Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/* U3memcpy.S: UltraSparc-III optimized memcpy.
3 *
4 * Copyright (C) 1999, 2000, 2004 David S. Miller (davem@redhat.com)
5 */
6
7#ifdef __KERNEL__
8#include <linux/linkage.h>
9#include <asm/visasm.h>
10#include <asm/asi.h>
11#define GLOBAL_SPARE %g7
12#else
13#define ASI_BLK_P 0xf0
14#define FPRS_FEF 0x04
15#ifdef MEMCPY_DEBUG
16#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs; \
17 clr %g1; clr %g2; clr %g3; subcc %g0, %g0, %g0;
18#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
19#else
20#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs
21#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
22#endif
23#define GLOBAL_SPARE %g5
24#endif
25
26#ifndef EX_LD
27#define EX_LD(x,y) x
28#endif
29#ifndef EX_LD_FP
30#define EX_LD_FP(x,y) x
31#endif
32
33#ifndef EX_ST
34#define EX_ST(x,y) x
35#endif
36#ifndef EX_ST_FP
37#define EX_ST_FP(x,y) x
38#endif
39
40#ifndef LOAD
41#define LOAD(type,addr,dest) type [addr], dest
42#endif
43
44#ifndef STORE
45#define STORE(type,src,addr) type src, [addr]
46#endif
47
48#ifndef STORE_BLK
49#define STORE_BLK(src,addr) stda src, [addr] ASI_BLK_P
50#endif
51
52#ifndef FUNC_NAME
53#define FUNC_NAME U3memcpy
54#endif
55
56#ifndef PREAMBLE
57#define PREAMBLE
58#endif
59
60#ifndef XCC
61#define XCC xcc
62#endif
63
64 .register %g2,#scratch
65 .register %g3,#scratch
66
67 /* Special/non-trivial issues of this code:
68 *
69 * 1) %o5 is preserved from VISEntryHalf to VISExitHalf
70 * 2) Only low 32 FPU registers are used so that only the
71 * lower half of the FPU register set is dirtied by this
72 * code. This is especially important in the kernel.
73 * 3) This code never prefetches cachelines past the end
74 * of the source buffer.
75 */
76
77 .text
78#ifndef EX_RETVAL
79#define EX_RETVAL(x) x
80__restore_fp:
81 VISExitHalf
82 retl
83 nop
84ENTRY(U3_retl_o2_plus_g2_plus_g1_plus_1_fp)
85 add %g1, 1, %g1
86 add %g2, %g1, %g2
87 ba,pt %xcc, __restore_fp
88 add %o2, %g2, %o0
89ENDPROC(U3_retl_o2_plus_g2_plus_g1_plus_1_fp)
90ENTRY(U3_retl_o2_plus_g2_fp)
91 ba,pt %xcc, __restore_fp
92 add %o2, %g2, %o0
93ENDPROC(U3_retl_o2_plus_g2_fp)
94ENTRY(U3_retl_o2_plus_g2_plus_8_fp)
95 add %g2, 8, %g2
96 ba,pt %xcc, __restore_fp
97 add %o2, %g2, %o0
98ENDPROC(U3_retl_o2_plus_g2_plus_8_fp)
99ENTRY(U3_retl_o2)
100 retl
101 mov %o2, %o0
102ENDPROC(U3_retl_o2)
103ENTRY(U3_retl_o2_plus_1)
104 retl
105 add %o2, 1, %o0
106ENDPROC(U3_retl_o2_plus_1)
107ENTRY(U3_retl_o2_plus_4)
108 retl
109 add %o2, 4, %o0
110ENDPROC(U3_retl_o2_plus_4)
111ENTRY(U3_retl_o2_plus_8)
112 retl
113 add %o2, 8, %o0
114ENDPROC(U3_retl_o2_plus_8)
115ENTRY(U3_retl_o2_plus_g1_plus_1)
116 add %g1, 1, %g1
117 retl
118 add %o2, %g1, %o0
119ENDPROC(U3_retl_o2_plus_g1_plus_1)
120ENTRY(U3_retl_o2_fp)
121 ba,pt %xcc, __restore_fp
122 mov %o2, %o0
123ENDPROC(U3_retl_o2_fp)
124ENTRY(U3_retl_o2_plus_o3_sll_6_plus_0x80_fp)
125 sll %o3, 6, %o3
126 add %o3, 0x80, %o3
127 ba,pt %xcc, __restore_fp
128 add %o2, %o3, %o0
129ENDPROC(U3_retl_o2_plus_o3_sll_6_plus_0x80_fp)
130ENTRY(U3_retl_o2_plus_o3_sll_6_plus_0x40_fp)
131 sll %o3, 6, %o3
132 add %o3, 0x40, %o3
133 ba,pt %xcc, __restore_fp
134 add %o2, %o3, %o0
135ENDPROC(U3_retl_o2_plus_o3_sll_6_plus_0x40_fp)
136ENTRY(U3_retl_o2_plus_GS_plus_0x10)
137 add GLOBAL_SPARE, 0x10, GLOBAL_SPARE
138 retl
139 add %o2, GLOBAL_SPARE, %o0
140ENDPROC(U3_retl_o2_plus_GS_plus_0x10)
141ENTRY(U3_retl_o2_plus_GS_plus_0x08)
142 add GLOBAL_SPARE, 0x08, GLOBAL_SPARE
143 retl
144 add %o2, GLOBAL_SPARE, %o0
145ENDPROC(U3_retl_o2_plus_GS_plus_0x08)
146ENTRY(U3_retl_o2_and_7_plus_GS)
147 and %o2, 7, %o2
148 retl
149 add %o2, GLOBAL_SPARE, %o0
150ENDPROC(U3_retl_o2_and_7_plus_GS)
151ENTRY(U3_retl_o2_and_7_plus_GS_plus_8)
152 add GLOBAL_SPARE, 8, GLOBAL_SPARE
153 and %o2, 7, %o2
154 retl
155 add %o2, GLOBAL_SPARE, %o0
156ENDPROC(U3_retl_o2_and_7_plus_GS_plus_8)
157#endif
158
159 .align 64
160
161 /* The cheetah's flexible spine, oversized liver, enlarged heart,
162 * slender muscular body, and claws make it the swiftest hunter
163 * in Africa and the fastest animal on land. Can reach speeds
164 * of up to 2.4GB per second.
165 */
166
167 .globl FUNC_NAME
168 .type FUNC_NAME,#function
169FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
170 srlx %o2, 31, %g2
171 cmp %g2, 0
172
173 /* software trap 5 "Range Check" if dst >= 0x80000000 */
174 tne %xcc, 5
175 PREAMBLE
176 mov %o0, %o4
177
178 /* if len == 0 */
179 cmp %o2, 0
180 be,pn %XCC, end_return
181 or %o0, %o1, %o3
182
183 /* if len < 16 */
184 cmp %o2, 16
185 blu,a,pn %XCC, less_than_16
186 or %o3, %o2, %o3
187
188 /* if len < 192 */
189 cmp %o2, (3 * 64)
190 blu,pt %XCC, less_than_192
191 andcc %o3, 0x7, %g0
192
193 /* Clobbers o5/g1/g2/g3/g7/icc/xcc. We must preserve
194 * o5 from here until we hit VISExitHalf.
195 */
196 VISEntryHalf
197
198 /* Is 'dst' already aligned on an 64-byte boundary? */
199 andcc %o0, 0x3f, %g2
200 be,pt %XCC, 2f
201
202 /* Compute abs((dst & 0x3f) - 0x40) into %g2. This is the number
203 * of bytes to copy to make 'dst' 64-byte aligned. We pre-
204 * subtract this from 'len'.
205 */
206 sub %o0, %o1, GLOBAL_SPARE
207 sub %g2, 0x40, %g2
208 sub %g0, %g2, %g2
209 sub %o2, %g2, %o2
210 andcc %g2, 0x7, %g1
211 be,pt %icc, 2f
212 and %g2, 0x38, %g2
213
2141: subcc %g1, 0x1, %g1
215 EX_LD_FP(LOAD(ldub, %o1 + 0x00, %o3), U3_retl_o2_plus_g2_plus_g1_plus_1)
216 EX_ST_FP(STORE(stb, %o3, %o1 + GLOBAL_SPARE), U3_retl_o2_plus_g2_plus_g1_plus_1)
217 bgu,pt %XCC, 1b
218 add %o1, 0x1, %o1
219
220 add %o1, GLOBAL_SPARE, %o0
221
2222: cmp %g2, 0x0
223 and %o1, 0x7, %g1
224 be,pt %icc, 3f
225 alignaddr %o1, %g0, %o1
226
227 EX_LD_FP(LOAD(ldd, %o1, %f4), U3_retl_o2_plus_g2)
2281: EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f6), U3_retl_o2_plus_g2)
229 add %o1, 0x8, %o1
230 subcc %g2, 0x8, %g2
231 faligndata %f4, %f6, %f0
232 EX_ST_FP(STORE(std, %f0, %o0), U3_retl_o2_plus_g2_plus_8)
233 be,pn %icc, 3f
234 add %o0, 0x8, %o0
235
236 EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f4), U3_retl_o2_plus_g2)
237 add %o1, 0x8, %o1
238 subcc %g2, 0x8, %g2
239 faligndata %f6, %f4, %f2
240 EX_ST_FP(STORE(std, %f2, %o0), U3_retl_o2_plus_g2_plus_8)
241 bne,pt %icc, 1b
242 add %o0, 0x8, %o0
243
2443: LOAD(prefetch, %o1 + 0x000, #one_read)
245 LOAD(prefetch, %o1 + 0x040, #one_read)
246 andn %o2, (0x40 - 1), GLOBAL_SPARE
247 LOAD(prefetch, %o1 + 0x080, #one_read)
248 LOAD(prefetch, %o1 + 0x0c0, #one_read)
249 LOAD(prefetch, %o1 + 0x100, #one_read)
250 EX_LD_FP(LOAD(ldd, %o1 + 0x000, %f0), U3_retl_o2)
251 LOAD(prefetch, %o1 + 0x140, #one_read)
252 EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2), U3_retl_o2)
253 LOAD(prefetch, %o1 + 0x180, #one_read)
254 EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4), U3_retl_o2)
255 LOAD(prefetch, %o1 + 0x1c0, #one_read)
256 faligndata %f0, %f2, %f16
257 EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6), U3_retl_o2)
258 faligndata %f2, %f4, %f18
259 EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8), U3_retl_o2)
260 faligndata %f4, %f6, %f20
261 EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10), U3_retl_o2)
262 faligndata %f6, %f8, %f22
263
264 EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12), U3_retl_o2)
265 faligndata %f8, %f10, %f24
266 EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14), U3_retl_o2)
267 faligndata %f10, %f12, %f26
268 EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0), U3_retl_o2)
269
270 subcc GLOBAL_SPARE, 0x80, GLOBAL_SPARE
271 add %o1, 0x40, %o1
272 bgu,pt %XCC, 1f
273 srl GLOBAL_SPARE, 6, %o3
274 ba,pt %xcc, 2f
275 nop
276
277 .align 64
2781:
279 EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2), U3_retl_o2_plus_o3_sll_6_plus_0x80)
280 faligndata %f12, %f14, %f28
281 EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4), U3_retl_o2_plus_o3_sll_6_plus_0x80)
282 faligndata %f14, %f0, %f30
283 EX_ST_FP(STORE_BLK(%f16, %o0), U3_retl_o2_plus_o3_sll_6_plus_0x80)
284 EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6), U3_retl_o2_plus_o3_sll_6_plus_0x40)
285 faligndata %f0, %f2, %f16
286 add %o0, 0x40, %o0
287
288 EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8), U3_retl_o2_plus_o3_sll_6_plus_0x40)
289 faligndata %f2, %f4, %f18
290 EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10), U3_retl_o2_plus_o3_sll_6_plus_0x40)
291 faligndata %f4, %f6, %f20
292 EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12), U3_retl_o2_plus_o3_sll_6_plus_0x40)
293 subcc %o3, 0x01, %o3
294 faligndata %f6, %f8, %f22
295 EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14), U3_retl_o2_plus_o3_sll_6_plus_0x80)
296
297 faligndata %f8, %f10, %f24
298 EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0), U3_retl_o2_plus_o3_sll_6_plus_0x80)
299 LOAD(prefetch, %o1 + 0x1c0, #one_read)
300 faligndata %f10, %f12, %f26
301 bg,pt %XCC, 1b
302 add %o1, 0x40, %o1
303
304 /* Finally we copy the last full 64-byte block. */
3052:
306 EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2), U3_retl_o2_plus_o3_sll_6_plus_0x80)
307 faligndata %f12, %f14, %f28
308 EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4), U3_retl_o2_plus_o3_sll_6_plus_0x80)
309 faligndata %f14, %f0, %f30
310 EX_ST_FP(STORE_BLK(%f16, %o0), U3_retl_o2_plus_o3_sll_6_plus_0x80)
311 EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6), U3_retl_o2_plus_o3_sll_6_plus_0x40)
312 faligndata %f0, %f2, %f16
313 EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8), U3_retl_o2_plus_o3_sll_6_plus_0x40)
314 faligndata %f2, %f4, %f18
315 EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10), U3_retl_o2_plus_o3_sll_6_plus_0x40)
316 faligndata %f4, %f6, %f20
317 EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12), U3_retl_o2_plus_o3_sll_6_plus_0x40)
318 faligndata %f6, %f8, %f22
319 EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14), U3_retl_o2_plus_o3_sll_6_plus_0x40)
320 faligndata %f8, %f10, %f24
321 cmp %g1, 0
322 be,pt %XCC, 1f
323 add %o0, 0x40, %o0
324 EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0), U3_retl_o2_plus_o3_sll_6_plus_0x40)
3251: faligndata %f10, %f12, %f26
326 faligndata %f12, %f14, %f28
327 faligndata %f14, %f0, %f30
328 EX_ST_FP(STORE_BLK(%f16, %o0), U3_retl_o2_plus_o3_sll_6_plus_0x40)
329 add %o0, 0x40, %o0
330 add %o1, 0x40, %o1
331 membar #Sync
332
333 /* Now we copy the (len modulo 64) bytes at the end.
334 * Note how we borrow the %f0 loaded above.
335 *
336 * Also notice how this code is careful not to perform a
337 * load past the end of the src buffer.
338 */
339 and %o2, 0x3f, %o2
340 andcc %o2, 0x38, %g2
341 be,pn %XCC, 2f
342 subcc %g2, 0x8, %g2
343 be,pn %XCC, 2f
344 cmp %g1, 0
345
346 sub %o2, %g2, %o2
347 be,a,pt %XCC, 1f
348 EX_LD_FP(LOAD(ldd, %o1 + 0x00, %f0), U3_retl_o2_plus_g2)
349
3501: EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f2), U3_retl_o2_plus_g2)
351 add %o1, 0x8, %o1
352 subcc %g2, 0x8, %g2
353 faligndata %f0, %f2, %f8
354 EX_ST_FP(STORE(std, %f8, %o0), U3_retl_o2_plus_g2_plus_8)
355 be,pn %XCC, 2f
356 add %o0, 0x8, %o0
357 EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f0), U3_retl_o2_plus_g2)
358 add %o1, 0x8, %o1
359 subcc %g2, 0x8, %g2
360 faligndata %f2, %f0, %f8
361 EX_ST_FP(STORE(std, %f8, %o0), U3_retl_o2_plus_g2_plus_8)
362 bne,pn %XCC, 1b
363 add %o0, 0x8, %o0
364
365 /* If anything is left, we copy it one byte at a time.
366 * Note that %g1 is (src & 0x3) saved above before the
367 * alignaddr was performed.
368 */
3692:
370 cmp %o2, 0
371 add %o1, %g1, %o1
372 VISExitHalf
373 be,pn %XCC, end_return
374 sub %o0, %o1, %o3
375
376 andcc %g1, 0x7, %g0
377 bne,pn %icc, 90f
378 andcc %o2, 0x8, %g0
379 be,pt %icc, 1f
380 nop
381 EX_LD(LOAD(ldx, %o1, %o5), U3_retl_o2)
382 EX_ST(STORE(stx, %o5, %o1 + %o3), U3_retl_o2)
383 add %o1, 0x8, %o1
384 sub %o2, 8, %o2
385
3861: andcc %o2, 0x4, %g0
387 be,pt %icc, 1f
388 nop
389 EX_LD(LOAD(lduw, %o1, %o5), U3_retl_o2)
390 EX_ST(STORE(stw, %o5, %o1 + %o3), U3_retl_o2)
391 add %o1, 0x4, %o1
392 sub %o2, 4, %o2
393
3941: andcc %o2, 0x2, %g0
395 be,pt %icc, 1f
396 nop
397 EX_LD(LOAD(lduh, %o1, %o5), U3_retl_o2)
398 EX_ST(STORE(sth, %o5, %o1 + %o3), U3_retl_o2)
399 add %o1, 0x2, %o1
400 sub %o2, 2, %o2
401
4021: andcc %o2, 0x1, %g0
403 be,pt %icc, end_return
404 nop
405 EX_LD(LOAD(ldub, %o1, %o5), U3_retl_o2)
406 ba,pt %xcc, end_return
407 EX_ST(STORE(stb, %o5, %o1 + %o3), U3_retl_o2)
408
409 .align 64
410 /* 16 <= len < 192 */
411less_than_192:
412 bne,pn %XCC, 75f
413 sub %o0, %o1, %o3
414
41572:
416 andn %o2, 0xf, GLOBAL_SPARE
417 and %o2, 0xf, %o2
4181: subcc GLOBAL_SPARE, 0x10, GLOBAL_SPARE
419 EX_LD(LOAD(ldx, %o1 + 0x00, %o5), U3_retl_o2_plus_GS_plus_0x10)
420 EX_LD(LOAD(ldx, %o1 + 0x08, %g1), U3_retl_o2_plus_GS_plus_0x10)
421 EX_ST(STORE(stx, %o5, %o1 + %o3), U3_retl_o2_plus_GS_plus_0x10)
422 add %o1, 0x8, %o1
423 EX_ST(STORE(stx, %g1, %o1 + %o3), U3_retl_o2_plus_GS_plus_0x08)
424 bgu,pt %XCC, 1b
425 add %o1, 0x8, %o1
42673: andcc %o2, 0x8, %g0
427 be,pt %XCC, 1f
428 nop
429 sub %o2, 0x8, %o2
430 EX_LD(LOAD(ldx, %o1, %o5), U3_retl_o2_plus_8)
431 EX_ST(STORE(stx, %o5, %o1 + %o3), U3_retl_o2_plus_8)
432 add %o1, 0x8, %o1
4331: andcc %o2, 0x4, %g0
434 be,pt %XCC, 1f
435 nop
436 sub %o2, 0x4, %o2
437 EX_LD(LOAD(lduw, %o1, %o5), U3_retl_o2_plus_4)
438 EX_ST(STORE(stw, %o5, %o1 + %o3), U3_retl_o2_plus_4)
439 add %o1, 0x4, %o1
4401: cmp %o2, 0
441 be,pt %XCC, end_return
442 nop
443 ba,pt %xcc, 90f
444 nop
445
44675:
447 andcc %o0, 0x7, %g1
448 sub %g1, 0x8, %g1
449 be,pn %icc, 2f
450 sub %g0, %g1, %g1
451 sub %o2, %g1, %o2
452
4531: subcc %g1, 1, %g1
454 EX_LD(LOAD(ldub, %o1, %o5), U3_retl_o2_plus_g1_plus_1)
455 EX_ST(STORE(stb, %o5, %o1 + %o3), U3_retl_o2_plus_g1_plus_1)
456 bgu,pt %icc, 1b
457 add %o1, 1, %o1
458
4592: add %o1, %o3, %o0
460 andcc %o1, 0x7, %g1
461 bne,pt %icc, 8f
462 sll %g1, 3, %g1
463
464 cmp %o2, 16
465 bgeu,pt %icc, 72b
466 nop
467 ba,a,pt %xcc, 73b
468
4698: mov 64, %o3
470 andn %o1, 0x7, %o1
471 EX_LD(LOAD(ldx, %o1, %g2), U3_retl_o2)
472 sub %o3, %g1, %o3
473 andn %o2, 0x7, GLOBAL_SPARE
474 sllx %g2, %g1, %g2
4751: EX_LD(LOAD(ldx, %o1 + 0x8, %g3), U3_retl_o2_and_7_plus_GS)
476 subcc GLOBAL_SPARE, 0x8, GLOBAL_SPARE
477 add %o1, 0x8, %o1
478 srlx %g3, %o3, %o5
479 or %o5, %g2, %o5
480 EX_ST(STORE(stx, %o5, %o0), U3_retl_o2_and_7_plus_GS_plus_8)
481 add %o0, 0x8, %o0
482 bgu,pt %icc, 1b
483 sllx %g3, %g1, %g2
484
485 srl %g1, 3, %g1
486 andcc %o2, 0x7, %o2
487 be,pn %icc, end_return
488 add %o1, %g1, %o1
489 ba,pt %xcc, 90f
490 sub %o0, %o1, %o3
491
492 .align 64
493 /* 0 < len < 16 */
494less_than_16:
495 andcc %o3, 0x3, %g0
496 bne,pn %XCC, 90f
497 sub %o0, %o1, %o3
498
4991:
500 subcc %o2, 4, %o2
501 EX_LD(LOAD(lduw, %o1, %g1), U3_retl_o2_plus_4)
502 EX_ST(STORE(stw, %g1, %o1 + %o3), U3_retl_o2_plus_4)
503 bgu,pt %XCC, 1b
504 add %o1, 4, %o1
505
506end_return:
507 retl
508 mov EX_RETVAL(%o4), %o0
509
510 .align 32
51190:
512 subcc %o2, 1, %o2
513 EX_LD(LOAD(ldub, %o1, %g1), U3_retl_o2_plus_1)
514 EX_ST(STORE(stb, %g1, %o1 + %o3), U3_retl_o2_plus_1)
515 bgu,pt %XCC, 90b
516 add %o1, 1, %o1
517 retl
518 mov EX_RETVAL(%o4), %o0
519
520 .size FUNC_NAME, .-FUNC_NAME
1/* U3memcpy.S: UltraSparc-III optimized memcpy.
2 *
3 * Copyright (C) 1999, 2000, 2004 David S. Miller (davem@redhat.com)
4 */
5
6#ifdef __KERNEL__
7#include <asm/visasm.h>
8#include <asm/asi.h>
9#define GLOBAL_SPARE %g7
10#else
11#define ASI_BLK_P 0xf0
12#define FPRS_FEF 0x04
13#ifdef MEMCPY_DEBUG
14#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs; \
15 clr %g1; clr %g2; clr %g3; subcc %g0, %g0, %g0;
16#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
17#else
18#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs
19#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
20#endif
21#define GLOBAL_SPARE %g5
22#endif
23
24#ifndef EX_LD
25#define EX_LD(x) x
26#endif
27#ifndef EX_LD_FP
28#define EX_LD_FP(x) x
29#endif
30
31#ifndef EX_ST
32#define EX_ST(x) x
33#endif
34#ifndef EX_ST_FP
35#define EX_ST_FP(x) x
36#endif
37
38#ifndef EX_RETVAL
39#define EX_RETVAL(x) x
40#endif
41
42#ifndef LOAD
43#define LOAD(type,addr,dest) type [addr], dest
44#endif
45
46#ifndef STORE
47#define STORE(type,src,addr) type src, [addr]
48#endif
49
50#ifndef STORE_BLK
51#define STORE_BLK(src,addr) stda src, [addr] ASI_BLK_P
52#endif
53
54#ifndef FUNC_NAME
55#define FUNC_NAME U3memcpy
56#endif
57
58#ifndef PREAMBLE
59#define PREAMBLE
60#endif
61
62#ifndef XCC
63#define XCC xcc
64#endif
65
66 .register %g2,#scratch
67 .register %g3,#scratch
68
69 /* Special/non-trivial issues of this code:
70 *
71 * 1) %o5 is preserved from VISEntryHalf to VISExitHalf
72 * 2) Only low 32 FPU registers are used so that only the
73 * lower half of the FPU register set is dirtied by this
74 * code. This is especially important in the kernel.
75 * 3) This code never prefetches cachelines past the end
76 * of the source buffer.
77 */
78
79 .text
80 .align 64
81
82 /* The cheetah's flexible spine, oversized liver, enlarged heart,
83 * slender muscular body, and claws make it the swiftest hunter
84 * in Africa and the fastest animal on land. Can reach speeds
85 * of up to 2.4GB per second.
86 */
87
88 .globl FUNC_NAME
89 .type FUNC_NAME,#function
90FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */
91 srlx %o2, 31, %g2
92 cmp %g2, 0
93 tne %xcc, 5
94 PREAMBLE
95 mov %o0, %o4
96 cmp %o2, 0
97 be,pn %XCC, 85f
98 or %o0, %o1, %o3
99 cmp %o2, 16
100 blu,a,pn %XCC, 80f
101 or %o3, %o2, %o3
102
103 cmp %o2, (3 * 64)
104 blu,pt %XCC, 70f
105 andcc %o3, 0x7, %g0
106
107 /* Clobbers o5/g1/g2/g3/g7/icc/xcc. We must preserve
108 * o5 from here until we hit VISExitHalf.
109 */
110 VISEntryHalf
111
112 /* Is 'dst' already aligned on an 64-byte boundary? */
113 andcc %o0, 0x3f, %g2
114 be,pt %XCC, 2f
115
116 /* Compute abs((dst & 0x3f) - 0x40) into %g2. This is the number
117 * of bytes to copy to make 'dst' 64-byte aligned. We pre-
118 * subtract this from 'len'.
119 */
120 sub %o0, %o1, GLOBAL_SPARE
121 sub %g2, 0x40, %g2
122 sub %g0, %g2, %g2
123 sub %o2, %g2, %o2
124 andcc %g2, 0x7, %g1
125 be,pt %icc, 2f
126 and %g2, 0x38, %g2
127
1281: subcc %g1, 0x1, %g1
129 EX_LD_FP(LOAD(ldub, %o1 + 0x00, %o3))
130 EX_ST_FP(STORE(stb, %o3, %o1 + GLOBAL_SPARE))
131 bgu,pt %XCC, 1b
132 add %o1, 0x1, %o1
133
134 add %o1, GLOBAL_SPARE, %o0
135
1362: cmp %g2, 0x0
137 and %o1, 0x7, %g1
138 be,pt %icc, 3f
139 alignaddr %o1, %g0, %o1
140
141 EX_LD_FP(LOAD(ldd, %o1, %f4))
1421: EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f6))
143 add %o1, 0x8, %o1
144 subcc %g2, 0x8, %g2
145 faligndata %f4, %f6, %f0
146 EX_ST_FP(STORE(std, %f0, %o0))
147 be,pn %icc, 3f
148 add %o0, 0x8, %o0
149
150 EX_LD_FP(LOAD(ldd, %o1 + 0x8, %f4))
151 add %o1, 0x8, %o1
152 subcc %g2, 0x8, %g2
153 faligndata %f6, %f4, %f2
154 EX_ST_FP(STORE(std, %f2, %o0))
155 bne,pt %icc, 1b
156 add %o0, 0x8, %o0
157
1583: LOAD(prefetch, %o1 + 0x000, #one_read)
159 LOAD(prefetch, %o1 + 0x040, #one_read)
160 andn %o2, (0x40 - 1), GLOBAL_SPARE
161 LOAD(prefetch, %o1 + 0x080, #one_read)
162 LOAD(prefetch, %o1 + 0x0c0, #one_read)
163 LOAD(prefetch, %o1 + 0x100, #one_read)
164 EX_LD_FP(LOAD(ldd, %o1 + 0x000, %f0))
165 LOAD(prefetch, %o1 + 0x140, #one_read)
166 EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2))
167 LOAD(prefetch, %o1 + 0x180, #one_read)
168 EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4))
169 LOAD(prefetch, %o1 + 0x1c0, #one_read)
170 faligndata %f0, %f2, %f16
171 EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6))
172 faligndata %f2, %f4, %f18
173 EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8))
174 faligndata %f4, %f6, %f20
175 EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10))
176 faligndata %f6, %f8, %f22
177
178 EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12))
179 faligndata %f8, %f10, %f24
180 EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14))
181 faligndata %f10, %f12, %f26
182 EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0))
183
184 subcc GLOBAL_SPARE, 0x80, GLOBAL_SPARE
185 add %o1, 0x40, %o1
186 bgu,pt %XCC, 1f
187 srl GLOBAL_SPARE, 6, %o3
188 ba,pt %xcc, 2f
189 nop
190
191 .align 64
1921:
193 EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2))
194 faligndata %f12, %f14, %f28
195 EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4))
196 faligndata %f14, %f0, %f30
197 EX_ST_FP(STORE_BLK(%f16, %o0))
198 EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6))
199 faligndata %f0, %f2, %f16
200 add %o0, 0x40, %o0
201
202 EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8))
203 faligndata %f2, %f4, %f18
204 EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10))
205 faligndata %f4, %f6, %f20
206 EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12))
207 subcc %o3, 0x01, %o3
208 faligndata %f6, %f8, %f22
209 EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14))
210
211 faligndata %f8, %f10, %f24
212 EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0))
213 LOAD(prefetch, %o1 + 0x1c0, #one_read)
214 faligndata %f10, %f12, %f26
215 bg,pt %XCC, 1b
216 add %o1, 0x40, %o1
217
218 /* Finally we copy the last full 64-byte block. */
2192:
220 EX_LD_FP(LOAD(ldd, %o1 + 0x008, %f2))
221 faligndata %f12, %f14, %f28
222 EX_LD_FP(LOAD(ldd, %o1 + 0x010, %f4))
223 faligndata %f14, %f0, %f30
224 EX_ST_FP(STORE_BLK(%f16, %o0))
225 EX_LD_FP(LOAD(ldd, %o1 + 0x018, %f6))
226 faligndata %f0, %f2, %f16
227 EX_LD_FP(LOAD(ldd, %o1 + 0x020, %f8))
228 faligndata %f2, %f4, %f18
229 EX_LD_FP(LOAD(ldd, %o1 + 0x028, %f10))
230 faligndata %f4, %f6, %f20
231 EX_LD_FP(LOAD(ldd, %o1 + 0x030, %f12))
232 faligndata %f6, %f8, %f22
233 EX_LD_FP(LOAD(ldd, %o1 + 0x038, %f14))
234 faligndata %f8, %f10, %f24
235 cmp %g1, 0
236 be,pt %XCC, 1f
237 add %o0, 0x40, %o0
238 EX_LD_FP(LOAD(ldd, %o1 + 0x040, %f0))
2391: faligndata %f10, %f12, %f26
240 faligndata %f12, %f14, %f28
241 faligndata %f14, %f0, %f30
242 EX_ST_FP(STORE_BLK(%f16, %o0))
243 add %o0, 0x40, %o0
244 add %o1, 0x40, %o1
245 membar #Sync
246
247 /* Now we copy the (len modulo 64) bytes at the end.
248 * Note how we borrow the %f0 loaded above.
249 *
250 * Also notice how this code is careful not to perform a
251 * load past the end of the src buffer.
252 */
253 and %o2, 0x3f, %o2
254 andcc %o2, 0x38, %g2
255 be,pn %XCC, 2f
256 subcc %g2, 0x8, %g2
257 be,pn %XCC, 2f
258 cmp %g1, 0
259
260 sub %o2, %g2, %o2
261 be,a,pt %XCC, 1f
262 EX_LD_FP(LOAD(ldd, %o1 + 0x00, %f0))
263
2641: EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f2))
265 add %o1, 0x8, %o1
266 subcc %g2, 0x8, %g2
267 faligndata %f0, %f2, %f8
268 EX_ST_FP(STORE(std, %f8, %o0))
269 be,pn %XCC, 2f
270 add %o0, 0x8, %o0
271 EX_LD_FP(LOAD(ldd, %o1 + 0x08, %f0))
272 add %o1, 0x8, %o1
273 subcc %g2, 0x8, %g2
274 faligndata %f2, %f0, %f8
275 EX_ST_FP(STORE(std, %f8, %o0))
276 bne,pn %XCC, 1b
277 add %o0, 0x8, %o0
278
279 /* If anything is left, we copy it one byte at a time.
280 * Note that %g1 is (src & 0x3) saved above before the
281 * alignaddr was performed.
282 */
2832:
284 cmp %o2, 0
285 add %o1, %g1, %o1
286 VISExitHalf
287 be,pn %XCC, 85f
288 sub %o0, %o1, %o3
289
290 andcc %g1, 0x7, %g0
291 bne,pn %icc, 90f
292 andcc %o2, 0x8, %g0
293 be,pt %icc, 1f
294 nop
295 EX_LD(LOAD(ldx, %o1, %o5))
296 EX_ST(STORE(stx, %o5, %o1 + %o3))
297 add %o1, 0x8, %o1
298
2991: andcc %o2, 0x4, %g0
300 be,pt %icc, 1f
301 nop
302 EX_LD(LOAD(lduw, %o1, %o5))
303 EX_ST(STORE(stw, %o5, %o1 + %o3))
304 add %o1, 0x4, %o1
305
3061: andcc %o2, 0x2, %g0
307 be,pt %icc, 1f
308 nop
309 EX_LD(LOAD(lduh, %o1, %o5))
310 EX_ST(STORE(sth, %o5, %o1 + %o3))
311 add %o1, 0x2, %o1
312
3131: andcc %o2, 0x1, %g0
314 be,pt %icc, 85f
315 nop
316 EX_LD(LOAD(ldub, %o1, %o5))
317 ba,pt %xcc, 85f
318 EX_ST(STORE(stb, %o5, %o1 + %o3))
319
320 .align 64
32170: /* 16 < len <= 64 */
322 bne,pn %XCC, 75f
323 sub %o0, %o1, %o3
324
32572:
326 andn %o2, 0xf, GLOBAL_SPARE
327 and %o2, 0xf, %o2
3281: subcc GLOBAL_SPARE, 0x10, GLOBAL_SPARE
329 EX_LD(LOAD(ldx, %o1 + 0x00, %o5))
330 EX_LD(LOAD(ldx, %o1 + 0x08, %g1))
331 EX_ST(STORE(stx, %o5, %o1 + %o3))
332 add %o1, 0x8, %o1
333 EX_ST(STORE(stx, %g1, %o1 + %o3))
334 bgu,pt %XCC, 1b
335 add %o1, 0x8, %o1
33673: andcc %o2, 0x8, %g0
337 be,pt %XCC, 1f
338 nop
339 sub %o2, 0x8, %o2
340 EX_LD(LOAD(ldx, %o1, %o5))
341 EX_ST(STORE(stx, %o5, %o1 + %o3))
342 add %o1, 0x8, %o1
3431: andcc %o2, 0x4, %g0
344 be,pt %XCC, 1f
345 nop
346 sub %o2, 0x4, %o2
347 EX_LD(LOAD(lduw, %o1, %o5))
348 EX_ST(STORE(stw, %o5, %o1 + %o3))
349 add %o1, 0x4, %o1
3501: cmp %o2, 0
351 be,pt %XCC, 85f
352 nop
353 ba,pt %xcc, 90f
354 nop
355
35675:
357 andcc %o0, 0x7, %g1
358 sub %g1, 0x8, %g1
359 be,pn %icc, 2f
360 sub %g0, %g1, %g1
361 sub %o2, %g1, %o2
362
3631: subcc %g1, 1, %g1
364 EX_LD(LOAD(ldub, %o1, %o5))
365 EX_ST(STORE(stb, %o5, %o1 + %o3))
366 bgu,pt %icc, 1b
367 add %o1, 1, %o1
368
3692: add %o1, %o3, %o0
370 andcc %o1, 0x7, %g1
371 bne,pt %icc, 8f
372 sll %g1, 3, %g1
373
374 cmp %o2, 16
375 bgeu,pt %icc, 72b
376 nop
377 ba,a,pt %xcc, 73b
378
3798: mov 64, %o3
380 andn %o1, 0x7, %o1
381 EX_LD(LOAD(ldx, %o1, %g2))
382 sub %o3, %g1, %o3
383 andn %o2, 0x7, GLOBAL_SPARE
384 sllx %g2, %g1, %g2
3851: EX_LD(LOAD(ldx, %o1 + 0x8, %g3))
386 subcc GLOBAL_SPARE, 0x8, GLOBAL_SPARE
387 add %o1, 0x8, %o1
388 srlx %g3, %o3, %o5
389 or %o5, %g2, %o5
390 EX_ST(STORE(stx, %o5, %o0))
391 add %o0, 0x8, %o0
392 bgu,pt %icc, 1b
393 sllx %g3, %g1, %g2
394
395 srl %g1, 3, %g1
396 andcc %o2, 0x7, %o2
397 be,pn %icc, 85f
398 add %o1, %g1, %o1
399 ba,pt %xcc, 90f
400 sub %o0, %o1, %o3
401
402 .align 64
40380: /* 0 < len <= 16 */
404 andcc %o3, 0x3, %g0
405 bne,pn %XCC, 90f
406 sub %o0, %o1, %o3
407
4081:
409 subcc %o2, 4, %o2
410 EX_LD(LOAD(lduw, %o1, %g1))
411 EX_ST(STORE(stw, %g1, %o1 + %o3))
412 bgu,pt %XCC, 1b
413 add %o1, 4, %o1
414
41585: retl
416 mov EX_RETVAL(%o4), %o0
417
418 .align 32
41990:
420 subcc %o2, 1, %o2
421 EX_LD(LOAD(ldub, %o1, %g1))
422 EX_ST(STORE(stb, %g1, %o1 + %o3))
423 bgu,pt %XCC, 90b
424 add %o1, 1, %o1
425 retl
426 mov EX_RETVAL(%o4), %o0
427
428 .size FUNC_NAME, .-FUNC_NAME