1 /* SPDX-License-Identifier: GPL-2.0-only */ !! 1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* !! 2 /* memcpy.S: Sparc optimized memcpy and memmove code
3 * linux/arch/arm/lib/memcpy.S !! 3 * Hand optimized from GNU libc's memcpy and memmove
4 * !! 4 * Copyright (C) 1991,1996 Free Software Foundation
5 * Author: Nicolas Pitre !! 5 * Copyright (C) 1995 Linus Torvalds (Linus.Torvalds@helsinki.fi)
6 * Created: Sep 28, 2005 !! 6 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
7 * Copyright: MontaVista Software, Inc. !! 7 * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be)
>> 8 * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
8 */ 9 */
9 10
10 #include <linux/linkage.h> !! 11 #include <asm/export.h>
11 #include <asm/assembler.h> !! 12 #define FUNC(x) \
12 #include <asm/unwind.h> !! 13 .globl x; \
13 !! 14 .type x,@function; \
14 #define LDR1W_SHIFT 0 !! 15 .align 4; \
15 #define STR1W_SHIFT 0 !! 16 x:
16 !! 17
17 .macro ldr1w ptr reg abort !! 18 /* Both these macros have to start with exactly the same insn */
18 W(ldr) \reg, [\ptr], #4 !! 19 #define MOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \
19 .endm !! 20 ldd [%src + (offset) + 0x00], %t0; \
20 !! 21 ldd [%src + (offset) + 0x08], %t2; \
21 .macro ldr4w ptr reg1 reg2 reg3 reg4 a !! 22 ldd [%src + (offset) + 0x10], %t4; \
22 ldmia \ptr!, {\reg1, \reg2, \reg3, \re !! 23 ldd [%src + (offset) + 0x18], %t6; \
23 .endm !! 24 st %t0, [%dst + (offset) + 0x00]; \
24 !! 25 st %t1, [%dst + (offset) + 0x04]; \
25 .macro ldr8w ptr reg1 reg2 reg3 reg4 r !! 26 st %t2, [%dst + (offset) + 0x08]; \
26 ldmia \ptr!, {\reg1, \reg2, \reg3, \re !! 27 st %t3, [%dst + (offset) + 0x0c]; \
27 .endm !! 28 st %t4, [%dst + (offset) + 0x10]; \
28 !! 29 st %t5, [%dst + (offset) + 0x14]; \
29 .macro ldr1b ptr reg cond=al abort !! 30 st %t6, [%dst + (offset) + 0x18]; \
30 ldrb\cond \reg, [\ptr], #1 !! 31 st %t7, [%dst + (offset) + 0x1c];
31 .endm !! 32
32 !! 33 #define MOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \
33 .macro str1w ptr reg abort !! 34 ldd [%src + (offset) + 0x00], %t0; \
34 W(str) \reg, [\ptr], #4 !! 35 ldd [%src + (offset) + 0x08], %t2; \
35 .endm !! 36 ldd [%src + (offset) + 0x10], %t4; \
36 !! 37 ldd [%src + (offset) + 0x18], %t6; \
37 .macro str8w ptr reg1 reg2 reg3 reg4 r !! 38 std %t0, [%dst + (offset) + 0x00]; \
38 stmia \ptr!, {\reg1, \reg2, \reg3, \re !! 39 std %t2, [%dst + (offset) + 0x08]; \
39 .endm !! 40 std %t4, [%dst + (offset) + 0x10]; \
40 !! 41 std %t6, [%dst + (offset) + 0x18];
41 .macro str1b ptr reg cond=al abort !! 42
42 strb\cond \reg, [\ptr], #1 !! 43 #define MOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \
43 .endm !! 44 ldd [%src - (offset) - 0x10], %t0; \
44 !! 45 ldd [%src - (offset) - 0x08], %t2; \
45 .macro enter regs:vararg !! 46 st %t0, [%dst - (offset) - 0x10]; \
46 UNWIND( .save {r0, \regs} ) !! 47 st %t1, [%dst - (offset) - 0x0c]; \
47 stmdb sp!, {r0, \regs} !! 48 st %t2, [%dst - (offset) - 0x08]; \
48 .endm !! 49 st %t3, [%dst - (offset) - 0x04];
49 !! 50
50 .macro exit regs:vararg !! 51 #define MOVE_LASTALIGNCHUNK(src, dst, offset, t0, t1, t2, t3) \
51 ldmfd sp!, {r0, \regs} !! 52 ldd [%src - (offset) - 0x10], %t0; \
52 .endm !! 53 ldd [%src - (offset) - 0x08], %t2; \
>> 54 std %t0, [%dst - (offset) - 0x10]; \
>> 55 std %t2, [%dst - (offset) - 0x08];
>> 56
>> 57 #define MOVE_SHORTCHUNK(src, dst, offset, t0, t1) \
>> 58 ldub [%src - (offset) - 0x02], %t0; \
>> 59 ldub [%src - (offset) - 0x01], %t1; \
>> 60 stb %t0, [%dst - (offset) - 0x02]; \
>> 61 stb %t1, [%dst - (offset) - 0x01];
53 62
54 .text 63 .text
>> 64 .align 4
55 65
56 /* Prototype: void *memcpy(void *dest, const v !! 66 FUNC(memmove)
57 !! 67 EXPORT_SYMBOL(memmove)
58 ENTRY(__memcpy) !! 68 cmp %o0, %o1
59 ENTRY(mmiocpy) !! 69 mov %o0, %g7
60 WEAK(memcpy) !! 70 bleu 9f
61 !! 71 sub %o0, %o1, %o4
62 #include "copy_template.S" !! 72
>> 73 add %o1, %o2, %o3
>> 74 cmp %o3, %o0
>> 75 bleu 0f
>> 76 andcc %o4, 3, %o5
>> 77
>> 78 add %o1, %o2, %o1
>> 79 add %o0, %o2, %o0
>> 80 sub %o1, 1, %o1
>> 81 sub %o0, 1, %o0
>> 82
>> 83 1: /* reverse_bytes */
>> 84
>> 85 ldub [%o1], %o4
>> 86 subcc %o2, 1, %o2
>> 87 stb %o4, [%o0]
>> 88 sub %o1, 1, %o1
>> 89 bne 1b
>> 90 sub %o0, 1, %o0
>> 91
>> 92 retl
>> 93 mov %g7, %o0
>> 94
>> 95 /* NOTE: This code is executed just for the cases,
>> 96 where %src (=%o1) & 3 is != 0.
>> 97 We need to align it to 4. So, for (%src & 3)
>> 98 1 we need to do ldub,lduh
>> 99 2 lduh
>> 100 3 just ldub
>> 101 so even if it looks weird, the branches
>> 102 are correct here. -jj
>> 103 */
>> 104 78: /* dword_align */
63 105
64 ENDPROC(memcpy) !! 106 andcc %o1, 1, %g0
65 ENDPROC(mmiocpy) !! 107 be 4f
66 ENDPROC(__memcpy) !! 108 andcc %o1, 2, %g0
>> 109
>> 110 ldub [%o1], %g2
>> 111 add %o1, 1, %o1
>> 112 stb %g2, [%o0]
>> 113 sub %o2, 1, %o2
>> 114 bne 3f
>> 115 add %o0, 1, %o0
>> 116 4:
>> 117 lduh [%o1], %g2
>> 118 add %o1, 2, %o1
>> 119 sth %g2, [%o0]
>> 120 sub %o2, 2, %o2
>> 121 b 3f
>> 122 add %o0, 2, %o0
>> 123
>> 124 FUNC(memcpy) /* %o0=dst %o1=src %o2=len */
>> 125 EXPORT_SYMBOL(memcpy)
>> 126
>> 127 sub %o0, %o1, %o4
>> 128 mov %o0, %g7
>> 129 9:
>> 130 andcc %o4, 3, %o5
>> 131 0:
>> 132 bne 86f
>> 133 cmp %o2, 15
>> 134
>> 135 bleu 90f
>> 136 andcc %o1, 3, %g0
>> 137
>> 138 bne 78b
>> 139 3:
>> 140 andcc %o1, 4, %g0
>> 141
>> 142 be 2f
>> 143 mov %o2, %g1
>> 144
>> 145 ld [%o1], %o4
>> 146 sub %g1, 4, %g1
>> 147 st %o4, [%o0]
>> 148 add %o1, 4, %o1
>> 149 add %o0, 4, %o0
>> 150 2:
>> 151 andcc %g1, 0xffffff80, %g0
>> 152 be 3f
>> 153 andcc %o0, 4, %g0
>> 154
>> 155 be 82f + 4
>> 156 5:
>> 157 MOVE_BIGCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5)
>> 158 MOVE_BIGCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5)
>> 159 MOVE_BIGCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5)
>> 160 MOVE_BIGCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5)
>> 161 sub %g1, 128, %g1
>> 162 add %o1, 128, %o1
>> 163 cmp %g1, 128
>> 164 bge 5b
>> 165 add %o0, 128, %o0
>> 166 3:
>> 167 andcc %g1, 0x70, %g4
>> 168 be 80f
>> 169 andcc %g1, 8, %g0
>> 170
>> 171 sethi %hi(80f), %o5
>> 172 srl %g4, 1, %o4
>> 173 add %g4, %o4, %o4
>> 174 add %o1, %g4, %o1
>> 175 sub %o5, %o4, %o5
>> 176 jmpl %o5 + %lo(80f), %g0
>> 177 add %o0, %g4, %o0
>> 178
>> 179 79: /* memcpy_table */
>> 180
>> 181 MOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g4, g5)
>> 182 MOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g4, g5)
>> 183 MOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g4, g5)
>> 184 MOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g4, g5)
>> 185 MOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g4, g5)
>> 186 MOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g4, g5)
>> 187 MOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g4, g5)
>> 188
>> 189 80: /* memcpy_table_end */
>> 190 be 81f
>> 191 andcc %g1, 4, %g0
>> 192
>> 193 ldd [%o1], %g2
>> 194 add %o0, 8, %o0
>> 195 st %g2, [%o0 - 0x08]
>> 196 add %o1, 8, %o1
>> 197 st %g3, [%o0 - 0x04]
>> 198
>> 199 81: /* memcpy_last7 */
>> 200
>> 201 be 1f
>> 202 andcc %g1, 2, %g0
>> 203
>> 204 ld [%o1], %g2
>> 205 add %o1, 4, %o1
>> 206 st %g2, [%o0]
>> 207 add %o0, 4, %o0
>> 208 1:
>> 209 be 1f
>> 210 andcc %g1, 1, %g0
>> 211
>> 212 lduh [%o1], %g2
>> 213 add %o1, 2, %o1
>> 214 sth %g2, [%o0]
>> 215 add %o0, 2, %o0
>> 216 1:
>> 217 be 1f
>> 218 nop
>> 219
>> 220 ldub [%o1], %g2
>> 221 stb %g2, [%o0]
>> 222 1:
>> 223 retl
>> 224 mov %g7, %o0
>> 225
>> 226 82: /* ldd_std */
>> 227 MOVE_BIGALIGNCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5)
>> 228 MOVE_BIGALIGNCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5)
>> 229 MOVE_BIGALIGNCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5)
>> 230 MOVE_BIGALIGNCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5)
>> 231 subcc %g1, 128, %g1
>> 232 add %o1, 128, %o1
>> 233 cmp %g1, 128
>> 234 bge 82b
>> 235 add %o0, 128, %o0
>> 236
>> 237 andcc %g1, 0x70, %g4
>> 238 be 84f
>> 239 andcc %g1, 8, %g0
>> 240
>> 241 sethi %hi(84f), %o5
>> 242 add %o1, %g4, %o1
>> 243 sub %o5, %g4, %o5
>> 244 jmpl %o5 + %lo(84f), %g0
>> 245 add %o0, %g4, %o0
>> 246
>> 247 83: /* amemcpy_table */
>> 248
>> 249 MOVE_LASTALIGNCHUNK(o1, o0, 0x60, g2, g3, g4, g5)
>> 250 MOVE_LASTALIGNCHUNK(o1, o0, 0x50, g2, g3, g4, g5)
>> 251 MOVE_LASTALIGNCHUNK(o1, o0, 0x40, g2, g3, g4, g5)
>> 252 MOVE_LASTALIGNCHUNK(o1, o0, 0x30, g2, g3, g4, g5)
>> 253 MOVE_LASTALIGNCHUNK(o1, o0, 0x20, g2, g3, g4, g5)
>> 254 MOVE_LASTALIGNCHUNK(o1, o0, 0x10, g2, g3, g4, g5)
>> 255 MOVE_LASTALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5)
>> 256
>> 257 84: /* amemcpy_table_end */
>> 258 be 85f
>> 259 andcc %g1, 4, %g0
>> 260
>> 261 ldd [%o1], %g2
>> 262 add %o0, 8, %o0
>> 263 std %g2, [%o0 - 0x08]
>> 264 add %o1, 8, %o1
>> 265 85: /* amemcpy_last7 */
>> 266 be 1f
>> 267 andcc %g1, 2, %g0
>> 268
>> 269 ld [%o1], %g2
>> 270 add %o1, 4, %o1
>> 271 st %g2, [%o0]
>> 272 add %o0, 4, %o0
>> 273 1:
>> 274 be 1f
>> 275 andcc %g1, 1, %g0
>> 276
>> 277 lduh [%o1], %g2
>> 278 add %o1, 2, %o1
>> 279 sth %g2, [%o0]
>> 280 add %o0, 2, %o0
>> 281 1:
>> 282 be 1f
>> 283 nop
>> 284
>> 285 ldub [%o1], %g2
>> 286 stb %g2, [%o0]
>> 287 1:
>> 288 retl
>> 289 mov %g7, %o0
>> 290
>> 291 86: /* non_aligned */
>> 292 cmp %o2, 6
>> 293 bleu 88f
>> 294 nop
>> 295
>> 296 save %sp, -96, %sp
>> 297 andcc %i0, 3, %g0
>> 298 be 61f
>> 299 andcc %i0, 1, %g0
>> 300 be 60f
>> 301 andcc %i0, 2, %g0
>> 302
>> 303 ldub [%i1], %g5
>> 304 add %i1, 1, %i1
>> 305 stb %g5, [%i0]
>> 306 sub %i2, 1, %i2
>> 307 bne 61f
>> 308 add %i0, 1, %i0
>> 309 60:
>> 310 ldub [%i1], %g3
>> 311 add %i1, 2, %i1
>> 312 stb %g3, [%i0]
>> 313 sub %i2, 2, %i2
>> 314 ldub [%i1 - 1], %g3
>> 315 add %i0, 2, %i0
>> 316 stb %g3, [%i0 - 1]
>> 317 61:
>> 318 and %i1, 3, %g2
>> 319 and %i2, 0xc, %g3
>> 320 and %i1, -4, %i1
>> 321 cmp %g3, 4
>> 322 sll %g2, 3, %g4
>> 323 mov 32, %g2
>> 324 be 4f
>> 325 sub %g2, %g4, %l0
>> 326
>> 327 blu 3f
>> 328 cmp %g3, 0x8
>> 329
>> 330 be 2f
>> 331 srl %i2, 2, %g3
>> 332
>> 333 ld [%i1], %i3
>> 334 add %i0, -8, %i0
>> 335 ld [%i1 + 4], %i4
>> 336 b 8f
>> 337 add %g3, 1, %g3
>> 338 2:
>> 339 ld [%i1], %i4
>> 340 add %i0, -12, %i0
>> 341 ld [%i1 + 4], %i5
>> 342 add %g3, 2, %g3
>> 343 b 9f
>> 344 add %i1, -4, %i1
>> 345 3:
>> 346 ld [%i1], %g1
>> 347 add %i0, -4, %i0
>> 348 ld [%i1 + 4], %i3
>> 349 srl %i2, 2, %g3
>> 350 b 7f
>> 351 add %i1, 4, %i1
>> 352 4:
>> 353 ld [%i1], %i5
>> 354 cmp %i2, 7
>> 355 ld [%i1 + 4], %g1
>> 356 srl %i2, 2, %g3
>> 357 bleu 10f
>> 358 add %i1, 8, %i1
>> 359
>> 360 ld [%i1], %i3
>> 361 add %g3, -1, %g3
>> 362 5:
>> 363 sll %i5, %g4, %g2
>> 364 srl %g1, %l0, %g5
>> 365 or %g2, %g5, %g2
>> 366 st %g2, [%i0]
>> 367 7:
>> 368 ld [%i1 + 4], %i4
>> 369 sll %g1, %g4, %g2
>> 370 srl %i3, %l0, %g5
>> 371 or %g2, %g5, %g2
>> 372 st %g2, [%i0 + 4]
>> 373 8:
>> 374 ld [%i1 + 8], %i5
>> 375 sll %i3, %g4, %g2
>> 376 srl %i4, %l0, %g5
>> 377 or %g2, %g5, %g2
>> 378 st %g2, [%i0 + 8]
>> 379 9:
>> 380 ld [%i1 + 12], %g1
>> 381 sll %i4, %g4, %g2
>> 382 srl %i5, %l0, %g5
>> 383 addcc %g3, -4, %g3
>> 384 or %g2, %g5, %g2
>> 385 add %i1, 16, %i1
>> 386 st %g2, [%i0 + 12]
>> 387 add %i0, 16, %i0
>> 388 bne,a 5b
>> 389 ld [%i1], %i3
>> 390 10:
>> 391 sll %i5, %g4, %g2
>> 392 srl %g1, %l0, %g5
>> 393 srl %l0, 3, %g3
>> 394 or %g2, %g5, %g2
>> 395 sub %i1, %g3, %i1
>> 396 andcc %i2, 2, %g0
>> 397 st %g2, [%i0]
>> 398 be 1f
>> 399 andcc %i2, 1, %g0
>> 400
>> 401 ldub [%i1], %g2
>> 402 add %i1, 2, %i1
>> 403 stb %g2, [%i0 + 4]
>> 404 add %i0, 2, %i0
>> 405 ldub [%i1 - 1], %g2
>> 406 stb %g2, [%i0 + 3]
>> 407 1:
>> 408 be 1f
>> 409 nop
>> 410 ldub [%i1], %g2
>> 411 stb %g2, [%i0 + 4]
>> 412 1:
>> 413 ret
>> 414 restore %g7, %g0, %o0
>> 415
>> 416 88: /* short_end */
>> 417
>> 418 and %o2, 0xe, %o3
>> 419 20:
>> 420 sethi %hi(89f), %o5
>> 421 sll %o3, 3, %o4
>> 422 add %o0, %o3, %o0
>> 423 sub %o5, %o4, %o5
>> 424 add %o1, %o3, %o1
>> 425 jmpl %o5 + %lo(89f), %g0
>> 426 andcc %o2, 1, %g0
>> 427
>> 428 MOVE_SHORTCHUNK(o1, o0, 0x0c, g2, g3)
>> 429 MOVE_SHORTCHUNK(o1, o0, 0x0a, g2, g3)
>> 430 MOVE_SHORTCHUNK(o1, o0, 0x08, g2, g3)
>> 431 MOVE_SHORTCHUNK(o1, o0, 0x06, g2, g3)
>> 432 MOVE_SHORTCHUNK(o1, o0, 0x04, g2, g3)
>> 433 MOVE_SHORTCHUNK(o1, o0, 0x02, g2, g3)
>> 434 MOVE_SHORTCHUNK(o1, o0, 0x00, g2, g3)
>> 435
>> 436 89: /* short_table_end */
>> 437
>> 438 be 1f
>> 439 nop
>> 440
>> 441 ldub [%o1], %g2
>> 442 stb %g2, [%o0]
>> 443 1:
>> 444 retl
>> 445 mov %g7, %o0
>> 446
>> 447 90: /* short_aligned_end */
>> 448 bne 88b
>> 449 andcc %o2, 8, %g0
>> 450
>> 451 be 1f
>> 452 andcc %o2, 4, %g0
>> 453
>> 454 ld [%o1 + 0x00], %g2
>> 455 ld [%o1 + 0x04], %g3
>> 456 add %o1, 8, %o1
>> 457 st %g2, [%o0 + 0x00]
>> 458 st %g3, [%o0 + 0x04]
>> 459 add %o0, 8, %o0
>> 460 1:
>> 461 b 81b
>> 462 mov %o2, %g1
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.