On Thu, Oct 04, 2012 at 06:42:47PM +0200, Steven Bosscher wrote:
> On Thu, Oct 4, 2012 at 6:33 PM, Jakub Jelinek wrote:
> > This patch fixes a few FAILs in the ix86 guality testsuite (mainly -Os),
> > by better disambiguating sp based VALUEs (which usually have no MEM_EXPR
> > and thus the alias Oracle can't be used for them) from frame pointer
> > based ones or global vars.
>
> Does this also help for some of the var-tracking compile-time-hog
> bugs? ISTR PR53958 was one with very long chains of sp based values...
Very unlikely. That PR has been reported against 4.7, which doesn't have
clobber_overlapping_mems at all, the patch could only affect compile times
if there were huge chains of sp based values and find_base_term appeared
anywhere significantly in the profiles.
That PR seems to create so huge VTA hash tables that var-tracking gives up,
unfortunately the current give up heuristics doesn't estimate the compile
time and compile memory needed accurately enough and lowering the limit
(which is a param)'s default would on the other side result in sane sized
programs to trigger no-VTA way too often.
Jakub
