forked from klauspost/reedsolomon
-
Notifications
You must be signed in to change notification settings - Fork 0
/
galoisAvx512_amd64.s
400 lines (313 loc) · 9.7 KB
/
galoisAvx512_amd64.s
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
//+build !noasm !appengine !gccgo
// Copyright 2015, Klaus Post, see LICENSE for details.
// Copyright 2019, Minio, Inc.
#define LOAD(OFFSET) \
MOVQ OFFSET(SI), BX \
VMOVDQU64 (BX)(R11*1), Z0 \
VPSRLQ $4, Z0, Z1 \ // high input
VPANDQ Z2, Z0, Z0 \ // low input
VPANDQ Z2, Z1, Z1 // high input
#define GALOIS_MUL(MUL_LO, MUL_HI, LO, HI, OUT) \
VPSHUFB Z0, MUL_LO, LO \ // mul low part
VPSHUFB Z1, MUL_HI, HI \ // mul high part
VPTERNLOGD $0x96, LO, HI, OUT
#define GALOIS(C1, C2, IN, LO, HI, OUT) \
VSHUFI64X2 $C1, IN, IN, LO \
VSHUFI64X2 $C2, IN, IN, HI \
GALOIS_MUL(LO, HI, LO, HI, OUT)
//
// Process single output row from a total of 8 input rows
//
// func _galMulAVX512Parallel81(in, out [][]byte, matrix *[matrixSize81]byte, addTo bool)
TEXT ·_galMulAVX512Parallel81(SB), 7, $0
MOVQ in+0(FP), SI
MOVQ 8(SI), R9 // R9: len(in)
SHRQ $6, R9 // len(in) / 64
TESTQ R9, R9
JZ done_avx512_parallel81
MOVQ matrix+48(FP), SI
VMOVDQU64 0x000(SI), Z16
VMOVDQU64 0x040(SI), Z17
VMOVDQU64 0x080(SI), Z18
VMOVDQU64 0x0c0(SI), Z19
// Initialize multiplication constants
VSHUFI64X2 $0x55, Z16, Z16, Z20
VSHUFI64X2 $0xaa, Z16, Z16, Z24
VSHUFI64X2 $0xff, Z16, Z16, Z28
VSHUFI64X2 $0x00, Z16, Z16, Z16
VSHUFI64X2 $0x55, Z17, Z17, Z21
VSHUFI64X2 $0xaa, Z17, Z17, Z25
VSHUFI64X2 $0xff, Z17, Z17, Z29
VSHUFI64X2 $0x00, Z17, Z17, Z17
VSHUFI64X2 $0x55, Z18, Z18, Z22
VSHUFI64X2 $0xaa, Z18, Z18, Z26
VSHUFI64X2 $0xff, Z18, Z18, Z30
VSHUFI64X2 $0x00, Z18, Z18, Z18
VSHUFI64X2 $0x55, Z19, Z19, Z23
VSHUFI64X2 $0xaa, Z19, Z19, Z27
VSHUFI64X2 $0xff, Z19, Z19, Z31
VSHUFI64X2 $0x00, Z19, Z19, Z19
MOVQ $15, BX
VPBROADCASTB BX, Z2
MOVB addTo+56(FP), AX
IMULQ $-0x1, AX
KMOVQ AX, K1
MOVQ in+0(FP), SI // SI: &in
MOVQ in_len+8(FP), AX // number of inputs
XORQ R11, R11
MOVQ out+24(FP), DX
MOVQ (DX), DX // DX: &out[0][0]
loopback_avx512_parallel81:
VMOVDQU64.Z (DX), K1, Z4
LOAD(0x00) // &in[0][0]
GALOIS_MUL(Z16, Z20, Z14, Z15, Z4)
CMPQ AX, $1
JE skip_avx512_parallel81
LOAD(0x18) // &in[1][0]
GALOIS_MUL(Z24, Z28, Z14, Z15, Z4)
CMPQ AX, $2
JE skip_avx512_parallel81
LOAD(0x30) // &in[2][0]
GALOIS_MUL(Z17, Z21, Z14, Z15, Z4)
CMPQ AX, $3
JE skip_avx512_parallel81
LOAD(0x48) // &in[3][0]
GALOIS_MUL(Z25, Z29, Z14, Z15, Z4)
CMPQ AX, $4
JE skip_avx512_parallel81
LOAD(0x60) // &in[4][0]
GALOIS_MUL(Z18, Z22, Z14, Z15, Z4)
CMPQ AX, $5
JE skip_avx512_parallel81
LOAD(0x78) // &in[5][0]
GALOIS_MUL(Z26, Z30, Z14, Z15, Z4)
CMPQ AX, $6
JE skip_avx512_parallel81
LOAD(0x90) // &in[6][0]
GALOIS_MUL(Z19, Z23, Z14, Z15, Z4)
CMPQ AX, $7
JE skip_avx512_parallel81
LOAD(0xa8) // &in[7][0]
GALOIS_MUL(Z27, Z31, Z14, Z15, Z4)
skip_avx512_parallel81:
VMOVDQU64 Z4, (DX)
ADDQ $64, R11 // in4+=64
ADDQ $64, DX // out+=64
SUBQ $1, R9
JNZ loopback_avx512_parallel81
done_avx512_parallel81:
VZEROUPPER
RET
//
// Process 2 output rows in parallel from a total of 8 input rows
//
// func _galMulAVX512Parallel82(in, out [][]byte, matrix *[matrixSize82]byte, addTo bool)
TEXT ·_galMulAVX512Parallel82(SB), 7, $0
MOVQ in+0(FP), SI
MOVQ 8(SI), R9 // R9: len(in)
SHRQ $6, R9 // len(in) / 64
TESTQ R9, R9
JZ done_avx512_parallel82
MOVQ matrix+48(FP), SI
VMOVDQU64 0x000(SI), Z16
VMOVDQU64 0x040(SI), Z17
VMOVDQU64 0x080(SI), Z18
VMOVDQU64 0x0c0(SI), Z19
VMOVDQU64 0x100(SI), Z20
VMOVDQU64 0x140(SI), Z21
VMOVDQU64 0x180(SI), Z22
VMOVDQU64 0x1c0(SI), Z23
// Initialize multiplication constants
VSHUFI64X2 $0x55, Z16, Z16, Z24
VSHUFI64X2 $0xaa, Z16, Z16, Z25
VSHUFI64X2 $0xff, Z16, Z16, Z26
VSHUFI64X2 $0x00, Z16, Z16, Z16
VSHUFI64X2 $0x55, Z20, Z20, Z27
VSHUFI64X2 $0xaa, Z20, Z20, Z28
VSHUFI64X2 $0xff, Z20, Z20, Z29
VSHUFI64X2 $0x00, Z20, Z20, Z20
VSHUFI64X2 $0x55, Z17, Z17, Z30
VSHUFI64X2 $0xaa, Z17, Z17, Z31
VSHUFI64X2 $0xff, Z17, Z17, Z11
VSHUFI64X2 $0x00, Z17, Z17, Z17
VSHUFI64X2 $0x55, Z21, Z21, Z8
VSHUFI64X2 $0xaa, Z21, Z21, Z9
VSHUFI64X2 $0xff, Z21, Z21, Z10
VSHUFI64X2 $0x00, Z21, Z21, Z21
MOVQ $15, BX
VPBROADCASTB BX, Z2
MOVB addTo+56(FP), AX
IMULQ $-0x1, AX
KMOVQ AX, K1
MOVQ in+0(FP), SI // SI: &in
MOVQ in_len+8(FP), AX // number of inputs
XORQ R11, R11
MOVQ out+24(FP), DX
MOVQ 24(DX), CX // CX: &out[1][0]
MOVQ (DX), DX // DX: &out[0][0]
loopback_avx512_parallel82:
VMOVDQU64.Z (DX), K1, Z4
VMOVDQU64.Z (CX), K1, Z5
LOAD(0x00) // &in[0][0]
GALOIS_MUL(Z16, Z24, Z14, Z15, Z4)
GALOIS_MUL(Z20, Z27, Z12, Z13, Z5)
CMPQ AX, $1
JE skip_avx512_parallel82
LOAD(0x18) // &in[1][0]
GALOIS_MUL(Z25, Z26, Z14, Z15, Z4)
GALOIS_MUL(Z28, Z29, Z12, Z13, Z5)
CMPQ AX, $2
JE skip_avx512_parallel82
LOAD(0x30) // &in[2][0]
GALOIS_MUL(Z17, Z30, Z14, Z15, Z4)
GALOIS_MUL(Z21, Z8, Z12, Z13, Z5)
CMPQ AX, $3
JE skip_avx512_parallel82
LOAD(0x48) // &in[3][0]
GALOIS_MUL(Z31, Z11, Z14, Z15, Z4)
GALOIS_MUL(Z9, Z10, Z12, Z13, Z5)
CMPQ AX, $4
JE skip_avx512_parallel82
LOAD(0x60) // &in[4][0]
GALOIS(0x00, 0x55, Z18, Z14, Z15, Z4)
GALOIS(0x00, 0x55, Z22, Z12, Z13, Z5)
CMPQ AX, $5
JE skip_avx512_parallel82
LOAD(0x78) // &in[5][0]
GALOIS(0xaa, 0xff, Z18, Z14, Z15, Z4)
GALOIS(0xaa, 0xff, Z22, Z12, Z13, Z5)
CMPQ AX, $6
JE skip_avx512_parallel82
LOAD(0x90) // &in[6][0]
GALOIS(0x00, 0x55, Z19, Z14, Z15, Z4)
GALOIS(0x00, 0x55, Z23, Z12, Z13, Z5)
CMPQ AX, $7
JE skip_avx512_parallel82
LOAD(0xa8) // &in[7][0]
GALOIS(0xaa, 0xff, Z19, Z14, Z15, Z4)
GALOIS(0xaa, 0xff, Z23, Z12, Z13, Z5)
skip_avx512_parallel82:
VMOVDQU64 Z4, (DX)
VMOVDQU64 Z5, (CX)
ADDQ $64, R11 // in4+=64
ADDQ $64, DX // out+=64
ADDQ $64, CX // out2+=64
SUBQ $1, R9
JNZ loopback_avx512_parallel82
done_avx512_parallel82:
VZEROUPPER
RET
//
// Process 4 output rows in parallel from a total of 8 input rows
//
// func _galMulAVX512Parallel84(in, out [][]byte, matrix *[matrixSize84]byte, addTo bool)
TEXT ·_galMulAVX512Parallel84(SB), 7, $0
MOVQ in+0(FP), SI
MOVQ 8(SI), R9 // R9: len(in)
SHRQ $6, R9 // len(in) / 64
TESTQ R9, R9
JZ done_avx512_parallel84
MOVQ matrix+48(FP), SI
VMOVDQU64 0x000(SI), Z16
VMOVDQU64 0x040(SI), Z17
VMOVDQU64 0x080(SI), Z18
VMOVDQU64 0x0c0(SI), Z19
VMOVDQU64 0x100(SI), Z20
VMOVDQU64 0x140(SI), Z21
VMOVDQU64 0x180(SI), Z22
VMOVDQU64 0x1c0(SI), Z23
VMOVDQU64 0x200(SI), Z24
VMOVDQU64 0x240(SI), Z25
VMOVDQU64 0x280(SI), Z26
VMOVDQU64 0x2c0(SI), Z27
VMOVDQU64 0x300(SI), Z28
VMOVDQU64 0x340(SI), Z29
VMOVDQU64 0x380(SI), Z30
VMOVDQU64 0x3c0(SI), Z31
MOVQ $15, BX
VPBROADCASTB BX, Z2
MOVB addTo+56(FP), AX
IMULQ $-0x1, AX
KMOVQ AX, K1
MOVQ in+0(FP), SI // SI: &in
MOVQ in_len+8(FP), AX // number of inputs
XORQ R11, R11
MOVQ out+24(FP), DX
MOVQ 24(DX), CX // CX: &out[1][0]
MOVQ 48(DX), R10 // R10: &out[2][0]
MOVQ 72(DX), R12 // R12: &out[3][0]
MOVQ (DX), DX // DX: &out[0][0]
loopback_avx512_parallel84:
VMOVDQU64.Z (DX), K1, Z4
VMOVDQU64.Z (CX), K1, Z5
VMOVDQU64.Z (R10), K1, Z6
VMOVDQU64.Z (R12), K1, Z7
LOAD(0x00) // &in[0][0]
GALOIS(0x00, 0x55, Z16, Z14, Z15, Z4)
GALOIS(0x00, 0x55, Z20, Z12, Z13, Z5)
GALOIS(0x00, 0x55, Z24, Z10, Z11, Z6)
GALOIS(0x00, 0x55, Z28, Z8, Z9, Z7)
CMPQ AX, $1
JE skip_avx512_parallel84
LOAD(0x18) // &in[1][0]
GALOIS(0xaa, 0xff, Z16, Z14, Z15, Z4)
GALOIS(0xaa, 0xff, Z20, Z12, Z13, Z5)
GALOIS(0xaa, 0xff, Z24, Z10, Z11, Z6)
GALOIS(0xaa, 0xff, Z28, Z8, Z9, Z7)
CMPQ AX, $2
JE skip_avx512_parallel84
LOAD(0x30) // &in[2][0]
GALOIS(0x00, 0x55, Z17, Z14, Z15, Z4)
GALOIS(0x00, 0x55, Z21, Z12, Z13, Z5)
GALOIS(0x00, 0x55, Z25, Z10, Z11, Z6)
GALOIS(0x00, 0x55, Z29, Z8, Z9, Z7)
CMPQ AX, $3
JE skip_avx512_parallel84
LOAD(0x48) // &in[3][0]
GALOIS(0xaa, 0xff, Z17, Z14, Z15, Z4)
GALOIS(0xaa, 0xff, Z21, Z12, Z13, Z5)
GALOIS(0xaa, 0xff, Z25, Z10, Z11, Z6)
GALOIS(0xaa, 0xff, Z29, Z8, Z9, Z7)
CMPQ AX, $4
JE skip_avx512_parallel84
LOAD(0x60) // &in[4][0]
GALOIS(0x00, 0x55, Z18, Z14, Z15, Z4)
GALOIS(0x00, 0x55, Z22, Z12, Z13, Z5)
GALOIS(0x00, 0x55, Z26, Z10, Z11, Z6)
GALOIS(0x00, 0x55, Z30, Z8, Z9, Z7)
CMPQ AX, $5
JE skip_avx512_parallel84
LOAD(0x78) // &in[5][0]
GALOIS(0xaa, 0xff, Z18, Z14, Z15, Z4)
GALOIS(0xaa, 0xff, Z22, Z12, Z13, Z5)
GALOIS(0xaa, 0xff, Z26, Z10, Z11, Z6)
GALOIS(0xaa, 0xff, Z30, Z8, Z9, Z7)
CMPQ AX, $6
JE skip_avx512_parallel84
LOAD(0x90) // &in[6][0]
GALOIS(0x00, 0x55, Z19, Z14, Z15, Z4)
GALOIS(0x00, 0x55, Z23, Z12, Z13, Z5)
GALOIS(0x00, 0x55, Z27, Z10, Z11, Z6)
GALOIS(0x00, 0x55, Z31, Z8, Z9, Z7)
CMPQ AX, $7
JE skip_avx512_parallel84
LOAD(0xa8) // &in[7][0]
GALOIS(0xaa, 0xff, Z19, Z14, Z15, Z4)
GALOIS(0xaa, 0xff, Z23, Z12, Z13, Z5)
GALOIS(0xaa, 0xff, Z27, Z10, Z11, Z6)
GALOIS(0xaa, 0xff, Z31, Z8, Z9, Z7)
skip_avx512_parallel84:
VMOVDQU64 Z4, (DX)
VMOVDQU64 Z5, (CX)
VMOVDQU64 Z6, (R10)
VMOVDQU64 Z7, (R12)
ADDQ $64, R11 // in4+=64
ADDQ $64, DX // out+=64
ADDQ $64, CX // out2+=64
ADDQ $64, R10 // out3+=64
ADDQ $64, R12 // out4+=64
SUBQ $1, R9
JNZ loopback_avx512_parallel84
done_avx512_parallel84:
VZEROUPPER
RET