Add recipe for pytorch (C++ interface only)

fxdiv.spec

@@ -0,0 +1,14 @@
+### RPM external fxdiv 2020-04-17
+
+%define commit b408327ac2a15ec3e43352421954f5b1967701d1
+
+Source0: git+https://github.com/Maratyszcza/FXdiv.git?obj=master/%{commit}&export=%{n}-%{realversion}&output=/%{n}-%{realversion}.tgz
+
+%prep
+%setup -n %{n}-%{realversion}
+
+%build
+
+%install
+mkdir -p %{i}/include
+cp -a include/fxdiv.h     %{i}/include/

pip/requirements.txt

@@ -312,6 +312,7 @@ toml==0.10.2
 tomli==2.0.1
 tomli-w==1.0.0
 tomlkit==0.11.4
+torch==1.13.1
 tornado==6.2
 tqdm==4.64.1
 traitlets==5.3.0

pip/torch.file

@@ -0,0 +1,72 @@
+## INCLUDE cuda-flags
+
+%define cuda_arch_float $(echo %{cuda_arch} | tr ' ' '\\n' | sed -E 's|([0-9])$|.\\1|' | tr '\\n' ' ')
+%define tag v%{realversion}
+%define branch release/1.13
+
+%define source0 git+https://github.com/pytorch/pytorch.git?obj=%{branch}/%{tag}&export=%{n}-%{realversion}&submodules=1&output=/%{n}-%{realversion}.tgz
+#Patch0: py3-torch-cpp-externsion-ppc64 - TODO: is it needed? 1st part doesn't apply
+
+BuildRequires: cmake gmake
+Requires: eigen fxdiv numactl openmpi protobuf psimd py3-astunparse py3-cffi py3-future py3-numpy py3-pip py3-protobuf py3-pybind11 py3-PyYAML py3-requests py3-setuptools py3-six py3-tqdm py3-typing-extensions py3-wheel python3 cuda cudnn OpenBLAS
+
+%define PipPreBuildPy \
+%if "%{?compiling_processes:set}" == "set" \
+export MAX_JOBS=%compiling_processes \
+%endif \
+export COLORIZE_OUTPUT=OFF \
+export BUILD_TEST=OFF \
+export USE_CUDA=ON \
+export TORCH_CUDA_ARCH_LIST="%{cuda_arch_float}" \
+ \
+export CUDNN_INCLUDE_DIR=${CUDNN_ROOT}/include \
+export CUDNN_LIBRARY=${CUDNN_ROOT}/lib64/libcudnn.so \
+ \
+export USE_NCCL=OFF \
+export USE_FBGEMM=OFF \
+export USE_KINETO=OFF \
+export USE_MAGMA=OFF \
+export USE_METAL=OFF \
+export USE_MPS=OFF \
+export USE_BREAKPAD=OFF \
+export USE_NNPACK=OFF \
+export USE_QNNPACK=OFF \
+export USE_PYTORCH_QNNPACK=OFF \
+export USE_XNNPACK=OFF \
+ \
+export USE_NUMA=ON \
+export NUMA_ROOT_DIR=${NUMACTL_ROOT} \
+ \
+export USE_NUMPY=ON \
+export USE_OPENMP=ON \
+export USE_QNNPACK=OFF \
+export USE_VALGRIND=OFF \
+export USE_XNNPACK=OFF \
+export USE_MKLDNN=OFF \
+export USE_DISTRIBUTED=OFF \
+export USE_MPI=ON \
+export USE_GLOO=OFF \
+export USE_TENSORPIPE=OFF \
+export ONNX_ML=ON \
+ \
+export PYTORCH_BUILD_VERSION=%{realversion} \
+export PYTORCH_BUILD_NUMBER=0 \
+export BLAS=OpenBLAS \
+export WITH_BLAS=open \
+ \
+export BUILD_CUSTOM_PROTOBUF=OFF \
+export USE_SYSTEM_EIGEN_INSTALL=ON \
+export pybind11_DIR=${PY3_PYBIND11_ROOT}/lib/python%{cms_python3_major_minor_version}/site-packages/pybind11/share/cmake/pybind11 \
+export pybind11_INCLUDE_DIR=${PY3_PYBIND11_ROOT}/lib/python%{cms_python3_major_minor_version}/site-packages/pybind11/include \
+export USE_SYSTEM_PYBIND11=ON \
+ \
+export USE_SYSTEM_PSIMD=ON \
+export USE_SYSTEM_FXDIV=ON \
+export USE_SYSTEM_BENCHMARK=ON \
+export CMAKE_PREFIX_PATH="%{cmake_prefix_path}"
+
+%define PipPrePost %{relocateConfig}lib/python%{cms_python3_major_minor_version}/site-packages/torch/share/cmake/ATen/ATenConfig.cmake
+
+# For ROCm, pre-build
+# NOTICE: can't build with both cuda and rocm
+# python @{_builddir}/tools/amd_build/build_amd.py

psimd.spec

@@ -0,0 +1,22 @@
+### RPM external psimd 2020-05-17
+%define commit 072586a71b55b7f8c584153d223e95687148a900
+
+Source0: git+https://github.com/Maratyszcza/psimd.git?obj=master/%{commit}&export=%{n}-%{realversion}&output=/%{n}-%{realversion}.tgz
+
+BuildRequires: gmake cmake
+
+%prep
+%setup -n %{n}-%{realversion}
+
+%build
+cd %{_builddir}
+rm -rf build
+mkdir build && cd build
+
+cmake ../%{n}-%{realversion} \
+    -DCMAKE_INSTALL_PREFIX=%{i} \
+    -DCMAKE_BUILD_TYPE=Release
+
+%install
+cd %{_builddir}/build
+make install

py3-torch-cpp-externsion-ppc64.patch

@@ -0,0 +1,75 @@
+diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
+index aed4fd5004..9ebaf18652 100644
+--- a/torch/utils/cpp_extension.py
++++ b/torch/utils/cpp_extension.py
+@@ -53,7 +53,7 @@ CUDA_GCC_VERSIONS = {
+     '11.1': (MINIMUM_GCC_VERSION, (10, 0, 0)),
+     '11.2': (MINIMUM_GCC_VERSION, (10, 0, 0)),
+     '11.3': (MINIMUM_GCC_VERSION, (10, 0, 0)),
+-    '11.4': ((6, 0, 0), (10, 0, 0))
++    '11.4': ((6, 0, 0), (11, 2, 0))
+ }
+
+ CUDA_CLANG_VERSIONS = {
+@@ -465,12 +465,17 @@ class BuildExtension(build_ext, object):
+         else:
+             original_compile = self.compiler._compile
+
+-        def append_std14_if_no_std_present(cflags) -> None:
++        def append_std14_if_no_std_present(cflags, with_cuda) -> None:
+             # NVCC does not allow multiple -std to be passed, so we avoid
+             # overriding the option if the user explicitly passed it.
++            if with_cuda:
++                std_val = 'c++14'
++            else:
++                std_val = 'gnu++14'
++
+             cpp_format_prefix = '/{}:' if self.compiler.compiler_type == 'msvc' else '-{}='
+             cpp_flag_prefix = cpp_format_prefix.format('std')
+-            cpp_flag = cpp_flag_prefix + 'c++14'
++            cpp_flag = cpp_flag_prefix + std_val
+             if not any(flag.startswith(cpp_flag_prefix) for flag in cflags):
+                 cflags.append(cpp_flag)
+
+@@ -511,11 +516,13 @@ class BuildExtension(build_ext, object):
+                         cflags = COMMON_HIPCC_FLAGS + cflags + _get_rocm_arch_flags(cflags)
+                     else:
+                         cflags = unix_cuda_flags(cflags)
++                    append_std14_if_no_std_present(cflags, True)
+                 elif isinstance(cflags, dict):
+                     cflags = cflags['cxx']
++                    append_std14_if_no_std_present(cflags, False)
+                 if IS_HIP_EXTENSION:
+                     cflags = COMMON_HIP_FLAGS + cflags
+-                append_std14_if_no_std_present(cflags)
++                    append_std14_if_no_std_present(cflags, False)
+
+                 original_compile(obj, src, ext, cc_args, cflags, pp_opts)
+             finally:
+@@ -564,7 +571,7 @@ class BuildExtension(build_ext, object):
+                 post_cflags = list(extra_postargs)
+             if IS_HIP_EXTENSION:
+                 post_cflags = COMMON_HIP_FLAGS + post_cflags
+-            append_std14_if_no_std_present(post_cflags)
++            append_std14_if_no_std_present(post_cflags, with_cuda)
+
+             cuda_post_cflags = None
+             cuda_cflags = None
+@@ -579,7 +586,7 @@ class BuildExtension(build_ext, object):
+                     cuda_post_cflags = COMMON_HIP_FLAGS + COMMON_HIPCC_FLAGS + cuda_post_cflags
+                 else:
+                     cuda_post_cflags = unix_cuda_flags(cuda_post_cflags)
+-                append_std14_if_no_std_present(cuda_post_cflags)
++                append_std14_if_no_std_present(cuda_post_cflags, with_cuda)
+                 cuda_cflags = [shlex.quote(f) for f in cuda_cflags]
+                 cuda_post_cflags = [shlex.quote(f) for f in cuda_post_cflags]
+
+@@ -710,7 +717,7 @@ class BuildExtension(build_ext, object):
+                 post_cflags = extra_postargs['cxx']
+             else:
+                 post_cflags = list(extra_postargs)
+-            append_std14_if_no_std_present(post_cflags)
++            append_std14_if_no_std_present(post_cflags, with_cuda)
+
+             cuda_post_cflags = None
+             cuda_cflags = None

python_tools.spec

@@ -16,6 +16,7 @@ Requires: py3-tensorflow
 Requires: py3-cmsml
 Requires: py3-law
 Requires: py3-protobuf
+Requires: py3-torch
 
 Requires: py3-tables
 Requires: py3-numexpr