https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118259
Bug ID: 118259
Summary: -O3 optimisation bug fixed with -fno-inline
Product: gcc
Version: 14.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: mjr19 at cam dot ac.uk
Target Milestone: ---
Created attachment 60012
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60012&action=edit
Module-free test case
The following code, which is a poor random number generator, is expected to
print
29 25
38 21
20 40
35 9
28 12
but with gfortran -O3 it produces
29 25
7 39
7 39
7 39
7 39
Tested on x86_64 and gfortran 13.2, 14.1, 15.0 (July 7th snapshot).
If the write statement is uncommented, it gives the expected result. If
compiled with "-O3 -fno-inline" it gives the expected result. "-Wall -Wextra"
makes no complaint. It appears to me that gfortran is attempting to optimise a
call to an impure subroutine too aggressively.
module spin
contains
subroutine myrand(rnd)
real::rnd
integer,save::seed=42
seed=iand(seed*1103515245+12345,z'7fffffff')
rnd=seed/2147483648.0
! write(*,*)'R',rnd
end subroutine myrand
subroutine myrand_v(rnd)
real::rnd(:)
integer i
do i=1,size(rnd)
call myrand(rnd(i))
end do
end subroutine myrand_v
end module spin
program test
use spin
implicit none
integer::index1,index2,i
! real::rnd(2) ! gives a different result
real::rnd(3)
do i=1,5
call myrand_v(rnd)
index1=int(50*rnd(1))
index2=int(50*rnd(2))
write(*,*)index1,index2
end do
end program test
I add as an attachment the same code but written to be module free. It behaves
in the same way.
