This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [trans-mem] verify_ssa failed with noreturn attribute function
- From: Richard Henderson <rth at redhat dot com>
- To: Aldy Hernandez <aldyh at redhat dot com>
- Cc: Pascal Felber <pascal dot felber at unine dot ch>, MARLIER Patrick <patrick dot marlier at unine dot ch>, gcc-patches at gcc dot gnu dot org
- Date: Thu, 17 Jun 2010 08:50:56 -0700
- Subject: Re: [trans-mem] verify_ssa failed with noreturn attribute function
- References: <4C0CD1FF.3080601@unine.ch> <20100608134559.GA3066@redhat.com> <4C0E690C.8010005@redhat.com> <2AD1C955-250B-44D6-87CD-0B1B72CAB7E6@unine.ch> <20100609202831.GA12470@redhat.com> <20100610111211.GB15138@redhat.com>
On 06/10/2010 04:12 AM, Aldy Hernandez wrote:
>> I don't understand why we only update the SSA on safe function calls in
>> ipa_tm_insert_gettmclone_call(), especially since
>> ipa_tm_insert_gettmclone_call() calls update_stmt() at the end, which
>> may possibly dirty the SSA web. IMO, we should always return true,
>> which is the approach I take below.
I don't recall the logic behind only updating if safe. It does
look wrong now though.
>> Another option is returning need_ssa_update_p(), or better yet, just
>> call update_ssa() only if need_ssa_update_p() returns true and get rid
>> of keeping track of need_ssa_rename altogether.
Those do look like attractive options.
>> * trans-mem.c (ipa_tm_insert_gettmclone_call): Always return true.
Ok.
r~