bcopy.S revision 1.2.40.3
1 /* $NetBSD: bcopy.S,v 1.2.40.3 2009/08/20 10:03:43 matt Exp $ */ 2 3 /* 4 * Mach Operating System 5 * Copyright (c) 1993 Carnegie Mellon University 6 * All Rights Reserved. 7 * 8 * Permission to use, copy, modify and distribute this software and its 9 * documentation is hereby granted, provided that both the copyright 10 * notice and this permission notice appear in all copies of the 11 * software, derivative works or modified versions, and any portions 12 * thereof, and that both notices appear in supporting documentation. 13 * 14 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 15 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR 16 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 17 * 18 * Carnegie Mellon requests users of this software to return to 19 * 20 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU 21 * School of Computer Science 22 * Carnegie Mellon University 23 * Pittsburgh PA 15213-3890 24 * 25 * any improvements or extensions that they make and grant Carnegie Mellon 26 * the rights to redistribute these changes. 27 */ 28 29 /* 30 * File: mips_bcopy.s 31 * Author: Chris Maeda 32 * Date: June 1993 33 * 34 * Fast copy routine. Derived from aligned_block_copy. 35 */ 36 37 38 #include <mips/asm.h> 39 #ifndef _LOCORE 40 #define _LOCORE /* XXX not really, just assembly-code source */ 41 #endif 42 #include <machine/endian.h> 43 44 45 #if defined(LIBC_SCCS) && !defined(lint) 46 /* RCSID("from: @(#)mips_bcopy.s 2.2 CMU 18/06/93") */ 47 RCSID("$NetBSD: bcopy.S,v 1.2.40.3 2009/08/20 10:03:43 matt Exp $") 48 #endif /* LIBC_SCCS and not lint */ 49 50 /* 51 * bcopy(caddr_t src, caddr_t dst, unsigned int len) 52 * 53 * a0 src address 54 * a1 dst address 55 * a2 length 56 */ 57 58 #if defined(MEMCOPY) || defined(MEMMOVE) 59 #ifdef MEMCOPY 60 #define FUNCTION memcpy 61 #else 62 #define FUNCTION memmove 63 #endif 64 #define SRCREG a1 65 #define DSTREG a0 66 #else 67 #define FUNCTION bcopy 68 #define SRCREG a0 69 #define DSTREG a1 70 #endif 71 72 #define SIZEREG a2 73 74 LEAF(FUNCTION) 75 .set noat 76 .set noreorder 77 78 #if defined(MEMCOPY) || defined(MEMMOVE) 79 /* set up return value, while we still can */ 80 move v0,DSTREG 81 #endif 82 /* 83 * Make sure we can copy forwards. 84 */ 85 sltu t0,SRCREG,DSTREG # t0 == SRCREG < DSTREG 86 bne t0,zero,6f # copy backwards 87 88 /* 89 * There are four alignment cases (with frequency) 90 * (Based on measurements taken with a DECstation 5000/200 91 * inside a Mach kernel.) 92 * 93 * aligned -> aligned (mostly) 94 * unaligned -> aligned (sometimes) 95 * aligned,unaligned -> unaligned (almost never) 96 * 97 * Note that we could add another case that checks if 98 * the destination and source are unaligned but the 99 * copy is alignable. eg if src and dest are both 100 * on a halfword boundary. 101 */ 102 andi t1,DSTREG,(SZREG-1) # get last bits of dest 103 bne t1,zero,3f # dest unaligned 104 andi t0,SRCREG,(SZREG-1) # get last bits of src 105 bne t0,zero,5f 106 107 /* 108 * Forward aligned->aligned copy, 8 words at a time. 109 */ 110 98: 111 li AT,-(SZREG*8) 112 and t0,SIZEREG,AT # count truncated to multiples 113 PTR_ADDU a3,SRCREG,t0 # run fast loop up to this addr 114 sltu AT,SRCREG,a3 # any work to do? 115 beq AT,zero,2f 116 PTR_SUBU SIZEREG,t0 117 118 /* 119 * loop body 120 */ 121 1: # cp 122 REG_L t3,(0*SZREG)(SRCREG) 123 REG_L v1,(1*SZREG)(SRCREG) 124 REG_L t0,(2*SZREG)(SRCREG) 125 REG_L t1,(3*SZREG)(SRCREG) 126 PTR_ADDU SRCREG,SZREG*8 127 REG_S t3,(0*SZREG)(DSTREG) 128 REG_S v1,(1*SZREG)(DSTREG) 129 REG_S t0,(2*SZREG)(DSTREG) 130 REG_S t1,(3*SZREG)(DSTREG) 131 REG_L t1,(-1*SZREG)(SRCREG) 132 REG_L t0,(-2*SZREG)(SRCREG) 133 REG_L v1,(-3*SZREG)(SRCREG) 134 REG_L t3,(-4*SZREG)(SRCREG) 135 PTR_ADDU DSTREG,SZREG*8 136 REG_S t1,(-1*SZREG)(DSTREG) 137 REG_S t0,(-2*SZREG)(DSTREG) 138 REG_S v1,(-3*SZREG)(DSTREG) 139 bne SRCREG,a3,1b 140 REG_S t3,(-4*SZREG)(DSTREG) 141 142 /* 143 * Copy a word at a time, no loop unrolling. 144 */ 145 2: # wordcopy 146 andi t2,SIZEREG,(SZREG-1) # get byte count / SZREG 147 PTR_SUBU t2,SIZEREG,t2 # t2 = words to copy * SZREG 148 beq t2,zero,3f 149 PTR_ADDU t0,SRCREG,t2 # stop at t0 150 PTR_SUBU SIZEREG,SIZEREG,t2 151 1: 152 REG_L t3,0(SRCREG) 153 PTR_ADDU SRCREG,SZREG 154 REG_S t3,0(DSTREG) 155 bne SRCREG,t0,1b 156 PTR_ADDU DSTREG,SZREG 157 158 3: # bytecopy 159 beq SIZEREG,zero,4f # nothing left to do? 160 nop 161 1: 162 lb t3,0(SRCREG) 163 PTR_ADDU SRCREG,1 164 sb t3,0(DSTREG) 165 PTR_SUBU SIZEREG,1 166 bgtz SIZEREG,1b 167 PTR_ADDU DSTREG,1 168 169 4: # copydone 170 j ra 171 nop 172 173 /* 174 * Copy from unaligned source to aligned dest. 175 */ 176 5: # destaligned 177 andi t0,SIZEREG,(SZREG-1) # t0 = bytecount mod SZREG 178 PTR_SUBU a3,SIZEREG,t0 # number of words to transfer 179 beq a3,zero,3b 180 nop 181 move SIZEREG,t0 # this many to do after we are done 182 PTR_ADDU a3,SRCREG,a3 # stop point 183 184 1: 185 REG_LHI t3,0(SRCREG) 186 REG_LLO t3,SZREG-1(SRCREG) 187 PTR_ADDI SRCREG,SZREG 188 REG_S t3,0(DSTREG) 189 bne SRCREG,a3,1b 190 PTR_ADDI DSTREG,SZREG 191 192 b 3b 193 nop 194 195 6: # backcopy -- based on above 196 PTR_ADDU SRCREG,SIZEREG 197 PTR_ADDU DSTREG,SIZEREG 198 andi t1,DSTREG,SZREG-1 # get last 3 bits of dest 199 bne t1,zero,3f 200 andi t0,SRCREG,SZREG-1 # get last 3 bits of src 201 bne t0,zero,5f 202 203 /* 204 * Forward aligned->aligned copy, 8*4 bytes at a time. 205 */ 206 li AT,(-8*SZREG) 207 and t0,SIZEREG,AT # count truncated to multiple of 32 208 beq t0,zero,2f # any work to do? 209 PTR_SUBU SIZEREG,t0 210 PTR_SUBU a3,SRCREG,t0 211 212 /* 213 * loop body 214 */ 215 1: # cp 216 REG_L t3,(-4*SZREG)(SRCREG) 217 REG_L v1,(-3*SZREG)(SRCREG) 218 REG_L t0,(-2*SZREG)(SRCREG) 219 REG_L t1,(-1*SZREG)(SRCREG) 220 PTR_SUBU SRCREG,8*SZREG 221 REG_S t3,(-4*SZREG)(DSTREG) 222 REG_S v1,(-3*SZREG)(DSTREG) 223 REG_S t0,(-2*SZREG)(DSTREG) 224 REG_S t1,(-1*SZREG)(DSTREG) 225 REG_L t1,(3*SZREG)(SRCREG) 226 REG_L t0,(2*SZREG)(SRCREG) 227 REG_L v1,(1*SZREG)(SRCREG) 228 REG_L t3,(0*SZREG)(SRCREG) 229 PTR_SUBU DSTREG,8*SZREG 230 REG_S t1,(3*SZREG)(DSTREG) 231 REG_S t0,(2*SZREG)(DSTREG) 232 REG_S v1,(1*SZREG)(DSTREG) 233 bne SRCREG,a3,1b 234 REG_S t3,(0*SZREG)(DSTREG) 235 236 /* 237 * Copy a word at a time, no loop unrolling. 238 */ 239 2: # wordcopy 240 andi t2,SIZEREG,SZREG-1 # get byte count / 4 241 PTR_SUBU t2,SIZEREG,t2 # t2 = number of words to copy 242 beq t2,zero,3f 243 PTR_SUBU t0,SRCREG,t2 # stop at t0 244 PTR_SUBU SIZEREG,SIZEREG,t2 245 1: 246 REG_L t3,-SZREG(SRCREG) 247 PTR_SUBU SRCREG,SZREG 248 REG_S t3,-SZREG(DSTREG) 249 bne SRCREG,t0,1b 250 PTR_SUBU DSTREG,SZREG 251 252 3: # bytecopy 253 beq SIZEREG,zero,4f # nothing left to do? 254 nop 255 1: 256 lb t3,-1(SRCREG) 257 PTR_SUBU SRCREG,1 258 sb t3,-1(DSTREG) 259 PTR_SUBU SIZEREG,1 260 bgtz SIZEREG,1b 261 PTR_SUBU DSTREG,1 262 263 4: # copydone 264 j ra 265 nop 266 267 /* 268 * Copy from unaligned source to aligned dest. 269 */ 270 5: # destaligned 271 andi t0,SIZEREG,SZREG-1 # t0 = bytecount mod 4 272 PTR_SUBU a3,SIZEREG,t0 # number of words to transfer 273 beq a3,zero,3b 274 nop 275 move SIZEREG,t0 # this many to do after we are done 276 PTR_SUBU a3,SRCREG,a3 # stop point 277 278 1: 279 REG_LHI t3,-SZREG(SRCREG) 280 REG_LLO t3,-1(SRCREG) 281 PTR_SUBU SRCREG,SZREG 282 REG_S t3,-SZREG(DSTREG) 283 bne SRCREG,a3,1b 284 PTR_SUBU DSTREG,SZREG 285 286 b 3b 287 nop 288 289 .set reorder 290 .set at 291 END(FUNCTION) 292
Indexes created Sat Sep 27 14:09:57 GMT 2025