| Matthias Weisser | d8834a1 | 2011-03-10 21:36:32 +0000 | [diff] [blame] | 1 | /* |
| 2 | * linux/arch/arm/lib/memset.S |
| 3 | * |
| 4 | * Copyright (C) 1995-2000 Russell King |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License version 2 as |
| 8 | * published by the Free Software Foundation. |
| 9 | * |
| 10 | * ASM optimised string functions |
| 11 | */ |
| 12 | #include <asm/assembler.h> |
| 13 | |
| 14 | .text |
| 15 | .align 5 |
| 16 | .word 0 |
| 17 | |
| 18 | 1: subs r2, r2, #4 @ 1 do we have enough |
| 19 | blt 5f @ 1 bytes to align with? |
| 20 | cmp r3, #2 @ 1 |
| 21 | strltb r1, [r0], #1 @ 1 |
| 22 | strleb r1, [r0], #1 @ 1 |
| 23 | strb r1, [r0], #1 @ 1 |
| 24 | add r2, r2, r3 @ 1 (r2 = r2 - (4 - r3)) |
| 25 | /* |
| 26 | * The pointer is now aligned and the length is adjusted. Try doing the |
| 27 | * memset again. |
| 28 | */ |
| 29 | |
| 30 | .globl memset |
| 31 | memset: |
| 32 | ands r3, r0, #3 @ 1 unaligned? |
| 33 | bne 1b @ 1 |
| 34 | /* |
| 35 | * we know that the pointer in r0 is aligned to a word boundary. |
| 36 | */ |
| 37 | orr r1, r1, r1, lsl #8 |
| 38 | orr r1, r1, r1, lsl #16 |
| 39 | mov r3, r1 |
| 40 | cmp r2, #16 |
| 41 | blt 4f |
| 42 | |
| 43 | #if ! CALGN(1)+0 |
| 44 | |
| 45 | /* |
| 46 | * We need an extra register for this loop - save the return address and |
| 47 | * use the LR |
| 48 | */ |
| 49 | str lr, [sp, #-4]! |
| 50 | mov ip, r1 |
| 51 | mov lr, r1 |
| 52 | |
| 53 | 2: subs r2, r2, #64 |
| 54 | stmgeia r0!, {r1, r3, ip, lr} @ 64 bytes at a time. |
| 55 | stmgeia r0!, {r1, r3, ip, lr} |
| 56 | stmgeia r0!, {r1, r3, ip, lr} |
| 57 | stmgeia r0!, {r1, r3, ip, lr} |
| 58 | bgt 2b |
| 59 | ldmeqfd sp!, {pc} @ Now <64 bytes to go. |
| 60 | /* |
| 61 | * No need to correct the count; we're only testing bits from now on |
| 62 | */ |
| 63 | tst r2, #32 |
| 64 | stmneia r0!, {r1, r3, ip, lr} |
| 65 | stmneia r0!, {r1, r3, ip, lr} |
| 66 | tst r2, #16 |
| 67 | stmneia r0!, {r1, r3, ip, lr} |
| 68 | ldr lr, [sp], #4 |
| 69 | |
| 70 | #else |
| 71 | |
| 72 | /* |
| 73 | * This version aligns the destination pointer in order to write |
| 74 | * whole cache lines at once. |
| 75 | */ |
| 76 | |
| 77 | stmfd sp!, {r4-r7, lr} |
| 78 | mov r4, r1 |
| 79 | mov r5, r1 |
| 80 | mov r6, r1 |
| 81 | mov r7, r1 |
| 82 | mov ip, r1 |
| 83 | mov lr, r1 |
| 84 | |
| 85 | cmp r2, #96 |
| 86 | tstgt r0, #31 |
| 87 | ble 3f |
| 88 | |
| 89 | and ip, r0, #31 |
| 90 | rsb ip, ip, #32 |
| 91 | sub r2, r2, ip |
| 92 | movs ip, ip, lsl #(32 - 4) |
| 93 | stmcsia r0!, {r4, r5, r6, r7} |
| 94 | stmmiia r0!, {r4, r5} |
| 95 | tst ip, #(1 << 30) |
| 96 | mov ip, r1 |
| 97 | strne r1, [r0], #4 |
| 98 | |
| 99 | 3: subs r2, r2, #64 |
| 100 | stmgeia r0!, {r1, r3-r7, ip, lr} |
| 101 | stmgeia r0!, {r1, r3-r7, ip, lr} |
| 102 | bgt 3b |
| 103 | ldmeqfd sp!, {r4-r7, pc} |
| 104 | |
| 105 | tst r2, #32 |
| 106 | stmneia r0!, {r1, r3-r7, ip, lr} |
| 107 | tst r2, #16 |
| 108 | stmneia r0!, {r4-r7} |
| 109 | ldmfd sp!, {r4-r7, lr} |
| 110 | |
| 111 | #endif |
| 112 | |
| 113 | 4: tst r2, #8 |
| 114 | stmneia r0!, {r1, r3} |
| 115 | tst r2, #4 |
| 116 | strne r1, [r0], #4 |
| 117 | /* |
| 118 | * When we get here, we've got less than 4 bytes to zero. We |
| 119 | * may have an unaligned pointer as well. |
| 120 | */ |
| 121 | 5: tst r2, #2 |
| 122 | strneb r1, [r0], #1 |
| 123 | strneb r1, [r0], #1 |
| 124 | tst r2, #1 |
| 125 | strneb r1, [r0], #1 |
| 126 | mov pc, lr |