And Sxy has a point, and I now remember this is why we did CAND/CIOR etc. I would say, this is a case where your change is not always a win. Any suggestion/pointer? Sun
On Mon, Jun 6, 2011 at 12:10 PM, Chandrasekhar Murthy <mur...@sgi.com> wrote: > Would replacing CAND/CIOR by LAND/LIOR work for the cases you are looking > at. > > > > Murthy > > > > ________________________________ > > From: shuxin yang [mailto:shuxin.ope...@gmail.com] > Sent: Monday, June 06, 2011 10:32 AM > To: Chandrasekhar Murthy > Cc: Sun Chan; open64-devel > > Subject: Re: [Open64-devel] patch to enable more if-conversion > > > > Hi: > (Just a general question about conditional and/or, not specific to > open64.) > > More often than not, I find it is more convenient to directly deal with > cand/cor at the level higher than CG. I don't remember all these situations > I ran into. Here is some examples: > > e.g.1: merge the if-then-else constructs > if (c1||c1) s1 then s2; > .. > if (c1||c2) s3 then s4; > > e.g.2 PRE complicate cand/cor > if (cand-cor) {..} > > > .. some stmts here > > if (cand-cor) {} > > e.g3 reorder the sub-expr in the cand/cor, suppose it is safe to to that > c1 || c2 || c3 => c3 || c2 || c1 > > I believe sometimes lowered cand-cor could be more convenient. However, > dealing with CFG is far more difficult than expression. If I were compiler > designer, I would keep cand/cor all the way down to the point right before > CG. What is your take? > > Thanks > Shuxin > > On Mon, Jun 6, 2011 at 8:42 AM, Chandrasekhar Murthy <mur...@sgi.com> wrote: > > As some of the files in lno (access_vector.cxx, array_bounds.cxx, etc..) are > processing these nodes, I think the lowering is done after LNO. > > Murthy > > -----Original Message----- > From: Sun Chan [mailto:sun.c...@gmail.com] > > Sent: Monday, June 06, 2011 8:20 AM > To: Chandrasekhar Murthy > Cc: David Coakley; open64-devel > Subject: Re: [Open64-devel] patch to enable more if-conversion > > Thx Murthy, does this mean we should keep these nodes till after LNO? > David, > Is your if-conv fix at mainopt? Or is it at preopt? From this > discussion so far, this should be done in mainopt, i.e. the lowering > into your new form ought to be done at mainopt, after LNO. > Sun > > On Mon, Jun 6, 2011 at 7:24 AM, Chandrasekhar Murthy <mur...@sgi.com> wrote: >> && and || were initially handled in the frontend. >> CAND/CIOR were added later in order to move the lowering to VHO. >> The lowering was deferred to later so that LNO could also process these >> nodes. >> >> Murthy >> >> -----Original Message----- >> From: Sun Chan [mailto:sun.c...@gmail.com] >> Sent: Saturday, June 04, 2011 9:54 PM >> To: David Coakley >> Cc: open64-devel >> Subject: Re: [Open64-devel] patch to enable more if-conversion >> >> I forgot why we did cand anymore. OTOH, I do remember lowering that >> with profile feedback info is tricky to get that right. Can you double >> check that you are not breaking profile info and consistency check >> with your changes? >> Sun >> >> On Fri, Jun 3, 2011 at 1:53 PM, David Coakley <dcoak...@gmail.com> wrote: >>> Could a gatekeeper please review the attached change that enables more >>> if-conversion? >>> >>> Here is the proposed log message: >>> >>> Convert: >>> if (a && b) >>> x = ... >>> To: >>> if (a) >>> if (b) >>> x = ... >>> To enable more if-conversion. >>> >>> Without this transformation, WHIRL-lowering creates control flow >>> that WOPT-if-conversion can't handle. With this transformation, >>> this now looks like any other if-conversion opportunity to WOPT. >>> So this transformation enables more if-conversion (i.e. replace >>> branch with conditional move). No if-conversion code was touched - >>> only lowering code. >>> >>> >>> >>> >>> I have also attached a small test program that demonstrates when the >>> transformation occurs. Thanks, >>> >>> -David Coakley / AMD Open Source Compiler Engineering >>> >>> >>> ------------------------------------------------------------------------------ >>> Simplify data backup and recovery for your virtual environment with >>> vRanger. >>> Installation's a snap, and flexible recovery options mean your data is >>> safe, >>> secure and there when you need it. Discover what all the cheering's >>> about. >>> Get your free trial download today. >>> http://p.sf.net/sfu/quest-dev2dev2 >>> _______________________________________________ >>> Open64-devel mailing list >>> Open64-devel@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/open64-devel >>> >>> >> >> >> ------------------------------------------------------------------------------ >> Simplify data backup and recovery for your virtual environment with >> vRanger. >> Installation's a snap, and flexible recovery options mean your data is >> safe, >> secure and there when you need it. Discover what all the cheering's about. >> Get your free trial download today. >> http://p.sf.net/sfu/quest-dev2dev2 >> _______________________________________________ >> Open64-devel mailing list >> Open64-devel@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/open64-devel >> > > ------------------------------------------------------------------------------ > Simplify data backup and recovery for your virtual environment with vRanger. > Installation's a snap, and flexible recovery options mean your data is safe, > secure and there when you need it. Discover what all the cheering's about. > Get your free trial download today. > http://p.sf.net/sfu/quest-dev2dev2 > _______________________________________________ > Open64-devel mailing list > Open64-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/open64-devel > > ------------------------------------------------------------------------------ EditLive Enterprise is the world's most technically advanced content authoring tool. Experience the power of Track Changes, Inline Image Editing and ensure content is compliant with Accessibility Checking. http://p.sf.net/sfu/ephox-dev2dev _______________________________________________ Open64-devel mailing list Open64-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/open64-devel
