Small error in KokkosBatched_Gemm_Serial_Imp.hpp with SerialGemm<Trans::Transpose,*,*>

[kliegeois]

It seems that in kokkos-kernels/src/batched/KokkosBatched_Gemm_Serial_Impl.hpp, when the matrix A is transposed, the code is currently using A.dimension_1() to get the number k instead of A.dimension_0().

The error seems to be in all the different combinations of the other template arguments (i.e. whether or not B is transposed and for CompactMKL, Blocked, and Unblocked).

@kyungjoo-kim

[kyungjoo-kim]

@kliegeois Thanks. You are right. The code is tested and used only for square blocks. So, the error does not pop up so far. I will fix it soon.

[srajama1]

Ping

[kyungjoo-kim]

#258 fixed this. done.

[srajama1]

Reopening it, and adding indevelop tag. We will close it when we do the master update. Otherwise it is hard to keep track of what are the fixes we did.