problem with cblas_cgemm3m when MKL_Domain_Set_Num_Threads(1, MKL_DOMAIN_BLAS)

[yang_w_1]

dear,

     I'm using cblas_cgemm3m to calculate matrix multiply matrix. when i use default mkl_domain_num_threads, api return correct result .

but when i use MKL_Domain_Set_Num_Threads(1, MKL_DOMAIN_BLAS),the result is wrong.

what's wrong with me?

mkl version i using is 10.3.7.

code below:

    int iRows = 24;
    int iCols = 25;
    int iLength = iRows*iCols;
    std::shared_ptr<std::complex<float> > pDataSrc(new std::complex<float>[iLength],std::default_delete<std::complex<float>[]>());
    std::shared_ptr<std::complex<float> > pDataRes(new std::complex<float>[iLength],std::default_delete<std::complex<float>[]>());
    //MKL_Domain_Set_Num_Threads(1, MKL_DOMAIN_BLAS);
    ReadBin(pDataSrc.get(),iLength*sizeof(std::complex<float>),"CORR_1.bin");
    MKL_Complex8 alpha;alpha.real = 1.0f;alpha.imag = 0.0f;
    MKL_Complex8 beta;beta.real = 0.0f;beta.imag = 0.0f;
    cblas_cgemm3m(CblasRowMajor, CblasNoTrans, CblasConjTrans,
        iRows,iRows,iCols,&alpha,pDataSrc.get(),iCols,pDataSrc.get(),iCols,&beta,pDataRes.get(),iRows);

[Gennady_F_Intel]

pls check the problem with the latest 11.3 version of MKL( you may take the eval version for 30 days). The similar symptoms were fixed in the past versions.