sci-libs/rocBLAS: add 5.4.2

Closes: https://github.com/gentoo/gentoo/pull/29319
Signed-off-by: Yiyang Wu <xgreenlandforwyy@gmail.com>
Signed-off-by: Benda Xu <heroxbd@gentoo.org>

6 files changed, 185 insertions, 0 deletions

diff --git a/sci-libs/rocBLAS/Manifest b/sci-libs/rocBLAS/Manifest
index 8e328c26a78e..3ba6938bbd3c 100644
--- a/sci-libs/rocBLAS/Manifest
+++ b/sci-libs/rocBLAS/Manifest
@@ -1,3 +1,5 @@
 DIST rocBLAS-5.0.2-Tensile-asm_full-navi22.tar.gz 1110323 BLAKE2B 80fab97c6f89fccf1f8829ad41baf15ad31bbbb313b48825841834c7ec1b9f8ece794aebdf9fda42b54a38f642beb1bc1e374f2f3b4e43d8ba9dfd1eb6e3ece5 SHA512 8f6946cf9d68afd195640ae7e0da8dc33341c67f28813813b5361a56531fa2af4453d075703ea52c66d93a1936cf74688b86aa2fa62b52a7ed5f03272290c232
+DIST rocBLAS-5.4.2-Tensile-asm_full-navi22.tar.gz 1057686 BLAKE2B aa755595bbcd48775713f1d94c18b9bfa6464cae97653dd5beda9e4ed600802b62c4159e6edef7fbd4c92e629bbda1c284e6c48e096c7eaf22bbaad86bdcdace SHA512 cdc928528e7ce5bcc3ae32ec18e3aee48533f24e77c1e42f6e0398858eefebd52b8812caabe01a0ef7fa552e00118b862e025f1675a9642b009f1e38efbcdd53
 DIST rocm-rocBLAS-5.0.2.tar.gz 13295178 BLAKE2B e1abfa06b9ac1545732465857839dae3f3708af49439a306a2b5b1b1b392101c38c95815f2c05a0d9ed29d41b154bc2e16a4f9203a37f5e2f68ee12cc4b44a06 SHA512 2f4a74f49350ede0debcf32229a6b6b95505615723c3f2bee16e4c169a3bb966dfb1971819028f47c525f571bea5662b892e63f655b0d1d2b8f20746bc5fe770
 DIST rocm-rocBLAS-5.1.3.tar.gz 13741546 BLAKE2B af10d8cb69bee44ef5aab2dc350e1d9f3b6fddb3a840c17230c4b55b2649a4724c9f2e58c1eafdb3e2716eea016a72a35c3ddb498881b1ce682d780baa8d8b07 SHA512 5ab71838fdd0e9c5848cbf28a19d113353b619a878d8c7d05f64feb32faae2054169c95ed6e9dd6b05a2e807b57229dd2c361c4d289b6e6f17c196558640890f
+DIST rocm-rocBLAS-5.4.2.tar.gz 15938434 BLAKE2B 14ebafa944fdac443800bb7f9b16f8ecf0f420b168d3c6534f68ad7d14bf058a4cc1673fce8f4b9be53e4a6c1cf05011e01853cd901bce0b59827d2aca4d029f SHA512 e62bb80457c1e89454885499bdce9d60beecd706806724418983c78c65c2ae303550f9670b5a6e71dae6a61c0e42b223ab01b36b8406430731ebcbff54c4a8f4
diff --git a/sci-libs/rocBLAS/files/rocBLAS-5.4.2-add-missing-header.patch b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-add-missing-header.patch
new file mode 100644
index 000000000000..cefad2bad61a
--- /dev/null
+++ b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-add-missing-header.patch
@@ -0,0 +1,14 @@
+Mitigate compilation issue.
+
+std::bad_alloc is defined in <new>
+--- a/clients/include/host_alloc.hpp
++++ b/clients/include/host_alloc.hpp
+@@ -22,6 +22,8 @@
+ 
+ #pragma once
+ 
++#include <new>
++
+ //!
+ //! @brief Host free memory w/o swap.  Returns kB or -1 if unknown.
+ //!
diff --git a/sci-libs/rocBLAS/files/rocBLAS-5.4.2-cpp_lib_filesystem.patch b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-cpp_lib_filesystem.patch
new file mode 100644
index 000000000000..688fe7eb5bc3
--- /dev/null
+++ b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-cpp_lib_filesystem.patch
@@ -0,0 +1,24 @@
+Index: rocBLAS-rocm-5.4.2/clients/common/utility.cpp
+===================================================================
+--- rocBLAS-rocm-5.4.2.orig/clients/common/utility.cpp
++++ rocBLAS-rocm-5.4.2/clients/common/utility.cpp
+@@ -38,6 +38,7 @@
+ #ifdef WIN32
+ #define strcasecmp(A, B) _stricmp(A, B)
+ 
++#include <version>
+ #ifdef __cpp_lib_filesystem
+ #include <filesystem>
+ namespace fs = std::filesystem;
+Index: rocBLAS-rocm-5.4.2/clients/include/singletons.hpp
+===================================================================
+--- rocBLAS-rocm-5.4.2.orig/clients/include/singletons.hpp
++++ rocBLAS-rocm-5.4.2/clients/include/singletons.hpp
+@@ -21,6 +21,6 @@
+  * ************************************************************************ */
+ 
+ // global for device memory padding see d_vector.hpp
+-
++#include<cstddef>
+ extern size_t g_DVEC_PAD;
+ void          d_vector_set_pad_length(size_t pad);
diff --git a/sci-libs/rocBLAS/files/rocBLAS-5.4.2-link-cblas.patch b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-link-cblas.patch
new file mode 100644
index 000000000000..a0a599ad0de8
--- /dev/null
+++ b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-link-cblas.patch
@@ -0,0 +1,12 @@
+Fix CPU blas function not found linking issue
+--- a/clients/CMakeLists.txt
++++ b/clients/CMakeLists.txt
+@@ -119,7 +119,7 @@ if( BUILD_CLIENTS_BENCHMARKS OR BUILD_CLIENTS_TESTS)
+       set( BLIS_CPP ../common/blis_interface.cpp )
+       set( BLAS_LIBRARY ${BUILD_DIR}/deps/blis/lib/libblis.a )
+     else()
+-      set( BLAS_LIBRARY "blas" )
++      set( BLAS_LIBRARY "blas -lcblas" )
+     endif()
+   else() # WIN32
+     set( BLAS_INCLUDE_DIR ${OPENBLAS_DIR}/include CACHE PATH "OpenBLAS library include path" )
diff --git a/sci-libs/rocBLAS/files/rocBLAS-5.4.2-unbundle-Tensile.patch b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-unbundle-Tensile.patch
new file mode 100644
index 000000000000..114c10502e68
--- /dev/null
+++ b/sci-libs/rocBLAS/files/rocBLAS-5.4.2-unbundle-Tensile.patch
@@ -0,0 +1,26 @@
+Index: rocBLAS-rocm-5.4.2/CMakeLists.txt
+===================================================================
+--- rocBLAS-rocm-5.4.2.orig/CMakeLists.txt
++++ rocBLAS-rocm-5.4.2/CMakeLists.txt
+@@ -138,21 +138,6 @@ if(NOT SKIP_LIBRARY)
+ 
+     set( Tensile_TEST_LOCAL_PATH "" CACHE PATH "Use local Tensile directory instead of fetching a GitHub branch" )
+ 
+-    include(virtualenv)
+-
+-    if (Tensile_TEST_LOCAL_PATH)
+-      virtualenv_install(${Tensile_TEST_LOCAL_PATH})
+-      message (STATUS "using local Tensile from ${Tensile_TEST_LOCAL_PATH}, copied to ${Tensile_ROOT}")
+-    else()
+-      # Use the virtual-env setup and download package from specified repot:
+-      set( tensile_fork "ROCmSoftwarePlatform" CACHE STRING "Tensile fork to use" )
+-      file (STRINGS "tensile_tag.txt" read_tensile_tag)
+-      set( tensile_tag ${read_tensile_tag} CACHE STRING "Tensile tag to download" )
+-      virtualenv_install("git+https://github.com/${tensile_fork}/Tensile.git@${tensile_tag}")
+-      message (STATUS "using GIT Tensile fork=${tensile_fork} from branch=${tensile_tag}")
+-    endif()
+-    message(STATUS "Adding ${VIRTUALENV_HOME_DIR} to CMAKE_PREFIX_PATH")
+-    list(APPEND CMAKE_PREFIX_PATH ${VIRTUALENV_HOME_DIR})
+     if (TENSILE_VERSION)
+       find_package(Tensile ${TENSILE_VERSION} EXACT REQUIRED HIP LLVM OpenMP PATHS "${INSTALLED_TENSILE_PATH}")
+     else()
diff --git a/sci-libs/rocBLAS/rocBLAS-5.4.2.ebuild b/sci-libs/rocBLAS/rocBLAS-5.4.2.ebuild
new file mode 100644
index 000000000000..28fad2ffbe68
--- /dev/null
+++ b/sci-libs/rocBLAS/rocBLAS-5.4.2.ebuild
@@ -0,0 +1,107 @@
+# Copyright 1999-2023 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+DOCS_BUILDER="doxygen"
+DOCS_DIR="docs"
+DOCS_DEPEND="media-gfx/graphviz"
+ROCM_VERSION=${PV}
+inherit cmake docs edo multiprocessing rocm
+
+DESCRIPTION="AMD's library for BLAS on ROCm"
+HOMEPAGE="https://github.com/ROCmSoftwarePlatform/rocBLAS"
+SRC_URI="https://github.com/ROCmSoftwarePlatform/rocBLAS/archive/rocm-${PV}.tar.gz -> rocm-${P}.tar.gz
+	https://media.githubusercontent.com/media/littlewu2508/littlewu2508.github.io/main/gentoo-distfiles/${PN}-5.4.2-Tensile-asm_full-navi22.tar.gz"
+S="${WORKDIR}/${PN}-rocm-${PV}"
+
+LICENSE="BSD"
+KEYWORDS="~amd64"
+SLOT="0/$(ver_cut 1-2)"
+IUSE="benchmark test"
+REQUIRED_USE="${ROCM_REQUIRED_USE}"
+RESTRICT="!test? ( test )"
+
+BDEPEND="
+	>=dev-util/rocm-cmake-5.3
+	dev-util/Tensile:${SLOT}
+"
+
+DEPEND="
+	dev-util/hip
+	dev-libs/msgpack
+	test? (
+		virtual/blas
+		dev-cpp/gtest
+		sys-libs/libomp
+	)
+	benchmark? (
+		virtual/blas
+		sys-libs/libomp
+	)
+"
+
+QA_FLAGS_IGNORED="/usr/lib64/rocblas/library/.*"
+
+PATCHES=(
+	"${FILESDIR}"/${PN}-5.4.2-cpp_lib_filesystem.patch
+	"${FILESDIR}"/${PN}-5.4.2-unbundle-Tensile.patch
+	"${FILESDIR}"/${PN}-5.4.2-add-missing-header.patch
+	"${FILESDIR}"/${PN}-5.4.2-link-cblas.patch
+	)
+
+src_prepare() {
+	cmake_src_prepare
+	cp -a "${WORKDIR}/asm_full/" library/src/blas3/Tensile/Logic/ || die
+	sed -e "s:,-rpath=.*\":\":" -i clients/CMakeLists.txt || die
+}
+
+src_configure() {
+	addpredict /dev/random
+	addpredict /dev/kfd
+	addpredict /dev/dri/
+
+	local mycmakeargs=(
+		-DCMAKE_SKIP_RPATH=On
+		-DBUILD_FILE_REORG_BACKWARD_COMPATIBILITY=OFF
+		-DROCM_SYMLINK_LIBS=OFF
+		-DAMDGPU_TARGETS="$(get_amdgpu_flags)"
+		-DTensile_LOGIC="asm_full"
+		-DTensile_COMPILER="hipcc"
+		-DTensile_LIBRARY_FORMAT="msgpack"
+		-DTensile_CODE_OBJECT_VERSION="V3"
+		-DTensile_TEST_LOCAL_PATH="${EPREFIX}/usr/share/Tensile"
+		-DTensile_ROOT="${EPREFIX}/usr/share/Tensile"
+		-DBUILD_WITH_TENSILE=ON
+		-DCMAKE_INSTALL_INCLUDEDIR="include/rocblas"
+		-DBUILD_CLIENTS_SAMPLES=OFF
+		-DBUILD_CLIENTS_TESTS=$(usex test ON OFF)
+		-DBUILD_CLIENTS_BENCHMARKS=$(usex benchmark ON OFF)
+		-DTensile_CPU_THREADS=$(makeopts_jobs)
+	)
+
+	CXX=hipcc cmake_src_configure
+}
+
+src_compile() {
+	docs_compile
+	cmake_src_compile
+}
+
+src_test() {
+	check_amdgpu
+	cd "${BUILD_DIR}"/clients/staging || die
+	export ROCBLAS_TEST_TIMEOUT=3600 ROCBLAS_TENSILE_LIBPATH="${BUILD_DIR}/Tensile/library"
+	export LD_LIBRARY_PATH="${BUILD_DIR}/clients:${BUILD_DIR}/library/src"
+	edob ./${PN,,}-test
+}
+
+src_install() {
+	cmake_src_install
+
+	if use benchmark; then
+		cd "${BUILD_DIR}" || die
+		dolib.a clients/librocblas_fortran_client.a
+		dobin clients/staging/rocblas-bench
+	fi
+}