On 07/13/2009 03:35 PM, Mark Wielaard wrote:
> This is really a shame, gcc is really a lot slower compiling with -O2.
> It would be good to find some other way to make pass 4 quicker by
> default.

Another approach mentioned on IRC is trying to simplify the code that we
present to GCC.

I took a stab at consolidating the probe locking code, patch attached.
It makes each probe body contain just a static array of locks needed,
and then call to a single function that actually loops over the locks.

This will only be a win for scripts with many distinct probe bodies,
like topsys.stp.  For that script, I see pass-4 time go from 5 seconds
to 4.4 seconds.  In runtime, it's hard to get consistent measurement,
but the average probe time is about 2-3% slower.

So, is that overhead worth 1/2 second compile time?  Can anyone see ways
to improve this patch, or suggest other areas that could benefit from focus?

Josh