QCS8550上是否有GPU burn的程序?

[weixin_32269179]

QCS8550上是否有GPU burn的程序?

[weixin_38498942]

没有官方释放的gpu压力测试的程序,但是目前QCS8550的GPU对opencl都能进行支持,所以我们可以使用以下的测试程序,将GPU的负载拉到100%, 这个程序需要手动编译:

#include <CL/cl.h>
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>

const char *kernelSource = 
    "__kernel void stress_test(__global float *a, __global float *b, __global float *c) {"
    "   int id = get_global_id(0);"
    "   float tmp = 0.0;"
    "   for (int i = 0; i < 1000; ++i) {"  // 增加计算量
    "       tmp += a[id] * b[id];"
    "   }"
    "   c[id] = tmp;"
    "}";

int main() {
    // Initialize OpenCL
    cl_platform_id platform;
    cl_device_id device;
    cl_context context;
    cl_command_queue queue;
    cl_program program;
    cl_kernel kernel;
    cl_mem a_mem, b_mem, c_mem;
    float *a, *b, *c;
    size_t globalSize, localSize;
    cl_int err;

    // Allocate memory
    globalSize = 1024 * 1024 * 2;  // 增加全局工作组大小
    localSize = 64;
    a = (float*)malloc(globalSize * sizeof(float));
    b = (float*)malloc(globalSize * sizeof(float));
    c = (float*)malloc(globalSize * sizeof(float));

    // Fill the arrays with random data
    srand(time(NULL));
    for(size_t i = 0; i < globalSize; i++) {
        a[i] = rand() / (float)RAND_MAX;
        b[i] = rand() / (float)RAND_MAX;
    }

    // Get platform and device information
    clGetPlatformIDs(1, &platform, NULL);
    clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);

    // Create context and command queue
    context = clCreateContext(0, 1, &device, NULL, NULL, &err);
    queue = clCreateCommandQueue(context, device, 0, &err);

    // Create program from source
    program = clCreateProgramWithSource(context, 1, &kernelSource, NULL, &err);
    clBuildProgram(program, 0, NULL, NULL, NULL, NULL);

    // Create kernel
    kernel = clCreateKernel(program, "stress_test", &err);

    // Create memory buffers
    a_mem = clCreateBuffer(context, CL_MEM_READ_ONLY, globalSize * sizeof(float), NULL, &err);
    b_mem = clCreateBuffer(context, CL_MEM_READ_ONLY, globalSize * sizeof(float), NULL, &err);
    c_mem = clCreateBuffer(context, CL_MEM_WRITE_ONLY, globalSize * sizeof(float), NULL, &err);

    // Copy data to buffers
    clEnqueueWriteBuffer(queue, a_mem, CL_TRUE, 0, globalSize * sizeof(float), a, 0, NULL, NULL);
    clEnqueueWriteBuffer(queue, b_mem, CL_TRUE, 0, globalSize * sizeof(float), b, 0, NULL, NULL);

    // Set kernel arguments
    clSetKernelArg(kernel, 0, sizeof(cl_mem), &a_mem);
    clSetKernelArg(kernel, 1, sizeof(cl_mem), &b_mem);
    clSetKernelArg(kernel, 2, sizeof(cl_mem), &c_mem);

    // Execute the kernel in an infinite loop to stress test the GPU
    while (1) {
        clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &globalSize, &localSize, 0, NULL, NULL);
        // Remove clFinish to avoid unnecessary synchronization
        // clFinish(queue); // Ensure the kernel execution is completed before launching another
    }

    // Clean up (this part will never be reached in this infinite loop example)
    clReleaseMemObject(a_mem);
    clReleaseMemObject(b_mem);
    clReleaseMemObject(c_mem);
    clReleaseProgram(program);
    clReleaseKernel(kernel);
    clReleaseCommandQueue(queue);
    clReleaseContext(context);
    free(a);
    free(b);
    free(c);

    return 0;
}