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
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
|
package ethash
/* DO NOT EDIT!!!
This code is version controlled at
https://github.com/ethereum/cpp-ethereum/blob/develop/libethash-cl/ethash_cl_miner_kernel.cl
If needed change it there first, then copy over here.
*/
const kernel = `
// author Tim Hughes <tim@twistedfury.com>
// Tested on Radeon HD 7850
// Hashrate: 15940347 hashes/s
// Bandwidth: 124533 MB/s
// search kernel should fit in <= 84 VGPRS (3 wavefronts)
#define THREADS_PER_HASH (128 / 16)
#define HASHES_PER_LOOP (GROUP_SIZE / THREADS_PER_HASH)
#define FNV_PRIME 0x01000193
__constant uint2 const Keccak_f1600_RC[24] = {
(uint2)(0x00000001, 0x00000000),
(uint2)(0x00008082, 0x00000000),
(uint2)(0x0000808a, 0x80000000),
(uint2)(0x80008000, 0x80000000),
(uint2)(0x0000808b, 0x00000000),
(uint2)(0x80000001, 0x00000000),
(uint2)(0x80008081, 0x80000000),
(uint2)(0x00008009, 0x80000000),
(uint2)(0x0000008a, 0x00000000),
(uint2)(0x00000088, 0x00000000),
(uint2)(0x80008009, 0x00000000),
(uint2)(0x8000000a, 0x00000000),
(uint2)(0x8000808b, 0x00000000),
(uint2)(0x0000008b, 0x80000000),
(uint2)(0x00008089, 0x80000000),
(uint2)(0x00008003, 0x80000000),
(uint2)(0x00008002, 0x80000000),
(uint2)(0x00000080, 0x80000000),
(uint2)(0x0000800a, 0x00000000),
(uint2)(0x8000000a, 0x80000000),
(uint2)(0x80008081, 0x80000000),
(uint2)(0x00008080, 0x80000000),
(uint2)(0x80000001, 0x00000000),
(uint2)(0x80008008, 0x80000000),
};
void keccak_f1600_round(uint2* a, uint r, uint out_size)
{
#if !__ENDIAN_LITTLE__
for (uint i = 0; i != 25; ++i)
a[i] = a[i].yx;
#endif
uint2 b[25];
uint2 t;
// Theta
b[0] = a[0] ^ a[5] ^ a[10] ^ a[15] ^ a[20];
b[1] = a[1] ^ a[6] ^ a[11] ^ a[16] ^ a[21];
b[2] = a[2] ^ a[7] ^ a[12] ^ a[17] ^ a[22];
b[3] = a[3] ^ a[8] ^ a[13] ^ a[18] ^ a[23];
b[4] = a[4] ^ a[9] ^ a[14] ^ a[19] ^ a[24];
t = b[4] ^ (uint2)(b[1].x << 1 | b[1].y >> 31, b[1].y << 1 | b[1].x >> 31);
a[0] ^= t;
a[5] ^= t;
a[10] ^= t;
a[15] ^= t;
a[20] ^= t;
t = b[0] ^ (uint2)(b[2].x << 1 | b[2].y >> 31, b[2].y << 1 | b[2].x >> 31);
a[1] ^= t;
a[6] ^= t;
a[11] ^= t;
a[16] ^= t;
a[21] ^= t;
t = b[1] ^ (uint2)(b[3].x << 1 | b[3].y >> 31, b[3].y << 1 | b[3].x >> 31);
a[2] ^= t;
a[7] ^= t;
a[12] ^= t;
a[17] ^= t;
a[22] ^= t;
t = b[2] ^ (uint2)(b[4].x << 1 | b[4].y >> 31, b[4].y << 1 | b[4].x >> 31);
a[3] ^= t;
a[8] ^= t;
a[13] ^= t;
a[18] ^= t;
a[23] ^= t;
t = b[3] ^ (uint2)(b[0].x << 1 | b[0].y >> 31, b[0].y << 1 | b[0].x >> 31);
a[4] ^= t;
a[9] ^= t;
a[14] ^= t;
a[19] ^= t;
a[24] ^= t;
// Rho Pi
b[0] = a[0];
b[10] = (uint2)(a[1].x << 1 | a[1].y >> 31, a[1].y << 1 | a[1].x >> 31);
b[7] = (uint2)(a[10].x << 3 | a[10].y >> 29, a[10].y << 3 | a[10].x >> 29);
b[11] = (uint2)(a[7].x << 6 | a[7].y >> 26, a[7].y << 6 | a[7].x >> 26);
b[17] = (uint2)(a[11].x << 10 | a[11].y >> 22, a[11].y << 10 | a[11].x >> 22);
b[18] = (uint2)(a[17].x << 15 | a[17].y >> 17, a[17].y << 15 | a[17].x >> 17);
b[3] = (uint2)(a[18].x << 21 | a[18].y >> 11, a[18].y << 21 | a[18].x >> 11);
b[5] = (uint2)(a[3].x << 28 | a[3].y >> 4, a[3].y << 28 | a[3].x >> 4);
b[16] = (uint2)(a[5].y << 4 | a[5].x >> 28, a[5].x << 4 | a[5].y >> 28);
b[8] = (uint2)(a[16].y << 13 | a[16].x >> 19, a[16].x << 13 | a[16].y >> 19);
b[21] = (uint2)(a[8].y << 23 | a[8].x >> 9, a[8].x << 23 | a[8].y >> 9);
b[24] = (uint2)(a[21].x << 2 | a[21].y >> 30, a[21].y << 2 | a[21].x >> 30);
b[4] = (uint2)(a[24].x << 14 | a[24].y >> 18, a[24].y << 14 | a[24].x >> 18);
b[15] = (uint2)(a[4].x << 27 | a[4].y >> 5, a[4].y << 27 | a[4].x >> 5);
b[23] = (uint2)(a[15].y << 9 | a[15].x >> 23, a[15].x << 9 | a[15].y >> 23);
b[19] = (uint2)(a[23].y << 24 | a[23].x >> 8, a[23].x << 24 | a[23].y >> 8);
b[13] = (uint2)(a[19].x << 8 | a[19].y >> 24, a[19].y << 8 | a[19].x >> 24);
b[12] = (uint2)(a[13].x << 25 | a[13].y >> 7, a[13].y << 25 | a[13].x >> 7);
b[2] = (uint2)(a[12].y << 11 | a[12].x >> 21, a[12].x << 11 | a[12].y >> 21);
b[20] = (uint2)(a[2].y << 30 | a[2].x >> 2, a[2].x << 30 | a[2].y >> 2);
b[14] = (uint2)(a[20].x << 18 | a[20].y >> 14, a[20].y << 18 | a[20].x >> 14);
b[22] = (uint2)(a[14].y << 7 | a[14].x >> 25, a[14].x << 7 | a[14].y >> 25);
b[9] = (uint2)(a[22].y << 29 | a[22].x >> 3, a[22].x << 29 | a[22].y >> 3);
b[6] = (uint2)(a[9].x << 20 | a[9].y >> 12, a[9].y << 20 | a[9].x >> 12);
b[1] = (uint2)(a[6].y << 12 | a[6].x >> 20, a[6].x << 12 | a[6].y >> 20);
// Chi
a[0] = bitselect(b[0] ^ b[2], b[0], b[1]);
a[1] = bitselect(b[1] ^ b[3], b[1], b[2]);
a[2] = bitselect(b[2] ^ b[4], b[2], b[3]);
a[3] = bitselect(b[3] ^ b[0], b[3], b[4]);
if (out_size >= 4)
{
a[4] = bitselect(b[4] ^ b[1], b[4], b[0]);
a[5] = bitselect(b[5] ^ b[7], b[5], b[6]);
a[6] = bitselect(b[6] ^ b[8], b[6], b[7]);
a[7] = bitselect(b[7] ^ b[9], b[7], b[8]);
a[8] = bitselect(b[8] ^ b[5], b[8], b[9]);
if (out_size >= 8)
{
a[9] = bitselect(b[9] ^ b[6], b[9], b[5]);
a[10] = bitselect(b[10] ^ b[12], b[10], b[11]);
a[11] = bitselect(b[11] ^ b[13], b[11], b[12]);
a[12] = bitselect(b[12] ^ b[14], b[12], b[13]);
a[13] = bitselect(b[13] ^ b[10], b[13], b[14]);
a[14] = bitselect(b[14] ^ b[11], b[14], b[10]);
a[15] = bitselect(b[15] ^ b[17], b[15], b[16]);
a[16] = bitselect(b[16] ^ b[18], b[16], b[17]);
a[17] = bitselect(b[17] ^ b[19], b[17], b[18]);
a[18] = bitselect(b[18] ^ b[15], b[18], b[19]);
a[19] = bitselect(b[19] ^ b[16], b[19], b[15]);
a[20] = bitselect(b[20] ^ b[22], b[20], b[21]);
a[21] = bitselect(b[21] ^ b[23], b[21], b[22]);
a[22] = bitselect(b[22] ^ b[24], b[22], b[23]);
a[23] = bitselect(b[23] ^ b[20], b[23], b[24]);
a[24] = bitselect(b[24] ^ b[21], b[24], b[20]);
}
}
// Iota
a[0] ^= Keccak_f1600_RC[r];
#if !__ENDIAN_LITTLE__
for (uint i = 0; i != 25; ++i)
a[i] = a[i].yx;
#endif
}
void keccak_f1600_no_absorb(ulong* a, uint in_size, uint out_size, uint isolate)
{
for (uint i = in_size; i != 25; ++i)
{
a[i] = 0;
}
#if __ENDIAN_LITTLE__
a[in_size] ^= 0x0000000000000001;
a[24-out_size*2] ^= 0x8000000000000000;
#else
a[in_size] ^= 0x0100000000000000;
a[24-out_size*2] ^= 0x0000000000000080;
#endif
// Originally I unrolled the first and last rounds to interface
// better with surrounding code, however I haven't done this
// without causing the AMD compiler to blow up the VGPR usage.
uint r = 0;
do
{
// This dynamic branch stops the AMD compiler unrolling the loop
// and additionally saves about 33% of the VGPRs, enough to gain another
// wavefront. Ideally we'd get 4 in flight, but 3 is the best I can
// massage out of the compiler. It doesn't really seem to matter how
// much we try and help the compiler save VGPRs because it seems to throw
// that information away, hence the implementation of keccak here
// doesn't bother.
if (isolate)
{
keccak_f1600_round((uint2*)a, r++, 25);
}
}
while (r < 23);
// final round optimised for digest size
keccak_f1600_round((uint2*)a, r++, out_size);
}
#define copy(dst, src, count) for (uint i = 0; i != count; ++i) { (dst)[i] = (src)[i]; }
#define countof(x) (sizeof(x) / sizeof(x[0]))
uint fnv(uint x, uint y)
{
return x * FNV_PRIME ^ y;
}
uint4 fnv4(uint4 x, uint4 y)
{
return x * FNV_PRIME ^ y;
}
uint fnv_reduce(uint4 v)
{
return fnv(fnv(fnv(v.x, v.y), v.z), v.w);
}
typedef union
{
ulong ulongs[32 / sizeof(ulong)];
uint uints[32 / sizeof(uint)];
} hash32_t;
typedef union
{
ulong ulongs[64 / sizeof(ulong)];
uint4 uint4s[64 / sizeof(uint4)];
} hash64_t;
typedef union
{
uint uints[128 / sizeof(uint)];
uint4 uint4s[128 / sizeof(uint4)];
} hash128_t;
hash64_t init_hash(__constant hash32_t const* header, ulong nonce, uint isolate)
{
hash64_t init;
uint const init_size = countof(init.ulongs);
uint const hash_size = countof(header->ulongs);
// sha3_512(header .. nonce)
ulong state[25];
copy(state, header->ulongs, hash_size);
state[hash_size] = nonce;
keccak_f1600_no_absorb(state, hash_size + 1, init_size, isolate);
copy(init.ulongs, state, init_size);
return init;
}
uint inner_loop_chunks(uint4 init, uint thread_id, __local uint* share, __global hash128_t const* g_dag, __global hash128_t const* g_dag1, __global hash128_t const* g_dag2, __global hash128_t const* g_dag3, uint isolate)
{
uint4 mix = init;
// share init0
if (thread_id == 0)
*share = mix.x;
barrier(CLK_LOCAL_MEM_FENCE);
uint init0 = *share;
uint a = 0;
do
{
bool update_share = thread_id == (a/4) % THREADS_PER_HASH;
#pragma unroll
for (uint i = 0; i != 4; ++i)
{
if (update_share)
{
uint m[4] = { mix.x, mix.y, mix.z, mix.w };
*share = fnv(init0 ^ (a+i), m[i]) % DAG_SIZE;
}
barrier(CLK_LOCAL_MEM_FENCE);
mix = fnv4(mix, *share>=3 * DAG_SIZE / 4 ? g_dag3[*share - 3 * DAG_SIZE / 4].uint4s[thread_id] : *share>=DAG_SIZE / 2 ? g_dag2[*share - DAG_SIZE / 2].uint4s[thread_id] : *share>=DAG_SIZE / 4 ? g_dag1[*share - DAG_SIZE / 4].uint4s[thread_id]:g_dag[*share].uint4s[thread_id]);
}
} while ((a += 4) != (ACCESSES & isolate));
return fnv_reduce(mix);
}
uint inner_loop(uint4 init, uint thread_id, __local uint* share, __global hash128_t const* g_dag, uint isolate)
{
uint4 mix = init;
// share init0
if (thread_id == 0)
*share = mix.x;
barrier(CLK_LOCAL_MEM_FENCE);
uint init0 = *share;
uint a = 0;
do
{
bool update_share = thread_id == (a/4) % THREADS_PER_HASH;
#pragma unroll
for (uint i = 0; i != 4; ++i)
{
if (update_share)
{
uint m[4] = { mix.x, mix.y, mix.z, mix.w };
*share = fnv(init0 ^ (a+i), m[i]) % DAG_SIZE;
}
barrier(CLK_LOCAL_MEM_FENCE);
mix = fnv4(mix, g_dag[*share].uint4s[thread_id]);
}
}
while ((a += 4) != (ACCESSES & isolate));
return fnv_reduce(mix);
}
hash32_t final_hash(hash64_t const* init, hash32_t const* mix, uint isolate)
{
ulong state[25];
hash32_t hash;
uint const hash_size = countof(hash.ulongs);
uint const init_size = countof(init->ulongs);
uint const mix_size = countof(mix->ulongs);
// keccak_256(keccak_512(header..nonce) .. mix);
copy(state, init->ulongs, init_size);
copy(state + init_size, mix->ulongs, mix_size);
keccak_f1600_no_absorb(state, init_size+mix_size, hash_size, isolate);
// copy out
copy(hash.ulongs, state, hash_size);
return hash;
}
hash32_t compute_hash_simple(
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
ulong nonce,
uint isolate
)
{
hash64_t init = init_hash(g_header, nonce, isolate);
hash128_t mix;
for (uint i = 0; i != countof(mix.uint4s); ++i)
{
mix.uint4s[i] = init.uint4s[i % countof(init.uint4s)];
}
uint mix_val = mix.uints[0];
uint init0 = mix.uints[0];
uint a = 0;
do
{
uint pi = fnv(init0 ^ a, mix_val) % DAG_SIZE;
uint n = (a+1) % countof(mix.uints);
#pragma unroll
for (uint i = 0; i != countof(mix.uints); ++i)
{
mix.uints[i] = fnv(mix.uints[i], g_dag[pi].uints[i]);
mix_val = i == n ? mix.uints[i] : mix_val;
}
}
while (++a != (ACCESSES & isolate));
// reduce to output
hash32_t fnv_mix;
for (uint i = 0; i != countof(fnv_mix.uints); ++i)
{
fnv_mix.uints[i] = fnv_reduce(mix.uint4s[i]);
}
return final_hash(&init, &fnv_mix, isolate);
}
typedef union
{
struct
{
hash64_t init;
uint pad; // avoid lds bank conflicts
};
hash32_t mix;
} compute_hash_share;
hash32_t compute_hash(
__local compute_hash_share* share,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
ulong nonce,
uint isolate
)
{
uint const gid = get_global_id(0);
// Compute one init hash per work item.
hash64_t init = init_hash(g_header, nonce, isolate);
// Threads work together in this phase in groups of 8.
uint const thread_id = gid % THREADS_PER_HASH;
uint const hash_id = (gid % GROUP_SIZE) / THREADS_PER_HASH;
hash32_t mix;
uint i = 0;
do
{
// share init with other threads
if (i == thread_id)
share[hash_id].init = init;
barrier(CLK_LOCAL_MEM_FENCE);
uint4 thread_init = share[hash_id].init.uint4s[thread_id % (64 / sizeof(uint4))];
barrier(CLK_LOCAL_MEM_FENCE);
uint thread_mix = inner_loop(thread_init, thread_id, share[hash_id].mix.uints, g_dag, isolate);
share[hash_id].mix.uints[thread_id] = thread_mix;
barrier(CLK_LOCAL_MEM_FENCE);
if (i == thread_id)
mix = share[hash_id].mix;
barrier(CLK_LOCAL_MEM_FENCE);
}
while (++i != (THREADS_PER_HASH & isolate));
return final_hash(&init, &mix, isolate);
}
hash32_t compute_hash_chunks(
__local compute_hash_share* share,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
__global hash128_t const* g_dag1,
__global hash128_t const* g_dag2,
__global hash128_t const* g_dag3,
ulong nonce,
uint isolate
)
{
uint const gid = get_global_id(0);
// Compute one init hash per work item.
hash64_t init = init_hash(g_header, nonce, isolate);
// Threads work together in this phase in groups of 8.
uint const thread_id = gid % THREADS_PER_HASH;
uint const hash_id = (gid % GROUP_SIZE) / THREADS_PER_HASH;
hash32_t mix;
uint i = 0;
do
{
// share init with other threads
if (i == thread_id)
share[hash_id].init = init;
barrier(CLK_LOCAL_MEM_FENCE);
uint4 thread_init = share[hash_id].init.uint4s[thread_id % (64 / sizeof(uint4))];
barrier(CLK_LOCAL_MEM_FENCE);
uint thread_mix = inner_loop_chunks(thread_init, thread_id, share[hash_id].mix.uints, g_dag, g_dag1, g_dag2, g_dag3, isolate);
share[hash_id].mix.uints[thread_id] = thread_mix;
barrier(CLK_LOCAL_MEM_FENCE);
if (i == thread_id)
mix = share[hash_id].mix;
barrier(CLK_LOCAL_MEM_FENCE);
}
while (++i != (THREADS_PER_HASH & isolate));
return final_hash(&init, &mix, isolate);
}
__attribute__((reqd_work_group_size(GROUP_SIZE, 1, 1)))
__kernel void ethash_hash_simple(
__global hash32_t* g_hashes,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
ulong start_nonce,
uint isolate
)
{
uint const gid = get_global_id(0);
g_hashes[gid] = compute_hash_simple(g_header, g_dag, start_nonce + gid, isolate);
}
__attribute__((reqd_work_group_size(GROUP_SIZE, 1, 1)))
__kernel void ethash_search_simple(
__global volatile uint* restrict g_output,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
ulong start_nonce,
ulong target,
uint isolate
)
{
uint const gid = get_global_id(0);
hash32_t hash = compute_hash_simple(g_header, g_dag, start_nonce + gid, isolate);
if (hash.ulongs[countof(hash.ulongs)-1] < target)
{
uint slot = min(convert_uint(MAX_OUTPUTS), convert_uint(atomic_inc(&g_output[0]) + 1));
g_output[slot] = gid;
}
}
__attribute__((reqd_work_group_size(GROUP_SIZE, 1, 1)))
__kernel void ethash_hash(
__global hash32_t* g_hashes,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
ulong start_nonce,
uint isolate
)
{
__local compute_hash_share share[HASHES_PER_LOOP];
uint const gid = get_global_id(0);
g_hashes[gid] = compute_hash(share, g_header, g_dag, start_nonce + gid, isolate);
}
__attribute__((reqd_work_group_size(GROUP_SIZE, 1, 1)))
__kernel void ethash_search(
__global volatile uint* restrict g_output,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
ulong start_nonce,
ulong target,
uint isolate
)
{
__local compute_hash_share share[HASHES_PER_LOOP];
uint const gid = get_global_id(0);
hash32_t hash = compute_hash(share, g_header, g_dag, start_nonce + gid, isolate);
if (as_ulong(as_uchar8(hash.ulongs[0]).s76543210) < target)
{
uint slot = min((uint)MAX_OUTPUTS, atomic_inc(&g_output[0]) + 1);
g_output[slot] = gid;
}
}
__attribute__((reqd_work_group_size(GROUP_SIZE, 1, 1)))
__kernel void ethash_hash_chunks(
__global hash32_t* g_hashes,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
__global hash128_t const* g_dag1,
__global hash128_t const* g_dag2,
__global hash128_t const* g_dag3,
ulong start_nonce,
uint isolate
)
{
__local compute_hash_share share[HASHES_PER_LOOP];
uint const gid = get_global_id(0);
g_hashes[gid] = compute_hash_chunks(share, g_header, g_dag, g_dag1, g_dag2, g_dag3,start_nonce + gid, isolate);
}
__attribute__((reqd_work_group_size(GROUP_SIZE, 1, 1)))
__kernel void ethash_search_chunks(
__global volatile uint* restrict g_output,
__constant hash32_t const* g_header,
__global hash128_t const* g_dag,
__global hash128_t const* g_dag1,
__global hash128_t const* g_dag2,
__global hash128_t const* g_dag3,
ulong start_nonce,
ulong target,
uint isolate
)
{
__local compute_hash_share share[HASHES_PER_LOOP];
uint const gid = get_global_id(0);
hash32_t hash = compute_hash_chunks(share, g_header, g_dag, g_dag1, g_dag2, g_dag3, start_nonce + gid, isolate);
if (as_ulong(as_uchar8(hash.ulongs[0]).s76543210) < target)
{
uint slot = min(convert_uint(MAX_OUTPUTS), convert_uint(atomic_inc(&g_output[0]) + 1));
g_output[slot] = gid;
}
}
`
|