From: Tobias Burnus <tburnus@baylibre.com>
To: Jakub Jelinek <jakub@redhat.com>, gcc-patches <gcc-patches@gcc.gnu.org>
Subject: [RFA/RFC] C++/OpenMP: Supporting (first)private for member variables [PR110347] - or VALUE_EXPR and gimplify
Date: Fri, 16 Feb 2024 16:40:13 +0100 [thread overview]
Message-ID: <4d72cb5b-a0d5-4a38-adf1-754a069c4a2c@baylibre.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 2669 bytes --]
The following works with PARALLEL but not with TARGET.
OpenMP states the following is supposed to work:
A = 5; // == this->A
B = 6; // == this->B
C[44] = 7; // == this->C; assume 'int C[100]'
#pragma <parallel|target> firstprivate(A,C) private(B)
{
A += 5; // Now: A is 10.
B = 7;
C[44] += 7; // Now C is 14
// It is unspecified what value this->{A,B,C} has
}
// {A,B,C[44]} == this->{A,B,C[44]} are still {5,6,7}
* * *
In the C++ FE, that's handled by creating a temporary variable:
v = create_temporary_var (TREE_TYPE (m));
with
SET_DECL_VALUE_EXPR (v, m);
DECL_OMP_PRIVATIZED_MEMBER(v)
where 'm' is, e.g., 'this->A' - and a bunch of
'if (DECL_OMP_PRIVATIZED_MEMBER(decl))'
in the g++ FE, only.
For PARALLEL, the VALUE_EXPR survives until omp-low.cc, which handles
this for (first)privatizing.
But for TARGET, in gimplify.cc, after the following call in
gimplify_omp_workshare
16813 gimple *g = gimplify_and_return_first (OMP_BODY (expr), &body);
the 'A' in the body will be turned into 'this->A'.
* * *
Thus, while there is after omplower the expected
#pragma omp target ... firstprivate(A)
and also
D.3081 = .omp_data_i->A; A= ...;
what actually gets used is
D.3084 = .omp_data_i->D.3046;
this = D.3084;
D.2996 = this->A;
which unsurprisingly breaks.
* * *
This can be "fixed" by using the following patch.
With that patch, the -fdump-tree-omplower looks fine. But it does then
fail with:
during RTL pass: expand
g2.cpp:11:7: internal compiler error: in make_decl_rtl, at varasm.cc:1443
for the 'A' with 'B = A' (where B is a non-member var) and 'A' is still
as the value expr 'this->A'.
--- a/gcc/gimplify.cc
+++ b/gcc/gimplify.cc
@@ -3285,12 +3285,15 @@ gimplify_var_or_parm_decl (tree *expr_p)
if (gimplify_omp_ctxp && omp_notice_variable (gimplify_omp_ctxp,
decl, true))
return GS_ALL_DONE;
+ if (!flag_openmp) // Assume: C++'s DECL_OMP_PRIVATIZED_MEMBER (decl)
+ {
/* If the decl is an alias for another expression, substitute it. */
if (DECL_HAS_VALUE_EXPR_P (decl))
{
*expr_p = unshare_expr (DECL_VALUE_EXPR (decl));
return GS_OK;
}
+ }
return GS_ALL_DONE;
}
* * *
Any idea / suggestion how to handle this best?
One way I see would be to add a lang-hook here to check for
DECL_OMP_PRIVATIZED_MEMBER, similar to the hack above. And
then ensure that the DECL_VALUE_EXPR points to the var decl
in the target region (i.e. some hacking in omp-low.cc).
I have no idea whether that would - nor whether that would be
the way forward. - Thoughts?
Tobias
[-- Attachment #2: testcase.cpp --]
[-- Type: text/x-c++src, Size: 1559 bytes --]
#if TEMPL
template <typename T>
#else
#define T int
#endif
#if PRIVATE
#define firstprivate private
#endif
struct t {
T A;
void f()
{
T B = 49;
A = 7;
#pragma omp parallel firstprivate(A) if(0) shared(B) default(none)
{
if (A != 7) __builtin_printf("ERROR 1b: %d (!= 7) inside omp parallel\n", A);
A = 5;
B = A;
}
if (A != 7) __builtin_printf("ERROR 1: %d (!= 7) omp parallel\n", A);
if (B != 5) __builtin_printf("ERROR 1a: %d\n", B);
A = 8; B = 49;
#pragma omp parallel firstprivate(A)if(0) shared(B) default(none)
{
if (A != 8) __builtin_printf("ERROR 1b: %d (!= 8) inside omp parallel\n", A);
A = 6;
B = A;
}
if (A != 8) __builtin_printf("ERROR 2: %d (!= 8) omp parallel\n", A);
if (B != 6) __builtin_printf("ERROR 2a: %d\n", B);
A = 8; B = 49;
#pragma omp target firstprivate(A) map(from:B) defaultmap(none)
{
if (A != 7) __builtin_printf("ERROR 2b: %d (!= 7) inside omp target\n", A);
A = 7;
B = A;
}
if (A != 8) __builtin_printf("ERROR 3: %d (!= 8) omp target\n", A);
if (B != 7) __builtin_printf("ERROR 3a: %d\n", B);
A = 9; B = 49;
#pragma omp target firstprivate(A) map(from:B) defaultmap(none)
{
if (A != 7) __builtin_printf("ERROR 3b: %d (!= 7) inside omp target\n", A);
A = 8;
B = A;
}
if (A != 9) __builtin_printf("ERROR 4: %d (!= 9) omp target\n", A); else __builtin_printf("OK\n");
if (B != 8) __builtin_printf("ERROR 4a: %d\n", B);
}
};
void bar() {
#if TEMPL
struct t<int> x;
#else
struct t x;
#endif
x.f();
}
int main()
{
bar();
}
next reply other threads:[~2024-02-16 15:40 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-02-16 15:40 Tobias Burnus [this message]
-- strict thread matches above, loose matches on Subject: below --
2024-02-16 15:15 Tobias Burnus
2024-02-16 15:42 ` Jakub Jelinek
2024-02-16 15:57 ` Jakub Jelinek
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=4d72cb5b-a0d5-4a38-adf1-754a069c4a2c@baylibre.com \
--to=tburnus@baylibre.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jakub@redhat.com \
/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).