Re: -fcse-follow-jumps and -fcse-skip-blocks and GCSE

[Daniel Berlin <dberlin at redhat dot com>]

kenner@vlsi1.ultra.nyu.edu (Richard Kenner) writes:

> When I added those two options to CSE, we knew they shouldn't be needed
> once a global CSE was written.  Well, we have one now and they are still
> there.  Are they needed?

Depends on how you look at it. They are still used in the non-GCSE
CSE passes. (look for cse_main in toplev.c)
> 
> Can somebody volunteer to do some benchmarking with those options on
> and off (note they default on in -O2) and see if they really help?

Funny you mention.
I recently, wanting to get a buttload of statistics for the new
register allocator, hooked up gdbm  to gcc, and had it store a lot of
these types of statistics that were dumped to files before, in a
database on a per-function basis, along with the RTL for that
function.

Made it a ton easier to do analysis on it, as i just used simple
python scripts.

I'll check out the stats on these options, and post the results.

> 
> If so, we need to find out why since they are expensive and should
> be deleted.