当前位置: 首页>>代码示例>>Java>>正文

Java JCuda.cudaMalloc方法代码示例

本文整理汇总了Java中jcuda.runtime.JCuda.cudaMalloc方法的典型用法代码示例。如果您正苦于以下问题：Java JCuda.cudaMalloc方法的具体用法？Java JCuda.cudaMalloc怎么用？Java JCuda.cudaMalloc使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类jcuda.runtime.JCuda的用法示例。


在下文中一共展示了JCuda.cudaMalloc方法的13个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Java代码示例。

示例1: main2

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
/**
 * @param args
 */
public static void main2(String[] args) {
	Pointer devPtr = new Pointer();
	JCuda.cudaMalloc(devPtr, 1024 * 1024 * 1024);
	logger.info("Pointer: "+devPtr);
	JCuda.cudaFree(devPtr);
}
开发者ID:naxos-simulator，项目名称:NaxosSimulator，代码行数:10，代码来源:CudaTest.java

示例2: Matrix

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
public Matrix(int rows, int cols) {
	this.dontFree = false;
	this.rows = rows;
	this.cols = cols;
	this.data_d = new Pointer();
	JCuda.cudaMalloc(data_d, rows*cols * Sizeof.FLOAT);
	CublasUtil.allocated.add(this);
}
开发者ID:tberg12，项目名称:murphy，代码行数:9，代码来源:CublasUtil.java

示例3: getrfGetriBatched

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
private static void getrfGetriBatched(List<Matrix> A, List<Matrix> B) {
	Pointer[] Apointers = new Pointer[A.size()];
	Pointer[] Bpointers = new Pointer[B.size()];
	for (int i=0; i<A.size(); ++i) {
		Apointers[i] = A.get(i).data_d;
		Bpointers[i] = B.get(i).data_d;
	}
	Pointer Apointers_d = new Pointer();
	JCuda.cudaMalloc(Apointers_d, A.size() * Sizeof.POINTER);
	JCuda.cudaMemcpy(Apointers_d, Pointer.to(Apointers), A.size() * Sizeof.POINTER, cudaMemcpyKind.cudaMemcpyHostToDevice);
	Pointer Bpointers_d = new Pointer();
	JCuda.cudaMalloc(Bpointers_d, B.size() * Sizeof.POINTER);
	JCuda.cudaMemcpy(Bpointers_d, Pointer.to(Bpointers), B.size() * Sizeof.POINTER, cudaMemcpyKind.cudaMemcpyHostToDevice);
	Pointer info_d = new Pointer();
	JCuda.cudaMalloc(info_d, A.size() * Sizeof.INT);
	Pointer pivots_d = new Pointer();
	JCuda.cudaMalloc(pivots_d, A.get(0).rows * A.size() * Sizeof.INT);
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
	
	JCublas2.cublasSgetrfBatched(cublasHandle, A.get(0).rows, Apointers_d, A.get(0).rows, pivots_d, info_d, A.size());
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
	
	JCublas2.cublasSgetriBatched(cublasHandle, A.get(0).rows, Apointers_d, A.get(0).rows, pivots_d, Bpointers_d, B.get(0).rows, info_d, A.size());
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
	
	JCuda.cudaFree(Apointers_d);
	JCuda.cudaFree(Bpointers_d);
	JCuda.cudaFree(info_d);
	JCuda.cudaFree(pivots_d);
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
}
开发者ID:tberg12，项目名称:murphy，代码行数:32，代码来源:CublasUtil.java

示例4: gemmBatched

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
private static void gemmBatched(float alpha, List<Matrix> A, List<Matrix> B, float beta, List<Matrix> C) {
	Pointer[] Apointers = new Pointer[A.size()];
	Pointer[] Bpointers = new Pointer[B.size()];
	Pointer[] Cpointers = new Pointer[C.size()];
	for (int i=0; i<A.size(); ++i) {
		Apointers[i] = A.get(i).data_d;
		Bpointers[i] = B.get(i).data_d;
		Cpointers[i] = C.get(i).data_d;
	}
	Pointer Apointers_d = new Pointer();
	JCuda.cudaMalloc(Apointers_d, A.size() * Sizeof.POINTER);
	JCuda.cudaMemcpy(Apointers_d, Pointer.to(Apointers), A.size() * Sizeof.POINTER, cudaMemcpyKind.cudaMemcpyHostToDevice);
	Pointer Bpointers_d = new Pointer();
	JCuda.cudaMalloc(Bpointers_d, B.size() * Sizeof.POINTER);
	JCuda.cudaMemcpy(Bpointers_d, Pointer.to(Bpointers), B.size() * Sizeof.POINTER, cudaMemcpyKind.cudaMemcpyHostToDevice);
	Pointer Cpointers_d = new Pointer();
	JCuda.cudaMalloc(Cpointers_d, C.size() * Sizeof.POINTER);
	JCuda.cudaMemcpy(Cpointers_d, Pointer.to(Cpointers), C.size() * Sizeof.POINTER, cudaMemcpyKind.cudaMemcpyHostToDevice);
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
	
	JCublas2.cublasSgemmBatched(cublasHandle, cublasOperation.CUBLAS_OP_N, cublasOperation.CUBLAS_OP_N, C.get(0).rows, C.get(0).cols, B.get(0).rows, Pointer.to(new float[] {alpha}), Apointers_d, A.get(0).rows, Bpointers_d, B.get(0).rows, Pointer.to(new float[] {beta}), Cpointers_d, C.get(0).rows, A.size());
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
	
	JCuda.cudaFree(Apointers_d);
	JCuda.cudaFree(Bpointers_d);
	JCuda.cudaFree(Cpointers_d);
	if (DEBUG_SYNC) JCudaDriver.cuCtxSynchronize();
}
开发者ID:tberg12，项目名称:murphy，代码行数:29，代码来源:CublasUtil.java

示例5: cudaMalloc

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
/**
 * Cuda malloc int.
 *
 * @param devPtr the dev ptr
 * @param size   the size
 * @return the int
 */
public static int cudaMalloc(final Pointer devPtr, final long size) {
  long startTime = System.nanoTime();
  final int result = JCuda.cudaMalloc(devPtr, size);
  CuDNN.log("cudaMalloc", result, devPtr, size);
  cudaMalloc_execution.accept((System.nanoTime() - startTime) / 1e9);
  handle(result);
  return result;
}
开发者ID:SimiaCryptus，项目名称:MindsEye，代码行数:16，代码来源:CuDNN.java

示例6: CUDAVolume3D

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
public CUDAVolume3D(int[] size2, float[] dim2, int inDim) {
	super(size2, dim2, inDim);
	int adaptedWidth = CUDAUtil.iDivUp(size[2], CUDAUtil.gridBlockSize[0]) * CUDAUtil.gridBlockSize[0];
	int adaptedHeight = CUDAUtil.iDivUp(size[1], CUDAUtil.gridBlockSize[1]) * CUDAUtil.gridBlockSize[1];
	int memorySize = adaptedWidth*adaptedHeight*size[0]* getInternalDimension() * Sizeof.FLOAT;
	deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);
	JCuda.cudaMemset(deviceX, 0, memorySize);
}
开发者ID:akmaier，项目名称:CONRAD，代码行数:10，代码来源:CUDAVolume3D.java

示例7: copyToDeviceMemory

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
/**
 * copies an int array to the device and returns a pointer to the memory.
 * @param data the int array
 * @return the pointer to the device memory
 */
public static CUdeviceptr copyToDeviceMemory(int [] data){
	int memorySize = data.length * Sizeof.INT;
	
	CUdeviceptr deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);
	JCuda.cudaMemcpy(deviceX, Pointer.to(data), memorySize, 
			cudaMemcpyKind.cudaMemcpyHostToDevice);
	return deviceX;
}
开发者ID:akmaier，项目名称:CONRAD，代码行数:15，代码来源:CUDAUtil.java

示例8: allocateSpace

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
/**
 * Allocates space on the CUDA device for a Volume3D 
 * @param vol the volume
 * @return the pointer to the memory
 */
public static CUdeviceptr allocateSpace(Volume3D vol){
	// We allocate too much memory as we parallelize along x and y direction and the memory must be a multiple along this direction internally.
	int adaptedWidth = iDivUp(vol.size[2], gridBlockSize[0]) * gridBlockSize[0];
	int adaptedHeight = iDivUp(vol.size[1], gridBlockSize[1]) * gridBlockSize[1];
	int memorySize = adaptedWidth*adaptedHeight*vol.size[0]* vol.getInternalDimension() * Sizeof.FLOAT;
	CUdeviceptr deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);
	return deviceX;
}
开发者ID:akmaier，项目名称:CONRAD，代码行数:15，代码来源:CUDAUtil.java

示例9: testGPUInstallation

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
/**
 * todo needs some more development/expansion
 */
private void testGPUInstallation(){
    try {
        jcuda.Pointer pointer = new jcuda.Pointer();
        JCuda.cudaMalloc(pointer, 4);
        JCuda.cudaFree(pointer);
    }
    catch (Exception e) {
        System.err.println("GPU/CUDA Installation Not Detected");
        System.err.println("Exiting HiCCUPS");
        System.exit(24);
    }
}
开发者ID:theaidenlab，项目名称:Juicebox，代码行数:16，代码来源:HiCCUPS_postproc.java

示例10: makeComplex

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
@Override
public void makeComplex(Volume3D vol)
{

	if (vol.getInternalDimension() == 2) return;
	if (vol.getInternalDimension() != 1) {
		fprintf("vol_make_comlex: Invalid dimension\n");
		return;
	}

	initCUDA();

	int adaptedWidth = CUDAUtil.iDivUp(vol.size[2], CUDAUtil.gridBlockSize[0]) * CUDAUtil.gridBlockSize[0];
	int adaptedHeight = CUDAUtil.iDivUp(vol.size[1], CUDAUtil.gridBlockSize[1]) * CUDAUtil.gridBlockSize[1];
	int memorySize = adaptedWidth*adaptedHeight*vol.size[0]* 2 * Sizeof.FLOAT;
	CUdeviceptr deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);
	JCuda.cudaMemset(deviceX, 0, memorySize);

	CUDAVolume3D cudaVol = (CUDAVolume3D) vol;

	// Calculate new grid size
	gridSize = getGrid(vol.size);

	Pointer sizePointer = CUDAUtil.copyToDeviceMemory(vol.size);

	CUfunction function = new CUfunction();
	JCudaDriver.cuModuleGetFunction(function, module,
			"_Z11makeComplexPfS_Pi");

	ArrayList<Object> arguments = new ArrayList<Object>();
	arguments.add(cudaVol.getDevicePointer());
	arguments.add(deviceX);
	arguments.add(sizePointer);

	callCUDAFunction(function, arguments);

	JCuda.cudaFree(sizePointer);
	JCuda.cudaFree(cudaVol.getDevicePointer());

	cudaVol.setDevicePointer(deviceX);

	float [][][] temp = new float [vol.size[0]][vol.size[1]][vol.size[2]*2];

	vol.data = null;
	vol.data = temp;
	temp = null;
	vol.in_dim = 2;

	return;

}
开发者ID:akmaier，项目名称:CONRAD，代码行数:53，代码来源:CUDAVolumeOperator.java

示例11: real

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
@Override
public int real(Volume3D vol)
{

	if (debug) {
		System.out.print("Called real ");
		printTime();
	}
	if (DEBUG_FLAG)
		fprintf("vol_real\n");

	if (vol.in_dim == 1) return(0);
	if (vol.in_dim != 2) {

		fprintf( "vol_real: Invalid dimension\n");
		return(-1);
	}

	int adaptedWidth = CUDAUtil.iDivUp(vol.size[2], CUDAUtil.gridBlockSize[0]) * CUDAUtil.gridBlockSize[0];
	int adaptedHeight = CUDAUtil.iDivUp(vol.size[1], CUDAUtil.gridBlockSize[1]) * CUDAUtil.gridBlockSize[1];
	int memorySize = adaptedWidth*adaptedHeight*vol.size[0]* 1 * Sizeof.FLOAT;
	CUdeviceptr deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);

	CUDAVolume3D cudaVol = (CUDAVolume3D) vol;

	// Calculate new grid size
	gridSize = getGrid(vol.size);

	Pointer sizePointer = CUDAUtil.copyToDeviceMemory(vol.size);

	CUfunction function = new CUfunction();
	JCudaDriver.cuModuleGetFunction(function, module,
			"_Z4realPfS_Pi");

	ArrayList<Object> arguments = new ArrayList<Object>();
	arguments.add(cudaVol.getDevicePointer());
	arguments.add(deviceX);
	arguments.add(sizePointer);
	if (debug) {
		System.out.print("Called init done ");
		printTime();
	}
	callCUDAFunction(function, arguments);
	if (debug) {
		System.out.print("CUDA done ");
		printTime();
	}

	JCuda.cudaFree(sizePointer);
	JCuda.cudaFree(cudaVol.getDevicePointer());

	cudaVol.setDevicePointer(deviceX);

	float [][][] temp = new float[vol.size[0]][vol.size[1]][vol.size[2]];
	vol.data = null;
	vol.data = temp;
	temp = null;
	vol.in_dim = 1;
	if (debug) {
		System.out.print("Clean up done ");
		printTime();
	}
	return(0);
}
开发者ID:akmaier，项目名称:CONRAD，代码行数:66，代码来源:CUDAVolumeOperator.java

示例12: imag

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
@Override
public int imag(Volume3D vol)
{

	if (debug) {
		System.out.print("Called real ");
		printTime();
	}
	if (DEBUG_FLAG)
		fprintf("vol_real\n");

	if (vol.in_dim == 1) return(0);
	if (vol.in_dim != 2) {

		fprintf( "vol_real: Invalid dimension\n");
		return(-1);
	}

	int adaptedWidth = CUDAUtil.iDivUp(vol.size[2], CUDAUtil.gridBlockSize[0]) * CUDAUtil.gridBlockSize[0];
	int adaptedHeight = CUDAUtil.iDivUp(vol.size[1], CUDAUtil.gridBlockSize[1]) * CUDAUtil.gridBlockSize[1];
	int memorySize = adaptedWidth*adaptedHeight*vol.size[0]* 1 * Sizeof.FLOAT;
	CUdeviceptr deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);

	CUDAVolume3D cudaVol = (CUDAVolume3D) vol;

	// Calculate new grid size
	gridSize = getGrid(vol.size);

	Pointer sizePointer = CUDAUtil.copyToDeviceMemory(vol.size);

	CUfunction function = new CUfunction();
	JCudaDriver.cuModuleGetFunction(function, module,
			"_Z4imagPfS_Pi");

	ArrayList<Object> arguments = new ArrayList<Object>();
	arguments.add(cudaVol.getDevicePointer());
	arguments.add(deviceX);
	arguments.add(sizePointer);
	if (debug) {
		System.out.print("Called init done ");
		printTime();
	}
	callCUDAFunction(function, arguments);
	if (debug) {
		System.out.print("CUDA done ");
		printTime();
	}

	JCuda.cudaFree(sizePointer);
	JCuda.cudaFree(cudaVol.getDevicePointer());

	cudaVol.setDevicePointer(deviceX);

	float [][][] temp = new float[vol.size[0]][vol.size[1]][vol.size[2]];
	vol.data = null;
	vol.data = temp;
	temp = null;
	vol.in_dim = 1;
	if (debug) {
		System.out.print("Clean up done ");
		printTime();
	}
	return(0);
}
开发者ID:akmaier，项目名称:CONRAD，代码行数:66，代码来源:CUDAVolumeOperator.java

示例13: abs

import jcuda.runtime.JCuda; //导入方法依赖的package包/类
@Override
public int abs(Volume3D vol)
{

	if (DEBUG_FLAG)
		fprintf("vol_abs\n");

	if (vol.in_dim != 1 && vol.in_dim != 2) {

		fprintf( "vol_abs: Invalid dimension\n");
		return(-1);
	}

	int adaptedWidth = CUDAUtil.iDivUp(vol.size[2], CUDAUtil.gridBlockSize[0]) * CUDAUtil.gridBlockSize[0];
	int adaptedHeight = CUDAUtil.iDivUp(vol.size[1], CUDAUtil.gridBlockSize[1]) * CUDAUtil.gridBlockSize[1];
	int memorySize = adaptedWidth*adaptedHeight*vol.size[0]* 1 * Sizeof.FLOAT;
	CUdeviceptr deviceX = new CUdeviceptr();
	JCuda.cudaMalloc(deviceX, memorySize);

	initCUDA();	
	
	CUDAVolume3D cudaVol = (CUDAVolume3D) vol;

	// Calculate new grid size
	gridSize = getGrid(vol.size);

	Pointer sizePointer = CUDAUtil.copyToDeviceMemory(vol.size);

	CUfunction function = new CUfunction();
	JCudaDriver.cuModuleGetFunction(function, module,
			"_Z3absPfS_Pii");

	ArrayList<Object> arguments = new ArrayList<Object>();
	arguments.add(cudaVol.getDevicePointer());
	arguments.add(deviceX);
	arguments.add(sizePointer);
	arguments.add(cudaVol.in_dim);
	if (debug) {
		System.out.print("Called init done ");
		printTime();
	}
	callCUDAFunction(function, arguments);
	if (debug) {
		System.out.print("CUDA done ");
		printTime();
	}

	JCuda.cudaFree(sizePointer);
	JCuda.cudaFree(cudaVol.getDevicePointer());

	cudaVol.setDevicePointer(deviceX);

	float [][][] temp = new float[vol.size[0]][vol.size[1]][vol.size[2]];
	vol.data = null;
	vol.data = temp;
	temp = null;
	vol.in_dim = 1;
	if (debug) {
		System.out.print("Clean up done ");
		printTime();
	}



	return(0);
}
开发者ID:akmaier，项目名称:CONRAD，代码行数:67，代码来源:CUDAVolumeOperator.java


注：本文中的jcuda.runtime.JCuda.cudaMalloc方法示例由纯净天空整理自Github/MSDocs等开源代码及文档管理平台，相关代码片段筛选自各路编程大神贡献的开源项目，源码版权归原作者所有，传播和使用请参考对应项目的License；未经允许，请勿转载。