1// Inferno's libkern/memmove-386.s 2// https://bitbucket.org/inferno-os/inferno-os/src/master/libkern/memmove-386.s 3// 4// Copyright © 1994-1999 Lucent Technologies Inc. All rights reserved. 5// Revisions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com). All rights reserved. 6// Portions Copyright 2009 The Go Authors. All rights reserved. 7// 8// Permission is hereby granted, free of charge, to any person obtaining a copy 9// of this software and associated documentation files (the "Software"), to deal 10// in the Software without restriction, including without limitation the rights 11// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 12// copies of the Software, and to permit persons to whom the Software is 13// furnished to do so, subject to the following conditions: 14// 15// The above copyright notice and this permission notice shall be included in 16// all copies or substantial portions of the Software. 17// 18// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 21// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 22// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 23// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 24// THE SOFTWARE. 25 26//go:build !plan9 27 28#include "go_asm.h" 29#include "textflag.h" 30 31// See memmove Go doc for important implementation constraints. 32 33// func memmove(to, from unsafe.Pointer, n uintptr) 34TEXT runtime·memmove(SB), NOSPLIT, $0-12 35 MOVL to+0(FP), DI 36 MOVL from+4(FP), SI 37 MOVL n+8(FP), BX 38 39 // REP instructions have a high startup cost, so we handle small sizes 40 // with some straightline code. The REP MOVSL instruction is really fast 41 // for large sizes. The cutover is approximately 1K. We implement up to 42 // 128 because that is the maximum SSE register load (loading all data 43 // into registers lets us ignore copy direction). 44tail: 45 // BSR+branch table make almost all memmove/memclr benchmarks worse. Not worth doing. 46 TESTL BX, BX 47 JEQ move_0 48 CMPL BX, $2 49 JBE move_1or2 50 CMPL BX, $4 51 JB move_3 52 JE move_4 53 CMPL BX, $8 54 JBE move_5through8 55 CMPL BX, $16 56 JBE move_9through16 57#ifdef GO386_softfloat 58 JMP nosse2 59#endif 60 CMPL BX, $32 61 JBE move_17through32 62 CMPL BX, $64 63 JBE move_33through64 64 CMPL BX, $128 65 JBE move_65through128 66 67nosse2: 68/* 69 * check and set for backwards 70 */ 71 CMPL SI, DI 72 JLS back 73 74/* 75 * forward copy loop 76 */ 77forward: 78 // If REP MOVSB isn't fast, don't use it 79 CMPB internal∕cpu·X86+const_offsetX86HasERMS(SB), $1 // enhanced REP MOVSB/STOSB 80 JNE fwdBy4 81 82 // Check alignment 83 MOVL SI, AX 84 ORL DI, AX 85 TESTL $3, AX 86 JEQ fwdBy4 87 88 // Do 1 byte at a time 89 MOVL BX, CX 90 REP; MOVSB 91 RET 92 93fwdBy4: 94 // Do 4 bytes at a time 95 MOVL BX, CX 96 SHRL $2, CX 97 ANDL $3, BX 98 REP; MOVSL 99 JMP tail 100 101/* 102 * check overlap 103 */ 104back: 105 MOVL SI, CX 106 ADDL BX, CX 107 CMPL CX, DI 108 JLS forward 109/* 110 * whole thing backwards has 111 * adjusted addresses 112 */ 113 114 ADDL BX, DI 115 ADDL BX, SI 116 STD 117 118/* 119 * copy 120 */ 121 MOVL BX, CX 122 SHRL $2, CX 123 ANDL $3, BX 124 125 SUBL $4, DI 126 SUBL $4, SI 127 REP; MOVSL 128 129 CLD 130 ADDL $4, DI 131 ADDL $4, SI 132 SUBL BX, DI 133 SUBL BX, SI 134 JMP tail 135 136move_1or2: 137 MOVB (SI), AX 138 MOVB -1(SI)(BX*1), CX 139 MOVB AX, (DI) 140 MOVB CX, -1(DI)(BX*1) 141 RET 142move_0: 143 RET 144move_3: 145 MOVW (SI), AX 146 MOVB 2(SI), CX 147 MOVW AX, (DI) 148 MOVB CX, 2(DI) 149 RET 150move_4: 151 // We need a separate case for 4 to make sure we write pointers atomically. 152 MOVL (SI), AX 153 MOVL AX, (DI) 154 RET 155move_5through8: 156 MOVL (SI), AX 157 MOVL -4(SI)(BX*1), CX 158 MOVL AX, (DI) 159 MOVL CX, -4(DI)(BX*1) 160 RET 161move_9through16: 162 MOVL (SI), AX 163 MOVL 4(SI), CX 164 MOVL -8(SI)(BX*1), DX 165 MOVL -4(SI)(BX*1), BP 166 MOVL AX, (DI) 167 MOVL CX, 4(DI) 168 MOVL DX, -8(DI)(BX*1) 169 MOVL BP, -4(DI)(BX*1) 170 RET 171move_17through32: 172 MOVOU (SI), X0 173 MOVOU -16(SI)(BX*1), X1 174 MOVOU X0, (DI) 175 MOVOU X1, -16(DI)(BX*1) 176 RET 177move_33through64: 178 MOVOU (SI), X0 179 MOVOU 16(SI), X1 180 MOVOU -32(SI)(BX*1), X2 181 MOVOU -16(SI)(BX*1), X3 182 MOVOU X0, (DI) 183 MOVOU X1, 16(DI) 184 MOVOU X2, -32(DI)(BX*1) 185 MOVOU X3, -16(DI)(BX*1) 186 RET 187move_65through128: 188 MOVOU (SI), X0 189 MOVOU 16(SI), X1 190 MOVOU 32(SI), X2 191 MOVOU 48(SI), X3 192 MOVOU -64(SI)(BX*1), X4 193 MOVOU -48(SI)(BX*1), X5 194 MOVOU -32(SI)(BX*1), X6 195 MOVOU -16(SI)(BX*1), X7 196 MOVOU X0, (DI) 197 MOVOU X1, 16(DI) 198 MOVOU X2, 32(DI) 199 MOVOU X3, 48(DI) 200 MOVOU X4, -64(DI)(BX*1) 201 MOVOU X5, -48(DI)(BX*1) 202 MOVOU X6, -32(DI)(BX*1) 203 MOVOU X7, -16(DI)(BX*1) 204 RET 205