mirror of
https://sourceware.org/git/glibc.git
synced 2024-12-03 10:21:05 +00:00
194 lines
5.0 KiB
ArmAsm
194 lines
5.0 KiB
ArmAsm
/* Optimized memchr implementation for PowerPC32/POWER7 using cmpb insn.
|
|
Copyright (C) 2010-2020 Free Software Foundation, Inc.
|
|
Contributed by Luis Machado <luisgpm@br.ibm.com>.
|
|
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
|
|
<https://www.gnu.org/licenses/>. */
|
|
|
|
#include <sysdep.h>
|
|
|
|
/* int [r3] memchr (char *s [r3], int byte [r4], int size [r5]) */
|
|
.machine power7
|
|
ENTRY (__memchr)
|
|
CALL_MCOUNT
|
|
dcbt 0,r3
|
|
clrrwi r8,r3,2
|
|
insrwi r4,r4,8,16 /* Replicate byte to word. */
|
|
|
|
/* Calculate the last acceptable address and check for possible
|
|
addition overflow by using satured math:
|
|
r7 = r3 + r5
|
|
r7 |= -(r7 < x) */
|
|
add r7,r3,r5
|
|
subfc r6,r3,r7
|
|
subfe r9,r9,r9
|
|
or r7,r7,r9
|
|
|
|
insrwi r4,r4,16,0
|
|
cmplwi r5,16
|
|
li r9, -1
|
|
rlwinm r6,r3,3,27,28 /* Calculate padding. */
|
|
addi r7,r7,-1
|
|
#ifdef __LITTLE_ENDIAN__
|
|
slw r9,r9,r6
|
|
#else
|
|
srw r9,r9,r6
|
|
#endif
|
|
ble L(small_range)
|
|
|
|
lwz r12,0(r8) /* Load word from memory. */
|
|
cmpb r3,r12,r4 /* Check for BYTEs in WORD1. */
|
|
and r3,r3,r9
|
|
clrlwi r5,r7,30 /* Byte count - 1 in last word. */
|
|
clrrwi r7,r7,2 /* Address of last word. */
|
|
cmplwi cr7,r3,0 /* If r3 == 0, no BYTEs have been found. */
|
|
bne cr7,L(done)
|
|
|
|
mtcrf 0x01,r8
|
|
/* Are we now aligned to a doubleword boundary? If so, skip to
|
|
the main loop. Otherwise, go through the alignment code. */
|
|
bt 29,L(loop_setup)
|
|
|
|
/* Handle WORD2 of pair. */
|
|
lwzu r12,4(r8)
|
|
cmpb r3,r12,r4
|
|
cmplwi cr7,r3,0
|
|
bne cr7,L(done)
|
|
|
|
L(loop_setup):
|
|
/* The last word we want to read in the loop below is the one
|
|
containing the last byte of the string, ie. the word at
|
|
(s + size - 1) & ~3, or r7. The first word read is at
|
|
r8 + 4, we read 2 * cnt words, so the last word read will
|
|
be at r8 + 4 + 8 * cnt - 4. Solving for cnt gives
|
|
cnt = (r7 - r8) / 8 */
|
|
sub r6,r7,r8
|
|
srwi r6,r6,3 /* Number of loop iterations. */
|
|
mtctr r6 /* Setup the counter. */
|
|
|
|
/* Main loop to look for BYTE in the string. Since
|
|
it's a small loop (8 instructions), align it to 32-bytes. */
|
|
.align 5
|
|
L(loop):
|
|
/* Load two words, compare and merge in a
|
|
single register for speed. This is an attempt
|
|
to speed up the byte-checking process for bigger strings. */
|
|
lwz r12,4(r8)
|
|
lwzu r11,8(r8)
|
|
cmpb r3,r12,r4
|
|
cmpb r9,r11,r4
|
|
or r6,r9,r3 /* Merge everything in one word. */
|
|
cmplwi cr7,r6,0
|
|
bne cr7,L(found)
|
|
bdnz L(loop)
|
|
|
|
/* We may have one more dword to read. */
|
|
cmplw r8,r7
|
|
beqlr
|
|
|
|
lwzu r12,4(r8)
|
|
cmpb r3,r12,r4
|
|
cmplwi cr6,r3,0
|
|
bne cr6,L(done)
|
|
blr
|
|
|
|
.align 4
|
|
L(found):
|
|
/* OK, one (or both) of the words contains BYTE. Check
|
|
the first word and decrement the address in case the first
|
|
word really contains BYTE. */
|
|
cmplwi cr6,r3,0
|
|
addi r8,r8,-4
|
|
bne cr6,L(done)
|
|
|
|
/* BYTE must be in the second word. Adjust the address
|
|
again and move the result of cmpb to r3 so we can calculate the
|
|
pointer. */
|
|
|
|
mr r3,r9
|
|
addi r8,r8,4
|
|
|
|
/* r3 has the output of the cmpb instruction, that is, it contains
|
|
0xff in the same position as BYTE in the original
|
|
word from the string. Use that to calculate the pointer.
|
|
We need to make sure BYTE is *before* the end of the range. */
|
|
L(done):
|
|
#ifdef __LITTLE_ENDIAN__
|
|
addi r0,r3,-1
|
|
andc r0,r0,r3
|
|
popcntw r0,r0 /* Count trailing zeros. */
|
|
#else
|
|
cntlzw r0,r3 /* Count leading zeros before the match. */
|
|
#endif
|
|
cmplw r8,r7 /* Are we on the last word? */
|
|
srwi r0,r0,3 /* Convert leading/trailing zeros to bytes. */
|
|
add r3,r8,r0
|
|
cmplw cr7,r0,r5 /* If on the last dword, check byte offset. */
|
|
bnelr
|
|
blelr cr7
|
|
li r3,0
|
|
blr
|
|
|
|
.align 4
|
|
L(null):
|
|
li r3,0
|
|
blr
|
|
|
|
/* Deals with size <= 16. */
|
|
.align 4
|
|
L(small_range):
|
|
cmplwi r5,0
|
|
beq L(null)
|
|
lwz r12,0(r8) /* Load word from memory. */
|
|
cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
|
|
and r3,r3,r9
|
|
cmplwi cr7,r3,0
|
|
clrlwi r5,r7,30 /* Byte count - 1 in last word. */
|
|
clrrwi r7,r7,2 /* Address of last word. */
|
|
cmplw r8,r7 /* Are we done already? */
|
|
bne cr7,L(done)
|
|
beqlr
|
|
|
|
lwzu r12,4(r8)
|
|
cmpb r3,r12,r4
|
|
cmplwi cr6,r3,0
|
|
cmplw r8,r7
|
|
bne cr6,L(done)
|
|
beqlr
|
|
|
|
lwzu r12,4(r8)
|
|
cmpb r3,r12,r4
|
|
cmplwi cr6,r3,0
|
|
cmplw r8,r7
|
|
bne cr6,L(done)
|
|
beqlr
|
|
|
|
lwzu r12,4(r8)
|
|
cmpb r3,r12,r4
|
|
cmplwi cr6,r3,0
|
|
cmplw r8,r7
|
|
bne cr6,L(done)
|
|
beqlr
|
|
|
|
lwzu r12,4(r8)
|
|
cmpb r3,r12,r4
|
|
cmplwi cr6,r3,0
|
|
bne cr6,L(done)
|
|
blr
|
|
|
|
END (__memchr)
|
|
weak_alias (__memchr, memchr)
|
|
libc_hidden_builtin_def (memchr)
|