Add AVX512 operations

config/opal_config_asm.m4

@@ -1374,7 +1374,7 @@ AC_MSG_ERROR([Can not continue.])
 
     # Check for RDTSCP support
     result=0
-    AS_IF([test "$opal_cv_asm_arch" = "OPAL_X86_64" || test "$opal_cv_asm_arch" = "OPAL_IA32"],
+    AS_IF([test "$opal_cv_asm_arch" = "X86_64" || test "$opal_cv_asm_arch" = "IA32"],
           [AC_MSG_CHECKING([for RDTSCP assembly support])
            AC_LANG_PUSH([C])
            AC_TRY_RUN([[

ompi/mca/op/Makefile.am

@@ -17,6 +17,8 @@
 # $HEADER$
 #
 
+AM_CPPFLAGS = $(LTDLINCL)
+
 # main library setup
 noinst_LTLIBRARIES = libmca_op.la
 libmca_op_la_SOURCES =

ompi/mca/op/avx/Makefile.am

@@ -0,0 +1,69 @@
+#
+# Copyright (c) 2019      The University of Tennessee and The University
+#                         of Tennessee Research Foundation.  All rights
+#                         reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+# This component provide support for the Advanced Vector Extensions (AVX)
+# available in recent versions of x86 processors.
+#
+# See https://github.com/open-mpi/ompi/wiki/devel-CreateComponent
+# for more details on how to make Open MPI components.
+
+# First, list all .h and .c sources.  It is necessary to list all .h
+# files so that they will be picked up in the distribution tarball.
+
+AM_CPPFLAGS=$(op_avx_CPPFLAGS)
+
+sources = op_avx_component.c op_avx_functions.c \
+	op_avx_functions.h op_avx.h
+
+# Open MPI components can be compiled two ways:
+#
+# 1. As a standalone dynamic shared object (DSO), sometimes called a
+# dynamically loadable library (DLL).
+#
+# 2. As a static library that is slurped up into the upper-level
+# libmpi library (regardless of whether libmpi is a static or dynamic
+# library).  This is called a "Libtool convenience library".
+#
+# The component needs to create an output library in this top-level
+# component directory, and named either mca_<type>_<name>.la (for DSO
+# builds) or libmca_<type>_<name>.la (for static builds).  The OMPI
+# build system will have set the
+# MCA_BUILD_ompi_<framework>_<component>_DSO AM_CONDITIONAL to indicate
+# which way this component should be built.
+
+if MCA_BUILD_ompi_op_avx_DSO
+component_noinst =
+component_install = mca_op_avx.la
+else
+component_install =
+component_noinst = libmca_op_avx.la
+endif
+
+# Specific information for DSO builds.
+#
+# The DSO should install itself in $(ompilibdir) (by default,
+# $prefix/lib/openmpi).
+
+mcacomponentdir = $(ompilibdir)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_op_avx_la_SOURCES = $(sources)
+mca_op_avx_la_CFLAGS = $(op_avx_CPPFLAGS)
+mca_op_avx_la_LDFLAGS = -module -avoid-version
+
+# Specific information for static builds.
+#
+# Note that we *must* "noinst"; the upper-layer Makefile.am's will
+# slurp in the resulting .la library into libmpi.
+
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_op_avx_la_SOURCES = $(sources)
+libmca_op_avx_la_CFLAGS = $(op_avx_CPPFLAGS)
+libmca_op_avx_la_LDFLAGS = -module -avoid-version

ompi/mca/op/avx/configure.m4

@@ -0,0 +1,67 @@
+# -*- shell-script -*-
+#
+# Copyright (c) 2019      The University of Tennessee and The University
+#                         of Tennessee Research Foundation.  All rights
+#                         reserved.
+#
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+# MCA_ompi_op_avx_CONFIG([action-if-can-compile],
+#		         [action-if-cant-compile])
+# ------------------------------------------------
+# We can always build, unless we were explicitly disabled.
+AC_DEFUN([MCA_ompi_op_avx_CONFIG],[
+    op_avx_support=0
+    OPAL_VAR_SCOPE_PUSH([op_avx_cflags_save op_avx_CPPFLAGS])
+
+    AS_IF([test "$opal_cv_asm_arch" = "X86_64"],
+          [AC_LANG_PUSH([C])
+
+           # Check for AVX512 support with default flags
+           op_avx_CPPFLAGS=""
+           AC_MSG_CHECKING([for AVX512 support (no additional flags)])
+           AC_LINK_IFELSE(
+               [AC_LANG_PROGRAM([[#include <immintrin.h>]],
+                                [[
+    __m512 vA, vB;
+    _mm512_add_ps(vA, vB)
+                                ]])],
+               [op_avx_support=1
+                AC_MSG_RESULT([yes])],
+               [AC_MSG_RESULT([no])])
+
+           AS_IF([test $op_avx_support -eq 0],
+                 [AC_MSG_CHECKING([for AVX512 support (with -march=skylake-avx512)])
+                  op_avx_cflags_save="$CFLAGS"
+                  CFLAGS="$CFLAGS -march=skylake-avx512"
+                  AC_LINK_IFELSE(
+	              [AC_LANG_PROGRAM([[#include <immintrin.h>]],
+                                       [[
+    __m512 vA, vB;
+    _mm512_add_ps(vA, vB)
+                                       ]])],
+                      [op_avx_support=1
+	               op_avx_CPPFLAGS="-march=skylake-avx512"
+                       AC_MSG_RESULT([yes])],
+                      [AC_MSG_RESULT([no])])
+                  CFLAGS="$op_avx_cflags_save"
+                 ])
+           AC_LANG_POP([C])
+          ])
+
+    AC_DEFINE_UNQUOTED([OMPI_MCA_OP_HAVE_AVX512],
+		       [$op_avx_support],
+		       [Whetever AVX512 is supported in the current compilation context])
+    AC_SUBST([op_avx_CPPFLAGS])
+    OPAL_VAR_SCOPE_POP
+    AS_IF([test $op_avx_support -eq 1],
+    	  [AC_CONFIG_FILES([ompi/mca/op/avx/Makefile])
+           [$1]],
+	  [$2])
+
+])dnl

ompi/mca/op/avx/op_avx.h

@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2019      The University of Tennessee and The University
+ *                         of Tennessee Research Foundation.  All rights
+ *                         reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#ifndef MCA_OP_AVX_EXPORT_H
+#define MCA_OP_AVX_EXPORT_H
+
+#include "ompi_config.h"
+
+#include "ompi/mca/mca.h"
+#include "opal/class/opal_object.h"
+
+#include "ompi/mca/op/op.h"
+
+BEGIN_C_DECLS
+
+/**
+ * Derive a struct from the base op component struct, allowing us to
+ * cache some component-specific information on our well-known
+ * component struct.
+ */
+typedef struct {
+    /** The base op component struct */
+    ompi_op_base_component_1_0_0_t super;
+
+    /* What follows is avx-component-specific cached information.  We
+       tend to use this scheme (caching information on the avx
+       component itself) instead of lots of individual global
+       variables for the component.  The following data fields are
+       avxs; replace them with whatever is relevant for your
+       component. */
+
+    /** A simple boolean indicating whether double precision is
+        supported. */
+    bool double_supported;
+} ompi_op_avx_component_t;
+
+/**
+ * Globally exported variable.  Note that it is a *avx* component
+ * (defined above), which has the ompi_op_base_component_t as its
+ * first member.  Hence, the MCA/op framework will find the data that
+ * it expects in the first memory locations, but then the component
+ * itself can cache additional information after that that can be used
+ * by both the component and modules.
+ */
+OMPI_DECLSPEC extern ompi_op_avx_component_t
+    mca_op_avx_component;
+
+END_C_DECLS
+
+#endif /* MCA_OP_AVX_EXPORT_H */