https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121452
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to Tobias Burnus from comment #1)
> !$omp for ordered(2)
> do i = 1, 5
> do j = 1, 5
Correction: 'for' → 'do', 'do i' → 'do 10 i' and 'do j' → 'do 20 j'.
This one actually fails already in the front end:
Error: !$OMP DO inner loops must be perfectly nested with ORDERED clause at
(1)
* * *
Draft patch - seems to fix three examples in comment 0 + SPECaccel's 463.swim
Todo:
- Fix Fortran testcase of comment 1 (with fixes mentioned above)
- check that it works
- check that it doesn't prevent intended diagnostic
- check whether there are other places where GIMPLE_OMP_STRUCTURED_BLOCK
needs to be handled (missed diagnostic vs. bogus diagnostic)
--- a/gcc/omp-low.cc
+++ b/gcc/omp-low.cc
@@ -3309,3 +3309,4 @@ check_omp_nesting_restrictions (gimple *stmt, omp_context
|| gimple_code (stmt) == GIMPLE_OMP_ATOMIC_STORE
- || gimple_code (stmt) == GIMPLE_OMP_SCAN)
+ || gimple_code (stmt) == GIMPLE_OMP_SCAN
+ || gimple_code (stmt) == GIMPLE_OMP_STRUCTURED_BLOCK)
return true;
@@ -3339,3 +3340,4 @@ check_omp_nesting_restrictions (gimple *stmt, omp_context
|| gimple_omp_for_kind (stmt) != GF_OMP_FOR_KIND_SIMD)
- && gimple_code (stmt) != GIMPLE_OMP_SCAN)
+ && gimple_code (stmt) != GIMPLE_OMP_SCAN
+ && gimple_code (stmt) != GIMPLE_OMP_STRUCTURED_BLOCK)
{
