sci-libs/rocFFT: bump version to 5.0.2

Enable benchmark and perfscript USE flags.

Closes: https://github.com/gentoo/gentoo/pull/24726
Package-Manager: Portage-3.0.30, Repoman-3.0.3
Signed-off-by: Yiyang Wu <xgreenlandforwyy@gmail.com>
Signed-off-by: Benda Xu <heroxbd@gentoo.org>

5 files changed, 219 insertions, 0 deletions

diff --git a/sci-libs/rocFFT/Manifest b/sci-libs/rocFFT/Manifest
index 8757f213cfe1..a6b401c669d5 100644
--- a/sci-libs/rocFFT/Manifest
+++ b/sci-libs/rocFFT/Manifest
@@ -1,2 +1,3 @@
 DIST rocFFT-4.0.0.tar.gz 772176 BLAKE2B f08db9487449e707ba814a66d23992f68c140017a38f72ee45b3b224fc3e46abd40b976aef645f4432601265b38cff8939456981c686360eba135349b5d28fb5 SHA512 db9a6b49162dfb7924aa5f5dc8d42c92a5a741e03c3c0b4ece2cfb72de7506a345f3d28d31a3d7755cfe78078c448c9f952ca671e472335a4301de8702916176
 DIST rocFFT-4.3.0.tar.gz 840748 BLAKE2B 236a5faa930751d35e501a9a32ecc8b4075448ef21a0ffb0bb7ec04f4ed1cb67b908aa153b5b424e0823d3e306cf375d7c01a77d687220ed79de6d374037e003 SHA512 765d41e4018a04a53e847c03de80eec830d3e8766187fb8cb493624afa2673175347c16e6eeeebd4ad7d88bb866e58f7e572708f22238723c710b98565a0cd3d
+DIST rocFFT-5.0.2.tar.gz 837508 BLAKE2B b74d0c6edd7b232f775bffefbca829d88d9d7cd9108c27a0698f45d7fe2760d40463bcda8c710e923874b3bbfb1e83e0fe8622db4dfa28afb29392b3e135e9a5 SHA512 d6810a92a10a27dbcac8a04e8fcd56c7fa198f208da626d8938d5ff28b6ab8b89a058c7a56b5666464165e8f6bd7c4c031a94315625a3613eb02a9b8eccd9eb9
diff --git a/sci-libs/rocFFT/files/rocFFT-5.0.2-add-math-header.patch b/sci-libs/rocFFT/files/rocFFT-5.0.2-add-math-header.patch
new file mode 100644
index 000000000000..1a35126f1d8c
--- /dev/null
+++ b/sci-libs/rocFFT/files/rocFFT-5.0.2-add-math-header.patch
@@ -0,0 +1,12 @@
+fix compilation issue: error: use of undeclared identifier 'ceil' 'floor'
+===================================================================
+--- rocFFT-rocm-5.0.2.orig/library/src/device/generator/stockham_gen_base.h
++++ rocFFT-rocm-5.0.2/library/src/device/generator/stockham_gen_base.h
+@@ -20,6 +20,7 @@
+ 
+ #pragma once
+ #include "stockham_gen.h"
++#include <math.h>
+ 
+ // Base class for stockham kernels.  Subclasses are responsible for
+ // different tiling types.
diff --git a/sci-libs/rocFFT/files/rocFFT-5.0.2-unbundle-sqlite.patch b/sci-libs/rocFFT/files/rocFFT-5.0.2-unbundle-sqlite.patch
new file mode 100644
index 000000000000..4da2b2092906
--- /dev/null
+++ b/sci-libs/rocFFT/files/rocFFT-5.0.2-unbundle-sqlite.patch
@@ -0,0 +1,54 @@
+Disable rocFFT fetching SQLite3 and link to system libsqlite.so
+--- rocFFT-rocm-5.0.2/library/src/CMakeLists.txt
++++ rocFFT-rocm-5.0.2/library/src/CMakeLists.txt
+@@ -134,31 +134,7 @@ set( rocfft_source
+   rtcsubprocess.cpp
+   )
+ 
+-# SQLite 3.36.0 enabled the backup API by default, which we need
+-# for cache serialization.  We also want to use a static SQLite,
+-# and distro static libraries aren't typically built
+-# position-independent.
+-include( FetchContent )
+-
+-# embed SQLite
+-FetchContent_Declare(sqlite_local
+-  URL https://sqlite.org/2021/sqlite-amalgamation-3360000.zip
+-  URL_HASH SHA256=999826fe4c871f18919fdb8ed7ec9dd8217180854dd1fe21eea96aed36186729
+-)
+-FetchContent_MakeAvailable(sqlite_local)
+-add_library( sqlite3 STATIC ${sqlite_local_SOURCE_DIR}/sqlite3.c )
+-set_target_properties( sqlite3 PROPERTIES
+-  C_VISIBILITY_PRESET "hidden"
+-  VISIBILITY_INLINES_HIDDEN ON
+-  POSITION_INDEPENDENT_CODE ON
+-  )
+-
+-# we don't need extensions, and omitting them from SQLite removes the
+-# need for dlopen/dlclose from within rocFFT
+-target_compile_options(
+-  sqlite3
+-  PRIVATE -DSQLITE_OMIT_LOAD_EXTENSION
+-)
++find_package( SQLite3 REQUIRED )
+ 
+ set_property(
+   SOURCE rtc.cpp
+@@ -170,7 +146,7 @@ prepend_path( ".." rocfft_headers_public
+ add_library( rocfft ${rocfft_source} ${relative_rocfft_headers_public} )
+ add_library( roc::rocfft ALIAS rocfft )
+ 
+-target_link_libraries( rocfft PRIVATE stockham_gen sqlite3 )
++target_link_libraries( rocfft PRIVATE stockham_gen ${SQLite3_LIBRARIES} )
+ if( NOT WIN32 )
+   target_link_libraries( rocfft PRIVATE -lstdc++fs )
+ endif()
+@@ -203,7 +179,7 @@ endif()
+ target_include_directories( rocfft
+   PRIVATE $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}/library/src/include>
+   $<BUILD_INTERFACE:${PROJECT_BINARY_DIR}/library/src/device>
+-  ${sqlite_local_SOURCE_DIR}
++  ${SQLite3_LIBRARIES}
+   PUBLIC  $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}/library/include>
+   $<BUILD_INTERFACE:${PROJECT_BINARY_DIR}/include>
+   $<INSTALL_INTERFACE:include/rocFFT>
diff --git a/sci-libs/rocFFT/metadata.xml b/sci-libs/rocFFT/metadata.xml
index 3274277df8c4..825327b204eb 100644
--- a/sci-libs/rocFFT/metadata.xml
+++ b/sci-libs/rocFFT/metadata.xml
@@ -16,4 +16,8 @@
   <upstream>
     <remote-id type="github">ROCmSoftwarePlatform/rocFFT</remote-id>
   </upstream>
+    <use>
+        <flag name="benchmark">Build and install [dyna-]rocfft-rider.</flag>
+        <flag name="perfscripts">Install rocfft-perf performance visualization suite.</flag>
+    </use>
 </pkgmetadata>
diff --git a/sci-libs/rocFFT/rocFFT-5.0.2.ebuild b/sci-libs/rocFFT/rocFFT-5.0.2.ebuild
new file mode 100644
index 000000000000..8d0cb72d3ed8
--- /dev/null
+++ b/sci-libs/rocFFT/rocFFT-5.0.2.ebuild
@@ -0,0 +1,148 @@
+# Copyright 1999-2022 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{8..10} )
+
+inherit cmake check-reqs multiprocessing python-r1
+
+DESCRIPTION="Next generation FFT implementation for ROCm"
+HOMEPAGE="https://github.com/ROCmSoftwarePlatform/rocFFT"
+SRC_URI="https://github.com/ROCmSoftwarePlatform/rocFFT/archive/rocm-${PV}.tar.gz -> rocFFT-${PV}.tar.gz"
+
+LICENSE="MIT"
+KEYWORDS="~amd64"
+SLOT="0/$(ver_cut 1-2)"
+
+# RDEPEND: perfscripts? dev-python/plotly[${PYTHON_USEDEP}] # currently masked by arch/amd64/x32/package.mask
+RDEPEND="
+perfscripts? (
+	>=media-gfx/asymptote-2.61
+	dev-texlive/texlive-latex
+	dev-tex/latexmk
+	sys-apps/texinfo
+	dev-python/sympy[${PYTHON_USEDEP}]
+	dev-python/numpy[${PYTHON_USEDEP}]
+	dev-python/scipy[${PYTHON_USEDEP}]
+	dev-python/pandas[${PYTHON_USEDEP}] )
+${PYTHON_DEPS}"
+
+DEPEND="dev-util/hip:${SLOT}
+	${PYTHON_DEPS}"
+
+BDEPEND="
+	test? ( dev-cpp/gtest dev-libs/boost
+	>=sci-libs/fftw-3
+	>=dev-util/cmake-3.22
+	dev-util/rocm-cmake:${SLOT}
+)"
+
+CHECKREQS_DISK_BUILD="7G"
+
+IUSE="benchmark perfscripts test"
+REQUIRED_USE="perfscripts? ( benchmark ) ${PYTHON_REQUIRED_USE}"
+
+RESTRICT="!test? ( test )"
+
+S="${WORKDIR}/rocFFT-rocm-${PV}"
+
+PATCHES=(
+	"${FILESDIR}/${PN}-4.2.0-add-functional-header.patch"
+	"${FILESDIR}/${PN}-5.0.2-unbundle-sqlite.patch"
+	"${FILESDIR}/${PN}-5.0.2-add-math-header.patch" )
+
+required_mem() {
+	if use test; then
+		echo "52G"
+	else
+		if [[ -n "${AMDGPU_TARGETS}" ]]; then
+			local NARCH=$(($(awk -F";" '{print NF-1}' <<< "${AMDGPU_TARGETS}" || die)+1)) # count how many archs user specified in ${AMDGPU_TARGETS}
+		else
+			local NARCH=7 # The default number of AMDGPU_TARGETS for rocFFT-4.3.0. May change in the future.
+		fi
+		echo "$(($(makeopts_jobs)*${NARCH}*25+2200))M" # A linear function estimating how much memory required
+	fi
+}
+
+pkg_pretend() {
+	return # leave the disk space check to pkg_setup phase
+}
+
+pkg_setup() {
+	export CHECKREQS_MEMORY=$(required_mem)
+	check-reqs_pkg_setup
+	python_setup
+}
+
+src_prepare() {
+	sed -e "s/PREFIX rocfft//" \
+		-e "/rocm_install_symlink_subdir/d" \
+		-e "/<INSTALL_INTERFACE/s,include,include/rocFFT," \
+		-i library/src/CMakeLists.txt || die
+
+	sed -e "/rocm_install_symlink_subdir/d" \
+		-e "$!N;s:PREFIX\n[ ]*rocfft:# PREFIX rocfft\n:;P;D" \
+		-i library/src/device/CMakeLists.txt || die
+
+	if use perfscripts; then
+		pushd scripts/perf || die
+		sed -e "/\/opt\/rocm/d" -e "/rocmversion/s,rocm_info.strip(),\"${PV}\"," -i perflib/specs.py || dir
+		sed -e "/^top/,+1d" -i rocfft-perf suites.py || die
+		sed -e "s,perflib,${PN}_perflib,g" -i rocfft-perf suites.py perflib/*.py || die
+		sed -e "/^top = /s,__file__).*$,\"${EPREFIX}/usr/share/${PN}-perflib\")," -i perflib/pdf.py perflib/generators.py || die
+		popd
+	fi
+
+	cmake_src_prepare
+}
+
+src_configure() {
+	# Grant access to the device
+	addwrite /dev/kfd
+	addpredict /dev/dri/
+
+	# Compiler to use
+	export CXX=hipcc
+
+	local mycmakeargs=(
+		-Wno-dev
+		-DCMAKE_INSTALL_INCLUDEDIR="include/rocFFT/"
+		-DCMAKE_SKIP_RPATH=ON
+		-DBUILD_CLIENTS_TESTS=$(usex test ON OFF)
+		-DBUILD_CLIENTS_SELFTEST=$(usex test ON OFF)
+		-DBUILD_CLIENTS_RIDER=$(usex benchmark ON OFF)
+		${AMDGPU_TARGETS+-DAMDGPU_TARGETS="${AMDGPU_TARGETS}"}
+	)
+
+	cmake_src_configure
+}
+
+src_test() {
+	addwrite /dev/kfd
+	addpredict /dev/dri
+	cd "${BUILD_DIR}/clients/staging" || die
+	einfo "Running rocfft-test"
+	LD_LIBRARY_PATH=${BUILD_DIR}/library/src/:${BUILD_DIR}/library/src/device ./rocfft-test || die
+
+	einfo "Running rocfft-selftest"
+	LD_LIBRARY_PATH=${BUILD_DIR}/library/src/:${BUILD_DIR}/library/src/device ./rocfft-selftest || die
+}
+
+src_install() {
+	cmake_src_install
+
+	if use benchmark; then
+		cd "${BUILD_DIR}"/clients/staging || die
+		dobin *rider
+	fi
+
+	if use perfscripts; then
+		cd "${S}"/scripts/perf || die
+		python_foreach_impl python_doexe rocfft-perf
+		python_moduleinto ${PN}_perflib
+		python_foreach_impl python_domodule perflib/*.py
+		insinto /usr/share/${PN}-perflib
+		doins *.asy suites.py
+	fi
+}