package mior.model;
  
  import java.util.Arrays;
  
  import mcmas.core.MCMCommandQueue;
  import mcmas.core.MCMCommandQueueProperty;
  import mcmas.core.MCMContext;
  import mcmas.core.MCMEvent;
  import mcmas.core.MCMKernel;
  import mcmas.core.MCMMem;
  import mcmas.core.MCMProgram;
  import mcmas.core.MCMUtils;
  
  import org.jocl.Pointer;
  import org.perf4j.slf4j.Slf4JStopWatch;
  import org.slf4j.Logger;
  import org.slf4j.LoggerFactory;
  
  
  public class OCLParaCSRModel2 extends AbstractMiorModel {
  	
  	/**
  	 * Java view
  	 */
  	
  	private final int nbMM;
  	private final int nbOM;
  	
  	private final MiorMM [] mmList;
  	private final MiorOM [] omList;
  	private final MiorWorld [] worlds;
  	private final int [] associations;
  	
  	/**
  	 * OpenCL implementation
  	 */
  	final MCMContext context;
  	final MCMCommandQueue queue;
  	final MCMProgram program;
  	
  	final MCMKernel topoKernel;
  	final MCMKernel autoliveKernel;
  	
  	final MCMMem mmMem;
  	final MCMMem omMem;
  	final MCMMem worldsMem;
  	
  	/*final OCLMem mmCSRMem;
  	final OCLMem omCSRMem;*/
  	final MCMMem associationsMem;
  	
  	final MCMMem partsMem;
  	
  	/*
  	private final int [] mmCSR;
  	private final int [] omCSR;
  	*/
  	
  	private final int nbSim;
  	private int blockSize;
  	
  	private final static Logger logger = LoggerFactory.getLogger(OCLParaCSRModel2.class);
  	private final Slf4JStopWatch watch = new Slf4JStopWatch(logger);
  	
  	private final static String MODEL_SOURCE = "kernels/mior_model_multisim2.cl";
  	
  	public OCLParaCSRModel2(int nbMM, int nbOM, int nbSim) {
  		this.nbSim = nbSim;
  		this.nbMM = nbMM;
  		this.nbOM = nbOM;
  		this.blockSize = Math.max(nbOM, nbMM);
  		
  		// Allocate Java MIOR structures
  		this.mmList = new MiorMM[nbSim * nbMM];
  		this.omList = new MiorOM[nbSim * nbOM];
  		this.worlds = new MiorWorld[nbSim];
  		this.associations = new int[nbSim * nbMM * nbOM];
  		//final Random rand = new Random(); TODO: Remove
  		
  		// Initialise models
  		for (int i = 0; i < nbSim; i++) {
  			this.worlds[i] = new MiorWorld(nbMM, nbOM);
  		}
  		
  		MiorUtils.initRandomMMArray(mmList, worlds[0].width);
  		MiorUtils.initRandomOMArray(omList, worlds[0].width);
  		Arrays.fill(associations, -1);
  		
  		// OpenCL allocations
  		watch.start("OpenCL setup");
  		
  		this.context = new MCMContext();
  		this.queue = context.createCommandQueue(MCMCommandQueueProperty.ENABLE_PROFILING);
  		
  		watch.stop();
  		
  		watch.start("Program compilation");
  		this.program = MCMUtils.compileFile(context, MODEL_SOURCE, " -DNB_MM=" + nbMM + " -DNB_OM=" + nbOM);// + " -g -O0");
  		watch.stop();
  		
  		watch.start("Kernel allocation");
  		
  		this.topoKernel = program.createKernel("topology");
  		this.autoliveKernel = program.createKernel("autolive");
  		
  		watch.stop();
  		watch.start("Memory allocation");
  		
  		this.mmMem = context.newBuffer().Using(mmList).b();
  		this.omMem = context.newBuffer().Using(omList).b();
  		this.worldsMem = context.newBuffer().Using(worlds).b();
  		
  		/*this.mmCSR = allocateStorage(nbOM * nbSim, nbMM);
  		this.mmCSRMem = context.newBuffer().Using(mmCSR).b();
  		
  		this.omCSR = allocateStorage(nbMM * nbSim, nbOM);
  		this.omCSRMem = context.newBuffer().Using(omCSR).b();
  		*/
  		
  		this.associationsMem = context.newBuffer().Using(associations).b();
  		
  		this.partsMem = context.newBuffer().Using(new int[associations.length]).b();
  		watch.stop();
  	}
  	
  	@Override
  	public int getNbSimulations() {
  		return nbSim;
  	}
  	
  	@Override
  	protected void resetImpl() {
  		// Initialize models
  		for (int i = 0; i < nbSim; i++) {
  			this.worlds[i] = new MiorWorld(nbMM, nbOM);
  		}
  		
  		MiorUtils.initRandomMMArray(mmList, worlds[0].width);
  		MiorUtils.initRandomOMArray(omList, worlds[0].width);
  		Arrays.fill(associations, -1);
  		
  		queue.enqueueWriteBuffer(mmMem, mmList, 0, mmMem.getSize());
  		queue.enqueueWriteBuffer(omMem, omList, 0, omMem.getSize());
  		queue.enqueueWriteBuffer(worldsMem, worlds, 0, worldsMem.getSize());
  		//queue.enqueueWriteBuffer(mmCSRMem, Pointer.to(mmCSR), 0, mmCSRMem.getSize());
  		//queue.enqueueWriteBuffer(omCSRMem, Pointer.to(omCSR), 0, omCSRMem.getSize());
  		queue.enqueueWriteBuffer(associationsMem, Pointer.to(associations), 0, associationsMem.getSize());
  		queue.finish();
  	}
  	
  	@Override
  	public void setBlockSize(int blockSize) {
  		this.blockSize = blockSize;
  	}
  	
  	
  	@Override
  	public int getBlockSize() {
  		return blockSize;
  	}
  	
  	@Override
  	protected void doTopologyImpl() {
  		// System.out.println(topoKernel.getArgumentsNumber());
  		topoKernel.setArguments(mmMem, omMem, worldsMem, associationsMem);
  		
  		MCMEvent event = queue.enqueueKernel(topoKernel, 2,
  				new long[] { nbSim * nbMM , nbOM}
  		);
  		
  		MCMEvent.waitFor(event);
  		MCMUtils.printEventStats("topology", event);
  		
  		if (! isBatchModeEnabled()) {
  			//queue.blockingReadBuffer(mmCSRMem, Pointer.to(mmCSR), 0, mmCSRMem.getSize());
  			//queue.blockingReadBuffer(omCSRMem, Pointer.to(omCSR), 0, omCSRMem.getSize());
  			queue.blockingReadBuffer(associationsMem, Pointer.to(associations), 0, associationsMem.getSize());
  		}
  	}
  	
  	@Override
  	protected void doLiveImpl() {
  		throw new UnsupportedOperationException();
  	}
  	
  	@Override
  	public void doAutoLive() {
  		resetImpl();
  		doTopologyImpl();
  		
  		autoliveKernel.setArguments(mmMem, omMem, worldsMem, associationsMem, partsMem);
  		
  		/*
  		OCLEvent event = queue.enqueueKernel(autoliveKernel, 2,
  				new long[] { nbSim, Math.max(nbOM, nbMM) }
  		);*/
  		
  		if (blockSize < Math.max(nbOM, nbMM)) {
  			throw new RuntimeException("blockSize (" + blockSize + ") is too small to execute the simulation");
  		}
  		
  		//System.out.println("nbAgents: " + nbAgents);
  		MCMEvent event = queue.enqueue1DKernel(autoliveKernel, nbSim * blockSize, blockSize);
  		
  		MCMEvent.waitFor(event);
  		MCMUtils.printEventStats("autolive", event);
  		
  		onSimulationFinished();
  	}
  	
  	@Override
  	protected void onSimulationFinished() {
  		//if (isBatchModeEnabled()) {
  			queue.blockingReadBuffer(mmMem, mmList, 0, mmMem.getSize());
  			queue.blockingReadBuffer(omMem, omList, 0, omMem.getSize());
  			queue.blockingReadBuffer(worldsMem, worlds, 0, worldsMem.getSize());
  			queue.blockingReadBuffer(associationsMem, Pointer.to(associations), 0, associationsMem.getSize());
  		//}
  		
  		super.onSimulationFinished();
  	}
  	
  	@Override
  	protected void releaseImpl() {
  		partsMem.release();
  		
  		associationsMem.release();
  		//omCSRMem.release();
  		//mmCSRMem.release();
  		
  		worldsMem.release();
  		omMem.release();
  		mmMem.release();
  		
  		autoliveKernel.release();
  		topoKernel.release();
  		
  		program.release();
  		queue.release();
  		context.release();
  	}
  
  	@Override
  	public MiorWorld getWorld() {
  		return worlds[0];
  	}
  
  	@Override
  	public MiorOM[] getOMList() {
  		return Arrays.copyOfRange(omList, 0, nbOM);
  	}
  
  	@Override
  	public MiorMM[] getMMList() {
  		return Arrays.copyOfRange(mmList, 0, nbMM);
  	}
  
  	@Override
  	public boolean isAccessible(int iMM, int iOM) {
  		return associations[iMM * nbOM + iOM] != -1;
  	}
  
  	@Override
  	public boolean isAccessible(int iMM, int iOM, int iSim) {
  		return associations[iSim * nbMM * nbOM + iMM * nbOM + iOM] != -1;
  	}
  	
  }