xref: /openbmc/linux/arch/x86/lib/memset_64.S (revision 83a7a2ad)
1/* Copyright 2002 Andi Kleen, SuSE Labs */
2
3#include <linux/linkage.h>
4#include <asm/dwarf2.h>
5
6/*
7 * ISO C memset - set a memory block to a byte value.
8 *
9 * rdi   destination
10 * rsi   value (char)
11 * rdx   count (bytes)
12 *
13 * rax   original destination
14 */
15	.section .altinstr_replacement, "ax", @progbits
16.Lmemset_c:
17	movq %rdi,%r9
18	movl %edx,%r8d
19	andl $7,%r8d
20	movl %edx,%ecx
21	shrl $3,%ecx
22	/* expand byte value  */
23	movzbl %sil,%esi
24	movabs $0x0101010101010101,%rax
25	mulq %rsi		/* with rax, clobbers rdx */
26	rep stosq
27	movl %r8d,%ecx
28	rep stosb
29	movq %r9,%rax
30	ret
31.Lmemset_e:
32	.previous
33
34ENTRY(memset)
35ENTRY(__memset)
36	CFI_STARTPROC
37	movq %rdi,%r10
38	movq %rdx,%r11
39
40	/* expand byte value  */
41	movzbl %sil,%ecx
42	movabs $0x0101010101010101,%rax
43	mul    %rcx		/* with rax, clobbers rdx */
44
45	/* align dst */
46	movl  %edi,%r9d
47	andl  $7,%r9d
48	jnz  .Lbad_alignment
49	CFI_REMEMBER_STATE
50.Lafter_bad_alignment:
51
52	movl %r11d,%ecx
53	shrl $6,%ecx
54	jz	 .Lhandle_tail
55
56	.p2align 4
57.Lloop_64:
58	decl   %ecx
59	movq  %rax,(%rdi)
60	movq  %rax,8(%rdi)
61	movq  %rax,16(%rdi)
62	movq  %rax,24(%rdi)
63	movq  %rax,32(%rdi)
64	movq  %rax,40(%rdi)
65	movq  %rax,48(%rdi)
66	movq  %rax,56(%rdi)
67	leaq  64(%rdi),%rdi
68	jnz    .Lloop_64
69
70	/* Handle tail in loops. The loops should be faster than hard
71	   to predict jump tables. */
72	.p2align 4
73.Lhandle_tail:
74	movl	%r11d,%ecx
75	andl    $63&(~7),%ecx
76	jz 		.Lhandle_7
77	shrl	$3,%ecx
78	.p2align 4
79.Lloop_8:
80	decl   %ecx
81	movq  %rax,(%rdi)
82	leaq  8(%rdi),%rdi
83	jnz    .Lloop_8
84
85.Lhandle_7:
86	movl	%r11d,%ecx
87	andl	$7,%ecx
88	jz      .Lende
89	.p2align 4
90.Lloop_1:
91	decl    %ecx
92	movb 	%al,(%rdi)
93	leaq	1(%rdi),%rdi
94	jnz     .Lloop_1
95
96.Lende:
97	movq	%r10,%rax
98	ret
99
100	CFI_RESTORE_STATE
101.Lbad_alignment:
102	cmpq $7,%r11
103	jbe	.Lhandle_7
104	movq %rax,(%rdi)	/* unaligned store */
105	movq $8,%r8
106	subq %r9,%r8
107	addq %r8,%rdi
108	subq %r8,%r11
109	jmp .Lafter_bad_alignment
110.Lfinal:
111	CFI_ENDPROC
112ENDPROC(memset)
113ENDPROC(__memset)
114
115	/* Some CPUs run faster using the string instructions.
116	   It is also a lot simpler. Use this when possible */
117
118#include <asm/cpufeature.h>
119
120	.section .altinstructions,"a"
121	.align 8
122	.quad memset
123	.quad .Lmemset_c
124	.word X86_FEATURE_REP_GOOD
125	.byte .Lfinal - memset
126	.byte .Lmemset_e - .Lmemset_c
127	.previous
128