extern "C" {
#include "sph/sph_blake.h"
#include "sph/sph_bmw.h"
#include "sph/sph_skein.h"
#include "sph/sph_keccak.h"
#include "sph/sph_cubehash.h"
#include "lyra2/Lyra2.h"
}

#include "miner.h"
#include "cuda_helper.h"

#include <math.h>

static uint64_t *d_hash[MAX_GPUS];
static uint64_t* d_matrix[MAX_GPUS];

extern void blake256_cpu_init(int thr_id, uint32_t threads);
extern void blake256_cpu_hash_80(const int thr_id, const uint32_t threads, const uint32_t startNonce, uint64_t *Hash, int order);
extern void blake256_cpu_setBlock_80(uint32_t *pdata);
extern void keccak256_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNonce, uint64_t *d_outputHash, int order);
extern void keccak256_cpu_init(int thr_id, uint32_t threads);
extern void keccak256_cpu_free(int thr_id);
extern void blakeKeccak256_cpu_hash_80(const int thr_id, const uint32_t threads, const uint32_t startNonce, uint64_t *Hash, int order);
extern void blakeKeccakcube256_cpu_hash_80(const int thr_id, const uint32_t threads, const uint32_t startNonce, uint64_t *Hash, int order);

extern void skein256_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNonce, uint64_t *d_outputHash, int order);
extern void skein256_cpu_init(int thr_id, uint32_t threads);
extern void cubehash256_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNounce, uint64_t *d_hash, int order);

extern void lyra2v2_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNonce, uint64_t *d_outputHash, int order);
extern void lyra2v2_cpu_init(int thr_id, uint32_t threads, uint64_t* d_matrix);

//extern void bmw256_setTarget(const void *ptarget);
extern void bmw256_cpu_init(int thr_id, uint32_t threads);
extern void bmw256_cpu_free(int thr_id);
extern void bmw256_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNounce, uint64_t *g_hash, uint32_t *resultnonces, uint32_t Target);
extern void bmw256_cpu_hash_32(int thr_id, uint32_t threads, uint32_t startNounce, uint64_t *g_hash, uint32_t *resultnonces, uint32_t Target, uint32_t **result);

void lyra2v2_hash(void *state, const void *input)
{
	uint32_t hashA[8], hashB[8];

	sph_blake256_context      ctx_blake;
	sph_keccak256_context     ctx_keccak;
	sph_skein256_context      ctx_skein;
	sph_bmw256_context        ctx_bmw;
	sph_cubehash256_context   ctx_cube;

	sph_blake256_set_rounds(14);

	sph_blake256_init(&ctx_blake);
	sph_blake256(&ctx_blake, input, 80);
	sph_blake256_close(&ctx_blake, hashA);

	sph_keccak256_init(&ctx_keccak);
	sph_keccak256(&ctx_keccak, hashA, 32);
	sph_keccak256_close(&ctx_keccak, hashB);

	sph_cubehash256_init(&ctx_cube);
	sph_cubehash256(&ctx_cube, hashB, 32);
	sph_cubehash256_close(&ctx_cube, hashA);

	LYRA2(hashB, 32, hashA, 32, hashA, 32, 1, 4, 4);

	sph_skein256_init(&ctx_skein);
	sph_skein256(&ctx_skein, hashB, 32);
	sph_skein256_close(&ctx_skein, hashA);

	sph_cubehash256_init(&ctx_cube);
	sph_cubehash256(&ctx_cube, hashA, 32);
	sph_cubehash256_close(&ctx_cube, hashB);

	sph_bmw256_init(&ctx_bmw);
	sph_bmw256(&ctx_bmw, hashB, 32);
	sph_bmw256_close(&ctx_bmw, hashA);

	memcpy(state, hashA, 32);
}

#ifdef _DEBUG
#define TRACE(algo) { \
	if (max_nonce == 1 && pdata[19] <= 1) { \
		uint32_t* debugbuf = NULL; \
		cudaMallocHost(&debugbuf, 32); \
		cudaMemcpy(debugbuf, d_hash[thr_id], 32, cudaMemcpyDeviceToHost); \
		printf("lyra2 %s %08x %08x %08x %08x...%08x... ¥n", algo, swab32(debugbuf[0]), swab32(debugbuf[1]), \
			swab32(debugbuf[2]), swab32(debugbuf[3]), swab32(debugbuf[7])); \
		cudaFreeHost(debugbuf); \
	} \
}
#else
#define TRACE(algo) {}
#endif

static bool init[MAX_GPUS] = { 0 };
static uint32_t throughput[MAX_GPUS] = { 0 };

extern "C" int scanhash_lyra2v2(int thr_id, struct work* work, uint32_t max_nonce, unsigned long *hashes_done)
{
	uint32_t *pdata = work->data;
	uint32_t *ptarget = work->target;
	const uint32_t first_nonce = pdata[19];
	if (opt_benchmark)
		ptarget[7] = 0x000f;

	if (!init[thr_id])
	{
		int dev_id = device_map[thr_id];
		cudaDeviceProp props;
		cudaGetDeviceProperties(&props, dev_id);

		int intensity = 0;
		// Pascal
		if (strstr(props.name, "1080")) intensity = 22;
		else if (strstr(props.name, "1070")) intensity = 21;
		// Maxwell
		else if (strstr(props.name, "TITAN X")) intensity = 21;
		else if (strstr(props.name, "980")) intensity = 21;
		else if (strstr(props.name, "970")) intensity = 20;
		else if (strstr(props.name, "960")) intensity = 20;
		else if (strstr(props.name, "950")) intensity = 19;
		else if (strstr(props.name, "750 Ti")) intensity = 19;
		else if (strstr(props.name, "750")) intensity = 18;
		// Kepler〜Fermi
		else if (strstr(props.name, "TITAN Z")) intensity = 20;
		else if (strstr(props.name, "TITAN")) intensity = 19;
		else if (strstr(props.name, "780")) intensity = 19;
		else if (strstr(props.name, "760")) intensity = 18;
		else if (strstr(props.name, "730")) intensity = 16;
		else if (strstr(props.name, "720")) intensity = 15;
		else if (strstr(props.name, "710")) intensity = 16;
		else if (strstr(props.name, "690")) intensity = 20;
		else if (strstr(props.name, "680")) intensity = 19;
		else if (strstr(props.name, "660")) intensity = 18;
		else if (strstr(props.name, "650 Ti")) intensity = 18;
		else if (strstr(props.name, "640")) intensity = 17;
		else if (strstr(props.name, "630")) intensity = 16;
		else if (strstr(props.name, "620")) intensity = 15;

		else if (strstr(props.name, "90")) intensity = 18;	//590
		else if (strstr(props.name, "80")) intensity = 18;	//480 580
		else if (strstr(props.name, "70")) intensity = 18;	//470 570 670 770
		else if (strstr(props.name, "65")) intensity = 17;	//465
		else if (strstr(props.name, "60")) intensity = 17;	//460 560
		else if (strstr(props.name, "55")) intensity = 17;	//555
		else if (strstr(props.name, "50")) intensity = 17;	//450 550Ti 650
		else if (strstr(props.name, "45")) intensity = 16;	//545
		else if (strstr(props.name, "40")) intensity = 15;	//440
		else if (strstr(props.name, "30")) intensity = 15;	//430 530
		else if (strstr(props.name, "20")) intensity = 14;	//420 520
		else if (strstr(props.name, "10")) intensity = 14;	//510 610

		if (intensity != 0 && opt_eco_mode) intensity -= 3.0;

		if (intensity == 0)
		{
			intensity = (device_sm[dev_id] > 500 && !is_windows()) ? 20 : 18;
			throughput[thr_id] = cuda_default_throughput(dev_id, 1UL << (int)intensity);
		}
		else
		{
			//uint32_t adds = 0;
			//	double d = floor(intensity);

			/*	if ((intensity - d) > 0.0) {
					adds = (uint32_t)floor((intensity - d) * (1 << (int)(d - 10.0)) * 1024;
					throughput = (1 << (int)d) + adds;
					gpulog(LOG_INFO, thr_id, "Adding %u threads to intensity %u, %u cuda threads",
					adds, (int)d, throughput);
					}
					else if (gpus_intensity[n] != (1 << (int)intensity)) {
					throughput = (1 << (int)intensity);
					applog(LOG_INFO, "Intensity set to %u, %u cuda threads",
					v, gpus_intensity[n]);
					}
					*/
			uint32_t temp = 1UL << intensity;
			throughput[thr_id] = cuda_default_throughput(dev_id, temp);

			if (temp == throughput[thr_id])
			{
				gpulog(LOG_INFO, thr_id, "Intensity set to %u, %u cuda threads",
					intensity, throughput[thr_id]);
			}
		}
		cudaSetDevice(dev_id);
		if (opt_cudaschedule == -1 && gpu_threads == 1) {
			cudaDeviceReset();
			// reduce cpu usage
			cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);
			CUDA_LOG_ERROR();
		}
		cudaDeviceSetCacheConfig(cudaFuncCachePreferL1);

		//blake256_cpu_init(thr_id, throughput);
		//keccak256_cpu_init(thr_id,throughput);
		skein256_cpu_init(thr_id, throughput[thr_id]);
		bmw256_cpu_init(thr_id, throughput[thr_id]);

		// SM 3 implentation requires a bit more memory
		//if (device_sm[dev_id] < 300 || cuda_arch[dev_id] < 300)
		//	matrix_sz = 16 * sizeof(uint64_t) * 4 * 4;
		//else
		size_t matrix_sz = sizeof(uint64_t) * 4 * 4;
		CUDA_SAFE_CALL(cudaMalloc(&d_matrix[thr_id], matrix_sz * throughput[thr_id]));
		lyra2v2_cpu_init(thr_id, throughput[thr_id], d_matrix[thr_id]);

		CUDA_SAFE_CALL(cudaMalloc(&d_hash[thr_id], (size_t)32 * throughput[thr_id]));

		api_set_throughput(thr_id, throughput[thr_id]);
		init[thr_id] = true;
	}
	else throughput[thr_id] = min(throughput[thr_id], max_nonce - first_nonce);

	uint32_t endiandata[20];
	for (int k = 0; k < 20; k++)
		be32enc(&endiandata[k], pdata[k]);

	blake256_cpu_setBlock_80(pdata);
	//bmw256_setTarget(ptarget);

	//uint32_t *vhash64[2];
	do {
		int order = 0;
		uint32_t foundNonces[2] = { 0, 0 };

		blakeKeccak256_cpu_hash_80(thr_id, throughput[thr_id], pdata[19], d_hash[thr_id], order++);
		//blakeKeccakcube256_cpu_hash_80(thr_id, throughput, pdata[19], d_hash[thr_id], order++);
		TRACE("blake  :");
		//keccak256_cpu_hash_32(thr_id, throughput, pdata[19], d_hash[thr_id], order++);
		TRACE("keccak :");
		cubehash256_cpu_hash_32(thr_id, throughput[thr_id], pdata[19], d_hash[thr_id], order++);
		TRACE("cube   :");
		lyra2v2_cpu_hash_32(thr_id, throughput[thr_id], pdata[19], d_hash[thr_id], order++);
		TRACE("lyra2  :");
		skein256_cpu_hash_32(thr_id, throughput[thr_id], pdata[19], d_hash[thr_id], order++);
		TRACE("skein  :");
		cubehash256_cpu_hash_32(thr_id, throughput[thr_id], pdata[19], d_hash[thr_id], order++);
		TRACE("cube   :");

		bmw256_cpu_hash_32(thr_id, throughput[thr_id], pdata[19], d_hash[thr_id], foundNonces, ptarget[7]);
		//bmw256_cpu_hash_32(thr_id, throughput, pdata[19], d_hash[thr_id], foundNonces, ptarget[7], vhash64);

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


		/*if (foundNonces[1] != 0)
		{
			if (fulltest(vhash64[0], ptarget))
			{
				gpulog(LOG_WARNING, thr_id, "result two foundNonces!");
				pdata[19] = foundNonces[1];
				pdata[21] = foundNonces[0];
				work_set_target_ratio(work, vhash64[0]);
				if (bn_hash_target_ratio(vhash64[1], ptarget) > work->shareratio) {
					work_set_target_ratio(work, vhash64[1]);
				}
				return 2;
			}
		}
		if (foundNonces[0] != 0)
		{
			if (fulltest(vhash64[0], ptarget))
			{
				gpulog(LOG_WARNING, thr_id, "result one foundNonce!");
				pdata[19] = foundNonces[0];
				work_set_target_ratio(work, vhash64[0]);
				return 1;
			}
		}*/

		if (foundNonces[0] != 0)
		{
			uint32_t vhash64[8];
			be32enc(&endiandata[19], foundNonces[0]);
			lyra2v2_hash(vhash64, endiandata);
			if (vhash64[7] <= ptarget[7] && fulltest(vhash64, ptarget))
			{
				int res = 1;
				work_set_target_ratio(work, vhash64);
				pdata[19] = foundNonces[0];
				// check if there was another one...
				if (foundNonces[1] != 0)
				{
					be32enc(&endiandata[19], foundNonces[1]);
					lyra2v2_hash(vhash64, endiandata);
					pdata[21] = foundNonces[1];
					xchg(pdata[19], pdata[21]);
					if (bn_hash_target_ratio(vhash64, ptarget) > work->shareratio) {
						work_set_target_ratio(work, vhash64);
					}
					res++;
				}
				return res;
			}
			else if (vhash64[7] > ptarget[7])
			{
				gpulog(LOG_WARNING, thr_id, "result for %08x does not validate on CPU!", foundNonces[0]);
			}
		}

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

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

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

// cleanup
extern "C" void free_lyra2v2(int thr_id)
{
	if (!init[thr_id])
		return;

	cudaThreadSynchronize();

	cudaFree(d_hash[thr_id]);
	cudaFree(d_matrix[thr_id]);

	bmw256_cpu_free(thr_id);
	//keccak256_cpu_free(thr_id);

	init[thr_id] = false;

	cudaDeviceSynchronize();
}