Re: possible gcc optimizer issue

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Thu, Sep 06, 2018 at 09:55:54AM -0400, Greg Troxel wrote:
> I found that pkgsrc/geography/geos has test failures on netbsd-8
> earmv7hf-el (RPI3).  Investigating, and asking others on pkgsrc-users to
> test:
> 
> ok	netbsd-8	amd64
> ok	netbsd-7	i386
> ok	netbsd-7	amd64
> ok      netbsd-current	sparc64
> 
> fail    netbsd-8        earmv7hf-el
> fail    netbsd-current	earmv7hf-el
> 
> better*  netbsd-8        earmv7hf-el	-O0, instead of the default -O2
> 
> * by better, one test suite that had many failures with -O2 passes, and
>   there is another error in a different suite that I currently believe
>   is a different issue, and also seems more debuggable.
> 
> The geos tests pass on Linux arm and arm64.
> 
> So, it looks like either
> 
>   gcc on netbsd/arm is buggy

gcc on arm does have bugs, actually. See for example patch-src_ocpndc.cpp
in pkgsrc/geography/opencpn/patches/

Also, I got problems when building with some more agressive optimisation
flags, even in base (for example, -mcpu=cortex-a7 -mfpu=neon-vfpv4).
I may even have a PR open on this.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--