Changes for ROCm-5.2.0 changes and new recipe rocwmma (#31667)

* Changes for ROCm-5.2.0 changes and new recipe rocwmma

* modify the maintainers for hipify-clang

* address review comments

* update the rocwmma new recipe as per latest syntax

* fix style errors

* modify the patch file to provide the details about the patch

* fix style errors

var/spack/repos/builtin/packages/hipify-clang/0001-install-hipify-clang-in-bin-dir-and-llvm-clangs-head.patch

@@ -0,0 +1,38 @@
+From 5af7f4a3dcb24d935208b3d598ea7e864b729fb7 Mon Sep 17 00:00:00 2001
+From: sreenivasa murthy kolam <sreenivasamurthy.kolam@amd.com>
+Date: Sat, 13 Aug 2022 04:51:21 +0000
+Subject: [PATCH] install hipify-clang in bin dir and llvm-clangs headers in
+ the include headers this will fix the issue
+ https://github.com/spack/spack/issues/30711
+
+---
+ CMakeLists.txt | 6 +++---
+ 1 file changed, 3 insertions(+), 3 deletions(-)
+
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index 6ba9a54..b624f81 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -101,16 +101,16 @@ endif()
+ 
+ set(HIPIFY_INSTALL_PATH ${CMAKE_INSTALL_PREFIX})
+ 
+-install(TARGETS hipify-clang DESTINATION ${CMAKE_INSTALL_PREFIX})
++install(TARGETS hipify-clang DESTINATION ${CMAKE_INSTALL_PREFIX}/bin)
+ install(
+     DIRECTORY ${CMAKE_SOURCE_DIR}/bin/
+-    DESTINATION ${CMAKE_INSTALL_PREFIX}
++    DESTINATION ${CMAKE_INSTALL_PREFIX}/bin
+     USE_SOURCE_PERMISSIONS
+     PATTERN "hipify-perl"
+     PATTERN "*.sh")
+ install(
+     DIRECTORY ${LLVM_DIR}/../../clang/${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}.${LLVM_VERSION_PATCH}/
+-    DESTINATION ${CMAKE_INSTALL_PREFIX}
++    DESTINATION ${CMAKE_INSTALL_PREFIX}/include
+     COMPONENT clang-resource-headers
+     FILES_MATCHING
+     PATTERN "*.h"
+-- 
+2.18.4
+

var/spack/repos/builtin/packages/hipify-clang/package.py

@@ -15,10 +15,11 @@ class HipifyClang(CMakePackage):
     url = "https://github.com/ROCm-Developer-Tools/HIPIFY/archive/rocm-5.1.3.tar.gz"
     tags = ["rocm"]
 
-    maintainers = ["srekolam", "arjun-raj-kuppala"]
+    maintainers = ["srekolam", "renjithravindrankannath"]
 
     version("master", branch="master")
 
+    version("5.2.0", sha256="dcd5f44daceb984bb654a209e78debf81e1cdeaf9202444a1e110b45ad6c3f4f")
     version("5.1.3", sha256="6354b08b8ab2f4c481398fb768652bae00bb78c4cec7a11d5f6c7e4cb831ddf1")
     version("5.1.0", sha256="ba792294cbdcc880e0f02e38ee352dff8d4a2c183430e13d1c5ed176bd46cfc5")
     version("5.0.2", sha256="812bccfeb044483a1c7df89f45843afcb28d8146f348c792f082b693cbff3984")
@@ -82,6 +83,10 @@ class HipifyClang(CMakePackage):
         values=("Release", "Debug", "RelWithDebInfo"),
         description="CMake build type",
     )
+    # the patch was added to install the targets in the correct directory structure
+    # this will fix the issue https://github.com/spack/spack/issues/30711
+
+    patch("0001-install-hipify-clang-in-bin-dir-and-llvm-clangs-head.patch", when="@5.1.0:")
 
     depends_on("cmake@3.5:", type="build")
     for ver in [
@@ -101,6 +106,7 @@ class HipifyClang(CMakePackage):
         "5.0.2",
         "5.1.0",
         "5.1.3",
+        "5.2.0",
         "master",
     ]:
         depends_on("llvm-amdgpu@" + ver, when="@" + ver)

var/spack/repos/builtin/packages/rocm-openmp-extras/package.py

@@ -29,6 +29,7 @@
     "64669ac448c439e89ec4b6e0506158e6d9b5a3edfae05882aee4c9bcd5f187b7",
     "e69fe0c933cb30daafe49d9f1df71fe16f387e0287bba921995feeefdf9ac262",
     "8bab3d621343f419b29043ac0cb56e062f114991dc3ec1e33e786f771deecc8f",
+    "20e21312816272222d1f427ea72a99a9a67077078552f5e2638a40860d161d25",
 ]
 
 devlib = [
@@ -45,6 +46,7 @@
     "49cfa8f8fc276ba27feef40546788a2aabe259a924a97af8bef24e295d19aa5e",
     "47dbcb41fb4739219cadc9f2b5f21358ed2f9895ce786d2f7a1b2c4fd044d30f",
     "c41958560ec29c8bf91332b9f668793463904a2081c330c0d828bf2f91d4f04e",
+    "901674bc941115c72f82c5def61d42f2bebee687aefd30a460905996f838e16c",
 ]
 
 llvm = [
@@ -61,6 +63,7 @@
     "99a14394b406263576ed3d8d10334de7c78d42b349109f375d178b11492eecaf",
     "db5d45c4a7842a908527c1b7b8d4a40c688225a41d23cfa382eab23edfffdd10",
     "d236a2064363c0278f7ba1bb2ff1545ee4c52278c50640e8bb2b9cfef8a2f128",
+    "0f892174111b78a02d1a00f8f46d9f80b9abb95513a7af38ecf2a5a0882fe87f",
 ]
 
 flang = [
@@ -77,6 +80,7 @@
     "9b9a53150009ff58bd0ab665b970dbebc51be891343fd5dc8e77a2133ac44333",
     "d95e36f3b93097ab6fb319c744ddc71cd94af0c358accc1e5224c2bbd431266d",
     "d7847b5c6e1344dc0b4723dbe76a859257b4c242644dedb34e425f07738530d4",
+    "20f48cac9b58496230fa2428eba4e15ec0a6e92d429569b154a328b7a8c5da17",
 ]
 
 extras = [
@@ -93,6 +97,7 @@
     "a4affb77bebaafb6f8d22c51d66aa6fa05381ec54cc1e14a4b10e0f3dc00157f",
     "c3a2a83d8f586ee765df96a692ebe010631446f700273fa31738ea260dfc35f7",
     "2e3151a47d77166d071213af2a1691487691aae0abd5c1718d818a6d7d09cb2d",
+    "817c2e8975e56a8875ff56f9d1ea34d5e7e50f1b541b7f1236e3e5c8d9eee47f",
 ]
 
 versions = [
@@ -109,6 +114,7 @@
     "5.0.2",
     "5.1.0",
     "5.1.3",
+    "5.2.0",
 ]
 versions_dict = dict()  # type: Dict[str,Dict[str,str]]
 components = ["aomp", "devlib", "llvm", "flang", "extras"]
@@ -126,10 +132,11 @@ class RocmOpenmpExtras(Package):
     """OpenMP support for ROCm LLVM."""
 
     homepage = tools_url + "/aomp"
-    url = tools_url + "/aomp/archive/rocm-5.1.3.tar.gz"
+    url = tools_url + "/aomp/archive/rocm-5.2.0.tar.gz"
     tags = ["rocm"]
 
-    maintainers = ["srekolam", "arjun-raj-kuppala", "estewart08"]
+    maintainers = ["srekolam", "renjithravindrankannath", "estewart08"]
+    version("5.2.0", sha256=versions_dict["5.2.0"]["aomp"])
     version("5.1.3", sha256=versions_dict["5.1.3"]["aomp"])
     version("5.1.0", sha256=versions_dict["5.1.0"]["aomp"])
     version("5.0.2", sha256=versions_dict["5.0.2"]["aomp"])
@@ -166,6 +173,7 @@ class RocmOpenmpExtras(Package):
         "5.0.2",
         "5.1.0",
         "5.1.3",
+        "5.2.0",
     ]:
         depends_on("hsakmt-roct@" + ver, when="@" + ver)
         depends_on("comgr@" + ver, when="@" + ver)
@@ -258,13 +266,13 @@ def patch(self):
         else:
             plugin = "/plugins/amdgpu/CMakeLists.txt"
 
+        if self.spec.version < Version("5.2.0"):
             filter_file(
                 "{ROCM_DIR}/amdgcn/bitcode",
                 "{DEVICE_LIBS_DIR}",
                 aomp_extras.format(src) + "/aompextras/CMakeLists.txt",
                 libomptarget.format(src) + "/deviceRTLs/amdgcn/CMakeLists.txt",
             )
-
         # Libm moved into llvm-project in 4.5.0
         if self.spec.version < Version("4.5.0"):
             filter_file(
@@ -289,7 +297,7 @@ def patch(self):
                 libomptarget.format(src) + "/deviceRTLs/amdgcn/CMakeLists.txt",
             )
 
-        if self.spec.version >= Version("4.5.0"):
+        if self.spec.version <= Version("5.1.3"):
             filter_file(
                 "{ROCM_DIR}/amdgcn/bitcode",
                 "{DEVICE_LIBS_DIR}",

var/spack/repos/builtin/packages/rocwmma/package.py

@@ -0,0 +1,85 @@
+# Copyright 2013-2022 Lawrence Livermore National Security, LLC and other
+# Spack Project Developers. See the top-level COPYRIGHT file for details.
+#
+# SPDX-License-Identifier: (Apache-2.0 OR MIT)
+
+import itertools
+
+from spack.package import *
+
+
+class Rocwmma(CMakePackage):
+    """AMD's C++ library for accelerating mixed precision matrix multiplication
+    and accumulation (MFMA) operations leveraging specialized GPU matrix cores.
+    rocWMMA provides a C++ API to facilitate breaking down matrix multiply-accumulate
+    problems into fragments and using them in block-wise operations that are
+    distributed in parallel across GPU wavefronts. The API is a header library
+    of GPU device code meaning that matrix core acceleration may be compiled directly
+    into your kernel device code. This can benefit from compiler optimization in the
+    generation of kernel assembly, and does not incur additional overhead costs of
+    linking to external runtime libraries or having to launch separate kernels."""
+
+    homepage = "https://github.com/ROCmSoftwarePlatform/rocWMMA"
+    git = "https://github.com/ROCmSoftwarePlatform/rocWMMA.git"
+    url = "https://github.com/ROCmSoftwarePlatform/rocWMMA/archive/refs/tags/rocm-5.2.0.tar.gz"
+    tags = ["rocm"]
+
+    maintainers = ["srekolam"]
+
+    version("5.2.0", sha256="257ccd1cf2bc1d8064e72e78d276ef7446b2cb7e2dec05ff8331bb44eff2b7cb")
+
+    # gfx908:xnack-;gfx90a:xnack-;gfx90a:xnack+
+    # are only targets currently supported for @5.2.0
+    # releases
+
+    amdgpu_targets = ("gfx908:xnack-", "gfx90a", "gfx90a:xnack-", "gfx90a:xnack+")
+    variant("amdgpu_target", values=auto_or_any_combination_of(*amdgpu_targets))
+    variant(
+        "build_type",
+        default="Release",
+        values=("Release", "Debug", "RelWithDebInfo"),
+        description="CMake build type",
+    )
+
+    depends_on("cmake@3.16:", type="build", when="@5.2.0:")
+    depends_on("cmake@3.5:", type="build")
+
+    depends_on("googletest@1.10.0:", type="test")
+
+    for ver in ["5.2.0"]:
+        depends_on("rocm-cmake@%s:" % ver, type="build", when="@" + ver)
+        depends_on("llvm-amdgpu@" + ver, type="build", when="@" + ver)
+        depends_on("hip@" + ver, when="@" + ver)
+        depends_on("rocblas@" + ver, type="build", when="@" + ver)
+
+        depends_on("rocm-openmp-extras@" + ver, type="build", when="@" + ver)
+
+    for tgt in itertools.chain(["auto"], amdgpu_targets):
+        depends_on("rocblas amdgpu_target={0}".format(tgt), when="amdgpu_target={0}".format(tgt))
+
+    def setup_build_environment(self, env):
+        env.set("CXX", self.spec["hip"].hipcc)
+
+    def cmake_args(self):
+        args = [
+            self.define("ROCWMMA_BUILD_TESTS", "ON"),
+            self.define("ROCWMMA_BUILD_VALIDATION_TESTS", "ON"),
+            self.define("ROCWMMA_BUILD_BENCHMARK_TESTS", "ON"),
+            self.define("ROCWMMA_BUILD_SAMPLES", "ON"),
+            self.define("ROCWMMA_BUILD_DOCS", "OFF"),
+            self.define("ROCWMMA_BUILD_ASSEMBLY", "OFF"),
+        ]
+        args.extend(
+            [
+                "-DOpenMP_CXX_FLAGS=-fopenmp=libomp",
+                "-DOpenMP_CXX_LIB_NAMES=libomp",
+                "-DOpenMP_libomp_LIBRARY={0}/lib/libomp.so".format(
+                    self.spec["rocm-openmp-extras"].prefix
+                ),
+            ]
+        )
+        tgt = self.spec.variants["amdgpu_target"]
+        if "auto" not in tgt:
+            args.append(self.define_from_variant("AMDGPU_TARGETS", "amdgpu_target"))
+
+        return args