Blame sysdeps/powerpc/powerpc32/strncmp.S

Packit 6c4009
/* Optimized strcmp implementation for PowerPC32.
Packit 6c4009
   Copyright (C) 2003-2018 Free Software Foundation, Inc.
Packit 6c4009
   This file is part of the GNU C Library.
Packit 6c4009
Packit 6c4009
   The GNU C Library is free software; you can redistribute it and/or
Packit 6c4009
   modify it under the terms of the GNU Lesser General Public
Packit 6c4009
   License as published by the Free Software Foundation; either
Packit 6c4009
   version 2.1 of the License, or (at your option) any later version.
Packit 6c4009
Packit 6c4009
   The GNU C Library is distributed in the hope that it will be useful,
Packit 6c4009
   but WITHOUT ANY WARRANTY; without even the implied warranty of
Packit 6c4009
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
Packit 6c4009
   Lesser General Public License for more details.
Packit 6c4009
Packit 6c4009
   You should have received a copy of the GNU Lesser General Public
Packit 6c4009
   License along with the GNU C Library; if not, see
Packit 6c4009
   <http://www.gnu.org/licenses/>.  */
Packit 6c4009
Packit 6c4009
#include <sysdep.h>
Packit 6c4009
Packit 6c4009
/* See strlen.s for comments on how the end-of-string testing works.  */
Packit 6c4009
Packit 6c4009
/* int [r3] strncmp (const char *s1 [r3], const char *s2 [r4], size_t size [r5])  */
Packit 6c4009
Packit 6c4009
EALIGN (strncmp, 4, 0)
Packit 6c4009
Packit 6c4009
#define rTMP2	r0
Packit 6c4009
#define rRTN	r3
Packit 6c4009
#define rSTR1	r3	/* first string arg */
Packit 6c4009
#define rSTR2	r4	/* second string arg */
Packit 6c4009
#define rN	r5	/* max string length */
Packit 6c4009
#define rWORD1	r6	/* current word in s1 */
Packit 6c4009
#define rWORD2	r7	/* current word in s2 */
Packit 6c4009
#define rFEFE	r8	/* constant 0xfefefeff (-0x01010101) */
Packit 6c4009
#define r7F7F	r9	/* constant 0x7f7f7f7f */
Packit 6c4009
#define rNEG	r10	/* ~(word in s1 | 0x7f7f7f7f) */
Packit 6c4009
#define rBITDIF	r11	/* bits that differ in s1 & s2 words */
Packit 6c4009
#define rTMP	r12
Packit 6c4009
Packit 6c4009
	dcbt	0,rSTR1
Packit 6c4009
	or	rTMP, rSTR2, rSTR1
Packit 6c4009
	lis	r7F7F, 0x7f7f
Packit 6c4009
	dcbt	0,rSTR2
Packit 6c4009
	clrlwi.	rTMP, rTMP, 30
Packit 6c4009
	cmplwi	cr1, rN, 0
Packit 6c4009
	lis	rFEFE, -0x101
Packit 6c4009
	bne	L(unaligned)
Packit 6c4009
/* We are word aligned so set up for two loops.  first a word
Packit 6c4009
   loop, then fall into the byte loop if any residual.  */
Packit 6c4009
	srwi.	rTMP, rN, 2
Packit 6c4009
	clrlwi	rN, rN, 30
Packit 6c4009
	addi	rFEFE, rFEFE, -0x101
Packit 6c4009
	addi	r7F7F, r7F7F, 0x7f7f
Packit 6c4009
	cmplwi	cr1, rN, 0
Packit 6c4009
	beq	L(unaligned)
Packit 6c4009
Packit 6c4009
	mtctr	rTMP	/* Power4 wants mtctr 1st in dispatch group.  */
Packit 6c4009
	lwz	rWORD1, 0(rSTR1)
Packit 6c4009
	lwz	rWORD2, 0(rSTR2)
Packit 6c4009
	b	L(g1)
Packit 6c4009
Packit 6c4009
L(g0):
Packit 6c4009
	lwzu	rWORD1, 4(rSTR1)
Packit 6c4009
	bne-	cr1, L(different)
Packit 6c4009
	lwzu	rWORD2, 4(rSTR2)
Packit 6c4009
L(g1):	add	rTMP, rFEFE, rWORD1
Packit 6c4009
	nor	rNEG, r7F7F, rWORD1
Packit 6c4009
	bdz	L(tail)
Packit 6c4009
	and.	rTMP, rTMP, rNEG
Packit 6c4009
	cmpw	cr1, rWORD1, rWORD2
Packit 6c4009
	beq+	L(g0)
Packit 6c4009
Packit 6c4009
/* OK. We've hit the end of the string. We need to be careful that
Packit 6c4009
   we don't compare two strings as different because of gunk beyond
Packit 6c4009
   the end of the strings...  */
Packit 6c4009
Packit 6c4009
#ifdef __LITTLE_ENDIAN__
Packit 6c4009
L(endstring):
Packit 6c4009
	slwi	rTMP, rTMP, 1
Packit 6c4009
	addi    rTMP2, rTMP, -1
Packit 6c4009
	andc    rTMP2, rTMP2, rTMP
Packit 6c4009
	and	rWORD2, rWORD2, rTMP2		/* Mask off gunk.  */
Packit 6c4009
	and	rWORD1, rWORD1, rTMP2
Packit 6c4009
	rlwinm	rTMP2, rWORD2, 8, 0xffffffff	/* Byte reverse word.  */
Packit 6c4009
	rlwinm	rTMP, rWORD1, 8, 0xffffffff
Packit 6c4009
	rlwimi	rTMP2, rWORD2, 24, 0, 7
Packit 6c4009
	rlwimi	rTMP, rWORD1, 24, 0, 7
Packit 6c4009
	rlwimi	rTMP2, rWORD2, 24, 16, 23
Packit 6c4009
	rlwimi	rTMP, rWORD1, 24, 16, 23
Packit 6c4009
	xor.	rBITDIF, rTMP, rTMP2
Packit 6c4009
	sub	rRTN, rTMP, rTMP2
Packit 6c4009
	bgelr+
Packit 6c4009
	ori	rRTN, rTMP2, 1
Packit 6c4009
	blr
Packit 6c4009
Packit 6c4009
L(different):
Packit 6c4009
	lwz	rWORD1, -4(rSTR1)
Packit 6c4009
	rlwinm	rTMP2, rWORD2, 8, 0xffffffff	/* Byte reverse word.  */
Packit 6c4009
	rlwinm	rTMP, rWORD1, 8, 0xffffffff
Packit 6c4009
	rlwimi	rTMP2, rWORD2, 24, 0, 7
Packit 6c4009
	rlwimi	rTMP, rWORD1, 24, 0, 7
Packit 6c4009
	rlwimi	rTMP2, rWORD2, 24, 16, 23
Packit 6c4009
	rlwimi	rTMP, rWORD1, 24, 16, 23
Packit 6c4009
	xor.	rBITDIF, rTMP, rTMP2
Packit 6c4009
	sub	rRTN, rTMP, rTMP2
Packit 6c4009
	bgelr+
Packit 6c4009
	ori	rRTN, rTMP2, 1
Packit 6c4009
	blr
Packit 6c4009
Packit 6c4009
#else
Packit 6c4009
L(endstring):
Packit 6c4009
	and	rTMP, r7F7F, rWORD1
Packit 6c4009
	beq	cr1, L(equal)
Packit 6c4009
	add	rTMP, rTMP, r7F7F
Packit 6c4009
	xor.	rBITDIF, rWORD1, rWORD2
Packit 6c4009
	andc	rNEG, rNEG, rTMP
Packit 6c4009
	blt-	L(highbit)
Packit 6c4009
	cntlzw	rBITDIF, rBITDIF
Packit 6c4009
	cntlzw	rNEG, rNEG
Packit 6c4009
	addi	rNEG, rNEG, 7
Packit 6c4009
	cmpw	cr1, rNEG, rBITDIF
Packit 6c4009
	sub	rRTN, rWORD1, rWORD2
Packit 6c4009
	bgelr+	cr1
Packit 6c4009
L(equal):
Packit 6c4009
	li	rRTN, 0
Packit 6c4009
	blr
Packit 6c4009
Packit 6c4009
L(different):
Packit 6c4009
	lwz	rWORD1, -4(rSTR1)
Packit 6c4009
	xor.	rBITDIF, rWORD1, rWORD2
Packit 6c4009
	sub	rRTN, rWORD1, rWORD2
Packit 6c4009
	bgelr+
Packit 6c4009
L(highbit):
Packit 6c4009
	ori	rRTN, rWORD2, 1
Packit 6c4009
	blr
Packit 6c4009
#endif
Packit 6c4009
Packit 6c4009
/* Oh well.  In this case, we just do a byte-by-byte comparison.  */
Packit 6c4009
	.align 4
Packit 6c4009
L(tail):
Packit 6c4009
	and.	rTMP, rTMP, rNEG
Packit 6c4009
	cmpw	cr1, rWORD1, rWORD2
Packit 6c4009
	bne-	L(endstring)
Packit 6c4009
	addi	rSTR1, rSTR1, 4
Packit 6c4009
	bne-	cr1, L(different)
Packit 6c4009
	addi	rSTR2, rSTR2, 4
Packit 6c4009
	cmplwi	cr1, rN, 0
Packit 6c4009
L(unaligned):
Packit 6c4009
	mtctr   rN	/* Power4 wants mtctr 1st in dispatch group */
Packit 6c4009
	bgt	cr1, L(uz)
Packit 6c4009
L(ux):
Packit 6c4009
	li	rRTN, 0
Packit 6c4009
	blr
Packit 6c4009
	.align 4
Packit 6c4009
L(uz):
Packit 6c4009
	lbz	rWORD1, 0(rSTR1)
Packit 6c4009
	lbz	rWORD2, 0(rSTR2)
Packit 6c4009
	nop
Packit 6c4009
	b	L(u1)
Packit 6c4009
L(u0):
Packit 6c4009
	lbzu	rWORD2, 1(rSTR2)
Packit 6c4009
L(u1):
Packit 6c4009
	bdz	L(u3)
Packit 6c4009
	cmpwi	cr1, rWORD1, 0
Packit 6c4009
	cmpw	rWORD1, rWORD2
Packit 6c4009
	beq-	cr1, L(u3)
Packit 6c4009
	lbzu	rWORD1, 1(rSTR1)
Packit 6c4009
	bne-	L(u2)
Packit 6c4009
	lbzu	rWORD2, 1(rSTR2)
Packit 6c4009
	bdz	L(u3)
Packit 6c4009
	cmpwi	cr1, rWORD1, 0
Packit 6c4009
	cmpw	rWORD1, rWORD2
Packit 6c4009
	bne-	L(u3)
Packit 6c4009
	lbzu	rWORD1, 1(rSTR1)
Packit 6c4009
	bne+	cr1, L(u0)
Packit 6c4009
Packit 6c4009
L(u2):	lbzu	rWORD1, -1(rSTR1)
Packit 6c4009
L(u3):	sub	rRTN, rWORD1, rWORD2
Packit 6c4009
	blr
Packit 6c4009
END (strncmp)
Packit 6c4009
libc_hidden_builtin_def (strncmp)