[dragonfly.git] / lib / libc / x86_64 / string / strcpy.S

/*
 * Written by J.T. Conklin <jtc@acorntoolworks.com>
 * Public domain.
 *
 * $NetBSD: strcpy.S,v 1.3 2004/07/19 20:04:41 drochner Exp $
 * $FreeBSD: src/lib/libc/amd64/string/strcpy.S,v 1.3 2008/11/02 01:10:54 peter Exp $
 */

#include <machine/asm.h>

/*
 * This strcpy implementation copies a byte at a time until the
 * source pointer is aligned to a word boundary, it then copies by
 * words until it finds a word containing a zero byte, and finally
 * copies by bytes until the end of the string is reached.
 *
 * While this may result in unaligned stores if the source and
 * destination pointers are unaligned with respect to each other,
 * it is still faster than either byte copies or the overhead of
 * an implementation suitable for machines with strict alignment
 * requirements.
 */

ENTRY(strcpy)
	movq	%rdi,%rax
	movabsq	$0x0101010101010101,%r8
	movabsq	$0x8080808080808080,%r9

	/*
	 * Align source to a word boundary.
	 * Consider unrolling loop?
	 */
.Lalign:
	testb	$7,%sil
	je	.Lword_aligned
	movb	(%rsi),%dl
	incq	%rsi
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl
	jne	.Lalign
	ret

	.p2align 4
.Lloop:
	movq	%rdx,(%rdi)
	addq	$8,%rdi
.Lword_aligned:
	movq	(%rsi),%rdx
	movq	%rdx,%rcx
	addq	$8,%rsi
	subq	%r8,%rcx
	testq	%r9,%rcx
	je	.Lloop

	/*
	 * In rare cases, the above loop may exit prematurely. We must
	 * return to the loop if none of the bytes in the word equal 0.
	 */

	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 1st byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 2nd byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 3rd byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 4th byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 5th byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 6th byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 7th byte == 0? */
	je	.Ldone

	shrq	$8,%rdx
	movb	%dl,(%rdi)
	incq	%rdi
	testb	%dl,%dl		/* 8th byte == 0? */
	jne	.Lword_aligned

.Ldone:
	ret
END(strcpy)
Commit	Line	Data
f4a040fe PA	1	/*
	2	* Written by J.T. Conklin <jtc@acorntoolworks.com>
	3	* Public domain.
	4	*
	5	* $NetBSD: strcpy.S,v 1.3 2004/07/19 20:04:41 drochner Exp $
	6	* $FreeBSD: src/lib/libc/amd64/string/strcpy.S,v 1.3 2008/11/02 01:10:54 peter Exp $
	7	*/
	8
	9	#include <machine/asm.h>
	10
	11	/*
	12	* This strcpy implementation copies a byte at a time until the
	13	* source pointer is aligned to a word boundary, it then copies by
	14	* words until it finds a word containing a zero byte, and finally
	15	* copies by bytes until the end of the string is reached.
	16	*
	17	* While this may result in unaligned stores if the source and
	18	* destination pointers are unaligned with respect to each other,
	19	* it is still faster than either byte copies or the overhead of
	20	* an implementation suitable for machines with strict alignment
	21	* requirements.
	22	*/
	23
	24	ENTRY(strcpy)
	25	movq %rdi,%rax
	26	movabsq $0x0101010101010101,%r8
	27	movabsq $0x8080808080808080,%r9
	28
	29	/*
	30	* Align source to a word boundary.
	31	* Consider unrolling loop?
	32	*/
	33	.Lalign:
	34	testb $7,%sil
	35	je .Lword_aligned
	36	movb (%rsi),%dl
	37	incq %rsi
	38	movb %dl,(%rdi)
	39	incq %rdi
	40	testb %dl,%dl
	41	jne .Lalign
	42	ret
	43
	44	.p2align 4
	45	.Lloop:
	46	movq %rdx,(%rdi)
	47	addq $8,%rdi
	48	.Lword_aligned:
	49	movq (%rsi),%rdx
	50	movq %rdx,%rcx
	51	addq $8,%rsi
	52	subq %r8,%rcx
	53	testq %r9,%rcx
	54	je .Lloop
	55
	56	/*
	57	* In rare cases, the above loop may exit prematurely. We must
	58	* return to the loop if none of the bytes in the word equal 0.
	59	*/
	60
	61	movb %dl,(%rdi)
	62	incq %rdi
	63	testb %dl,%dl /* 1st byte == 0? */
	64	je .Ldone
65
66	shrq $8,%rdx
67	movb %dl,(%rdi)
68	incq %rdi
69	testb %dl,%dl /* 2nd byte == 0? */
70	je .Ldone
71
72	shrq $8,%rdx
73	movb %dl,(%rdi)
74	incq %rdi
75	testb %dl,%dl /* 3rd byte == 0? */
76	je .Ldone
77
78	shrq $8,%rdx
79	movb %dl,(%rdi)
80	incq %rdi
81	testb %dl,%dl /* 4th byte == 0? */
82	je .Ldone
83
84	shrq $8,%rdx
85	movb %dl,(%rdi)
86	incq %rdi
87	testb %dl,%dl /* 5th byte == 0? */
88	je .Ldone
89
90	shrq $8,%rdx
91	movb %dl,(%rdi)
92	incq %rdi
93	testb %dl,%dl /* 6th byte == 0? */
94	je .Ldone
95
96	shrq $8,%rdx
97	movb %dl,(%rdi)
98	incq %rdi
99	testb %dl,%dl /* 7th byte == 0? */
100	je .Ldone
101
102	shrq $8,%rdx
103	movb %dl,(%rdi)
104	incq %rdi
105	testb %dl,%dl /* 8th byte == 0? */
106	jne .Lword_aligned
107
108	.Ldone:
109	ret
110	END(strcpy)