mirror of
https://sourceware.org/git/glibc.git
synced 2024-12-04 10:50:07 +00:00
e23d3d2690
Optimization is achieved on 8 byte aligned strings with double word comparison using cmpb instruction. On unaligned strings loop unrolling is applied for Power7 gain.
196 lines
6.4 KiB
ArmAsm
196 lines
6.4 KiB
ArmAsm
/* Optimized strcmp implementation for Power7 using 'cmpb' instruction
|
|
Copyright (C) 2014 Free Software Foundation, Inc.
|
|
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
|
|
License along with the GNU C Library; if not, see
|
|
<http://www.gnu.org/licenses/>. */
|
|
|
|
/* The optimization is achieved here through cmpb instruction.
|
|
8byte aligned strings are processed with double word comparision
|
|
and unaligned strings are handled effectively with loop unrolling
|
|
technique */
|
|
|
|
#include <sysdep.h>
|
|
|
|
/* int [r3] strcmp (const char *s1 [r3], const char *s2 [r4]) */
|
|
|
|
EALIGN (strcmp, 4, 0)
|
|
CALL_MCOUNT 2
|
|
|
|
or r9, r3, r4
|
|
rldicl. r10, r9, 0, 61 /* are s1 and s2 8 byte aligned..? */
|
|
bne cr0, L(process_unaligned_bytes)
|
|
|
|
/* process input parameters on double word aligned boundary */
|
|
ld r9, 0(r4) /* load s2 at offset=0 */
|
|
li r10, 0 /* load mask=0 */
|
|
cmpb r10, r9, r10 /* compare bytes at s2 with mask */
|
|
cmpdi cr7, r10, 0 /* is NULL found ..? is end of string HIT */
|
|
bne cr7, L(process_unaligned_bytes) /* process byte by byte */
|
|
|
|
ld r10, 0(r3) /* load s1 at offset=0 */
|
|
li r8, 0 /* load mask=0 */
|
|
cmpb r8, r10, r8 /* compare bytes at s1 with mask */
|
|
cmpdi cr7, r8, 0 /* is NULL found ..? is end of string HIT */
|
|
bne cr7, L(process_unaligned_bytes) /* process byte by byte */
|
|
|
|
/*s1 and s2 does not contain NULL now , so compare all 8 bytes in a GO */
|
|
cmpb r9, r10, r9 /* compare s1 and s2 */
|
|
cmpdi cr7, r9, -1 /* compare result with 0xFFFFFFFFFFFFFFFF */
|
|
bne cr7, L(process_unaligned_bytes) /* s1,s2 mismatch found */
|
|
|
|
addi r5, r3, 8 /* save next offset of s2 */
|
|
addi r11, r4, 8 /* save next offset of s1 */
|
|
ld r8, 8(r4) /* load s2 at offset=8 */
|
|
li r9, 0 /* load mask=0 */
|
|
cmpb r9, r8, r9 /* compare bytes at s2 with mask */
|
|
cmpdi cr7, r9, 0 /* NULL found ..? */
|
|
bne cr7, L(processBytes)/* update input and process bytes one by one */
|
|
|
|
mr r9, r4 /* save s2 */
|
|
li r10, 0 /* load mask=0 */
|
|
|
|
ld r7, 8(r3) /* load s1 at offset=8 */
|
|
cmpb r6, r7, r10 /* compare bytes at s1 with mask */
|
|
cmpdi cr7, r6, 0 /* is NULL found */
|
|
bne cr7, L(processBytes)/* mismatch, so process one by one */
|
|
|
|
L(unrollDword):
|
|
cmpb r8, r7, r8 /* compare s1 and s2 */
|
|
cmpdi cr7, r8, -1 /* compare result with 0xFFFFFFFFFFFFFFFF */
|
|
bne cr7, L(processBytes)/* mismatch with s1 and s2 */
|
|
|
|
addi r5, r3, 16 /* save offset=16 of s1 */
|
|
addi r4, r9, 16 /* save offset=16 of s2 */
|
|
ld r8, 16(r9) /* load s2 at offset=16 */
|
|
cmpb r7, r8, r10 /* compare bytes at s2 with mask */
|
|
cmpdi cr7, r7, 0 /* NULL found ..? */
|
|
bne cr7, L(update2processBytes)
|
|
|
|
ld r7, 16(r3) /* load s1 at offset=16 */
|
|
cmpb r6, r7, r10 /* check s1 for end of string */
|
|
cmpdi cr7, r6, 0 /* end of s1 ?,then handle byte by byte */
|
|
bne 7,L(update2processBytes)
|
|
|
|
cmpb r8, r7, r8 /* compare s1 and s2 double words */
|
|
cmpdi cr7, r8, -1 /* compare results with 0xFFFFFFFFFFFFFFFF */
|
|
bne cr7,L(update2processBytes)
|
|
|
|
addi r5, r3, 24 /* update s1 to offset=24 */
|
|
addi r4, r9, 24 /* update s2 to offset=24 */
|
|
|
|
ld r8, 24(r9) /* load s2 */
|
|
cmpb r7, r8, r10 /* compare s2 for NULL */
|
|
cmpdi cr7, r7, 0 /* verify if s2 is ending now */
|
|
bne cr7,L(update2processBytes)
|
|
|
|
ld r7, 24(r3) /* load s1 at offset=24 */
|
|
cmpb r6, r7, r10 /* verify for NULL */
|
|
cmpdi cr7, r6, 0 /* is NULL found */
|
|
bne cr7, L(update2processBytes)
|
|
|
|
cmpb r8, r7, r8 /* compare s1 and s2 */
|
|
cmpdi cr7, r8, -1 /* are s1 and s2 same ..? */
|
|
bne cr7, L(update2processBytes)
|
|
|
|
addi r7, r9, 32 /* update s2 to next double word */
|
|
addi r3, r3, 32 /* update s1 to next double word */
|
|
|
|
ld r8, 32(r9) /* load s2 */
|
|
mr r4, r7 /* save s2 */
|
|
cmpb r6, r8, r10 /* compare s2 with NULL */
|
|
cmpdi cr7, r6, 0 /* end of s2 ..? */
|
|
bne cr7, L(process_unaligned_bytes)
|
|
|
|
ld r6, 0(r3) /* load and compare s1 for NULL */
|
|
cmpb r5, r6, r10
|
|
cmpdi cr7, r5, 0
|
|
bne cr7, L(process_unaligned_bytes)
|
|
|
|
cmpb r8, r6, r8 /* compare s1 and s2 */
|
|
cmpdi cr7, r8, -1
|
|
bne cr7, L(process_unaligned_bytes)
|
|
|
|
addi r5, r3, 8 /* increment s1 and d2 here */
|
|
addi r11, r9, 40
|
|
|
|
ld r8, 40(r9) /* process s2 now */
|
|
cmpb r9, r8, r10
|
|
cmpdi cr7, r9, 0
|
|
bne cr7, L(processBytes)
|
|
|
|
mr r9, r7
|
|
ld r7, 8(r3) /* process s1 now */
|
|
cmpb r6, r7, r10
|
|
cmpdi cr7, r6, 0
|
|
beq cr7, L(unrollDword) /* unroll to compare s1 and s2 */
|
|
|
|
L(processBytes):
|
|
mr r4, r11 /* update input params */
|
|
mr r3, r5
|
|
|
|
.p2align 4
|
|
L(process_unaligned_bytes):
|
|
lbz r9, 0(r3) /* load byte from s1 */
|
|
lbz r10, 0(r4) /* load byte from s2 */
|
|
cmpdi cr7, r9, 0 /* compare *s1 with NULL */
|
|
beq cr7, L(diffOfNULL) /* if *s1 is NULL , return *s1 - *s2 */
|
|
cmplw cr7, r9, r10 /* compare *s1 and *s2 */
|
|
bne cr7, L(ComputeDiff) /* branch to compute difference and return */
|
|
|
|
lbz r9, 1(r3) /* load next byte from s1 */
|
|
lbz r10, 1(r4) /* load next byte from s2 */
|
|
cmpdi cr7, r9, 0 /* compare *s1 with NULL */
|
|
beq cr7, L(diffOfNULL) /* if *s1 is NULL , return *s1 - *s2 */
|
|
cmplw cr7, r9, r10 /* compare *s1 and *s2 */
|
|
bne cr7, L(ComputeDiff) /* branch to compute difference and return */
|
|
|
|
lbz r9, 2(r3) /* unroll 3rd byte here */
|
|
lbz r10, 2(r4)
|
|
cmpdi cr7, r9, 0
|
|
beq cr7, L(diffOfNULL)
|
|
cmplw cr7, r9, r10
|
|
bne 7, L(ComputeDiff)
|
|
|
|
lbz r9, 3(r3) /* unroll 4th byte now */
|
|
lbz r10, 3(r4)
|
|
addi r3, r3, 4 /* increment s1 by unroll factor */
|
|
cmpdi cr7, r9, 0
|
|
cmplw cr6, 9, r10
|
|
beq cr7, L(diffOfNULL)
|
|
addi r4, r4, 4 /* increment s2 by unroll factor */
|
|
beq cr6, L(process_unaligned_bytes) /* unroll byte processing */
|
|
|
|
.p2align 4
|
|
L(ComputeDiff):
|
|
extsw r9, r9
|
|
subf r10, r10, r9 /* compute s1 - s2 */
|
|
extsw r3, r10
|
|
blr /* return */
|
|
|
|
.p2align 4
|
|
L(diffOfNULL):
|
|
li r9, 0
|
|
subf r10, r10, r9 /* compute s1 - s2 */
|
|
extsw r3, r10 /* sign extend result */
|
|
blr /* return */
|
|
|
|
.p2align 4
|
|
L(update2processBytes):
|
|
mr r3, r5 /* update and proceed */
|
|
b L(process_unaligned_bytes)
|
|
|
|
END (strcmp)
|
|
libc_hidden_builtin_def (strcmp)
|