Const correctness of dgbtrs and friends in mkl_lapack.h?

[Rhys_Ulerich]

Any reason why the C interface to LAPACK's dgbtrf doesn't provide a const correct argument for ab and ipiv?

void dgbtrs( char* trans, MKL_INT* n, MKL_INT* kl, MKL_INT* ku, MKL_INT* nrhs, double* ab, MKL_INT* ldab, MKL_INT* ipiv, double* b, MKL_INT* ldb, MKL_INT* info );

It doesn't seem like the factored matrix or its pivots should be touched at all during the solve process.

[Artem_V_Intel]

Quoting - Rhys Ulerich

Any reason why the C interface to LAPACK's dgbtrf doesn't provide a const correct argument for ab and ipiv?

void dgbtrs( char* trans, MKL_INT* n, MKL_INT* kl, MKL_INT* ku, MKL_INT* nrhs, double* ab, MKL_INT* ldab, MKL_INT* ipiv, double* b, MKL_INT* ldb, MKL_INT* info );

It doesn't seem like the factored matrix or its pivots should be touched at all during the solve process.

Hello,

It is guaranteed that the content of arguments that used only as input parameters will not be changed during function execution. 'const' modifier is redundant here.

Thanks,
Art

[Vladimir_Koldakov__I · Accepted solution]

Hi Rhys,

Your question is not entirely clear to me. So I'll try to answer in general.

Arrays ab and ipiv are as returned by dgbtrf on the entry dbgtrs.

Arrays ab and ipiv are not changed within dgbtrs. The same is for n, kl, ku, nrhs, ldab and ldb arguments. All of them have not "const" specifier. You can't find "const" anywhere in the mkl_lapack.h. Of course this is not good enough and may be fixed in the further MKLreleases. In any case it has not impact on the functionality of the interfaces. See also CLAPACK thread: http://software.intel.com/en-us/forums/showthread.php?t=61234

Thanks,

Vladimir

[Rhys_Ulerich]

Quoting - Vladimir Koldakov (Intel)

Arrays ab and ipiv are not changed within dgbtrs. The same is for n, kl, ku, nrhs, ldab and ldb arguments. All of them have not "const" specifier. You can't find "const" anywhere in the mkl_lapack.h. Of course this is not good enough and may be fixed in the further MKLreleases. In any case it has not impact on the functionality of the interfaces. See also CLAPACK thread: http://software.intel.com/en-us/forums/showthread.php?t=61234

Thanks Vladimir, that helps.

[Rhys_Ulerich]

Quoting - Artem Vorobiev (Intel)

It is guaranteed that the content of arguments that used only as input parameters will not be changed during function execution. 'const' modifier is redundant here.

Thanks for the response Art. The guarentee that the input parameters won't be changed is exactly what const means. It's not redundant because including const in the interface would cause the compiler to enforce this semantic.

[mfactor]

The need to pass via reference is a Fortran/C one, but I agree that they should be const.