| lh | 9ed821d | 2023-04-07 01:36:19 -0700 | [diff] [blame^] | 1 | /* memcpy.S: optimised assembly memcpy |
| 2 | * |
| 3 | * Copyright (C) 2003, 2004 Red Hat, Inc. All Rights Reserved. |
| 4 | * Written by David Howells (dhowells@redhat.com) |
| 5 | * |
| 6 | * This library is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Library General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This library is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Library General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Library General Public |
| 17 | * License along with this library; if not, write to the Free |
| 18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 19 | */ |
| 20 | |
| 21 | #include <features.h> |
| 22 | |
| 23 | .text |
| 24 | .p2align 4 |
| 25 | |
| 26 | ############################################################################### |
| 27 | # |
| 28 | # void *memcpy(void *to, const char *from, size_t count) |
| 29 | # |
| 30 | # - NOTE: must not use any stack. exception detection performs function return |
| 31 | # to caller's fixup routine, aborting the remainder of the copy |
| 32 | # |
| 33 | ############################################################################### |
| 34 | .globl memcpy |
| 35 | .type memcpy,@function |
| 36 | memcpy: |
| 37 | or.p gr8,gr9,gr4 |
| 38 | orcc gr10,gr0,gr0,icc3 |
| 39 | or.p gr10,gr4,gr4 |
| 40 | beqlr icc3,#0 |
| 41 | |
| 42 | # optimise based on best common alignment for to, from & count |
| 43 | andicc.p gr4,#0x1f,gr0,icc0 |
| 44 | setlos #8,gr11 |
| 45 | andicc.p gr4,#0x0f,gr0,icc1 |
| 46 | beq icc0,#0,memcpy_32 |
| 47 | andicc.p gr4,#0x07,gr0,icc0 |
| 48 | beq icc1,#0,memcpy_16 |
| 49 | andicc.p gr4,#0x03,gr0,icc1 |
| 50 | beq icc0,#0,memcpy_8 |
| 51 | andicc.p gr4,#0x01,gr0,icc0 |
| 52 | beq icc1,#0,memcpy_4 |
| 53 | setlos.p #1,gr11 |
| 54 | beq icc0,#0,memcpy_2 |
| 55 | |
| 56 | # do byte by byte copy |
| 57 | sub.p gr8,gr11,gr3 |
| 58 | sub gr9,gr11,gr9 |
| 59 | 0: ldubu.p @(gr9,gr11),gr4 |
| 60 | subicc gr10,#1,gr10,icc0 |
| 61 | stbu.p gr4,@(gr3,gr11) |
| 62 | bne icc0,#2,0b |
| 63 | bralr |
| 64 | |
| 65 | # do halfword by halfword copy |
| 66 | memcpy_2: |
| 67 | setlos #2,gr11 |
| 68 | sub.p gr8,gr11,gr3 |
| 69 | sub gr9,gr11,gr9 |
| 70 | 0: lduhu.p @(gr9,gr11),gr4 |
| 71 | subicc gr10,#2,gr10,icc0 |
| 72 | sthu.p gr4,@(gr3,gr11) |
| 73 | bne icc0,#2,0b |
| 74 | bralr |
| 75 | |
| 76 | # do word by word copy |
| 77 | memcpy_4: |
| 78 | setlos #4,gr11 |
| 79 | sub.p gr8,gr11,gr3 |
| 80 | sub gr9,gr11,gr9 |
| 81 | 0: ldu.p @(gr9,gr11),gr4 |
| 82 | subicc gr10,#4,gr10,icc0 |
| 83 | stu.p gr4,@(gr3,gr11) |
| 84 | bne icc0,#2,0b |
| 85 | bralr |
| 86 | |
| 87 | # do double-word by double-word copy |
| 88 | memcpy_8: |
| 89 | sub.p gr8,gr11,gr3 |
| 90 | sub gr9,gr11,gr9 |
| 91 | 0: lddu.p @(gr9,gr11),gr4 |
| 92 | subicc gr10,#8,gr10,icc0 |
| 93 | stdu.p gr4,@(gr3,gr11) |
| 94 | bne icc0,#2,0b |
| 95 | bralr |
| 96 | |
| 97 | # do quad-word by quad-word copy |
| 98 | memcpy_16: |
| 99 | sub.p gr8,gr11,gr3 |
| 100 | sub gr9,gr11,gr9 |
| 101 | 0: lddu @(gr9,gr11),gr4 |
| 102 | lddu.p @(gr9,gr11),gr6 |
| 103 | subicc gr10,#16,gr10,icc0 |
| 104 | stdu gr4,@(gr3,gr11) |
| 105 | stdu.p gr6,@(gr3,gr11) |
| 106 | bne icc0,#2,0b |
| 107 | bralr |
| 108 | |
| 109 | # do eight-word by eight-word copy |
| 110 | memcpy_32: |
| 111 | sub.p gr8,gr11,gr3 |
| 112 | sub gr9,gr11,gr9 |
| 113 | 0: lddu @(gr9,gr11),gr4 |
| 114 | lddu @(gr9,gr11),gr6 |
| 115 | lddu @(gr9,gr11),gr12 |
| 116 | lddu.p @(gr9,gr11),gr14 |
| 117 | subicc gr10,#32,gr10,icc0 |
| 118 | stdu gr4,@(gr3,gr11) |
| 119 | stdu gr6,@(gr3,gr11) |
| 120 | stdu gr12,@(gr3,gr11) |
| 121 | stdu.p gr14,@(gr3,gr11) |
| 122 | bne icc0,#2,0b |
| 123 | bralr |
| 124 | |
| 125 | .size memcpy, .-memcpy |
| 126 | |
| 127 | libc_hidden_def(memcpy) |