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
|
/ This Source Code Form is subject to the terms of the Mozilla Public
/ License, v. 2.0. If a copy of the MPL was not distributed with this
/ file, You can obtain one at http://mozilla.org/MPL/2.0/.
/ ------------------------------------------------------------------------
/
/ Implementation of s_mpv_mul_set_vec which exploits
/ the 64X64->128 bit unsigned multiply instruction.
/
/ ------------------------------------------------------------------------
/ r = a * digit, r and a are vectors of length len
/ returns the carry digit
/ r and a are 64 bit aligned.
/
/ uint64_t
/ s_mpv_mul_set_vec64(uint64_t *r, uint64_t *a, int len, uint64_t digit)
/
.text; .align 16; .globl s_mpv_mul_set_vec64; .type s_mpv_mul_set_vec64, @function; s_mpv_mul_set_vec64:
xorq %rax, %rax / if (len == 0) return (0)
testq %rdx, %rdx
jz .L17
movq %rdx, %r8 / Use r8 for len; %rdx is used by mul
xorq %r9, %r9 / cy = 0
.L15:
cmpq $8, %r8 / 8 - len
jb .L16
movq 0(%rsi), %rax / rax = a[0]
movq 8(%rsi), %r11 / prefetch a[1]
mulq %rcx / p = a[0] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 0(%rdi) / r[0] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 16(%rsi), %r11 / prefetch a[2]
mulq %rcx / p = a[1] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 8(%rdi) / r[1] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 24(%rsi), %r11 / prefetch a[3]
mulq %rcx / p = a[2] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 16(%rdi) / r[2] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 32(%rsi), %r11 / prefetch a[4]
mulq %rcx / p = a[3] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 24(%rdi) / r[3] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 40(%rsi), %r11 / prefetch a[5]
mulq %rcx / p = a[4] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 32(%rdi) / r[4] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 48(%rsi), %r11 / prefetch a[6]
mulq %rcx / p = a[5] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 40(%rdi) / r[5] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 56(%rsi), %r11 / prefetch a[7]
mulq %rcx / p = a[6] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 48(%rdi) / r[6] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
mulq %rcx / p = a[7] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 56(%rdi) / r[7] = lo(p)
movq %rdx, %r9 / cy = hi(p)
addq $64, %rsi
addq $64, %rdi
subq $8, %r8
jz .L17
jmp .L15
.L16:
movq 0(%rsi), %rax
mulq %rcx / p = a[0] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 0(%rdi) / r[0] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
movq 8(%rsi), %rax
mulq %rcx / p = a[1] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 8(%rdi) / r[1] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
movq 16(%rsi), %rax
mulq %rcx / p = a[2] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 16(%rdi) / r[2] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
movq 24(%rsi), %rax
mulq %rcx / p = a[3] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 24(%rdi) / r[3] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
movq 32(%rsi), %rax
mulq %rcx / p = a[4] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 32(%rdi) / r[4] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
movq 40(%rsi), %rax
mulq %rcx / p = a[5] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 40(%rdi) / r[5] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
movq 48(%rsi), %rax
mulq %rcx / p = a[6] * digit
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 48(%rdi) / r[6] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L17
.L17:
movq %r9, %rax
ret
.size s_mpv_mul_set_vec64, .-s_mpv_mul_set_vec64
/ ------------------------------------------------------------------------
/
/ Implementation of s_mpv_mul_add_vec which exploits
/ the 64X64->128 bit unsigned multiply instruction.
/
/ ------------------------------------------------------------------------
/ r += a * digit, r and a are vectors of length len
/ returns the carry digit
/ r and a are 64 bit aligned.
/
/ uint64_t
/ s_mpv_mul_add_vec64(uint64_t *r, uint64_t *a, int len, uint64_t digit)
/
.text; .align 16; .globl s_mpv_mul_add_vec64; .type s_mpv_mul_add_vec64, @function; s_mpv_mul_add_vec64:
xorq %rax, %rax / if (len == 0) return (0)
testq %rdx, %rdx
jz .L27
movq %rdx, %r8 / Use r8 for len; %rdx is used by mul
xorq %r9, %r9 / cy = 0
.L25:
cmpq $8, %r8 / 8 - len
jb .L26
movq 0(%rsi), %rax / rax = a[0]
movq 0(%rdi), %r10 / r10 = r[0]
movq 8(%rsi), %r11 / prefetch a[1]
mulq %rcx / p = a[0] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[0]
movq 8(%rdi), %r10 / prefetch r[1]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 0(%rdi) / r[0] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 16(%rsi), %r11 / prefetch a[2]
mulq %rcx / p = a[1] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[1]
movq 16(%rdi), %r10 / prefetch r[2]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 8(%rdi) / r[1] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 24(%rsi), %r11 / prefetch a[3]
mulq %rcx / p = a[2] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[2]
movq 24(%rdi), %r10 / prefetch r[3]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 16(%rdi) / r[2] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 32(%rsi), %r11 / prefetch a[4]
mulq %rcx / p = a[3] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[3]
movq 32(%rdi), %r10 / prefetch r[4]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 24(%rdi) / r[3] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 40(%rsi), %r11 / prefetch a[5]
mulq %rcx / p = a[4] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[4]
movq 40(%rdi), %r10 / prefetch r[5]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 32(%rdi) / r[4] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 48(%rsi), %r11 / prefetch a[6]
mulq %rcx / p = a[5] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[5]
movq 48(%rdi), %r10 / prefetch r[6]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 40(%rdi) / r[5] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
movq 56(%rsi), %r11 / prefetch a[7]
mulq %rcx / p = a[6] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[6]
movq 56(%rdi), %r10 / prefetch r[7]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 48(%rdi) / r[6] = lo(p)
movq %rdx, %r9 / cy = hi(p)
movq %r11, %rax
mulq %rcx / p = a[7] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[7]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 56(%rdi) / r[7] = lo(p)
movq %rdx, %r9 / cy = hi(p)
addq $64, %rsi
addq $64, %rdi
subq $8, %r8
jz .L27
jmp .L25
.L26:
movq 0(%rsi), %rax
movq 0(%rdi), %r10
mulq %rcx / p = a[0] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[0]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 0(%rdi) / r[0] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
movq 8(%rsi), %rax
movq 8(%rdi), %r10
mulq %rcx / p = a[1] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[1]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 8(%rdi) / r[1] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
movq 16(%rsi), %rax
movq 16(%rdi), %r10
mulq %rcx / p = a[2] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[2]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 16(%rdi) / r[2] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
movq 24(%rsi), %rax
movq 24(%rdi), %r10
mulq %rcx / p = a[3] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[3]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 24(%rdi) / r[3] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
movq 32(%rsi), %rax
movq 32(%rdi), %r10
mulq %rcx / p = a[4] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[4]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 32(%rdi) / r[4] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
movq 40(%rsi), %rax
movq 40(%rdi), %r10
mulq %rcx / p = a[5] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[5]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 40(%rdi) / r[5] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
movq 48(%rsi), %rax
movq 48(%rdi), %r10
mulq %rcx / p = a[6] * digit
addq %r10, %rax
adcq $0, %rdx / p += r[6]
addq %r9, %rax
adcq $0, %rdx / p += cy
movq %rax, 48(%rdi) / r[6] = lo(p)
movq %rdx, %r9 / cy = hi(p)
decq %r8
jz .L27
.L27:
movq %r9, %rax
ret
.size s_mpv_mul_add_vec64, .-s_mpv_mul_add_vec64
|
