On 1/20/2017 3:19 PM, Philip Reames via llvm-dev wrote:> > If someone can produce an *IR* reproducer which they believe is > incorrect, I'm happy to take a look at the PRE code. I will not have > the time to reproduce this from C or work through the interaction of > other passes; I will only look at this if there is a bug with PRE > specific IR reproducer attached. >https://llvm.org/bugs/show_bug.cgi?id=31651 has an IR testcase. -Eli -- Employee of Qualcomm Innovation Center, Inc. Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
Hi Philip, On 01/21/2017 01:03 AM, Friedman, Eli wrote:> On 1/20/2017 3:19 PM, Philip Reames via llvm-dev wrote: >> >> If someone can produce an *IR* reproducer which they believe is >> incorrect, I'm happy to take a look at the PRE code. I will not have >> the time to reproduce this from C or work through the interaction of >> other passes; I will only look at this if there is a bug with PRE >> specific IR reproducer attached. >> > > https://llvm.org/bugs/show_bug.cgi?id=31651 has an IR testcase.As Eli said, in the PR there is IR that I think is miscompiled if you run gvn on it. In the PR and earlier in this thread I pasted some C-like code just to describe what I think is happening since I thought it was easier than pasting llvm assembler, but the pre_gvn.ll file in PR31651 can indeed be used to reproduce the problem by just running gvn on it. Great if you could have a look! Mikael> > -Eli >
Will try to get to this by end of week. Feel free to ping me if I don't respond by then. Philip On 01/22/2017 11:54 PM, Mikael Holmén wrote:> Hi Philip, > > On 01/21/2017 01:03 AM, Friedman, Eli wrote: >> On 1/20/2017 3:19 PM, Philip Reames via llvm-dev wrote: >>> >>> If someone can produce an *IR* reproducer which they believe is >>> incorrect, I'm happy to take a look at the PRE code. I will not have >>> the time to reproduce this from C or work through the interaction of >>> other passes; I will only look at this if there is a bug with PRE >>> specific IR reproducer attached. >>> >> >> https://llvm.org/bugs/show_bug.cgi?id=31651 has an IR testcase. > > As Eli said, in the PR there is IR that I think is miscompiled if you > run gvn on it. > > In the PR and earlier in this thread I pasted some C-like code just to > describe what I think is happening since I thought it was easier than > pasting llvm assembler, but the pre_gvn.ll file in PR31651 can indeed > be used to reproduce the problem by just running gvn on it. > > Great if you could have a look! > Mikael > >> >> -Eli >>