public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
From: "burnus at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug fortran/104949] [OpenMP] omp target: firstprivate of allocatable array – only descriptor firstprivatized
Date: Thu, 31 Mar 2022 18:12:39 +0000	[thread overview]
Message-ID: <bug-104949-4-hYmz5E9Fte@http.gcc.gnu.org/bugzilla/> (raw)
In-Reply-To: <bug-104949-4@http.gcc.gnu.org/bugzilla/>

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104949

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> ---
The following addition to testcase is needed.

! -------------------------------

!$omp parallel default(A)
!$omp master
  if (any (A /= [1,2,3,4,5])) error stop
  A(:) = [99,88,77,66,55]
!$omp end master
!$omp end parallel
if (any (A /= [1,2,3,4,5])) error stop

!$omp target defaultmap(firstprivate)
  if (any (A /= [1,2,3,4,5])) error stop
  A(:) = [99,88,77,66,55]
!$omp end target
if (any (A /= [1,2,3,4,5])) error stop

! -------------------------------

Reason: A different code path for
(a) defaultmap(firstprivate)  and  default(firstprivate)
    For them gfc_omp_finish_clause is invoked
(b) explicit firstprivate(a)
    Handled via gfc_trans_omp_clauses -> gfc_trans_omp_variable_list.
    but gfc_omp_finish_clause is not called (should it?)

The 'parallel' variant works – and for both variants the
gfc_omp_clause_copy_ctor is invoked for both via lower_rec_input_clauses()'s
the block which follows the 'do_firstprivate:' label.

Both 'target' variants only create 'firstprivate(a)' - but additionally
'firstprivate(a.data)' is needed, including doing a pointer attach.

  *  *  * 

Currently, we support either code using attach like:

  map(force_to:var) map(force_to:*var.data [len...]) map(attach_detach:var.data
[bias: 0])

or pointer-set/pointer assign like for

  map(to:var.data [len:...])
  map(to:var [pointer set, len: 64])
  map(alloc: var.data [pointer assign, bias: 0])

In my understanding, either code requires that 'var.data' can be found as
host->device lookup, which does not work for firstprivate. And just internally
adding 'a' to the host->device mapping for firstprivate purpose internally does
not handle the following.

a = 5
!$omp target enter data map(to: a)
a = 7

! adding internally 'a' (host)->'a'(firstprivate) mapping to find the 'a.var'
! for the usage in attach/pointer assign does not work as 'a' already exists.

!$omp target firstprivate(a) 
  ! a should be now 7
!$omp end firstprivate
!$omp target
   ! a is 5
!$omp end target


For the usage in
  type t
    integer, allocatable :: a, b
  end type t
  type(t) :: var
  !$omp target firstprivate(var)
    ...
multiple attachments are needed, i.e. just using firstprivate(a)
firstprivate(b) pointer_attach (a.data) and working with index i, i+1, i+2 does
not work.

  reply	other threads:[~2022-03-31 18:12 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-03-16  8:45 [Bug fortran/104949] New: " burnus at gcc dot gnu.org
2022-03-31 18:12 ` burnus at gcc dot gnu.org [this message]
2022-05-23  8:54 ` [Bug fortran/104949] " cvs-commit at gcc dot gnu.org
2022-05-23 10:05 ` burnus at gcc dot gnu.org
2023-03-24 16:16 ` cvs-commit at gcc dot gnu.org

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=bug-104949-4-hYmz5E9Fte@http.gcc.gnu.org/bugzilla/ \
    --to=gcc-bugzilla@gcc.gnu.org \
    --cc=gcc-bugs@gcc.gnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).