1/*
2 * linux/arch/alpha/lib/memset.S
3 *
4 * This is an efficient (and small) implementation of the C library "memset()"
5 * function for the alpha.
6 *
7 *	(C) Copyright 1996 Linus Torvalds
8 *
9 * This routine is "moral-ware": you are free to use it any way you wish, and
10 * the only obligation I put on you is a moral one: if you make any improvements
11 * to the routine, please send me your improvements for me to use similarly.
12 *
13 * The scheduling comments are according to the EV5 documentation (and done by
14 * hand, so they might well be incorrect, please do tell me about it..)
15 */
16
17	.set noat
18	.set noreorder
19.text
20	.globl memset
21	.globl __memset
22	.globl ___memset
23	.globl __memsetw
24	.globl __constant_c_memset
25
26	.ent ___memset
27.align 5
28___memset:
29	.frame $30,0,$26,0
30	.prologue 0
31
32	and $17,255,$1		/* E1 */
33	insbl $17,1,$17		/* .. E0 */
34	bis $17,$1,$17		/* E0 (p-c latency, next cycle) */
35	sll $17,16,$1		/* E1 (p-c latency, next cycle) */
36
37	bis $17,$1,$17		/* E0 (p-c latency, next cycle) */
38	sll $17,32,$1		/* E1 (p-c latency, next cycle) */
39	bis $17,$1,$17		/* E0 (p-c latency, next cycle) */
40	ldq_u $31,0($30)	/* .. E1 */
41
42.align 5
43__constant_c_memset:
44	addq $18,$16,$6		/* E0 */
45	bis $16,$16,$0		/* .. E1 */
46	xor $16,$6,$1		/* E0 */
47	ble $18,end		/* .. E1 */
48
49	bic $1,7,$1		/* E0 */
50	beq $1,within_one_quad	/* .. E1 (note EV5 zero-latency forwarding) */
51	and $16,7,$3		/* E0 */
52	beq $3,aligned		/* .. E1 (note EV5 zero-latency forwarding) */
53
54	ldq_u $4,0($16)		/* E0 */
55	bis $16,$16,$5		/* .. E1 */
56	insql $17,$16,$2	/* E0 */
57	subq $3,8,$3		/* .. E1 */
58
59	addq $18,$3,$18		/* E0	$18 is new count ($3 is negative) */
60	mskql $4,$16,$4		/* .. E1 (and possible load stall) */
61	subq $16,$3,$16		/* E0 	$16 is new aligned destination */
62	bis $2,$4,$1		/* .. E1 */
63
64	bis $31,$31,$31		/* E0 */
65	ldq_u $31,0($30)	/* .. E1 */
66	stq_u $1,0($5)		/* E0 */
67	bis $31,$31,$31		/* .. E1 */
68
69.align 4
70aligned:
71	sra $18,3,$3		/* E0 */
72	and $18,7,$18		/* .. E1 */
73	bis $16,$16,$5		/* E0 */
74	beq $3,no_quad		/* .. E1 */
75
76.align 3
77loop:
78	stq $17,0($5)		/* E0 */
79	subq $3,1,$3		/* .. E1 */
80	addq $5,8,$5		/* E0 */
81	bne $3,loop		/* .. E1 */
82
83no_quad:
84	bis $31,$31,$31		/* E0 */
85	beq $18,end		/* .. E1 */
86	ldq $7,0($5)		/* E0 */
87	mskqh $7,$6,$2		/* .. E1 (and load stall) */
88
89	insqh $17,$6,$4		/* E0 */
90	bis $2,$4,$1		/* .. E1 */
91	stq $1,0($5)		/* E0 */
92	ret $31,($26),1		/* .. E1 */
93
94.align 3
95within_one_quad:
96	ldq_u $1,0($16)		/* E0 */
97	insql $17,$16,$2	/* E1 */
98	mskql $1,$16,$4		/* E0 (after load stall) */
99	bis $2,$4,$2		/* E0 */
100
101	mskql $2,$6,$4		/* E0 */
102	mskqh $1,$6,$2		/* .. E1 */
103	bis $2,$4,$1		/* E0 */
104	stq_u $1,0($16)		/* E0 */
105
106end:
107	ret $31,($26),1		/* E1 */
108	.end ___memset
109
110	.align 5
111	.ent __memsetw
112__memsetw:
113	.prologue 0
114
115	inswl $17,0,$1		/* E0 */
116	inswl $17,2,$2		/* E0 */
117	inswl $17,4,$3		/* E0 */
118	or $1,$2,$1		/* .. E1 */
119	inswl $17,6,$4		/* E0 */
120	or $1,$3,$1		/* .. E1 */
121	or $1,$4,$17		/* E0 */
122	br __constant_c_memset	/* .. E1 */
123
124	.end __memsetw
125
126memset = ___memset
127__memset = ___memset
128