2 * linux/arch/alpha/memset.S
4 * This is an efficient (and small) implementation of the C library "memset()"
5 * function for the alpha.
7 * (C) Copyright 1996 Linus Torvalds
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.
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..)
22 .globl __constant_c_memset
29 and $17,255,$1 /* E1 */
30 insbl $17,1,$17 /* .. E0 */
31 bis $17,$1,$17 /* E0 (p-c latency, next cycle) */
32 sll $17,16,$1 /* E1 (p-c latency, next cycle) */
34 bis $17,$1,$17 /* E0 (p-c latency, next cycle) */
35 sll $17,32,$1 /* E1 (p-c latency, next cycle) */
36 bis $17,$1,$17 /* E0 (p-c latency, next cycle) */
37 ldq_u $31,0($30) /* .. E1 */
41 addq $18,$16,$6 /* E0 */
42 bis $16,$16,$0 /* .. E1 */
43 xor $16,$6,$1 /* E0 */
44 ble $18,end /* .. E1 */
47 beq $1,within_one_quad /* .. E1 (note EV5 zero-latency forwarding) */
49 beq $3,aligned /* .. E1 (note EV5 zero-latency forwarding) */
51 ldq_u $4,0($16) /* E0 */
52 bis $16,$16,$5 /* .. E1 */
53 insql $17,$16,$2 /* E0 */
54 subq $3,8,$3 /* .. E1 */
56 addq $18,$3,$18 /* E0 $18 is new count ($3 is negative) */
57 mskql $4,$16,$4 /* .. E1 (and possible load stall) */
58 subq $16,$3,$16 /* E0 $16 is new aligned destination */
59 bis $2,$4,$1 /* .. E1 */
61 bis $31,$31,$31 /* E0 */
62 ldq_u $31,0($30) /* .. E1 */
63 stq_u $1,0($5) /* E0 */
64 bis $31,$31,$31 /* .. E1 */
69 and $18,7,$18 /* .. E1 */
70 bis $16,$16,$5 /* E0 */
71 beq $3,no_quad /* .. E1 */
75 stq $17,0($5) /* E0 */
76 subq $3,1,$3 /* .. E1 */
78 bne $3,loop /* .. E1 */
81 bis $31,$31,$31 /* E0 */
82 beq $18,end /* .. E1 */
84 mskqh $7,$6,$2 /* .. E1 (and load stall) */
86 insqh $17,$6,$4 /* E0 */
87 bis $2,$4,$1 /* .. E1 */
89 ret $31,($26),1 /* .. E1 */
93 ldq_u $1,0($16) /* E0 */
94 insql $17,$16,$2 /* E1 */
95 mskql $1,$16,$4 /* E0 (after load stall) */
98 mskql $2,$6,$4 /* E0 */
99 mskqh $1,$6,$2 /* .. E1 */
100 bis $2,$4,$1 /* E0 */
101 stq_u $1,0($16) /* E0 */
104 ret $31,($26),1 /* E1 */
112 inswl $17,0,$1 /* E0 */
113 inswl $17,2,$2 /* E0 */
114 inswl $17,4,$3 /* E0 */
115 or $1,$2,$1 /* .. E1 */
116 inswl $17,6,$4 /* E0 */
117 or $1,$3,$1 /* .. E1 */
118 or $1,$4,$17 /* E0 */
119 br __constant_c_memset /* .. E1 */
124 .weak memset; memset = __memset
126 .weakext memset, __memset