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
|
// Copyright 2013 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// +build !math_big_pure_go,ppc64 !math_big_pure_go,ppc64le
#include "textflag.h"
// This file provides fast assembly versions for the elementary
// arithmetic operations on vectors implemented in arith.go.
// func mulWW(x, y Word) (z1, z0 Word)
TEXT ·mulWW(SB), NOSPLIT, $0
MOVD x+0(FP), R4
MOVD y+8(FP), R5
MULHDU R4, R5, R6
MULLD R4, R5, R7
MOVD R6, z1+16(FP)
MOVD R7, z0+24(FP)
RET
// func addVV(z, y, y []Word) (c Word)
// z[i] = x[i] + y[i] for all i, carrying
TEXT ·addVV(SB), NOSPLIT, $0
MOVD z_len+8(FP), R7 // R7 = z_len
MOVD x+24(FP), R8 // R8 = x[]
MOVD y+48(FP), R9 // R9 = y[]
MOVD z+0(FP), R10 // R10 = z[]
// If z_len = 0, we are done
CMP R0, R7
MOVD R0, R4
BEQ done
// Process the first iteration out of the loop so we can
// use MOVDU and avoid 3 index registers updates.
MOVD 0(R8), R11 // R11 = x[i]
MOVD 0(R9), R12 // R12 = y[i]
ADD $-1, R7 // R7 = z_len - 1
ADDC R12, R11, R15 // R15 = x[i] + y[i], set CA
CMP R0, R7
MOVD R15, 0(R10) // z[i]
BEQ final // If z_len was 1, we are done
SRD $2, R7, R5 // R5 = z_len/4
CMP R0, R5
MOVD R5, CTR // Set up loop counter
BEQ tail // If R5 = 0, we can't use the loop
// Process 4 elements per iteration. Unrolling this loop
// means a performance trade-off: we will lose performance
// for small values of z_len (0.90x in the worst case), but
// gain significant performance as z_len increases (up to
// 1.45x).
loop:
MOVD 8(R8), R11 // R11 = x[i]
MOVD 16(R8), R12 // R12 = x[i+1]
MOVD 24(R8), R14 // R14 = x[i+2]
MOVDU 32(R8), R15 // R15 = x[i+3]
MOVD 8(R9), R16 // R16 = y[i]
MOVD 16(R9), R17 // R17 = y[i+1]
MOVD 24(R9), R18 // R18 = y[i+2]
MOVDU 32(R9), R19 // R19 = y[i+3]
ADDE R11, R16, R20 // R20 = x[i] + y[i] + CA
ADDE R12, R17, R21 // R21 = x[i+1] + y[i+1] + CA
ADDE R14, R18, R22 // R22 = x[i+2] + y[i+2] + CA
ADDE R15, R19, R23 // R23 = x[i+3] + y[i+3] + CA
MOVD R20, 8(R10) // z[i]
MOVD R21, 16(R10) // z[i+1]
MOVD R22, 24(R10) // z[i+2]
MOVDU R23, 32(R10) // z[i+3]
ADD $-4, R7 // R7 = z_len - 4
BC 16, 0, loop // bdnz
// We may have more elements to read
CMP R0, R7
BEQ final
// Process the remaining elements, one at a time
tail:
MOVDU 8(R8), R11 // R11 = x[i]
MOVDU 8(R9), R16 // R16 = y[i]
ADD $-1, R7 // R7 = z_len - 1
ADDE R11, R16, R20 // R20 = x[i] + y[i] + CA
CMP R0, R7
MOVDU R20, 8(R10) // z[i]
BEQ final // If R7 = 0, we are done
MOVDU 8(R8), R11
MOVDU 8(R9), R16
ADD $-1, R7
ADDE R11, R16, R20
CMP R0, R7
MOVDU R20, 8(R10)
BEQ final
MOVD 8(R8), R11
MOVD 8(R9), R16
ADDE R11, R16, R20
MOVD R20, 8(R10)
final:
ADDZE R4 // Capture CA
done:
MOVD R4, c+72(FP)
RET
// func subVV(z, x, y []Word) (c Word)
// z[i] = x[i] - y[i] for all i, carrying
TEXT ·subVV(SB), NOSPLIT, $0
MOVD z_len+8(FP), R7 // R7 = z_len
MOVD x+24(FP), R8 // R8 = x[]
MOVD y+48(FP), R9 // R9 = y[]
MOVD z+0(FP), R10 // R10 = z[]
// If z_len = 0, we are done
CMP R0, R7
MOVD R0, R4
BEQ done
// Process the first iteration out of the loop so we can
// use MOVDU and avoid 3 index registers updates.
MOVD 0(R8), R11 // R11 = x[i]
MOVD 0(R9), R12 // R12 = y[i]
ADD $-1, R7 // R7 = z_len - 1
SUBC R12, R11, R15 // R15 = x[i] - y[i], set CA
CMP R0, R7
MOVD R15, 0(R10) // z[i]
BEQ final // If z_len was 1, we are done
SRD $2, R7, R5 // R5 = z_len/4
CMP R0, R5
MOVD R5, CTR // Set up loop counter
BEQ tail // If R5 = 0, we can't use the loop
// Process 4 elements per iteration. Unrolling this loop
// means a performance trade-off: we will lose performance
// for small values of z_len (0.92x in the worst case), but
// gain significant performance as z_len increases (up to
// 1.45x).
loop:
MOVD 8(R8), R11 // R11 = x[i]
MOVD 16(R8), R12 // R12 = x[i+1]
MOVD 24(R8), R14 // R14 = x[i+2]
MOVDU 32(R8), R15 // R15 = x[i+3]
MOVD 8(R9), R16 // R16 = y[i]
MOVD 16(R9), R17 // R17 = y[i+1]
MOVD 24(R9), R18 // R18 = y[i+2]
MOVDU 32(R9), R19 // R19 = y[i+3]
SUBE R16, R11, R20 // R20 = x[i] - y[i] + CA
SUBE R17, R12, R21 // R21 = x[i+1] - y[i+1] + CA
SUBE R18, R14, R22 // R22 = x[i+2] - y[i+2] + CA
SUBE R19, R15, R23 // R23 = x[i+3] - y[i+3] + CA
MOVD R20, 8(R10) // z[i]
MOVD R21, 16(R10) // z[i+1]
MOVD R22, 24(R10) // z[i+2]
MOVDU R23, 32(R10) // z[i+3]
ADD $-4, R7 // R7 = z_len - 4
BC 16, 0, loop // bdnz
// We may have more elements to read
CMP R0, R7
BEQ final
// Process the remaining elements, one at a time
tail:
MOVDU 8(R8), R11 // R11 = x[i]
MOVDU 8(R9), R16 // R16 = y[i]
ADD $-1, R7 // R7 = z_len - 1
SUBE R16, R11, R20 // R20 = x[i] - y[i] + CA
CMP R0, R7
MOVDU R20, 8(R10) // z[i]
BEQ final // If R7 = 0, we are done
MOVDU 8(R8), R11
MOVDU 8(R9), R16
ADD $-1, R7
SUBE R16, R11, R20
CMP R0, R7
MOVDU R20, 8(R10)
BEQ final
MOVD 8(R8), R11
MOVD 8(R9), R16
SUBE R16, R11, R20
MOVD R20, 8(R10)
final:
ADDZE R4
XOR $1, R4
done:
MOVD R4, c+72(FP)
RET
// func addVW(z, x []Word, y Word) (c Word)
TEXT ·addVW(SB), NOSPLIT, $0
MOVD z+0(FP), R10 // R10 = z[]
MOVD x+24(FP), R8 // R8 = x[]
MOVD y+48(FP), R4 // R4 = y = c
MOVD z_len+8(FP), R11 // R11 = z_len
CMP R0, R11 // If z_len is zero, return
BEQ done
// We will process the first iteration out of the loop so we capture
// the value of c. In the subsequent iterations, we will rely on the
// value of CA set here.
MOVD 0(R8), R20 // R20 = x[i]
ADD $-1, R11 // R11 = z_len - 1
ADDC R20, R4, R6 // R6 = x[i] + c
CMP R0, R11 // If z_len was 1, we are done
MOVD R6, 0(R10) // z[i]
BEQ final
// We will read 4 elements per iteration
SRD $2, R11, R9 // R9 = z_len/4
DCBT (R8)
CMP R0, R9
MOVD R9, CTR // Set up the loop counter
BEQ tail // If R9 = 0, we can't use the loop
loop:
MOVD 8(R8), R20 // R20 = x[i]
MOVD 16(R8), R21 // R21 = x[i+1]
MOVD 24(R8), R22 // R22 = x[i+2]
MOVDU 32(R8), R23 // R23 = x[i+3]
ADDZE R20, R24 // R24 = x[i] + CA
ADDZE R21, R25 // R25 = x[i+1] + CA
ADDZE R22, R26 // R26 = x[i+2] + CA
ADDZE R23, R27 // R27 = x[i+3] + CA
MOVD R24, 8(R10) // z[i]
MOVD R25, 16(R10) // z[i+1]
MOVD R26, 24(R10) // z[i+2]
MOVDU R27, 32(R10) // z[i+3]
ADD $-4, R11 // R11 = z_len - 4
BC 16, 0, loop // bdnz
// We may have some elements to read
CMP R0, R11
BEQ final
tail:
MOVDU 8(R8), R20
ADDZE R20, R24
ADD $-1, R11
MOVDU R24, 8(R10)
CMP R0, R11
BEQ final
MOVDU 8(R8), R20
ADDZE R20, R24
ADD $-1, R11
MOVDU R24, 8(R10)
CMP R0, R11
BEQ final
MOVD 8(R8), R20
ADDZE R20, R24
MOVD R24, 8(R10)
final:
ADDZE R0, R4 // c = CA
done:
MOVD R4, c+56(FP)
RET
// func subVW(z, x []Word, y Word) (c Word)
TEXT ·subVW(SB), NOSPLIT, $0
MOVD z+0(FP), R10 // R10 = z[]
MOVD x+24(FP), R8 // R8 = x[]
MOVD y+48(FP), R4 // R4 = y = c
MOVD z_len+8(FP), R11 // R11 = z_len
CMP R0, R11 // If z_len is zero, return
BEQ done
// We will process the first iteration out of the loop so we capture
// the value of c. In the subsequent iterations, we will rely on the
// value of CA set here.
MOVD 0(R8), R20 // R20 = x[i]
ADD $-1, R11 // R11 = z_len - 1
SUBC R4, R20, R6 // R6 = x[i] - c
CMP R0, R11 // If z_len was 1, we are done
MOVD R6, 0(R10) // z[i]
BEQ final
// We will read 4 elements per iteration
SRD $2, R11, R9 // R9 = z_len/4
DCBT (R8)
CMP R0, R9
MOVD R9, CTR // Set up the loop counter
BEQ tail // If R9 = 0, we can't use the loop
// The loop here is almost the same as the one used in s390x, but
// we don't need to capture CA every iteration because we've already
// done that above.
loop:
MOVD 8(R8), R20
MOVD 16(R8), R21
MOVD 24(R8), R22
MOVDU 32(R8), R23
SUBE R0, R20
SUBE R0, R21
SUBE R0, R22
SUBE R0, R23
MOVD R20, 8(R10)
MOVD R21, 16(R10)
MOVD R22, 24(R10)
MOVDU R23, 32(R10)
ADD $-4, R11
BC 16, 0, loop // bdnz
// We may have some elements to read
CMP R0, R11
BEQ final
tail:
MOVDU 8(R8), R20
SUBE R0, R20
ADD $-1, R11
MOVDU R20, 8(R10)
CMP R0, R11
BEQ final
MOVDU 8(R8), R20
SUBE R0, R20
ADD $-1, R11
MOVDU R20, 8(R10)
CMP R0, R11
BEQ final
MOVD 8(R8), R20
SUBE R0, R20
MOVD R20, 8(R10)
final:
// Capture CA
SUBE R4, R4
NEG R4, R4
done:
MOVD R4, c+56(FP)
RET
TEXT ·shlVU(SB), NOSPLIT, $0
BR ·shlVU_g(SB)
TEXT ·shrVU(SB), NOSPLIT, $0
BR ·shrVU_g(SB)
// func mulAddVWW(z, x []Word, y, r Word) (c Word)
TEXT ·mulAddVWW(SB), NOSPLIT, $0
MOVD z+0(FP), R10 // R10 = z[]
MOVD x+24(FP), R8 // R8 = x[]
MOVD y+48(FP), R9 // R9 = y
MOVD r+56(FP), R4 // R4 = r = c
MOVD z_len+8(FP), R11 // R11 = z_len
CMP R0, R11
BEQ done
MOVD 0(R8), R20
ADD $-1, R11
MULLD R9, R20, R6 // R6 = z0 = Low-order(x[i]*y)
MULHDU R9, R20, R7 // R7 = z1 = High-order(x[i]*y)
ADDC R4, R6 // R6 = z0 + r
ADDZE R7 // R7 = z1 + CA
CMP R0, R11
MOVD R7, R4 // R4 = c
MOVD R6, 0(R10) // z[i]
BEQ done
// We will read 4 elements per iteration
SRD $2, R11, R14 // R14 = z_len/4
DCBT (R8)
CMP R0, R14
MOVD R14, CTR // Set up the loop counter
BEQ tail // If R9 = 0, we can't use the loop
loop:
MOVD 8(R8), R20 // R20 = x[i]
MOVD 16(R8), R21 // R21 = x[i+1]
MOVD 24(R8), R22 // R22 = x[i+2]
MOVDU 32(R8), R23 // R23 = x[i+3]
MULLD R9, R20, R24 // R24 = z0[i]
MULHDU R9, R20, R20 // R20 = z1[i]
ADDC R4, R24 // R24 = z0[i] + c
ADDZE R20 // R7 = z1[i] + CA
MULLD R9, R21, R25
MULHDU R9, R21, R21
ADDC R20, R25
ADDZE R21
MULLD R9, R22, R26
MULHDU R9, R22, R22
MULLD R9, R23, R27
MULHDU R9, R23, R23
ADDC R21, R26
ADDZE R22
MOVD R24, 8(R10) // z[i]
MOVD R25, 16(R10) // z[i+1]
ADDC R22, R27
ADDZE R23,R4 // update carry
MOVD R26, 24(R10) // z[i+2]
MOVDU R27, 32(R10) // z[i+3]
ADD $-4, R11 // R11 = z_len - 4
BC 16, 0, loop // bdnz
// We may have some elements to read
CMP R0, R11
BEQ done
// Process the remaining elements, one at a time
tail:
MOVDU 8(R8), R20 // R20 = x[i]
MULLD R9, R20, R24 // R24 = z0[i]
MULHDU R9, R20, R25 // R25 = z1[i]
ADD $-1, R11 // R11 = z_len - 1
ADDC R4, R24
ADDZE R25
MOVDU R24, 8(R10) // z[i]
CMP R0, R11
MOVD R25, R4 // R4 = c
BEQ done // If R11 = 0, we are done
MOVDU 8(R8), R20
MULLD R9, R20, R24
MULHDU R9, R20, R25
ADD $-1, R11
ADDC R4, R24
ADDZE R25
MOVDU R24, 8(R10)
CMP R0, R11
MOVD R25, R4
BEQ done
MOVD 8(R8), R20
MULLD R9, R20, R24
MULHDU R9, R20, R25
ADD $-1, R11
ADDC R4, R24
ADDZE R25
MOVD R24, 8(R10)
MOVD R25, R4
done:
MOVD R4, c+64(FP)
RET
// func addMulVVW(z, x []Word, y Word) (c Word)
TEXT ·addMulVVW(SB), NOSPLIT, $0
MOVD z+0(FP), R10 // R10 = z[]
MOVD x+24(FP), R8 // R8 = x[]
MOVD y+48(FP), R9 // R9 = y
MOVD z_len+8(FP), R22 // R22 = z_len
MOVD R0, R3 // R3 will be the index register
CMP R0, R22
MOVD R0, R4 // R4 = c = 0
MOVD R22, CTR // Initialize loop counter
BEQ done
loop:
MOVD (R8)(R3), R20 // Load x[i]
MOVD (R10)(R3), R21 // Load z[i]
MULLD R9, R20, R6 // R6 = Low-order(x[i]*y)
MULHDU R9, R20, R7 // R7 = High-order(x[i]*y)
ADDC R21, R6 // R6 = z0
ADDZE R7 // R7 = z1
ADDC R4, R6 // R6 = z0 + c + 0
ADDZE R7, R4 // c += z1
MOVD R6, (R10)(R3) // Store z[i]
ADD $8, R3
BC 16, 0, loop // bdnz
done:
MOVD R4, c+56(FP)
RET
|
