%global pkgvers 1
%global scdate0 20240328
%global sctags0 v2.8.0
%global source0 https://bitbucket.org/icl/magma.git

%global sshort0 %{expand:%%{lua:print(('%{schash0}'):sub(1,8))}}

%global vcu_maj 12
%global vcu_min 4

Name:           magma
Version:        2.8.0
Release:        %{scdate0}.%{pkgvers}.cu%{vcu_maj}_%{vcu_min}%{?dist}
Summary:        Collection of next generation linear algebra (LA) GPU accelerated libraries
License:        BSD

URL:            http://icl.cs.utk.edu/magma/software

Patch0:         magma-sover.patch

BuildRequires:  cmake doxygen git python3
BuildRequires:  openblas-devel openblas-openmp eigen3-devel

%global have_cuda_gcc 1
%ifarch aarch64
%global gpu_target_arch "sm_50 sm_52 sm_53 sm_60 sm_61 sm_62 sm_70 sm_75 sm_80 sm_86 sm_90"
%else
%global gpu_target_arch "sm_50 sm_52 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90"
%endif

%if %{have_cuda_gcc}
%if (0%{?fedora} > 34) || (0%{?rhel} > 8)
BuildRequires:  cuda-gcc-c++
%endif
%endif
BuildRequires:  cuda-nvcc-%{vcu_maj}-%{vcu_min}
BuildRequires:  cuda-cudart-devel-%{vcu_maj}-%{vcu_min}
BuildRequires:  cuda-nvml-devel-%{vcu_maj}-%{vcu_min}
BuildRequires:  cuda-profiler-api-%{vcu_maj}-%{vcu_min}
BuildRequires:  libcublas-devel-%{vcu_maj}-%{vcu_min}
BuildRequires:  libcusparse-devel-%{vcu_maj}-%{vcu_min}
BuildRequires:  libnccl-devel libnccl

Requires: libcublas-%{vcu_maj}-%{vcu_min}
Requires: libcusparse-%{vcu_maj}-%{vcu_min}

%define __cmake_in_source_build 1

%description
collection of next generation linear algebra (LA) GPU accelerated libraries 

%package        devel
Summary:        Development files for the library
Requires:       %{name} = %{version}-%{release}

%description    devel
This package contains the development library and header files.


%prep
%setup -T -c -n %{name}
git clone --depth 1 -n -b %{sctags0} %{source0} .
git reset --hard %{sctags0}
git log --format=fuller
%patch0 -p0 -b .sover~


%build
# disable testing
sed -i '/testing/d' Makefile.internal
sed -i '/include_directories( testing )/d'  CMakeLists.txt
# python3
for f in `find . -name '*.py'`;
do
  sed -i -e 's|#!/usr/bin/env python|#!/usr/bin/python3|' $f
done

# set all cuda arch
echo "FORT=true" > make.inc
echo "GPU_TARGET=%{gpu_target_arch}" >> make.inc
# generate
make generate

mkdir -p build
pushd build
export LD_LIBRARY_PATH="/usr/local/cuda-%{vcu_maj}.%{vcu_min}/%{_lib}/"
%cmake .. \
       -DCMAKE_SKIP_RPATH=ON \
       -DCMAKE_BUILD_TYPE=RelWithDebInfo \
       -DCMAKE_VERBOSE_MAKEFILE=OFF \
       -DCMAKE_CXX_STANDARD=14 \
       -DCMAKE_CXX_FLAGS="-I/usr/local/cuda-%{vcu_maj}.%{vcu_min}/include" \
       -DBUILD_SHARED_LIBS=ON \
       -DUSE_FORTRAN=OFF \
       -DMAGMA_ENABLE_CUDA=ON \
       -DCUDA_GENERATION="Auto" \
       -DCUDA_PROPAGATE_HOST_FLAGS=OFF \
       -DCUDA_CFLAGS="-fPIC -Wno-deprecated-gpu-targets -I/usr/local/cuda-%{vcu_maj}.%{vcu_min}/include --fatbin-options --compress-all" \
       -DGPU_TARGET=%{gpu_target_arch} \
       -DCMAKE_CUDA_PROPAGATE_HOST_FLAGS=OFF \
       -DCMAKE_CUDA_COMPILER=/usr/local/cuda-%{vcu_maj}.%{vcu_min}/bin/nvcc \
       -DCMAKE_CUDA_FLAGS="-fPIC -Wno-deprecated-gpu-targets -I/usr/local/cuda-%{vcu_maj}.%{vcu_min}/include --fatbin-options --compress-all" \
%if %{have_cuda_gcc}
%if (0%{?fedora} > 34) || (0%{?rhel} > 8)
       -DCMAKE_CUDA_HOST_COMPILER="%{_bindir}/cuda-gcc" \
%endif
%endif
       -DLAPACK_LIBRARIES="%{_libdir}/libopenblaso.so"

make %{?_smp_mflags}
popd

pushd docs
doxygen
popd


%install
pushd build
make install DESTDIR=%{buildroot} INSTALL="install -p" CPPROG="cp -p"
popd
mv -f %{buildroot}/usr/lib %{buildroot}/%{_libdir}


%ldconfig_scriptlets


%files
%doc README
%license COPYRIGHT
%{_libdir}/lib*.so.*


%files devel
%doc docs/html
%{_includedir}/*
%{_libdir}/lib*.so
%{_libdir}/pkgconfig/%{name}.pc


%changelog
* Sun Dec 23 2018 Balint Cristian <cristian.balint@gmail.com>
- git update releases