%global upstreamname rocWMMA
%global rocm_release 6.0
%global rocm_patch 2
%global rocm_version %{rocm_release}.%{rocm_patch}

%global toolchain rocm
# hipcc does not support some clang flags
%global build_cxxflags %(echo %{optflags} | sed -e 's/-fstack-protector-strong/-Xarch_host -fstack-protector-strong/' -e 's/-fcf-protection/-Xarch_host -fcf-protection/')
# there is no debug package
%global debug_package %{nil}

# It is necessary to use this with a local build
# export QA_RPATHS=0xff
%bcond_with test

# When we are not using the system clang
%bcond_without compat_build
%global llvm_maj_ver 17


# may run out of memory for both compile and link
# Use fine tuned cmake ROCWMMA_PARALLEL_COMPILE|LINK_JOBS switches
%global _smp_mflags %{nil}

Name:           rocwmma
Version:        %{rocm_version}
Release:        %autorelease
Summary:        ROCm Matrix Multiple and Accumulate library
Url:            https://github.com/ROCm/%{upstreamname}
License:        MIT

Source0:        %{url}/archive/rocm-%{rocm_version}.tar.gz#/%{upstreamname}-%{rocm_version}.tar.gz
Patch0:         0001-add-link-and-compile-pools-for-rocwmma.patch

BuildRequires:  cmake
%if %{with compat_build}
BuildRequires:  libomp%{llvm_maj_ver}-devel
%else
BuildRequires:  libomp-devel
%endif
BuildRequires:  ninja-build
BuildRequires:  rocm-cmake
BuildRequires:  rocm-comgr-devel
BuildRequires:  rocm-hip-devel
BuildRequires:  rocm-rpm-macros
BuildRequires:  rocm-runtime-devel

%if %{with test}
BuildRequires:  rocblas-devel
BuildRequires:  gtest-devel
%endif

# Only x86_64 works right now:
ExclusiveArch:  x86_64

%description
rocWMMA is a C++ library for accelerating mixed-precision matrix
multiply-accumulate (MMA) operations leveraging AMD GPU hardware.
rocWMMA makes it easier to break down MMA problems into fragments
and distribute block-wise MMA operations in parallel across GPU
wavefronts. Our API consists of a header library, that you can
use to compile MMA acceleration directly into GPU kernel device
code. This can benefit from compiler optimization in the
generation of kernel assembly, and doesn't incur additional
overhead costs of linking to external runtime libraries or having
to launch separate kernels.

%package devel
Summary:        Libraries and headers for %{name}
Provides:       %{name}-static = %{version}-%{release}

%description devel
%{summary}

%if %{with test}
%package test
Summary:        Tests for %{name}

%description test
%{summary}
%endif

%prep
%autosetup -p1 -n %{upstreamname}-rocm-%{version}

%build

# Real cores, No hyperthreading
COMPILE_JOBS=`cat /proc/cpuinfo | grep -m 1 'cpu cores' | awk '{ print $4 }'`
if [ ${COMPILE_JOBS}x = x ]; then
    COMPILE_JOBS=1
fi
# Take into account memmory usage per core, do not thrash real memory
BUILD_MEM=2
MEM_KB=0
MEM_KB=`cat /proc/meminfo | grep MemTotal | awk '{ print $2 }'`
MEM_MB=`eval "expr ${MEM_KB} / 1024"`
MEM_GB=`eval "expr ${MEM_MB} / 1024"`
COMPILE_JOBS_MEM=`eval "expr 1 + ${MEM_GB} / ${BUILD_MEM}"`
if [ "$COMPILE_JOBS_MEM" -lt "$COMPILE_JOBS" ]; then
    COMPILE_JOBS=$COMPILE_JOBS_MEM
fi
LINK_MEM=16
LINK_JOBS=`eval "expr 1 + ${MEM_GB} / ${LINK_MEM}"`

%cmake -G Ninja \
       -DBUILD_FILE_REORG_BACKWARD_COMPATIBILITY=OFF \
       -DROCM_SYMLINK_LIBS=OFF \
       -DHIP_PLATFORM=amd \
       -DROCWMMA_PARALLEL_COMPILE_JOBS=$COMPILE_JOBS \
       -DROCWMMA_PARALLEL_LINK_JOBS=$LINK_JOBS \
       -DROCWMMA_BUILD_SAMPLES=OFF \
       -DROCWMMA_VALIDATE_WITH_ROCBLAS=OFF \
       -DROCWMMA_BENCHMARK_WITH_ROCBLAS=OFF \
%if %{with test}
       -DROCWMMA_BUILD_TESTS=ON
%else
       -DROCWMMA_BUILD_TESTS=OFF
%endif

%cmake_build

%install
%cmake_install


%files devel
%dir %{_includedir}/%{name}/
%license LICENSE.md
%exclude %{_docdir}/%{name}/LICENSE.md
%{_includedir}/%{name}/*

%if %{with test}
%files test
%{_bindir}/*
%endif

%changelog
%autochangelog