This is the mail archive of the mailing list for the GNU Fortran project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] Full float128, third iteration

On Thu, Sep 09, 2010 at 10:58:10PM +0200, FX wrote:
> Thanks to everyone who tested the two previous patches on their systems, I hope we're now close to the final integration (at least from the technical point of view!).


quad looks good on i686-*-freebsd.  The testsuite is showing

                === gfortran Summary ===

# of expected passes            36598
# of unexpected failures        39
# of unexpected successes       18
# of expected failures          50
# of unsupported tests          186
/usr/home/kargl/gcc/obj4x/gcc/testsuite/gfortran/../../gfortran  version 4.6.0 20100910 (experimental) (GCC) 

The failures caused by quad support are gfortran.dg/promotion.f90
and gfortran.dg/default_format_denormal_2.f90.  For promotion.f90,
I changed the testcase to

! { dg-do run { target i?86-*-* x86_64-*-* } }
! { dg-require-effective-target ilp32 }
! { dg-options "-fdefault-integer-8 -fdefault-real-8" }
program a
   logical l
   integer i
   real x
   double precision d
   if (kind(l) /= 8) print*, 'l', kind(l)
   if (kind(i) /= 8) print*, 'i', kind(i)
   if (kind(x) /= 8) print*, 'x', kind(x)
   if (kind(d) /= 8) print*, 'd', kind(d)
end program a

with the result

laptop:kargl[244] gfc4x -o z -fdefault-integer-8 -fdefault-real-8 promotion.f90 
laptop:kargl[245] ./z
 d                   16

so, 'double precision' is being promoted to REAL(16).  This is
not a big deal to me.  Anyone using the above option should be

default_format_denormal_2.f90 fails with

ptop:kargl[251] gfc4x -o z default_format_denormal_2.f90
/tmp/ccZbSAFl.o(.text+0x245): In function `__test_default_format_MOD_test_rl':
: undefined reference to `huge_valq'
/tmp/ccZbSAFl.o(.text+0x637): In function `__test_default_format_MOD_test_rl':
: undefined reference to `huge_valq'
collect2: ld returned 1 exit status

huge_valq is a builtin function, which probably means that TFmode
isn't quite right of i686-*-freebsd.

Anyway, quad support that you've implemented looks very nice.
Thanks for everything you've done!

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]