<div dir="ltr"><br><div class="gmail_extra"><br><br><div class="gmail_quote">On Wed, Jun 25, 2014 at 8:41 AM, Mark Heffernan <span dir="ltr"><<a href="mailto:meheff@google.com" target="_blank">meheff@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="">On Wed, Jun 25, 2014 at 8:30 AM, Eli Bendersky <<a href="mailto:eliben@google.com">eliben@google.com</a>> wrote:<br>
> test/Bitcode/Output/upgrade-loop-metadata.ll.script: line 1:<br>
> /media/SSD1/llvm/llvm_svn_rw/test/Bitcode/upgrade-loop-metadata.ll.bc: No<br>
> such file or directory<br>
> FileCheck error: '-' is empty.<br>
<br>
</div>Hmm. Not sure how to include a binary file in an svn diff, so the<br>
file is attached.<br>
<span class="HOEnZb"><font color="#888888"><br></font></span></blockquote><div><br></div><div>LLVM r211710 & r211711. Clang r211712</div><div><br></div><div>Eli</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<span class="HOEnZb"><font color="#888888">
Mark<br>
</font></span><div class="HOEnZb"><div class="h5"><br>
><br>
> --<br>
><br>
> ********************<br>
> Testing Time: 74.68s<br>
> ********************<br>
> Failing Tests (1):<br>
> LLVM :: Bitcode/upgrade-loop-metadata.ll<br>
><br>
><br>
><br>
> Eli<br>
><br>
><br>
>><br>
>> Thanks,<br>
>> Mark<br>
>><br>
>> On Tue, Jun 24, 2014 at 6:40 PM, Tyler Nowicki <<a href="mailto:tnowicki@apple.com">tnowicki@apple.com</a>> wrote:<br>
>> > Hi Mark,<br>
>> ><br>
>> > I also have a weak preference for the llvm.loop.unroll/vectorize option<br>
>> > because I would like to add metadata for interleaving and I think<br>
>> > interleave.enable would look better than interleaver.enable. Otherwise LGTM!<br>
>> ><br>
>> > Tyler<br>
>> ><br>
>> > On Jun 24, 2014, at 4:28 PM, Hal Finkel <<a href="mailto:hfinkel@anl.gov">hfinkel@anl.gov</a>> wrote:<br>
>> ><br>
>> >> ----- Original Message -----<br>
>> >>> From: "Mark Heffernan" <<a href="mailto:meheff@google.com">meheff@google.com</a>><br>
>> >>> To: "Hal Finkel" <<a href="mailto:hfinkel@anl.gov">hfinkel@anl.gov</a>><br>
>> >>> Cc: <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a>, "cfe-commits" <<a href="mailto:cfe-commits@cs.uiuc.edu">cfe-commits@cs.uiuc.edu</a>>,<br>
>> >>> "Pekka Jääskeläinen"<br>
>> >>> <<a href="mailto:pekka.jaaskelainen@tut.fi">pekka.jaaskelainen@tut.fi</a>><br>
>> >>> Sent: Tuesday, June 24, 2014 4:18:30 PM<br>
>> >>> Subject: Re: [PATCH] Change loop unroll and vectorizer metadata prefix<br>
>> >>> to 'llvm.loop.'<br>
>> >>><br>
>> >>> The attached patch includes a note in ReleaseNotes.rst and code to<br>
>> >>> autoupgrade llvm.vectorizer.* metadata strings to<br>
>> >>> llvm.loop.vectorizer.* when reading assembly and bitcode. PTAL.<br>
>> >><br>
>> >> Okay, thanks. That's better. However, I just noticed that the proposed<br>
>> >> mapping:<br>
>> >><br>
>> >>> llvm.vectorizer.* => llvm.loop.vectorizer.*<br>
>> >>> llvm.loopunroll.* => llvm.loop.unroll.*<br>
>> >><br>
>> >> is not grammatically consistent. I think that we should have<br>
>> >> ("llvm.loop.vectorize" and "llvm.loop.unroll") or ("llvm.loop.vectorizer"<br>
>> >> and "llvm.loop.unroller"). I have a weak preference for the first pair, but<br>
>> >> I care more about consistency than the choice.<br>
>> >><br>
>> >> -Hal<br>
>> >><br>
>> >>><br>
>> >>> Thanks!<br>
>> >>> Mark<br>
>> >>><br>
>> >>> On Tue, Jun 24, 2014 at 9:53 AM, Hal Finkel <<a href="mailto:hfinkel@anl.gov">hfinkel@anl.gov</a>> wrote:<br>
>> >>>> ----- Original Message -----<br>
>> >>>>> From: "Mark Heffernan" <<a href="mailto:meheff@google.com">meheff@google.com</a>><br>
>> >>>>> To: "Hal Finkel" <<a href="mailto:hfinkel@anl.gov">hfinkel@anl.gov</a>><br>
>> >>>>> Cc: <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a>, "cfe-commits"<br>
>> >>>>> <<a href="mailto:cfe-commits@cs.uiuc.edu">cfe-commits@cs.uiuc.edu</a>>, "Pekka Jääskeläinen"<br>
>> >>>>> <<a href="mailto:pekka.jaaskelainen@tut.fi">pekka.jaaskelainen@tut.fi</a>><br>
>> >>>>> Sent: Tuesday, June 24, 2014 11:18:33 AM<br>
>> >>>>> Subject: Re: [PATCH] Change loop unroll and vectorizer metadata<br>
>> >>>>> prefix to 'llvm.loop.'<br>
>> >>>>><br>
>> >>>>> Thanks for the comments. I'll add something to the release notes.<br>
>> >>>>><br>
>> >>>>> On Tue, Jun 24, 2014 at 3:52 AM, Hal Finkel <<a href="mailto:hfinkel@anl.gov">hfinkel@anl.gov</a>><br>
>> >>>>> wrote:<br>
>> >>>>>> Moreover, for metadata that was supported by the 3.4 release,<br>
>> >>>>>> autoupgrade support should be added. Please do this before you<br>
>> >>>>>> commit.<br>
>> >>>>><br>
>> >>>>> By autoupgrade do you mean be permissive about what is accepted?<br>
>> >>>>> That<br>
>> >>>>> is, accept llvm.vectorizer.* and llvm.loop.vectorizer.* but only<br>
>> >>>>> emit<br>
>> >>>>> the llvm.loop form from the FE. Or do you mean try to rename the<br>
>> >>>>> metadata as soon as it parsed? I can see how to do the first case<br>
>> >>>>> (be<br>
>> >>>>> permissive) easily, but for the second (rename metadata) do you<br>
>> >>>>> have<br>
>> >>>>> any suggestions for the best place to cut in (I'm fairly new to<br>
>> >>>>> the<br>
>> >>>>> code base)?<br>
>> >>>><br>
>> >>>> No, I mean the second. I believe that most of the current logic for<br>
>> >>>> this is in lib/IR/AutoUpgrade.cpp<br>
>> >>>><br>
>> >>>> -Hal<br>
>> >>>><br>
>> >>>>><br>
>> >>>>> Thanks!<br>
>> >>>>> Mark<br>
>> >>>>><br>
>> >>>>>><br>
>> >>>>>> -Hal<br>
>> >>>>>><br>
>> >>>>>>><br>
>> >>>>>>> On Tue, Jun 24, 2014 at 12:30 AM, Mark Heffernan<br>
>> >>>>>>> <<a href="mailto:meheff@google.com">meheff@google.com</a>><br>
>> >>>>>>> wrote:<br>
>> >>>>>>>> These patches rename the loop unrolling and loop vectorizer<br>
>> >>>>>>>> metadata<br>
>> >>>>>>>> such that they have a common 'llvm.loop.' prefix. Metadata<br>
>> >>>>>>>> name<br>
>> >>>>>>>> changes:<br>
>> >>>>>>>><br>
>> >>>>>>>> llvm.vectorizer.* => llvm.loop.vectorizer.*<br>
>> >>>>>>>> llvm.loopunroll.* => llvm.loop.unroll.*<br>
>> >>>>>>>><br>
>> >>>>>>>> This was a suggestion from an earlier review<br>
>> >>>>>>>> (<a href="http://reviews.llvm.org/D4090" target="_blank">http://reviews.llvm.org/D4090</a>) which added the loop<br>
>> >>>>>>>> unrolling<br>
>> >>>>>>>> metadata. Two patches are attached. One for front-end and<br>
>> >>>>>>>> one<br>
>> >>>>>>>> for<br>
>> >>>>>>>> optimizer.<br>
>> >>>>>>>><br>
>> >>>>>>>> Mark<br>
>> >>>>>>><br>
>> >>>>>>><br>
>> >>>>>>><br>
>> >>>>>>> --<br>
>> >>>>>>> --PJ<br>
>> >>>>>>> _______________________________________________<br>
>> >>>>>>> cfe-commits mailing list<br>
>> >>>>>>> <a href="mailto:cfe-commits@cs.uiuc.edu">cfe-commits@cs.uiuc.edu</a><br>
>> >>>>>>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits</a><br>
>> >>>>>>><br>
>> >>>>>><br>
>> >>>>>> --<br>
>> >>>>>> Hal Finkel<br>
>> >>>>>> Assistant Computational Scientist<br>
>> >>>>>> Leadership Computing Facility<br>
>> >>>>>> Argonne National Laboratory<br>
>> >>>>><br>
>> >>>><br>
>> >>>> --<br>
>> >>>> Hal Finkel<br>
>> >>>> Assistant Computational Scientist<br>
>> >>>> Leadership Computing Facility<br>
>> >>>> Argonne National Laboratory<br>
>> >>><br>
>> >><br>
>> >> --<br>
>> >> Hal Finkel<br>
>> >> Assistant Computational Scientist<br>
>> >> Leadership Computing Facility<br>
>> >> Argonne National Laboratory<br>
>> >><br>
>> >> _______________________________________________<br>
>> >> llvm-commits mailing list<br>
>> >> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>> >> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
>> ><br>
><br>
><br>
</div></div></blockquote></div><br></div></div>