llvm dev - Sep 2015 - Codegen difference between Asserts and No-Asserts mode

Hi,

          What is the guarantee that the codegen is same across Release and
Release+Asserts build (except the known differences)?

I assume there is no guarantee and it is possible to drop a bug like this.
Personally I feel it is very expensive to debug and it depends on each
scenario.

Most of the bots test the Release+Asserts mode and any thoughts on how we
can catch this sort of a problem with minimal set of changes to build
infrastructure?

 

--Sumanth G

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150923/26661e98/attachment.html>

Hi Sumanth,

I don't know to what you're referring, but the output from Release vs.
Release+Asserts should be identical. There might be differences in the IR value
names, but if that is affecting the final result, that's a bug.

 -Hal

----- Original Message -----
> From: "Sumanth Gundapaneni via llvm-dev" <llvm-dev at
lists.llvm.org>
> To: llvmdev at cs.uiuc.edu, llvm-dev at lists.llvm.org
> Sent: Wednesday, September 23, 2015 4:44:05 PM
> Subject: [llvm-dev] Codegen difference between Asserts and No-Asserts mode
> 
> Hi,
> 
> What is the guarantee that the codegen is same across Release and
> Release+Asserts build (except the known differences)?
> 
> I assume there is no guarantee and it is possible to drop a bug like
> this. Personally I feel it is very expensive to debug and it depends
> on each scenario.
> 
> Most of the bots test the Release+Asserts mode and any thoughts on
> how we can catch this sort of a problem with minimal set of changes
> to build infrastructure?
> 
> --Sumanth G
> 
> 
> 
> 
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> 
-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory

I imagine that the Clang "Check Compile Flow Consistency" tool
(tools/clang/utils/check_cfc) would be a good fit for this sort of
testing.  It's currently used to verify that adding -g does not affect
codegen or that compiling via a .s file produces the same codegen as a
directly emitted object, so it may not be too much of a stretch to use it
to verify that two different sets of compiler binaries produce identical
codegen.  Obviously you'd need a bot that was building both sets of
executables first.  Unfortunately, I don't have any time to look at doing
this myself but I expect it would be worthwhile thing to do if someone
wanted to pick this up.

-Greg


On 23 September 2015 at 22:48, Hal Finkel via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hi Sumanth,
>
> I don't know to what you're referring, but the output from Release
vs.
> Release+Asserts should be identical. There might be differences in the IR
> value names, but if that is affecting the final result, that's a bug.
>
>  -Hal
>
> ----- Original Message -----
> > From: "Sumanth Gundapaneni via llvm-dev" <llvm-dev at
lists.llvm.org>
> > To: llvmdev at cs.uiuc.edu, llvm-dev at lists.llvm.org
> > Sent: Wednesday, September 23, 2015 4:44:05 PM
> > Subject: [llvm-dev] Codegen difference between Asserts and No-Asserts
> mode
> >
> > Hi,
> >
> > What is the guarantee that the codegen is same across Release and
> > Release+Asserts build (except the known differences)?
> >
> > I assume there is no guarantee and it is possible to drop a bug like
> > this. Personally I feel it is very expensive to debug and it depends
> > on each scenario.
> >
> > Most of the bots test the Release+Asserts mode and any thoughts on
> > how we can catch this sort of a problem with minimal set of changes
> > to build infrastructure?
> >
> > --Sumanth G
> >
> >
> >
> >
> > _______________________________________________
> > LLVM Developers mailing list
> > llvm-dev at lists.llvm.org
> > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> >
>
> --
> Hal Finkel
> Assistant Computational Scientist
> Leadership Computing Facility
> Argonne National Laboratory
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150928/e797a6a4/attachment.html>