CUDA中block和thread的合理划分配置

CUDA并行编程的基本思路是把一个很大的任务划分成N个简单重复的操作，创建N个线程分别执行执行，每个网格（Grid）可以最多创建65535个线程块，每个线程块（Block）一般最多可以创建512个并行线程，在第一个CUDA程序中对核函数的调用是：

addKernel<<<1, size>>>(dev_c, dev_a, dev_b);

这里的<<<>>>运算符内是核函数的执行参数，告诉编译器运行时如何启动核函数，用于说明内核函数中的线程数量，以及线程是如何组织的。

<<<>>>运算符完整的执行配置参数形式是<<<Dg, Db, Ns, S>>>

• 参数Dg用于定义整个grid的维度和尺寸，即一个grid有多少个block。为dim3类型。Dim3 Dg(Dg.x, Dg.y, 1)表示grid中每行有Dg.x个block，每列有Dg.y个block，第三维恒为1(目前一个核函数只有一个grid)。整个grid中共有Dg.x*Dg.y个block，其中Dg.x和Dg.y最大值为65535。

• 参数Db用于定义一个block的维度和尺寸，即一个block有多少个thread。为dim3类型。Dim3 Db(Db.x, Db.y, Db.z)表示整个block中每行有Db.x个thread，每列有Db.y个thread，高度为Db.z。Db.x和Db.y最大值为512，Db.z最大值为62。 一个block中共有Db.x*Db.y*Db.z个thread。计算能力为1.0,1.1的硬件该乘积的最大值为768，计算能力为1.2,1.3的硬件支持的最大值为1024。

• 参数Ns是一个可选参数，用于设置每个block除了静态分配的shared Memory以外，最多能动态分配的shared memory大小，单位为byte。不需要动态分配时该值为0或省略不写。

• 参数S是一个cudaStream_t类型的可选参数，初始值为零，表示该核函数处在哪个流之中。

在第一个CUDA程序中使用了1个线程块，每个线程块包含size个并行线程，每个线程的索引是threadIdx.x。

也可以选择创建size个线程块，每个线程块包含1个线程，核函数的调用更改为：

addKernel<<<size, 1>>>(dev_c, dev_a, dev_b);

线程的索引更改为blockIdx.x。完整程序如下：

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <stdio.h>

cudaError_t addWithCuda(int *c, const int *a, const int *b, unsigned int size);

__global__ void addKernel(int *c, const int *a, const int *b)
{
	int i = blockIdx.x;
	c[i] = a[i] + b[i];
}

int main()
{
	const int arraySize = 5;
	const int a[arraySize] = { 1, 2, 3, 4, 5 };
	const int b[arraySize] = { 10, 20, 30, 40, 50 };
	int c[arraySize] = { 0 };

	// Add vectors in parallel.
	cudaError_t cudaStatus = addWithCuda(c, a, b, arraySize);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "addWithCuda failed!");
		return 1;
	}

	printf("{1,2,3,4,5} + {10,20,30,40,50} = {%d,%d,%d,%d,%d}
",
		c[0], c[1], c[2], c[3], c[4]);

	// cudaDeviceReset must be called before exiting in order for profiling and
	// tracing tools such as Nsight and Visual Profiler to show complete traces.
	cudaStatus = cudaDeviceReset();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaDeviceReset failed!");
		return 1;
	}
	getchar();
	return 0;
}

// Helper function for using CUDA to add vectors in parallel.
cudaError_t addWithCuda(int *c, const int *a, const int *b, unsigned int size)
{
	int *dev_a = 0;
	int *dev_b = 0;
	int *dev_c = 0;
	cudaError_t cudaStatus;

	// Choose which GPU to run on, change this on a multi-GPU system.
	cudaStatus = cudaSetDevice(0);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaSetDevice failed!  Do you have a CUDA-capable GPU installed?");
		goto Error;
	}

	// Allocate GPU buffers for three vectors (two input, one output)    .
	cudaStatus = cudaMalloc((void**)&dev_c, size * sizeof(int));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
		goto Error;
	}

	cudaStatus = cudaMalloc((void**)&dev_a, size * sizeof(int));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
		goto Error;
	}

	cudaStatus = cudaMalloc((void**)&dev_b, size * sizeof(int));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
		goto Error;
	}

	// Copy input vectors from host memory to GPU buffers.
	cudaStatus = cudaMemcpy(dev_a, a, size * sizeof(int), cudaMemcpyHostToDevice);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
		goto Error;
	}

	cudaStatus = cudaMemcpy(dev_b, b, size * sizeof(int), cudaMemcpyHostToDevice);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
		goto Error;
	}

	// Launch a kernel on the GPU with one thread for each element.
	//addKernel<<<1, size>>>(dev_c, dev_a, dev_b);
	addKernel << <size, 1 >> > (dev_c, dev_a, dev_b);

	// Check for any errors launching the kernel
	cudaStatus = cudaGetLastError();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "addKernel launch failed: %s
", cudaGetErrorString(cudaStatus));
		goto Error;
	}

	// cudaDeviceSynchronize waits for the kernel to finish, and returns
	// any errors encountered during the launch.
	cudaStatus = cudaDeviceSynchronize();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching addKernel!
", cudaStatus);
		goto Error;
	}

	// Copy output vector from GPU buffer to host memory.
	cudaStatus = cudaMemcpy(c, dev_c, size * sizeof(int), cudaMemcpyDeviceToHost);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
		goto Error;
	}

Error:
	cudaFree(dev_c);
	cudaFree(dev_a);
	cudaFree(dev_b);

	return cudaStatus;
}

执行结果一致：

更普遍的情况是需要创建多个线程块，每个线程块包含多个并行线程，这种情况下线程索引的计算为：

int tid=threadIdx.x+blockIdx.x*blockDim.x;

blockIdx代表线程块在网格中的索引值，blockDim代表线程块的尺寸大小，另外还有gridDim代表网格的尺寸大小。

如果有N个并行的任务，我们希望每个线程块固定包含6个并行的线程，则可以使用以下的核函数调用：

addKernel<<<(N+5)/6, 6>>>(dev_c, dev_a, dev_b);

把第一个CUDA程序的向量个数增加到15个，修改成以上调用方式：

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <stdio.h>

cudaError_t addWithCuda(int *c, const int *a, const int *b, unsigned int size);

__global__ void addKernel(int *c, const int *a, const int *b)
{
	int i = threadIdx.x + blockIdx.x*blockDim.x;
	if (i < 15)
		c[i] = a[i] + b[i];
}

int main()
{
	const int arraySize = 15;
	const int a[arraySize] = { 1, 2, 3, 4, 5,6,7,8,9,10,11,12,13,14,15 };
	const int b[arraySize] = { 10, 20, 30, 40, 50,60,70,80,90,100,110,120,130,140,150 };
	int c[arraySize] = { 0 };

	// Add vectors in parallel.
	cudaError_t cudaStatus = addWithCuda(c, a, b, arraySize);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "addWithCuda failed!");
		return 1;
	}
	printf("{ 1, 2, 3, 4, 5,6,7,8,9,10,11,12,13,14,15}+
{ 10, 20, 30, 40, 50,60,70,80,90,100,110,120,130,140,150}=
{%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,%d}
",
		c[0], c[1], c[2], c[3], c[4], c[5], c[6], c[7], c[8], c[9], c[10], c[11], c[12], c[13], c[14]);

	// cudaDeviceReset must be called before exiting in order for profiling and
	// tracing tools such as Nsight and Visual Profiler to show complete traces.
	cudaStatus = cudaDeviceReset();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaDeviceReset failed!");
		return 1;
	}
	getchar();
	return 0;
}

// Helper function for using CUDA to add vectors in parallel.
cudaError_t addWithCuda(int *c, const int *a, const int *b, unsigned int size)
{
	int *dev_a = 0;
	int *dev_b = 0;
	int *dev_c = 0;
	cudaError_t cudaStatus;

	// Choose which GPU to run on, change this on a multi-GPU system.
	cudaStatus = cudaSetDevice(0);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaSetDevice failed!  Do you have a CUDA-capable GPU installed?");
		goto Error;
	}

	// Allocate GPU buffers for three vectors (two input, one output)    .
	cudaStatus = cudaMalloc((void**)&dev_c, size * sizeof(int));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
		goto Error;
	}

	cudaStatus = cudaMalloc((void**)&dev_a, size * sizeof(int));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
		goto Error;
	}

	cudaStatus = cudaMalloc((void**)&dev_b, size * sizeof(int));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc failed!");
		goto Error;
	}

	// Copy input vectors from host memory to GPU buffers.
	cudaStatus = cudaMemcpy(dev_a, a, size * sizeof(int), cudaMemcpyHostToDevice);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
		goto Error;
	}

	cudaStatus = cudaMemcpy(dev_b, b, size * sizeof(int), cudaMemcpyHostToDevice);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
		goto Error;
	}

	// Launch a kernel on the GPU with one thread for each element.
	//addKernel<<<1, size>>>(dev_c, dev_a, dev_b);
	addKernel << <(size + 5) / 6, 6 >> > (dev_c, dev_a, dev_b);

	// Check for any errors launching the kernel
	cudaStatus = cudaGetLastError();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "addKernel launch failed: %s
", cudaGetErrorString(cudaStatus));
		goto Error;
	}

	// cudaDeviceSynchronize waits for the kernel to finish, and returns
	// any errors encountered during the launch.
	cudaStatus = cudaDeviceSynchronize();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching addKernel!
", cudaStatus);
		goto Error;
	}

	// Copy output vector from GPU buffer to host memory.
	cudaStatus = cudaMemcpy(c, dev_c, size * sizeof(int), cudaMemcpyDeviceToHost);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
		goto Error;
	}

Error:
	cudaFree(dev_c);
	cudaFree(dev_a);
	cudaFree(dev_b);

	return cudaStatus;
}

执行结果：

以下CUDA和OpenCV混合编程，对一幅图像上每个像素点的颜色执行一次运算，生成一幅规则的图形。

新建了一个 dim3类型的变量grid(DIM, DIM)，代表一个二维的网格，尺寸大小是DIM*DIM个线程块:

#include "cuda_runtime.h"  
#include <highgui.hpp>  

using namespace cv;

#define DIM 600   //图像长宽  

__global__ void kernel(unsigned char *ptr)
{
	// map from blockIdx to pixel position  
	int x = blockIdx.x;
	int y = blockIdx.y;
	int offset = x + y * gridDim.x;

	//BGR设置
	ptr[offset * 3 + 0] = 999 * x*y % 255;
	ptr[offset * 3 + 1] = 99 * x*x*y*y % 255;
	ptr[offset * 3 + 2] = 9 * offset*offset % 255;
}

// globals needed by the update routine  
struct DataBlock
{
	unsigned char   *dev_bitmap;
};

int main(void)
{
	DataBlock   data;
	cudaError_t error;

	Mat image = Mat(DIM, DIM, CV_8UC3, Scalar::all(0));

	data.dev_bitmap = image.data;
	unsigned char    *dev_bitmap;

	error = cudaMalloc((void**)&dev_bitmap, 3 * image.cols*image.rows);
	data.dev_bitmap = dev_bitmap;

	dim3    grid(DIM, DIM);
	//DIM*DIM个线程块
	kernel <<<grid, 1 >>> (dev_bitmap);

	error = cudaMemcpy(image.data, dev_bitmap,
		3 * image.cols*image.rows,
		cudaMemcpyDeviceToHost);

	error = cudaFree(dev_bitmap);

	imshow("CUDA Grid/Block/Thread)", image);
	waitKey();
}

执行效果：