp), %a2 | stride
move.l (5*4+16,%sp), %d1 | height
moveq.l #24, %d0 | shift amount
1:
movem.l (%a1), %d3-%d5
add.l %a2, %a1
LEFT24_PW %d3, %d4
lsl.l %d0, %d4
lsr.l #8, %d5
or.l %d5, %d4
movem.l %d3-%d4, (%a0)
add.l %a2, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d5/%a2
lea.l (5*4,%sp), %sp
rts
.po8_2:
lea.l (-3*4,%sp), %sp
movem.l %d2-%d4, (%sp) | save some registers
movem.l (3*4+12,%sp), %d0-%d1 | stride, height
1:
movem.l (%a1), %d2-%d4
add.l %d0, %a1
swap %d2
swap %d3
move.w %d3, %d2
swap %d4
move.w %d4, %d3
movem.l %d2-%d3, (%a0)
add.l %d0, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d4
lea.l (3*4,%sp), %sp
rts
.po8_1:
lea.l (-5*4,%sp), %sp
movem.l %d2-%d5/%a2, (%sp) | save some registers
move.l (5*4+12,%sp), %a2 | stride
move.l (5*4+16,%sp), %d1 | height
moveq.l #24, %d0 | shift amount
1:
movem.l (%a1), %d3-%d5
add.l %a2, %a1
LEFT8_PW %d3, %d4
lsl.l #8, %d4
lsr.l %d0, %d5
or.l %d5, %d4
movem.l %d3-%d4, (%a0)
add.l %a2, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d5/%a2
lea.l (5*4,%sp), %sp
rts
.po8_0:
movem.l (12,%sp), %d0-%d1 | stride, height
subq.l #4, %d0 | adjust for increment within the loop
1:
move.l (%a1)+, (%a0)+
move.l (%a1), (%a0)
add.l %d0, %a0
add.l %d0, %a1
subq.l #1, %d1
bne.s 1b
rts
/*****************************************************************************/
.align 2
.global MC_put_o_16
.type MC_put_o_16, @function
MC_put_o_16:
lea.l (-7*4,%sp), %sp
movem.l %d2-%d7/%a2, (%sp) | save some registers
movem.l (7*4+4,%sp), %a0-%a2| dest, source, stride
move.l (7*4+16,%sp), %d1 | height
move.l %a1, %d0
and.l #3, %d0
sub.l %d0, %a1
jmp.l (2, %pc, %d0.l*4)
bra.w .po16_0
bra.w .po16_1
bra.w .po16_2
| last table entry coincides with target
.po16_3:
moveq.l #24, %d0 | shift amount
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
LEFT24_PW %d3, %d4
LEFT24_PW %d4, %d5
LEFT24_PW %d5, %d6
lsl.l %d0, %d6
lsr.l #8, %d7
or.l %d7, %d6
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d7/%a2
lea.l (7*4,%sp), %sp
rts
.po16_2:
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
swap %d3
swap %d4
move.w %d4, %d3
swap %d5
move.w %d5, %d4
swap %d6
move.w %d6, %d5
swap %d7
move.w %d7, %d6
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d7/%a2
lea.l (7*4,%sp), %sp
rts
.po16_1:
moveq.l #24, %d0 | shift amount
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
LEFT8_PW %d3, %d4
LEFT8_PW %d4, %d5
LEFT8_PW %d5, %d6
lsl.l #8, %d6
lsr.l %d0, %d7
or.l %d7, %d6
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d7/%a2
lea.l (7*4,%sp), %sp
rts
.po16_0:
1:
movem.l (%a1), %d3-%d6
add.l %a2, %a1
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %d1
bne.s 1b
movem.l (%sp), %d2-%d7/%a2
lea.l (7*4,%sp), %sp
rts
/*****************************************************************************/
.macro AVG_PW dW1, dW2 | needs %d0 == 24, clobbers %d1, %d2,
move.l \dW1, %d1 | changes dW1, keeps dW2
lsl.l #8, \dW1
move.l \dW2, %d2
lsr.l %d0, %d2
or.l %d2, \dW1
move.l %d1, %d2
eor.l \dW1, %d1
and.l %d2, \dW1
move.l #0xfefefefe, %d2
and.l %d1, %d2
eor.l %d2, %d1
lsr.l #1, %d2
add.l %d2, \dW1
add.l %d1, \dW1
.endm
/*****************************************************************************/
.align 2
.global MC_put_x_8
.type MC_put_x_8, @function
MC_put_x_8:
lea.l (-6*4,%sp), %sp
movem.l %d2-%d6/%a2, (%sp) | save some registers
movem.l (6*4+4,%sp), %a0-%a2| dest, source, stride
move.l (6*4+16,%sp), %d6 | height
move.l %a1, %d0
and.l #3, %d0
sub.l %d0, %a1
jmp.l (2, %pc, %d0.l*4)
bra.w .px8_0
bra.w .px8_1
bra.w .px8_2
| last table entry coincides with target
.px8_3:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d5
add.l %a2, %a1
LEFT24_PW %d3, %d4
LEFT24_PW %d4, %d5
lsl.l %d0, %d5
AVG_PW %d3, %d4
AVG_PW %d4, %d5
movem.l %d3-%d4, (%a0)
add.l %a2, %a0
subq.l #1, %d6
bne.s 1b
movem.l (%sp), %d2-%d6/%a2
lea.l (6*4,%sp), %sp
rts
.px8_2:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d5
add.l %a2, %a1
swap %d3
swap %d4
move.w %d4, %d3
swap %d5
move.w %d5, %d4
AVG_PW %d3, %d4
AVG_PW %d4, %d5
movem.l %d3-%d4, (%a0)
add.l %a2, %a0
subq.l #1, %d6
bne.s 1b
movem.l (%sp), %d2-%d6/%a2
lea.l (6*4,%sp), %sp
rts
.px8_1:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d5
add.l %a2, %a1
LEFT8_PW %d3, %d4
LEFT8_PW %d4, %d5
lsl.l #8, %d5
AVG_PW %d3, %d4
AVG_PW %d4, %d5
movem.l %d3-%d4, (%a0)
add.l %a2, %a0
subq.l #1, %d6
bne.s 1b
movem.l (%sp), %d2-%d6/%a2
lea.l (6*4,%sp), %sp
rts
.px8_0:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d5
add.l %a2, %a1
AVG_PW %d3, %d4
AVG_PW %d4, %d5
movem.l %d3-%d4, (%a0)
add.l %a2, %a0
subq.l #1, %d6
bne.s 1b
movem.l (%sp), %d2-%d6/%a2
lea.l (6*4,%sp), %sp
rts
/*****************************************************************************/
.align 2
.global MC_put_x_16
.type MC_put_x_16, @function
MC_put_x_16:
lea.l (-8*4,%sp), %sp
movem.l %d2-%d7/%a2-%a3, (%sp) | save some registers
movem.l (8*4+4,%sp), %a0-%a3 | dest, source, stride, height
move.l %a1, %d0
and.l #3, %d0
sub.l %d0, %a1
jmp.l (2, %pc, %d0.l*4)
bra.w .px16_0
bra.w .px16_1
bra.w .px16_2
| last table entry coincides with target
.px16_3:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
LEFT24_PW %d3, %d4
LEFT24_PW %d4, %d5
LEFT24_PW %d5, %d6
LEFT24_PW %d6, %d7
lsl.l %d0, %d7
AVG_PW %d3, %d4
AVG_PW %d4, %d5
AVG_PW %d5, %d6
AVG_PW %d6, %d7
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %a3
tst.l %a3
bne.w 1b
movem.l (%sp), %d2-%d7/%a2-%a3
lea.l (8*4,%sp), %sp
rts
.px16_2:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
swap %d3
swap %d4
move.w %d4, %d3
swap %d5
move.w %d5, %d4
swap %d6
move.w %d6, %d5
swap %d7
move.w %d7, %d6
AVG_PW %d3, %d4
AVG_PW %d4, %d5
AVG_PW %d5, %d6
AVG_PW %d6, %d7
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %a3
tst.l %a3
bne.w 1b
movem.l (%sp), %d2-%d7/%a2-%a3
lea.l (8*4,%sp), %sp
rts
.px16_1:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
LEFT8_PW %d3, %d4
LEFT8_PW %d4, %d5
LEFT8_PW %d5, %d6
LEFT8_PW %d6, %d7
lsl.l #8, %d7
AVG_PW %d3, %d4
AVG_PW %d4, %d5
AVG_PW %d5, %d6
AVG_PW %d6, %d7
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %a3
tst.l %a3
bne.w 1b
movem.l (%sp), %d2-%d7/%a2-%a3
lea.l (8*4,%sp), %sp
rts
.px16_0:
moveq.l #24, %d0
1:
movem.l (%a1), %d3-%d7
add.l %a2, %a1
AVG_PW %d3, %d4
AVG_PW %d4, %d5
AVG_PW %d5, %d6
AVG_PW %d6, %d7
movem.l %d3-%d6, (%a0)
add.l %a2, %a0
subq.l #1, %a3
tst.l %a3
bne.w 1b
movem.l (%sp), %d2-%d7/%a2-%a3
lea.l (8*4,%sp), %sp
rts
Reference in a new issue View git blame Copy permalink