CUDA: 共享存储器实现矩阵相乘

共享存储器使用__shared__限定词分配。

　　相关阅读：

       CUDA编程接口：运行初始化与设备存储器

　　CUDA编程接口：使用nvcc编译器的兼容性

　　CUDA编程接口：如何用nvcc编译CUDA程序

　　CUDA编程模型：存储器层次和异构编程

　　CUDA编程模型：内核与线程层次概述

　　正如在前面的文章提到的，共享存储器应当比全局存储器更快，详细内容将在后续文章中介绍。任何用访问共享存储器取代访问全局存储器的机会应当被发掘，如下面的矩阵相乘例子展示的那样。 下面的代码是矩阵相乘的一个直接的实现，没有利用到共享存储器。每个线程读入A的一行和B的一列，然后计算C中对应的元素，如图1所示。这样，A读了B.width次，B读了A.height次。

#include<stdio.h>
#include<cutil.h>
#include<cuda_runtime.h>
#pragma comment(lib, "cudart.lib")
#pragma comment(lib, "cutil32.lib")
bool InitCUDA(void)
{
 int count = 0;
 int i = 0;
 cudaGetDeviceCount(&count);
 if(count == 0)
 {
  fprintf(stderr, "There is no device.\n");
  return false;
 }
 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.\n");
  return false;
 }
 cudaSetDevice(i);
 printf("CUDA initialized.\n");
 return true;  
}
///
#define THREAD_NUM 256
#define MATRIX_SIZE 100 //方阵阶
float a[MATRIX_SIZE][MATRIX_SIZE], b[MATRIX_SIZE][MATRIX_SIZE], c[MATRIX_SIZE][MATRIX_SIZE];
void Random_Number(float source[][MATRIX_SIZE])
{
 for(int i = 0; i < MATRIX_SIZE; i++)
 {
  for(int j = 0; j < MATRIX_SIZE; j++)
  {
   source[i][j] = float(rand() % 10) / 10;
  }
 }
}
__global__ static void Matrix_Mul(const float* source1, const float* source2, float* result)
{
 const int tid = threadIdx.x;
 const int bid = blockIdx.x;
 const int id = bid * blockDim.x + tid;
 const int row = id / MATRIX_SIZE;
 const int col = id % MATRIX_SIZE;
 if(row < MATRIX_SIZE && col < MATRIX_SIZE)
 {
  float t = 0;
  for(int i = 0; i < MATRIX_SIZE; i++)
  {
   t += source1[row * MATRIX_SIZE + i] * source2[i * MATRIX_SIZE + col];
  }
  result[row * MATRIX_SIZE + col] = t;
 }
}
int main(int argc,char* argv[])
{
 InitCUDA();//初始化CUDA
 float *ac, *bc, *cc;
 size_t lda, ldb, ldc;
 srand(0);
 Random_Number(a);
 Random_Number(b);
 unsigned int timer = 0;
 CUT_SAFE_CALL( cutCreateTimer(&timer));
 CUT_SAFE_CALL( cutStartTimer(timer));//define in cutil.h
 //到显存中申请空间,其中lda,ldb,ldc为显存中的行间距
 cudaMallocPitch((void**)&ac, &lda, sizeof(float) * MATRIX_SIZE, MATRIX_SIZE);
 cudaMallocPitch((void**)&bc, &ldb, sizeof(float) * MATRIX_SIZE, MATRIX_SIZE);
 cudaMallocPitch((void**)&cc, &ldc, sizeof(float) * MATRIX_SIZE, MATRIX_SIZE);
 //复制内存元素到显存
 cudaMemcpy2D(ac, lda, a, sizeof(float) * MATRIX_SIZE, sizeof(float) * MATRIX_SIZE, MATRIX_SIZE, cudaMemcpyHostToDevice);
 cudaMemcpy2D(bc, ldb, b, sizeof(float) * MATRIX_SIZE, sizeof(float) * MATRIX_SIZE, MATRIX_SIZE, cudaMemcpyHostToDevice);
 int BLOCK_NUM = (MATRIX_SIZE + THREAD_NUM - 1) / THREAD_NUM;
 Matrix_Mul<<<BLOCK_NUM * MATRIX_SIZE, THREAD_NUM, 0>>>(ac, bc, cc);
 //把处理结果从显存中取出,其中c对应sizeof(float)*MATRIX_SIZE,cc对应ldc
 cudaMemcpy2D(c, sizeof(float) * MATRIX_SIZE, cc, ldc, sizeof(float) * MATRIX_SIZE, MATRIX_SIZE, cudaMemcpyDeviceToHost);
 cutStopTimer(timer);
 printf("Processing time: %f (ms)\n", cutGetTimerValue(timer));
 for(int j = 0; j < MATRIX_SIZE; j++)
 {
  float temp = 0;
  for(int i = 0; i < MATRIX_SIZE; i++)
  {
   temp += a[0][i] * b[i][j];
  }
  printf ("CPU:c[0][%d]=%f\n", j, temp);
  printf ("GPU:c[0][%d]=%f\n", j, c[0][j]);
 }
 system("PAUSE");
 return 0;
}

 　　

　　▲图1 没有使用共享存储器的矩阵相乘

　　下面的例子代码利用了共享存储器实现矩阵相乘。本实现中，每个线程块负责计算一个小方阵Csub，Csub是C的一部分，而块内的每个线程计算Csub的一个元素。如图2所示。Csub等于两个长方形矩阵的乘积：A的子矩阵尺寸是(A.width,block_size)，行索引与Csub相同，B的子矩阵的尺寸是(block_size,A.width)，列索引与Csub相同。为了满足设备的资源，两个长方形的子矩阵分割为尺寸为block_size的方阵，Csub是这些方阵积的和。每次乘法的计算是这样的，首先从全局存储器中将二个对应的方阵载入共享存储器中，载入的方式是一个线程载入一个矩阵元素，然后一个线程计算乘积的一个元素。每个线程积累每次乘法的结果并写入寄存器中，结束后，再写入全局存储器。

　　采用这种将计算分块的方式，利用了快速的共享存储器，节约了许多全局存储器带宽，因为在全局存储器中，A只被读了(B.width/block_size)次同时B读了(A.height/block_size)次。

　　前面代码中的Matrix 类型增加了一个stride域，这样子矩阵能够用同样的类型有效表示。__device__函数(相关阅读的文章中提及)用于读写元素和从矩阵中建立子矩阵。

#define BLOCK_SIZE 16
// Matrices are stored in row-major order:
// M(row, col) = *(M.elements + row * M.stride + col)
typedef struct {
	int width;
	int height; int stride;
	float* elements;
}
Matrix; 
// Get a matrix element 
__device__ float GetElement(const Matrix A, int row, int col)
{ 
	return A.elements[row * A.stride + col];
} 
// Set a matrix element 
__device__ void SetElement(Matrix A, int row, int col, float value)
{
	A.elements[row * A.stride + col] = value;
} 
// Get the BLOCK_SIZExBLOCK_SIZE sub-matrix Asub of A that is
// located col sub-matrices to the right and row sub-matrices down 
// from the upper-left corner of A 
__device__ Matrix GetSubMatrix(Matrix A, int row, int col) {
	Matrix Asub;
	Asub.width = BLOCK_SIZE; 
	Asub.height = BLOCK_SIZE;
	Asub.stride = A.stride;
	Asub.elements = &A.elements[A.stride * BLOCK_SIZE * row + BLOCK_SIZE * col];
	return Asub;
} 
// Thread block size #define BLOCK_SIZE 16 
// Forward declaration of the matrix multiplication kernel 
__global__ void MatMulKernel(const Matrix, const Matrix, Matrix); 
// Matrix multiplication - Host code 
// Matrix dimensions are assumed to be multiples of BLOCK_SIZE
void MatMul(const Matrix A, const Matrix B, Matrix C) 
{
	// Load A and B to device memory 
	Matrix d_A; 
	d_A.width = d_A.stride = A.width;
	d_A.height = A.height;
	size_t size = A.width * A.height * sizeof(float);
	cudaMalloc((void**)&d_A.elements, size);
	cudaMemcpy(d_A.elements, A.elements, size, cudaMemcpyHostToDevice);
	Matrix d_B;
	d_B.width = d_B.stride = B.width;
	d_B.height = B.height;
	size = B.width * B.height * sizeof(float);   
	cudaMalloc((void**)&d_B.elements, size);
	cudaMemcpy(d_B.elements, B.elements, size,cudaMemcpyHostToDevice);
	// Allocate C in device memory 
	Matrix d_C;
	d_C.width = d_C.stride = C.width;
	d_C.height = C.height;
	size = C.width * C.height * sizeof(float);
	cudaMalloc((void**)&d_C.elements, size);
	// Invoke kernel 
	dim3 dimBlock(BLOCK_SIZE, BLOCK_SIZE);
	dim3 dimGrid(B.width / dimBlock.x, A.height / dimBlock.y);
	MatMulKernel<<<dimGrid, dimBlock>>>(d_A, d_B, d_C);
	// Read C from device memory 
	cudaMemcpy(C.elements, d_C.elements, size, cudaMemcpyDeviceToHost);
	// Free device memory 
	cudaFree(d_A.elements); 
	cudaFree(d_B.elements);
	cudaFree(d_C.elements);
}
// Matrix multiplication kernel called by MatMul() 
__global__ void MatMulKernel(Matrix A, Matrix B, Matrix C) {
	// Block row and column 
	int blockRow = blockIdx.y;
	int blockCol = blockIdx.x;
	// Each thread block computes one sub-matrix Csub of C
	Matrix Csub = GetSubMatrix(C, blockRow, blockCol);
	// Each thread computes one element of Csub 
	// by accumulating results into Cvalue
	float Cvalue = 0;
	// Thread row and column within Csub
	int row = threadIdx.y;
	int col = threadIdx.x;
	// Loop over all the sub-matrices of A and B that are
	// required to compute Csub
	// Multiply each pair of sub-matrices together
	// and accumulate the results
	for (int m = 0; m < (A.width / BLOCK_SIZE); ++m)
	{
		// Get sub-matrix Asub of A
		Matrix Asub = GetSubMatrix(A, blockRow, m);
		// Get sub-matrix Bsub of B 
		Matrix Bsub = GetSubMatrix(B, m, blockCol);
		// Shared memory used to store Asub and Bsub respectively
		__shared__ float As[BLOCK_SIZE][BLOCK_SIZE];
		__shared__ float Bs[BLOCK_SIZE][BLOCK_SIZE];
		// Load Asub and Bsub from device memory to shared memory
		// Each thread loads one element of each sub-matrix
		As[row][col] = GetElement(Asub, row, col);
		Bs[row][col] = GetElement(Bsub, row, col);
		// Synchronize to make sure the sub-matrices are loaded
		// before starting the computation 
		__syncthreads();
		// Multiply Asub and Bsub together
		for (int e = 0; e < BLOCK_SIZE; ++e)
			Cvalue += As[row][e] * Bs[e][col];
		// Synchronize to make sure that the preceding
		// computation is done before loading two new
		// sub-matrices of A and B in the next iteration
		__syncthreads(); }
	// Write Csub to device memory
	// Each thread writes one element
	SetElement(Csub, row, col, Cvalue); 
}

　　

　　▲图2 使用共享存储器的矩阵相乘

http://cuda.it168.com/a2011/1207/1285/000001285186.shtml