https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
Bug ID: 86735
Summary: [8/9 Regression] Bad wrong-code bug with
"-march=skylake-avx512 -Ofast"
Product: gcc
Version: 8.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: janus at gcc dot gnu.org
Target Milestone: ---
Consider this simple test case:
program avx512_bug
implicit none
integer, dimension(:), allocatable :: list
integer, parameter :: maxList = 32
integer :: i
allocate(list(1:maxList))
do i=1,maxList
list(i) = i
end do
call test(list)
contains
subroutine test(a)
integer, dimension(:), intent(in) :: a
integer, parameter :: maxSamples = 16
integer, dimension(1:maxSamples) :: iarr
integer :: i, cnt, nbSamples
cnt = ubound(a,1)
nbSamples = min(maxSamples, cnt)
do i=1,nbSamples
iarr(i) = i
end do
print *, iarr(1:nbSamples)
print *, a(iarr(1:nbSamples))
end subroutine
end
It should print the sequence from 1 to 16 twice, which it does with gfortran
versions up to 7.
With 8.2.0 and trunk, however, the second line has varying random numbers when
the code is compiled with "-march=skylake-avx512 -Ofast".
-march=skylake works, -O3 works as well.
