| Message ID | CAO9iq9HYU+LD=7X6vSxc-t26693zsoBHFtZ77hWH3DUiFRp33A@mail.gmail.com |
|---|---|
| State | New |
| Headers | show |
PING? FWIW, Rainer Orth confirmed that the fixed testcase now passes on Solaris/SPARC. On Tue, Aug 23, 2016 at 4:40 PM, Janne Blomqvist <blomqvist.janne@gmail.com> wrote: > Hi, > > the attached patch fixes PR 77261. The problem is that with the new > PRNG, the random sequence for real(16) and lesser kinds is not the > same. And then on targets that have real(16) but not real(10) the > testcase fails. Fixed by essentially running the tests only when > real(10) is available. > > Regtested on x86_64-pc-linux-gnu, Ok for trunk? > > 2016-08-23 Janne Blomqvist <jb@gcc.gnu.org> > > PR fortran/77261 > * gfortran.dg/random_3.f90: Run testcase only when real(10) is > available. > > > -- > Janne Blomqvist
Dear Janne, That's fine for trunk. Thanks for the patch. Paul On 29 August 2016 at 09:05, Janne Blomqvist <blomqvist.janne@gmail.com> wrote: > PING? > > FWIW, Rainer Orth confirmed that the fixed testcase now passes on Solaris/SPARC. > > On Tue, Aug 23, 2016 at 4:40 PM, Janne Blomqvist > <blomqvist.janne@gmail.com> wrote: >> Hi, >> >> the attached patch fixes PR 77261. The problem is that with the new >> PRNG, the random sequence for real(16) and lesser kinds is not the >> same. And then on targets that have real(16) but not real(10) the >> testcase fails. Fixed by essentially running the tests only when >> real(10) is available. >> >> Regtested on x86_64-pc-linux-gnu, Ok for trunk? >> >> 2016-08-23 Janne Blomqvist <jb@gcc.gnu.org> >> >> PR fortran/77261 >> * gfortran.dg/random_3.f90: Run testcase only when real(10) is >> available. >> >> >> -- >> Janne Blomqvist > > > > -- > Janne Blomqvist
diff --git a/gcc/testsuite/gfortran.dg/random_3.f90 b/gcc/testsuite/gfortran.dg/random_3.f90 index 8e087c4..0491000 100644 --- a/gcc/testsuite/gfortran.dg/random_3.f90 +++ b/gcc/testsuite/gfortran.dg/random_3.f90 @@ -1,10 +1,10 @@ ! { dg-do run } ! { dg-require-effective-target fortran_large_real } -! Check that the random_seed for real(10) or real(16) exists and that -! real(8) and real(10) or real(16) random number generators +! Check that the random_seed for real(10) exists and that +! real(8) and real(10) random number generators ! return the same sequence of values. ! Mostly copied from random_2.f90 -program random_4 +program random_3 integer,parameter :: k = selected_real_kind (precision (0.0_8) + 1) integer, dimension(:), allocatable :: seed @@ -13,17 +13,23 @@ program random_4 real, parameter :: delta = 1.d-10 integer n - call random_seed (size=n) - allocate (seed(n)) - call random_seed (get=seed) - ! Test both array valued and scalar routines. - call random_number(r8) - call random_number (r8(10)) + ! Run the test only if real(10) is available. With the current + ! xorshift1024* PRNG the real(16) generator uses two uint64_t values + ! for every real(16) value generated, and hence the sequences won't + ! be the same as with real(4,8,10). + if (k == 10) then + call random_seed (size=n) + allocate (seed(n)) + call random_seed (get=seed) + ! Test both array valued and scalar routines. + call random_number(r8) + call random_number (r8(10)) - ! Reset the seed and get the real(8) values. - call random_seed (put=seed) - call random_number(r10) - call random_number (r10(10)) + ! Reset the seed and get the real(8) values. + call random_seed (put=seed) + call random_number(r10) + call random_number (r10(10)) - if (any ((r8 - r10) .gt. delta)) call abort -end program random_4 + if (any ((r8 - r10) .gt. delta)) call abort + end if +end program random_3
Hi, the attached patch fixes PR 77261. The problem is that with the new PRNG, the random sequence for real(16) and lesser kinds is not the same. And then on targets that have real(16) but not real(10) the testcase fails. Fixed by essentially running the tests only when real(10) is available. Regtested on x86_64-pc-linux-gnu, Ok for trunk? 2016-08-23 Janne Blomqvist <jb@gcc.gnu.org> PR fortran/77261 * gfortran.dg/random_3.f90: Run testcase only when real(10) is available.