https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126279
Bug ID: 126279
Summary: [OpenMP] Fortran & multiple uses_allocators clauses
mishandled
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: openmp, wrong-code
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: burnus at gcc dot gnu.org
CC: cltang at gcc dot gnu.org
Target Milestone: ---
Created attachment 65057
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65057&action=edit
Patch to fix 'memspace' conversion + extend testcase
The following is with mainline r17-2423-g6f754f345e0621 and the attached patch
applied.
Using:
!$omp target uses_allocators( &
traits(t), memspace(omp_high_bw_mem_space) : my; &
omp_default_mem_alloc, & ! predefined allocator (= 0B)
omp_null_allocator; & ! ignored
my2; & ! = default mem space, default traits
traits(t2) : my3;
memspace(omp_large_cap_mem_space) : my4)
block
type(c_ptr) :: c1, c2, c3, c4
c1 = omp_alloc (c_sizeof(x), my)
c2 = omp_alloc (c_sizeof(x), my2)
c3 = omp_alloc (c_sizeof(x), my3)
c4 = omp_alloc (c_sizeof(x), my4)
Gives as shown in the debugger the calls
omp_alloc (size=4, allocator=140737488339600)
omp_alloc (size=8, allocator=140737488339600)
omp_alloc (size=12, allocator=140737488339600)
omp_alloc (size=16, allocator=140737488339600)
That's with
memspace = omp_large_cap_mem_space
i.e. matches 'my4'. Accordingly, 'c3' is allocated with
allocator_data->alignment == 8 – instead of with an alignment of 1024,
such that the alignment check fails at runtime!
EXPECTED: 4 different allocators are used - and c3 has an alignment of 1024
bit.
I think the original + gimple dump are okay.
Besides the issue of above, I noticed in the omp-lower dump
the following:
D.5004 = (unsigned long) &D.5002;
uses_allocator_descr.49[2] = D.5004;
This one is pretty bad as there is no 'D.5002 = ...' in the code.
