brintos

brintos / llvm-project-archived public Read only

0
0
Text · 3.9 KiB · c03d471 Raw
135 lines · plain
1/*2 * memchr - scan memory for a character3 *4 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.5 * See https://llvm.org/LICENSE.txt for license information.6 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception7 */8 9/*10   Written by Dave Gilbert <david.gilbert@linaro.org>11 12   This __memchr_arm routine is optimised on a Cortex-A9 and should work on13   all ARMv7 processors.   It has a fast past for short sizes, and has14   an optimised path for large data sets; the worst case is finding the15   match early in a large data set.16 17 */18 19@ 2011-02-07 david.gilbert@linaro.org20@    Extracted from local git a5b438d86121@ 2011-07-14 david.gilbert@linaro.org22@    Import endianness fix from local git ea786f1b23@ 2011-12-07 david.gilbert@linaro.org24@    Removed unneeded cbz from align loop25 26	.syntax unified27	.arch armv7-a28 29@ this lets us check a flag in a 00/ff byte easily in either endianness30#ifdef __ARMEB__31#define CHARTSTMASK(c) 1<<(31-(c*8))32#else33#define CHARTSTMASK(c) 1<<(c*8)34#endif35	.text36	.thumb37 38@ ---------------------------------------------------------------------------39	.thumb_func40	.align 241	.p2align 4,,1542	.global __memchr_arm43	.type __memchr_arm,%function44__memchr_arm:45	@ r0 = start of memory to scan46	@ r1 = character to look for47	@ r2 = length48	@ returns r0 = pointer to character or NULL if not found49	and	r1,r1,#0xff	@ Don't think we can trust the caller to actually pass a char50 51	cmp	r2,#16		@ If it's short don't bother with anything clever52	blt	20f53 54	tst	r0, #7		@ If it's already aligned skip the next bit55	beq	10f56 57	@ Work up to an aligned point585:59	ldrb	r3, [r0],#160	subs	r2, r2, #161	cmp	r3, r162	beq	50f		@ If it matches exit found63	tst	r0, #764	bne	5b		@ If not aligned yet then do next byte65 6610:67	@ At this point, we are aligned, we know we have at least 8 bytes to work with68	push	{r4,r5,r6,r7}69	orr	r1, r1, r1, lsl #8	@ expand the match word across to all bytes70	orr	r1, r1, r1, lsl #1671	bic	r4, r2, #7	@ Number of double words to work with72	mvns	r7, #0		@ all F's73	movs	r3, #074 7515:76	ldmia	r0!,{r5,r6}77	subs	r4, r4, #878	eor	r5,r5, r1	@ Get it so that r5,r6 have 00's where the bytes match the target79	eor	r6,r6, r180	uadd8	r5, r5, r7	@ Parallel add 0xff - sets the GE bits for anything that wasn't 081	sel	r5, r3, r7	@ bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION82	uadd8	r6, r6, r7	@ Parallel add 0xff - sets the GE bits for anything that wasn't 083	sel	r6, r5, r7	@ chained....bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION84	cbnz	r6, 60f85	bne	15b		@ (Flags from the subs above) If not run out of bytes then go around again86 87	pop	{r4,r5,r6,r7}88	and	r1,r1,#0xff	@ Get r1 back to a single character from the expansion above89	and	r2,r2,#7	@ Leave the count remaining as the number after the double words have been done90 9120:92	cbz	r2, 40f		@ 0 length or hit the end already then not found93 9421:  @ Post aligned section, or just a short call95	ldrb	r3,[r0],#196	subs	r2,r2,#197	eor	r3,r3,r1	@ r3 = 0 if match - doesn't break flags from sub98	cbz	r3, 50f99	bne	21b		@ on r2 flags100 10140:102	movs	r0,#0		@ not found103	bx	lr104 10550:106	subs	r0,r0,#1	@ found107	bx	lr108 10960:  @ We're here because the fast path found a hit - now we have to track down exactly which word it was110	@ r0 points to the start of the double word after the one that was tested111	@ r5 has the 00/ff pattern for the first word, r6 has the chained value112	cmp	r5, #0113	itte	eq114	moveq	r5, r6		@ the end is in the 2nd word115	subeq	r0,r0,#3	@ Points to 2nd byte of 2nd word116	subne	r0,r0,#7	@ or 2nd byte of 1st word117 118	@ r0 currently points to the 3rd byte of the word containing the hit119	tst	r5, # CHARTSTMASK(0)	@ 1st character120	bne	61f121	adds	r0,r0,#1122	tst	r5, # CHARTSTMASK(1)	@ 2nd character123	ittt	eq124	addeq	r0,r0,#1125	tsteq	r5, # (3<<15)		@ 2nd & 3rd character126	@ If not the 3rd must be the last one127	addeq	r0,r0,#1128 12961:130	pop	{r4,r5,r6,r7}131	subs	r0,r0,#1132	bx	lr133 134	.size	__memchr_arm, . - __memchr_arm135