From a1e5154a93ba7b8557a27b47223a7bbc552210d9 Mon Sep 17 00:00:00 2001
From: misaka00251 <liuxin@iscas.ac.cn>
Date: Fri, 21 Feb 2025 00:47:22 +0800
Subject: [PATCH] Enable tensile
---
rocblas.spec | 95 +++++++++++++++++++++++++++-------------------------
1 file changed, 49 insertions(+), 46 deletions(-)
diff --git a/rocblas.spec b/rocblas.spec
index 67a8d65..b3e6bbc 100644
--- a/rocblas.spec
+++ b/rocblas.spec
@@ -7,9 +7,6 @@
# 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/')
-# $gpu will be evaluated in the loops below
-%global _vpath_builddir %{_vendor}-%{_target_os}-build-${gpu}
-
%bcond_with debug
%if %{with debug}
%global build_type DEBUG
@@ -33,7 +30,7 @@
%endif
# Tensile blocks general enablement of rocBLAS.
-%bcond_with tensile
+%bcond_without tensile
%if %{with tensile}
%global build_tensile ON
%else
@@ -42,10 +39,17 @@
# Why Build ID is 8 bytes
%global buildid 1
+# or just
+#global debug_package %{nil}
+
+# Compression type and level for source/binary package payloads.
+# "w7T0.xzdio" xz level 7 using %%{getncpus} threads
+%define _source_payload w7T0.xzdio
+%define _binary_payload w7T0.xzdio
Name: rocblas
Version: %{rocm_version}
-Release: 1
+Release: 2
Summary: BLAS implementation for ROCm
Url: https://github.com/ROCmSoftwarePlatform/%{upstreamname}
License: MIT AND BSD-3-Clause
@@ -62,7 +66,6 @@ BuildRequires: rocm-compilersupport-macros
BuildRequires: rocm-hip-devel
BuildRequires: rocm-runtime-devel
BuildRequires: rocm-rpm-macros
-BuildRequires: rocm-rpm-macros-modules
%if %{with tensile}
BuildRequires: msgpack-devel
@@ -82,7 +85,7 @@ BuildRequires: rocminfo
BuildRequires: rocm-smi-devel
%endif
-Requires: rocm-rpm-macros-modules
+Provides: rocblas = %{version}-%{release}
ExclusiveArch: x86_64 riscv64
@@ -111,6 +114,10 @@ Requires: %{name}%{?_isa} = %{version}-%{release}
%autosetup -p1 -n %{upstreamname}-rocm-%{version}
sed -i -e 's@set( BLAS_LIBRARY "blas" )@set( BLAS_LIBRARY "cblas" )@' clients/CMakeLists.txt
sed -i -e 's@target_link_libraries( rocblas-test PRIVATE ${BLAS_LIBRARY} ${GTEST_BOTH_LIBRARIES} roc::rocblas )@target_link_libraries( rocblas-test PRIVATE cblas ${GTEST_BOTH_LIBRARIES} roc::rocblas )@' clients/gtest/CMakeLists.txt
+
+# no git in this build
+sed -i -e 's@find_package(Git REQUIRED)@find_package(Git)@' library/CMakeLists.txt
+
%build
# With compat llvm the system clang is wrong
@@ -120,50 +127,43 @@ export TENSILE_ROCM_OFFLOAD_BUNDLER_PATH=${CLANG_PATH}/clang-offload-bundler
# Work around problem with koji's ld
export HIPCC_LINK_FLAGS_APPEND=-fuse-ld=lld
-for gpu in %{rocm_gpu_list}
-do
- module load rocm/$gpu
- # We can't compile all the targets at once
- export ROCM_GPUS='gfx1030;gfx1100;gfx1101;gfx1102'
-
- %cmake \
- -DCMAKE_CXX_COMPILER=hipcc \
- -DCMAKE_C_COMPILER=hipcc \
- -DCMAKE_LINKER=%rocmllvm_bindir/ld.lld \
- -DCMAKE_AR=%rocmllvm_bindir/llvm-ar \
- -DCMAKE_RANLIB=%rocmllvm_bindir/llvm-ranlib \
- -DCMAKE_BUILD_TYPE=%{build_type} \
- -DCMAKE_PREFIX_PATH=%{rocmllvm_cmakedir}/.. \
- -DCMAKE_SKIP_RPATH=ON \
- -DBUILD_FILE_REORG_BACKWARD_COMPATIBILITY=OFF \
- -DROCM_SYMLINK_LIBS=OFF \
- -DHIP_PLATFORM=amd \
- -DAMDGPU_TARGETS=${ROCM_GPUS} \
- -DCMAKE_INSTALL_LIBDIR=$ROCM_LIB \
- -DCMAKE_INSTALL_BINDIR=$ROCM_BIN \
- -DBUILD_CLIENTS_BENCHMARKS=%{build_test} \
- -DBUILD_CLIENTS_TESTS=%{build_test} \
- -DBUILD_WITH_TENSILE=%{build_tensile} \
- -DBUILD_CLIENTS_TESTS_OPENMP=OFF \
- -DBUILD_FORTRAN_CLIENTS=OFF \
- -DBLAS_LIBRARY=cblas \
- -DBUILD_OFFLOAD_COMPRESS=%{build_compress} \
- -DBUILD_WITH_HIPBLASLT=OFF \
- -DTensile_COMPILER=hipcc \
+# We can't compile all the targets at once
+export ROCM_GPUS='gfx1030;gfx1100;gfx1101;gfx1102'
+
+%cmake \
+ -DCMAKE_CXX_COMPILER=hipcc \
+ -DCMAKE_C_COMPILER=hipcc \
+ -DCMAKE_LINKER=%rocmllvm_bindir/ld.lld \
+ -DCMAKE_AR=%rocmllvm_bindir/llvm-ar \
+ -DCMAKE_RANLIB=%rocmllvm_bindir/llvm-ranlib \
+ -DCMAKE_BUILD_TYPE=%{build_type} \
+ -DCMAKE_PREFIX_PATH=%{rocmllvm_cmakedir}/.. \
+ -DCMAKE_SKIP_RPATH=ON \
+ -DBUILD_FILE_REORG_BACKWARD_COMPATIBILITY=OFF \
+ -DROCM_SYMLINK_LIBS=OFF \
+ -DHIP_PLATFORM=amd \
+ -DAMDGPU_TARGETS=${ROCM_GPUS} \
+ -DCMAKE_INSTALL_LIBDIR=%_libdir \
+ -DBUILD_CLIENTS_BENCHMARKS=%{build_test} \
+ -DBUILD_CLIENTS_TESTS=%{build_test} \
+ -DBUILD_CLIENTS_TESTS_OPENMP=OFF \
+ -DBUILD_FORTRAN_CLIENTS=OFF \
+ -DBLAS_LIBRARY=cblas \
+ -DBUILD_OFFLOAD_COMPRESS=%{build_compress} \
+ -DBUILD_WITH_HIPBLASLT=OFF \
+ -DTensile_COMPILER=hipcc \
+ -DBUILD_WITH_TENSILE=%{build_tensile} \
+ -DTensile_DIR=${TP}/cmake \
%if %{buildid}
- -DCMAKE_SHARED_LINKER_FLAGS="-Wl,--build-id=sha1" \
+ -DCMAKE_SHARED_LINKER_FLAGS="-Wl,--build-id=sha1" \
%endif
- -DBUILD_WITH_PIP=OFF
+ -DBUILD_WITH_PIP=OFF
+
+%cmake_build
- %cmake_build
- module purge
-done
%install
-for gpu in %{rocm_gpu_list}
-do
- %cmake_install
-done
+%cmake_install
echo s@%{buildroot}@@ > br.sed
find %{buildroot}%{_libdir} -name '*.so.*.[0-9]' | sed -f br.sed > %{name}.files
@@ -196,6 +196,9 @@ fi
%endif
%changelog
+* Fri Feb 21 2025 misaka00251 <liuxin@iscas.ac.cn> - 6.3.0-2
+- Enable tensile
+
* Wed Dec 18 2024 misaka00251 <liuxin@iscas.ac.cn> - 6.3.0-1
- Update to 6.3.0
--
GitLab