[5.2.X] TRMM functions do not have correct correspondence in rocBLAS

[emankov]

rocBLAS TRMM functions rocblas_strmm, rocblas_dtrmm, rocblas_ctrmm, rocblas_ztrmm do not match neither cublas TRMM functions, nor cublas TRMM _v2 functions.

For instance:

cublasStrmm:

void CUBLASWINAPI cublasStrmm(char side,
                              char uplo,
                              char transa,
                              char diag,
                              int m,
                              int n,
                              float alpha,
                              const float* A,
                              int lda,
                              float* B,
                              int ldb);

cublasStrmm_v2:

CUBLASAPI cublasStatus_t CUBLASWINAPI cublasStrmm_v2(cublasHandle_t handle,
                                                     cublasSideMode_t side,
                                                     cublasFillMode_t uplo,
                                                     cublasOperation_t trans,
                                                     cublasDiagType_t diag,
                                                     int m,
                                                     int n,
                                                     const float* alpha, /* host or device pointer */
                                                     const float* A,
                                                     int lda,
                                                     const float* B,
                                                     int ldb,
                                                     float* C,
                                                     int ldc);

rocblas_strmm:

ROCBLAS_EXPORT rocblas_status rocblas_strmm(rocblas_handle    handle,
                                            rocblas_side      side,
                                            rocblas_fill      uplo,
                                            rocblas_operation transA,
                                            rocblas_diagonal  diag,
                                            rocblas_int       m,
                                            rocblas_int       n,
                                            const float*      alpha,
                                            const float*      A,
                                            rocblas_int       lda,
                                            float*            B,
                                            rocblas_int       ldb);

The same goes for hipBLAS analogues hipblasStrmm, hipblasDtrmm, hipblasCtrmm, hipblasZtrmm (ROCm/hipBLAS#524).

So, the above 4 rocBLAS and 4 hipBLAS functions are marked as HIP UNSUPPORTED.

[Solution]
As far as rocBLAS doesn't support v1 BLAS functions, populate rocblas TRMM functions with two missing arguments: float* C and int ldc and revise functions' logic.

[daineAMD]

Hi @emankov,

For trmm functions, cuBLAS differs from the traditional BLAS API as cuBLAS includes two extra parameters, C and ldc (cuBLAS trmm doc, lapack doc). rocBLAS has typically followed the BLAS API as close as possible, resulting in this difference.

We have, however, implemented rocblas_xtrmm_outofplace functions which have the 2 extra parameters being requested. This function supports out-of-place and in-place functionality for trmm. We do not plan on changing the current rocblas_xtrmm API as the rocblas_xtrmm_outofplace API will take care of this functionality.

I hope this helps,
Daine

[emankov]

Thank you, @daineAMD!

I've renamed rocblas_xtrmm functions to rocblas_xtrmm_outofplace. Now they are always hipified from cublasXtrmm_v2 and from cublasXtrmm if cublas_v2.h is included in the source CUDA being hipified.

So closing this issue as ERRONEOUS.