[obvious] Fix for ICE in tree-nested.cc
Checks
Context |
Check |
Description |
linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_check--master-aarch64 |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_build--master-arm |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_check--master-arm |
success
|
Testing passed
|
Commit Message
I'm planning to push the attached patch to mainline in a couple days,
unless someone disputes that it's "obvious" or has some other objection.
I think it qualifies because the code immediately following the loop
with the bad initializer is already conditionalized on GIMPLE_OMP_TARGET
in the same way.
I think this patch is a candidate for backporting to the GCC 14 branch
once it is unfrozen after the release, too.
-Sandra
From 001d166819525dcef43b980298c2f85af77cd8ad Mon Sep 17 00:00:00 2001
From: Sandra Loosemore <sloosemore@baylibre.com>
Date: Mon, 22 Apr 2024 18:24:25 +0000
Subject: [PATCH, obvious] OpenMP: Fix for ICE in tree-nested.cc.
Use gimple_omp_target_clauses() instead of gimple_omp_taskreg_clauses()
when stmt is GIMPLE_OMP_TARGET, to avoid an as_a<> ICE. The code
immediately following this is already conditionalized in the same way.
gcc/ChangeLog
* tree-nested.cc (convert_tramp_reference_stmt): Use the correct
accessor for GIMPLE_OMP_TARGET clauses.
---
gcc/tree-nested.cc | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
@@ -2906,9 +2906,11 @@ convert_tramp_reference_stmt (gimple_stmt_iterator *gsi, bool *handled_ops_p,
continue;
decl = i ? get_chain_decl (info) : info->frame_decl;
/* Don't add CHAIN.* or FRAME.* twice. */
- for (c = gimple_omp_taskreg_clauses (stmt);
- c;
- c = OMP_CLAUSE_CHAIN (c))
+ if (gimple_code (stmt) == GIMPLE_OMP_TARGET)
+ c = gimple_omp_target_clauses (stmt);
+ else
+ c = gimple_omp_taskreg_clauses (stmt);
+ for (; c; c = OMP_CLAUSE_CHAIN (c))
if ((OMP_CLAUSE_CODE (c) == OMP_CLAUSE_FIRSTPRIVATE
|| OMP_CLAUSE_CODE (c) == OMP_CLAUSE_SHARED)
&& OMP_CLAUSE_DECL (c) == decl)
--
2.25.1