[Bug target/120450] GCN offloading: ICE 'during RTL pass: expand', 'in assign_temp, at function.cc:990' for C++ 'std::flat_map', 'std::flat_multimap'

[tschwinge at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120450

Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=122268
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2025-10-20
             Status|UNCONFIRMED                 |NEW

--- Comment #4 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
Via commit r16-4212-gf256a13f8aed833fe964a2ba541b7b30ad9b4a76 "c++, gimplify:
Implement C++26 P2795R5 - Erroneous behavior for uninitialized reads
[PR114457]", 'libgomp.c++/target-std__flat_map-concurrent.C' and
'libgomp.c++/target-std__flat_multimap-concurrent.C' now PASS on trunk (but
presumably still ICE on releases/gcc-15 branch).  And in return, we've got
'libgomp.c++/target-flex-101.C' ICEing for GCN and nvptx offloading
compilation: PR122268 "[OpenMP][offload:gcn/nvptx lto1] ICE (during RTL pass:
expand) in assign_temp, at function.cc:990 for libgomp.c++/target-flex-101.C
(c++/16.0.0/flat_map:185:1) since r16-4212-gf256a13f8aed83".

The "simple '-O0' reproducer" posted above still ICEs, too -- now for both GCN
and nvptx offloading compilation.

Chances are that it's all the same underlying issue.  (I've not yet had a
chance to look into Richi's suggestions.)