A recent patch increased GCC's memory consumption in some cases!
Kenneth Zadeck
zadeck@naturalbridge.com
Tue Apr 17 17:02:00 GMT 2007
Paolo Bonzini wrote:
> Jan Hubicka wrote:
>> Hi,
>> this is memory report for current tree-ssa branch relative to mainline
>> from this morning.
>
> Kenny, I think we should disable UR/LIVE (and just use LR) at -O1.
>
This will require the we change the references for DF_LR_* to
df_get_live_* for the passes that run at -01. Asside from that and the
obvious changes to pass_df_initialize_opt, this should not be hard.
> (Besides that, -O0 is still a little too memory hungry.) :-( :-(
>
> The only testcase which is "out of control" at -O2/-O3 insn-attrtab.c
> and some of them are now consuming *less* memory.
>
I thought that you were going to try to get rid of the ur bitvectors by
combining them with the live bitvectors. This would improve the o2/o3
space and might even be a little bit faster.
> Opinions?
>
> Paolo
More information about the Gcc-regression
mailing list