/*
  * This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
  * the Free Software Foundation, either version 3 of the License, or
  * any later version.
  *
  * This program is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  * GNU General Public License for more details.
  *
  * You should have received a copy of the GNU General Public License
  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  *
  * Additional permission under GNU GPL version 3 section 7
  *
  * If you modify this Program, or any covered work, by linking or combining
  * it with OpenSSL (or a modified version of that library), containing parts
  * covered by the terms of OpenSSL License and SSLeay License, the licensors
  * of this Program grant you additional permission to convey the resulting work.
  *
  */

#include <assert.h>
#include <cmath>
#include <chrono>
#include <thread>
#include "console.h"

#ifdef _WIN32
#include <windows.h>

void thd_setaffinity(std::thread::native_handle_type h, uint64_t cpu_id)
{
	SetThreadAffinityMask(h, 1ULL << cpu_id);
}

#else
#include <pthread.h>

#if defined(__APPLE__)
#include <mach/thread_policy.h>
#include <mach/thread_act.h>
#define SYSCTL_CORE_COUNT   "machdep.cpu.core_count"
#endif

void thd_setaffinity(std::thread::native_handle_type h, uint64_t cpu_id)
{
#if defined(__APPLE__)
	thread_port_t mach_thread;
	thread_affinity_policy_data_t policy = { cpu_id };
	mach_thread = pthread_mach_thread_np(h);
	thread_policy_set(mach_thread, THREAD_AFFINITY_POLICY, (thread_policy_t)&policy, 1);
#else
	cpu_set_t mn;
	CPU_ZERO(&mn);
	CPU_SET(cpu_id, &mn);
	pthread_setaffinity_np(h, sizeof(cpu_set_t), &mn);
#endif
}
#endif // _WIN32

#include "executor.h"
#include "minethd.h"
#include "jconf.h"
#include "crypto/cryptonight.h"

telemetry::telemetry(size_t iThd)
{
	ppHashCounts = new uint64_t*[iThd];
	ppTimestamps = new uint64_t*[iThd];
	iBucketTop = new uint32_t[iThd];

	for (size_t i = 0; i < iThd; i++)
	{
		ppHashCounts[i] = new uint64_t[iBucketSize];
		ppTimestamps[i] = new uint64_t[iBucketSize];
		iBucketTop[i] = 0;
		memset(ppHashCounts[0], 0, sizeof(uint64_t) * iBucketSize);
		memset(ppTimestamps[0], 0, sizeof(uint64_t) * iBucketSize);
	}
}

double telemetry::calc_telemetry_data(size_t iLastMilisec, size_t iThread)
{
	using namespace std::chrono;
	uint64_t iTimeNow = time_point_cast<milliseconds>(high_resolution_clock::now()).time_since_epoch().count();

	uint64_t iEarliestHashCnt = 0;
	uint64_t iEarliestStamp = 0;
	uint64_t iLastestStamp = 0;
	uint64_t iLastestHashCnt = 0;
	bool bHaveFullSet = false;

	//Start at 1, buckettop points to next empty
	for (size_t i = 1; i < iBucketSize; i++)
	{
		size_t idx = (iBucketTop[iThread] - i) & iBucketMask; //overflow expected here

		if (ppTimestamps[iThread][idx] == 0)
			break; //That means we don't have the data yet

		if (iLastestStamp == 0)
		{
			iLastestStamp = ppTimestamps[iThread][idx];
			iLastestHashCnt = ppHashCounts[iThread][idx];
		}

		if (iTimeNow - ppTimestamps[iThread][idx] > iLastMilisec)
		{
			bHaveFullSet = true;
			break; //We are out of the requested time period
		}

		iEarliestStamp = ppTimestamps[iThread][idx];
		iEarliestHashCnt = ppHashCounts[iThread][idx];
	}

	if (!bHaveFullSet || iEarliestStamp == 0 || iLastestStamp == 0)
		return nan("");

	double fHashes, fTime;
	fHashes = iLastestHashCnt - iEarliestHashCnt;
	fTime = iLastestStamp - iEarliestStamp;
	fTime /= 1000.0;

	return fHashes / fTime;
}

void telemetry::push_perf_value(size_t iThd, uint64_t iHashCount, uint64_t iTimestamp)
{
	size_t iTop = iBucketTop[iThd];
	ppHashCounts[iThd][iTop] = iHashCount;
	ppTimestamps[iThd][iTop] = iTimestamp;

	iBucketTop[iThd] = (iTop + 1) & iBucketMask;
}

minethd::minethd(miner_work& pWork, size_t iNo, GpuContext* ctx)
{
	oWork = pWork;
	bQuit = 0;
	iThreadNo = (uint8_t)iNo;
	iJobNo = 0;
	iHashCount = 0;
	iTimestamp = 0;
	pGpuCtx = ctx;

	oWorkThd = std::thread(&minethd::work_main, this);
}

bool minethd::init_gpus()
{
	size_t i, n = jconf::inst()->GetThreadCount();

	printer::inst()->print_msg(L1, "Compiling code and initializing GPUs. This will take a while...");
	vGpuData.resize(n);

	jconf::thd_cfg cfg;
	for(i = 0; i < n; i++)
	{
		jconf::inst()->GetThreadConfig(i, cfg);
		vGpuData[i].deviceIdx = cfg.index;
		vGpuData[i].rawIntensity = cfg.intensity;
		vGpuData[i].workSize = cfg.w_size;
	}

	printer::inst()->print_msg(L0, "Test GPU 0");
	
	return InitOpenCL(vGpuData.data(), n, jconf::inst()->GetPlatformIdx()) == ERR_SUCCESS;
}

std::atomic<uint64_t> minethd::iGlobalJobNo;
std::atomic<uint64_t> minethd::iConsumeCnt; //Threads get jobs as they are initialized
minethd::miner_work minethd::oGlobalWork;
uint64_t minethd::iThreadCount = 0;
std::vector<GpuContext> minethd::vGpuData;

std::vector<minethd*>* minethd::thread_starter(miner_work& pWork)
{
	iGlobalJobNo = 0;
	iConsumeCnt = 0;
	std::vector<minethd*>* pvThreads = new std::vector<minethd*>;

	size_t i, n = jconf::inst()->GetThreadCount();
	pvThreads->reserve(n);

	jconf::thd_cfg cfg;
	for (i = 0; i < n; i++)
	{
		jconf::inst()->GetThreadConfig(i, cfg);
		minethd* thd = new minethd(pWork, i, &vGpuData[i]);

		if(cfg.cpu_aff >= 0)
		{
#if defined(__APPLE__)
			printer::inst()->print_msg(L1, "WARNING on MacOS thread affinity is only advisory.");
#endif
			thd_setaffinity(thd->oWorkThd.native_handle(), cfg.cpu_aff);
		}

		pvThreads->push_back(thd);
		if(cfg.cpu_aff >= 0)
			printer::inst()->print_msg(L1, "Starting GPU thread, affinity: %d.", (int)cfg.cpu_aff);
		else
			printer::inst()->print_msg(L1, "Starting GPU thread, no affinity.");
	}

	iThreadCount = n;
	return pvThreads;
}

void minethd::switch_work(miner_work& pWork)
{
	// iConsumeCnt is a basic lock-like polling mechanism just in case we happen to push work
	// faster than threads can consume them. This should never happen in real life.
	// Pool cant physically send jobs faster than every 250ms or so due to net latency.

	while (iConsumeCnt.load(std::memory_order_seq_cst) < iThreadCount)
		std::this_thread::sleep_for(std::chrono::milliseconds(100));

	oGlobalWork = pWork;
	iConsumeCnt.store(0, std::memory_order_seq_cst);
	iGlobalJobNo++;
}

void minethd::consume_work()
{
	memcpy(&oWork, &oGlobalWork, sizeof(miner_work));
	iJobNo++;
	iConsumeCnt++;

	if(!oWork.bStall)
	{
		pGpuCtx->Nonce = calc_start_nonce(oWork.iResumeCnt);
		XMRSetJob(pGpuCtx, oWork.bWorkBlob, oWork.iWorkSize, oWork.iTarget);
	}
}

void minethd::work_main()
{
	uint64_t iCount = 0;
	iConsumeCnt++;
	while (bQuit == 0)
	{
		if (oWork.bStall)
		{
			/*  We are stalled here because the executor didn't find a job for us yet,
			    either because of network latency, or a socket problem. Since we are
			    raison d'etre of this software it us sensible to just wait until we have something*/

			while (iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
				std::this_thread::sleep_for(std::chrono::milliseconds(100));

			consume_work();
			continue;
		}

		assert(sizeof(job_result::sJobID) == sizeof(pool_job::sJobID));

		while(iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
		{
			cl_uint results[0x100] = { 0 };

			XMRRunJob(pGpuCtx, results);

			for(size_t i = 0; i < results[0xFF]; i++)
			{
				executor::inst()->push_event(ex_event(job_result(oWork.sJobID, oWork.bWorkBlob,
					oWork.iWorkSize, oWork.iTarget, results[i]), oWork.iPoolId));
			}

			iCount += pGpuCtx->rawIntensity;
			using namespace std::chrono;
			uint64_t iStamp = time_point_cast<milliseconds>(high_resolution_clock::now()).time_since_epoch().count();
			iHashCount.store(iCount, std::memory_order_relaxed);
			iTimestamp.store(iStamp, std::memory_order_relaxed);
			std::this_thread::yield();
		}

		consume_work();
	}
}