Date: Tue, 12 Feb 2013 10:31:29 GMT From: Anton Shterenlikht <mexas@bristol.ac.uk> To: mexas@bristol.ac.uk, sgk@troutmask.apl.washington.edu Cc: freebsd-current@freebsd.org, bsam@passap.ru Subject: Re: 7+ days of dogfood Message-ID: <201302121031.r1CAVTIa098618@mech-cluster241.men.bris.ac.uk> In-Reply-To: <20130210190048.GA77774@troutmask.apl.washington.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
From sgk@troutmask.apl.washington.edu Sun Feb 10 19:06:10 2013 On Sun, Feb 10, 2013 at 06:51:23PM +0000, Anton Shterenlikht wrote: > Date: Sun, 10 Feb 2013 08:44:24 -0800 > From: Steve Kargl <sgk@troutmask.apl.washington.edu> > To: Boris Samorodov <bsam@passap.ru> > Subject: Re: 7+ days of dogfood > > > > FFLAGS = -O2 -pipe -march=native -mtune=native > > > > I don't like using "=" for FLAGS at make.conf... > > > > > FFLAGS+= -funroll-loops -ftree-vectorize > > > > ... as well as defaults overriding. > > > > FFLAGS are the options used while compiling Fortran. Having > spent years contributing to and testing gfortran, I am fairly > comfortable with these options. > > Do you recommend these FFLAGS settings to all > users of gfortran (via lang/gcc4x)? > Yes. > For example, I use blas, lapack, slatec, atlas, etc. > Should I use these FFLAGS too? Without knowing what options you currently use, I cannot make a concrete recommendation other than I've extensive tested the above options with my own codes and the Polyhedron testsuite. Oh, I do recommend that one should avoid -ffast-math (unless one really understands what it actually does). You probably don't care, as you don't use ia64, but just for the record, on ia64 r244834 with # cat /etc/make.conf|grep FFLAGS FFLAGS = -O2 -pipe -march=native -mtune=native FFLAGS+= -funroll-loops -ftree-vectorize # I get building lapack: gfortran47 -O2 -pipe -march=native -mtune=native -funroll-loops -ftree-vectorize -pg -c slatms.f -o slatms.o gfortran47: error: unrecognized command line option '-march=native' gfortran47: error: unrecognized argument in option '-mtune=native' gfortran47: note: valid arguments to '-mtune=' are: itanium2 mckinley This box is CPU: Madison (1500 MHz Itanium 2) so I guess I can try "-mtune=itanium2" and probably have to drop "-march" completely. Anyway, thank you for the recommendation. Anton -- Steve
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201302121031.r1CAVTIa098618>