1 /*
2 * M7memcpy: Optimized SPARC M7 memcpy
3 *
4 * Copyright (c) 2016, Oracle and/or its affil
5 */
6
7 .file "M7memcpy.S"
8
9 /*
10 * memcpy(s1, s2, len)
11 *
12 * Copy s2 to s1, always copy n bytes.
13 * Note: this C code does not work for overlap
14 *
15 * Fast assembler language version of the foll
16 * which represents the `standard' for the C-l
17 *
18 * void *
19 * memcpy(void *s, const void *s0, size_t
20 * {
21 * if (n != 0) {
22 * char *s1 = s;
23 * const char *s2 = s0;
24 * do {
25 * *s1++ = *s2++;
26 * } while (--n != 0);
27 * }
28 * return (s);
29 * }
30 *
31 *
32 * SPARC T7/M7 Flow :
33 *
34 * if (count < SMALL_MAX) {
35 * if count < SHORTCOPY (SHORTC
36 * copy bytes; exit with dst addr
37 * if src & dst aligned on word boundary but
38 * copy with ldw/stw; branch to finish_up
39 * if src & dst aligned on long word boundar
40 * copy with ldx/stx; branch to finish_up
41 * if src & dst not aligned and length <= SH
42 * copy bytes; exit with dst addr
43 * move enough bytes to get src to word boun
44 * if dst now on word boundary
45 * move_words:
46 * copy words; branch to finish_up
47 * if dst now on half word boundary
48 * load words, shift half words, store wor
49 * if dst on byte 1
50 * load words, shift 3 bytes, store words;
51 * if dst on byte 3
52 * load words, shift 1 byte, store words;
53 * finish_up:
54 * copy bytes; exit with dst addr
55 * } else {
56 * move bytes until dst is on long word boun
57 * if( src is on long word boundary ) {
58 * if (count < MED_MAX) {
59 * finish_long:
60 * copy with ldx/stx in 8-way unrolled l
61 * copy final 0-63 bytes; exit with dst
62 * } else { s
63 * align dst on 64 byte boundary; for ma
64 * prefetch src data to L2 cache; let HW
65 * Use BIS (block initializing store) to
66 * lines from memory. But pre-store firs
67 * ST_CHUNK lines in advance of the rest
68 * gives time for replacement cache line
69 * excess STQ and Miss Buffer filling. R
70 * then finish up storing before going t
71 * }
72 * } else {
73 * if src is word aligned and count < MED_
74 * move words in 8-way unrolled loop
75 * move final 0-31 bytes; exit with dst
76 * if count < MED_UMAX
77 * use alignaddr/faligndata combined wit
78 * unrolled loop to move data.
79 * go to unalign_done
80 * else
81 * setup alignaddr for faligndata instru
82 * align dst on 64 byte boundary; prefet
83 * loadx8, falign, block-store, prefetch
84 * (only use block-init-store when src/d
85 * unalign_done:
86 * move remaining bytes for unaligned ca
87 * }
88 *
89 */
90
91 #include <asm/visasm.h>
92 #include <asm/asi.h>
93
94 #if !defined(EX_LD) && !defined(EX_ST)
95 #define NON_USER_COPY
96 #endif
97
98 #ifndef EX_LD
99 #define EX_LD(x,y) x
100 #endif
101 #ifndef EX_LD_FP
102 #define EX_LD_FP(x,y) x
103 #endif
104
105 #ifndef EX_ST
106 #define EX_ST(x,y) x
107 #endif
108 #ifndef EX_ST_FP
109 #define EX_ST_FP(x,y) x
110 #endif
111
112 #ifndef EX_RETVAL
113 #define EX_RETVAL(x) x
114 #endif
115
116 #ifndef LOAD
117 #define LOAD(type,addr,dest) type [addr], d
118 #endif
119
120 #ifndef STORE
121 #define STORE(type,src,addr) type src, [add
122 #endif
123
124 /*
125 * ASI_BLK_INIT_QUAD_LDD_P/ASI_BLK_INIT_QUAD_L
126 * line as "least recently used" which means i
127 * active, it has a high probability of being
128 * between the first initializing store and th
129 * Thus, we use ASI_ST_BLKINIT_MRU_P/ASI_ST_BL
130 * marks the cache line as "most recently used
131 * but the last cache line
132 */
133 #ifndef STORE_ASI
134 #ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
135 #define STORE_ASI ASI_BLK_INIT_QUAD_LDD_
136 #else
137 #define STORE_ASI 0x80 /* ASI
138 #endif
139 #endif
140
141 #ifndef STORE_MRU_ASI
142 #ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
143 #define STORE_MRU_ASI ASI_ST_BLKINIT_MRU_P
144 #else
145 #define STORE_MRU_ASI 0x80 /* ASI
146 #endif
147 #endif
148
149 #ifndef STORE_INIT
150 #define STORE_INIT(src,addr) stxa src, [add
151 #endif
152
153 #ifndef STORE_INIT_MRU
154 #define STORE_INIT_MRU(src,addr) stxa s
155 #endif
156
157 #ifndef FUNC_NAME
158 #define FUNC_NAME M7memcpy
159 #endif
160
161 #ifndef PREAMBLE
162 #define PREAMBLE
163 #endif
164
165 #define BLOCK_SIZE 64
166 #define SHORTCOPY 3
167 #define SHORTCHECK 14
168 #define SHORT_LONG 64 /* max copy fo
169 /* must be at
170 #define SMALL_MAX 128
171 #define MED_UMAX 1024 /* max copy fo
172 #define MED_WMAX 1024 /* max copy fo
173 #define MED_MAX 1024 /* max copy fo
174 #define ST_CHUNK 24 /* ST_CHUNK -
175 #define ALIGN_PRE 24 /* distance fo
176
177 .register %g2,#scratch
178
179 .section ".text"
180 .global FUNC_NAME
181 .type FUNC_NAME, #function
182 .align 16
183 FUNC_NAME:
184 srlx %o2, 31, %g2
185 cmp %g2, 0
186 tne %xcc, 5
187 PREAMBLE
188 mov %o0, %g1 ! save
189 brz,pn %o2, .Lsmallx
190 cmp %o2, 3
191 ble,pn %icc, .Ltiny_cp
192 cmp %o2, 19
193 ble,pn %icc, .Lsmall_cp
194 or %o0, %o1, %g2
195 cmp %o2, SMALL_MAX
196 bl,pn %icc, .Lmedium_cp
197 nop
198
199 .Lmedium:
200 neg %o0, %o5
201 andcc %o5, 7, %o5 ! byte
202 brz,pt %o5, .Ldst_aligned_on_8
203
204 ! %o5 has the bytes to be written in p
205 sub %o2, %o5, %o2
206 sub %o1, %o0, %o1 ! %o1
207 7: ! dst
208 add %o1, %o0, %o4
209 EX_LD(LOAD(ldub, %o4, %o4), memcpy_ret
210 subcc %o5, 1, %o5
211 EX_ST(STORE(stb, %o4, %o0), memcpy_ret
212 bgu,pt %xcc, 7b
213 add %o0, 1, %o0 ! adva
214 add %o1, %o0, %o1 ! rest
215 .Ldst_aligned_on_8:
216 andcc %o1, 7, %o5
217 brnz,pt %o5, .Lsrc_dst_unaligned_on_8
218 nop
219
220 .Lsrc_dst_aligned_on_8:
221 ! check if we are copying MED_MAX or m
222 set MED_MAX, %o3
223 cmp %o2, %o3 ! limi
224 bgu,pn %xcc, .Llarge_align8_copy
225 nop
226
227 /*
228 * Special case for handling when src and dest
229 * and total data to move is less than MED_MAX
230 */
231 .Lmedlong:
232 subcc %o2, 63, %o2 ! adju
233 ble,pn %xcc, .Lmedl63 ! skip
234 nop
235 .Lmedl64:
236 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
237 subcc %o2, 64, %o2 ! decr
238 EX_ST(STORE(stx, %o4, %o0), memcpy_ret
239 EX_LD(LOAD(ldx, %o1+8, %o3), memcpy_re
240 EX_ST(STORE(stx, %o3, %o0+8), memcpy_r
241 EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_r
242 EX_ST(STORE(stx, %o4, %o0+16), memcpy_
243 EX_LD(LOAD(ldx, %o1+24, %o3), memcpy_r
244 EX_ST(STORE(stx, %o3, %o0+24), memcpy_
245 EX_LD(LOAD(ldx, %o1+32, %o4), memcpy_r
246 EX_ST(STORE(stx, %o4, %o0+32), memcpy_
247 EX_LD(LOAD(ldx, %o1+40, %o3), memcpy_r
248 add %o1, 64, %o1 ! incr
249 EX_ST(STORE(stx, %o3, %o0+40), memcpy_
250 EX_LD(LOAD(ldx, %o1-16, %o4), memcpy_r
251 add %o0, 64, %o0 ! incr
252 EX_ST(STORE(stx, %o4, %o0-16), memcpy_
253 EX_LD(LOAD(ldx, %o1-8, %o3), memcpy_re
254 bgu,pt %xcc, .Lmedl64 ! repe
255 EX_ST(STORE(stx, %o3, %o0-8), memcpy_
256 .Lmedl63:
257 addcc %o2, 32, %o2 ! adju
258 ble,pt %xcc, .Lmedl31 ! to s
259 nop
260 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
261 sub %o2, 32, %o2 ! decr
262 EX_ST(STORE(stx, %o4, %o0), memcpy_ret
263 EX_LD(LOAD(ldx, %o1+8, %o3), memcpy_re
264 add %o1, 32, %o1 ! incr
265 EX_ST(STORE(stx, %o3, %o0+8), memcpy_r
266 EX_LD(LOAD(ldx, %o1-16, %o4), memcpy_r
267 add %o0, 32, %o0 ! incr
268 EX_ST(STORE(stx, %o4, %o0-16), memcpy_
269 EX_LD(LOAD(ldx, %o1-8, %o3), memcpy_re
270 EX_ST(STORE(stx, %o3, %o0-8), memcpy_r
271 .Lmedl31:
272 addcc %o2, 16, %o2 ! adju
273 ble,pt %xcc, .Lmedl15 ! skip
274 nop !
275 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
276 add %o1, 16, %o1 ! incr
277 EX_ST(STORE(stx, %o4, %o0), memcpy_ret
278 sub %o2, 16, %o2 ! decr
279 EX_LD(LOAD(ldx, %o1-8, %o3), memcpy_re
280 add %o0, 16, %o0 ! incr
281 EX_ST(STORE(stx, %o3, %o0-8), memcpy_r
282 .Lmedl15:
283 addcc %o2, 15, %o2 ! rest
284 bz,pt %xcc, .Lsmallx ! exit if fini
285 cmp %o2, 8
286 blt,pt %xcc, .Lmedw7 ! skip
287 tst %o2
288 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
289 add %o1, 8, %o1 ! incr
290 add %o0, 8, %o0 ! incr
291 subcc %o2, 8, %o2 ! decr
292 bnz,pn %xcc, .Lmedw7
293 EX_ST(STORE(stx, %o4, %o0-8), memcpy_
294 retl
295 mov EX_RETVAL(%g1), %o0 ! rest
296
297 .align 16
298 .Lsrc_dst_unaligned_on_8:
299 ! DST is 8-byte aligned, src is not
300 2:
301 andcc %o1, 0x3, %o5 ! test
302 bnz,pt %xcc, .Lunalignsetup ! bran
303 nop
304
305 /*
306 * Handle all cases where src and dest are ali
307 * boundaries. Use unrolled loops for better p
308 * This option wins over standard large data m
309 * source and destination is in cache for.Lmed
310 * to short data moves.
311 */
312 set MED_WMAX, %o3
313 cmp %o2, %o3 ! limi
314 bge,pt %xcc, .Lunalignrejoin ! othe
315 nop
316
317 subcc %o2, 31, %o2 ! adju
318 ! for
319 ble,pt %xcc, .Lmedw31 ! skip
320 .Lmedw32:
321 EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_
322 sllx %o4, 32, %o5
323 EX_LD(LOAD(ld, %o1+4, %o4), memcpy_ret
324 or %o4, %o5, %o5
325 EX_ST(STORE(stx, %o5, %o0), memcpy_ret
326 subcc %o2, 32, %o2 ! decr
327 EX_LD(LOAD(ld, %o1+8, %o4), memcpy_ret
328 sllx %o4, 32, %o5
329 EX_LD(LOAD(ld, %o1+12, %o4), memcpy_re
330 or %o4, %o5, %o5
331 EX_ST(STORE(stx, %o5, %o0+8), memcpy_r
332 add %o1, 32, %o1 ! incr
333 EX_LD(LOAD(ld, %o1-16, %o4), memcpy_re
334 sllx %o4, 32, %o5
335 EX_LD(LOAD(ld, %o1-12, %o4), memcpy_re
336 or %o4, %o5, %o5
337 EX_ST(STORE(stx, %o5, %o0+16), memcpy_
338 add %o0, 32, %o0 ! incr
339 EX_LD(LOAD(ld, %o1-8, %o4), memcpy_ret
340 sllx %o4, 32, %o5
341 EX_LD(LOAD(ld, %o1-4, %o4), memcpy_ret
342 or %o4, %o5, %o5
343 bgu,pt %xcc, .Lmedw32 ! repe
344 EX_ST(STORE(stx, %o5, %o0-8), memcpy_
345 .Lmedw31:
346 addcc %o2, 31, %o2 ! rest
347
348 bz,pt %xcc, .Lsmallx ! exit if fini
349 nop
350 cmp %o2, 16
351 blt,pt %xcc, .Lmedw15
352 nop
353 EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_
354 sllx %o4, 32, %o5
355 subcc %o2, 16, %o2 ! decr
356 EX_LD(LOAD(ld, %o1+4, %o4), memcpy_ret
357 or %o4, %o5, %o5
358 EX_ST(STORE(stx, %o5, %o0), memcpy_ret
359 add %o1, 16, %o1 ! incr
360 EX_LD(LOAD(ld, %o1-8, %o4), memcpy_ret
361 add %o0, 16, %o0 ! incr
362 sllx %o4, 32, %o5
363 EX_LD(LOAD(ld, %o1-4, %o4), memcpy_ret
364 or %o4, %o5, %o5
365 EX_ST(STORE(stx, %o5, %o0-8), memcpy_r
366 .Lmedw15:
367 bz,pt %xcc, .Lsmallx ! exit if fini
368 cmp %o2, 8
369 blt,pn %xcc, .Lmedw7 ! skip
370 tst %o2
371 EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_
372 subcc %o2, 8, %o2 ! decr
373 EX_ST(STORE(stw, %o4, %o0), memcpy_ret
374 add %o1, 8, %o1 ! incr
375 EX_LD(LOAD(ld, %o1-4, %o3), memcpy_ret
376 add %o0, 8, %o0 ! incr
377 EX_ST(STORE(stw, %o3, %o0-4), memcpy_r
378 bz,pt %xcc, .Lsmallx ! exit if fini
379 .Lmedw7: ! coun
380 cmp %o2, 4 ! chec
381 blt,pn %xcc, .Lsmallleft3 ! skip
382 nop !
383 EX_LD(LOAD(ld, %o1, %o4), memcpy_retl_
384 add %o1, 4, %o1 ! incr
385 add %o0, 4, %o0 ! incr
386 subcc %o2, 4, %o2 ! decr
387 bnz .Lsmallleft3
388 EX_ST(STORE(stw, %o4, %o0-4), memcpy_
389 retl
390 mov EX_RETVAL(%g1), %o0
391
392 .align 16
393 .Llarge_align8_copy: ! Src
394 ! align dst to 64 byte boundary
395 andcc %o0, 0x3f, %o3 ! %o3
396 brz,pn %o3, .Laligned_to_64
397 andcc %o0, 8, %o3 ! odd
398 brz,pt %o3, .Laligned_to_16
399 nop
400 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
401 sub %o2, 8, %o2
402 add %o1, 8, %o1 ! incr
403 add %o0, 8, %o0 ! incr
404 EX_ST(STORE(stx, %o4, %o0-8), memcpy_r
405 .Laligned_to_16:
406 andcc %o0, 16, %o3 ! pair
407 brz,pt %o3, .Laligned_to_32
408 nop
409 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
410 sub %o2, 16, %o2
411 EX_ST(STORE(stx, %o4, %o0), memcpy_ret
412 add %o1, 16, %o1 ! incr
413 EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_re
414 add %o0, 16, %o0 ! incr
415 EX_ST(STORE(stx, %o4, %o0-8), memcpy_r
416 .Laligned_to_32:
417 andcc %o0, 32, %o3 ! four
418 brz,pt %o3, .Laligned_to_64
419 nop
420 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
421 sub %o2, 32, %o2
422 EX_ST(STORE(stx, %o4, %o0), memcpy_ret
423 EX_LD(LOAD(ldx, %o1+8, %o4), memcpy_re
424 EX_ST(STORE(stx, %o4, %o0+8), memcpy_r
425 EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_r
426 EX_ST(STORE(stx, %o4, %o0+16), memcpy_
427 add %o1, 32, %o1 ! incr
428 EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_re
429 add %o0, 32, %o0 ! incr
430 EX_ST(STORE(stx, %o4, %o0-8), memcpy_r
431 .Laligned_to_64:
432 !
433 ! Using block init store (BIS) instructi
434 ! lines from memory. Use ST_CHUNK stores
435 ! line (similar to prefetching) to avoid
436 ! Gives existing cache lines time to be
437 ! Initial stores using MRU version of BI
438 ! cache until we are ready to store fina
439 ! Then store last element using the LRU
440 !
441 andn %o2, 0x3f, %o5 ! %o5
442 and %o2, 0x3f, %o2 ! resi
443 !
444 ! We use STORE_MRU_ASI for the first sev
445 ! followed by STORE_ASI (mark as LRU) fo
446 ! mixed approach reduces the probability
447 ! before we finish setting it, while min
448 ! other cached values during a large mem
449 !
450 ! ST_CHUNK batches up initial BIS operat
451 ! to allow multiple requests to not be b
452 ! the store miss buffer. Then the matchi
453 ! BIS operations are executed.
454 !
455
456 sub %o0, 8, %o0 ! adju
457 .Lalign_loop:
458 cmp %o5, ST_CHUNK*64
459 blu,pt %xcc, .Lalign_loop_fin
460 mov ST_CHUNK,%o3
461 .Lalign_loop_start:
462 prefetch [%o1 + (ALIGN_PRE * BLOCK_SIZ
463 subcc %o3, 1, %o3
464 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
465 add %o1, 64, %o1
466 add %o0, 8, %o0
467 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
468 bgu %xcc,.Lalign_loop_start
469 add %o0, 56, %o0
470
471 mov ST_CHUNK,%o3
472 sllx %o3, 6, %o4 ! ST_C
473 sub %o1, %o4, %o1 ! rese
474 sub %o0, %o4, %o0 ! rese
475
476 .Lalign_loop_rest:
477 EX_LD(LOAD(ldx, %o1+8, %o4), memcpy_re
478 add %o0, 16, %o0
479 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
480 EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_r
481 add %o0, 8, %o0
482 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
483 subcc %o3, 1, %o3
484 EX_LD(LOAD(ldx, %o1+24, %o4), memcpy_r
485 add %o0, 8, %o0
486 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
487 EX_LD(LOAD(ldx, %o1+32, %o4), memcpy_r
488 add %o0, 8, %o0
489 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
490 EX_LD(LOAD(ldx, %o1+40, %o4), memcpy_r
491 add %o0, 8, %o0
492 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
493 EX_LD(LOAD(ldx, %o1+48, %o4), memcpy_r
494 add %o1, 64, %o1
495 add %o0, 8, %o0
496 EX_ST(STORE_INIT_MRU(%o4, %o0), memcpy
497 add %o0, 8, %o0
498 EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_re
499 sub %o5, 64, %o5
500 bgu %xcc,.Lalign_loop_rest
501 ! mark cache line as LRU
502 EX_ST(STORE_INIT(%o4, %o0), memcpy_re
503
504 cmp %o5, ST_CHUNK*64
505 bgu,pt %xcc, .Lalign_loop_start
506 mov ST_CHUNK,%o3
507
508 cmp %o5, 0
509 beq .Lalign_done
510 nop
511 .Lalign_loop_fin:
512 EX_LD(LOAD(ldx, %o1, %o4), memcpy_retl
513 EX_ST(STORE(stx, %o4, %o0+8), memcpy_r
514 EX_LD(LOAD(ldx, %o1+8, %o4), memcpy_re
515 EX_ST(STORE(stx, %o4, %o0+8+8), memcpy
516 EX_LD(LOAD(ldx, %o1+16, %o4), memcpy_r
517 EX_ST(STORE(stx, %o4, %o0+8+16), memcp
518 subcc %o5, 64, %o5
519 EX_LD(LOAD(ldx, %o1+24, %o4), memcpy_r
520 EX_ST(STORE(stx, %o4, %o0+8+24), memcp
521 EX_LD(LOAD(ldx, %o1+32, %o4), memcpy_r
522 EX_ST(STORE(stx, %o4, %o0+8+32), memcp
523 EX_LD(LOAD(ldx, %o1+40, %o4), memcpy_r
524 EX_ST(STORE(stx, %o4, %o0+8+40), memcp
525 EX_LD(LOAD(ldx, %o1+48, %o4), memcpy_r
526 add %o1, 64, %o1
527 EX_ST(STORE(stx, %o4, %o0+8+48), memcp
528 add %o0, 64, %o0
529 EX_LD(LOAD(ldx, %o1-8, %o4), memcpy_re
530 bgu %xcc,.Lalign_loop_fin
531 EX_ST(STORE(stx, %o4, %o0), memcpy_re
532
533 .Lalign_done:
534 add %o0, 8, %o0 ! rest
535 membar #StoreStore
536 sub %o2, 63, %o2 ! adju
537 ba .Lmedl63 ! in .
538 nop
539
540 .align 16
541 ! Dst is on 8 byte boundary; src is no
542 .Lunalignsetup:
543 .Lunalignrejoin:
544 mov %g1, %o3 ! save %g1 as
545 #ifdef NON_USER_COPY
546 VISEntryHalfFast(.Lmedium_vis_entry_fa
547 #else
548 VISEntryHalf
549 #endif
550 mov %o3, %g1 ! restore %g1
551
552 set MED_UMAX, %o3
553 cmp %o2, %o3 ! check for.Lm
554 bge,pt %xcc,.Lunalign_large
555 prefetch [%o1 + (4 * BLOCK_SIZE)], 20
556 andn %o2, 0x3f, %o5 ! %o5
557 and %o2, 0x3f, %o2 ! resi
558 cmp %o2, 8 ! Insu
559 bgt .Lunalign_adjust ! end
560 andn %o1, 0x7, %o4 ! %o4
561 add %o2, 64, %o2 ! adju
562 sub %o5, 64, %o5 ! earl
563 .Lunalign_adjust:
564 alignaddr %o1, %g0, %g0 ! gene
565 add %o1, %o5, %o1 ! adva
566 EX_LD_FP(LOAD(ldd, %o4, %f0), memcpy_r
567 .Lunalign_loop:
568 EX_LD_FP(LOAD(ldd, %o4+8, %f2), memcpy
569 faligndata %f0, %f2, %f16
570 EX_LD_FP(LOAD(ldd, %o4+16, %f4), memcp
571 subcc %o5, BLOCK_SIZE, %o5
572 EX_ST_FP(STORE(std, %f16, %o0), memcpy
573 faligndata %f2, %f4, %f18
574 EX_LD_FP(LOAD(ldd, %o4+24, %f6), memcp
575 EX_ST_FP(STORE(std, %f18, %o0+8), memc
576 faligndata %f4, %f6, %f20
577 EX_LD_FP(LOAD(ldd, %o4+32, %f8), memcp
578 EX_ST_FP(STORE(std, %f20, %o0+16), mem
579 faligndata %f6, %f8, %f22
580 EX_LD_FP(LOAD(ldd, %o4+40, %f10), memc
581 EX_ST_FP(STORE(std, %f22, %o0+24), mem
582 faligndata %f8, %f10, %f24
583 EX_LD_FP(LOAD(ldd, %o4+48, %f12), memc
584 EX_ST_FP(STORE(std, %f24, %o0+32), mem
585 faligndata %f10, %f12, %f26
586 EX_LD_FP(LOAD(ldd, %o4+56, %f14), memc
587 add %o4, BLOCK_SIZE, %o4
588 EX_ST_FP(STORE(std, %f26, %o0+40), mem
589 faligndata %f12, %f14, %f28
590 EX_LD_FP(LOAD(ldd, %o4, %f0), memcpy_r
591 EX_ST_FP(STORE(std, %f28, %o0+48), mem
592 faligndata %f14, %f0, %f30
593 EX_ST_FP(STORE(std, %f30, %o0+56), mem
594 add %o0, BLOCK_SIZE, %o0
595 bgu,pt %xcc, .Lunalign_loop
596 prefetch [%o4 + (5 * BLOCK_SIZE)], 20
597 ba .Lunalign_done
598 nop
599
600 .Lunalign_large:
601 andcc %o0, 0x3f, %o3 ! is d
602 bz %xcc, .Lunalignsrc
603 sub %o3, 64, %o3 ! %o3
604 neg %o3 ! byte
605 sub %o2, %o3, %o2 ! upda
606 ! Move bytes according to source align
607 andcc %o1, 0x1, %o5
608 bnz %xcc, .Lunalignbyte ! chec
609 nop
610 andcc %o1, 2, %o5 ! chec
611 bnz %xcc, .Lunalignhalf
612 nop
613 ! Src is word aligned
614 .Lunalignword:
615 EX_LD_FP(LOAD(ld, %o1, %o4), memcpy_re
616 add %o1, 8, %o1 ! incr
617 EX_ST_FP(STORE(stw, %o4, %o0), memcpy_
618 subcc %o3, 8, %o3 ! decr
619 EX_LD_FP(LOAD(ld, %o1-4, %o4), memcpy_
620 add %o0, 8, %o0 ! incr
621 bnz %xcc, .Lunalignword
622 EX_ST_FP(STORE(stw, %o4, %o0-4), memc
623 ba .Lunalignsrc
624 nop
625
626 ! Src is half-word aligned
627 .Lunalignhalf:
628 EX_LD_FP(LOAD(lduh, %o1, %o4), memcpy_
629 sllx %o4, 32, %o5 ! shif
630 EX_LD_FP(LOAD(lduw, %o1+2, %o4), memcp
631 or %o4, %o5, %o5
632 sllx %o5, 16, %o5
633 EX_LD_FP(LOAD(lduh, %o1+6, %o4), memcp
634 or %o4, %o5, %o5
635 EX_ST_FP(STORE(stx, %o5, %o0), memcpy_
636 add %o1, 8, %o1
637 subcc %o3, 8, %o3
638 bnz %xcc, .Lunalignhalf
639 add %o0, 8, %o0
640 ba .Lunalignsrc
641 nop
642
643 ! Src is Byte aligned
644 .Lunalignbyte:
645 sub %o0, %o1, %o0 ! shar
646 .Lunalignbyte_loop:
647 EX_LD_FP(LOAD(ldub, %o1, %o4), memcpy_
648 sllx %o4, 56, %o5
649 EX_LD_FP(LOAD(lduh, %o1+1, %o4), memcp
650 sllx %o4, 40, %o4
651 or %o4, %o5, %o5
652 EX_LD_FP(LOAD(lduh, %o1+3, %o4), memcp
653 sllx %o4, 24, %o4
654 or %o4, %o5, %o5
655 EX_LD_FP(LOAD(lduh, %o1+5, %o4), memcp
656 sllx %o4, 8, %o4
657 or %o4, %o5, %o5
658 EX_LD_FP(LOAD(ldub, %o1+7, %o4), memcp
659 or %o4, %o5, %o5
660 add %o0, %o1, %o0
661 EX_ST_FP(STORE(stx, %o5, %o0), memcpy_
662 sub %o0, %o1, %o0
663 subcc %o3, 8, %o3
664 bnz %xcc, .Lunalignbyte_loop
665 add %o1, 8, %o1
666 add %o0,%o1, %o0 ! rest
667
668 ! Destination is now block (64 byte al
669 .Lunalignsrc:
670 andn %o2, 0x3f, %o5 ! %o5
671 and %o2, 0x3f, %o2 ! resi
672 add %o2, 64, %o2 ! Insu
673 sub %o5, 64, %o5 ! end
674
675 andn %o1, 0x7, %o4 ! %o4
676 alignaddr %o1, %g0, %g0 ! gene
677 add %o1, %o5, %o1 ! adva
678
679 EX_LD_FP(LOAD(ldd, %o4, %f14), memcpy_
680 add %o4, 8, %o4
681 .Lunalign_sloop:
682 EX_LD_FP(LOAD(ldd, %o4, %f16), memcpy_
683 faligndata %f14, %f16, %f0
684 EX_LD_FP(LOAD(ldd, %o4+8, %f18), memcp
685 faligndata %f16, %f18, %f2
686 EX_LD_FP(LOAD(ldd, %o4+16, %f20), memc
687 faligndata %f18, %f20, %f4
688 EX_ST_FP(STORE(std, %f0, %o0), memcpy_
689 subcc %o5, 64, %o5
690 EX_LD_FP(LOAD(ldd, %o4+24, %f22), memc
691 faligndata %f20, %f22, %f6
692 EX_ST_FP(STORE(std, %f2, %o0+8), memcp
693 EX_LD_FP(LOAD(ldd, %o4+32, %f24), memc
694 faligndata %f22, %f24, %f8
695 EX_ST_FP(STORE(std, %f4, %o0+16), memc
696 EX_LD_FP(LOAD(ldd, %o4+40, %f26), memc
697 faligndata %f24, %f26, %f10
698 EX_ST_FP(STORE(std, %f6, %o0+24), memc
699 EX_LD_FP(LOAD(ldd, %o4+48, %f28), memc
700 faligndata %f26, %f28, %f12
701 EX_ST_FP(STORE(std, %f8, %o0+32), memc
702 add %o4, 64, %o4
703 EX_LD_FP(LOAD(ldd, %o4-8, %f30), memcp
704 faligndata %f28, %f30, %f14
705 EX_ST_FP(STORE(std, %f10, %o0+40), mem
706 EX_ST_FP(STORE(std, %f12, %o0+48), mem
707 add %o0, 64, %o0
708 EX_ST_FP(STORE(std, %f14, %o0-8), memc
709 fsrc2 %f30, %f14
710 bgu,pt %xcc, .Lunalign_sloop
711 prefetch [%o4 + (8 * BLOCK_SIZE)], 20
712
713 .Lunalign_done:
714 ! Handle trailing bytes, 64 to 127
715 ! Dest long word aligned, Src not long
716 cmp %o2, 15
717 bleu %xcc, .Lunalign_short
718
719 andn %o2, 0x7, %o5 ! %o5
720 and %o2, 0x7, %o2 ! resi
721 add %o2, 8, %o2
722 sub %o5, 8, %o5 ! insu
723 andn %o1, 0x7, %o4 ! %o4
724 add %o1, %o5, %o1 ! adva
725 EX_LD_FP(LOAD(ldd, %o4, %f0), memcpy_r
726 .Lunalign_by8:
727 EX_LD_FP(LOAD(ldd, %o4+8, %f2), memcpy
728 add %o4, 8, %o4
729 faligndata %f0, %f2, %f16
730 subcc %o5, 8, %o5
731 EX_ST_FP(STORE(std, %f16, %o0), memcpy
732 fsrc2 %f2, %f0
733 bgu,pt %xcc, .Lunalign_by8
734 add %o0, 8, %o0
735
736 .Lunalign_short:
737 #ifdef NON_USER_COPY
738 VISExitHalfFast
739 #else
740 VISExitHalf
741 #endif
742 ba .Lsmallrest
743 nop
744
745 /*
746 * This is a special case of nested memcpy. Th
747 * calls unaligned memcpy back to back without
748 * traps(context switch) to save/restore FP re
749 * memcpy without this trap sequence we will h
750 * the normal integer load/store method in thi
751 */
752
753 #ifdef NON_USER_COPY
754 .Lmedium_vis_entry_fail_cp:
755 or %o0, %o1, %g2
756 #endif
757 .Lmedium_cp:
758 LOAD(prefetch, %o1 + 0x40, #n_reads_st
759 andcc %g2, 0x7, %g0
760 bne,pn %xcc, .Lmedium_unaligned_cp
761 nop
762
763 .Lmedium_noprefetch_cp:
764 andncc %o2, 0x20 - 1, %o5
765 be,pn %xcc, 2f
766 sub %o2, %o5, %o2
767 1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3), memc
768 EX_LD(LOAD(ldx, %o1 + 0x08, %g2), memc
769 EX_LD(LOAD(ldx, %o1 + 0x10, %g7), memc
770 EX_LD(LOAD(ldx, %o1 + 0x18, %o4), memc
771 add %o1, 0x20, %o1
772 subcc %o5, 0x20, %o5
773 EX_ST(STORE(stx, %o3, %o0 + 0x00), mem
774 EX_ST(STORE(stx, %g2, %o0 + 0x08), mem
775 EX_ST(STORE(stx, %g7, %o0 + 0x10), mem
776 EX_ST(STORE(stx, %o4, %o0 + 0x18), mem
777 bne,pt %xcc, 1b
778 add %o0, 0x20, %o0
779 2: andcc %o2, 0x18, %o5
780 be,pt %xcc, 3f
781 sub %o2, %o5, %o2
782 1: EX_LD(LOAD(ldx, %o1 + 0x00, %o3), memc
783 add %o1, 0x08, %o1
784 add %o0, 0x08, %o0
785 subcc %o5, 0x08, %o5
786 bne,pt %xcc, 1b
787 EX_ST(STORE(stx, %o3, %o0 - 0x08), me
788 3: brz,pt %o2, .Lexit_cp
789 cmp %o2, 0x04
790 bl,pn %xcc, .Ltiny_cp
791 nop
792 EX_LD(LOAD(lduw, %o1 + 0x00, %o3), mem
793 add %o1, 0x04, %o1
794 add %o0, 0x04, %o0
795 subcc %o2, 0x04, %o2
796 bne,pn %xcc, .Ltiny_cp
797 EX_ST(STORE(stw, %o3, %o0 - 0x04), me
798 ba,a,pt %xcc, .Lexit_cp
799
800 .Lmedium_unaligned_cp:
801 /* First get dest 8 byte aligned. */
802 sub %g0, %o0, %o3
803 and %o3, 0x7, %o3
804 brz,pt %o3, 2f
805 sub %o2, %o3, %o2
806
807 1: EX_LD(LOAD(ldub, %o1 + 0x00, %g2), mem
808 add %o1, 1, %o1
809 subcc %o3, 1, %o3
810 add %o0, 1, %o0
811 bne,pt %xcc, 1b
812 EX_ST(STORE(stb, %g2, %o0 - 0x01), me
813 2:
814 and %o1, 0x7, %o3
815 brz,pn %o3, .Lmedium_noprefetch_cp
816 sll %o3, 3, %o3
817 mov 64, %g2
818 sub %g2, %o3, %g2
819 andn %o1, 0x7, %o1
820 EX_LD(LOAD(ldx, %o1 + 0x00, %o4), memc
821 sllx %o4, %o3, %o4
822 andn %o2, 0x08 - 1, %o5
823 sub %o2, %o5, %o2
824
825 1: EX_LD(LOAD(ldx, %o1 + 0x08, %g3), memc
826 add %o1, 0x08, %o1
827 subcc %o5, 0x08, %o5
828 srlx %g3, %g2, %g7
829 or %g7, %o4, %g7
830 EX_ST(STORE(stx, %g7, %o0 + 0x00), mem
831 add %o0, 0x08, %o0
832 bne,pt %xcc, 1b
833 sllx %g3, %o3, %o4
834 srl %o3, 3, %o3
835 add %o1, %o3, %o1
836 brz,pn %o2, .Lexit_cp
837 nop
838 ba,pt %xcc, .Lsmall_unaligned_cp
839
840 .Ltiny_cp:
841 EX_LD(LOAD(ldub, %o1 + 0x00, %o3), mem
842 subcc %o2, 1, %o2
843 be,pn %xcc, .Lexit_cp
844 EX_ST(STORE(stb, %o3, %o0 + 0x00), me
845 EX_LD(LOAD(ldub, %o1 + 0x01, %o3), mem
846 subcc %o2, 1, %o2
847 be,pn %xcc, .Lexit_cp
848 EX_ST(STORE(stb, %o3, %o0 + 0x01), me
849 EX_LD(LOAD(ldub, %o1 + 0x02, %o3), mem
850 ba,pt %xcc, .Lexit_cp
851 EX_ST(STORE(stb, %o3, %o0 + 0x02), me
852
853 .Lsmall_cp:
854 andcc %g2, 0x3, %g0
855 bne,pn %xcc, .Lsmall_unaligned_cp
856 andn %o2, 0x4 - 1, %o5
857 sub %o2, %o5, %o2
858 1:
859 EX_LD(LOAD(lduw, %o1 + 0x00, %o3), mem
860 add %o1, 0x04, %o1
861 subcc %o5, 0x04, %o5
862 add %o0, 0x04, %o0
863 bne,pt %xcc, 1b
864 EX_ST(STORE(stw, %o3, %o0 - 0x04), me
865 brz,pt %o2, .Lexit_cp
866 nop
867 ba,a,pt %xcc, .Ltiny_cp
868
869 .Lsmall_unaligned_cp:
870 1: EX_LD(LOAD(ldub, %o1 + 0x00, %o3), mem
871 add %o1, 1, %o1
872 add %o0, 1, %o0
873 subcc %o2, 1, %o2
874 bne,pt %xcc, 1b
875 EX_ST(STORE(stb, %o3, %o0 - 0x01), me
876 ba,a,pt %xcc, .Lexit_cp
877
878 .Lsmallrest:
879 tst %o2
880 bz,pt %xcc, .Lsmallx
881 cmp %o2, 4
882 blt,pn %xcc, .Lsmallleft3
883 nop
884 sub %o2, 3, %o2
885 .Lsmallnotalign4:
886 EX_LD(LOAD(ldub, %o1, %o3), memcpy_ret
887 subcc %o2, 4, %o2 ! redu
888 EX_ST(STORE(stb, %o3, %o0), memcpy_ret
889 EX_LD(LOAD(ldub, %o1+1, %o3), memcpy_r
890 add %o1, 4, %o1 ! adva
891 EX_ST(STORE(stb, %o3, %o0+1), memcpy_r
892 EX_LD(LOAD(ldub, %o1-2, %o3), memcpy_r
893 add %o0, 4, %o0 ! adva
894 EX_ST(STORE(stb, %o3, %o0-2), memcpy_r
895 EX_LD(LOAD(ldub, %o1-1, %o3), memcpy_r
896 bgu,pt %xcc, .Lsmallnotalign4 ! loop
897 EX_ST(STORE(stb, %o3, %o0-1), memcpy_r
898 addcc %o2, 3, %o2 ! rest
899 bz,pt %xcc, .Lsmallx
900 .Lsmallleft3: ! 1, 2
901 subcc %o2, 1, %o2
902 EX_LD(LOAD(ldub, %o1, %o3), memcpy_ret
903 bz,pt %xcc, .Lsmallx
904 EX_ST(STORE(stb, %o3, %o0), memcpy_ret
905 EX_LD(LOAD(ldub, %o1+1, %o3), memcpy_r
906 subcc %o2, 1, %o2
907 bz,pt %xcc, .Lsmallx
908 EX_ST(STORE(stb, %o3, %o0+1), memcpy_r
909 EX_LD(LOAD(ldub, %o1+2, %o3), memcpy_r
910 EX_ST(STORE(stb, %o3, %o0+2), memcpy_r
911 .Lsmallx:
912 retl
913 mov EX_RETVAL(%g1), %o0
914 .Lsmallfin:
915 tst %o2
916 bnz,pn %xcc, .Lsmallleft3
917 nop
918 retl
919 mov EX_RETVAL(%g1), %o0 ! rest
920 .Lexit_cp:
921 retl
922 mov EX_RETVAL(%g1), %o0
923 .size FUNC_NAME, .-FUNC_NAME
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.