nvprof  是一个可执行文件，使用everything搜索可以看到它在：

执行命令：

nvprof exe_name

如何在windows 下使用，可以参看: windows下使用nvcc和nvprof。

示例

#include <cuda_runtime.h>
#include <device_launch_parameters.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <iostream>

static void CheckCudaErrorAux(const char*, unsigned, const char*, cudaError_t);
#define CUDA_CHECK_RETURN(value) CheckCudaErrorAux(__FILE__,__LINE__, #value, value)

// 设备函数
__device__ float add(const float x, const float y)
{
	return x + y;
}

__global__ void addFromGPU(float* A, float* B, float* C, const int N)
{
	int blockId = blockIdx.x;
	int id = blockId * blockDim.x + threadIdx.x;
	if (id >= N)
	{
		return;
	}

	C[id] = add(A[id], B[id]);
}

void initialData(float* addr, int nCount)
{
	for (size_t i = 0; i < nCount; i++)
	{
		addr[i] = (float)(rand() & 0xFFF) / 100.f;
	}
}

int main()
{
	int iElemntCount = 4096 * 10;
	size_t stBytesCount = iElemntCount * sizeof(float); // 字节数

	// 分配主机内存和设备内存并初始化
	float* fpHost_A = new float[iElemntCount];
	float* fpHost_B = new float[iElemntCount];
	float* fpHost_C = new float[iElemntCount];
	memset(fpHost_A, 0, stBytesCount);
	memset(fpHost_B, 0, stBytesCount);
	memset(fpHost_C, 0, stBytesCount);
	float* fpDevice_A, * fpDevice_B, * fpDevice_C;
	CUDA_CHECK_RETURN(cudaMalloc((void**)&fpDevice_A, stBytesCount));
	CUDA_CHECK_RETURN(cudaMalloc((void**)&fpDevice_B, stBytesCount));
	CUDA_CHECK_RETURN(cudaMalloc((void**)&fpDevice_C, stBytesCount));

	CUDA_CHECK_RETURN(cudaMemset(fpDevice_C, 0, stBytesCount));

	srand(666);
	initialData(fpHost_A, iElemntCount);
	initialData(fpHost_B, iElemntCount);

	CUDA_CHECK_RETURN(cudaMemcpy(fpDevice_A, fpHost_A, stBytesCount, cudaMemcpyHostToDevice));
	CUDA_CHECK_RETURN(cudaMemcpy(fpDevice_B, fpHost_B, stBytesCount, cudaMemcpyHostToDevice));

	dim3 block(32);
	dim3 grid((iElemntCount + block.x - 1) / block.x);

	addFromGPU <<<grid, block >>> (fpDevice_A, fpDevice_B, fpDevice_C, iElemntCount);

	cudaFree(fpDevice_A);
	cudaFree(fpDevice_B);
	cudaFree(fpDevice_C);

	delete[]fpHost_A;
	delete[]fpHost_B;
	delete[]fpHost_C;

	fpHost_A = nullptr;
	fpHost_B = nullptr;
	fpHost_C = nullptr;
	printf("***********finish**************\n");

	return 0;
}

static void CheckCudaErrorAux(const char* file, unsigned line, const char* statement, cudaError_t err)
{
	if (err == cudaSuccess)
		return;
	std::cerr << statement << " returned: " << cudaGetErrorName(err) << "  \t : " << cudaGetErrorString(err) << "(" << err << ") at " << file << ":" << line << std::endl;
	exit(1);
}

分析结果：

这个就包含 命令耗时、调用次数、平均用时、最小用时、最大用时、命令名。