OpenCL-Puffer ist leer

Question

Ich versuche zu lernen, wie OpenCL mit diesem Tutorial https://anteru.net/blog/2012/11/04/2016/index.html zu verwenden, aber ich glaube nicht, dass die Werte in den Float-Puffer auf nichts festgelegt werden. und wenn ich den Puffer am Ende lese, ist es alles 0s mit einer Ansicht Dezimalzahlen in wissenschaftlicher Notation wie es voller Zufallsspeicher ist. Ich werde den Code unten posten. Was der Kernel macht, ist in 3 Parameter zu nehmen, float buffer x, float buffer y und float a. const int i = get_global_id(0) und y[i] += a * x[i]; das Problem ist (ich denke), dass ich nie irgendwelche Zahlen in aBuffer oder bBuffer setzen, so dass die Multiplikationen und die Addition nichts bedeuten. aber eine seltsame Sache ist, dass, wenn ich den Kernel machen dies tun y[i] += a; es immer noch die gleiche Leistung gibt, wenn ich denke, es wäre 2.

main.cpp geben würde:

#include <iostream> 
#include <vector> 


#ifdef __APPLE__ 
#include "OpenCL/opencl.h" 
#else 
#include "CL/cl.h" 
#endif 


using namespace std; 




int main(int argc, const char * argv[]) { 
    cl_uint platformIdCount = 0; 
    clGetPlatformIDs(0, nullptr, &platformIdCount); 

    vector<cl_platform_id> platformIds(platformIdCount); 
    clGetPlatformIDs(platformIdCount, platformIds.data(), nullptr); 
    cout << "Platforms " << platformIdCount << endl; 

    cl_uint deviceIdCount = 0; 


    clGetDeviceIDs(platformIds[0], CL_DEVICE_TYPE_GPU, 0, nullptr, &deviceIdCount); 
    cout << "Devices " << deviceIdCount << endl; 

    vector<cl_device_id> deviceIds(deviceIdCount); 
    clGetDeviceIDs(platformIds[0], CL_DEVICE_TYPE_GPU, deviceIdCount, deviceIds.data(), nullptr); 


    const cl_context_properties contextProperties[] = { 

     CL_CONTEXT_PLATFORM, 
     reinterpret_cast<cl_context_properties>(platformIds[0]), 
    0,0 
}; 
    cl_int error = 0; 
    cl_context context = clCreateContext(contextProperties, deviceIdCount, deviceIds.data(), nullptr, nullptr, &error); 

    error = 0; 

    cl_mem aBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float) * (64), nullptr, &error); 
    cl_mem bBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float) * (64), nullptr, &error); 



    cl_program program; 

    clBuildProgram(program, deviceIdCount, deviceIds.data(), nullptr, nullptr, nullptr); 

    cl_kernel kernel1 = clCreateKernel(program, "SAXPY", &error); 

    clSetKernelArg(kernel1, 0, sizeof(cl_mem), aBuffer); 
    clSetKernelArg(kernel1, 1, sizeof(cl_mem), bBuffer); 
    static const float two = 2.0f; 
    clSetKernelArg(kernel1, 2, sizeof(float),&two); 

    const size_t globalWorkSize [] = {64,0,0}; 
    cl_command_queue queue; 
    clEnqueueNDRangeKernel(queue, kernel1, 1, nullptr, globalWorkSize, nullptr, 0, nullptr, nullptr); 



    float done[64]; 
    clEnqueueReadBuffer(queue, bBuffer, CL_TRUE, 0, sizeof(float)*64, done, 0, nullptr, nullptr); 
    for (int a = 0; a < 64; a++) { 
     cout << done[a] << endl; 
    } 




    clReleaseContext(context); 
    return 0; 
} 

.cl Datei:

kernel void SAXPY(__global float* x,__global float* y, float a){ 
    const int i = get_global_id(0); 
    //y[i] = 2.0f; 
    y[i] += a * x[i]; 
} 

Answer 1

Erstens, wenn Kernel-Argument festlegen, müssen Sie pass pointer auf das Speicherobjekt:

clSetKernelArg(kernel1, 0, sizeof(cl_mem), &aBuffer); // &aBuffer, not aBuffer 
clSetKernelArg(kernel1, 1, sizeof(cl_mem), &bBuffer); // &bBuffer, not bBuffer 

Zweitens

, nicht wahr create command queue:

cl_command_queue queue = clCreateCommandQueue(context, deviceIds[0], 0, nullptr); 

Drittens Sie nicht nennen clCreateProgramWithSource() vor clBuildProgram().

Darüber hinaus versuchen, die cl_mem Objekte zu initialisieren:

cl_float* mem = (cl_float*) malloc(sizeof(cl_float)*64); 
for(int i=0; i<64; i++) 
    mem[i] = i; 

cl_mem aBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(cl_float) * (64), mem, &error); 
cl_mem bBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(cl_float) * (64), mem, &error); 

free(mem); 

Festcode:

#include <iostream> 
#include <vector> 

#ifdef __APPLE__ 
#include <OpenCL/cl.h> 
#else 
#include "CL/cl.h" 
#endif 

using namespace std; 


int main(int argc, const char * argv[]) { 
    cl_uint platformIdCount = 0; 
    clGetPlatformIDs(0, nullptr, &platformIdCount); 

    vector<cl_platform_id> platformIds(platformIdCount); 
    clGetPlatformIDs(platformIdCount, platformIds.data(), nullptr); 

    cl_uint deviceIdCount = 0; 
    clGetDeviceIDs(platformIds[0], CL_DEVICE_TYPE_GPU, 0, nullptr, &deviceIdCount); 

    vector<cl_device_id> deviceIds(deviceIdCount); 
    clGetDeviceIDs(platformIds[0], CL_DEVICE_TYPE_GPU, deviceIdCount, deviceIds.data(), nullptr); 

    const cl_context_properties contextProperties[] = { 

     CL_CONTEXT_PLATFORM, 
     (cl_context_properties)platformIds[0], 
     0 
    }; 

    cl_int error = 0; 
    cl_context context = clCreateContext(contextProperties, 1, &deviceIds[0], [](const char* errinfo, const void* private_info, size_t cb, void* user_data) -> void { 
     /* context-creation and runtime error handler */ 
     cout << "Context error: " << errinfo << endl; 
    }, nullptr, &error); 


    cl_float* mem = (cl_float*) malloc(sizeof(cl_float)*64); 
    for(int i=0; i<64; i++) 
     mem[i] = i; 

    cl_mem aBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(cl_float) * (64), mem, &error); 
    cl_mem bBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(cl_float) * (64), mem, &error); 

    free(mem); 

    cl_program program; 
    string src = "__kernel void SAXPY(__global float* x, __global float* y, float a){" 
        "size_t i=get_global_id(0);" 
        "y[i]=a*x[i];" 
       "}"; 

    const char* sources[] = {src.c_str()}; 
    const size_t lens[] = {src.length()}; 

    program = clCreateProgramWithSource(context, 1, sources, lens, &error); 
    clBuildProgram(program, 1, &deviceIds[0], nullptr, nullptr, nullptr); 

    cl_kernel kernel1 = clCreateKernel(program, "SAXPY", &error); 
    clSetKernelArg(kernel1, 0, sizeof(cl_mem), &aBuffer); 
    clSetKernelArg(kernel1, 1, sizeof(cl_mem), &bBuffer); 
    static const float two = 2.0f; 
    clSetKernelArg(kernel1, 2, sizeof(float),&two); 

    const size_t globalWorkSize [] = {64,0,0}; 
    cl_command_queue queue = clCreateCommandQueue(context, deviceIds[0], 0, nullptr); 

    clEnqueueNDRangeKernel(queue, kernel1, 1, nullptr, globalWorkSize, nullptr, 0, nullptr, nullptr); 

    float done[64]; 
    clEnqueueReadBuffer(queue, bBuffer, CL_TRUE, 0, sizeof(float)*64, done, 0, nullptr, nullptr); 
    for (int a = 0; a < 64; a++) 
     cout << done[a] << endl; 

    clReleaseContext(context); 
    return 0; 
}