ccminer-gostd-lite/qubit/luffa.cu

/*
 * luffa 80 algo (Introduced by Doomcoin)
 */
extern "C" {
#include "sph/sph_luffa.h"
}

#include "miner.h"

#include "cuda_helper.h"

static uint32_t *d_hash[MAX_GPUS];

extern void qubit_luffa512_cpu_init(int thr_id, uint32_t threads);
extern void qubit_luffa512_cpu_setBlock_80(void *pdata);
extern void qubit_luffa512_cpu_hash_80(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_hash, int order);

extern "C" void luffa_hash(void *state, const void *input)
{
	uint8_t _ALIGN(64) hash[64];

	sph_luffa512_context ctx_luffa;

	sph_luffa512_init(&ctx_luffa);
	sph_luffa512 (&ctx_luffa, input, 80);
	sph_luffa512_close(&ctx_luffa, (void*) hash);

	memcpy(state, hash, 32);
}

static bool init[MAX_GPUS] = { 0 };

extern "C" int scanhash_luffa(int thr_id, struct work* work, uint32_t max_nonce, unsigned long *hashes_done)
{
	uint32_t _ALIGN(64) endiandata[20];
	uint32_t *pdata = work->data;
	uint32_t *ptarget = work->target;
	const uint32_t first_nonce = pdata[19];
	uint32_t throughput = cuda_default_throughput(thr_id, 1U << 21);
	if (init[thr_id]) throughput = min(throughput, max_nonce - first_nonce);

	if (opt_benchmark)
		ptarget[7] = 0x0000f;

	if (!init[thr_id])
	{
		cudaSetDevice(device_map[thr_id]);
		if (opt_cudaschedule == -1 && gpu_threads == 1) {
			cudaDeviceReset();
			// reduce cpu usage
			cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);
			CUDA_LOG_ERROR();
		}
		gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput);

		CUDA_SAFE_CALL(cudaMalloc(&d_hash[thr_id], (size_t) 64 * throughput));

		qubit_luffa512_cpu_init(thr_id, throughput);
		cuda_check_cpu_init(thr_id, throughput);

		init[thr_id] = true;
	}

	for (int k=0; k < 19; k++)
		be32enc(&endiandata[k], pdata[k]);

	qubit_luffa512_cpu_setBlock_80((void*)endiandata);
	cuda_check_cpu_setTarget(ptarget);

	do {
		qubit_luffa512_cpu_hash_80(thr_id, (int) throughput, pdata[19], d_hash[thr_id], 0);

		*hashes_done = pdata[19] - first_nonce + throughput;

		work->nonces[0] = cuda_check_hash(thr_id, throughput, pdata[19], d_hash[thr_id]);
		if (work->nonces[0] != UINT32_MAX)
		{
			const uint32_t Htarg = ptarget[7];
			uint32_t _ALIGN(64) vhash[8];
			be32enc(&endiandata[19], work->nonces[0]);
			luffa_hash(vhash, endiandata);

			if (vhash[7] <= Htarg && fulltest(vhash, ptarget)) {
				work->valid_nonces = 1;
				work_set_target_ratio(work, vhash);
				work->nonces[1] = cuda_check_hash_suppl(thr_id, throughput, pdata[19], d_hash[thr_id], 1);
				if (work->nonces[1] != 0) {
					be32enc(&endiandata[19], work->nonces[1]);
					luffa_hash(vhash, endiandata);
					bn_set_target_ratio(work, vhash, 1);
					work->valid_nonces++;
					pdata[19] = max(work->nonces[0], work->nonces[1]) + 1;
				} else {
					pdata[19] = work->nonces[0] + 1; // cursor
				}
				return work->valid_nonces;
			}
			else if (vhash[7] > Htarg) {
				gpu_increment_reject(thr_id);
				if (!opt_quiet)
				gpulog(LOG_WARNING, thr_id, "result for %08x does not validate on CPU!", work->nonces[0]);
				pdata[19] = work->nonces[0] + 1;
				continue;
			}
		}

		if ((uint64_t)throughput + pdata[19] >= max_nonce) {
			pdata[19] = max_nonce;
			break;
		}
		pdata[19] += throughput;

	} while (!work_restart[thr_id].restart);

	*hashes_done = pdata[19] - first_nonce;
	return 0;
}

// cleanup
extern "C" void free_luffa(int thr_id)
{
	if (!init[thr_id])
		return;

	cudaThreadSynchronize();

	cudaFree(d_hash[thr_id]);

	cuda_check_cpu_free(thr_id);

	init[thr_id] = false;
	cudaDeviceSynchronize();
}