2007-06-03 21:20:06 +00:00
|
|
|
/* Optimized memcpy implementation for PowerPC64.
|
2011-04-23 01:34:32 +00:00
|
|
|
Copyright (C) 2003, 2006, 2007, 2011 Free Software Foundation, Inc.
|
2007-06-03 21:20:06 +00:00
|
|
|
This file is part of the GNU C Library.
|
|
|
|
|
|
|
|
The GNU C Library 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.
|
|
|
|
|
|
|
|
The GNU C Library 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
|
2012-02-09 23:18:22 +00:00
|
|
|
License along with the GNU C Library; if not, see
|
|
|
|
<http://www.gnu.org/licenses/>. */
|
2007-06-03 21:20:06 +00:00
|
|
|
|
|
|
|
#include <sysdep.h>
|
|
|
|
#include <bp-sym.h>
|
|
|
|
#include <bp-asm.h>
|
|
|
|
|
|
|
|
/* __ptr_t [r3] memcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
|
|
|
|
Returns 'dst'.
|
|
|
|
|
|
|
|
Memcpy handles short copies (< 32-bytes) using a binary move blocks
|
|
|
|
(no loops) of lwz/stw. The tail (remaining 1-3) bytes is handled
|
|
|
|
with the appropriate combination of byte and halfword load/stores.
|
|
|
|
There is minimal effort to optimize the alignment of short moves.
|
|
|
|
The 64-bit implementations of POWER3 and POWER4 do a reasonable job
|
|
|
|
of handling unligned load/stores that do not cross 32-byte boundries.
|
|
|
|
|
|
|
|
Longer moves (>= 32-bytes) justify the effort to get at least the
|
|
|
|
destination doubleword (8-byte) aligned. Further optimization is
|
|
|
|
posible when both source and destination are doubleword aligned.
|
|
|
|
Each case has a optimized unrolled loop.
|
|
|
|
|
|
|
|
For POWER6 unaligned loads will take a 20+ cycle hicup for any
|
|
|
|
L1 cache miss that crosses a 32- or 128-byte boundary. Store
|
|
|
|
is more forgiving and does not take a hicup until page or
|
|
|
|
segment boundaries. So we require doubleword alignment for
|
|
|
|
the source but may take a risk and only require word alignment
|
|
|
|
for the destination. */
|
|
|
|
|
|
|
|
.machine "power6"
|
|
|
|
EALIGN (BP_SYM (memcpy), 7, 0)
|
|
|
|
CALL_MCOUNT 3
|
|
|
|
|
|
|
|
cmpldi cr1,5,31
|
|
|
|
neg 0,3
|
|
|
|
std 3,-16(1)
|
|
|
|
std 31,-8(1)
|
|
|
|
andi. 11,3,7 /* check alignement of dst. */
|
|
|
|
clrldi 0,0,61 /* Number of bytes until the 1st doubleword of dst. */
|
|
|
|
clrldi 10,4,61 /* check alignement of src. */
|
|
|
|
cmpldi cr6,5,8
|
|
|
|
ble- cr1,.L2 /* If move < 32 bytes use short move code. */
|
|
|
|
mtcrf 0x01,0
|
|
|
|
cmpld cr6,10,11
|
|
|
|
srdi 9,5,3 /* Number of full double words remaining. */
|
|
|
|
beq .L0
|
|
|
|
|
|
|
|
subf 5,0,5
|
|
|
|
/* Move 0-7 bytes as needed to get the destination doubleword alligned.
|
|
|
|
Duplicate some code to maximize fall-throught and minimize agen delays. */
|
|
|
|
1: bf 31,2f
|
|
|
|
lbz 6,0(4)
|
|
|
|
stb 6,0(3)
|
|
|
|
bf 30,5f
|
|
|
|
lhz 6,1(4)
|
|
|
|
sth 6,1(3)
|
|
|
|
bf 29,0f
|
|
|
|
lwz 6,3(4)
|
|
|
|
stw 6,3(3)
|
|
|
|
b 0f
|
|
|
|
5:
|
|
|
|
bf 29,0f
|
|
|
|
lwz 6,1(4)
|
|
|
|
stw 6,1(3)
|
|
|
|
b 0f
|
|
|
|
|
|
|
|
2: bf 30,4f
|
|
|
|
lhz 6,0(4)
|
|
|
|
sth 6,0(3)
|
|
|
|
bf 29,0f
|
|
|
|
lwz 6,2(4)
|
|
|
|
stw 6,2(3)
|
|
|
|
b 0f
|
|
|
|
|
|
|
|
4: bf 29,0f
|
|
|
|
lwz 6,0(4)
|
|
|
|
stw 6,0(3)
|
|
|
|
0:
|
|
|
|
/* Add the number of bytes until the 1st doubleword of dst to src and dst. */
|
|
|
|
add 4,4,0
|
|
|
|
add 3,3,0
|
|
|
|
|
|
|
|
clrldi 10,4,61 /* check alignement of src again. */
|
|
|
|
srdi 9,5,3 /* Number of full double words remaining. */
|
|
|
|
|
|
|
|
/* Copy doublewords from source to destination, assumpting the
|
|
|
|
destination is aligned on a doubleword boundary.
|
|
|
|
|
|
|
|
At this point we know there are at least 25 bytes left (32-7) to copy.
|
|
|
|
The next step is to determine if the source is also doubleword aligned.
|
|
|
|
If not branch to the unaligned move code at .L6. which uses
|
|
|
|
a load, shift, store strategy.
|
|
|
|
|
|
|
|
Otherwise source and destination are doubleword aligned, and we can
|
|
|
|
the optimized doubleword copy loop. */
|
|
|
|
.align 4
|
|
|
|
.L0:
|
|
|
|
clrldi 11,5,61
|
|
|
|
andi. 0,5,0x78
|
|
|
|
srdi 12,5,7 /* Number of 128-byte blocks to move. */
|
|
|
|
cmpldi cr1,11,0 /* If the tail is 0 bytes */
|
|
|
|
bne- cr6,.L6 /* If source is not DW aligned. */
|
|
|
|
|
|
|
|
/* Move doublewords where destination and source are DW aligned.
|
|
|
|
Use a unrolled loop to copy 16 doublewords (128-bytes) per iteration.
|
2011-04-23 01:34:32 +00:00
|
|
|
If the copy is not an exact multiple of 128 bytes, 1-15
|
2007-06-03 21:20:06 +00:00
|
|
|
doublewords are copied as needed to set up the main loop. After
|
|
|
|
the main loop exits there may be a tail of 1-7 bytes. These byte
|
|
|
|
are copied a word/halfword/byte at a time as needed to preserve
|
|
|
|
alignment.
|
|
|
|
|
|
|
|
For POWER6 the L1 is store-through and the L2 is store-in. The
|
|
|
|
L2 is clocked at half CPU clock so we can store 16 bytes every
|
|
|
|
other cycle. POWER6 also has a load/store bypass so we can do
|
|
|
|
load, load, store, store every 2 cycles.
|
|
|
|
|
|
|
|
The following code is sensitive to cache line alignment. Do not
|
|
|
|
make any change with out first making sure thay don't result in
|
|
|
|
splitting ld/std pairs across a cache line. */
|
|
|
|
|
|
|
|
mtcrf 0x02,5
|
|
|
|
mtcrf 0x01,5
|
|
|
|
cmpldi cr5,12,1
|
|
|
|
beq L(das_loop)
|
|
|
|
|
|
|
|
bf 25,4f
|
|
|
|
.align 3
|
|
|
|
ld 6,0(4)
|
|
|
|
ld 7,8(4)
|
|
|
|
mr 11,4
|
|
|
|
mr 10,3
|
|
|
|
std 6,0(3)
|
|
|
|
std 7,8(3)
|
|
|
|
ld 6,16(4)
|
|
|
|
ld 7,24(4)
|
|
|
|
std 6,16(3)
|
|
|
|
std 7,24(3)
|
|
|
|
ld 6,0+32(4)
|
|
|
|
ld 7,8+32(4)
|
|
|
|
addi 4,4,64
|
|
|
|
addi 3,3,64
|
|
|
|
std 6,0+32(10)
|
|
|
|
std 7,8+32(10)
|
|
|
|
ld 6,16+32(11)
|
|
|
|
ld 7,24+32(11)
|
|
|
|
std 6,16+32(10)
|
|
|
|
std 7,24+32(10)
|
|
|
|
4:
|
|
|
|
mr 10,3
|
|
|
|
bf 26,2f
|
|
|
|
ld 6,0(4)
|
|
|
|
ld 7,8(4)
|
|
|
|
mr 11,4
|
|
|
|
nop
|
|
|
|
std 6,0(3)
|
|
|
|
std 7,8(3)
|
|
|
|
ld 6,16(4)
|
|
|
|
ld 7,24(4)
|
|
|
|
addi 4,4,32
|
|
|
|
std 6,16(3)
|
|
|
|
std 7,24(3)
|
|
|
|
addi 3,3,32
|
|
|
|
6:
|
|
|
|
nop
|
|
|
|
bf 27,5f
|
|
|
|
ld 6,0+32(11)
|
|
|
|
ld 7,8+32(11)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 3,3,16
|
|
|
|
std 6,0+32(10)
|
|
|
|
std 7,8+32(10)
|
|
|
|
bf 28,L(das_loop_s)
|
|
|
|
ld 0,16+32(11)
|
|
|
|
addi 4,4,8
|
|
|
|
addi 3,3,8
|
|
|
|
std 0,16+32(10)
|
|
|
|
blt cr5,L(das_tail)
|
|
|
|
b L(das_loop)
|
|
|
|
.align 3
|
|
|
|
5:
|
|
|
|
nop
|
|
|
|
bf 28,L(das_loop_s)
|
|
|
|
ld 6,32(11)
|
|
|
|
addi 4,4,8
|
|
|
|
addi 3,3,8
|
|
|
|
std 6,32(10)
|
|
|
|
blt cr5,L(das_tail)
|
|
|
|
b L(das_loop)
|
|
|
|
.align 3
|
|
|
|
2:
|
|
|
|
mr 11,4
|
|
|
|
bf 27,1f
|
|
|
|
ld 6,0(4)
|
|
|
|
ld 7,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 3,3,16
|
|
|
|
std 6,0(10)
|
|
|
|
std 7,8(10)
|
|
|
|
bf 28,L(das_loop_s)
|
|
|
|
ld 0,16(11)
|
|
|
|
addi 4,11,24
|
|
|
|
addi 3,10,24
|
|
|
|
std 0,16(10)
|
|
|
|
blt cr5,L(das_tail)
|
|
|
|
b L(das_loop)
|
|
|
|
.align 3
|
|
|
|
1:
|
|
|
|
nop
|
|
|
|
bf 28,L(das_loop_s)
|
|
|
|
ld 6,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
addi 3,3,8
|
|
|
|
std 6,0(10)
|
|
|
|
L(das_loop_s):
|
|
|
|
nop
|
|
|
|
blt cr5,L(das_tail)
|
|
|
|
.align 4
|
|
|
|
L(das_loop):
|
|
|
|
ld 6,0(4)
|
|
|
|
ld 7,8(4)
|
|
|
|
mr 10,3
|
|
|
|
mr 11,4
|
|
|
|
std 6,0(3)
|
|
|
|
std 7,8(3)
|
|
|
|
addi 12,12,-1
|
|
|
|
nop
|
|
|
|
ld 8,16(4)
|
|
|
|
ld 0,24(4)
|
|
|
|
std 8,16(3)
|
|
|
|
std 0,24(3)
|
|
|
|
|
|
|
|
ld 6,0+32(4)
|
|
|
|
ld 7,8+32(4)
|
|
|
|
std 6,0+32(3)
|
|
|
|
std 7,8+32(3)
|
|
|
|
ld 8,16+32(4)
|
|
|
|
ld 0,24+32(4)
|
|
|
|
std 8,16+32(3)
|
|
|
|
std 0,24+32(3)
|
|
|
|
|
|
|
|
ld 6,0+64(11)
|
|
|
|
ld 7,8+64(11)
|
|
|
|
std 6,0+64(10)
|
|
|
|
std 7,8+64(10)
|
|
|
|
ld 8,16+64(11)
|
|
|
|
ld 0,24+64(11)
|
|
|
|
std 8,16+64(10)
|
|
|
|
std 0,24+64(10)
|
|
|
|
|
|
|
|
ld 6,0+96(11)
|
|
|
|
ld 7,8+96(11)
|
|
|
|
addi 4,4,128
|
|
|
|
addi 3,3,128
|
|
|
|
std 6,0+96(10)
|
|
|
|
std 7,8+96(10)
|
|
|
|
ld 8,16+96(11)
|
|
|
|
ld 0,24+96(11)
|
|
|
|
std 8,16+96(10)
|
|
|
|
std 0,24+96(10)
|
|
|
|
ble cr5,L(das_loop_e)
|
|
|
|
|
|
|
|
mtctr 12
|
|
|
|
.align 4
|
|
|
|
L(das_loop2):
|
|
|
|
ld 6,0(4)
|
|
|
|
ld 7,8(4)
|
|
|
|
mr 10,3
|
|
|
|
mr 11,4
|
|
|
|
std 6,0(3)
|
|
|
|
std 7,8(3)
|
|
|
|
ld 8,16(4)
|
|
|
|
ld 0,24(4)
|
|
|
|
std 8,16(3)
|
|
|
|
std 0,24(3)
|
|
|
|
|
|
|
|
ld 6,0+32(4)
|
|
|
|
ld 7,8+32(4)
|
|
|
|
std 6,0+32(3)
|
|
|
|
std 7,8+32(3)
|
|
|
|
ld 8,16+32(4)
|
|
|
|
ld 0,24+32(4)
|
|
|
|
std 8,16+32(3)
|
|
|
|
std 0,24+32(3)
|
|
|
|
|
|
|
|
ld 6,0+64(11)
|
|
|
|
ld 7,8+64(11)
|
|
|
|
std 6,0+64(10)
|
|
|
|
std 7,8+64(10)
|
|
|
|
ld 8,16+64(11)
|
|
|
|
ld 0,24+64(11)
|
|
|
|
std 8,16+64(10)
|
|
|
|
std 0,24+64(10)
|
|
|
|
|
|
|
|
ld 6,0+96(11)
|
|
|
|
ld 7,8+96(11)
|
|
|
|
addi 4,4,128
|
|
|
|
addi 3,3,128
|
|
|
|
std 6,0+96(10)
|
|
|
|
std 7,8+96(10)
|
|
|
|
ld 8,16+96(11)
|
|
|
|
ld 0,24+96(11)
|
|
|
|
std 8,16+96(10)
|
|
|
|
std 0,24+96(10)
|
|
|
|
bdnz L(das_loop2)
|
|
|
|
L(das_loop_e):
|
|
|
|
/* Check of a 1-7 byte tail, return if none. */
|
|
|
|
bne cr1,L(das_tail2)
|
|
|
|
/* Return original dst pointer. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(das_tail):
|
|
|
|
beq cr1,0f
|
|
|
|
|
|
|
|
L(das_tail2):
|
|
|
|
/* At this point we have a tail of 0-7 bytes and we know that the
|
|
|
|
destiniation is double word aligned. */
|
|
|
|
4: bf 29,2f
|
|
|
|
lwz 6,0(4)
|
|
|
|
stw 6,0(3)
|
|
|
|
bf 30,5f
|
|
|
|
lhz 6,4(4)
|
|
|
|
sth 6,4(3)
|
|
|
|
bf 31,0f
|
|
|
|
lbz 6,6(4)
|
|
|
|
stb 6,6(3)
|
|
|
|
b 0f
|
|
|
|
5: bf 31,0f
|
|
|
|
lbz 6,4(4)
|
|
|
|
stb 6,4(3)
|
|
|
|
b 0f
|
|
|
|
|
|
|
|
2: bf 30,1f
|
|
|
|
lhz 6,0(4)
|
|
|
|
sth 6,0(3)
|
|
|
|
bf 31,0f
|
|
|
|
lbz 6,2(4)
|
|
|
|
stb 6,2(3)
|
|
|
|
b 0f
|
|
|
|
|
|
|
|
1: bf 31,0f
|
|
|
|
lbz 6,0(4)
|
|
|
|
stb 6,0(3)
|
|
|
|
0:
|
|
|
|
/* Return original dst pointer. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
|
|
|
|
/* Copy up to 31 bytes. This divided into two cases 0-8 bytes and 9-31
|
|
|
|
bytes. Each case is handled without loops, using binary (1,2,4,8)
|
|
|
|
tests.
|
|
|
|
|
|
|
|
In the short (0-8 byte) case no attempt is made to force alignment
|
|
|
|
of either source or destination. The hardware will handle the
|
|
|
|
unaligned load/stores with small delays for crossing 32- 128-byte,
|
|
|
|
and 4096-byte boundaries. Since these short moves are unlikely to be
|
|
|
|
unaligned or cross these boundaries, the overhead to force
|
|
|
|
alignment is not justified.
|
|
|
|
|
|
|
|
The longer (9-31 byte) move is more likely to cross 32- or 128-byte
|
|
|
|
boundaries. Since only loads are sensitive to the 32-/128-byte
|
|
|
|
boundaries it is more important to align the source then the
|
|
|
|
destination. If the source is not already word aligned, we first
|
|
|
|
move 1-3 bytes as needed. Since we are only word aligned we don't
|
|
|
|
use double word load/stores to insure that all loads are aligned.
|
|
|
|
While the destination and stores may still be unaligned, this
|
|
|
|
is only an issue for page (4096 byte boundary) crossing, which
|
|
|
|
should be rare for these short moves. The hardware handles this
|
|
|
|
case automatically with a small (~20 cycle) delay. */
|
|
|
|
.align 4
|
|
|
|
.L2:
|
|
|
|
mtcrf 0x01,5
|
|
|
|
neg 8,4
|
|
|
|
clrrdi 11,4,2
|
|
|
|
andi. 0,8,3
|
|
|
|
ble cr6,.LE8 /* Handle moves of 0-8 bytes. */
|
|
|
|
/* At least 9 bytes left. Get the source word aligned. */
|
|
|
|
cmpldi cr1,5,16
|
|
|
|
mr 10,5
|
|
|
|
mr 12,4
|
|
|
|
cmpldi cr6,0,2
|
|
|
|
beq L(dus_tail) /* If the source is already word aligned skip this. */
|
|
|
|
/* Copy 1-3 bytes to get source address word aligned. */
|
|
|
|
lwz 6,0(11)
|
|
|
|
subf 10,0,5
|
|
|
|
add 12,4,0
|
|
|
|
blt cr6,5f
|
|
|
|
srdi 7,6,16
|
|
|
|
bgt cr6,3f
|
|
|
|
sth 6,0(3)
|
|
|
|
b 7f
|
|
|
|
.align 4
|
|
|
|
3:
|
|
|
|
stb 7,0(3)
|
|
|
|
sth 6,1(3)
|
|
|
|
b 7f
|
|
|
|
.align 4
|
|
|
|
5:
|
|
|
|
stb 6,0(3)
|
|
|
|
7:
|
|
|
|
cmpldi cr1,10,16
|
|
|
|
add 3,3,0
|
|
|
|
mtcrf 0x01,10
|
|
|
|
.align 4
|
|
|
|
L(dus_tail):
|
|
|
|
/* At least 6 bytes left and the source is word aligned. This allows
|
|
|
|
some speculative loads up front. */
|
|
|
|
/* We need to special case the fall-through because the biggest delays
|
|
|
|
are due to address computation not being ready in time for the
|
|
|
|
AGEN. */
|
|
|
|
lwz 6,0(12)
|
|
|
|
lwz 7,4(12)
|
|
|
|
blt cr1,L(dus_tail8)
|
|
|
|
cmpldi cr0,10,24
|
|
|
|
L(dus_tail16): /* Move 16 bytes. */
|
|
|
|
stw 6,0(3)
|
|
|
|
stw 7,4(3)
|
|
|
|
lwz 6,8(12)
|
|
|
|
lwz 7,12(12)
|
|
|
|
stw 6,8(3)
|
|
|
|
stw 7,12(3)
|
|
|
|
/* Move 8 bytes more. */
|
|
|
|
bf 28,L(dus_tail16p8)
|
|
|
|
cmpldi cr1,10,28
|
|
|
|
lwz 6,16(12)
|
|
|
|
lwz 7,20(12)
|
|
|
|
stw 6,16(3)
|
|
|
|
stw 7,20(3)
|
|
|
|
/* Move 4 bytes more. */
|
|
|
|
bf 29,L(dus_tail16p4)
|
|
|
|
lwz 6,24(12)
|
|
|
|
stw 6,24(3)
|
|
|
|
addi 12,12,28
|
|
|
|
addi 3,3,28
|
|
|
|
bgt cr1,L(dus_tail2)
|
|
|
|
/* exactly 28 bytes. Return original dst pointer and exit. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(dus_tail16p8): /* less then 8 bytes left. */
|
|
|
|
beq cr1,L(dus_tailX) /* exactly 16 bytes, early exit. */
|
|
|
|
cmpldi cr1,10,20
|
|
|
|
bf 29,L(dus_tail16p2)
|
|
|
|
/* Move 4 bytes more. */
|
|
|
|
lwz 6,16(12)
|
|
|
|
stw 6,16(3)
|
|
|
|
addi 12,12,20
|
|
|
|
addi 3,3,20
|
|
|
|
bgt cr1,L(dus_tail2)
|
|
|
|
/* exactly 20 bytes. Return original dst pointer and exit. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(dus_tail16p4): /* less then 4 bytes left. */
|
|
|
|
addi 12,12,24
|
|
|
|
addi 3,3,24
|
|
|
|
bgt cr0,L(dus_tail2)
|
|
|
|
/* exactly 24 bytes. Return original dst pointer and exit. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(dus_tail16p2): /* 16 bytes moved, less then 4 bytes left. */
|
|
|
|
addi 12,12,16
|
|
|
|
addi 3,3,16
|
|
|
|
b L(dus_tail2)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(dus_tail8): /* Move 8 bytes. */
|
|
|
|
/* r6, r7 already loaded speculatively. */
|
|
|
|
cmpldi cr1,10,8
|
|
|
|
cmpldi cr0,10,12
|
|
|
|
bf 28,L(dus_tail4)
|
|
|
|
.align 2
|
|
|
|
stw 6,0(3)
|
|
|
|
stw 7,4(3)
|
|
|
|
/* Move 4 bytes more. */
|
|
|
|
bf 29,L(dus_tail8p4)
|
|
|
|
lwz 6,8(12)
|
|
|
|
stw 6,8(3)
|
|
|
|
addi 12,12,12
|
|
|
|
addi 3,3,12
|
|
|
|
bgt cr0,L(dus_tail2)
|
|
|
|
/* exactly 12 bytes. Return original dst pointer and exit. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(dus_tail8p4): /* less then 4 bytes left. */
|
|
|
|
addi 12,12,8
|
|
|
|
addi 3,3,8
|
|
|
|
bgt cr1,L(dus_tail2)
|
|
|
|
/* exactly 8 bytes. Return original dst pointer and exit. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(dus_tail4): /* Move 4 bytes. */
|
|
|
|
/* r6 already loaded speculatively. If we are here we know there is
|
|
|
|
more then 4 bytes left. So there is no need to test. */
|
|
|
|
addi 12,12,4
|
|
|
|
stw 6,0(3)
|
|
|
|
addi 3,3,4
|
|
|
|
L(dus_tail2): /* Move 2-3 bytes. */
|
|
|
|
bf 30,L(dus_tail1)
|
|
|
|
lhz 6,0(12)
|
|
|
|
sth 6,0(3)
|
|
|
|
bf 31,L(dus_tailX)
|
|
|
|
lbz 7,2(12)
|
|
|
|
stb 7,2(3)
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
L(dus_tail1): /* Move 1 byte. */
|
|
|
|
bf 31,L(dus_tailX)
|
|
|
|
lbz 6,0(12)
|
|
|
|
stb 6,0(3)
|
|
|
|
L(dus_tailX):
|
|
|
|
/* Return original dst pointer. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
|
|
|
|
/* Special case to copy 0-8 bytes. */
|
|
|
|
.align 4
|
|
|
|
.LE8:
|
|
|
|
mr 12,4
|
|
|
|
bne cr6,L(dus_4)
|
|
|
|
/* Exactly 8 bytes. We may cross a 32-/128-byte boundry and take a ~20
|
|
|
|
cycle delay. This case should be rare and any attempt to avoid this
|
|
|
|
would take most of 20 cycles any way. */
|
|
|
|
ld 6,0(4)
|
|
|
|
std 6,0(3)
|
|
|
|
/* Return original dst pointer. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(dus_4):
|
|
|
|
bf 29,L(dus_tail2)
|
|
|
|
lwz 6,0(4)
|
|
|
|
stw 6,0(3)
|
|
|
|
bf 30,L(dus_5)
|
|
|
|
lhz 7,4(4)
|
|
|
|
sth 7,4(3)
|
|
|
|
bf 31,L(dus_0)
|
|
|
|
lbz 8,6(4)
|
|
|
|
stb 8,6(3)
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
.align 4
|
|
|
|
L(dus_5):
|
|
|
|
bf 31,L(dus_0)
|
|
|
|
lbz 6,4(4)
|
|
|
|
stb 6,4(3)
|
|
|
|
L(dus_0):
|
|
|
|
/* Return original dst pointer. */
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
.L6:
|
|
|
|
cfi_offset(31,-8)
|
|
|
|
mr 12,4
|
|
|
|
mr 31,5
|
|
|
|
/* Copy doublewords where the destination is aligned but the source is
|
|
|
|
not. Use aligned doubleword loads from the source, shifted to realign
|
|
|
|
the data, to allow aligned destination stores. */
|
|
|
|
addi 11,9,-1 /* loop DW count is one less than total */
|
|
|
|
subf 5,10,12 /* Move source addr to previous full double word. */
|
|
|
|
cmpldi cr5, 10, 2
|
|
|
|
cmpldi cr0, 10, 4
|
|
|
|
mr 4,3
|
|
|
|
srdi 8,11,2 /* calculate the 32 byte loop count */
|
|
|
|
ld 6,0(5) /* pre load 1st full doubleword. */
|
|
|
|
mtcrf 0x01,11
|
|
|
|
cmpldi cr6,9,4
|
|
|
|
mtctr 8
|
|
|
|
ld 7,8(5) /* pre load 2nd full doubleword. */
|
|
|
|
bge cr0, L(du4_do)
|
|
|
|
blt cr5, L(du1_do)
|
|
|
|
beq cr5, L(du2_do)
|
|
|
|
b L(du3_do)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du1_do):
|
|
|
|
bf 30,L(du1_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 8
|
|
|
|
srdi 8,7, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 8
|
|
|
|
srdi 8,6, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du1_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du1_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 8
|
|
|
|
srdi 8,7, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du1_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du1_loop)
|
|
|
|
.align 4
|
|
|
|
L(du1_1dw):
|
|
|
|
sldi 0,6, 8
|
|
|
|
srdi 8,7, 64-8
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du1_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du1_loop):
|
|
|
|
sldi 0,6, 8
|
|
|
|
srdi 8,7, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 8
|
|
|
|
srdi 8,6, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 8
|
|
|
|
srdi 8,7, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 8
|
|
|
|
srdi 8,6, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du1_loop)
|
|
|
|
.align 4
|
|
|
|
L(du1_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 8
|
|
|
|
srdi 8,7, 64-8
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du2_do):
|
|
|
|
bf 30,L(du2_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 16
|
|
|
|
srdi 8,7, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 16
|
|
|
|
srdi 8,6, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du2_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du2_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 16
|
|
|
|
srdi 8,7, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du2_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du2_loop)
|
|
|
|
.align 4
|
|
|
|
L(du2_1dw):
|
|
|
|
sldi 0,6, 16
|
|
|
|
srdi 8,7, 64-16
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du2_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du2_loop):
|
|
|
|
sldi 0,6, 16
|
|
|
|
srdi 8,7, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 16
|
|
|
|
srdi 8,6, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 16
|
|
|
|
srdi 8,7, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 16
|
|
|
|
srdi 8,6, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du2_loop)
|
|
|
|
.align 4
|
|
|
|
L(du2_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 16
|
|
|
|
srdi 8,7, 64-16
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du3_do):
|
|
|
|
bf 30,L(du3_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 24
|
|
|
|
srdi 8,7, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 24
|
|
|
|
srdi 8,6, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du3_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du3_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 24
|
|
|
|
srdi 8,7, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du3_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du3_loop)
|
|
|
|
.align 4
|
|
|
|
L(du3_1dw):
|
|
|
|
sldi 0,6, 24
|
|
|
|
srdi 8,7, 64-24
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du3_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du3_loop):
|
|
|
|
sldi 0,6, 24
|
|
|
|
srdi 8,7, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 24
|
|
|
|
srdi 8,6, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 24
|
|
|
|
srdi 8,7, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 24
|
|
|
|
srdi 8,6, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du3_loop)
|
|
|
|
.align 4
|
|
|
|
L(du3_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 24
|
|
|
|
srdi 8,7, 64-24
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du4_do):
|
|
|
|
cmpldi cr5, 10, 6
|
|
|
|
beq cr0, L(du4_dox)
|
|
|
|
blt cr5, L(du5_do)
|
|
|
|
beq cr5, L(du6_do)
|
|
|
|
b L(du7_do)
|
|
|
|
L(du4_dox):
|
|
|
|
bf 30,L(du4_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 32
|
|
|
|
srdi 8,7, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 32
|
|
|
|
srdi 8,6, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du4_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du4_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 32
|
|
|
|
srdi 8,7, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du4_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du4_loop)
|
|
|
|
.align 4
|
|
|
|
L(du4_1dw):
|
|
|
|
sldi 0,6, 32
|
|
|
|
srdi 8,7, 64-32
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du4_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du4_loop):
|
|
|
|
sldi 0,6, 32
|
|
|
|
srdi 8,7, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 32
|
|
|
|
srdi 8,6, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 32
|
|
|
|
srdi 8,7, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 32
|
|
|
|
srdi 8,6, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du4_loop)
|
|
|
|
.align 4
|
|
|
|
L(du4_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 32
|
|
|
|
srdi 8,7, 64-32
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du5_do):
|
|
|
|
bf 30,L(du5_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 40
|
|
|
|
srdi 8,7, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 40
|
|
|
|
srdi 8,6, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du5_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du5_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 40
|
|
|
|
srdi 8,7, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du5_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du5_loop)
|
|
|
|
.align 4
|
|
|
|
L(du5_1dw):
|
|
|
|
sldi 0,6, 40
|
|
|
|
srdi 8,7, 64-40
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du5_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du5_loop):
|
|
|
|
sldi 0,6, 40
|
|
|
|
srdi 8,7, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 40
|
|
|
|
srdi 8,6, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 40
|
|
|
|
srdi 8,7, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 40
|
|
|
|
srdi 8,6, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du5_loop)
|
|
|
|
.align 4
|
|
|
|
L(du5_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 40
|
|
|
|
srdi 8,7, 64-40
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du6_do):
|
|
|
|
bf 30,L(du6_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 48
|
|
|
|
srdi 8,7, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 48
|
|
|
|
srdi 8,6, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du6_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du6_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 48
|
|
|
|
srdi 8,7, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du6_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du6_loop)
|
|
|
|
.align 4
|
|
|
|
L(du6_1dw):
|
|
|
|
sldi 0,6, 48
|
|
|
|
srdi 8,7, 64-48
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du6_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du6_loop):
|
|
|
|
sldi 0,6, 48
|
|
|
|
srdi 8,7, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 48
|
|
|
|
srdi 8,6, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 48
|
|
|
|
srdi 8,7, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 48
|
|
|
|
srdi 8,6, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du6_loop)
|
|
|
|
.align 4
|
|
|
|
L(du6_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 48
|
|
|
|
srdi 8,7, 64-48
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du7_do):
|
|
|
|
bf 30,L(du7_1dw)
|
|
|
|
|
|
|
|
/* there are at least two DWs to copy */
|
|
|
|
sldi 0,6, 56
|
|
|
|
srdi 8,7, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 56
|
|
|
|
srdi 8,6, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,8(4)
|
|
|
|
addi 4,4,16
|
|
|
|
addi 5,5,32
|
|
|
|
blt cr6,L(du7_fini) /* if total DWs = 3, then bypass loop */
|
|
|
|
bf 31,L(du7_loop)
|
|
|
|
/* there is a third DW to copy */
|
|
|
|
sldi 0,6, 56
|
|
|
|
srdi 8,7, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
addi 4,4,8
|
|
|
|
beq cr6,L(du7_fini) /* if total DWs = 4, then bypass loop */
|
|
|
|
b L(du7_loop)
|
|
|
|
.align 4
|
|
|
|
L(du7_1dw):
|
|
|
|
sldi 0,6, 56
|
|
|
|
srdi 8,7, 64-56
|
|
|
|
addi 5,5,16
|
|
|
|
or 0,0,8
|
|
|
|
bf 31,L(du7_loop)
|
|
|
|
mr 6,7
|
|
|
|
ld 7,0(5)
|
|
|
|
addi 5,5,8
|
|
|
|
std 0,0(4)
|
|
|
|
addi 4,4,8
|
|
|
|
.align 4
|
|
|
|
/* copy 32 bytes at a time */
|
|
|
|
L(du7_loop):
|
|
|
|
sldi 0,6, 56
|
|
|
|
srdi 8,7, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,0(5)
|
|
|
|
std 0,0(4)
|
|
|
|
sldi 0,7, 56
|
|
|
|
srdi 8,6, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,8(5)
|
|
|
|
std 0,8(4)
|
|
|
|
sldi 0,6, 56
|
|
|
|
srdi 8,7, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
ld 6,16(5)
|
|
|
|
std 0,16(4)
|
|
|
|
sldi 0,7, 56
|
|
|
|
srdi 8,6, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
ld 7,24(5)
|
|
|
|
std 0,24(4)
|
|
|
|
addi 5,5,32
|
|
|
|
addi 4,4,32
|
|
|
|
bdnz+ L(du7_loop)
|
|
|
|
.align 4
|
|
|
|
L(du7_fini):
|
|
|
|
/* calculate and store the final DW */
|
|
|
|
sldi 0,6, 56
|
|
|
|
srdi 8,7, 64-56
|
|
|
|
or 0,0,8
|
|
|
|
std 0,0(4)
|
|
|
|
b L(du_done)
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
L(du_done):
|
|
|
|
rldicr 0,31,0,60
|
|
|
|
mtcrf 0x01,31
|
|
|
|
beq cr1,0f /* If the tail is 0 bytes we are done! */
|
|
|
|
|
|
|
|
add 3,3,0
|
|
|
|
add 12,12,0
|
|
|
|
/* At this point we have a tail of 0-7 bytes and we know that the
|
|
|
|
destiniation is double word aligned. */
|
|
|
|
4: bf 29,2f
|
|
|
|
lwz 6,0(12)
|
|
|
|
addi 12,12,4
|
|
|
|
stw 6,0(3)
|
|
|
|
addi 3,3,4
|
|
|
|
2: bf 30,1f
|
|
|
|
lhz 6,0(12)
|
|
|
|
addi 12,12,2
|
|
|
|
sth 6,0(3)
|
|
|
|
addi 3,3,2
|
|
|
|
1: bf 31,0f
|
|
|
|
lbz 6,0(12)
|
|
|
|
stb 6,0(3)
|
|
|
|
0:
|
|
|
|
/* Return original dst pointer. */
|
|
|
|
ld 31,-8(1)
|
|
|
|
ld 3,-16(1)
|
|
|
|
blr
|
|
|
|
END_GEN_TB (BP_SYM (memcpy),TB_TOCLESS)
|
|
|
|
libc_hidden_builtin_def (memcpy)
|