forked from cbuchner1/ccminer
/
cuda_lyra2v2.cu
380 lines (303 loc) · 8.91 KB
/
cuda_lyra2v2.cu
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
#include <stdio.h>
#include <stdint.h>
#include <memory.h>
#ifdef __INTELLISENSE__
/* just for vstudio code colors */
#define __CUDA_ARCH__ 500
#endif
#define TPB52 10
#define TPB50 16
#include "cuda_lyra2v2_sm3.cuh"
#if __CUDA_ARCH__ >= 500
#include "cuda_lyra2_vectors.h"
#define Nrow 4
#define Ncol 4
#define u64type uint2
#define vectype uint28
#define memshift 3
__device__ vectype *DMatrix;
__device__ __forceinline__
void Gfunc_v5(uint2 &a, uint2 &b, uint2 &c, uint2 &d)
{
a += b; d ^= a; d = SWAPUINT2(d);
c += d; b ^= c; b = ROR24(b);
a += b; d ^= a; d = ROR16(d);
c += d; b ^= c; b = ROR2(b, 63);
}
__device__ __forceinline__
void round_lyra_v5(vectype* s)
{
Gfunc_v5(s[0].x, s[1].x, s[2].x, s[3].x);
Gfunc_v5(s[0].y, s[1].y, s[2].y, s[3].y);
Gfunc_v5(s[0].z, s[1].z, s[2].z, s[3].z);
Gfunc_v5(s[0].w, s[1].w, s[2].w, s[3].w);
Gfunc_v5(s[0].x, s[1].y, s[2].z, s[3].w);
Gfunc_v5(s[0].y, s[1].z, s[2].w, s[3].x);
Gfunc_v5(s[0].z, s[1].w, s[2].x, s[3].y);
Gfunc_v5(s[0].w, s[1].x, s[2].y, s[3].z);
}
__device__ __forceinline__
void reduceDuplex(vectype state[4], uint32_t thread)
{
vectype state1[3];
uint32_t ps1 = (Nrow * Ncol * memshift * thread);
uint32_t ps2 = (memshift * (Ncol-1) + memshift * Ncol + Nrow * Ncol * memshift * thread);
#pragma unroll 4
for (int i = 0; i < Ncol; i++)
{
uint32_t s1 = ps1 + i*memshift;
uint32_t s2 = ps2 - i*memshift;
#pragma unroll
for (int j = 0; j < 3; j++)
state1[j] = __ldg4(&(DMatrix+s1)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
state[j] ^= state1[j];
round_lyra_v5(state);
#pragma unroll
for (int j = 0; j < 3; j++)
state1[j] ^= state[j];
#pragma unroll
for (int j = 0; j < 3; j++)
(DMatrix + s2)[j] = state1[j];
}
}
__device__ __forceinline__
void reduceDuplex50(vectype state[4], uint32_t thread)
{
uint32_t ps1 = (Nrow * Ncol * memshift * thread);
uint32_t ps2 = (memshift * (Ncol - 1) + memshift * Ncol + Nrow * Ncol * memshift * thread);
#pragma unroll 4
for (int i = 0; i < Ncol; i++)
{
uint32_t s1 = ps1 + i*memshift;
uint32_t s2 = ps2 - i*memshift;
#pragma unroll
for (int j = 0; j < 3; j++)
state[j] ^= __ldg4(&(DMatrix + s1)[j]);
round_lyra_v5(state);
#pragma unroll
for (int j = 0; j < 3; j++)
(DMatrix + s2)[j] = __ldg4(&(DMatrix + s1)[j]) ^ state[j];
}
}
__device__ __forceinline__
void reduceDuplexRowSetupV2(const int rowIn, const int rowInOut, const int rowOut, vectype state[4], uint32_t thread)
{
vectype state2[3], state1[3];
uint32_t ps1 = (memshift * Ncol * rowIn + Nrow * Ncol * memshift * thread);
uint32_t ps2 = (memshift * Ncol * rowInOut + Nrow * Ncol * memshift * thread);
uint32_t ps3 = (memshift * (Ncol-1) + memshift * Ncol * rowOut + Nrow * Ncol * memshift * thread);
for (int i = 0; i < Ncol; i++)
{
uint32_t s1 = ps1 + i*memshift;
uint32_t s2 = ps2 + i*memshift;
uint32_t s3 = ps3 - i*memshift;
#if __CUDA_ARCH__ == 500
#pragma unroll
for (int j = 0; j < 3; j++)
state[j] = state[j] ^ (__ldg4(&(DMatrix + s1)[j]) + __ldg4(&(DMatrix + s2)[j]));
round_lyra_v5(state);
#pragma unroll
for (int j = 0; j < 3; j++)
state1[j] = __ldg4(&(DMatrix + s1)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
state2[j] = __ldg4(&(DMatrix + s2)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
{
state1[j] ^= state[j];
(DMatrix + s3)[j] = state1[j];
}
#else /* 5.2 */
#pragma unroll
for (int j = 0; j < 3; j++)
state1[j] = __ldg4(&(DMatrix + s1)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
state2[j] = __ldg4(&(DMatrix + s2)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
{
vectype tmp = state1[j] + state2[j];
state[j] ^= tmp;
}
round_lyra_v5(state);
#pragma unroll
for (int j = 0; j < 3; j++)
{
state1[j] ^= state[j];
(DMatrix + s3)[j] = state1[j];
}
#endif
((uint2*)state2)[0] ^= ((uint2*)state)[11];
#pragma unroll
for (int j = 0; j < 11; j++)
((uint2*)state2)[j+1] ^= ((uint2*)state)[j];
#pragma unroll
for (int j = 0; j < 3; j++)
(DMatrix + s2)[j] = state2[j];
}
}
__device__ __forceinline__
void reduceDuplexRowtV2(const int rowIn, const int rowInOut, const int rowOut, vectype* state, uint32_t thread)
{
vectype state1[3],state2[3];
uint32_t ps1 = (memshift * Ncol * rowIn + Nrow * Ncol * memshift * thread);
uint32_t ps2 = (memshift * Ncol * rowInOut + Nrow * Ncol * memshift * thread);
uint32_t ps3 = (memshift * Ncol * rowOut + Nrow * Ncol * memshift * thread);
for (int i = 0; i < Ncol; i++)
{
uint32_t s1 = ps1 + i*memshift;
uint32_t s2 = ps2 + i*memshift;
uint32_t s3 = ps3 + i*memshift;
#pragma unroll
for (int j = 0; j < 3; j++)
state1[j] = __ldg4(&(DMatrix + s1)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
state2[j] = __ldg4(&(DMatrix + s2)[j]);
#pragma unroll
for (int j = 0; j < 3; j++)
state1[j] += state2[j];
#pragma unroll
for (int j = 0; j < 3; j++)
state[j] ^= state1[j];
round_lyra_v5(state);
((uint2*)state2)[0] ^= ((uint2*)state)[11];
#pragma unroll
for (int j = 0; j < 11; j++)
((uint2*)state2)[j + 1] ^= ((uint2*)state)[j];
#if __CUDA_ARCH__ == 500
if (rowInOut != rowOut)
{
#pragma unroll
for (int j = 0; j < 3; j++)
(DMatrix + s3)[j] ^= state[j];
}
if (rowInOut == rowOut)
{
#pragma unroll
for (int j = 0; j < 3; j++)
state2[j] ^= state[j];
}
#else
if (rowInOut != rowOut)
{
#pragma unroll
for (int j = 0; j < 3; j++)
(DMatrix + s3)[j] ^= state[j];
} else {
#pragma unroll
for (int j = 0; j < 3; j++)
state2[j] ^= state[j];
}
#endif
#pragma unroll
for (int j = 0; j < 3; j++)
(DMatrix + s2)[j] = state2[j];
}
}
#if __CUDA_ARCH__ == 500
__global__ __launch_bounds__(TPB50, 1)
#else
__global__ __launch_bounds__(TPB52, 1)
#endif
void lyra2v2_gpu_hash_32(uint32_t threads, uint32_t startNounce, uint2 *outputHash)
{
const uint32_t thread = (blockDim.x * blockIdx.x + threadIdx.x);
vectype state[4];
uint28 blake2b_IV[2];
if (threadIdx.x == 0) {
((uint16*)blake2b_IV)[0] = make_uint16(
0xf3bcc908, 0x6a09e667, 0x84caa73b, 0xbb67ae85,
0xfe94f82b, 0x3c6ef372, 0x5f1d36f1, 0xa54ff53a,
0xade682d1, 0x510e527f, 0x2b3e6c1f, 0x9b05688c,
0xfb41bd6b, 0x1f83d9ab, 0x137e2179, 0x5be0cd19
);
}
if (thread < threads)
{
((uint2*)state)[0] = __ldg(&outputHash[thread]);
((uint2*)state)[1] = __ldg(&outputHash[thread + threads]);
((uint2*)state)[2] = __ldg(&outputHash[thread + 2 * threads]);
((uint2*)state)[3] = __ldg(&outputHash[thread + 3 * threads]);
state[1] = state[0];
state[2] = ((blake2b_IV)[0]);
state[3] = ((blake2b_IV)[1]);
for (int i = 0; i<12; i++)
round_lyra_v5(state);
((uint2*)state)[0].x ^= 0x20;
((uint2*)state)[1].x ^= 0x20;
((uint2*)state)[2].x ^= 0x20;
((uint2*)state)[3].x ^= 0x01;
((uint2*)state)[4].x ^= 0x04;
((uint2*)state)[5].x ^= 0x04;
((uint2*)state)[6].x ^= 0x80;
((uint2*)state)[7].y ^= 0x01000000;
for (int i = 0; i<12; i++)
round_lyra_v5(state);
uint32_t ps1 = (memshift * (Ncol - 1) + Nrow * Ncol * memshift * thread);
for (int i = 0; i < Ncol; i++)
{
const uint32_t s1 = ps1 - memshift * i;
DMatrix[s1] = state[0];
DMatrix[s1+1] = state[1];
DMatrix[s1+2] = state[2];
round_lyra_v5(state);
}
reduceDuplex50(state, thread);
reduceDuplexRowSetupV2(1, 0, 2, state, thread);
reduceDuplexRowSetupV2(2, 1, 3, state, thread);
uint32_t rowa;
int prev=3;
for (int i = 0; i < 4; i++)
{
rowa = ((uint2*)state)[0].x & 3;
reduceDuplexRowtV2(prev, rowa, i, state, thread);
prev=i;
}
const uint32_t shift = (memshift * Ncol * rowa + Nrow * Ncol * memshift * thread);
#pragma unroll
for (int j = 0; j < 3; j++)
state[j] ^= __ldg4(&(DMatrix + shift)[j]);
for (int i = 0; i < 12; i++)
round_lyra_v5(state);
outputHash[thread] = ((uint2*)state)[0];
outputHash[thread + threads] = ((uint2*)state)[1];
outputHash[thread + 2 * threads] = ((uint2*)state)[2];
outputHash[thread + 3 * threads] = ((uint2*)state)[3];
}
}
#else
#include "cuda_helper.h"
__device__ void* DMatrix;
__global__ void lyra2v2_gpu_hash_32(uint32_t threads, uint32_t startNounce, uint2 *outputHash) {}
#endif
__host__
void lyra2v2_cpu_init(int thr_id, uint32_t threads, uint64_t *d_matrix)
{
// just assign the device pointer allocated in main loop
cudaMemcpyToSymbol(DMatrix, &d_matrix, sizeof(uint64_t*), 0, cudaMemcpyHostToDevice);
}
__host__
void lyra2v2_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNounce, uint64_t *d_outputHash, int order)
{
uint32_t tpb;
if (device_sm[device_map[thr_id]] < 350)
tpb = TPB30;
else if (device_sm[device_map[thr_id]] == 350)
tpb = TPB35;
else if (device_sm[device_map[thr_id]] == 500)
tpb = TPB50;
else
tpb = TPB52;
dim3 grid((threads + tpb - 1) / tpb);
dim3 block(tpb);
if (device_sm[device_map[thr_id]] >= 500)
lyra2v2_gpu_hash_32 <<<grid, block>>> (threads, startNounce, (uint2*)d_outputHash);
else
lyra2v2_gpu_hash_32_v3 <<<grid, block>>> (threads, startNounce, (uint2*)d_outputHash);
//MyStreamSynchronize(NULL, order, thr_id);
}