From 67a5e3d825fde17b7b00906ce42b0bd8cafebc4e Mon Sep 17 00:00:00 2001
From: admin <weikou2014>
Date: 星期四, 18 八月 2022 18:22:14 +0800
Subject: [PATCH] '完善'
---
ConsoleApplication/OpenclTest.cpp | 617 ++++++++++++++++++-------------------------------------
1 files changed, 206 insertions(+), 411 deletions(-)
diff --git a/ConsoleApplication/OpenclTest.cpp b/ConsoleApplication/OpenclTest.cpp
index 673c0b4..ee51a5f 100644
--- a/ConsoleApplication/OpenclTest.cpp
+++ b/ConsoleApplication/OpenclTest.cpp
@@ -1,431 +1,226 @@
-#include<stdio.h>
-#include<stdlib.h>
-#include<iostream>
-#include<CL/cl.h>
-#include <chrono>
-#include "OpenCLExcuter.h"
-#include "ImgUtil.h"
-using namespace std;
-void printDevice();
-uchar* createDemoData(int line);
-uchar* createTemplateData(int line);
+#include <stdio.h>
+#include <stdlib.h>
+#include <CL/cl.h>
-int LINE_NUMBER = 10;
-int RE = _NUMBER_L2_TOTAL_NUMBER;
+const int ARRAY_SIZE = 1000;
+/*
+1. 创建平台
+2. 创建设备
+3. 根据设备创建上下文
+*/
+cl_context CreateContext(cl_device_id* device) {
+ cl_int errNum;
+ cl_uint numPlateforms;
+ cl_platform_id firstPlatformId;
+ cl_context context = NULL;
+ errNum = clGetPlatformIDs(1, &firstPlatformId, &numPlateforms);
-
-
-#define CHECK_ERRORS(ERR) \
- if(ERR != CL_SUCCESS){ \
- std::cerr << "OpenCL error code" << ERR << "file: " << __FILE__ << "line: " << __LINE__ << ".\nExiting..." << std::endl; \
- exit(1); \
+ if (errNum != CL_SUCCESS || numPlateforms <= 0) {
+ printf("Failed to find any OpenCL platforms.\n");
+ return NULL;
}
-int main1() {
- //printDevice();
- //return 1;
- auto starttime = std::chrono::steady_clock::now();
+ errNum = clGetDeviceIDs(firstPlatformId, CL_DEVICE_TYPE_GPU, 1, device, NULL);
+ if (errNum != CL_SUCCESS) {
+ printf("There is no GPU , trying CPU...\n");
+ errNum = clGetDeviceIDs(firstPlatformId, CL_DEVICE_TYPE_CPU, 1, device, NULL);
- cl_int error;
- cl_uint num_of_platforms = 0;
- cl_platform_id platforms;
+ }
+ if (errNum != CL_SUCCESS) {
+ printf("There is NO CPU or GPU\n");
+ return NULL;
+ }
+ context = clCreateContext(NULL, 1, device, NULL, NULL, &errNum);
+ if (errNum != CL_SUCCESS) {
+ printf("Create context error\n");
+ return NULL;
+ }
+ return context;
+}
+/*
+@ 在上下文可用的第一个设备中创建命令队列
+*/
- cl_device_id devices;
+cl_command_queue CreateCommandQueue(cl_context context, cl_device_id device) {
+ cl_int errNum;
+ cl_command_queue commandQueue = NULL;
+ //commandQueue = clCreateCommandQueue(context, device, 0, NULL);
+ // OpenCL 2.0 的用法
+ commandQueue = clCreateCommandQueueWithProperties(context, device, 0, NULL);
+ if (commandQueue == NULL) {
+ printf("Failed to create commandQueue for device 0\n");
+ return NULL;
+ }
+ return commandQueue;
+}
- cl_context context;
+char* ReadKernelSourceFile(const char* filename, size_t* length) {
+ FILE* file = NULL;
+ size_t sourceLenth;
+ char* sourceString;
+ int ret;
+ fopen_s(&file, filename, "rb");
- FILE* program_handle;
- size_t program_size;
- char* program_buffer;
+ if (file == NULL) {
+ printf("%s at %d : Can't open %s \n", __FILE__, __LINE__ - 2, filename);
+ return NULL;
+ }
+ //重定位到文件末尾
+ fseek(file, 0, SEEK_END);
+ sourceLenth = ftell(file);
+ //重定位到文件开头
+ fseek(file, 0, SEEK_SET);
+ sourceString = (char*)malloc(sourceLenth + 1);
+ sourceString[0] = '\0';
+ ret = fread(sourceString, sourceLenth, 1, file);
+ if (ret == 0) {
+ printf("%s at %d : Cant't read source %s\n", __FILE__, __LINE__ - 2, filename);
+ return NULL;
+ }
+ fclose(file);
+ if (length != 0) {
+ *length = sourceLenth;
+ }
+ sourceString[sourceLenth] = '\0';
+ return sourceString;
+
+
+}
+cl_program CreateProgram(cl_context context, cl_device_id device, const char* filename) {
+ cl_int errNum;
cl_program program;
+ //记录大小的数据类型
+ size_t program_length;
+ char* const source = ReadKernelSourceFile(filename, &program_length);
+ program = clCreateProgramWithSource(context, 1, (const char**)&source, NULL, NULL);
- size_t log_size;
- char* program_log;
-
- char kernel_name[] = "createBuffer";
- cl_kernel kernel;
-
- cl_command_queue queue;
- //获取平台
- error = clGetPlatformIDs(1, &platforms, &num_of_platforms);
- if (error != 0) {
- printf("Get platform failed!");
- return -1;
+ if (program == NULL) {
+ printf("Failed to creae CL program from source.\n");
+ return NULL;
}
- //获取设备
- error = clGetDeviceIDs(platforms, CL_DEVICE_TYPE_GPU, 1, &devices, NULL);
- if (error != 0) {
- printf("Get device failed!");
- return -1;
+ errNum = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
+ if (errNum != CL_SUCCESS) {
+ char buildLog[16384];
+ clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(buildLog), buildLog, NULL);
+ printf("Error in kernel : %s \n", buildLog);
+ clReleaseProgram(program);
+ return NULL;
+ }
+ return program;
+}
+/*
+@ 创建内存对象
+*/
+bool CreateMemObjects(cl_context context, cl_mem memObjects[3], float* a, float* b) {
+ memObjects[0] = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float) * ARRAY_SIZE, a, NULL);
+ memObjects[1] = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float) * ARRAY_SIZE, b, NULL);
+ memObjects[2] = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * ARRAY_SIZE, NULL, NULL);
+ if (memObjects[0] == NULL || memObjects[1] == NULL || memObjects[2] == NULL) {
+ printf("Error creating memeory objects.\n");
+ return false;
+
+ }
+ return true;
+}
+/*
+@ 清楚OpenCL资源
+*/
+void CleanUp(cl_context context, cl_command_queue commandQueue, cl_program program, cl_kernel kernel, cl_mem memObjects[3]) {
+ for (int i = 0; i < 3; i++) {
+ if (memObjects[i] != 0) {
+ clReleaseMemObject(memObjects[i]);
+ }
}
- std::cout << " 获取设备：" << std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::steady_clock::now() - starttime).count() << std::endl;
-
- //创建上下文
- context = clCreateContext(NULL, 1, &devices, NULL, NULL, &error);
- if (error != 0) {
- printf("Creat context failed!");
- return -1;
+ if (commandQueue != 0) {
+ clReleaseCommandQueue(commandQueue);
}
- std::cout << " 创建上下文：" << std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::steady_clock::now() - starttime).count() << std::endl;
- //创建程序；注意要用"rb"
- fopen_s(&program_handle,"kernel.cl", "rb");
- if (program_handle == NULL) {
- printf("The kernle can not be opened!");
- return -1;
- }
- fseek(program_handle, 0, SEEK_END);
- program_size = ftell(program_handle);
- rewind(program_handle);
-
- program_buffer = (char*)malloc(program_size + 1);
- program_buffer[program_size] = '\0';
- error = fread(program_buffer, sizeof(char), program_size, program_handle);
- if (error == 0) {
- printf("Read kernel failed!");
- return -1;
- }
- fclose(program_handle);
- program = clCreateProgramWithSource(context, 1, (const char**)&program_buffer,
- &program_size, &error);
- if (error < 0) {
- printf("Couldn't create the program!");
- return -1;
- }
- //编译程序
- error = clBuildProgram(program, 1, &devices, NULL, NULL, NULL);
- if (error < 0) {
- //确定日志文件的大小
- clGetProgramBuildInfo(program, devices, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
- program_log = (char*)malloc(log_size + 1);
- program_log[log_size] = '\0';
- //读取日志
- clGetProgramBuildInfo(program, devices, CL_PROGRAM_BUILD_LOG,
- log_size + 1, program_log, NULL);
- printf("%s\n", program_log);
- free(program_log);
- return -1;
- }
- free(program_buffer);
-
- //创建命令队列
- queue = clCreateCommandQueue(context, devices, CL_QUEUE_PROFILING_ENABLE, &error);
- if (error < 0) {
- printf("Coudn't create the command queue");
- return -1;
- }
- //----------程序正文开始---------
-
- for (int n = 0;n < 100;n++)
- {
-
- auto startexectime = std::chrono::steady_clock::now();
-
-
-
- //创建内核
- kernel = clCreateKernel(program, "createBuffer", &error);
- if (kernel == NULL) {
- printf("Couldn't create kernel!\n");
- return -1;
- }
- //初始化参数
- const int size = 70000;
- float result[size];
- float a_in[size];
- float b_in[size];
- for (int i = 0; i < size; i++) {
- a_in[i] = i+n;
- b_in[i] = i * 2.0;
- }
- //创建缓存对象
- cl_mem memObject1 = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float) * size, a_in, &error);
- if (error < 0) {
- printf("Creat memObject1 failed!\n");
- return -1;
- }
- cl_mem memObject2 = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
- sizeof(float) * size, b_in, &error);
- if (error < 0) {
- printf("Creat memObject2 failed!\n");
- return -1;
- }
- cl_mem memObject3 = clCreateBuffer(context, CL_MEM_WRITE_ONLY,
- sizeof(float) * size, NULL, &error);
- if (error < 0) {
- printf("Creat memObject3 failed!\n");
- return -1;
- }
- //设置内核参数
- error = clSetKernelArg(kernel, 0, sizeof(cl_mem), &memObject1);
- error |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &memObject2);
- error |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &memObject3);
- if (error != CL_SUCCESS) {
- printf("Error setting kernel arguments!\n");
- return -1;
- }
- //执行内核
- size_t globalWorkSize[1] = { size };
- size_t localWorkSize[1] = { 1 };
-
- const int work_dim = 2;
-
- error = clEnqueueNDRangeKernel(queue, kernel, work_dim, NULL, globalWorkSize,
- localWorkSize, 0, NULL, NULL);
- if (error != CL_SUCCESS) {
- printf("Error queuing kernel for execution!\n");
- return -1;
- }
-
- //读取执行结果
- error = clEnqueueReadBuffer(queue, memObject3, CL_TRUE, 0, size * sizeof(float),
- result, 0, NULL, NULL);
- if (error != CL_SUCCESS) {
- printf("Error reading result buffer!\n");
- return -1;
- }
- //显示结果
- for (int i = 0; i < 2; i++) {
- printf("%f ", result[i]);
- }
-
+ if (kernel != 0) {
clReleaseKernel(kernel);
- clReleaseMemObject(memObject1);
- clReleaseMemObject(memObject2);
- clReleaseMemObject(memObject3);
-
-
- std::cout << " 执行结束：" << std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::steady_clock::now() - startexectime).count() << std::endl;
}
+ if (program != 0) {
+ clReleaseProgram(program);
+ }
+ if (context != 0) {
+ clReleaseContext(context);
+ }
+}
- //-------------程序正文结束
+/*main函数*/
+int main__(int argc, char** agrv) {
+ cl_context context = 0;
+ cl_command_queue commandQueue = 0;
+ cl_program program = 0;
+ cl_device_id device = 0;
+ cl_kernel kernel = 0;
+ cl_mem memObjects[3] = { 0,0,0 };
+ cl_int errNum;
+ //创建OpenCL上下文
+ context = CreateContext(&device);
+ if (context == NULL) {
+ printf("Failed to create OpenCL context\n");
+ return 1;
+ }
+ //获得OpenCL设备，并创建命令队列
+ commandQueue = CreateCommandQueue(context, device);
+ if (commandQueue == NULL) {
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ //创建OpenCL程序
+ program = CreateProgram(context, device, "kernel.cl");
+ if (program == NULL) {
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ kernel = clCreateKernel(program, "vector_add", NULL);
+ if (kernel == NULL) {
+ printf("Failed to create kernel\n");
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ //创建OpenCL内存对象
+ float result[ARRAY_SIZE];
+ float a[ARRAY_SIZE];
+ float b[ARRAY_SIZE];
+ for (int i = 0; i < ARRAY_SIZE; i++) {
+ a[i] = (float)i;
+ b[i] = (float)(i * 2);
+ }
+ if (!CreateMemObjects(context, memObjects, a, b)) {
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ //设置内核参数
+ errNum = clSetKernelArg(kernel, 0, sizeof(cl_mem), &memObjects[0]);
+ errNum |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &memObjects[1]);
+ errNum |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &memObjects[2]);
-
- //释放资源
- clReleaseDevice(devices);
- clReleaseContext(context);
- clReleaseProgram(program);
- clReleaseCommandQueue(queue);
-
- std::cout << " 耗时：" << std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::steady_clock::now() - starttime).count() << std::endl;
+ if (errNum != CL_SUCCESS) {
+ printf("Error setting kernel arguments.\n");
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ size_t globalWorkSize[1] = { ARRAY_SIZE };
+ size_t localWorkSize[1] = { 1 };
+ //执行内核
+ errNum = clEnqueueNDRangeKernel(commandQueue, kernel, 1, NULL, globalWorkSize, localWorkSize, 0, NULL, NULL);
+ if (errNum != CL_SUCCESS) {
+ printf("Error queueing kernel for execution\n");
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ //将计算的结果拷贝到主机上
+ errNum = clEnqueueReadBuffer(commandQueue, memObjects[2], CL_TRUE, 0, ARRAY_SIZE * sizeof(float), result, 0, NULL, NULL);
+ if (errNum != CL_SUCCESS) {
+ printf("Error reading result buffer.\n");
+ CleanUp(context, commandQueue, program, kernel, memObjects);
+ return 1;
+ }
+ for (int i = 0; i < ARRAY_SIZE; i++) {
+ printf("i=%d:%f\n", i, result[i]);
+ }
+ printf("Executed program succesfully\n");
+ CleanUp(context, commandQueue, program, kernel, memObjects);
return 0;
}
-
-
-int main3() {
- ImgUtil::init();
- OpenCLExcuter* op = new OpenCLExcuter();
- int dataCount = 100;
-
-
- int rows = 8* dataCount;
- int cols = 5 * LINE_NUMBER * RE;
- uchar* b_in = createTemplateData(dataCount);
- uchar* a_in = createDemoData(dataCount);
-
- //printf("输入1: \n");
- for (int r = 0;r < rows;r++) {
- for (int c = 0;c < cols;c++) {
- // printf("%d ", a_in[r*cols+c]);
- }
- // printf("\n");
- }
-
- //printf("输入2: \n");
- for (int r = 0;r < rows;r++) {
- for (int c = 0;c < cols;c++) {
- // printf("%d ", b_in[r * cols + c]);
- }
- //printf("\n");
- }
-
- /*
- int rows = 10;
- int cols = 10;
- uchar* b_in = (uchar*)malloc(sizeof(uchar*)*400);
- uchar* a_in = (uchar*)malloc(sizeof(uchar*) * 400);
- for (int i = 0;i < 400;i++) {
- b_in[i] = i;
- a_in[i] = i + 1;
- }
- */
-
- op->init();
- for (int i = 0;i < 100;i++)
- {
- op->recognition_numbers(a_in, b_in, rows, cols ,_NUMBER_L2_WIDTH, _NUMBER_L2_HEIGHT, _NUMBER_L2_TOTAL_NUMBER);
- }
- op->destory();
-
- return 1;
-}
-
-uchar* createDemoData(int line) {
- unsigned char* data = (unsigned char*)malloc(sizeof(unsigned char) * (_NUMBER_L2_HEIGHT * line) * 5 * LINE_NUMBER * RE);
- int outLineDataCount = 8 * 5 * LINE_NUMBER * RE;
- int inLineDataCount = 5 * LINE_NUMBER * RE;
- for (int l = 0;l < line;l++)
- {
- int outLineCount = outLineDataCount * l;
- for (int re = 0;re < RE;re++) {
-
-
-
- for (int n = 0;n < LINE_NUMBER;n++)
- {
- for (int r = 0;r < 8;r++) {
- int intLineCount = inLineDataCount * r;
-
- for (int c = 0;c < 5;c++) {
- uchar value = ImgUtil::NUMS_LEVEL2[re%10].data.ptr(r)[c];
- int index = outLineCount;
- index += intLineCount;
- int x = re * LINE_NUMBER * 5 + n * 5 + c;
- index += x;
- data[index] = value > 40 ? 1 : 0;
- }
- }
- }
- }
- }
- return data;
-}
-
-uchar* createTemplateData(int line) {
- unsigned char* data = (unsigned char*)malloc(sizeof(unsigned char) * (_NUMBER_L2_HEIGHT * line) * _NUMBER_L2_WIDTH * LINE_NUMBER * RE);
- int outLineDataCount = _NUMBER_L2_HEIGHT * _NUMBER_L2_WIDTH * LINE_NUMBER * RE;
- int inLineDataCount = _NUMBER_L2_WIDTH * LINE_NUMBER * RE;
- for (int l = 0;l < line;l++)
- {
- int outLineCount = outLineDataCount * l;
- for (int re = 0;re < RE;re++) {
- for (int n = 0;n < LINE_NUMBER;n++)
- {
- for (int r = 0;r < _NUMBER_L2_HEIGHT;r++) {
- int intLineCount = inLineDataCount * r;
-
- for (int c = 0;c < _NUMBER_L2_WIDTH;c++) {
- uchar value = ImgUtil::NUMS_LEVEL2[n].data.ptr(r)[c];
- int index = outLineCount;
- index += intLineCount;
- int x = re * LINE_NUMBER * _NUMBER_L2_WIDTH + n * _NUMBER_L2_WIDTH + c;
- index += x;
- data[index] = value > 40 ? 1 : 0;
- }
- }
- }
- }
- }
- return data;
-}
-
-
-void printDevice() {
-
- cl_int err = CL_SUCCESS;
-
- // 1. 获取当前设备所有支持OpenCL的平台的数量
- cl_uint num_of_platforms = 0;
- err = clGetPlatformIDs(0, 0, &num_of_platforms);
- CHECK_ERRORS(err);
-
- // 2. 获取当前设备所有支持OpenCL的平台的信息
- cl_platform_id* platforms = new cl_platform_id[num_of_platforms];
- err = clGetPlatformIDs(num_of_platforms, platforms, 0);
- CHECK_ERRORS(err);
-
- cl_char platform_names[10][50];
- // 3. 打印平台信息
- cout << "平台信息：\n";
- for (cl_uint i = 0; i < num_of_platforms; i++)
- {
- // 获取平台字符串的长度
- size_t platform_name_length = 0;
- err = clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME, 0, 0, &platform_name_length);
- CHECK_ERRORS(err);
-
- // 获取平台字符串
- char* platform_name = new char[platform_name_length];
- err = clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME, platform_name_length, platform_name, 0);
- CHECK_ERRORS(err);
-
- cout << " [" << i << "] " << platform_name << endl;
-
- // 备份platform name
- if (i < 10) {
- memset(platform_names[i], 0, 50);
- memcpy(platform_names[i], platform_name, platform_name_length);
- }
-
- delete[] platform_name;
- }
-
- // 4. 查询各平台设备数量
- struct {
- cl_device_type type;
- const char* name;
- cl_uint count;
- }devices[] = {
- {CL_DEVICE_TYPE_GPU, "CL_DEVICE_TYPE_GPU", 0}, // GPU
- {CL_DEVICE_TYPE_CPU, "CL_DEVICE_TYPE_CPU", 0}, // CPU
- {CL_DEVICE_TYPE_ACCELERATOR, "CL_DEVICE_TYPE_ACCELERATOR", 0} // 加速器
- };
-
- // 5. 遍历查询各个平台下拥有的设备数量
- for (cl_int j = 0; j < num_of_platforms; j++) {
- cl_platform_id platform = platforms[j];
- cout << "\nplatform:" << platform_names[j] << endl;
-
- // 6. 遍历查询GPU、CPU、ACCELERATOR设备的数量
- for (cl_int i = 0; i < (sizeof(devices) / sizeof(devices[0])); i++)
- {
- err = clGetDeviceIDs(platform, devices[i].type, 0, 0, &devices[i].count);
- if (err == CL_DEVICE_NOT_FOUND) {
- devices[i].count = 0;
- err = CL_SUCCESS;
- }
- CHECK_ERRORS(err);
-
- cout << "\tdevices:" << devices[i].name
- << "\tcount:" << devices[i].count;
-
- if (devices[i].count != 0) {
- // 7. 遍历查询GPU、CPU、ACCELERATOR 所有设备的信息
- cl_device_id* device = new cl_device_id[devices[i].count];
- err = clGetDeviceIDs(platform, devices[i].type, devices[i].count, device, 0);
-
- cout << "\t\tdevice name:";
- for (cl_int k = 0; k < devices[i].count; k++)
- {
- // 8. 获取和打印各个设备的name
- size_t length = 0;
- cl_device_id each_device = device[k];
- err = clGetDeviceInfo(each_device, CL_DEVICE_NAME, 0, 0, &length);
- CHECK_ERRORS(err);
-
- char* value = new char[length];
- err = clGetDeviceInfo(each_device, CL_DEVICE_NAME, length, value, 0);
- CHECK_ERRORS(err);
- cout << value << " ";
-
- // 9. 获取和打印各个设备的version
- err = clGetDeviceInfo(each_device, CL_DEVICE_VERSION, 0, 0, &length);
- CHECK_ERRORS(err);
-
- char* version = new char[length];
- err = clGetDeviceInfo(each_device, CL_DEVICE_VERSION, length, version, 0);
- CHECK_ERRORS(err);
- cout << version << " ";
-
- delete[] value;
- delete[] version;
- }
-
- delete[] device;
- }
- cout << endl;
- }
- }
-
- delete[] platforms;
-}
\ No newline at end of file
--
Gitblit v1.8.0