Question

我正在尝试实现INTEL MKL的dft api以在CentOS 7上测试其速度。编译成功，但是运行时出现分段错误。但是，该代码已在Windows上的Visual Studio 2017中成功运行。 Windows上的结果如下：result on windows

#include <opencv2/core/core.hpp>
#include <opencv2/opencv.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <execinfo.h>
#include <cxxabi.h>
#include <omp.h>
#include <complex>
#include "mkl_dfti.h"

int main() {
    MKL_LONG len[2] = { 1080, 961 }, status;
    float x_in[1080][1920];
    DFTI_DESCRIPTOR_HANDLE fft;
    status = DftiCreateDescriptor(&fft, DFTI_SINGLE, DFTI_REAL, 2, len);
    status = DftiSetValue(fft, DFTI_PLACEMENT, DFTI_NOT_INPLACE);
    status = DftiCommitDescriptor(fft);

    //float x[1080* 2000];
    std::complex<float> x_out[1080][961];
    for (int i = 0; i < 10; i++) {
        double totalcputime = (double)cv::getTickCount();
        //std::cout << status << std::endl;
        status = DftiComputeForward(fft, x_in, x_out);
        //std::cout << status << std::endl;
        totalcputime = ((double)cv::getTickCount() - totalcputime) / cv::getTickFrequency();
        std::cout << "MKL-DFT Time: " << totalcputime << std::endl;
    }
    cv::Mat sizedimage = cv::Mat::zeros(1080, 1920, CV_32FC1);
    cv::Mat opencvtransform = cv::Mat(1080, 1920 / 2 + 1, CV_32FC1);
    for (int i = 0; i < 10; i++) {
        double totalcputime = (double)cv::getTickCount();
        cv::dft(sizedimage, opencvtransform);
        totalcputime = ((double)cv::getTickCount() - totalcputime) / cv::getTickFrequency();
        std::cout << "opencv-DFT Time: " << totalcputime << std::endl;
    }

    return 0;
}

我已经使用GDB调试了我的代码，它为我提供了以下信息：

Program received signal SIGSEGV, Segmentation fault.
0x00000000004012b8 in main () at comparison.cpp:25
25              status = DftiCreateDescriptor(&fft, DFTI_SINGLE, DFTI_REAL, 2, len);

以下参数成功编译了文件：

g++ comparison.cpp `pkg-config opencv --cflags --libs` -lmkl_rt -g

有人知道这个错误的原因吗？

Answer 1

您能检查MKL 2019 u4的问题吗？

我略微重新设计了代码，以通过删除opencv条目并动态分配输入/输出数组来检查最新的mkl 2019是否存在一些问题

#include <iostream>
#include <stdio.h>
#include <stdlib.h>
//#include <execinfo.h>
//#include <cxxabi.h>
#include <assert.h>
#include <omp.h>
#include <complex>
#include "mkl.h"

#define N1 1080
#define N2 961
#define N3 1920

int main() 
{
//    MKL_LONG len[2] = { 1080, 961 }, status;
    MKL_LONG status;
        MKL_LONG len[2];
        len[0] = N1;
    len[1] = N2;

    //float x_in[1080][1920];
    float* x_in = (float*)mkl_malloc(N1*N3*sizeof(float), 64);       
    assert(NULL != x_in);

    DFTI_DESCRIPTOR_HANDLE fft;
    status = DftiCreateDescriptor(&fft, DFTI_SINGLE, DFTI_REAL, 2, len);
    if (0 != status){       
        std::cout << "\t DftiCreateDescriptor Error : " << status << std::endl;
    }
    status = DftiSetValue(fft, DFTI_PLACEMENT, DFTI_NOT_INPLACE);
    if (0 != status){       
        std::cout << "\t DftiSetValue Error : " << status << std::endl;
    }
    status = DftiCommitDescriptor(fft);
    if (0 != status){       
        std::cout << "\t DftiCommitDescriptor Error : " << status << std::endl;
    }

    double t1,texec;
   // std::complex<float> x_out[1080][961];
    MKL_Complex8* x_out = (MKL_Complex8*)mkl_malloc(N1*N2*sizeof(MKL_Complex8), 64);         

    t1 = dsecnd();
    for (int i = 0; i < 10; i++) {

        t1 = dsecnd();
            status = DftiComputeForward(fft, x_in, x_out);
            if (0 != status){std::cout << "\t DftiComputeForward Error : " << status << std::endl;}
            texec = dsecnd() - t1;
            std::cout << "MKL-DFT Time: " << texec << std::endl;
        }

    status = DftiFreeDescriptor(&fft);
    if (0 != status){       
        std::cout << "\t DftiFreeDescriptor Error : " << status << std::endl;
    }

    return 0;
}

这是我看到的输出：

]$ ./a.out
MKL-DFT Time: 0.00725237
MKL-DFT Time: 0.00381843
MKL-DFT Time: 0.00362679
MKL-DFT Time: 0.0021284
MKL-DFT Time: 0.00221884
MKL-DFT Time: 0.00215556
MKL-DFT Time: 0.00211133
MKL-DFT Time: 0.002133
MKL-DFT Time: 0.00212184
MKL-DFT Time: 0.00215306

Answer 2

您还可以尝试设置MKL_VERBOSE模式以查看所有所需的运行时详细信息：出口MKL_VERBOSE = 1 这是fft调用的mkl详细输出：

 ./a.out
MKL_VERBOSE Intel(R) MKL 2019.0 Update 4 Product build 20190411 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions (Intel(R) AVX) enabled processors, Lnx 2.80GHz intel_thread
.........
MKL_VERBOSE FFT(srfo1080:961:961x961:1:1,pack:ccs,tLim:20,desc:0x1b4df40) 3.83ms CNR:OFF Dyn:1 FastMM:1 TID:0  NThr:20
MKL-DFT Time: 0.0038483

在Linux上实施Intel MKL的DFT时出现“分段故障”

2 个答案: