uvos
uvos
Hi @daineAMD Thank you for the detailed comment on this matter and for: `The request for rocblas_hgemv() has also been noted and I can discuss with the team about whether...
Hi @daineAMD, Thank you for your examples, this has been useful in determining when to use gemv in my code to work around this issue an when not. Since this...
Hi @daineAMD, Thank you for your quick update. > In the meantime, I've mocked up some changes to potentially allow users to opt-in to using gemv kernels from `rocblas_gemm_ex()` calls...
Hi @daineAMD, Given the elapsed time, i would like to inquire as to if a direction forward with this has been decided internally.
Hi @daineAMD, That sounds great, thank you for working on resolving this! If nothing speaks against it may make sense to have such proposals public if possible somewhere (maybe a...
I'd like to stress that this issue has been pending since _March of 2022_ and causes these (really quite common, since using gemmex for fp16 gemv is the fastest pattern...
@daineAMD unfortionatly i can __not__ confirm that this issue is solved by the commit mentioned: ``` % rocblas-bench --version rocBLAS info: client (OPENMP) reduced omp_set_num_threads to 46 rocBLAS info: Using...
Ah i see, yes i missed that. Thank you!
This is still very mutch an issue and should be reopened
yeah i second this, its pretty silly having to use hipify at compiletime on a first party lib.