[gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/files/, dev-util/Tensile/

["Patrick Lauer" <patrick@gentoo.org>]

commit:     7fb8049f065a12165c5664e74b7cb422f1346949
Author:     Patrick Lauer <patrick <AT> gentoo <DOT> org>
AuthorDate: Sat Nov  9 21:04:45 2024 +0000
Commit:     Patrick Lauer <patrick <AT> gentoo <DOT> org>
CommitDate: Sat Nov  9 21:08:24 2024 +0000
URL:        https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=7fb8049f

dev-util/Tensile: Fix runtime behaviour

Similar to https://github.com/ROCm/Tensile/pull/1898

Fixes rocBLAS failing, no idea why above code was
merged but fell out during refactoring and why is
everything so horrible

Signed-off-by: Patrick Lauer <patrick <AT> gentoo.org>

 dev-util/Tensile/Tensile-6.1.1-r1.ebuild           | 140 +++++++++++++++++++++
 .../files/Tensile-6.1.1-ignore-asm-cap.patch       |  11 ++
 2 files changed, 151 insertions(+)

diff --git a/dev-util/Tensile/Tensile-6.1.1-r1.ebuild b/dev-util/Tensile/Tensile-6.1.1-r1.ebuild
new file mode 100644
index 000000000000..f10408b60b93
--- /dev/null
+++ b/dev-util/Tensile/Tensile-6.1.1-r1.ebuild
@@ -0,0 +1,140 @@
+# Copyright 1999-2024 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{10..12} )
+DISTUTILS_USE_PEP517=setuptools
+ROCM_VERSION=${PV}
+LLVM_COMPAT=( 18 )
+
+inherit cmake distutils-r1 llvm-r1 prefix rocm
+
+DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions"
+HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile"
+SRC_URI="https://github.com/ROCmSoftwarePlatform/Tensile/archive/rocm-${PV}.tar.gz -> rocm-Tensile-${PV}.tar.gz"
+S="${WORKDIR}/${PN}-rocm-${PV}"
+
+LICENSE="MIT"
+SLOT="0/$(ver_cut 1-2)"
+KEYWORDS="~amd64"
+IUSE="client test"
+REQUIRED_USE="client? ( ${ROCM_REQUIRED_USE} )"
+
+RESTRICT="!test? ( test )"
+
+RDEPEND="${PYTHON_DEPS}
+	client? ( dev-libs/boost )
+	>=dev-cpp/msgpack-cxx-6.0.0
+	dev-python/pyyaml[${PYTHON_USEDEP}]
+	dev-python/msgpack[${PYTHON_USEDEP}]
+	dev-python/joblib[${PYTHON_USEDEP}]
+	=dev-util/hip-6*
+	>=dev-util/rocm-smi-4.3.0
+	$(llvm_gen_dep '
+		sys-devel/clang:${LLVM_SLOT}
+	')
+"
+DEPEND="${RDEPEND}"
+BDEPEND="
+	test? (
+		dev-python/pytest-forked[${PYTHON_USEDEP}]
+		dev-python/pytest-xdist[${PYTHON_USEDEP}]
+		dev-python/filelock[${PYTHON_USEDEP}]
+		dev-python/joblib[${PYTHON_USEDEP}]
+	)
+"
+
+distutils_enable_tests pytest
+
+PATCHES=(
+	"${FILESDIR}"/${PN}-4.3.0-output-commands.patch
+	"${FILESDIR}"/${PN}-5.4.2-fix-arch-parse.patch
+	"${FILESDIR}"/${PN}-5.4.2-use-ninja.patch
+	"${FILESDIR}"/${PN}-6.1.1-fix-msgpack-dependency.patch
+	"${FILESDIR}"/${PN}-6.0.2-expand-isa-compatibility.patch
+	"${FILESDIR}"/${PN}-6.1.1-ignore-asm-cap.patch
+)
+
+CMAKE_USE_DIR="${S}/${PN}/Source"
+
+src_prepare() {
+	distutils-r1_src_prepare
+	sed -e "s,\@LLVM_PATH\@,$(get_llvm_prefix),g" \
+		"${FILESDIR}"/${PN}-5.7.1-gentoopath.patch > "${S}"/gentoopath.patch || die
+	eapply $(prefixify_ro "${S}"/gentoopath.patch)
+
+	pushd ${PN} || die
+
+	sed -e "/ROCM_SMI_ROOT/s,lib,$(get_libdir)," \
+		-i Source/cmake/FindROCmSMI.cmake || die
+	sed -r -e "/TENSILE_USE_LLVM/s/ON/OFF/" \
+		-i Source/CMakeLists.txt || die
+
+	# ${Tensile_ROOT}/bin does not exists; call command directly
+	sed -e "s,\${Tensile_ROOT}/bin/,,g" -i cmake/TensileConfig.cmake || die
+
+	local Tensile_share_dir="\"${EPREFIX}/usr/share/${PN}\""
+	sed -e "/HipClangVersion/s/0.0.0/$(hipconfig -v)/" -i Common.py || die
+
+	sed -e "s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir},g" \
+		-i ReplacementKernels.py Common.py ${PN}.py || die
+
+	sed -e "s|os\.path\.dirname.*$|\"${EPREFIX}/usr/share/Tensile/Source\", end='')|" -i __init__.py || die
+
+	popd || die
+
+	sed -e "/package_data/d" -e "/data_files/d" -i setup.py || die
+	use client && PATCHES= cmake_src_prepare  # do not apply patches again in cmake_src_prepare
+}
+
+src_configure() {
+	rocm_use_hipcc
+
+	distutils-r1_src_configure
+	if use client; then
+		local mycmakeargs=(
+			-DCMAKE_SKIP_RPATH=ON
+			-DTENSILE_USE_MSGPACK=ON
+			-DTENSILE_USE_LLVM=ON
+			-DTensile_LIBRARY_FORMAT=msgpack
+			-DAMDGPU_TARGETS="$(get_amdgpu_flags)"
+		)
+		cmake_src_configure
+	fi
+}
+
+src_compile() {
+	distutils-r1_src_compile
+	use client && cmake_src_compile
+}
+
+python_install() {
+	distutils-r1_python_install
+
+	python_moduleinto Tensile
+	pushd Tensile || die
+	python_domodule Components
+	python_newexe Utilities/merge.py ${PN}-merge
+}
+
+src_install() {
+	distutils-r1_src_install
+
+	pushd ${PN} || die
+	insinto /usr/share/${PN}
+	doins -r Configs Perf Source CustomKernels
+	insinto /usr/$(get_libdir)/cmake/${PN}
+	doins cmake/*.cmake
+
+	if use client; then
+		pushd "${BUILD_DIR}" || die
+		dobin client/tensile_client
+	fi
+}
+
+# Test suite fails to start without this
+python_test() {
+	export ROCM_PATH="${EPREFIX}/usr"
+	epytest
+}

diff --git a/dev-util/Tensile/files/Tensile-6.1.1-ignore-asm-cap.patch b/dev-util/Tensile/files/Tensile-6.1.1-ignore-asm-cap.patch
new file mode 100644
index 000000000000..b0ad3b276fba
--- /dev/null
+++ b/dev-util/Tensile/files/Tensile-6.1.1-ignore-asm-cap.patch
@@ -0,0 +1,11 @@
+--- a/Tensile/Common.py	2024-11-09 20:38:11.186998794 -0000
++++ b/Tensile/Common.py	2024-11-09 20:43:18.524804845 -0000
+@@ -2030,7 +2030,7 @@
+     if len(compilerVer) >= 2:
+       ignoreCacheCheck = ignoreCacheCheck or \
+                          compilerVer[0] < 5 or \
+-                         (compilerVer[0] == 5 and compilerVer[1] <= 2) 
++                         (compilerVer[0] == 6 and compilerVer[1] <= 1)
+       
+     if not derivedAsmCaps["SupportedISA"] and CACHED_ASM_CAPS[isaVersion]["SupportedISA"]:
+       printWarning("Architecture {} not supported by ROCm {}".format(isaVersion, globalParameters['HipClangVersion']))