commit: e1704cb77eb176882afe45c9b65875e2d56c4034
Author: Patrick Lauer <patrick <AT> gentoo <DOT> org>
AuthorDate: Fri Feb 7 15:36:15 2025 +0000
Commit: Patrick Lauer <patrick <AT> gentoo <DOT> org>
CommitDate: Sat Feb 8 10:07:08 2025 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=e1704cb7
dev-util/Tensile: add 6.3.2
Signed-off-by: Patrick Lauer <patrick <AT> gentoo.org>
dev-util/Tensile/Manifest | 1 +
dev-util/Tensile/Tensile-6.3.2.ebuild | 140 ++++++++++++++++++++++++++++++++++
2 files changed, 141 insertions(+)
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
index a0bbe00c22e7..d6587e93f1f8 100644
--- a/dev-util/Tensile/Manifest
+++ b/dev-util/Tensile/Manifest
@@ -4,3 +4,4 @@ DIST rocm-Tensile-5.4.2.tar.gz 14413805 BLAKE2B
ca1c13785164439f6fe83f16b36b7a0b
DIST rocm-Tensile-5.7.1.tar.gz 12396002 BLAKE2B
ec2c2bc3f07fe22abd800b642c998ea80b9e4e2d0d28c4763d574dbd2fcb14c735223ddc7571d4ab3a48f0bd819cc835c16df9882049c5bf56f25ef729e6d654
SHA512
8f356f36cfec74ee9b000d1180cf52077dc5c595e5c5a40792fc3eb13f3139980c7440f9bc912218f946cf8137e2481d23fe39f110511c1bb55fe85a933ffd36
DIST rocm-Tensile-6.1.1.tar.gz 12675289 BLAKE2B
dfe3698d7dda14d5b20a6fe1ef6535ac9737ac330b71678a3a94e939adc2350e2dcf33e58c711ee90dae543efba59fe210bfa39fc69b63910ec4537db3f209bf
SHA512
cf8b014b1ff7cd5d45236b8a9a6f5604b02e34a9245a163c672847b535d565151bf7e3a5127601681e80e6fe26964a0e544594044efe2fe68e39b0f1128f96e2
DIST rocm-Tensile-6.3.0.tar.gz 12724004 BLAKE2B
83887efe8ab4102b9d33bae1f97a245d40bcc186635760de509079218628fd4d57072c6c2c1726fec6d8603d40e760e37fa1a443eead290d121cac099ae779ff
SHA512
4c80a269236690545d8bc9090b799c5b4095ed9152a1e19ff8b2e5decd588545641ccdf3c6c469b12f4de29b3f4c2821a6e4714460771250b05339ac32f57c13
+DIST rocm-Tensile-6.3.2.tar.gz 12723907 BLAKE2B
aaabae87664340f3e044ac5fdc68331084cd597466f294d60f8ed3d523c0f1ebc0c6e3e353585d4bb85f4fe30c644b1a70f2123b6f62d67a6b7a1904f3d820bb
SHA512
df8888d7250f9e3dc6818590c82a3e1dc3ff787e0eac29e45969ef408db92261d39faca29f5ac7a6bae5e5b614365ff6e37d087956697c25c65d9d7e4c1ba74d
diff --git a/dev-util/Tensile/Tensile-6.3.2.ebuild
b/dev-util/Tensile/Tensile-6.3.2.ebuild
new file mode 100644
index 000000000000..2b811c879756
--- /dev/null
+++ b/dev-util/Tensile/Tensile-6.3.2.ebuild
@@ -0,0 +1,140 @@
+# Copyright 1999-2025 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{10..13} )
+DISTUTILS_USE_PEP517=setuptools
+ROCM_VERSION=${PV}
+LLVM_COMPAT=( 19 )
+
+inherit cmake distutils-r1 llvm-r1 prefix rocm
+
+DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions"
+HOMEPAGE="https://github.com/ROCm/Tensile"
+SRC_URI="https://github.com/ROCm/Tensile/archive/rocm-${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? ( ${ROCM_REQUIRED_USE} )"
+
+# tests can freeze machine depending on gpu/kernel
+RESTRICT="test"
+
+RDEPEND="${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[${PYTHON_USEDEP}]
+ dev-util/hip:${SLOT}
+ dev-util/rocm-smi:${SLOT}
+ $(llvm_gen_dep '
+ llvm-core/clang:${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}"/${PN}-5.4.2-fix-arch-parse.patch
+ "${FILESDIR}"/${PN}-6.3.0-use-ninja.patch
+ "${FILESDIR}"/${PN}-6.1.1-fix-msgpack-dependency.patch
+ "${FILESDIR}"/${PN}-6.0.2-expand-isa-compatibility.patch
+)
+
+CMAKE_USE_DIR="${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_domodule Utilities
+ python_domodule TensileCreateLib
+}
+
+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
+}