impl. transpose matvec, matmul

[t-hishinuma]

I want to give modern and intuitive transposition information. But I have no idea how to implement it easily.

First, we create the following function as a prototype

matmul(A,B,C) // C=AB
matmul_TNN(A, B, C); // C=A^TB
matvec(A,x,y); // y = Ax
matvec_T(A, x, y); // y=A^Tx

This interface is not beautiful. However, it has the following advantages

• It does not affect other functions.
• Easy to trace with logger
• Simple to implement FFI in the future.
• When beautiful ideas appear in the future, these functions can be implemented wrapping it.

[t-hishinuma]

matmul_TNN(A, B, C); // C=A^TB

Transpose C is not implemented. Therefore, change as follows.

matmul_TN(A, B, C); // C=A^T B^N

[t-hishinuma]

cusparse SpMM has bug? It need to take a lot of time..

I impl. GEMV and SpMV at 0.16.0 I impl. GEMM and SpMM at 0.16.1

this isssue is separlated: #115 #114