public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [Patch] OpenMP: Handle DECL_ASSEMBLER_NAME with 'declare variant'
@ 2024-02-13 14:54 Tobias Burnus
  2024-02-13 15:08 ` Jakub Jelinek
  0 siblings, 1 reply; 6+ messages in thread
From: Tobias Burnus @ 2024-02-13 14:54 UTC (permalink / raw)
  To: gcc-patches, Jakub Jelinek; +Cc: Paul-Antoine Arras

[-- Attachment #1: Type: text/plain, Size: 578 bytes --]

Inomp_resolve_declare_variant, a code path generates a new decl for the 
base function – in doing so, it ignores the assembler name. As the 
included Fortran example shows, this will lead to a linker error. 
Solution: Also copy over the assembler name. Comments, suggestions, 
remarks before I commit it? Tobias PS: As a fallout of some testing, 
motivated by the original testcase, I have filled a couple of 
declare-variant and context-selector PRs: 113904 (dyn. 
user={condition(...)}), 113905 (multiple users of variant funcs), 113906 
(construct={...} lacks constructs).

[-- Attachment #2: fix-decl-var.diff --]
[-- Type: text/x-patch, Size: 3028 bytes --]

OpenMP: Handle DECL_ASSEMBLER_NAME with 'declare variant'

gcc/ChangeLog:

	* omp-general.cc (omp_resolve_declare_variant): When building the decl
	for the base variant, honor also the assembler name.

gcc/testsuite/ChangeLog:

	* gfortran.dg/gomp/declare-variant-20.f90: New test.

 gcc/omp-general.cc                                 |  2 +
 .../gfortran.dg/gomp/declare-variant-20.f90        | 62 ++++++++++++++++++++++
 2 files changed, 64 insertions(+)

diff --git a/gcc/omp-general.cc b/gcc/omp-general.cc
index 2e31a3f9290..bc92a170e96 100644
--- a/gcc/omp-general.cc
+++ b/gcc/omp-general.cc
@@ -2630,6 +2630,8 @@ omp_resolve_declare_variant (tree base)
       (*slot)->variants = entry.variants;
       tree alt = build_decl (DECL_SOURCE_LOCATION (base), FUNCTION_DECL,
 			     DECL_NAME (base), TREE_TYPE (base));
+      if (DECL_ASSEMBLER_NAME_SET_P (base))
+	SET_DECL_ASSEMBLER_NAME (alt, DECL_ASSEMBLER_NAME (base));
       DECL_ARTIFICIAL (alt) = 1;
       DECL_IGNORED_P (alt) = 1;
       TREE_STATIC (alt) = 1;
diff --git a/gcc/testsuite/gfortran.dg/gomp/declare-variant-20.f90 b/gcc/testsuite/gfortran.dg/gomp/declare-variant-20.f90
new file mode 100644
index 00000000000..c7050a22365
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/gomp/declare-variant-20.f90
@@ -0,0 +1,62 @@
+! { dg-additional-options "-fdump-tree-gimple-asmname" }
+
+! This tests that mangled names, i.e. DECL_NAME != DECL_ASSEMBLER_NAME
+! are properly handled
+
+! This test case failed before with:
+!   undefined reference to `foo'
+! as the actual symbol is __m_MOD_foo
+
+! NOTE 1: This test relies  on late resolution of condition,
+! which is here enforced via the always_false_flag variable.
+!
+! NOTE 2: Using a variable is an OpenMP 5.1 feature that is/was not supported
+! when this test case was created, cf. PR middle-end/113904
+
+module m
+  implicit none (type, external)
+  logical :: always_false_flag = .false.
+contains
+  integer function variant1() result(res)
+    res = 1
+  end function
+
+  integer function variant2() result(res)
+    res = 2
+  end function
+
+  integer function variant3() result(res)
+    res = 3
+  end function
+
+  integer function foo() result(res)
+    !$omp  declare variant(variant1) match(construct={teams})
+    !$omp  declare variant(variant2) match(construct={parallel})
+    !$omp  declare variant(variant3) match(user={condition(always_false_flag)},construct={target})
+    res = 99
+  end
+end module m
+
+program main
+  use m
+  implicit none (type, external)
+  integer :: r1, r2, r3
+
+  r1 = foo()
+  if (r1 /= 99) stop 1
+
+  !$omp parallel if (.false.)
+    r2 = foo()
+    if (r2 /= 2) stop 2
+  !$omp end parallel
+
+  !$omp teams num_teams(1)
+    r3 = foo()
+    if (r3 /= 1) stop 3
+  !$omp end teams
+
+end program 
+
+! { dg-final { scan-tree-dump-times "r1 = __m_MOD_foo \\(\\);" 1 "gimple" } }
+! { dg-final { scan-tree-dump-times "r2 = __m_MOD_variant2 \\(\\);" 1 "gimple" } }
+! { dg-final { scan-tree-dump-times "r3 = __m_MOD_variant1 \\(\\);" 1 "gimple" } }

^ permalink raw reply	[flat|nested] 6+ messages in thread
end of thread, other threads:[~2024-03-04 12:58 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-02-13 14:54 [Patch] OpenMP: Handle DECL_ASSEMBLER_NAME with 'declare variant' Tobias Burnus
2024-02-13 15:08 ` Jakub Jelinek
2024-02-13 17:31   ` [Patch] OpenMP: Reject non-const 'condition' trait in Fortran (was: [Patch] OpenMP: Handle DECL_ASSEMBLER_NAME with 'declare variant') Tobias Burnus
2024-02-13 17:48     ` Jakub Jelinek
2024-02-13 20:03       ` [Patch] OpenMP: Reject non-const 'condition' trait in Fortran Tobias Burnus
2024-03-04 12:58     ` [Patch] OpenMP: Reject non-const 'condition' trait in Fortran (was: [Patch] OpenMP: Handle DECL_ASSEMBLER_NAME with 'declare variant') Thomas Schwinge

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).