Devin Matthews
Devin Matthews
Correct me if I'm wrong, but isn't `gemmt` just `syr2k` with some stuff deleted?
@rvdg Yes, and that is an operation like that I would very much like to see as well. But from an implementation standpoint, assuming `gemmt` as a given, starting from...
@gmargari could you create a separate issue for this?
I think, of the currently-available configs, that ThunderX2 should perform best on N2. The SVE kernels are tuned for 256+ bit so I think you really want a neon kernel....
@fgvanzee yes. The updates in question are the A -= L D L^T updates for A11, A21, and A22 (depending on variant). @poulson I implemented C = A D B...
I suggest making any change in behavior configurable. Maybe a `BLIS_CBLAS_STOP_ON_ERROR` env. variable?
@fgvanzee MKL and OpenBLAS do not normally halt in `xerbla`. I can't say why not halting is important behavior as a call to xerbla indicates a serious error but that...
> @devinamatthews But we're not talking about xerbla(). We're talking about cblas_xerbla(). Sure the latter should call the former in a sane world? But seriously it should be pretty easy...
@grisuthedragon the calling program could just call `setenv()` before the first error.
@jeffhammond `sched_yield` is too heavyweight and not portable enough to be used all the time. I will update #82 with a general framework for config-specific behavior and then we can...