This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] alternative hirate for builtin_expert
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: Dehao Chen <dehao at google dot com>
- Cc: Jan Hubicka <hubicka at ucw dot cz>, Rong Xu <xur at google dot com>, Ramana Radhakrishnan <ramrad01 at arm dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>, David Li <davidxl at google dot com>
- Date: Fri, 4 Oct 2013 23:23:25 +0200
- Subject: Re: [PATCH] alternative hirate for builtin_expert
- Authentication-results: sourceware.org; auth=none
- References: <CAF1bQ=Q_eiCz8AYkU-c=Xu7quQ0vy8eP=pi9QZt+QOF4vicnsg at mail dot gmail dot com> <20131002160827 dot GB7181 at kam dot mff dot cuni dot cz> <CAF1bQ=Snvzzmm=AEr-crrKhopRugKbcExH+wFCyCcgZZHz+Uow at mail dot gmail dot com> <20131002213102 dot GE7181 at kam dot mff dot cuni dot cz> <CAF1bQ=R+oNesEedf_8-oVtQBgWSZMvEegtY-Z_LgsMfVj4qpqg at mail dot gmail dot com> <524EBD0D dot 9060608 at arm dot com> <CAF1bQ=Rg2pxmH5Cwakei=Z_hmGfXvpKgN1RxUosBtKWGzdwgmg at mail dot gmail dot com> <CAO2gOZUEuZmWSMSP2fVo=caQgLbJ3c+kCw-5ya6UKEuyYVKOfw at mail dot gmail dot com> <20131004185416 dot GC7530 at atrey dot karlin dot mff dot cuni dot cz> <CAO2gOZXigCZ-JeziCiJbfgbA7Ynx-odvHU7C4N23jWA74qJ75Q at mail dot gmail dot com>
> On Fri, Oct 4, 2013 at 11:54 AM, Jan Hubicka <firstname.lastname@example.org> wrote:
> >> I looked at this problem. Bug updated
> >> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58619
> >> This is a bug when updating block during tree-inline. Basically, it is
> >> legal for *n to be NULL. E.g. When gimple_block(id->gimple_call) is
> >> NULL, remap_blocks_to_null will be called to set *n to NULL.
> > The NULL in gimple_block (gimple_call) comes from the call introduced by ipa-split?
> That is correct.
> > I remember that ipa-split used to try to put the call into block since we was ICEing
> > in similar ways previously, too. Perhaps this has changed with new BLOCK representation?
> The new BLOCK representation does not change this. I think it makes
> sense to leave the block of newly introduced call_stmt as NULL because
> when it's inlined back, we don't want to add additional block layers.
You are right, it may be result of Jakub's changes in the area (to improve debug info
after inlining back). I guess the patch makes sense then.
> > Honza