| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* strlen.S: Sparc optimized strlen code | 
 | 2 |  * Hand optimized from GNU libc's strlen | 
 | 3 |  * Copyright (C) 1991,1996 Free Software Foundation | 
 | 4 |  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) | 
 | 5 |  * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz) | 
 | 6 |  */ | 
 | 7 |  | 
 | 8 | #define LO_MAGIC 0x01010101 | 
 | 9 | #define HI_MAGIC 0x80808080 | 
 | 10 |  | 
 | 11 | 0: | 
 | 12 | 	ldub	[%o0], %o5 | 
 | 13 | 	cmp	%o5, 0 | 
 | 14 | 	be	1f | 
 | 15 | 	 add	%o0, 1, %o0 | 
 | 16 | 	andcc	%o0, 3, %g0 | 
 | 17 | 	be	4f | 
 | 18 | 	 or	%o4, %lo(HI_MAGIC), %o3 | 
 | 19 | 	ldub	[%o0], %o5 | 
 | 20 | 	cmp	%o5, 0 | 
 | 21 | 	be	2f | 
 | 22 | 	 add	%o0, 1, %o0 | 
 | 23 | 	andcc	%o0, 3, %g0 | 
 | 24 | 	be	5f | 
 | 25 | 	 sethi	%hi(LO_MAGIC), %o4 | 
 | 26 | 	ldub	[%o0], %o5 | 
 | 27 | 	cmp	%o5, 0 | 
 | 28 | 	be	3f | 
 | 29 | 	 add	%o0, 1, %o0 | 
 | 30 | 	b	8f | 
 | 31 | 	 or	%o4, %lo(LO_MAGIC), %o2 | 
 | 32 | 1: | 
 | 33 | 	retl | 
 | 34 | 	 mov	0, %o0 | 
 | 35 | 2: | 
 | 36 | 	retl | 
 | 37 | 	 mov	1, %o0 | 
 | 38 | 3: | 
 | 39 | 	retl | 
 | 40 | 	 mov	2, %o0 | 
 | 41 |  | 
 | 42 | 	.align 4 | 
 | 43 | 	.global strlen | 
 | 44 | strlen: | 
 | 45 | 	mov	%o0, %o1 | 
 | 46 | 	andcc	%o0, 3, %g0 | 
 | 47 | 	bne	0b | 
 | 48 | 	 sethi	%hi(HI_MAGIC), %o4 | 
 | 49 | 	or	%o4, %lo(HI_MAGIC), %o3 | 
 | 50 | 4: | 
 | 51 | 	sethi	%hi(LO_MAGIC), %o4 | 
 | 52 | 5: | 
 | 53 | 	or	%o4, %lo(LO_MAGIC), %o2 | 
 | 54 | 8: | 
 | 55 | 	ld	[%o0], %o5 | 
 | 56 | 2: | 
 | 57 | 	sub	%o5, %o2, %o4 | 
 | 58 | 	andcc	%o4, %o3, %g0 | 
 | 59 | 	be	8b | 
 | 60 | 	 add	%o0, 4, %o0 | 
 | 61 |  | 
 | 62 | 	/* Check every byte. */ | 
 | 63 | 	srl	%o5, 24, %g5 | 
 | 64 | 	andcc	%g5, 0xff, %g0 | 
 | 65 | 	be	1f | 
 | 66 | 	 add	%o0, -4, %o4 | 
 | 67 | 	srl	%o5, 16, %g5 | 
 | 68 | 	andcc	%g5, 0xff, %g0 | 
 | 69 | 	be	1f | 
 | 70 | 	 add	%o4, 1, %o4 | 
 | 71 | 	srl	%o5, 8, %g5 | 
 | 72 | 	andcc	%g5, 0xff, %g0 | 
 | 73 | 	be	1f | 
 | 74 | 	 add	%o4, 1, %o4 | 
 | 75 | 	andcc	%o5, 0xff, %g0 | 
 | 76 | 	bne,a	2b | 
 | 77 | 	 ld	[%o0], %o5 | 
 | 78 | 	add	%o4, 1, %o4 | 
 | 79 | 1: | 
 | 80 | 	retl | 
 | 81 | 	 sub	%o4, %o1, %o0 |