If you are using the profiling mechanism in the compiler or if
you are instrumenting and reading in at the same place, the ID's
will be identical. Note that the profiling mechanism of the compiler
is on the IR, so it might not corresponding to a BB (e.g. at High
Whirl, one Whirl node, say, CAND, might represent multiple BBs)
If you are profiling the binary, you need to do -O0 to gather
your profiling and read that back right after Whirl generation,
there you have best chance to match the BB_id.
Sun
> -----Original Message-----
> From: Steve Wayne Haga [mailto:stevhaga@xxxxxxxxxxxx]
> Sent: Monday, July 02, 2001 10:12 AM
> To: pro64-support@xxxxxxxxxxx
> Cc: Steve Wayne Haga; Alan P. Levicki
> Subject: Recovering intermediate representation
>
>
> I am performing profiling, and after generating the profiling
> statistics,
> wish to rerun the compiler and use those statistics. It
> seems unsafe to
> just assume that all of the Basic Blocks will retain the same
> BB_id numbers
> when rerun. Is there a way to restart compilation from the
> intermediate
> representation, rather than from the original C file, so that I can
> guarentee the same numbering?
>
> Sincerely,
> Steve Haga
>
>
>
|