The most relevant of these is r222856 which reverted a set of broken optimizations in r210006. I'd be happy with just including r222856. Optionally, we could take r222868 and r222871 which correctly restore the optimization. On Mon, Dec 29, 2014 at 2:39 PM, Chandler Carruth <chandlerc at google.com> wrote:> Hey Tom, > > David has pointed out to me that we have some *really* bad bugs in > instcombine in 3.5 that are starting to seriously impact users. > http://llvm.org/PR22050 is a great example. I'm actually surprised LLVM > itself doesn't trip over this bug, and I suspect an innocent change to LLVM > could result in that. > > Do you think it's possible to spin a rc3 with a couple more patches to fix > this? I think David has already curated the patches and would be willing to > ensure they merge cleanly. I'm going to help review them as OK. My hope is > that this won't require a huge delay in 3.5.1, but what do you think? > > -Chandler > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net> wrote: > >> Hi, >> >> I've just tagged 3.5.1-rc2, so testing can begin. Remember to use the 3.5 >> branch of the test-suite. If you have any questions, let me know. >> >> -Tom >> _______________________________________________ >> LLVM Developers mailing list >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >> > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20141229/c781cf36/attachment.html>
On Mon, Dec 29, 2014 at 11:52 AM, David Majnemer <david.majnemer at gmail.com> wrote:> The most relevant of these is r222856 which reverted a set of broken > optimizations in r210006. I'd be happy with just including r222856.FWIW, I've reviewed this and it is a very safe patch. I would approve it even at this late stage because (for example) they have exactly zero impact on different platforms, and so should require minimal testing. I think the only question is whether this creates an unreasonable disruption of the 3.5.1 release, and that's really a question for Tom. Optionally, we could take r222868 and r222871 which correctly restore the> optimization.These I don't think are reasonable at this stage IMO. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20141229/00ccb125/attachment.html>
On Mon, Dec 29, 2014 at 02:52:37PM -0500, David Majnemer wrote:> The most relevant of these is r222856 which reverted a set of broken > optimizations in r210006. I'd be happy with just including r222856.Go ahead and merge this one only. If there is an -rc3, it will be included. -Tom> Optionally, we could take r222868 and r222871 which correctly restore the > optimization. > > On Mon, Dec 29, 2014 at 2:39 PM, Chandler Carruth <chandlerc at google.com> > wrote: > > > Hey Tom, > > > > David has pointed out to me that we have some *really* bad bugs in > > instcombine in 3.5 that are starting to seriously impact users. > > http://llvm.org/PR22050 is a great example. I'm actually surprised LLVM > > itself doesn't trip over this bug, and I suspect an innocent change to LLVM > > could result in that. > > > > Do you think it's possible to spin a rc3 with a couple more patches to fix > > this? I think David has already curated the patches and would be willing to > > ensure they merge cleanly. I'm going to help review them as OK. My hope is > > that this won't require a huge delay in 3.5.1, but what do you think? > > > > -Chandler > > > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net> wrote: > > > >> Hi, > >> > >> I've just tagged 3.5.1-rc2, so testing can begin. Remember to use the 3.5 > >> branch of the test-suite. If you have any questions, let me know. > >> > >> -Tom > >> _______________________________________________ > >> LLVM Developers mailing list > >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > >> > > > >
In FreeBSD, I opted to import r222856 on top of 3.5.0 (see here <https://svnweb.freebsd.org/base?view=revision&revision=275160>). This has the lowest impact, and fixed the most problematic case (libapr) for us. I would really advise against doing a 3.5.1 release with this bug unfixed. :-) -Dimitry> On 29 Dec 2014, at 20:52, David Majnemer <david.majnemer at gmail.com> wrote: > > The most relevant of these is r222856 which reverted a set of broken optimizations in r210006. I'd be happy with just including r222856. Optionally, we could take r222868 and r222871 which correctly restore the optimization. > > On Mon, Dec 29, 2014 at 2:39 PM, Chandler Carruth <chandlerc at google.com <mailto:chandlerc at google.com>> wrote: > Hey Tom, > > David has pointed out to me that we have some *really* bad bugs in instcombine in 3.5 that are starting to seriously impact users. http://llvm.org/PR22050 <http://llvm.org/PR22050> is a great example. I'm actually surprised LLVM itself doesn't trip over this bug, and I suspect an innocent change to LLVM could result in that. > > Do you think it's possible to spin a rc3 with a couple more patches to fix this? I think David has already curated the patches and would be willing to ensure they merge cleanly. I'm going to help review them as OK. My hope is that this won't require a huge delay in 3.5.1, but what do you think? > > -Chandler > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net <mailto:tom at stellard.net>> wrote: > Hi, > > I've just tagged 3.5.1-rc2, so testing can begin. Remember to use the 3.5 > branch of the test-suite. If you have any questions, let me know. > > -Tom > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu <mailto:LLVMdev at cs.uiuc.edu> http://llvm.cs.uiuc.edu <http://llvm.cs.uiuc.edu/> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev> > > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20141231/820e31a2/attachment.html>
Done in r225255. On Mon, Dec 29, 2014 at 1:46 PM, Tom Stellard <tom at stellard.net> wrote:> On Mon, Dec 29, 2014 at 02:52:37PM -0500, David Majnemer wrote: > > The most relevant of these is r222856 which reverted a set of broken > > optimizations in r210006. I'd be happy with just including r222856. > > Go ahead and merge this one only. If there is an -rc3, it will be > included. > > -Tom > > > Optionally, we could take r222868 and r222871 which correctly restore the > > optimization. > > > > On Mon, Dec 29, 2014 at 2:39 PM, Chandler Carruth <chandlerc at google.com> > > wrote: > > > > > Hey Tom, > > > > > > David has pointed out to me that we have some *really* bad bugs in > > > instcombine in 3.5 that are starting to seriously impact users. > > > http://llvm.org/PR22050 is a great example. I'm actually surprised > LLVM > > > itself doesn't trip over this bug, and I suspect an innocent change to > LLVM > > > could result in that. > > > > > > Do you think it's possible to spin a rc3 with a couple more patches to > fix > > > this? I think David has already curated the patches and would be > willing to > > > ensure they merge cleanly. I'm going to help review them as OK. My > hope is > > > that this won't require a huge delay in 3.5.1, but what do you think? > > > > > > -Chandler > > > > > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net> > wrote: > > > > > >> Hi, > > >> > > >> I've just tagged 3.5.1-rc2, so testing can begin. Remember to use > the 3.5 > > >> branch of the test-suite. If you have any questions, let me know. > > >> > > >> -Tom > > >> _______________________________________________ > > >> LLVM Developers mailing list > > >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > > >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > > >> > > > > > > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150106/ee2661d1/attachment.html>
On Mon, Dec 29, 2014 at 02:52:37PM -0500, David Majnemer wrote:> The most relevant of these is r222856 which reverted a set of broken > optimizations in r210006. I'd be happy with just including r222856. > Optionally, we could take r222868 and r222871 which correctly restore the > optimization. >Hi David, I planning on doing a 3.5.2 release, so I think r222868 and r222871 can be merged to the 3.5 branch too if Chandler approves. -Tom> On Mon, Dec 29, 2014 at 2:39 PM, Chandler Carruth <chandlerc at google.com> > wrote: > > > Hey Tom, > > > > David has pointed out to me that we have some *really* bad bugs in > > instcombine in 3.5 that are starting to seriously impact users. > > http://llvm.org/PR22050 is a great example. I'm actually surprised LLVM > > itself doesn't trip over this bug, and I suspect an innocent change to LLVM > > could result in that. > > > > Do you think it's possible to spin a rc3 with a couple more patches to fix > > this? I think David has already curated the patches and would be willing to > > ensure they merge cleanly. I'm going to help review them as OK. My hope is > > that this won't require a huge delay in 3.5.1, but what do you think? > > > > -Chandler > > > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net> wrote: > > > >> Hi, > >> > >> I've just tagged 3.5.1-rc2, so testing can begin. Remember to use the 3.5 > >> branch of the test-suite. If you have any questions, let me know. > >> > >> -Tom > >> _______________________________________________ > >> LLVM Developers mailing list > >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > >> > > > >
Hi I noticed that the announce of release 3.5.1 is there one week ago (http://lists.cs.uiuc.edu/pipermail/llvm-announce/2015-January/000056.html) and the final binaries are also available, or to be precise, downloadable at http://llvm.org/releases/3.5.1/ . However I could not found the freebsd and darwin binaries the last time I checked there. Just to be kind of curious, were they missing by mistake? Anyway, I am not urge about that. I have found the brand-new release 3.6.0-rc2 (which has full kinds of binaries as usual, from http://llvm.org/pre-releases/3.6.0/ ) and switched to it already. -- Cheers, Chilledheart On Mon, Jan 19, 2015, at 22:50, Tom Stellard wrote:> On Mon, Dec 29, 2014 at 02:52:37PM -0500, David Majnemer wrote: > > The most relevant of these is r222856 which reverted a set of broken > > optimizations in r210006. I'd be happy with just including r222856. > > Optionally, we could take r222868 and r222871 which correctly restore the > > optimization. > > > > Hi David, > > I planning on doing a 3.5.2 release, so I think r222868 and r222871 can > be > merged to the 3.5 branch too if Chandler approves. > > -Tom > > > On Mon, Dec 29, 2014 at 2:39 PM, Chandler Carruth <chandlerc at google.com> > > wrote: > > > > > Hey Tom, > > > > > > David has pointed out to me that we have some *really* bad bugs in > > > instcombine in 3.5 that are starting to seriously impact users. > > > http://llvm.org/PR22050 is a great example. I'm actually surprised LLVM > > > itself doesn't trip over this bug, and I suspect an innocent change to LLVM > > > could result in that. > > > > > > Do you think it's possible to spin a rc3 with a couple more patches to fix > > > this? I think David has already curated the patches and would be willing to > > > ensure they merge cleanly. I'm going to help review them as OK. My hope is > > > that this won't require a huge delay in 3.5.1, but what do you think? > > > > > > -Chandler > > > > > > On Fri, Dec 19, 2014 at 2:35 PM, Tom Stellard <tom at stellard.net> wrote: > > > > > >> Hi, > > >> > > >> I've just tagged 3.5.1-rc2, so testing can begin. Remember to use the 3.5 > > >> branch of the test-suite. If you have any questions, let me know. > > >> > > >> -Tom > > >> _______________________________________________ > > >> LLVM Developers mailing list > > >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > > >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev > > >> > > > > > > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev