GCC 16.1 was officially released today:
2 Likes
does anyone have a ./configure line that has worked? I am not being able to compile gcc
1 Like
Jerry guided me to compile GCC a few weeks ago.
I had created a gcc-dev directory with three subdirs:
trunk/(gcc sources)objdir/(empty)usr/(empty)
First, in trunk/:
$ ./contrib/download_prerequisites
$ sudo dnf install gcc-c++ glibc-devel.i686
From objdir/, I had in my script:
readonly repertoire=/home/my_login/gcc_dev
../trunk/configure --prefix="${repertoire}"/usr --enable-languages=c,c++,fortran --enable-multilib --enable-libgomp --disable-bootstrap
make -j10 1>/dev/null && make install 1>/dev/null
Drink you coffee… 
Finally I tested my snippets with:
gcc_dev/my_snippets$ LD_LIBRARY_PATH=/home/my_login/gcc_dev/usr/lib:/home/my_login/gcc_dev/usr/lib64
gcc_dev/my_snippets$ export LD_LIBRARY_PATH
gcc_dev/my_snippets$ ../usr/bin/gfortran -fcoarray=lib collective.f90 -lcaf_shmem
I should update the Fortran Wiki page with those info.
1 Like
For adventurers only, GFortran 16.1 has arrived in Fedora Rawhide (development distribution):
Next steps should be Fedora 44 Testing, then Stable.
It has just arrived in Fedora Linux 44:
$ gfortran --version
GNU Fortran (GCC) 16.1.1 20260501 (Red Hat 16.1.1-1)
Copyright © 2026 Free Software Foundation, Inc.
1 Like
Does gfortran 16 have an entirely new front-end? What kinds of problems should we watch for with new front end code?
The gfortran front end is an evolution of what has existed for many years. So you shouldn’t have to change anything. (Modulo bug fixes. One I know about is that Jerry D and I worked to get certain INQUIRE return values corrected.) The new front ends over the past couple of major releases are Algol 68 and COBOL.
2 Likes
DO CONCURRENT can speedup like openMP in gfortran 16.1?
I’ve had little success with GNU accelerating do concurrent, if you have good experiences please do tell
2 Likes
the solution was to do an out of source build…
1 Like
Indeed @jorgeg , that’s an advice that Jerry gave me, after my first attempts.
1 Like
Jerry is smart! thanks so much
2 Likes
That is definitely a prerequisite to be a GFortran developer 
. I guess he compiles GCC every day, so you can trust him.
1 Like
Another news about GCC (17 ?):
I don’t know if Fortran will be concerned, but if it is a back-end… LFortran has already a WebAssembly back-end.
1 Like
I did some comparative bench marking of gfortran and Intel ifx.
Here is the source code of the main program.
! include ‘precision_module.f90’
! include ‘integer_kind_module.f90’
! Included in the timing_module
include ‘timing_module.f90’
program ch3403
use precision_module
use timing_module
implicit none
real (dp) :: fortran_internal_pi
real (dp) :: partial_pi
real (dp) :: coarray_pi
real (dp) :: width
real (dp) :: total_sum
real (dp) :: x
real (dp), codimension [ * ] :: partial_sum
integer :: n_intervals
integer :: i
integer :: j
integer :: current_image
integer :: n_images
fortran_internal_pi = 4.0_dp*atan(1.0_dp)
n_images = num_images()
current_image = this_image()
if (current_image==1) then
print *, ’ Number of images = ', n_images
call start_timing()
end if
n_intervals = 100000
do j = 1, 5
width = 1.0_dp/real(n_intervals, dp)
total_sum = 0.0_dp
partial_sum = 0.0_dp
do i = current_image, n_intervals, n_images
x = (real(i,dp)-0.5_dp)width
partial_sum = partial_sum + f(x)
end do
partial_sum = partial_sumwidth
sync all
if (current_image==1) then
do i = 1, n_images
total_sum = total_sum + partial_sum[ i ]
end do
coarray_pi = total_sum
print 100, n_intervals, time_difference()
print 110, coarray_pi, abs(coarray_pi-fortran_internal_pi)
end if
n_intervals = n_intervals*10
sync all
end do
if (current_image==1) then
call end_timing()
end if
100 format (’ n intervals = ‘, i12, ’ time =’, f18.6)
110 format (’ pi = ', f20.16, /, ’ difference = ', f20.16)
contains
real (dp) function f(x)
implicit none
real (dp), intent (in) :: x
f = 4.0_dp/(1.0_dp+x*x)
end function
end program
Here are the timing details.
| Intervals | Average of 10 runs | ||
|---|---|---|---|
| gfortran | 100,000 | 0.000408 | |
| gfortran | 1,000,000 | 0.000907 | |
| gfortran | 10,000,000 | 0.001538 | |
| gfortran | 100,000,000 | 0.009646 | |
| gfortran | 1,000,000,000 | 0.098619 | |
| gfortran | Total | 0.111455 | |
| Intel ifx | 100,000 | 0.000275 | |
| Intel ifx | 1,000,000 | 0.000289 | |
| Intel ifx | 10,000,000 | 0.001027 | |
| Intel ifx | 100,000,000 | 0.007884 | |
| Intel ifx | 1,000,000,000 | 0.108436 | |
| Intel ifx | Total | 0.118361 | |
| gfortran | 16.1.1 20260501 | ||
| ifx | 2026.0.0 | ||
| Fedora | Fedora Linux 44 | Workstation edition | |
| Redhat | 16.1.1-1 | ||
| Processor | Intel I9 | 10980XE | |
| hyper-v | |||
| memory | 32 GB | ||
| cores | 16 |
The source files are on our web sites.
There is a tar file of all the source files.
2 Likes
GCC 16 has appeared in the Ubuntu repositories. I installed it on my Linux Mint 21.3 (Ubuntu 22.04 LTS Jammy) system with sudo apt install
2 Likes
What is the status of coarray support in gfortran 16? I’m curious about both threads-based and image-based capabilities, and possibly the combination of the two. Is there a gfortran link that has this information?
The page https://gcc.gnu.org/wiki/Coarray is totally outdated (2016).
I am no expert on CAF, so I asked ChatGPT to construct a test program for as many Coarrays features as it can. The output is below, so you can check its completeness yourself.
Gfortran 16.1 on Fedora 44 VM (6 CPUs) passes all tests. Interestingly, ‘ifx’ 2025.3.2 segfaults on test_sync_images_memory() procedure. I have yet to check the newest, 2026 version.
Coarray test program (created by ChatGPT 5.5)
module reducers
contains
pure function add_int(a, b) result(c)
integer, intent(in) :: a, b
integer :: c
c = a + b
end function add_int
end module reducers
program caf_feature_test
use reducers
use iso_fortran_env, only : atomic_int_kind, event_type, lock_type, team_type
implicit none
integer :: me, nimg, i
integer :: scalar[*]
integer, allocatable :: arr(:)[:]
integer(atomic_int_kind) :: acount[*]
type(lock_type) :: lk[*]
type(event_type) :: ev[*]
type(team_type) :: even_odd_team
integer :: team_no
me = this_image()
nimg = num_images()
if (me == 1) then
print *, "========================================"
print *, "Coarray Fortran feature test"
print *, "num_images() =", nimg
print *, "========================================"
end if
sync all
call test_images_and_coindexed_access()
call test_allocatable_coarray()
call test_sync_images()
call test_sync_images_memory()
call test_atomic()
call test_lock()
call test_events()
call test_collectives()
call test_teams()
sync all
if (me == 1) print *, "All tests finished."
contains
subroutine say(name)
character(*), intent(in) :: name
sync all
if (me == 1) then
print *
print *, "---- ", trim(name), " ----"
end if
sync all
end subroutine say
subroutine test_images_and_coindexed_access()
call say("images and coindexed access")
scalar = 1000 + me
sync all
if (me == 1) then
do i = 1, nimg
print *, "scalar on image", i, "=", scalar[i]
end do
end if
sync all
end subroutine test_images_and_coindexed_access
subroutine test_allocatable_coarray()
call say("allocatable coarray")
allocate(arr(3)[*])
arr = [me, 10*me, 100*me]
sync all
if (me == 1) then
do i = 1, nimg
print *, "arr(:, image", i, ") =", arr(:)[i]
end do
end if
sync all
deallocate(arr)
end subroutine test_allocatable_coarray
subroutine test_sync_images()
integer :: left, right
call say("sync images / remote write")
left = merge(me - 1, nimg, me > 1)
right = merge(me + 1, 1, me < nimg)
scalar = -1
sync all
scalar[right] = me
sync all
print *, "image", me, "received scalar =", scalar, &
"from left neighbour", left
sync all
end subroutine test_sync_images
subroutine test_sync_images_memory()
integer :: left, right
integer, allocatable :: partners(:)
call say("sync images and sync memory")
left = merge(me - 1, nimg, me > 1)
right = merge(me + 1, 1, me < nimg)
scalar = -1
sync all
! Each image writes to its right neighbour.
scalar[right] = me
! Ensure remote write is completed before synchronization.
sync memory
! Synchronize with both neighbours, so synchronization is matched.
if (nimg == 1) then
sync all
else if (nimg == 2) then
sync images(left)
else
partners = [left, right]
sync images(partners)
end if
print *, "image", me, "received scalar =", scalar, &
"from left neighbour", left
sync all
end subroutine test_sync_images_memory
subroutine test_atomic()
integer(atomic_int_kind) :: old
call say("atomic add / atomic ref")
acount = 0_atomic_int_kind
sync all
call atomic_add(acount[1], 1_atomic_int_kind)
sync all
if (me == 1) then
call atomic_ref(old, acount)
print *, "atomic counter on image 1 =", old, "expected", nimg
end if
sync all
end subroutine test_atomic
subroutine test_lock()
integer :: old
call say("locks")
scalar = 0
sync all
lock(lk[1])
old = scalar[1]
scalar[1] = old + 1
unlock(lk[1])
sync all
if (me == 1) then
print *, "locked counter =", scalar, "expected", nimg
end if
sync all
end subroutine test_lock
subroutine test_events()
integer :: src
call say("events")
if (nimg < 2) then
if (me == 1) print *, "Skipping event test: need at least 2 images."
sync all
return
end if
if (me == 1) then
scalar = 12345
event post(ev[2])
else if (me == 2) then
event wait(ev)
src = scalar[1]
print *, "image 2 received event; scalar[1] =", src
end if
sync all
end subroutine test_events
subroutine test_collectives()
integer :: x, y
integer :: result
character(len=40) :: msg
call say("collective subroutines")
x = me
y = x
call co_sum(y)
if (me == 1) print *, "co_sum(image index) =", y, "expected", nimg*(nimg+1)/2
y = x
call co_max(y)
if (me == 1) print *, "co_max(image index) =", y, "expected", nimg
y = x
call co_min(y)
if (me == 1) print *, "co_min(image index) =", y, "expected", 1
msg = ""
if (me == 1) msg = "hello from image 1"
call co_broadcast(msg, source_image=1)
print *, "image", me, "after co_broadcast:", trim(msg)
result = me
call co_reduce(result, add_int)
if (me == 1) print *, "co_reduce(add_int) =", result
sync all
end subroutine test_collectives
subroutine test_teams()
integer :: local_sum
call say("teams and change team")
if (nimg < 2) then
if (me == 1) print *, "Skipping team test: need at least 2 images."
sync all
return
end if
! Team 1: odd original images.
! Team 2: even original images.
team_no = merge(1, 2, mod(me,2) == 1)
form team(team_no, even_odd_team)
change team(even_odd_team)
local_sum = this_image()
call co_sum(local_sum)
if (this_image() == 1) then
print *, "original image", me, &
"is leader of team", team_number(), &
"with num_images() =", num_images(), &
"and local co_sum =", local_sum
end if
sync all
end team
sync all
end subroutine test_teams
end program caf_feature_test
2 Likes
When will gfortran 16 be available to Ubuntu systems? My sad little dialogue with it this morning is
sudo apt install gfortran
[sudo] password for john:
gfortran is already the newest version (4:15.2.0-4ubuntu1).
Summary:
Upgrading: 0, Installing: 0, Removing: 0, Not Upgrading: 1495