| Alexei Fedorov | 2381d2e | 2020-09-01 15:38:32 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2020, Arm Limited. All rights reserved. |
| 3 | * |
| 4 | * SPDX-License-Identifier: BSD-3-Clause |
| 5 | */ |
| 6 | |
| 7 | #include <asm_macros.S> |
| 8 | |
| 9 | .syntax unified |
| 10 | .global memset |
| 11 | |
| 12 | /* ----------------------------------------------------------------------- |
| 13 | * void *memset(void *dst, int val, size_t count) |
| 14 | * |
| 15 | * Copy the value of 'val' (converted to an unsigned char) into |
| 16 | * each of the first 'count' characters of the object pointed to by 'dst'. |
| 17 | * |
| 18 | * Returns the value of 'dst'. |
| 19 | * ----------------------------------------------------------------------- |
| 20 | */ |
| 21 | func memset |
| 22 | mov r12, r0 /* keep r0 */ |
| 23 | tst r0, #3 |
| 24 | beq aligned /* 4-bytes aligned */ |
| 25 | |
| 26 | /* Unaligned 'dst' */ |
| 27 | unaligned: |
| 28 | subs r2, r2, #1 |
| 29 | strbhs r1, [r12], #1 |
| 30 | bxls lr /* return if 0 */ |
| 31 | tst r12, #3 |
| 32 | bne unaligned /* continue while unaligned */ |
| 33 | |
| 34 | /* 4-bytes aligned */ |
| 35 | aligned:bfi r1, r1, #8, #8 /* propagate 'val' */ |
| 36 | bfi r1, r1, #16, #16 |
| 37 | |
| 38 | mov r3, r1 |
| 39 | |
| 40 | cmp r2, #16 |
| 41 | blo less_16 /* < 16 */ |
| 42 | |
| 43 | push {r4, lr} |
| 44 | mov r4, r1 |
| 45 | mov lr, r1 |
| 46 | |
| 47 | write_32: |
| 48 | subs r2, r2, #32 |
| 49 | stmiahs r12!, {r1, r3, r4, lr} |
| 50 | stmiahs r12!, {r1, r3, r4, lr} |
| 51 | bhi write_32 /* write 32 bytes in a loop */ |
| 52 | popeq {r4, pc} /* return if 0 */ |
| 53 | lsls r2, r2, #28 /* C = r2[4]; N = r2[3]; Z = r2[3:0] */ |
| 54 | stmiacs r12!, {r1, r3, r4, lr} /* write 16 bytes */ |
| 55 | popeq {r4, pc} /* return if 16 */ |
| 56 | stmiami r12!, {r1, r3} /* write 8 bytes */ |
| 57 | lsls r2, r2, #2 /* C = r2[2]; N = r2[1]; Z = r2[1:0] */ |
| 58 | strcs r1, [r12], #4 /* write 4 bytes */ |
| 59 | popeq {r4, pc} /* return if 8 or 4 */ |
| 60 | strhmi r1, [r12], #2 /* write 2 bytes */ |
| 61 | lsls r2, r2, #1 /* N = Z = r2[0] */ |
| 62 | strbmi r1, [r12] /* write 1 byte */ |
| 63 | pop {r4, pc} |
| 64 | |
| 65 | less_16:lsls r2, r2, #29 /* C = r2[3]; N = r2[2]; Z = r2[2:0] */ |
| 66 | stmiacs r12!, {r1, r3} /* write 8 bytes */ |
| 67 | bxeq lr /* return if 8 */ |
| 68 | strmi r1, [r12], #4 /* write 4 bytes */ |
| 69 | lsls r2, r2, #2 /* C = r2[1]; N = Z = r2[0] */ |
| 70 | strhcs r1, [r12], #2 /* write 2 bytes */ |
| 71 | strbmi r1, [r12] /* write 1 byte */ |
| 72 | bx lr |
| 73 | |
| 74 | endfunc memset |