Thread-MPI fixes for i386 llvm & simplification of atomics
authorSander Pronk <pronk@kth.se>
Sat, 29 Dec 2012 17:29:23 +0000 (18:29 +0100)
committerSander Pronk <pronk@kth.se>
Sat, 29 Dec 2012 17:29:23 +0000 (18:29 +0100)
commit648a6fa785d3c50f5ff29e2038cd80843208d89e
treec732f830550171b33a2d83ff297d2a858dc2d52b
parent492378e3403d6cf4e9021a258ef9fb02488da4cb
Thread-MPI fixes for i386 llvm & simplification of atomics

llvm on 32-bit x86 now uses the gcc-style intrinsics rather than
inline assembly; this fixes http://redmine.gromacs.org/issues/1098

Also, atomics use in thread-MPI is simplified to use only add-return
rather than fetch-add and add-return.

Change-Id: Id7edba926957b0d708e59ffd42db7c2b38edbdfc
Fixes: 1098
include/thread_mpi/atomic/gcc_x86.h
src/gmxlib/thread_mpi/collective.c
src/gmxlib/thread_mpi/once.c
src/gmxlib/thread_mpi/tmpi_init.c