2007-06-03 21:20:06 +00:00
|
|
|
/* Optimized strcmp implementation for PowerPC64.
|
2013-01-02 19:01:50 +00:00
|
|
|
Copyright (C) 2003-2013 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>
|
|
|
|
|
|
|
|
|
|
/* int [r3] memcmp (const char *s1 [r3], const char *s2 [r4], size_t size [r5]) */
|
|
|
|
|
|
2007-07-12 18:38:01 +00:00
|
|
|
.machine power4
|
2007-06-03 21:20:06 +00:00
|
|
|
EALIGN (BP_SYM(memcmp), 4, 0)
|
|
|
|
|
CALL_MCOUNT
|
|
|
|
|
|
|
|
|
|
#define rTMP r0
|
|
|
|
|
#define rRTN r3
|
|
|
|
|
#define rSTR1 r3 /* first string arg */
|
|
|
|
|
#define rSTR2 r4 /* second string arg */
|
|
|
|
|
#define rN r5 /* max string length */
|
|
|
|
|
#define rWORD1 r6 /* current word in s1 */
|
|
|
|
|
#define rWORD2 r7 /* current word in s2 */
|
|
|
|
|
#define rWORD3 r8 /* next word in s1 */
|
|
|
|
|
#define rWORD4 r9 /* next word in s2 */
|
|
|
|
|
#define rWORD5 r10 /* next word in s1 */
|
|
|
|
|
#define rWORD6 r11 /* next word in s2 */
|
|
|
|
|
#define rBITDIF r12 /* bits that differ in s1 & s2 words */
|
|
|
|
|
#define rWORD7 r30 /* next word in s1 */
|
|
|
|
|
#define rWORD8 r31 /* next word in s2 */
|
|
|
|
|
|
|
|
|
|
xor rTMP, rSTR2, rSTR1
|
|
|
|
|
cmplwi cr6, rN, 0
|
|
|
|
|
cmplwi cr1, rN, 12
|
|
|
|
|
clrlwi. rTMP, rTMP, 30
|
|
|
|
|
clrlwi rBITDIF, rSTR1, 30
|
|
|
|
|
cmplwi cr5, rBITDIF, 0
|
|
|
|
|
beq- cr6, L(zeroLength)
|
|
|
|
|
dcbt 0,rSTR1
|
|
|
|
|
dcbt 0,rSTR2
|
|
|
|
|
/* If less than 8 bytes or not aligned, use the unaligned
|
|
|
|
|
byte loop. */
|
|
|
|
|
blt cr1, L(bytealigned)
|
|
|
|
|
stwu 1,-64(1)
|
|
|
|
|
cfi_adjust_cfa_offset(64)
|
|
|
|
|
stw r31,48(1)
|
|
|
|
|
cfi_offset(31,(48-64))
|
|
|
|
|
stw r30,44(1)
|
|
|
|
|
cfi_offset(30,(44-64))
|
|
|
|
|
bne L(unaligned)
|
|
|
|
|
/* At this point we know both strings have the same alignment and the
|
|
|
|
|
compare length is at least 8 bytes. rBITDIF contains the low order
|
|
|
|
|
2 bits of rSTR1 and cr5 contains the result of the logical compare
|
|
|
|
|
of rBITDIF to 0. If rBITDIF == 0 then we are already word
|
|
|
|
|
aligned and can perform the word aligned loop.
|
|
|
|
|
|
|
|
|
|
Otherwise we know the two strings have the same alignment (but not
|
|
|
|
|
yet word aligned). So we force the string addresses to the next lower
|
|
|
|
|
word boundary and special case this first word using shift left to
|
2013-01-07 17:20:53 +00:00
|
|
|
eliminate bits preceding the first byte. Since we want to join the
|
2007-06-03 21:20:06 +00:00
|
|
|
normal (word aligned) compare loop, starting at the second word,
|
|
|
|
|
we need to adjust the length (rN) and special case the loop
|
|
|
|
|
versioning for the first word. This insures that the loop count is
|
|
|
|
|
correct and the first word (shifted) is in the expected register pair. */
|
|
|
|
|
.align 4
|
|
|
|
|
L(samealignment):
|
|
|
|
|
clrrwi rSTR1, rSTR1, 2
|
|
|
|
|
clrrwi rSTR2, rSTR2, 2
|
|
|
|
|
beq cr5, L(Waligned)
|
|
|
|
|
add rN, rN, rBITDIF
|
|
|
|
|
slwi r11, rBITDIF, 3
|
|
|
|
|
srwi rTMP, rN, 4 /* Divide by 16 */
|
|
|
|
|
andi. rBITDIF, rN, 12 /* Get the word remainder */
|
|
|
|
|
lwz rWORD1, 0(rSTR1)
|
|
|
|
|
lwz rWORD2, 0(rSTR2)
|
|
|
|
|
cmplwi cr1, rBITDIF, 8
|
|
|
|
|
cmplwi cr7, rN, 16
|
|
|
|
|
clrlwi rN, rN, 30
|
|
|
|
|
beq L(dPs4)
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
bgt cr1, L(dPs3)
|
|
|
|
|
beq cr1, L(dPs2)
|
|
|
|
|
|
|
|
|
|
/* Remainder is 4 */
|
|
|
|
|
.align 3
|
|
|
|
|
L(dsP1):
|
|
|
|
|
slw rWORD5, rWORD1, r11
|
|
|
|
|
slw rWORD6, rWORD2, r11
|
|
|
|
|
cmplw cr5, rWORD5, rWORD6
|
|
|
|
|
blt cr7, L(dP1x)
|
|
|
|
|
/* Do something useful in this cycle since we have to branch anyway. */
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
b L(dP1e)
|
|
|
|
|
/* Remainder is 8 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dPs2):
|
|
|
|
|
slw rWORD5, rWORD1, r11
|
|
|
|
|
slw rWORD6, rWORD2, r11
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
blt cr7, L(dP2x)
|
|
|
|
|
/* Do something useful in this cycle since we have to branch anyway. */
|
|
|
|
|
lwz rWORD7, 4(rSTR1)
|
|
|
|
|
lwz rWORD8, 4(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
b L(dP2e)
|
|
|
|
|
/* Remainder is 12 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dPs3):
|
|
|
|
|
slw rWORD3, rWORD1, r11
|
|
|
|
|
slw rWORD4, rWORD2, r11
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
b L(dP3e)
|
|
|
|
|
/* Count is a multiple of 16, remainder is 0 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dPs4):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
slw rWORD1, rWORD1, r11
|
|
|
|
|
slw rWORD2, rWORD2, r11
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
b L(dP4e)
|
|
|
|
|
|
|
|
|
|
/* At this point we know both strings are word aligned and the
|
|
|
|
|
compare length is at least 8 bytes. */
|
|
|
|
|
.align 4
|
|
|
|
|
L(Waligned):
|
|
|
|
|
andi. rBITDIF, rN, 12 /* Get the word remainder */
|
|
|
|
|
srwi rTMP, rN, 4 /* Divide by 16 */
|
|
|
|
|
cmplwi cr1, rBITDIF, 8
|
|
|
|
|
cmplwi cr7, rN, 16
|
|
|
|
|
clrlwi rN, rN, 30
|
|
|
|
|
beq L(dP4)
|
|
|
|
|
bgt cr1, L(dP3)
|
|
|
|
|
beq cr1, L(dP2)
|
|
|
|
|
|
|
|
|
|
/* Remainder is 4 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dP1):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
/* Normally we'd use rWORD7/rWORD8 here, but since we might exit early
|
|
|
|
|
(8-15 byte compare), we want to use only volatile registers. This
|
|
|
|
|
means we can avoid restoring non-volatile registers since we did not
|
|
|
|
|
change any on the early exit path. The key here is the non-early
|
|
|
|
|
exit path only cares about the condition code (cr5), not about which
|
|
|
|
|
register pair was used. */
|
|
|
|
|
lwz rWORD5, 0(rSTR1)
|
|
|
|
|
lwz rWORD6, 0(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD5, rWORD6
|
|
|
|
|
blt cr7, L(dP1x)
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
L(dP1e):
|
|
|
|
|
lwz rWORD3, 8(rSTR1)
|
|
|
|
|
lwz rWORD4, 8(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
lwz rWORD5, 12(rSTR1)
|
|
|
|
|
lwz rWORD6, 12(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
bne cr0, L(dLcr0)
|
|
|
|
|
|
|
|
|
|
lwzu rWORD7, 16(rSTR1)
|
|
|
|
|
lwzu rWORD8, 16(rSTR2)
|
|
|
|
|
bne cr1, L(dLcr1)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bdnz L(dLoop)
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
.align 3
|
|
|
|
|
L(dP1x):
|
|
|
|
|
slwi. r12, rN, 3
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
subfic rN, r12, 32 /* Shift count is 32 - (rN * 8). */
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bne L(d00)
|
|
|
|
|
li rRTN, 0
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/* Remainder is 8 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dP2):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
lwz rWORD5, 0(rSTR1)
|
|
|
|
|
lwz rWORD6, 0(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
blt cr7, L(dP2x)
|
|
|
|
|
lwz rWORD7, 4(rSTR1)
|
|
|
|
|
lwz rWORD8, 4(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
L(dP2e):
|
|
|
|
|
lwz rWORD1, 8(rSTR1)
|
|
|
|
|
lwz rWORD2, 8(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
lwz rWORD3, 12(rSTR1)
|
|
|
|
|
lwz rWORD4, 12(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
addi rSTR1, rSTR1, 4
|
|
|
|
|
addi rSTR2, rSTR2, 4
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
b L(dLoop2)
|
|
|
|
|
/* Again we are on a early exit path (16-23 byte compare), we want to
|
|
|
|
|
only use volatile registers and avoid restoring non-volatile
|
|
|
|
|
registers. */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dP2x):
|
|
|
|
|
lwz rWORD3, 4(rSTR1)
|
|
|
|
|
lwz rWORD4, 4(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD3, rWORD4
|
|
|
|
|
slwi. r12, rN, 3
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
addi rSTR1, rSTR1, 4
|
|
|
|
|
addi rSTR2, rSTR2, 4
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
subfic rN, r12, 32 /* Shift count is 32 - (rN * 8). */
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bne L(d00)
|
|
|
|
|
li rRTN, 0
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/* Remainder is 12 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dP3):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
lwz rWORD3, 0(rSTR1)
|
|
|
|
|
lwz rWORD4, 0(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
L(dP3e):
|
|
|
|
|
lwz rWORD5, 4(rSTR1)
|
|
|
|
|
lwz rWORD6, 4(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
blt cr7, L(dP3x)
|
|
|
|
|
lwz rWORD7, 8(rSTR1)
|
|
|
|
|
lwz rWORD8, 8(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
lwz rWORD1, 12(rSTR1)
|
|
|
|
|
lwz rWORD2, 12(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
addi rSTR1, rSTR1, 8
|
|
|
|
|
addi rSTR2, rSTR2, 8
|
|
|
|
|
bne cr1, L(dLcr1)
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
b L(dLoop1)
|
|
|
|
|
/* Again we are on a early exit path (24-31 byte compare), we want to
|
|
|
|
|
only use volatile registers and avoid restoring non-volatile
|
|
|
|
|
registers. */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dP3x):
|
|
|
|
|
lwz rWORD1, 8(rSTR1)
|
|
|
|
|
lwz rWORD2, 8(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD1, rWORD2
|
|
|
|
|
slwi. r12, rN, 3
|
|
|
|
|
bne cr1, L(dLcr1)
|
|
|
|
|
addi rSTR1, rSTR1, 8
|
|
|
|
|
addi rSTR2, rSTR2, 8
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
subfic rN, r12, 32 /* Shift count is 32 - (rN * 8). */
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bne L(d00)
|
|
|
|
|
li rRTN, 0
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/* Count is a multiple of 16, remainder is 0 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dP4):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
lwz rWORD1, 0(rSTR1)
|
|
|
|
|
lwz rWORD2, 0(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
L(dP4e):
|
|
|
|
|
lwz rWORD3, 4(rSTR1)
|
|
|
|
|
lwz rWORD4, 4(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
lwz rWORD5, 8(rSTR1)
|
|
|
|
|
lwz rWORD6, 8(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
lwzu rWORD7, 12(rSTR1)
|
|
|
|
|
lwzu rWORD8, 12(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bne cr0, L(dLcr0)
|
|
|
|
|
bne cr1, L(dLcr1)
|
|
|
|
|
bdz- L(d24) /* Adjust CTR as we start with +4 */
|
|
|
|
|
/* This is the primary loop */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dLoop):
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
L(dLoop1):
|
|
|
|
|
lwz rWORD3, 8(rSTR1)
|
|
|
|
|
lwz rWORD4, 8(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
L(dLoop2):
|
|
|
|
|
lwz rWORD5, 12(rSTR1)
|
|
|
|
|
lwz rWORD6, 12(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bne cr0, L(dLcr0)
|
|
|
|
|
L(dLoop3):
|
|
|
|
|
lwzu rWORD7, 16(rSTR1)
|
|
|
|
|
lwzu rWORD8, 16(rSTR2)
|
|
|
|
|
bne- cr1, L(dLcr1)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
bdnz+ L(dLoop)
|
|
|
|
|
|
|
|
|
|
L(dL4):
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
L(d44):
|
|
|
|
|
bne cr0, L(dLcr0)
|
|
|
|
|
L(d34):
|
|
|
|
|
bne cr1, L(dLcr1)
|
|
|
|
|
L(d24):
|
|
|
|
|
bne cr6, L(dLcr6)
|
|
|
|
|
L(d14):
|
|
|
|
|
slwi. r12, rN, 3
|
|
|
|
|
bne cr5, L(dLcr5)
|
|
|
|
|
L(d04):
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
subfic rN, r12, 32 /* Shift count is 32 - (rN * 8). */
|
|
|
|
|
beq L(zeroLength)
|
|
|
|
|
/* At this point we have a remainder of 1 to 3 bytes to compare. Since
|
|
|
|
|
we are aligned it is safe to load the whole word, and use
|
|
|
|
|
shift right to eliminate bits beyond the compare length. */
|
|
|
|
|
L(d00):
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
srw rWORD1, rWORD1, rN
|
|
|
|
|
srw rWORD2, rWORD2, rN
|
|
|
|
|
cmplw rWORD1,rWORD2
|
|
|
|
|
li rRTN,0
|
|
|
|
|
beqlr
|
|
|
|
|
li rRTN,1
|
|
|
|
|
bgtlr
|
|
|
|
|
li rRTN,-1
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
|
L(dLcr0):
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bgtlr cr0
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
.align 4
|
|
|
|
|
L(dLcr1):
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bgtlr cr1
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
.align 4
|
|
|
|
|
L(dLcr6):
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bgtlr cr6
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
.align 4
|
|
|
|
|
L(dLcr5):
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
L(dLcr5x):
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
bgtlr cr5
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
|
L(bytealigned):
|
|
|
|
|
cfi_adjust_cfa_offset(-64)
|
|
|
|
|
mtctr rN /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
|
|
|
|
|
/* We need to prime this loop. This loop is swing modulo scheduled
|
|
|
|
|
to avoid pipe delays. The dependent instruction latencies (load to
|
|
|
|
|
compare to conditional branch) is 2 to 3 cycles. In this loop each
|
|
|
|
|
dispatch group ends in a branch and takes 1 cycle. Effectively
|
|
|
|
|
the first iteration of the loop only serves to load operands and
|
|
|
|
|
branches based on compares are delayed until the next loop.
|
|
|
|
|
|
|
|
|
|
So we must precondition some registers and condition codes so that
|
|
|
|
|
we don't exit the loop early on the first iteration. */
|
|
|
|
|
|
|
|
|
|
lbz rWORD1, 0(rSTR1)
|
|
|
|
|
lbz rWORD2, 0(rSTR2)
|
|
|
|
|
bdz- L(b11)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
lbz rWORD3, 1(rSTR1)
|
|
|
|
|
lbz rWORD4, 1(rSTR2)
|
|
|
|
|
bdz- L(b12)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
lbzu rWORD5, 2(rSTR1)
|
|
|
|
|
lbzu rWORD6, 2(rSTR2)
|
|
|
|
|
bdz- L(b13)
|
|
|
|
|
.align 4
|
|
|
|
|
L(bLoop):
|
|
|
|
|
lbzu rWORD1, 1(rSTR1)
|
|
|
|
|
lbzu rWORD2, 1(rSTR2)
|
|
|
|
|
bne- cr0, L(bLcr0)
|
|
|
|
|
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bdz- L(b3i)
|
|
|
|
|
|
|
|
|
|
lbzu rWORD3, 1(rSTR1)
|
|
|
|
|
lbzu rWORD4, 1(rSTR2)
|
|
|
|
|
bne- cr1, L(bLcr1)
|
|
|
|
|
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
bdz- L(b2i)
|
|
|
|
|
|
|
|
|
|
lbzu rWORD5, 1(rSTR1)
|
|
|
|
|
lbzu rWORD6, 1(rSTR2)
|
|
|
|
|
bne- cr6, L(bLcr6)
|
|
|
|
|
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
bdnz+ L(bLoop)
|
|
|
|
|
|
|
|
|
|
/* We speculatively loading bytes before we have tested the previous
|
|
|
|
|
bytes. But we must avoid overrunning the length (in the ctr) to
|
|
|
|
|
prevent these speculative loads from causing a segfault. In this
|
|
|
|
|
case the loop will exit early (before the all pending bytes are
|
|
|
|
|
tested. In this case we must complete the pending operations
|
|
|
|
|
before returning. */
|
|
|
|
|
L(b1i):
|
|
|
|
|
bne- cr0, L(bLcr0)
|
|
|
|
|
bne- cr1, L(bLcr1)
|
|
|
|
|
b L(bx56)
|
|
|
|
|
.align 4
|
|
|
|
|
L(b2i):
|
|
|
|
|
bne- cr6, L(bLcr6)
|
|
|
|
|
bne- cr0, L(bLcr0)
|
|
|
|
|
b L(bx34)
|
|
|
|
|
.align 4
|
|
|
|
|
L(b3i):
|
|
|
|
|
bne- cr1, L(bLcr1)
|
|
|
|
|
bne- cr6, L(bLcr6)
|
|
|
|
|
b L(bx12)
|
|
|
|
|
.align 4
|
|
|
|
|
L(bLcr0):
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgtlr cr0
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
L(bLcr1):
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgtlr cr1
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
L(bLcr6):
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgtlr cr6
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
L(b13):
|
|
|
|
|
bne- cr0, L(bx12)
|
|
|
|
|
bne- cr1, L(bx34)
|
|
|
|
|
L(bx56):
|
|
|
|
|
sub rRTN, rWORD5, rWORD6
|
|
|
|
|
blr
|
|
|
|
|
nop
|
|
|
|
|
L(b12):
|
|
|
|
|
bne- cr0, L(bx12)
|
|
|
|
|
L(bx34):
|
|
|
|
|
sub rRTN, rWORD3, rWORD4
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
L(b11):
|
|
|
|
|
L(bx12):
|
|
|
|
|
sub rRTN, rWORD1, rWORD2
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
.align 4
|
|
|
|
|
L(zeroLengthReturn):
|
|
|
|
|
|
|
|
|
|
L(zeroLength):
|
|
|
|
|
li rRTN, 0
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
cfi_adjust_cfa_offset(64)
|
|
|
|
|
.align 4
|
|
|
|
|
/* At this point we know the strings have different alignment and the
|
|
|
|
|
compare length is at least 8 bytes. rBITDIF contains the low order
|
|
|
|
|
2 bits of rSTR1 and cr5 contains the result of the logical compare
|
|
|
|
|
of rBITDIF to 0. If rBITDIF == 0 then rStr1 is word aligned and can
|
|
|
|
|
perform the Wunaligned loop.
|
|
|
|
|
|
|
|
|
|
Otherwise we know that rSTR1 is not aready word aligned yet.
|
|
|
|
|
So we can force the string addresses to the next lower word
|
|
|
|
|
boundary and special case this first word using shift left to
|
2013-01-07 17:20:53 +00:00
|
|
|
eliminate bits preceding the first byte. Since we want to join the
|
2007-06-03 21:20:06 +00:00
|
|
|
normal (Wualigned) compare loop, starting at the second word,
|
|
|
|
|
we need to adjust the length (rN) and special case the loop
|
|
|
|
|
versioning for the first W. This insures that the loop count is
|
|
|
|
|
correct and the first W (shifted) is in the expected resister pair. */
|
|
|
|
|
#define rSHL r29 /* Unaligned shift left count. */
|
|
|
|
|
#define rSHR r28 /* Unaligned shift right count. */
|
|
|
|
|
#define rB r27 /* Left rotation temp for rWORD2. */
|
|
|
|
|
#define rD r26 /* Left rotation temp for rWORD4. */
|
|
|
|
|
#define rF r25 /* Left rotation temp for rWORD6. */
|
|
|
|
|
#define rH r24 /* Left rotation temp for rWORD8. */
|
|
|
|
|
#define rA r0 /* Right rotation temp for rWORD2. */
|
|
|
|
|
#define rC r12 /* Right rotation temp for rWORD4. */
|
|
|
|
|
#define rE r0 /* Right rotation temp for rWORD6. */
|
|
|
|
|
#define rG r12 /* Right rotation temp for rWORD8. */
|
|
|
|
|
L(unaligned):
|
|
|
|
|
stw r29,40(r1)
|
|
|
|
|
cfi_offset(r29,(40-64))
|
|
|
|
|
clrlwi rSHL, rSTR2, 30
|
|
|
|
|
stw r28,36(r1)
|
|
|
|
|
cfi_offset(r28,(36-64))
|
|
|
|
|
beq cr5, L(Wunaligned)
|
|
|
|
|
stw r27,32(r1)
|
|
|
|
|
cfi_offset(r27,(32-64))
|
|
|
|
|
/* Adjust the logical start of rSTR2 to compensate for the extra bits
|
|
|
|
|
in the 1st rSTR1 W. */
|
|
|
|
|
sub r27, rSTR2, rBITDIF
|
|
|
|
|
/* But do not attempt to address the W before that W that contains
|
|
|
|
|
the actual start of rSTR2. */
|
|
|
|
|
clrrwi rSTR2, rSTR2, 2
|
|
|
|
|
stw r26,28(r1)
|
|
|
|
|
cfi_offset(r26,(28-64))
|
|
|
|
|
/* Compute the left/right shift counts for the unalign rSTR2,
|
|
|
|
|
compensating for the logical (W aligned) start of rSTR1. */
|
|
|
|
|
clrlwi rSHL, r27, 30
|
|
|
|
|
clrrwi rSTR1, rSTR1, 2
|
|
|
|
|
stw r25,24(r1)
|
|
|
|
|
cfi_offset(r25,(24-64))
|
|
|
|
|
slwi rSHL, rSHL, 3
|
|
|
|
|
cmplw cr5, r27, rSTR2
|
|
|
|
|
add rN, rN, rBITDIF
|
|
|
|
|
slwi r11, rBITDIF, 3
|
|
|
|
|
stw r24,20(r1)
|
|
|
|
|
cfi_offset(r24,(20-64))
|
|
|
|
|
subfic rSHR, rSHL, 32
|
|
|
|
|
srwi rTMP, rN, 4 /* Divide by 16 */
|
|
|
|
|
andi. rBITDIF, rN, 12 /* Get the W remainder */
|
|
|
|
|
/* We normally need to load 2 Ws to start the unaligned rSTR2, but in
|
|
|
|
|
this special case those bits may be discarded anyway. Also we
|
|
|
|
|
must avoid loading a W where none of the bits are part of rSTR2 as
|
|
|
|
|
this may cross a page boundary and cause a page fault. */
|
|
|
|
|
li rWORD8, 0
|
|
|
|
|
blt cr5, L(dus0)
|
|
|
|
|
lwz rWORD8, 0(rSTR2)
|
|
|
|
|
la rSTR2, 4(rSTR2)
|
|
|
|
|
slw rWORD8, rWORD8, rSHL
|
|
|
|
|
|
|
|
|
|
L(dus0):
|
|
|
|
|
lwz rWORD1, 0(rSTR1)
|
|
|
|
|
lwz rWORD2, 0(rSTR2)
|
|
|
|
|
cmplwi cr1, rBITDIF, 8
|
|
|
|
|
cmplwi cr7, rN, 16
|
|
|
|
|
srw rG, rWORD2, rSHR
|
|
|
|
|
clrlwi rN, rN, 30
|
|
|
|
|
beq L(duPs4)
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
or rWORD8, rG, rWORD8
|
|
|
|
|
bgt cr1, L(duPs3)
|
|
|
|
|
beq cr1, L(duPs2)
|
|
|
|
|
|
|
|
|
|
/* Remainder is 4 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(dusP1):
|
|
|
|
|
slw rB, rWORD2, rSHL
|
|
|
|
|
slw rWORD7, rWORD1, r11
|
|
|
|
|
slw rWORD8, rWORD8, r11
|
|
|
|
|
bge cr7, L(duP1e)
|
|
|
|
|
/* At this point we exit early with the first word compare
|
|
|
|
|
complete and remainder of 0 to 3 bytes. See L(du14) for details on
|
|
|
|
|
how we handle the remaining bytes. */
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
slwi. rN, rN, 3
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
cmplw cr7, rN, rSHR
|
|
|
|
|
beq L(duZeroReturn)
|
|
|
|
|
li rA, 0
|
|
|
|
|
ble cr7, L(dutrim)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
b L(dutrim)
|
|
|
|
|
/* Remainder is 8 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duPs2):
|
|
|
|
|
slw rH, rWORD2, rSHL
|
|
|
|
|
slw rWORD5, rWORD1, r11
|
|
|
|
|
slw rWORD6, rWORD8, r11
|
|
|
|
|
b L(duP2e)
|
|
|
|
|
/* Remainder is 12 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duPs3):
|
|
|
|
|
slw rF, rWORD2, rSHL
|
|
|
|
|
slw rWORD3, rWORD1, r11
|
|
|
|
|
slw rWORD4, rWORD8, r11
|
|
|
|
|
b L(duP3e)
|
|
|
|
|
/* Count is a multiple of 16, remainder is 0 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duPs4):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
or rWORD8, rG, rWORD8
|
|
|
|
|
slw rD, rWORD2, rSHL
|
|
|
|
|
slw rWORD1, rWORD1, r11
|
|
|
|
|
slw rWORD2, rWORD8, r11
|
|
|
|
|
b L(duP4e)
|
|
|
|
|
|
|
|
|
|
/* At this point we know rSTR1 is word aligned and the
|
|
|
|
|
compare length is at least 8 bytes. */
|
|
|
|
|
.align 4
|
|
|
|
|
L(Wunaligned):
|
|
|
|
|
stw r27,32(r1)
|
|
|
|
|
cfi_offset(r27,(32-64))
|
|
|
|
|
clrrwi rSTR2, rSTR2, 2
|
|
|
|
|
stw r26,28(r1)
|
|
|
|
|
cfi_offset(r26,(28-64))
|
|
|
|
|
srwi rTMP, rN, 4 /* Divide by 16 */
|
|
|
|
|
stw r25,24(r1)
|
|
|
|
|
cfi_offset(r25,(24-64))
|
|
|
|
|
andi. rBITDIF, rN, 12 /* Get the W remainder */
|
|
|
|
|
stw r24,20(r1)
|
2010-08-12 16:19:19 +00:00
|
|
|
cfi_offset(r24,(20-64))
|
2007-06-03 21:20:06 +00:00
|
|
|
slwi rSHL, rSHL, 3
|
|
|
|
|
lwz rWORD6, 0(rSTR2)
|
|
|
|
|
lwzu rWORD8, 4(rSTR2)
|
|
|
|
|
cmplwi cr1, rBITDIF, 8
|
|
|
|
|
cmplwi cr7, rN, 16
|
|
|
|
|
clrlwi rN, rN, 30
|
|
|
|
|
subfic rSHR, rSHL, 32
|
|
|
|
|
slw rH, rWORD6, rSHL
|
|
|
|
|
beq L(duP4)
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
bgt cr1, L(duP3)
|
|
|
|
|
beq cr1, L(duP2)
|
|
|
|
|
|
|
|
|
|
/* Remainder is 4 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duP1):
|
|
|
|
|
srw rG, rWORD8, rSHR
|
|
|
|
|
lwz rWORD7, 0(rSTR1)
|
|
|
|
|
slw rB, rWORD8, rSHL
|
|
|
|
|
or rWORD8, rG, rH
|
|
|
|
|
blt cr7, L(duP1x)
|
|
|
|
|
L(duP1e):
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
slw rD, rWORD2, rSHL
|
|
|
|
|
or rWORD2, rA, rB
|
|
|
|
|
lwz rWORD3, 8(rSTR1)
|
|
|
|
|
lwz rWORD4, 8(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
srw rC, rWORD4, rSHR
|
|
|
|
|
slw rF, rWORD4, rSHL
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
or rWORD4, rC, rD
|
|
|
|
|
lwz rWORD5, 12(rSTR1)
|
|
|
|
|
lwz rWORD6, 12(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
srw rE, rWORD6, rSHR
|
|
|
|
|
slw rH, rWORD6, rSHL
|
|
|
|
|
bne cr0, L(duLcr0)
|
|
|
|
|
or rWORD6, rE, rF
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
b L(duLoop3)
|
|
|
|
|
.align 4
|
|
|
|
|
/* At this point we exit early with the first word compare
|
|
|
|
|
complete and remainder of 0 to 3 bytes. See L(du14) for details on
|
|
|
|
|
how we handle the remaining bytes. */
|
|
|
|
|
L(duP1x):
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
slwi. rN, rN, 3
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
cmplw cr7, rN, rSHR
|
|
|
|
|
beq L(duZeroReturn)
|
|
|
|
|
li rA, 0
|
|
|
|
|
ble cr7, L(dutrim)
|
|
|
|
|
ld rWORD2, 8(rSTR2)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
b L(dutrim)
|
|
|
|
|
/* Remainder is 8 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duP2):
|
|
|
|
|
srw rE, rWORD8, rSHR
|
|
|
|
|
lwz rWORD5, 0(rSTR1)
|
|
|
|
|
or rWORD6, rE, rH
|
|
|
|
|
slw rH, rWORD8, rSHL
|
|
|
|
|
L(duP2e):
|
|
|
|
|
lwz rWORD7, 4(rSTR1)
|
|
|
|
|
lwz rWORD8, 4(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
srw rG, rWORD8, rSHR
|
|
|
|
|
slw rB, rWORD8, rSHL
|
|
|
|
|
or rWORD8, rG, rH
|
|
|
|
|
blt cr7, L(duP2x)
|
|
|
|
|
lwz rWORD1, 8(rSTR1)
|
|
|
|
|
lwz rWORD2, 8(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
slw rD, rWORD2, rSHL
|
|
|
|
|
or rWORD2, rA, rB
|
|
|
|
|
lwz rWORD3, 12(rSTR1)
|
|
|
|
|
lwz rWORD4, 12(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
srw rC, rWORD4, rSHR
|
|
|
|
|
slw rF, rWORD4, rSHL
|
|
|
|
|
or rWORD4, rC, rD
|
|
|
|
|
addi rSTR1, rSTR1, 4
|
|
|
|
|
addi rSTR2, rSTR2, 4
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
b L(duLoop2)
|
|
|
|
|
.align 4
|
|
|
|
|
L(duP2x):
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
addi rSTR1, rSTR1, 4
|
|
|
|
|
addi rSTR2, rSTR2, 4
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
slwi. rN, rN, 3
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
cmplw cr7, rN, rSHR
|
|
|
|
|
beq L(duZeroReturn)
|
|
|
|
|
li rA, 0
|
|
|
|
|
ble cr7, L(dutrim)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
b L(dutrim)
|
|
|
|
|
|
|
|
|
|
/* Remainder is 12 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duP3):
|
|
|
|
|
srw rC, rWORD8, rSHR
|
|
|
|
|
lwz rWORD3, 0(rSTR1)
|
|
|
|
|
slw rF, rWORD8, rSHL
|
|
|
|
|
or rWORD4, rC, rH
|
|
|
|
|
L(duP3e):
|
|
|
|
|
lwz rWORD5, 4(rSTR1)
|
|
|
|
|
lwz rWORD6, 4(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
srw rE, rWORD6, rSHR
|
|
|
|
|
slw rH, rWORD6, rSHL
|
|
|
|
|
or rWORD6, rE, rF
|
|
|
|
|
lwz rWORD7, 8(rSTR1)
|
|
|
|
|
lwz rWORD8, 8(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr1, L(duLcr1)
|
|
|
|
|
srw rG, rWORD8, rSHR
|
|
|
|
|
slw rB, rWORD8, rSHL
|
|
|
|
|
or rWORD8, rG, rH
|
|
|
|
|
blt cr7, L(duP3x)
|
|
|
|
|
lwz rWORD1, 12(rSTR1)
|
|
|
|
|
lwz rWORD2, 12(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
slw rD, rWORD2, rSHL
|
|
|
|
|
or rWORD2, rA, rB
|
|
|
|
|
addi rSTR1, rSTR1, 8
|
|
|
|
|
addi rSTR2, rSTR2, 8
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
b L(duLoop1)
|
|
|
|
|
.align 4
|
|
|
|
|
L(duP3x):
|
|
|
|
|
addi rSTR1, rSTR1, 8
|
|
|
|
|
addi rSTR2, rSTR2, 8
|
|
|
|
|
bne cr1, L(duLcr1)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
slwi. rN, rN, 3
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
cmplw cr7, rN, rSHR
|
|
|
|
|
beq L(duZeroReturn)
|
|
|
|
|
li rA, 0
|
|
|
|
|
ble cr7, L(dutrim)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
b L(dutrim)
|
|
|
|
|
|
|
|
|
|
/* Count is a multiple of 16, remainder is 0 */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duP4):
|
|
|
|
|
mtctr rTMP /* Power4 wants mtctr 1st in dispatch group */
|
|
|
|
|
srw rA, rWORD8, rSHR
|
|
|
|
|
lwz rWORD1, 0(rSTR1)
|
|
|
|
|
slw rD, rWORD8, rSHL
|
|
|
|
|
or rWORD2, rA, rH
|
|
|
|
|
L(duP4e):
|
|
|
|
|
lwz rWORD3, 4(rSTR1)
|
|
|
|
|
lwz rWORD4, 4(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
srw rC, rWORD4, rSHR
|
|
|
|
|
slw rF, rWORD4, rSHL
|
|
|
|
|
or rWORD4, rC, rD
|
|
|
|
|
lwz rWORD5, 8(rSTR1)
|
|
|
|
|
lwz rWORD6, 8(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
bne cr0, L(duLcr0)
|
|
|
|
|
srw rE, rWORD6, rSHR
|
|
|
|
|
slw rH, rWORD6, rSHL
|
|
|
|
|
or rWORD6, rE, rF
|
|
|
|
|
lwzu rWORD7, 12(rSTR1)
|
|
|
|
|
lwzu rWORD8, 12(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr1, L(duLcr1)
|
|
|
|
|
srw rG, rWORD8, rSHR
|
|
|
|
|
slw rB, rWORD8, rSHL
|
|
|
|
|
or rWORD8, rG, rH
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bdz- L(du24) /* Adjust CTR as we start with +4 */
|
|
|
|
|
/* This is the primary loop */
|
|
|
|
|
.align 4
|
|
|
|
|
L(duLoop):
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
slw rD, rWORD2, rSHL
|
|
|
|
|
or rWORD2, rA, rB
|
|
|
|
|
L(duLoop1):
|
|
|
|
|
lwz rWORD3, 8(rSTR1)
|
|
|
|
|
lwz rWORD4, 8(rSTR2)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
srw rC, rWORD4, rSHR
|
|
|
|
|
slw rF, rWORD4, rSHL
|
|
|
|
|
or rWORD4, rC, rD
|
|
|
|
|
L(duLoop2):
|
|
|
|
|
lwz rWORD5, 12(rSTR1)
|
|
|
|
|
lwz rWORD6, 12(rSTR2)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
bne cr0, L(duLcr0)
|
|
|
|
|
srw rE, rWORD6, rSHR
|
|
|
|
|
slw rH, rWORD6, rSHL
|
|
|
|
|
or rWORD6, rE, rF
|
|
|
|
|
L(duLoop3):
|
|
|
|
|
lwzu rWORD7, 16(rSTR1)
|
|
|
|
|
lwzu rWORD8, 16(rSTR2)
|
|
|
|
|
cmplw cr0, rWORD1, rWORD2
|
|
|
|
|
bne- cr1, L(duLcr1)
|
|
|
|
|
srw rG, rWORD8, rSHR
|
|
|
|
|
slw rB, rWORD8, rSHL
|
|
|
|
|
or rWORD8, rG, rH
|
|
|
|
|
bdnz+ L(duLoop)
|
|
|
|
|
|
|
|
|
|
L(duL4):
|
|
|
|
|
bne cr1, L(duLcr1)
|
|
|
|
|
cmplw cr1, rWORD3, rWORD4
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
cmplw cr6, rWORD5, rWORD6
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
cmplw cr5, rWORD7, rWORD8
|
|
|
|
|
L(du44):
|
|
|
|
|
bne cr0, L(duLcr0)
|
|
|
|
|
L(du34):
|
|
|
|
|
bne cr1, L(duLcr1)
|
|
|
|
|
L(du24):
|
|
|
|
|
bne cr6, L(duLcr6)
|
|
|
|
|
L(du14):
|
|
|
|
|
slwi. rN, rN, 3
|
|
|
|
|
bne cr5, L(duLcr5)
|
|
|
|
|
/* At this point we have a remainder of 1 to 3 bytes to compare. We use
|
|
|
|
|
shift right to eliminate bits beyond the compare length.
|
|
|
|
|
|
|
|
|
|
However it may not be safe to load rWORD2 which may be beyond the
|
|
|
|
|
string length. So we compare the bit length of the remainder to
|
|
|
|
|
the right shift count (rSHR). If the bit count is less than or equal
|
|
|
|
|
we do not need to load rWORD2 (all significant bits are already in
|
|
|
|
|
rB). */
|
|
|
|
|
cmplw cr7, rN, rSHR
|
|
|
|
|
beq L(duZeroReturn)
|
|
|
|
|
li rA, 0
|
|
|
|
|
ble cr7, L(dutrim)
|
|
|
|
|
lwz rWORD2, 4(rSTR2)
|
|
|
|
|
srw rA, rWORD2, rSHR
|
|
|
|
|
.align 4
|
|
|
|
|
L(dutrim):
|
|
|
|
|
lwz rWORD1, 4(rSTR1)
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
subfic rN, rN, 32 /* Shift count is 32 - (rN * 8). */
|
|
|
|
|
or rWORD2, rA, rB
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
lwz r29,40(r1)
|
|
|
|
|
srw rWORD1, rWORD1, rN
|
|
|
|
|
srw rWORD2, rWORD2, rN
|
|
|
|
|
lwz r28,36(r1)
|
|
|
|
|
lwz r27,32(r1)
|
|
|
|
|
cmplw rWORD1,rWORD2
|
|
|
|
|
li rRTN,0
|
|
|
|
|
beq L(dureturn26)
|
|
|
|
|
li rRTN,1
|
|
|
|
|
bgt L(dureturn26)
|
|
|
|
|
li rRTN,-1
|
|
|
|
|
b L(dureturn26)
|
|
|
|
|
.align 4
|
|
|
|
|
L(duLcr0):
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgt cr0, L(dureturn29)
|
|
|
|
|
lwz r29,40(r1)
|
|
|
|
|
lwz r28,36(r1)
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
b L(dureturn27)
|
|
|
|
|
.align 4
|
|
|
|
|
L(duLcr1):
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgt cr1, L(dureturn29)
|
|
|
|
|
lwz r29,40(r1)
|
|
|
|
|
lwz r28,36(r1)
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
b L(dureturn27)
|
|
|
|
|
.align 4
|
|
|
|
|
L(duLcr6):
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgt cr6, L(dureturn29)
|
|
|
|
|
lwz r29,40(r1)
|
|
|
|
|
lwz r28,36(r1)
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
b L(dureturn27)
|
|
|
|
|
.align 4
|
|
|
|
|
L(duLcr5):
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
li rRTN, 1
|
|
|
|
|
bgt cr5, L(dureturn29)
|
|
|
|
|
lwz r29,40(r1)
|
|
|
|
|
lwz r28,36(r1)
|
|
|
|
|
li rRTN, -1
|
|
|
|
|
b L(dureturn27)
|
|
|
|
|
.align 3
|
|
|
|
|
L(duZeroReturn):
|
|
|
|
|
li rRTN,0
|
|
|
|
|
.align 4
|
|
|
|
|
L(dureturn):
|
|
|
|
|
lwz r31,48(1)
|
|
|
|
|
lwz r30,44(1)
|
|
|
|
|
L(dureturn29):
|
|
|
|
|
lwz r29,40(r1)
|
|
|
|
|
lwz r28,36(r1)
|
|
|
|
|
L(dureturn27):
|
|
|
|
|
lwz r27,32(r1)
|
|
|
|
|
L(dureturn26):
|
|
|
|
|
lwz r26,28(r1)
|
|
|
|
|
L(dureturn25):
|
|
|
|
|
lwz r25,24(r1)
|
|
|
|
|
lwz r24,20(r1)
|
|
|
|
|
lwz 1,0(1)
|
|
|
|
|
blr
|
|
|
|
|
END (BP_SYM (memcmp))
|
|
|
|
|
|
|
|
|
|
libc_hidden_builtin_def (memcmp)
|
|
|
|
|
weak_alias (memcmp, bcmp)
|
