[PATCH] D24925: [testsuite] turn fp-contract off for ARM because output checking is not flexible enough

[Hal Finkel via llvm-commits]

----- Original Message -----
> From: "Renato Golin" <renato.golin at linaro.org>
> To: "Stephen Canon" <scanon at apple.com>
> Cc: reviews+D24925+public+fb3ab83c5103fedd at reviews.llvm.org, "Sanjay Patel" <spatel at rotateright.com>, "Abe Skolnik"
> <a.skolnik at samsung.com>, "sebpop llvm" <sebpop.llvm at gmail.com>, "yaxun liu" <yaxun.liu at amd.com>, "Hal Finkel"
> <hfinkel at anl.gov>, "Matthias Braun" <matze at braunis.de>, "Amara Emerson" <amara.emerson at arm.com>, "Chad Rosier"
> <mcrosier at codeaurora.org>, "Sam Parker" <sam.parker at arm.com>, "LLVM Commits" <llvm-commits at lists.llvm.org>
> Sent: Monday, September 26, 2016 2:13:55 PM
> Subject: Re: [PATCH] D24925: [testsuite] turn fp-contract off for ARM because output checking is not flexible enough
> 
> On 26 September 2016 at 20:07, Stephen Canon <scanon at apple.com>
> wrote:
> > I agree with Hal, but would go a little bit further: we should
> > really be
> > running these tests twice.  First, for platforms with
> > FLOAT_EVAL_METHOD=0 in
> > -ffp-contract=off with a bit-exact result requirement, and second
> > for all
> > platforms with -ffp-contract=on and only checking that the
> > appropriate error
> > bound is satisfied (i.e. *no* comparison to previous results, only
> > the
> > established bound).
> 
> I agree we should test both, but given we only have hardware to test
> one, testing what the user will see is the only one we *can* do.

This is getting off topic a bit, but given the number of miscompiles we find by self hosting that don't show up in the test suite, and some other experience, I estimate that our test suite should be at least an order of magnitude larger than it is currently. If our current hardware can't handle the test suite duplicating some of the tests using floating-point computation in some different configurations (e.g. FMAs off vs. FMAs on, not to mention -ffp-contract and -ffast-math) we're already in trouble.

 -Hal

> 
> cheers,
> --renato
> 

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