Powerpc atomics: Force usage of powerpc assembly.

The builtins used by default on Power have been
shown to perform poorly. For the time being, force
all compilers to use the inline assembly until
atomic builtins catch-up.

This changes the defaults for all compilers sans xl, including:
gcc, clang, and pgi to use the assembly.

Previously, all of the above were using C11 or
the gcc builtins.

Bonus:
Add a configure flag to force Power machines to use
the builtins/C11, depending on what is available. This
will make future testing easier.

Signed-off-by: Austen Lauria <[email protected]>

Author: awlauria

config/opal_config_asm.m4

@@ -1192,6 +1192,18 @@ AC_DEFUN([OPAL_CONFIG_ASM],[
                 AC_MSG_ERROR([Could not determine PowerPC word size: $ac_cv_sizeof_long])
             fi
             OPAL_GCC_INLINE_ASSIGN='"1: li %0,0" : "=&r"(ret)'
+
+            # See the following github PR and some performance numbers/discussion:
+            # https://github.com/open-mpi/ompi/pull/8649
+            AC_MSG_CHECKING([$opal_cv_asm_arch: Checking if force gcc atomics requested])
+            if test $force_gcc_atomics_ppc = 0 ; then
+                AC_MSG_RESULT([no])
+                opal_cv_asm_builtin="BUILTIN_NO"
+            else
+                AC_MSG_RESULT([Yes])
+                AC_MSG_WARN([$opal_cv_asm_arch: gcc atomics have been known to perform poorly on powerpc.])
+            fi
+
             ;;
         *)
 	    if test $opal_cv_have___atomic = "yes" ; then

config/opal_configure_options.m4

@@ -84,6 +84,14 @@ else
     WANT_BRANCH_PROBABILITIES=0
 fi
 
+AC_ARG_ENABLE([builtin-atomics-for-ppc],[AS_HELP_STRING([--enable-builtin-atomics-for-ppc],
+                  [POWER architectures only: Force use of builtin atomics if available. This could either be gcc builtins or C11 atomics, depending on what is available on your system. Enabling this is known to cause poor performance in atomic operations on Power machines. (default: disabled)])])
+if test "x$enable_builtin_atomics_for_ppc" = "xyes" ; then
+force_gcc_atomics_ppc=1
+else
+force_gcc_atomics_ppc=0
+fi
+
 
 #
 # Memory debugging