This is maybe not the right place for a bug report, sorry for this.
But I think there is a minor bug, related to cublas_dsymm, in the
fortran wrapper (NON-THUNKING interface) provided by the
cudatoolkit (i checked version 2.3 and 2.2 ).
file fortran.c.
5104 cublasDsymm (*side, *uplo, *m, *m, *alpha, A, *lda, B, *ldb, *beta, C,
5105 *ldc);
should be:
5104 cublasDsymm (*side, *uplo, *m, *n, *alpha, A, *lda, B, *ldb, *beta, C,
5105 *ldc);
so the 4-th parameter should be *n and not *m.
Loriano Storchi