1 /* 2 * linux/arch/alpha/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 __constant_c_memset 22 .ent __memset 23 .align 5 24 __memset: 25 .frame $30,0,$26,0 26 .prologue 0 27 28 zapnot $17,1,$17 /* E0 */ 29 sll $17,8,$1 /* E1 (p-c latency, next cycle) */ 30 bis $17,$1,$17 /* E0 (p-c latency, next cycle) */ 31 sll $17,16,$1 /* E1 (p-c latency, next cycle) */ 32 33 bis $17,$1,$17 /* E0 (p-c latency, next cycle) */ 34 sll $17,32,$1 /* E1 (p-c latency, next cycle) */ 35 bis $17,$1,$17 /* E0 (p-c latency, next cycle) */ 36 ldq_u $31,0($30) /* .. E1 */ 37 38 .align 5 39 __constant_c_memset: 40 addq $18,$16,$6 /* E0 */ 41 bis $16,$16,$0 /* .. E1 */ 42 xor $16,$6,$1 /* E0 */ 43 ble $18,end /* .. E1 */ 44 45 bic $1,7,$1 /* E0 */ 46 beq $1,within_one_quad /* .. E1 (note EV5 zero-latency forwarding) */ 47 and $16,7,$3 /* E0 */ 48 beq $3,aligned /* .. E1 (note EV5 zero-latency forwarding) */ 49 50 ldq_u $4,0($16) /* E0 */ 51 bis $16,$16,$5 /* .. E1 */ 52 insql $17,$16,$2 /* E0 */ 53 subq $3,8,$3 /* .. E1 */ 54 55 addq $18,$3,$18 /* E0 $18 is new count ($3 is negative) */ 56 mskql $4,$16,$4 /* .. E1 (and possible load stall) */ 57 subq $16,$3,$16 /* E0 $16 is new aligned destination */ 58 bis $2,$4,$1 /* .. E1 */ 59 60 bis $31,$31,$31 /* E0 */ 61 ldq_u $31,0($30) /* .. E1 */ 62 stq_u $1,0($5) /* E0 */ 63 bis $31,$31,$31 /* .. E1 */ 64 65 .align 4 66 aligned: 67 sra $18,3,$3 /* E0 */ 68 and $18,7,$18 /* .. E1 */ 69 bis $16,$16,$5 /* E0 */ 70 beq $3,no_quad /* .. E1 */ 71 72 .align 3 73 loop: 74 stq $17,0($5) /* E0 */ 75 subq $3,1,$3 /* .. E1 */ 76 addq $5,8,$5 /* E0 */ 77 bne $3,loop /* .. E1 */ 78 79 no_quad: 80 bis $31,$31,$31 /* E0 */ 81 beq $18,end /* .. E1 */ 82 ldq $7,0($5) /* E0 */ 83 mskqh $7,$6,$2 /* .. E1 (and load stall) */ 84 85 insqh $17,$6,$4 /* E0 */ 86 bis $2,$4,$1 /* .. E1 */ 87 stq $1,0($5) /* E0 */ 88 ret $31,($26),1 /* .. E1 */ 89 90 .align 3 91 within_one_quad: 92 ldq_u $1,0($16) /* E0 */ 93 insql $17,$16,$2 /* E1 */ 94 mskql $1,$16,$4 /* E0 (after load stall) */ 95 bis $2,$4,$2 /* E0 */ 96 97 mskql $2,$6,$4 /* E0 */ 98 mskqh $1,$6,$2 /* .. E1 */ 99 bis $2,$4,$1 /* E0 */ 100 stq_u $1,0($16) /* E0 */ 101 102 end: 103 ret $31,($26),1 /* E1 */ 104 .end __memset 105 106 .weakext memset, __memset