[Flang][OpenMP] Fix COPYIN of derived types with allocatable components at -O3 (#196063) COPYIN of threadprivate derived types with allocatable components segfaults at -O3 because the OpenMP runtime zero-fills per-thread storage, leaving allocatable component descriptors with invalid metadata. This patch skips the copy on the master thread (where source and destination alias) and uses temporary_lhs assignment on worker threads so the runtime initializes descriptors before the deep copy. Assisted-by: Claude Opus 4.6 Fixes : [https://github.com/llvm/llvm-project/issues/196134](https://github.com/llvm/llvm-project/issues/196134) Minimal reprducing test-case : ``` program repro_o3_segv use omp_lib implicit none integer, parameter :: NT = 4 integer :: i, rc integer :: nThreads(NT), nThreads1(NT) type structure_1 integer, allocatable :: a(:) end type type structure_2 type(structure_1) :: struc1 end type type(structure_2), save :: struc2(2) !$omp threadprivate(struc2) rc = 0 nThreads = -999 nThreads1 = -999 ! Keep this: dynamic teams can change codegen/runtime behavior. call omp_set_dynamic(.true.) call omp_set_num_threads(NT) allocate(struc2(1)%struc1%a(2)) struc2(1)%struc1%a(1) = 1 struc2(1)%struc1%a(2) = 2 !$omp parallel copyin(struc2) if (omp_get_thread_num() == NT-1) then ! Keep branch shape from original. ! struc2(1)%struc1%a(1) = 3 end if !$omp barrier allocate(struc2(2)%struc1%a(3)) nThreads(omp_get_thread_num()+1) = struc2(1)%struc1%a(1) struc2(2)%struc1%a(1) = omp_get_thread_num()+1 struc2(2)%struc1%a(2) = omp_get_thread_num()+2 !$omp end parallel do i = 1, NT if (nThreads(i) /= 1) rc = 1 end do struc2(2)%struc1%a(2) = -1 !$omp parallel copyin(struc2) nThreads(omp_get_thread_num()+1) = struc2(2)%struc1%a(1) nThreads1(omp_get_thread_num()+1) = struc2(2)%struc1%a(2) !$omp end parallel do i = 1, NT if (nThreads1(i) /= -1) rc = 1 end do if (rc /= 0) stop 1 deallocate(struc2(1)%struc1%a) deallocate(struc2(2)%struc1%a) end program ``` [>./flang -fopenmp -fopenmp-version=50 copyin_derrived_alloct.f90 -O3 > ./a.out Segmentation fault (core dumped) Co-authored-by: Chandra Ghale <ghale@pe34genoa.hpc.amslabs.hpecorp.net>
Welcome to the LLVM project!
This repository contains the source code for LLVM, a toolkit for the construction of highly optimized compilers, optimizers, and run-time environments.
The LLVM project has multiple components. The core of the project is itself called “LLVM”. This contains all of the tools, libraries, and header files needed to process intermediate representations and convert them into object files. Tools include an assembler, disassembler, bitcode analyzer, and bitcode optimizer.
C-like languages use the Clang frontend. This component compiles C, C++, Objective-C, and Objective-C++ code into LLVM bitcode -- and from there into object files, using LLVM.
Other components include: the libc++ C++ standard library, the LLD linker, and more.
Consult the Getting Started with LLVM page for information on building and running LLVM.
For information on how to contribute to the LLVM project, please take a look at the Contributing to LLVM guide.
Join the LLVM Discourse forums, Discord chat, LLVM Office Hours or Regular sync-ups.
The LLVM project has adopted a code of conduct for participants to all modes of communication within the project.