gcc-patches@gcc.gnu.org
[Top] [All Lists]

Re: [PATCH][Revised3] Fix PR41313 with dual approach

Subject: Re: [PATCH][Revised3] Fix PR41313 with dual approach
From: Jakub Jelinek
Date: Thu, 8 Oct 2009 18:35:31 +0200
On Thu, Oct 08, 2009 at 09:26:09AM -0700, Mike Stump wrote:
> I think it is wrong to use just flag_profile_use here.  That alters  
> codegen and there is an invariant that codegen can't be altered between 
> generation and use.  So, at a minimum, the profile generate flag would 
> have to be used as well.  But, it isn't clear to me why these fail.  I 
> think there is some other more basic issue going on, and that should be 
> the flag used, not the profile flag.  Conservatively, absent the correct 
> flag to use, true would be safer to use, that would just always turn off 
> partitioning.

FAIL: gcc.dg/tree-prof/bb-reorg.c compilation,  -fprofile-use -D_PROFILE_USE
UNRESOLVED: gcc.dg/tree-prof/bb-reorg.c execution,    -fprofile-use 
-D_PROFILE_USE
FAIL: gcc.dg/tree-prof/pr34999.c compilation,  -fprofile-use -D_PROFILE_USE
UNRESOLVED: gcc.dg/tree-prof/pr34999.c execution,    -fprofile-use 
-D_PROFILE_USE

has been failing on many targets until my 4.5 changes, those aren't
regressions (at least not recent).

BTW, the patch that was posted violates code formatting rules in many ways.

        Jakub

<Prev in Thread] Current Thread [Next in Thread>