From 4f1a76a0d161223a5eff21662e4d87176f2b7e4e Mon Sep 17 00:00:00 2001
From: AMD Toolchain Support
 <73240730+amd-toolchain-support@users.noreply.github.com>
Date: Fri, 19 Mar 2021 05:39:20 +0530
Subject: [PATCH] fixing the perf regression issues with OpenMPI v4.0.x till
 v4.1.0 for x86_64 (#22350)

---
 .../packages/openmpi/opal_assembly_arch.patch | 27 +++++++++++++++++++
 .../repos/builtin/packages/openmpi/package.py |  3 +++
 2 files changed, 30 insertions(+)
 create mode 100644 var/spack/repos/builtin/packages/openmpi/opal_assembly_arch.patch

diff --git a/var/spack/repos/builtin/packages/openmpi/opal_assembly_arch.patch b/var/spack/repos/builtin/packages/openmpi/opal_assembly_arch.patch
new file mode 100644
index 00000000000..4fc3112567f
--- /dev/null
+++ b/var/spack/repos/builtin/packages/openmpi/opal_assembly_arch.patch
@@ -0,0 +1,27 @@
+diff --git a/opal/include/opal/sys/gcc_builtin/atomic.h b/opal/include/opal/sys/gcc_builtin/atomic.h
+index d85ff02bd6a..a465fdae5db 100644
+--- a/opal/include/opal/sys/gcc_builtin/atomic.h
++++ b/opal/include/opal/sys/gcc_builtin/atomic.h
+@@ -13,8 +13,8 @@
+  * Copyright (c) 2011      Sandia National Laboratories. All rights reserved.
+  * Copyright (c) 2014-2017 Los Alamos National Security, LLC. All rights
+  *                         reserved.
+- * Copyright (c) 2016-2017 Research Organization for Information Science
+- *                         and Technology (RIST). All rights reserved.
++ * Copyright (c) 2016-2021 Research Organization for Information Science
++ *                         and Technology (RIST).  All rights reserved.
+  * Copyright (c) 2018      Triad National Security, LLC. All rights
+  *                         reserved.
+  * $COPYRIGHT$
+@@ -61,9 +61,8 @@ static inline void opal_atomic_rmb(void)
+ {
+ #if OPAL_ASSEMBLY_ARCH == OPAL_X86_64
+     /* work around a bug in older gcc versions where ACQUIRE seems to get
+-     * treated as a no-op instead of being equivalent to
+-     * __asm__ __volatile__("": : :"memory") */
+-    __atomic_thread_fence (__ATOMIC_SEQ_CST);
++     * treated as a no-op instead */
++    __asm__ __volatile__("": : :"memory");
+ #else
+     __atomic_thread_fence (__ATOMIC_ACQUIRE);
+ #endif
diff --git a/var/spack/repos/builtin/packages/openmpi/package.py b/var/spack/repos/builtin/packages/openmpi/package.py
index 2b9d8aa0af7..c0c532f187a 100644
--- a/var/spack/repos/builtin/packages/openmpi/package.py
+++ b/var/spack/repos/builtin/packages/openmpi/package.py
@@ -188,6 +188,9 @@ class Openmpi(AutotoolsPackage):
     # The second patch was applied starting version v4.0.0 and backported to
     # v2.x, v3.0.x, and v3.1.x.
     patch('use_mpi_tkr_sizeof/step_2.patch', when='@1.8.4:2.1.3,3:3.0.1')
+    # To fix performance regressions introduced while fixing a bug in older
+    # gcc versions on x86_64, Refs. open-mpi/ompi#8603
+    patch('opal_assembly_arch.patch', when='@4.0.0:4.1.1')
 
     variant(
         'fabrics',