You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
That would be very nice and quite easy. The solution I thought for this (which is still interesting for profiling) is to create a custom mlir::Location that takes a Julia callstack snapshot without actually solving the caller symbols (until requested).
Also, with this the Ops.with_debug would be become useless and can be removed. It just calls stacktrace and converts each StackFrame into a MLIR.IR.Location which is extremely slow.
with_debug still has use, since that gets the full stacktrace and not just the line which called it (e.g. we would now see +(TracedRArray, TracedRArray) but not who calls +)
Currently we get location info in the Ops file. This will let us populate the location info from where the corresponding
Ops.<...>
is being made.Essentially
gets converted into
with the location info of the current line automatically populated.
The text was updated successfully, but these errors were encountered: