[all-commits] [llvm/llvm-project] fc14d0: [flang][cuda] Preserve allocator index when splitt...

[Valentin Clement (バレンタイン クレメン) via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: fc14d058e00f93b2a6416547fb3f8f94cd3ed9eb
      https://github.com/llvm/llvm-project/commit/fc14d058e00f93b2a6416547fb3f8f94cd3ed9eb
  Author: Valentin Clement (バレンタイン クレメン) <clementval at gmail.com>
  Date:   2026-09-22 (Tue, 22 Sep 2026)

  Changed paths:
    M flang/lib/Optimizer/Builder/MutableBox.cpp
    M flang/test/HLFIR/separate-allocatable-assign.fir

  Log Message:
  -----------
  [flang][cuda] Preserve allocator index when splitting host-accessible realloc assigns (#225537)

After #224195, shape-changing assignment to a pinned/managed/unified
allocatable inlined cuf.alloc/cuf.free but rebuilt the descriptor
without
allocator_idx. Implicit DEALLOCATE then used the default allocator and
called libc free() on CUDA storage.

Pass the CUDA allocator index through MutablePropertyWriter so fir.embox
keeps pinned=1, managed=3, and unified=4.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications