Parallel Computing

gsl library optimization error

Hi all,

as a follow up to my previous post on the gsl library coredumping on the Xeon Phi there is good news and bad news.

The good news is: with icc v15 the coredumps are gone

The bad news is: there are other vectorization errors that seem to occur only when -mmic is used.

Consider the following program (distilled from the gsl-1.16 source code):

C++ Compiler 15.0, initializer syntax issue

The following example code produces the compiler error "c:/temp/test.cpp(12): error #453: protected function "C1::C1(int)" (declared at line 4) is not accessible through a "C1" pointer or object" using command line "icpc.exe -std=c++11 c:\temp\test.cpp".

class C1
{
protected:
    C1(int a)
    {
    }
};

class C2 : public C1
{
public:
    C2() : C1{2}
    {
    }
}; 

It compiles if we change initializer syntax in line

    C2() : C1{2}

to constructor syntax

C++ Compiler 15.0, type deduction issue

The following Example code produces the compiler error "c:/temp/test.cpp(15): error: no instance of function template "f2" matches the argument list" using command line "icpc.exe -std=c++11 c:\temp\test.cpp".

template <typename T, typename U>
auto f1(U const& u) -> decltype(T())
{
    return (T()+1.1) * u;
}

template <typename T, typename U>
auto f2(U const& u) -> decltype(f1<T>(u))
{
    return f1<T>(u);
}

int main()
{
    auto dummy = f2<double>(2);
    return 0;
}

 

mixing different icpc version on MIC

Hello,

I am doing the porting of a part (some libraries) of a code to the Xeon Phi architecture. I have some questions concerning compatibility between different ICPC versions.

May be this question should have been posted to the icpc forum ?

To reach the maximum performance, we plan to use icpc V 15.0.0.90 for the Xeon Phi parts of the code.

Finaly, some of the libraries used by the code will come compiled with ICPC V13.0.1 and others (the MIC ones) with icpc 15.0.0.90.

 

My questions are :

Intel MKL's call from mex-files are crashing MATLAB

Hello,

this is an issue already discussed but I have not been able to find a solution. I'm using Intel MKL 10.2 and MATLAB R2011b. My C function, which has many MKL BLAS and MKL Lapack calls, runs well when called from a C program.

When called from a mex file, MATLAB crashs. If I comment all the calls to mkl, it works (but output array are all zeros, as I don't perform calculations), so it is the call to Intel mkl that crashs matlab. I downlaoded yesterday the lastest version of Intel MKL but it did'nt solve the problem.

Subscribe to Parallel Computing