Re: [PATCH] D24481: make “#pragma STDC FP_CONTRACT” on by default

[Hal Finkel via cfe-commits]

----- Original Message -----
> From: "Matthias Braun" <mbraun at apple.com>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: "Steve Canon" <scanon at apple.com>, "yaxun liu" <yaxun.liu at amd.com>, "a skolnik" <a.skolnik at samsung.com>,
> cfe-commits at lists.llvm.org, clattner at apple.com, reviews+D24481+public+c0b8d50a922986a0 at reviews.llvm.org
> Sent: Wednesday, September 28, 2016 2:55:44 PM
> Subject: Re: [PATCH] D24481: make “#pragma STDC FP_CONTRACT” on by default
> 
> 
> If we do this we should at least be targetted and restrict it to the
> tests that need the flag. For example you could put:
> list(APPEND CFLAGS -ffp-contract=off) to
> MultiSource/Applications/oggenc/CMakeLists.txt
> and
> CFLAGS += -ffp-contract=off into
> MultiSource/Applications/oggenc/Makefile
> 

I think this is a reasonable idea, with the understanding that this is going to turn into somewhat of a game of Whac-A-Mole (because you'll potentially get different FMAs on different architectures, and so the set of affected programs will be somewhat architecture dependent). Nevertheless, it shouldn't be too bad.

 -Hal

> 
> that at least forces us to think about why a specific benchmark fails
> and maybe we can find a way to rather use fpcmp/set an
> absolution/relative tollerance for when comparing the results
> (though admittedly I don't see how we can do that in a case of
> oggenc where a .ogg file is produced).
> 
> 
> - Matthias
> 
> 
> 
> 
> 
> On Sep 23, 2016, at 2:53 PM, Hal Finkel via cfe-commits <
> cfe-commits at lists.llvm.org > wrote:
> 
> We currently have logic in the test suite that sets -ffp-contract=off
> on PowerPC (because the default for GCC and other compilers on
> PowerPC/Linux systems is essentially -ffp-contract=fast). We might
> just want to do this now for all platforms.
> 
> -Hal
> 
> ----- Original Message -----
> 
> 
> From: "Steve Canon" < scanon at apple.com >
> To: reviews+D24481+public+c0b8d50a922986a0 at reviews.llvm.org
> Cc: "a skolnik" < a.skolnik at samsung.com >, clattner at apple.com ,
> hfinkel at anl.gov , "yaxun liu" < yaxun.liu at amd.com >,
> cfe-commits at lists.llvm.org
> Sent: Friday, September 23, 2016 4:47:32 PM
> Subject: Re: [PATCH] D24481: make “#pragma STDC FP_CONTRACT” on by
> default
> 
> Without digging into them yet, these are almost caused by
> overly-sensitive tests that are erroneously expecting bit-exact
> results.
> 
> - Steve
> 
> Sent from my iPhone
> 
> 
> 
> On Sep 23, 2016, at 4:42 PM, Renato Golin < renato.golin at linaro.org >
> wrote:
> 
> rengolin added a subscriber: rengolin.
> rengolin added a comment.
> 
> Folks, this commit has broken both AArch64 test-suite buildbots:
> 
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/3162
> 
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-quick/builds/10449
> 
> I have reverted in r282289, let me know if you need help testing on
> AArch64.
> 
> 
> Repository:
> rL LLVM
> 
> https://reviews.llvm.org/D24481
> 
> 
> 
> 
> 
> --
> Hal Finkel
> Lead, Compiler Technology and Programming Languages
> Leadership Computing Facility
> Argonne National Laboratory
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
> 

-- 
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory