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