cuda编程入门示例16

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <cuda.h>
#include <cuda_runtime.h>

#define BLOCK_SIZE 16
static void HandleError(cudaError_t err, const char *file, int line)
{
	if (err != cudaSuccess)
	{
		printf("%s in %s at line %d\n", cudaGetErrorString(err), file, line);
		exit(EXIT_FAILURE);
	}
}
#define HANDLE_ERROR( err ) (HandleError( err, __FILE__, __LINE__ ))

#define HANDLE_NULL( a ) {if ((a) == NULL) { \
	printf("Host memory failed in %s at line %d\n", \
	__FILE__, __LINE__); \
	exit(EXIT_FAILURE); }}

static void GenerateNumbers(int *number, int size)
{
	for (int i = 0; i < size; i++)
	{
		number[i] = rand() % 10;
	}
}

static bool InitCUDA()
{
	int count;

	cudaGetDeviceCount(&count);
	if (count == 0)
	{
		fprintf(stderr, "There is no device.\n");
		return false;
	}

	int i;
	for (i = 0; i < count; i++)
	{
		cudaDeviceProp prop;
		if (cudaGetDeviceProperties(&prop, i) == cudaSuccess)
		{
			if (prop.major >= 1)
			{
				break;
			}
		}
	}

	if (i >= count)
	{
		fprintf(stderr, "There is no device supporting CUDA 1.x.\n");
		return false;
	}

	cudaSetDevice(i);

	return true;
}

//多个block， 每个block内256个线程，每一个块内的线程得到一个累加结果，每一个块负责将
//块内所有线程的累加结果累加起来，最后每一个块的threadIdx.x = 0将该块得到的结果进行汇总。
//块内用共享内存。global memory 原子操作
__global__ static void sumOfSquares(int *num, int size, long long* result, clock_t* time)
{
	extern __shared__ int temp[];
	const int tid = threadIdx.x;
	const int bid = blockIdx.x;
	int i;

	temp[tid] = 0;
	if (tid == 0) time[bid] = clock();
	for (i = tid + bid * blockDim.x; i < size; i += gridDim.x * blockDim.x)
	{
		temp[tid] += num[i] * num[i];
	}
	__syncthreads();

	int offset = blockDim.x >> 1;
	while (offset > 0)
	{
		if (tid < offset)
		{
			temp[tid] += temp[tid + offset];
		}
		__syncthreads();
		offset >>= 1;
	}

	if (tid == 0)
	{
		//result[bid] = temp[0];
		atomicAdd(result, temp[0]);
		time[bid + gridDim.x] = clock();
	}
}


int main(int argc, char *argv[])
{
	if (!InitCUDA())
	{
		return -1;
	}
	printf("CUDA initialized.\n");

	//const int block_num = 32;
	const int block_num = 16;
	const int  thread_num = 256;
	const int DATA_SIZE = 1024 * 16 * 8;
	int data[DATA_SIZE];
	GenerateNumbers(data, DATA_SIZE);

	int* gpudata;
	long long *result;
	clock_t* devTime, gpuTime[2 * block_num], cpuTime;
	HANDLE_ERROR(cudaMalloc((void **)&gpudata, sizeof(int)* DATA_SIZE));
	HANDLE_ERROR(cudaMalloc((void**)&result, sizeof(*result)));
	HANDLE_ERROR(cudaMalloc((void**)&devTime, sizeof(clock_t)* 2 * block_num));
	HANDLE_ERROR(cudaMemcpy(gpudata, data, sizeof(int)* DATA_SIZE, cudaMemcpyHostToDevice));
	HANDLE_ERROR(cudaMemset(result, 0, sizeof(*result)));

	sumOfSquares << <block_num, thread_num, thread_num * sizeof(int) >> >(gpudata, DATA_SIZE, result, devTime);

	long long sumCPU = 0, gpuSum;
	HANDLE_ERROR(cudaMemcpy(&gpuSum, result, sizeof(int), cudaMemcpyDeviceToHost));
	HANDLE_ERROR(cudaMemcpy(gpuTime, devTime, sizeof(clock_t)* 2 * block_num, cudaMemcpyDeviceToHost));

	cudaFree(gpudata);
	cudaFree(result);
	cudaFree(devTime);

	clock_t minStart = gpuTime[0];
	clock_t maxEnd = gpuTime[block_num];
	for (int i = 1; i < block_num; i++)
	{
		if (minStart > gpuTime[i])
		{
			minStart = gpuTime[i];
		}
		if (maxEnd < gpuTime[i + block_num])
		{
			maxEnd = gpuTime[i + block_num];
		}
	}
	printf("sum (GPU): %ld, time: %d\n", gpuSum, maxEnd - minStart);

	sumCPU = 0;
	cpuTime = clock();
	for (int i = 0; i < DATA_SIZE; i++)
	{
		sumCPU += data[i] * data[i];
	}
	cpuTime = clock() - cpuTime;
	printf("sum (CPU): %ld, time: %d\n", sumCPU, cpuTime);
	printf("Result %s\n", gpuSum == sumCPU ? "OK" : "Wrong");

	//remember to release the device
	cudaDeviceReset();

	return 0;
}