我是OpenCL的初学者。我一直在尝试编写矩阵乘法代码。 它工作正常只有它给垃圾值作为C数组的输出。我无法修复错误。 任何帮助将不胜感激。
这是主机和内核代码。
#include <CL/cl.h>
#include <iostream>
#include <cstdio>
#include <fstream>
#include <stdlib.h>
#include <assert.h>
#include <string.h>
using namespace std;
#define SUCCESS 0
#define FAILURE 1
// Function to convert file name into a string
int convertToString(const char *filename, std::string &s)
{
size_t size;
char *str;
std::fstream f(filename, (std::fstream::in | std::fstream::binary));
if (f.is_open())
{
size_t fileSize;
f.seekg(0, std::fstream::end);
size = fileSize = (size_t)f.tellg();
f.seekg(0, std::fstream::beg);
str = new char[size + 1];
if (!str)
{
f.close();
return 0;
}
f.read(str, fileSize);
f.close();
str[size] = '\0';
s = str;
delete[] str;
return 0;
}
cout << "Error: failed to open file\n:" << filename << endl;
return FAILURE;
}
int main()
{
cl_uint status;
cl_int *error;
int A[9] = {1, 1, 1, 1, 1, 1, 1, 1, 1};
int B[9] = {2, 2, 2, 2, 2, 2, 2, 2, 2};
int C[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
// Setting up platforms
cl_platform_id platform = NULL;
cl_uint numPlatforms = 0;
// Getting no of platforms
status = clGetPlatformIDs(0, NULL, &numPlatforms);
if (status != CL_SUCCESS)
{
cout << "\nUnable to query platforms";
return 0;
}
// Get the platform
if (numPlatforms > 0)
{
cl_platform_id*platforms=
cl_platform_id*)malloc(numPlatforms*sizeof(cl_platform_id));
status = clGetPlatformIDs(numPlatforms, platforms, NULL);
platform = platforms[0];
free(platforms);
}
cl_uint numDevices = 0;
cl_device_id *devices = NULL;
status =
clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 0, devices, &numDevices);
if (numDevices == 0)
{
cout << "No GPU device available! Choosing CPU.\n";
status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_CPU, 0, devices,
&numDevices);
devices = (cl_device_id *)malloc(numDevices * sizeof(cl_device_id));
status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_CPU, numDevices,
devices, NULL);
}
else
{
devices = (cl_device_id *)malloc(numDevices * sizeof(cl_device_id));
status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, numDevices,
devices, NULL);
if (status == 0)
{
cout << "Device error!";
return 0;
}
}
// Creating contexts
cl_context context =
clCreateContext(NULL, 1, devices, NULL, NULL, (cl_int *)status);
if (status != CL_SUCCESS)
{
cout << status;
}
// Creating command queues
cl_command_queue command =
clCreateCommandQueue(context, devices[0], 0, NULL);
// if(error!=CL_SUCCESS)
//{
// cout<<error;
//}
// Creating buffers
cl_mem bufferA = clCreateBuffer(context, CL_MEM_READ_ONLY,
3 * 3 * sizeof(int), NULL, NULL);
cl_mem bufferB = clCreateBuffer(context, CL_MEM_READ_ONLY,
3 * 3 * sizeof(int), NULL, NULL);
cl_mem bufferC = clCreateBuffer(context, CL_MEM_WRITE_ONLY,
3 * 3 * sizeof(int), NULL, NULL);
status = clEnqueueWriteBuffer(command, bufferA, CL_TRUE, 0, 9 * sizeof(int),
(void *)A, 0, NULL, NULL);
status = clEnqueueWriteBuffer(command, bufferB, CL_TRUE, 0, 9 * sizeof(int),
(void *)B, 0, NULL, NULL);
// status=clEnqueueReadBuffer(command,bufferA,CL_TRUE,0,9*sizeof(int),(void*)C,0,NULL,NULL);
const char *filename = "kernel.cl";
string sourceStr;
status = convertToString(filename, sourceStr);
const char *source = sourceStr.c_str();
size_t sourceSize[] = {strlen(source)};
cl_program program =
clCreateProgramWithSource(context, 1, &source, sourceSize, NULL);
status = clBuildProgram(program, numDevices, 0, NULL, NULL, NULL);
cl_kernel myKernel = clCreateKernel(program, "multiply", NULL);
// Setting kernel arguments
clSetKernelArg(myKernel, 0, sizeof(cl_mem), &bufferC);
clSetKernelArg(myKernel, 1, sizeof(cl_mem), &bufferA);
clSetKernelArg(myKernel, 2, sizeof(cl_mem), &bufferB);
size_t localws[2] = {9, 9};
size_t globalws[2] = {3, 3};
status = clEnqueueNDRangeKernel(command, myKernel, 2, NULL, globalws,
localws, 0, NULL, NULL);
status = clEnqueueReadBuffer(command, bufferC, CL_TRUE, 0, 9 * sizeof(int),
(void *)C, 0, NULL, NULL);
for (int i = 0; i < 9; i++) cout << C[i] << " ";
status = clReleaseKernel(myKernel); // Release kernel.
status = clReleaseProgram(program); // Release program object.
status = clReleaseMemObject(bufferA); // Release mem object.
status = clReleaseMemObject(bufferB);
status = clReleaseMemObject(bufferC);
status = clReleaseCommandQueue(command); // Release Command queue.
status = clReleaseContext(context); // Release context.
}
内核代码：
__kernel void multiply(_global int outputC, _global int inputA,
_global int inputB)
{
int row = get_global_id(0);
int col = get_global_id(1);
int sum = 0;
for (int i = 0; i < 3; i++)
sum += inputA[row * 3 + 1] * inputB[i * 3 + col];
outputC[row + 3 + col] = sum;
}
答案 0 :(得分：1)
正如@ Marco13已经指出的那样，内核存在很多问题。
通过clcc之类的工具运行此内核时，您可以看到有许多编译错误开始：
> clcc matmul.cl
"/tmp/OCLu7FyFF.cl", line 1: error: identifier "_global" is undefined
__kernel void multiply(_global int outputC, _global int inputA,
^
"/tmp/OCLu7FyFF.cl", line 1: error: invalid combination of type specifiers
__kernel void multiply(_global int outputC, _global int inputA,
^
"/tmp/OCLu7FyFF.cl", line 1: error: identifier "_global" is undefined
__kernel void multiply(_global int outputC, _global int inputA,
^
"/tmp/OCLu7FyFF.cl", line 1: error: invalid combination of type specifiers
__kernel void multiply(_global int outputC, _global int inputA,
^
"/tmp/OCLu7FyFF.cl", line 2: error: identifier "_global" is undefined
_global int inputB)
^
"/tmp/OCLu7FyFF.cl", line 2: error: invalid combination of type specifiers
_global int inputB)
^
6 errors detected in the compilation of "/tmp/OCLu7FyFF.cl".
像clcc这样的工具对于及早发现错误非常有用。大多数供应商也有自己的独立内核编译器/检查器版本：例如英特尔有Kernel Builder，AMD的CodeXL包含一个静态内核分析器。另一种选择是直接从主机代码中检索内核编译错误，在clGetProgramBuildInfo返回clBuildProgram之后调用CL_BUILD_PROGRAM_FAILURE来检索编译器输出。
一旦修复了这些编译错误，看起来你的内核仍然没有达到预期效果：如上所述，输入和输出应该是指针，因为你将缓冲区传递给内核。此外，输入和输出数组的索引不正确：在for循环中inputA[row * 3 + 1]应为inputA[row * 3 + i]（i而不是1）。将结果保存到outputC时，我希望outputC[row * 3 + col]（row * 3）代替row + 3）。
我没有仔细查看主机代码，但我至少要确保，特别是刚开始使用OpenCL时，始终检查每个返回代码和错误。这将为您节省大量时间和挫折。
最后，如果您希望通过实践方法快速学习OpenCL，我强烈建议您参加Simon McIntosh-Smith和Tom Deakin的开源Hands-on OpenCL培训。它不需要很长时间，非常实用，并提供了许多有用的见解。优化矩阵乘法是逐步显示的用例之一。