https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98391
Bug ID: 98391
Summary: Wrong results with small matrix size
Product: gcc
Version: 10.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libgomp
Assignee: unassigned at gcc dot gnu.org
Reporter: mehdi.chinoune at hotmail dot com
CC: jakub at gcc dot gnu.org
Target Milestone: ---
$ cat matmul.f90
program main
implicit none
integer, parameter :: sp = selected_real_kind(6,37)
integer :: i, j, k
integer :: n
character(len=5) :: var
real(sp), allocatable :: a(:,:), b(:,:), c(:,:), cc(:,:)
integer :: t1, t2, t3
real(sp) :: tic
!
call get_command_argument( 1, var )
read( var, '(i5.3)') n
print*, n
!
allocate( a(n,n), b(n,n), c(n,n), cc(n,n) )
call random_number(a)
call random_number(b)
c = 0._sp
!
call system_clock(t1, tic)
cc = 0._sp
do j = 1, n
do k = 1, n
do i = 1, n
cc(i,j) = a(i,k)*b(k,j) + cc(i,j)
end do
end do
end do
call system_clock( t2 )
print*, "T1 = ", (t2-t1)/tic
!$acc parallel loop collapse(3) copyin(a,b) copy(c)
!$omp target teams distribute parallel do collapse(3) map( to:a,b) map(
tofrom:c )
do j = 1, n
do k = 1, n
do i = 1, n
c(i,j) = a(i,k)*b(k,j) + c(i,j)
end do
end do
end do
call system_clock(t3)
print*, "T2 = ", (t3-t2)/tic
!
print*, "Max(|C2-C1|) = ", maxval( abs(cc-c) )
print '(8(f8.6,x))', abs(cc-c)
!
end program main
$ gfortran-10 -fopenmp -foffload=nvptx-none matmul.f90 -o matmul.x
$ ./matmul.x 16
16
T1 = 0.00000000
T2 = 0.129999995
Max(|C2-C1|) = 1.94201875
1.431175 1.329625 0.592309 1.422428 0.535594 0.723758 0.365623 0.492199
0.108950 0.489844 1.100990 0.736585 1.182946 0.372061 0.185741 1.237249
0.629846 0.706569 0.913222 1.189827 0.617232 0.973922 0.290312 0.755902
0.886491 0.874653 1.642155 1.499317 1.506070 1.541327 1.228426 1.011003
0.967729 0.587790 0.353565 0.241555 0.283207 0.498868 0.628635 0.259884
0.469164 0.637377 0.428355 0.807801 0.409959 0.630940 0.384264 0.590675
0.805984 1.565366 0.850695 1.189638 1.602054 0.152907 0.648516 0.345356
0.261692 0.754292 0.995521 1.108027 0.837113 0.498883 1.011412 0.765647
0.741106 0.335977 0.721143 0.176006 0.169367 0.256758 0.311771 0.158323
0.401097 0.566629 0.787195 0.316571 0.179871 0.129046 0.326540 0.400939
1.898347 0.547532 1.015191 0.659491 0.223342 1.052599 1.158586 0.745466
0.491430 0.677603 0.854077 0.635704 0.475488 0.692096 0.749846 1.144690
0.712732 0.894958 0.874708 0.775818 0.674985 1.034302 1.350598 0.503635
0.082981 0.620052 0.888901 1.243073 1.147498 0.785260 0.260857 0.790078
0.571918 0.326864 0.090428 0.353512 0.167201 0.553904 0.378190 0.368513
0.238174 0.247996 0.038351 0.118958 0.041398 0.009377 0.049255 0.042368
1.942019 0.577147 1.252101 1.314539 0.845633 0.898366 1.743726 0.886740
0.674877 0.428364 0.431692 0.143040 0.551206 0.222548 0.058435 0.401517
0.172672 0.128829 0.063327 0.320080 0.120494 0.111500 0.341694 0.348082
0.408216 0.414768 0.706280 0.923327 0.265845 0.318310 0.082875 0.091210
1.165700 1.020596 0.282284 0.086616 0.350870 0.860692 0.299911 0.278247
0.756335 0.357223 0.633629 1.136913 0.815934 0.602087 0.790322 0.821650
1.006154 1.494015 0.950610 1.597616 0.962432 0.411526 1.268245 1.035166
0.755580 0.614513 1.165176 0.863545 0.478808 1.186599 1.245776 1.010780
1.108782 0.462348 0.980049 0.717201 1.594621 1.345250 1.080317 1.295755
0.406487 0.916595 1.325978 1.354560 1.209628 0.775334 1.611006 0.734939
0.604795 0.831590 0.316839 0.720204 0.739102 0.406983 0.807365 0.439079
0.654506 0.749812 1.219932 1.221502 0.534549 0.366653 0.989139 1.047143
0.643886 0.319602 0.621458 0.171583 0.052870 0.410973 0.266432 0.346806
0.124081 0.744631 0.876374 0.983852 1.121478 0.226588 0.390455 0.458977
0.646022 0.169284 0.593737 0.848242 1.612478 1.024757 1.028395 1.715509
1.202510 1.336850 1.219257 0.633523 0.988450 1.170813 0.852149 0.870198
It happens with both OpenACC and OpenMP.
