1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
|
dnl AMD64 mpn_sqr_basecase optimised for AMD bobcat.
dnl Copyright 2003-2005, 2007, 2008, 2011, 2012 Free Software Foundation, Inc.
dnl This file is part of the GNU MP Library.
dnl
dnl The GNU MP Library is free software; you can redistribute it and/or modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
C cycles/limb
C AMD K8,K9 4.5
C AMD K10 4.5
C AMD bd1 4.75
C AMD bobcat 5
C Intel P4 17.7
C Intel core2 5.5
C Intel NHM 5.43
C Intel SBR 3.92
C Intel atom 23
C VIA nano 5.63
C This sqr_basecase is based on mul_1 and addmul_1, since these both run at the
C multiply insn bandwidth, without any apparent loop branch exit pipeline
C replays experienced on K8. The structure is unusual: it falls into mul_1 in
C the same way for all n, then it splits into 4 different wind-down blocks and
C 4 separate addmul_1 loops.
C
C We have not tried using the same addmul_1 loops with a switch into feed-in
C code, as we do in other basecase implementations. Doing that could save
C substantial code volume, but would also probably add some overhead.
C TODO
C * Tune un < 4 code.
C * Perhaps implement a larger final corner (it is now 2 x 1).
C * Lots of space could be saved by replacing the "switch" code by gradual
C jumps out from mul_1 winddown code, perhaps with no added overhead.
C * Are the ALIGN(16) really necessary? They add about 25 bytes of padding.
ABI_SUPPORT(DOS64)
ABI_SUPPORT(STD64)
C Standard parameters
define(`rp', `%rdi')
define(`up', `%rsi')
define(`un_param', `%rdx')
C Standard allocations
define(`un', `%rbx')
define(`w0', `%r8')
define(`w1', `%r9')
define(`w2', `%r10')
define(`w3', `%r11')
define(`n', `%rbp')
define(`v0', `%rcx')
C Temp macro for allowing control over indexing.
C Define to return $1 for more conservative ptr handling.
define(`X',`$2')
dnl define(`X',`$1')
ASM_START()
TEXT
ALIGN(64)
PROLOGUE(mpn_sqr_basecase)
FUNC_ENTRY(3)
mov (up), %rax
cmp $2, R32(un_param)
jae L(ge2)
mul %rax
mov %rax, (rp)
mov %rdx, 8(rp)
FUNC_EXIT()
ret
L(ge2): mov (up), v0
jnz L(g2)
mul %rax
mov %rax, (rp)
mov 8(up), %rax
mov %rdx, w0
mul v0
add %rax, w0
mov %rdx, w1
adc $0, w1
mov 8(up), v0
mov (up), %rax
mul v0
add %rax, w0
mov w0, 8(rp)
mov %rdx, w0 C CAUTION: r8 realloc
adc $0, w0
mov 8(up), %rax
mul v0
add w1, w0
adc $0, %rdx
add w0, %rax
adc $0, %rdx
mov %rax, 16(rp)
mov %rdx, 24(rp)
FUNC_EXIT()
ret
L(g2): cmp $3, R32(un_param)
ja L(g3)
mul %rax
mov %rax, (rp)
mov %rdx, 8(rp)
mov 8(up), %rax
mul %rax
mov %rax, 16(rp)
mov %rdx, 24(rp)
mov 16(up), %rax
mul %rax
mov %rax, 32(rp)
mov %rdx, 40(rp)
mov (up), v0
mov 8(up), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov 16(up), %rax
mul v0
xor R32(w2), R32(w2)
add %rax, w1
adc %rdx, w2
mov 8(up), v0
mov 16(up), %rax
mul v0
xor R32(w3), R32(w3)
add %rax, w2
adc %rdx, w3
add w0, w0
adc w1, w1
adc w2, w2
adc w3, w3
mov $0, R32(v0)
adc v0, v0
add w0, 8(rp)
adc w1, 16(rp)
adc w2, 24(rp)
adc w3, 32(rp)
adc v0, 40(rp)
FUNC_EXIT()
ret
L(g3): push %rbx
push %rbp
mov 8(up), %rax
lea -24(rp,un_param,8), rp
lea -24(up,un_param,8), up
neg un_param
push un_param C for sqr_diag_addlsh1
lea (un_param), un
lea 3(un_param), n
mul v0
mov %rax, w2
mov %rdx, w3
jmp L(L3)
ALIGN(16)
L(top): mov w0, -16(rp,n,8)
add w1, w2
adc $0, w3
mov (up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov w2, -8(rp,n,8)
add w3, w0
adc $0, w1
mov 8(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
mov w0, (rp,n,8)
add w1, w2
adc $0, w3
L(L3): mov 16(up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov w2, 8(rp,n,8)
add w3, w0
adc $0, w1
mov 24(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add $4, n
js L(top)
mov w0, -16(rp,n,8)
add w1, w2
adc $0, w3
test n, n
jz L(r2)
cmp $2, R32(n)
ja L(r3)
jz L(r0)
L(r1): mov X((up,n,8),8(up)), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov w2, X(-8(rp,n,8),(rp))
add w3, w0
adc $0, w1
mov X(8(up,n,8),16(up)), %rax
mul v0
mov %rax, w2
mov %rdx, w3
mov w0, X((rp,n,8),8(rp))
add w1, w2
adc $0, w3
mov w2, X(8(rp,n,8),16(rp))
mov w3, X(16(rp,n,8),24(rp))
add $5, un
jmp L(to0)
L(r2): mov X((up,n,8),(up)), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov w2, X(-8(rp,n,8),-8(rp))
add w3, w0
adc $0, w1
mov X(8(up,n,8),8(up)), %rax
mul v0
mov %rax, w2
mov %rdx, w3
mov w0, X((rp,n,8),(rp))
add w1, w2
adc $0, w3
mov X(16(up,n,8),16(up)), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov w2, X(8(rp,n,8),8(rp))
add w3, w0
adc $0, w1
mov w0, X(16(rp,n,8),16(rp))
adc $0, w3
mov w1, X(24(rp,n,8),24(rp))
add $6, un
jmp L(to1)
L(r3): mov w2, X(-8(rp,n,8),16(rp))
mov w3, X((rp,n,8),24(rp))
add $3, un
jmp L(to2)
L(r0): mov X((up,n,8),16(up)), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov w2, X(-8(rp,n,8),8(rp))
add w3, w0
adc $0, w1
mov w0, X((rp,n,8),16(rp))
mov w1, X(8(rp,n,8),24(rp))
add $4, un
C jmp L(to3)
C fall through into main loop
L(outer):
mov un, n
mov (up,un,8), v0
mov 8(up,un,8), %rax
lea 8(rp), rp
mul v0
mov %rax, w2
mov %rdx, w3
jmp L(al3)
ALIGN(16)
L(ta3): add w0, -16(rp,n,8)
adc w1, w2
adc $0, w3
mov (up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, -8(rp,n,8)
adc w3, w0
adc $0, w1
mov 8(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add w0, (rp,n,8)
adc w1, w2
adc $0, w3
L(al3): mov 16(up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, 8(rp,n,8)
adc w3, w0
adc $0, w1
mov 24(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add $4, n
js L(ta3)
add w0, X(-16(rp,n,8),8(rp))
adc w1, w2
adc $0, w3
add w2, X(-8(rp,n,8),16(rp))
adc $0, w3
mov w3, X((rp,n,8),24(rp))
L(to2): mov un, n
cmp $-4, R32(un)
jnc L(end)
add $4, un
mov 8(up,n,8), v0
mov 16(up,n,8), %rax
lea 8(rp), rp
mul v0
mov %rax, w0
mov %rdx, w1
jmp L(al2)
ALIGN(16)
L(ta2): add w0, -16(rp,n,8)
adc w1, w2
adc $0, w3
mov (up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, -8(rp,n,8)
adc w3, w0
adc $0, w1
mov 8(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add w0, (rp,n,8)
adc w1, w2
adc $0, w3
mov 16(up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, 8(rp,n,8)
adc w3, w0
adc $0, w1
L(al2): mov 24(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add $4, n
js L(ta2)
add w0, X(-16(rp,n,8),8(rp))
adc w1, w2
adc $0, w3
add w2, X(-8(rp,n,8),16(rp))
adc $0, w3
mov w3, X((rp,n,8),24(rp))
L(to1): mov un, n
mov -16(up,un,8), v0
mov -8(up,un,8), %rax
lea 8(rp), rp
mul v0
mov %rax, w2
mov %rdx, w3
jmp L(al1)
ALIGN(16)
L(ta1): add w0, -16(rp,n,8)
adc w1, w2
adc $0, w3
L(al1): mov (up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, -8(rp,n,8)
adc w3, w0
adc $0, w1
mov 8(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add w0, (rp,n,8)
adc w1, w2
adc $0, w3
mov 16(up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, 8(rp,n,8)
adc w3, w0
adc $0, w1
mov 24(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add $4, n
js L(ta1)
add w0, X(-16(rp,n,8),8(rp))
adc w1, w2
adc $0, w3
add w2, X(-8(rp,n,8),16(rp))
adc $0, w3
mov w3, X((rp,n,8),24(rp))
L(to0): mov un, n
mov -8(up,un,8), v0
mov (up,un,8), %rax
lea 8(rp), rp
mul v0
mov %rax, w0
mov %rdx, w1
jmp L(al0)
ALIGN(16)
L(ta0): add w0, -16(rp,n,8)
adc w1, w2
adc $0, w3
mov (up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, -8(rp,n,8)
adc w3, w0
adc $0, w1
L(al0): mov 8(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add w0, (rp,n,8)
adc w1, w2
adc $0, w3
mov 16(up,n,8), %rax
mul v0
mov %rax, w0
mov %rdx, w1
add w2, 8(rp,n,8)
adc w3, w0
adc $0, w1
mov 24(up,n,8), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add $4, n
js L(ta0)
add w0, X(-16(rp,n,8),8(rp))
adc w1, w2
adc $0, w3
add w2, X(-8(rp,n,8),16(rp))
adc $0, w3
mov w3, X((rp,n,8),24(rp))
jmp L(outer)
L(end): mov X(8(up,un,8),(up)), v0
mov X(16(up,un,8),8(up)), %rax
mul v0
mov %rax, w0
mov %rdx, w1
mov X(24(up,un,8),16(up)), %rax
mul v0
mov %rax, w2
mov %rdx, w3
add w0, X(24(rp,un,8),16(rp))
adc w1, w2
adc $0, w3
add w2, X(32(rp,un,8),24(rp))
adc $0, w3
mov X(16(up,un,8),8(up)), v0
mov X(24(up,un,8),16(up)), %rax
mul v0
add %rax, w3
mov w3, X(40(rp,un,8),32(rp))
adc $0, %rdx
mov %rdx, X(48(rp,un,8),40(rp))
C sqr_diag_addlsh1
lea 16(up), up
lea 40(rp), rp
pop n
lea 2(n,n), n
mov (up,n,4), %rax
mul %rax
xor R32(w2), R32(w2)
mov 8(rp,n,8), w0
mov %rax, (rp,n,8)
jmp L(lm)
ALIGN(8)
L(tsd): add %rbx, w0
adc %rax, w1
mov w0, -8(rp,n,8)
mov 8(rp,n,8), w0
mov w1, (rp,n,8)
L(lm): mov 16(rp,n,8), w1
adc w0, w0
adc w1, w1
lea (%rdx,w2), %rbx
mov 8(up,n,4), %rax
setc R8(w2)
mul %rax
add $2, n
js L(tsd)
L(esd): add %rbx, w0
adc %rax, w1
mov w0, X(-8(rp,n,8),-8(rp))
mov w1, X((rp,n,8),(rp))
adc w2, %rdx
mov %rdx, X(8(rp,n,8),8(rp))
pop %rbp
pop %rbx
FUNC_EXIT()
ret
EPILOGUE()
|
