[tuning] Interface for Fin #3330
Conversation
averinevg
requested review from
alexandraBara,
DrizztDoUrden,
cderb and
bghimireamd
|
|
||
| protected: | ||
| ConvSolver(const miopen::solver::SolverBase* solver_base, | ||
| uint64_t solver_id, |
There was a problem hiding this comment.
[Suggestion] I would recommend using solver::Id instead of numeric ids. It would be hassle free to create this from either number or string.
|
@averinevg please resolve the conflicts and a few hip tidy issue. Thanks. |
| template <> | ||
| const auto& GetSolversInfo<ConvSolverInfo>() | ||
| { | ||
| static const std::unordered_map<std::string, ConvSolverInfo> solver_info = { |
There was a problem hiding this comment.
Wow, what a list! Is this generated, or maintained by hand? There are likely to be many additions coming with MX data types. Seems like it wouldn't be too hard to import it from a file that would be easier for external tools to interact with (e.g. json, csv).
In either case, it would be nice to have placeholders for the missing ordinals. That way we can know at a glance if an algo was deleted or otherwise expected to be missing. Otherwise, it looks like something may have been missed.
There was a problem hiding this comment.
This list is filled manually. It contains all the solvers for the specified primitives currently available in the library. This is needed to check that the interface provides correct data. However, we don't need such a large list; a dozen solvers would be sufficient. There is also no need to update it when new solvers are added.
This PR adds interface for
Fin