Date: Monday, April 10, 2023 @ 16:11:07
Author: felixonmars
Revision: 1444461
archrelease: copy trunk to community-staging-x86_64
Added:
python-pytorch/repos/community-staging-x86_64/87773.patch
(from rev 1444460, python-pytorch/trunk/87773.patch)
python-pytorch/repos/community-staging-x86_64/PKGBUILD
(from rev 1444460, python-pytorch/trunk/PKGBUILD)
python-pytorch/repos/community-staging-x86_64/disable-werror1.patch
(from rev 1444460, python-pytorch/trunk/disable-werror1.patch)
python-pytorch/repos/community-staging-x86_64/disable-werror2.patch
(from rev 1444460, python-pytorch/trunk/disable-werror2.patch)
python-pytorch/repos/community-staging-x86_64/disable-werror3.patch
(from rev 1444460, python-pytorch/trunk/disable-werror3.patch)
python-pytorch/repos/community-staging-x86_64/disable-werror4.patch
(from rev 1444460, python-pytorch/trunk/disable-werror4.patch)
python-pytorch/repos/community-staging-x86_64/ffmpeg4.4.patch
(from rev 1444460, python-pytorch/trunk/ffmpeg4.4.patch)
python-pytorch/repos/community-staging-x86_64/fix-building-for-torchvision.patch
(from rev 1444460, python-pytorch/trunk/fix-building-for-torchvision.patch)
python-pytorch/repos/community-staging-x86_64/fix_include_system.patch
(from rev 1444460, python-pytorch/trunk/fix_include_system.patch)
python-pytorch/repos/community-staging-x86_64/rocblas-batched.patch
(from rev 1444460, python-pytorch/trunk/rocblas-batched.patch)
python-pytorch/repos/community-staging-x86_64/test.py
(from rev 1444460, python-pytorch/trunk/test.py)
python-pytorch/repos/community-staging-x86_64/use-system-libuv.patch
(from rev 1444460, python-pytorch/trunk/use-system-libuv.patch)
Deleted:
python-pytorch/repos/community-staging-x86_64/87773.patch
python-pytorch/repos/community-staging-x86_64/PKGBUILD
python-pytorch/repos/community-staging-x86_64/disable-werror1.patch
python-pytorch/repos/community-staging-x86_64/disable-werror2.patch
python-pytorch/repos/community-staging-x86_64/disable-werror3.patch
python-pytorch/repos/community-staging-x86_64/disable-werror4.patch
python-pytorch/repos/community-staging-x86_64/ffmpeg4.4.patch
python-pytorch/repos/community-staging-x86_64/fix-building-for-torchvision.patch
python-pytorch/repos/community-staging-x86_64/fix_include_system.patch
python-pytorch/repos/community-staging-x86_64/rocblas-batched.patch
python-pytorch/repos/community-staging-x86_64/test.py
python-pytorch/repos/community-staging-x86_64/use-system-libuv.patch
------------------------------------+
87773.patch | 78 +--
PKGBUILD | 876 +++++++++++++++++------------------
disable-werror1.patch | 26 -
disable-werror2.patch | 26 -
disable-werror3.patch | 26 -
disable-werror4.patch | 30 -
ffmpeg4.4.patch | 102 ++--
fix-building-for-torchvision.patch | 50 -
fix_include_system.patch | 26 -
rocblas-batched.patch | 72 +-
test.py | 14
use-system-libuv.patch | 26 -
12 files changed, 676 insertions(+), 676 deletions(-)
Deleted: 87773.patch
===================================================================
--- 87773.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ 87773.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,39 +0,0 @@
-From 94465d6287e8f54c99f4b0b891a6c618bb80d7ce Mon Sep 17 00:00:00 2001
-From: Arfrever Frehtes Taifersar Arahesis <[email protected]>
-Date: Wed, 26 Oct 2022 07:32:59 +0200
-Subject: [PATCH] Support only GLog >=0.6.0
-
-Fixes https://github.com/pytorch/pytorch/issues/58054
----
- c10/util/Logging.cpp | 12 +-----------
- 1 file changed, 1 insertion(+), 11 deletions(-)
-
-diff --git a/c10/util/Logging.cpp b/c10/util/Logging.cpp
-index fe74e49548646..d8fce12ce4840 100644
---- a/c10/util/Logging.cpp
-+++ b/c10/util/Logging.cpp
-@@ -192,23 +192,13 @@ C10_DEFINE_int(
- google::GLOG_WARNING,
- "The minimum log level that caffe2 will output.");
-
--// Google glog's api does not have an external function that allows one to
check
--// if glog is initialized or not. It does have an internal function - so we
are
--// declaring it here. This is a hack but has been used by a bunch of others
too
--// (e.g. Torch).
--namespace google {
--namespace glog_internal_namespace_ {
--bool IsGoogleLoggingInitialized();
--} // namespace glog_internal_namespace_
--} // namespace google
--
- namespace c10 {
- namespace {
-
- void initGoogleLogging(char const* name) {
- #if !defined(_MSC_VER)
- // This trick can only be used on UNIX platforms
-- if (!::google::glog_internal_namespace_::IsGoogleLoggingInitialized())
-+ if (!::google::IsGoogleLoggingInitialized())
- #endif
- {
- ::google::InitGoogleLogging(name);
Copied: python-pytorch/repos/community-staging-x86_64/87773.patch (from rev
1444460, python-pytorch/trunk/87773.patch)
===================================================================
--- 87773.patch (rev 0)
+++ 87773.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,39 @@
+From 94465d6287e8f54c99f4b0b891a6c618bb80d7ce Mon Sep 17 00:00:00 2001
+From: Arfrever Frehtes Taifersar Arahesis <[email protected]>
+Date: Wed, 26 Oct 2022 07:32:59 +0200
+Subject: [PATCH] Support only GLog >=0.6.0
+
+Fixes https://github.com/pytorch/pytorch/issues/58054
+---
+ c10/util/Logging.cpp | 12 +-----------
+ 1 file changed, 1 insertion(+), 11 deletions(-)
+
+diff --git a/c10/util/Logging.cpp b/c10/util/Logging.cpp
+index fe74e49548646..d8fce12ce4840 100644
+--- a/c10/util/Logging.cpp
++++ b/c10/util/Logging.cpp
+@@ -192,23 +192,13 @@ C10_DEFINE_int(
+ google::GLOG_WARNING,
+ "The minimum log level that caffe2 will output.");
+
+-// Google glog's api does not have an external function that allows one to
check
+-// if glog is initialized or not. It does have an internal function - so we
are
+-// declaring it here. This is a hack but has been used by a bunch of others
too
+-// (e.g. Torch).
+-namespace google {
+-namespace glog_internal_namespace_ {
+-bool IsGoogleLoggingInitialized();
+-} // namespace glog_internal_namespace_
+-} // namespace google
+-
+ namespace c10 {
+ namespace {
+
+ void initGoogleLogging(char const* name) {
+ #if !defined(_MSC_VER)
+ // This trick can only be used on UNIX platforms
+- if (!::google::glog_internal_namespace_::IsGoogleLoggingInitialized())
++ if (!::google::IsGoogleLoggingInitialized())
+ #endif
+ {
+ ::google::InitGoogleLogging(name);
Deleted: PKGBUILD
===================================================================
--- PKGBUILD 2023-04-10 16:10:27 UTC (rev 1444460)
+++ PKGBUILD 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,438 +0,0 @@
-# Maintainer: Sven-Hendrik Haase <[email protected]>
-# Maintainer: Torsten Keßler <[email protected]>
-# Contributor: Stephen Zhang <zsrkmyn at gmail dot com>
-
-_pkgname=pytorch
-pkgbase="python-${_pkgname}"
-pkgname=("${pkgbase}" "${pkgbase}-opt" "${pkgbase}-cuda" "${pkgbase}-opt-cuda"
"${pkgbase}-rocm" "${pkgbase}-opt-rocm")
-pkgver=2.0.0
-_pkgver=2.0.0
-pkgrel=3
-_pkgdesc='Tensors and Dynamic neural networks in Python with strong GPU
acceleration'
-pkgdesc="${_pkgdesc}"
-arch=('x86_64')
-url="https://pytorch.org";
-license=('BSD')
-depends=('google-glog' 'gflags' 'opencv' 'openmp' 'nccl' 'pybind11' 'python'
'python-yaml' 'libuv'
- 'python-numpy' 'python-sympy' 'protobuf' 'ffmpeg4.4' 'python-future'
'qt5-base'
- 'intel-oneapi-mkl' 'python-typing_extensions')
-# Exclude the magma package here and add the corresponding {cuda, rocm/hip}
version
-# to makedepends of the split packages.
-# The magma package does not allow to build the cuda and rocm/hip code at the
same time,
-# so we need to work with the split packages magma-{cuda,hip}.
-makedepends=('python' 'python-setuptools' 'python-yaml' 'python-numpy' 'cmake'
'cuda'
- 'cudnn' 'git' 'rocm-hip-sdk' 'roctracer' 'miopen'
- 'ninja' 'pkgconfig' 'doxygen' 'vulkan-headers' 'shaderc')
-source=("${_pkgname}-${pkgver}::git+https://github.com/pytorch/pytorch.git#tag=v$_pkgver";
- # generated using parse-submodules
-
"${pkgname}-ARM_NEON_2_x86_SSE::git+https://github.com/intel/ARM_NEON_2_x86_SSE.git";
- "${pkgname}-FP16::git+https://github.com/Maratyszcza/FP16.git";
- "${pkgname}-FXdiv::git+https://github.com/Maratyszcza/FXdiv.git";
- "${pkgname}-NNPACK::git+https://github.com/Maratyszcza/NNPACK.git";
- "${pkgname}-PeachPy::git+https://github.com/malfet/PeachPy.git";
- "${pkgname}-QNNPACK::git+https://github.com/pytorch/QNNPACK";
-
"${pkgname}-VulkanMemoryAllocator::git+https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator.git";
- "${pkgname}-XNNPACK::git+https://github.com/google/XNNPACK.git";
- "${pkgname}-benchmark::git+https://github.com/google/benchmark.git";
- "${pkgname}-cpuinfo::git+https://github.com/pytorch/cpuinfo.git";
- "${pkgname}-cub::git+https://github.com/NVlabs/cub.git";
-
"${pkgname}-cudnn-frontend::git+https://github.com/NVIDIA/cudnn-frontend.git";
- "${pkgname}-cutlass::git+https://github.com/NVIDIA/cutlass.git";
- "${pkgname}-eigen::git+https://gitlab.com/libeigen/eigen.git";
- "${pkgname}-enum34::git+https://github.com/PeachPy/enum34.git";
- "${pkgname}-fbgemm::git+https://github.com/pytorch/fbgemm";
- "${pkgname}-fbjni::git+https://github.com/facebookincubator/fbjni.git";
- "${pkgname}-flatbuffers::git+https://github.com/google/flatbuffers.git";
- "${pkgname}-fmt::git+https://github.com/fmtlib/fmt.git";
- "${pkgname}-foxi::git+https://github.com/houseroad/foxi.git";
- "${pkgname}-gemmlowp::git+https://github.com/google/gemmlowp.git";
- "${pkgname}-gloo::git+https://github.com/facebookincubator/gloo";
- "${pkgname}-googletest::git+https://github.com/google/googletest.git";
- "${pkgname}-ideep::git+https://github.com/intel/ideep";
- "${pkgname}-ios-cmake::git+https://github.com/Yangqing/ios-cmake.git";
- "${pkgname}-ittapi::git+https://github.com/intel/ittapi.git";
- "${pkgname}-json::git+https://github.com/nlohmann/json.git";
- "${pkgname}-kineto::git+https://github.com/pytorch/kineto";
- "${pkgname}-nccl::git+https://github.com/NVIDIA/nccl";
- "${pkgname}-onnx-tensorrt::git+https://github.com/onnx/onnx-tensorrt";
- "${pkgname}-onnx::git+https://github.com/onnx/onnx.git";
- "${pkgname}-pocketfft::git+https://github.com/mreineck/pocketfft";
-
"${pkgname}-protobuf::git+https://github.com/protocolbuffers/protobuf.git";
- "${pkgname}-psimd::git+https://github.com/Maratyszcza/psimd.git";
-
"${pkgname}-pthreadpool::git+https://github.com/Maratyszcza/pthreadpool.git";
- "${pkgname}-pybind11::git+https://github.com/pybind/pybind11.git";
- "${pkgname}-six::git+https://github.com/benjaminp/six.git";
- "${pkgname}-sleef::git+https://github.com/shibatch/sleef";
- "${pkgname}-tbb::git+https://github.com/01org/tbb";
- "${pkgname}-tensorpipe::git+https://github.com/pytorch/tensorpipe.git";
- "${pkgname}-zstd::git+https://github.com/facebook/zstd.git";
- fix_include_system.patch
- use-system-libuv.patch
- fix-building-for-torchvision.patch
- 87773.patch
- disable-werror1.patch
- disable-werror2.patch
- disable-werror3.patch
- disable-werror4.patch
- ffmpeg4.4.patch
- rocblas-batched.patch)
-b2sums=('SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
- 'SKIP'
-
'77f85808e480bd37dfb5f072d565466ae30a8f827f49ef97591fc2fc03bea54944eb1adeaa4a1e3466518a5640f575eda88d15b4c4d549a6f41f0bf4f2cfb086'
-
'1f7ce593fa9fc62535ca1c3d85c996a73006cc614c7b7258160c3fc53cd52a1cfddcb18baf897f2e1223ecdfee52ca1471b91c9f845368ed6ac51b66f6e0e676'
-
'fdea0b815d7750a4233c1d4668593020da017aea43cf4cb63b4c00d0852c7d34f0333e618fcf98b8df2185313a2089b8c2e9fe8ec3cfb0bf693598f9c61461a8'
-
'0a8fc110a306e81beeb9ddfb3a1ddfd26aeda5e3f7adfb0f7c9bc3fd999c2dde62e0b407d3eca573097a53fd97329214e30e8767fb38d770197c7ec2b53daf18'
-
'844d0b7b39777492a6d456fa845d5399f673b4bb37b62473393449c9ad0c29dca3c33276dc3980f2e766680100335c0acfb69d51781b79575f4da112d9c4018c'
-
'985e331b2025e1ca5a4fba5188af0900f1f38bd0fd32c9173deb8bed7358af01e387d4654c7e0389e5f98b6f7cbed053226934d180b8b3b1270bdbbb36fc89b2'
-
'96de2729b29c7ce3e4fdd8008f575d24c2c3ef9f85d6217e607902d7b870ac71b9290fde71e87a68d75bb75ef28eacbf5ce04e071146809ccf1e76a03f97b479'
-
'eea86bbed0a37e1661035913536456f90e0cd1e687c7e4103011f0688bc8347b6fc2ff82019909c41e7c89ddbc3b80dde641e88abf406f4faebc71b0bb693d25'
-
'6286b05d5b5143f117363e3ce3c7d693910f53845aeb6f501b3eea64aa71778cb2d7dcd4ac945d5321ef23b4da02446e86dedc6a9b6a998df4a7f3b1ce50550a'
-
'232d2aca7cae8da511d1451890f8696d47da72276929ac5731a1a1a481d2a515fa7288bf33730d8ea2c892616551a74ca2439b53de6b1dfee156c30919120741')
-options=('!lto' '!debug')
-
-get_pyver () {
- python -c 'import sys; print(str(sys.version_info[0]) + "." +
str(sys.version_info[1]))'
-}
-
-prepare() {
- cd "${srcdir}/${_pkgname}-${pkgver}"
-
- # generated using parse-submodules
- git submodule init
-
- git config submodule."android/libs/fbjni".url "${srcdir}/${pkgname}"-fbjni
- git config submodule."third_party/NNPACK".url "${srcdir}/${pkgname}"-NNPACK
- git config submodule."third_party/NNPACK_deps/FP16".url
"${srcdir}/${pkgname}"-FP16
- git config submodule."third_party/NNPACK_deps/FXdiv".url
"${srcdir}/${pkgname}"-FXdiv
- git config submodule."third_party/NNPACK_deps/psimd".url
"${srcdir}/${pkgname}"-psimd
- git config submodule."third_party/NNPACK_deps/pthreadpool".url
"${srcdir}/${pkgname}"-pthreadpool
- git config submodule."third_party/QNNPACK".url "${srcdir}/${pkgname}"-QNNPACK
- git config submodule."third_party/VulkanMemoryAllocator".url
"${srcdir}/${pkgname}"-VulkanMemoryAllocator
- git config submodule."third_party/XNNPACK".url "${srcdir}/${pkgname}"-XNNPACK
- git config submodule."third_party/benchmark".url
"${srcdir}/${pkgname}"-benchmark
- git config submodule."third_party/cpuinfo".url "${srcdir}/${pkgname}"-cpuinfo
- git config submodule."third_party/cub".url "${srcdir}/${pkgname}"-cub
- git config submodule."third_party/cudnn_frontend".url
"${srcdir}/${pkgname}"-cudnn-frontend
- git config submodule."third_party/cutlass".url "${srcdir}/${pkgname}"-cutlass
- git config submodule."third_party/eigen".url "${srcdir}/${pkgname}"-eigen
- git config submodule."third_party/fbgemm".url "${srcdir}/${pkgname}"-fbgemm
- git config submodule."third_party/flatbuffers".url
"${srcdir}/${pkgname}"-flatbuffers
- git config submodule."third_party/fmt".url "${srcdir}/${pkgname}"-fmt
- git config submodule."third_party/foxi".url "${srcdir}/${pkgname}"-foxi
- git config submodule."third_party/gemmlowp/gemmlowp".url
"${srcdir}/${pkgname}"-gemmlowp
- git config submodule."third_party/gloo".url "${srcdir}/${pkgname}"-gloo
- git config submodule."third_party/googletest".url
"${srcdir}/${pkgname}"-googletest
- git config submodule."third_party/ideep".url "${srcdir}/${pkgname}"-ideep
- git config submodule."third_party/ios-cmake".url
"${srcdir}/${pkgname}"-ios-cmake
- git config submodule."third_party/ittapi".url "${srcdir}/${pkgname}"-ittapi
- git config submodule."third_party/kineto".url "${srcdir}/${pkgname}"-kineto
- git config submodule."third_party/nccl/nccl".url "${srcdir}/${pkgname}"-nccl
- git config submodule."third_party/neon2sse".url
"${srcdir}/${pkgname}"-ARM_NEON_2_x86_SSE
- git config submodule."third_party/nlohmann".url "${srcdir}/${pkgname}"-json
- git config submodule."third_party/onnx".url "${srcdir}/${pkgname}"-onnx
- git config submodule."third_party/onnx-tensorrt".url
"${srcdir}/${pkgname}"-onnx-tensorrt
- git config submodule."third_party/pocketfft".url
"${srcdir}/${pkgname}"-pocketfft
- git config submodule."third_party/protobuf".url
"${srcdir}/${pkgname}"-protobuf
- git config submodule."third_party/pybind11".url
"${srcdir}/${pkgname}"-pybind11
- git config submodule."third_party/python-enum".url
"${srcdir}/${pkgname}"-enum34
- git config submodule."third_party/python-peachpy".url
"${srcdir}/${pkgname}"-PeachPy
- git config submodule."third_party/python-six".url "${srcdir}/${pkgname}"-six
- git config submodule."third_party/sleef".url "${srcdir}/${pkgname}"-sleef
- git config submodule."third_party/tbb".url "${srcdir}/${pkgname}"-tbb
- git config submodule."third_party/tensorpipe".url
"${srcdir}/${pkgname}"-tensorpipe
- git config submodule."third_party/zstd".url "${srcdir}/${pkgname}"-zstd
-
- git -c protocol.file.allow=always submodule update --init --recursive
-
- # Fix include with GCC 12
- sed "1i#include <mutex>" -i
third_party/kineto/libkineto/src/RoctracerActivityApi.h
-
- # https://bugs.archlinux.org/task/64981
- patch -N torch/utils/cpp_extension.py "${srcdir}"/fix_include_system.patch
-
- # Use system libuv
- patch -Np1 -i "${srcdir}"/use-system-libuv.patch
-
- # fix https://github.com/pytorch/vision/issues/3695
- patch -Np1 -i "${srcdir}/fix-building-for-torchvision.patch"
-
- # Fix building against glog 0.6
- patch -Np1 -i "${srcdir}/87773.patch"
-
- # Disable -Werror
- patch -Np1 -d third_party/fbgemm -i "${srcdir}/disable-werror1.patch"
- patch -Np1 -d third_party/benchmark -i "${srcdir}/disable-werror2.patch"
- patch -Np1 -d third_party/ideep/mkl-dnn -i "${srcdir}/disable-werror3.patch"
- patch -Np1 -i "${srcdir}/disable-werror4.patch"
-
- # build against ffmpeg4.4
- patch -Np1 -i "${srcdir}/ffmpeg4.4.patch"
-
- # fix https://github.com/pytorch/pytorch/issues/97640
- patch -Np1 -i "${srcdir}/rocblas-batched.patch"
-
- cd "${srcdir}"
-
- cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-opt"
- cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-cuda"
- cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-opt-cuda"
- cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-rocm"
- cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-opt-rocm"
-}
-
-# Common build configuration, called in all package() functions.
-_prepare() {
- export VERBOSE=1
- export PYTORCH_BUILD_VERSION="${pkgver}"
- export PYTORCH_BUILD_NUMBER=1
-
- # Check tools/setup_helpers/cmake.py, setup.py and CMakeLists.txt for a list
of flags that can be set via env vars.
- export ATEN_NO_TEST=ON # do not build ATen tests
- export USE_MKLDNN=ON
- export BUILD_CUSTOM_PROTOBUF=OFF
- export BUILD_CAFFE2=ON
- export BUILD_CAFFE2_OPS=ON
- # export BUILD_SHARED_LIBS=OFF
- export USE_FFMPEG=ON
- export USE_GFLAGS=ON
- export USE_GLOG=ON
- export USE_VULKAN=ON
- export BUILD_BINARY=ON
- export USE_OBSERVERS=ON
- export USE_OPENCV=ON
- # export USE_SYSTEM_LIBS=ON # experimental, not all libs present in repos
- export USE_SYSTEM_NCCL=ON
- export NCCL_VERSION=$(pkg-config nccl --modversion)
- export NCCL_VER_CODE=$(sed -n 's/^#define NCCL_VERSION_CODE\s*\(.*\).*/\1/p'
/usr/include/nccl.h)
- # export BUILD_SPLIT_CUDA=ON # modern preferred build, but splits libs and
symbols, ABI break
- # export USE_FAST_NVCC=ON # parallel build with nvcc, spawns too many
processes
- export USE_CUPTI_SO=ON # make sure cupti.so is used as shared lib
- export CC=/usr/bin/gcc
- export CXX=/usr/bin/g++
- export CUDAHOSTCXX=/opt/cuda/bin/g++
- export CUDA_HOST_COMPILER="${CUDAHOSTCXX}"
- export CUDA_HOME=/opt/cuda
- # hide build-time CUDA devices
- export CUDA_VISIBLE_DEVICES=""
- export CUDNN_LIB_DIR=/usr/lib
- export CUDNN_INCLUDE_DIR=/usr/include
- export TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
- # CUDA arch 8.7 is not supported (needed by Jetson boards, etc.)
- export
TORCH_CUDA_ARCH_LIST="5.2;5.3;6.0;6.1;6.2;7.0;7.2;7.5;8.0;8.6;8.9;9.0;9.0+PTX"
#include latest PTX for future compat
- export OVERRIDE_TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST}"
- export PYTORCH_ROCM_ARCH="gfx803;gfx900;gfx906;gfx908;gfx90a;gfx1030"
-
- # Hack to make sure that the generated dnnl_config.h from onednn can be
inlcuded.
- export CXXFLAGS="${CXXFLAGS} -I
third_party/ideep/mkl-dnn/third_party/oneDNN/include/"
-}
-
-#
-# Important note on the missing build() function
-#
-# We build the pytorch packages for the different backends directly in
-# corresponding package() functions. This change became necessary when
-# merging the two different GPU backends (CUDA and ROCm) into one package.
-# Both share a dependency on the magma package but compiled against
-# different GPU backends. This leads to two incompatible magma-{cuda,hip}
-# packages that cannot be installed side-by-side.
-# Therefore, we need to separately add magma-{cuda,hip} as (make-) dependencies
-# of pytorch-{cuda,rocm}.
-
-_package() {
- # Prevent setup.py from re-running CMake and rebuilding
- sed -e 's/RUN_BUILD_DEPS = True/RUN_BUILD_DEPS = False/g' -i setup.py
-
- python setup.py install --root="${pkgdir}"/ --optimize=1 --skip-build
-
- install -Dm644 LICENSE "${pkgdir}/usr/share/licenses/${pkgname}/LICENSE"
-
- pytorchpath="usr/lib/python$(get_pyver)/site-packages/torch"
- install -d "${pkgdir}/usr/lib"
-
- # put CMake files in correct place
- mv "${pkgdir}/${pytorchpath}/share/cmake" "${pkgdir}/usr/lib/cmake"
-
- # put C++ API in correct place
- mv "${pkgdir}/${pytorchpath}/include" "${pkgdir}/usr/include"
- find "${pkgdir}/${pytorchpath}"/lib/ -type f,l \( -iname '*.so' -or -iname
'*.so*' \) -print0 | while read -rd $'\0' _lib; do
- mv "${_lib}" "${pkgdir}"/usr/lib/
- done
-
- # clean up duplicates
- rm -r "${pkgdir}/usr/include/pybind11"
- rm "${pkgdir}"/usr/include/*.h
-
- # Python module is hardcoded so look there at runtime
- ln -s /usr/include "${pkgdir}/${pytorchpath}/include"
- find "${pkgdir}"/usr/lib -maxdepth 1 -type f,l \( -iname '*.so' -or -iname
'*.so*' \) -print0 | while read -rd $'\0' _lib; do
- ln -s ${_lib#"$pkgdir"} "${pkgdir}/${pytorchpath}/lib/"
- done
-}
-
-package_python-pytorch() {
- pkgdesc="${_pkgdesc}"
-
- cd "${srcdir}/${_pkgname}-${pkgver}"
- echo "Building without cuda or rocm and without non-x86-64 optimizations"
- _prepare
- export USE_CUDA=0
- export USE_CUDNN=0
- export USE_ROCM=0
- echo "add_definitions(-march=x86-64)" >> cmake/MiscCheck.cmake
- # this horrible hack is necessary because the current release
- # ships inconsistent CMake which tries to build objects before
- # thier dependencies, build twice when dependencies are available
- python setup.py build || python setup.py build
-
- _package
-}
-
-package_python-pytorch-opt() {
- pkgdesc="${_pkgdesc} (with AVX2 CPU optimizations)"
- conflicts=(python-pytorch)
- provides=(python-pytorch)
-
- cd "${srcdir}/${_pkgname}-${pkgver}-opt"
- echo "Building without cuda or rocm and with non-x86-64 optimizations"
- _prepare
- export USE_CUDA=0
- export USE_CUDNN=0
- export USE_ROCM=0
- echo "add_definitions(-march=haswell)" >> cmake/MiscCheck.cmake
- # same horrible hack as above
- python setup.py build || python setup.py build
-
- _package
-}
-
-package_python-pytorch-cuda() {
- pkgdesc="${_pkgdesc} (with CUDA)"
- depends+=(cuda cudnn magma-cuda)
- conflicts=(python-pytorch)
- provides=(python-pytorch)
-
- cd "${srcdir}/${_pkgname}-${pkgver}-cuda"
- echo "Building with cuda and without non-x86-64 optimizations"
- _prepare
- export USE_CUDA=1
- export USE_CUDNN=1
- export USE_ROCM=0
- cd "${srcdir}/${_pkgname}-${pkgver}-cuda"
- echo "add_definitions(-march=x86-64)" >> cmake/MiscCheck.cmake
- # same horrible hack as above
- python setup.py build || python setup.py build
-
- _package
- # oneDNN from the repos conflicts with the version in the ideep submodule,
- # so we have to add the dependency after building the package.
- depends+=(onednn)
-}
-
-package_python-pytorch-opt-cuda() {
- pkgdesc="${_pkgdesc} (with CUDA and AVX2 CPU optimizations)"
- depends+=(cuda cudnn magma-cuda)
- conflicts=(python-pytorch)
- provides=(python-pytorch python-pytorch-cuda)
-
- cd "${srcdir}/${_pkgname}-${pkgver}-opt-cuda"
- echo "Building with cuda and with non-x86-64 optimizations"
- export USE_CUDA=1
- export USE_CUDNN=1
- export USE_ROCM=0
- _prepare
- echo "add_definitions(-march=haswell)" >> cmake/MiscCheck.cmake
- # same horrible hack as above
- python setup.py build || python setup.py build
-
- _package
- # see above
- depends+=(onednn)
-}
-
-package_python-pytorch-rocm() {
- pkgdesc="${_pkgdesc} (with ROCm)"
- depends+=(rocm-hip-sdk roctracer miopen magma-hip)
- conflicts=(python-pytorch)
- provides=(python-pytorch)
-
- cd "${srcdir}/${_pkgname}-${pkgver}-rocm"
- echo "Building with rocm and without non-x86-64 optimizations"
- _prepare
- export USE_CUDA=0
- export USE_CUDNN=0
- export USE_ROCM=1
- echo "add_definitions(-march=x86-64)" >> cmake/MiscCheck.cmake
- # Conversion of CUDA to ROCm source files
- python tools/amd_build/build_amd.py
- # same horrible hack as above
- python setup.py build || python setup.py build
-
- _package
- # see above
- depends+=(onednn)
-}
-
-package_python-pytorch-opt-rocm() {
- pkgdesc="${_pkgdesc} (with ROCm and AVX2 CPU optimizations)"
- depends+=(rocm-hip-sdk roctracer miopen magma-hip)
- conflicts=(python-pytorch)
- provides=(python-pytorch python-pytorch-rocm)
-
- cd "${srcdir}/${_pkgname}-${pkgver}-opt-rocm"
- echo "Building with rocm and with non-x86-64 optimizations"
- _prepare
- export USE_CUDA=0
- export USE_CUDNN=0
- export USE_ROCM=1
- echo "add_definitions(-march=haswell)" >> cmake/MiscCheck.cmake
- # Conversion of CUDA to ROCm source files
- python tools/amd_build/build_amd.py
- # same horrible hack as above
- python setup.py build || python setup.py build
-
- _package
- # see above
- depends+=(onednn)
-}
-
-# vim:set ts=2 sw=2 et:
Copied: python-pytorch/repos/community-staging-x86_64/PKGBUILD (from rev
1444460, python-pytorch/trunk/PKGBUILD)
===================================================================
--- PKGBUILD (rev 0)
+++ PKGBUILD 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,438 @@
+# Maintainer: Sven-Hendrik Haase <[email protected]>
+# Maintainer: Torsten Keßler <[email protected]>
+# Contributor: Stephen Zhang <zsrkmyn at gmail dot com>
+
+_pkgname=pytorch
+pkgbase="python-${_pkgname}"
+pkgname=("${pkgbase}" "${pkgbase}-opt" "${pkgbase}-cuda" "${pkgbase}-opt-cuda"
"${pkgbase}-rocm" "${pkgbase}-opt-rocm")
+pkgver=2.0.0
+_pkgver=2.0.0
+pkgrel=4
+_pkgdesc='Tensors and Dynamic neural networks in Python with strong GPU
acceleration'
+pkgdesc="${_pkgdesc}"
+arch=('x86_64')
+url="https://pytorch.org";
+license=('BSD')
+depends=('google-glog' 'gflags' 'opencv' 'openmp' 'nccl' 'pybind11' 'python'
'python-yaml' 'libuv'
+ 'python-numpy' 'python-sympy' 'protobuf' 'ffmpeg4.4' 'python-future'
'qt5-base'
+ 'intel-oneapi-mkl' 'python-typing_extensions')
+# Exclude the magma package here and add the corresponding {cuda, rocm/hip}
version
+# to makedepends of the split packages.
+# The magma package does not allow to build the cuda and rocm/hip code at the
same time,
+# so we need to work with the split packages magma-{cuda,hip}.
+makedepends=('python' 'python-setuptools' 'python-yaml' 'python-numpy' 'cmake'
'cuda'
+ 'cudnn' 'git' 'rocm-hip-sdk' 'roctracer' 'miopen'
+ 'ninja' 'pkgconfig' 'doxygen' 'vulkan-headers' 'shaderc')
+source=("${_pkgname}-${pkgver}::git+https://github.com/pytorch/pytorch.git#tag=v$_pkgver";
+ # generated using parse-submodules
+
"${pkgname}-ARM_NEON_2_x86_SSE::git+https://github.com/intel/ARM_NEON_2_x86_SSE.git";
+ "${pkgname}-FP16::git+https://github.com/Maratyszcza/FP16.git";
+ "${pkgname}-FXdiv::git+https://github.com/Maratyszcza/FXdiv.git";
+ "${pkgname}-NNPACK::git+https://github.com/Maratyszcza/NNPACK.git";
+ "${pkgname}-PeachPy::git+https://github.com/malfet/PeachPy.git";
+ "${pkgname}-QNNPACK::git+https://github.com/pytorch/QNNPACK";
+
"${pkgname}-VulkanMemoryAllocator::git+https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator.git";
+ "${pkgname}-XNNPACK::git+https://github.com/google/XNNPACK.git";
+ "${pkgname}-benchmark::git+https://github.com/google/benchmark.git";
+ "${pkgname}-cpuinfo::git+https://github.com/pytorch/cpuinfo.git";
+ "${pkgname}-cub::git+https://github.com/NVlabs/cub.git";
+
"${pkgname}-cudnn-frontend::git+https://github.com/NVIDIA/cudnn-frontend.git";
+ "${pkgname}-cutlass::git+https://github.com/NVIDIA/cutlass.git";
+ "${pkgname}-eigen::git+https://gitlab.com/libeigen/eigen.git";
+ "${pkgname}-enum34::git+https://github.com/PeachPy/enum34.git";
+ "${pkgname}-fbgemm::git+https://github.com/pytorch/fbgemm";
+ "${pkgname}-fbjni::git+https://github.com/facebookincubator/fbjni.git";
+ "${pkgname}-flatbuffers::git+https://github.com/google/flatbuffers.git";
+ "${pkgname}-fmt::git+https://github.com/fmtlib/fmt.git";
+ "${pkgname}-foxi::git+https://github.com/houseroad/foxi.git";
+ "${pkgname}-gemmlowp::git+https://github.com/google/gemmlowp.git";
+ "${pkgname}-gloo::git+https://github.com/facebookincubator/gloo";
+ "${pkgname}-googletest::git+https://github.com/google/googletest.git";
+ "${pkgname}-ideep::git+https://github.com/intel/ideep";
+ "${pkgname}-ios-cmake::git+https://github.com/Yangqing/ios-cmake.git";
+ "${pkgname}-ittapi::git+https://github.com/intel/ittapi.git";
+ "${pkgname}-json::git+https://github.com/nlohmann/json.git";
+ "${pkgname}-kineto::git+https://github.com/pytorch/kineto";
+ "${pkgname}-nccl::git+https://github.com/NVIDIA/nccl";
+ "${pkgname}-onnx-tensorrt::git+https://github.com/onnx/onnx-tensorrt";
+ "${pkgname}-onnx::git+https://github.com/onnx/onnx.git";
+ "${pkgname}-pocketfft::git+https://github.com/mreineck/pocketfft";
+
"${pkgname}-protobuf::git+https://github.com/protocolbuffers/protobuf.git";
+ "${pkgname}-psimd::git+https://github.com/Maratyszcza/psimd.git";
+
"${pkgname}-pthreadpool::git+https://github.com/Maratyszcza/pthreadpool.git";
+ "${pkgname}-pybind11::git+https://github.com/pybind/pybind11.git";
+ "${pkgname}-six::git+https://github.com/benjaminp/six.git";
+ "${pkgname}-sleef::git+https://github.com/shibatch/sleef";
+ "${pkgname}-tbb::git+https://github.com/01org/tbb";
+ "${pkgname}-tensorpipe::git+https://github.com/pytorch/tensorpipe.git";
+ "${pkgname}-zstd::git+https://github.com/facebook/zstd.git";
+ fix_include_system.patch
+ use-system-libuv.patch
+ fix-building-for-torchvision.patch
+ 87773.patch
+ disable-werror1.patch
+ disable-werror2.patch
+ disable-werror3.patch
+ disable-werror4.patch
+ ffmpeg4.4.patch
+ rocblas-batched.patch)
+b2sums=('SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+ 'SKIP'
+
'77f85808e480bd37dfb5f072d565466ae30a8f827f49ef97591fc2fc03bea54944eb1adeaa4a1e3466518a5640f575eda88d15b4c4d549a6f41f0bf4f2cfb086'
+
'1f7ce593fa9fc62535ca1c3d85c996a73006cc614c7b7258160c3fc53cd52a1cfddcb18baf897f2e1223ecdfee52ca1471b91c9f845368ed6ac51b66f6e0e676'
+
'fdea0b815d7750a4233c1d4668593020da017aea43cf4cb63b4c00d0852c7d34f0333e618fcf98b8df2185313a2089b8c2e9fe8ec3cfb0bf693598f9c61461a8'
+
'0a8fc110a306e81beeb9ddfb3a1ddfd26aeda5e3f7adfb0f7c9bc3fd999c2dde62e0b407d3eca573097a53fd97329214e30e8767fb38d770197c7ec2b53daf18'
+
'844d0b7b39777492a6d456fa845d5399f673b4bb37b62473393449c9ad0c29dca3c33276dc3980f2e766680100335c0acfb69d51781b79575f4da112d9c4018c'
+
'985e331b2025e1ca5a4fba5188af0900f1f38bd0fd32c9173deb8bed7358af01e387d4654c7e0389e5f98b6f7cbed053226934d180b8b3b1270bdbbb36fc89b2'
+
'96de2729b29c7ce3e4fdd8008f575d24c2c3ef9f85d6217e607902d7b870ac71b9290fde71e87a68d75bb75ef28eacbf5ce04e071146809ccf1e76a03f97b479'
+
'eea86bbed0a37e1661035913536456f90e0cd1e687c7e4103011f0688bc8347b6fc2ff82019909c41e7c89ddbc3b80dde641e88abf406f4faebc71b0bb693d25'
+
'6286b05d5b5143f117363e3ce3c7d693910f53845aeb6f501b3eea64aa71778cb2d7dcd4ac945d5321ef23b4da02446e86dedc6a9b6a998df4a7f3b1ce50550a'
+
'232d2aca7cae8da511d1451890f8696d47da72276929ac5731a1a1a481d2a515fa7288bf33730d8ea2c892616551a74ca2439b53de6b1dfee156c30919120741')
+options=('!lto' '!debug')
+
+get_pyver () {
+ python -c 'import sys; print(str(sys.version_info[0]) + "." +
str(sys.version_info[1]))'
+}
+
+prepare() {
+ cd "${srcdir}/${_pkgname}-${pkgver}"
+
+ # generated using parse-submodules
+ git submodule init
+
+ git config submodule."android/libs/fbjni".url "${srcdir}/${pkgname}"-fbjni
+ git config submodule."third_party/NNPACK".url "${srcdir}/${pkgname}"-NNPACK
+ git config submodule."third_party/NNPACK_deps/FP16".url
"${srcdir}/${pkgname}"-FP16
+ git config submodule."third_party/NNPACK_deps/FXdiv".url
"${srcdir}/${pkgname}"-FXdiv
+ git config submodule."third_party/NNPACK_deps/psimd".url
"${srcdir}/${pkgname}"-psimd
+ git config submodule."third_party/NNPACK_deps/pthreadpool".url
"${srcdir}/${pkgname}"-pthreadpool
+ git config submodule."third_party/QNNPACK".url "${srcdir}/${pkgname}"-QNNPACK
+ git config submodule."third_party/VulkanMemoryAllocator".url
"${srcdir}/${pkgname}"-VulkanMemoryAllocator
+ git config submodule."third_party/XNNPACK".url "${srcdir}/${pkgname}"-XNNPACK
+ git config submodule."third_party/benchmark".url
"${srcdir}/${pkgname}"-benchmark
+ git config submodule."third_party/cpuinfo".url "${srcdir}/${pkgname}"-cpuinfo
+ git config submodule."third_party/cub".url "${srcdir}/${pkgname}"-cub
+ git config submodule."third_party/cudnn_frontend".url
"${srcdir}/${pkgname}"-cudnn-frontend
+ git config submodule."third_party/cutlass".url "${srcdir}/${pkgname}"-cutlass
+ git config submodule."third_party/eigen".url "${srcdir}/${pkgname}"-eigen
+ git config submodule."third_party/fbgemm".url "${srcdir}/${pkgname}"-fbgemm
+ git config submodule."third_party/flatbuffers".url
"${srcdir}/${pkgname}"-flatbuffers
+ git config submodule."third_party/fmt".url "${srcdir}/${pkgname}"-fmt
+ git config submodule."third_party/foxi".url "${srcdir}/${pkgname}"-foxi
+ git config submodule."third_party/gemmlowp/gemmlowp".url
"${srcdir}/${pkgname}"-gemmlowp
+ git config submodule."third_party/gloo".url "${srcdir}/${pkgname}"-gloo
+ git config submodule."third_party/googletest".url
"${srcdir}/${pkgname}"-googletest
+ git config submodule."third_party/ideep".url "${srcdir}/${pkgname}"-ideep
+ git config submodule."third_party/ios-cmake".url
"${srcdir}/${pkgname}"-ios-cmake
+ git config submodule."third_party/ittapi".url "${srcdir}/${pkgname}"-ittapi
+ git config submodule."third_party/kineto".url "${srcdir}/${pkgname}"-kineto
+ git config submodule."third_party/nccl/nccl".url "${srcdir}/${pkgname}"-nccl
+ git config submodule."third_party/neon2sse".url
"${srcdir}/${pkgname}"-ARM_NEON_2_x86_SSE
+ git config submodule."third_party/nlohmann".url "${srcdir}/${pkgname}"-json
+ git config submodule."third_party/onnx".url "${srcdir}/${pkgname}"-onnx
+ git config submodule."third_party/onnx-tensorrt".url
"${srcdir}/${pkgname}"-onnx-tensorrt
+ git config submodule."third_party/pocketfft".url
"${srcdir}/${pkgname}"-pocketfft
+ git config submodule."third_party/protobuf".url
"${srcdir}/${pkgname}"-protobuf
+ git config submodule."third_party/pybind11".url
"${srcdir}/${pkgname}"-pybind11
+ git config submodule."third_party/python-enum".url
"${srcdir}/${pkgname}"-enum34
+ git config submodule."third_party/python-peachpy".url
"${srcdir}/${pkgname}"-PeachPy
+ git config submodule."third_party/python-six".url "${srcdir}/${pkgname}"-six
+ git config submodule."third_party/sleef".url "${srcdir}/${pkgname}"-sleef
+ git config submodule."third_party/tbb".url "${srcdir}/${pkgname}"-tbb
+ git config submodule."third_party/tensorpipe".url
"${srcdir}/${pkgname}"-tensorpipe
+ git config submodule."third_party/zstd".url "${srcdir}/${pkgname}"-zstd
+
+ git -c protocol.file.allow=always submodule update --init --recursive
+
+ # Fix include with GCC 12
+ sed "1i#include <mutex>" -i
third_party/kineto/libkineto/src/RoctracerActivityApi.h
+
+ # https://bugs.archlinux.org/task/64981
+ patch -N torch/utils/cpp_extension.py "${srcdir}"/fix_include_system.patch
+
+ # Use system libuv
+ patch -Np1 -i "${srcdir}"/use-system-libuv.patch
+
+ # fix https://github.com/pytorch/vision/issues/3695
+ patch -Np1 -i "${srcdir}/fix-building-for-torchvision.patch"
+
+ # Fix building against glog 0.6
+ patch -Np1 -i "${srcdir}/87773.patch"
+
+ # Disable -Werror
+ patch -Np1 -d third_party/fbgemm -i "${srcdir}/disable-werror1.patch"
+ patch -Np1 -d third_party/benchmark -i "${srcdir}/disable-werror2.patch"
+ patch -Np1 -d third_party/ideep/mkl-dnn -i "${srcdir}/disable-werror3.patch"
+ patch -Np1 -i "${srcdir}/disable-werror4.patch"
+
+ # build against ffmpeg4.4
+ patch -Np1 -i "${srcdir}/ffmpeg4.4.patch"
+
+ # fix https://github.com/pytorch/pytorch/issues/97640
+ patch -Np1 -i "${srcdir}/rocblas-batched.patch"
+
+ cd "${srcdir}"
+
+ cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-opt"
+ cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-cuda"
+ cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-opt-cuda"
+ cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-rocm"
+ cp -r "${_pkgname}-${pkgver}" "${_pkgname}-${pkgver}-opt-rocm"
+}
+
+# Common build configuration, called in all package() functions.
+_prepare() {
+ export VERBOSE=1
+ export PYTORCH_BUILD_VERSION="${pkgver}"
+ export PYTORCH_BUILD_NUMBER=1
+
+ # Check tools/setup_helpers/cmake.py, setup.py and CMakeLists.txt for a list
of flags that can be set via env vars.
+ export ATEN_NO_TEST=ON # do not build ATen tests
+ export USE_MKLDNN=ON
+ export BUILD_CUSTOM_PROTOBUF=OFF
+ export BUILD_CAFFE2=ON
+ export BUILD_CAFFE2_OPS=ON
+ # export BUILD_SHARED_LIBS=OFF
+ export USE_FFMPEG=ON
+ export USE_GFLAGS=ON
+ export USE_GLOG=ON
+ export USE_VULKAN=ON
+ export BUILD_BINARY=ON
+ export USE_OBSERVERS=ON
+ export USE_OPENCV=ON
+ # export USE_SYSTEM_LIBS=ON # experimental, not all libs present in repos
+ export USE_SYSTEM_NCCL=ON
+ export NCCL_VERSION=$(pkg-config nccl --modversion)
+ export NCCL_VER_CODE=$(sed -n 's/^#define NCCL_VERSION_CODE\s*\(.*\).*/\1/p'
/usr/include/nccl.h)
+ # export BUILD_SPLIT_CUDA=ON # modern preferred build, but splits libs and
symbols, ABI break
+ # export USE_FAST_NVCC=ON # parallel build with nvcc, spawns too many
processes
+ export USE_CUPTI_SO=ON # make sure cupti.so is used as shared lib
+ export CC=/usr/bin/gcc
+ export CXX=/usr/bin/g++
+ export CUDAHOSTCXX=/opt/cuda/bin/g++
+ export CUDA_HOST_COMPILER="${CUDAHOSTCXX}"
+ export CUDA_HOME=/opt/cuda
+ # hide build-time CUDA devices
+ export CUDA_VISIBLE_DEVICES=""
+ export CUDNN_LIB_DIR=/usr/lib
+ export CUDNN_INCLUDE_DIR=/usr/include
+ export TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
+ # CUDA arch 8.7 is not supported (needed by Jetson boards, etc.)
+ export
TORCH_CUDA_ARCH_LIST="5.2;5.3;6.0;6.1;6.2;7.0;7.2;7.5;8.0;8.6;8.9;9.0;9.0+PTX"
#include latest PTX for future compat
+ export OVERRIDE_TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST}"
+ export PYTORCH_ROCM_ARCH="gfx803;gfx900;gfx906;gfx908;gfx90a;gfx1030"
+
+ # Hack to make sure that the generated dnnl_config.h from onednn can be
inlcuded.
+ export CXXFLAGS="${CXXFLAGS} -I
third_party/ideep/mkl-dnn/third_party/oneDNN/include/"
+}
+
+#
+# Important note on the missing build() function
+#
+# We build the pytorch packages for the different backends directly in
+# corresponding package() functions. This change became necessary when
+# merging the two different GPU backends (CUDA and ROCm) into one package.
+# Both share a dependency on the magma package but compiled against
+# different GPU backends. This leads to two incompatible magma-{cuda,hip}
+# packages that cannot be installed side-by-side.
+# Therefore, we need to separately add magma-{cuda,hip} as (make-) dependencies
+# of pytorch-{cuda,rocm}.
+
+_package() {
+ # Prevent setup.py from re-running CMake and rebuilding
+ sed -e 's/RUN_BUILD_DEPS = True/RUN_BUILD_DEPS = False/g' -i setup.py
+
+ python setup.py install --root="${pkgdir}"/ --optimize=1 --skip-build
+
+ install -Dm644 LICENSE "${pkgdir}/usr/share/licenses/${pkgname}/LICENSE"
+
+ pytorchpath="usr/lib/python$(get_pyver)/site-packages/torch"
+ install -d "${pkgdir}/usr/lib"
+
+ # put CMake files in correct place
+ mv "${pkgdir}/${pytorchpath}/share/cmake" "${pkgdir}/usr/lib/cmake"
+
+ # put C++ API in correct place
+ mv "${pkgdir}/${pytorchpath}/include" "${pkgdir}/usr/include"
+ find "${pkgdir}/${pytorchpath}"/lib/ -type f,l \( -iname '*.so' -or -iname
'*.so*' \) -print0 | while read -rd $'\0' _lib; do
+ mv "${_lib}" "${pkgdir}"/usr/lib/
+ done
+
+ # clean up duplicates
+ rm -r "${pkgdir}/usr/include/pybind11"
+ rm "${pkgdir}"/usr/include/*.h
+
+ # Python module is hardcoded so look there at runtime
+ ln -s /usr/include "${pkgdir}/${pytorchpath}/include"
+ find "${pkgdir}"/usr/lib -maxdepth 1 -type f,l \( -iname '*.so' -or -iname
'*.so*' \) -print0 | while read -rd $'\0' _lib; do
+ ln -s ${_lib#"$pkgdir"} "${pkgdir}/${pytorchpath}/lib/"
+ done
+}
+
+package_python-pytorch() {
+ pkgdesc="${_pkgdesc}"
+
+ cd "${srcdir}/${_pkgname}-${pkgver}"
+ echo "Building without cuda or rocm and without non-x86-64 optimizations"
+ _prepare
+ export USE_CUDA=0
+ export USE_CUDNN=0
+ export USE_ROCM=0
+ echo "add_definitions(-march=x86-64)" >> cmake/MiscCheck.cmake
+ # this horrible hack is necessary because the current release
+ # ships inconsistent CMake which tries to build objects before
+ # thier dependencies, build twice when dependencies are available
+ python setup.py build || python setup.py build
+
+ _package
+}
+
+package_python-pytorch-opt() {
+ pkgdesc="${_pkgdesc} (with AVX2 CPU optimizations)"
+ conflicts=(python-pytorch)
+ provides=(python-pytorch)
+
+ cd "${srcdir}/${_pkgname}-${pkgver}-opt"
+ echo "Building without cuda or rocm and with non-x86-64 optimizations"
+ _prepare
+ export USE_CUDA=0
+ export USE_CUDNN=0
+ export USE_ROCM=0
+ echo "add_definitions(-march=haswell)" >> cmake/MiscCheck.cmake
+ # same horrible hack as above
+ python setup.py build || python setup.py build
+
+ _package
+}
+
+package_python-pytorch-cuda() {
+ pkgdesc="${_pkgdesc} (with CUDA)"
+ depends+=(cuda cudnn magma-cuda)
+ conflicts=(python-pytorch)
+ provides=(python-pytorch)
+
+ cd "${srcdir}/${_pkgname}-${pkgver}-cuda"
+ echo "Building with cuda and without non-x86-64 optimizations"
+ _prepare
+ export USE_CUDA=1
+ export USE_CUDNN=1
+ export USE_ROCM=0
+ cd "${srcdir}/${_pkgname}-${pkgver}-cuda"
+ echo "add_definitions(-march=x86-64)" >> cmake/MiscCheck.cmake
+ # same horrible hack as above
+ python setup.py build || python setup.py build
+
+ _package
+ # oneDNN from the repos conflicts with the version in the ideep submodule,
+ # so we have to add the dependency after building the package.
+ depends+=(onednn)
+}
+
+package_python-pytorch-opt-cuda() {
+ pkgdesc="${_pkgdesc} (with CUDA and AVX2 CPU optimizations)"
+ depends+=(cuda cudnn magma-cuda)
+ conflicts=(python-pytorch)
+ provides=(python-pytorch python-pytorch-cuda)
+
+ cd "${srcdir}/${_pkgname}-${pkgver}-opt-cuda"
+ echo "Building with cuda and with non-x86-64 optimizations"
+ export USE_CUDA=1
+ export USE_CUDNN=1
+ export USE_ROCM=0
+ _prepare
+ echo "add_definitions(-march=haswell)" >> cmake/MiscCheck.cmake
+ # same horrible hack as above
+ python setup.py build || python setup.py build
+
+ _package
+ # see above
+ depends+=(onednn)
+}
+
+package_python-pytorch-rocm() {
+ pkgdesc="${_pkgdesc} (with ROCm)"
+ depends+=(rocm-hip-sdk roctracer miopen magma-hip)
+ conflicts=(python-pytorch)
+ provides=(python-pytorch)
+
+ cd "${srcdir}/${_pkgname}-${pkgver}-rocm"
+ echo "Building with rocm and without non-x86-64 optimizations"
+ _prepare
+ export USE_CUDA=0
+ export USE_CUDNN=0
+ export USE_ROCM=1
+ echo "add_definitions(-march=x86-64)" >> cmake/MiscCheck.cmake
+ # Conversion of CUDA to ROCm source files
+ python tools/amd_build/build_amd.py
+ # same horrible hack as above
+ python setup.py build || python setup.py build
+
+ _package
+ # see above
+ depends+=(onednn)
+}
+
+package_python-pytorch-opt-rocm() {
+ pkgdesc="${_pkgdesc} (with ROCm and AVX2 CPU optimizations)"
+ depends+=(rocm-hip-sdk roctracer miopen magma-hip)
+ conflicts=(python-pytorch)
+ provides=(python-pytorch python-pytorch-rocm)
+
+ cd "${srcdir}/${_pkgname}-${pkgver}-opt-rocm"
+ echo "Building with rocm and with non-x86-64 optimizations"
+ _prepare
+ export USE_CUDA=0
+ export USE_CUDNN=0
+ export USE_ROCM=1
+ echo "add_definitions(-march=haswell)" >> cmake/MiscCheck.cmake
+ # Conversion of CUDA to ROCm source files
+ python tools/amd_build/build_amd.py
+ # same horrible hack as above
+ python setup.py build || python setup.py build
+
+ _package
+ # see above
+ depends+=(onednn)
+}
+
+# vim:set ts=2 sw=2 et:
Deleted: disable-werror1.patch
===================================================================
--- disable-werror1.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ disable-werror1.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,13 +0,0 @@
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index 58dcb9a..269a7a6 100644
---- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -144,7 +144,7 @@ if(MSVC)
- else(MSVC)
- string(APPEND CMAKE_CXX_FLAGS " -Wall")
- string(APPEND CMAKE_CXX_FLAGS " -Wextra")
-- string(APPEND CMAKE_CXX_FLAGS " -Werror")
-+ # string(APPEND CMAKE_CXX_FLAGS " -Werror")
- string(APPEND CMAKE_CXX_FLAGS " -Wno-deprecated-declarations")
- target_compile_options(fbgemm_avx2 PRIVATE
- "-m64" "-mavx2" "-mf16c" "-mfma")
Copied: python-pytorch/repos/community-staging-x86_64/disable-werror1.patch
(from rev 1444460, python-pytorch/trunk/disable-werror1.patch)
===================================================================
--- disable-werror1.patch (rev 0)
+++ disable-werror1.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,13 @@
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index 58dcb9a..269a7a6 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -144,7 +144,7 @@ if(MSVC)
+ else(MSVC)
+ string(APPEND CMAKE_CXX_FLAGS " -Wall")
+ string(APPEND CMAKE_CXX_FLAGS " -Wextra")
+- string(APPEND CMAKE_CXX_FLAGS " -Werror")
++ # string(APPEND CMAKE_CXX_FLAGS " -Werror")
+ string(APPEND CMAKE_CXX_FLAGS " -Wno-deprecated-declarations")
+ target_compile_options(fbgemm_avx2 PRIVATE
+ "-m64" "-mavx2" "-mf16c" "-mfma")
Deleted: disable-werror2.patch
===================================================================
--- disable-werror2.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ disable-werror2.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,13 +0,0 @@
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index b8852e4..cf1d447 100644
---- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -19,7 +19,7 @@ option(BENCHMARK_ENABLE_TESTING "Enable testing of the
benchmark library." ON)
- option(BENCHMARK_ENABLE_EXCEPTIONS "Enable the use of exceptions in the
benchmark library." ON)
- option(BENCHMARK_ENABLE_LTO "Enable link time optimisation of the benchmark
library." OFF)
- option(BENCHMARK_USE_LIBCXX "Build and test using libc++ as the standard
library." OFF)
--option(BENCHMARK_ENABLE_WERROR "Build Release candidates with -Werror." ON)
-+option(BENCHMARK_ENABLE_WERROR "Build Release candidates with -Werror." OFF)
- option(BENCHMARK_FORCE_WERROR "Build Release candidates with -Werror
regardless of compiler issues." OFF)
-
- if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "PGI")
Copied: python-pytorch/repos/community-staging-x86_64/disable-werror2.patch
(from rev 1444460, python-pytorch/trunk/disable-werror2.patch)
===================================================================
--- disable-werror2.patch (rev 0)
+++ disable-werror2.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,13 @@
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index b8852e4..cf1d447 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -19,7 +19,7 @@ option(BENCHMARK_ENABLE_TESTING "Enable testing of the
benchmark library." ON)
+ option(BENCHMARK_ENABLE_EXCEPTIONS "Enable the use of exceptions in the
benchmark library." ON)
+ option(BENCHMARK_ENABLE_LTO "Enable link time optimisation of the benchmark
library." OFF)
+ option(BENCHMARK_USE_LIBCXX "Build and test using libc++ as the standard
library." OFF)
+-option(BENCHMARK_ENABLE_WERROR "Build Release candidates with -Werror." ON)
++option(BENCHMARK_ENABLE_WERROR "Build Release candidates with -Werror." OFF)
+ option(BENCHMARK_FORCE_WERROR "Build Release candidates with -Werror
regardless of compiler issues." OFF)
+
+ if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "PGI")
Deleted: disable-werror3.patch
===================================================================
--- disable-werror3.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ disable-werror3.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,13 +0,0 @@
-diff --git a/cmake/platform.cmake b/cmake/platform.cmake
-index b588bcc23..81283ab70 100644
---- a/cmake/platform.cmake
-+++ b/cmake/platform.cmake
-@@ -110,7 +110,7 @@ if(MSVC)
- append(CMAKE_CCXX_FLAGS "/fp:precise")
- endif()
- elseif(UNIX OR MINGW)
-- append(CMAKE_CCXX_FLAGS "-Wall -Werror -Wno-unknown-pragmas")
-+ append(CMAKE_CCXX_FLAGS "-Wall -Wno-unknown-pragmas")
- if(DNNL_GRAPH_WITH_SYCL OR CMAKE_BASE_NAME STREQUAL "icx" OR
CMAKE_BASE_NAME STREQUAL "icpx")
- # When using Debug build mode CMake adds "-g" option without "-O0"
- # causing the warning. This probably happens because clang/gcc
compilers
Copied: python-pytorch/repos/community-staging-x86_64/disable-werror3.patch
(from rev 1444460, python-pytorch/trunk/disable-werror3.patch)
===================================================================
--- disable-werror3.patch (rev 0)
+++ disable-werror3.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,13 @@
+diff --git a/cmake/platform.cmake b/cmake/platform.cmake
+index b588bcc23..81283ab70 100644
+--- a/cmake/platform.cmake
++++ b/cmake/platform.cmake
+@@ -110,7 +110,7 @@ if(MSVC)
+ append(CMAKE_CCXX_FLAGS "/fp:precise")
+ endif()
+ elseif(UNIX OR MINGW)
+- append(CMAKE_CCXX_FLAGS "-Wall -Werror -Wno-unknown-pragmas")
++ append(CMAKE_CCXX_FLAGS "-Wall -Wno-unknown-pragmas")
+ if(DNNL_GRAPH_WITH_SYCL OR CMAKE_BASE_NAME STREQUAL "icx" OR
CMAKE_BASE_NAME STREQUAL "icpx")
+ # When using Debug build mode CMake adds "-g" option without "-O0"
+ # causing the warning. This probably happens because clang/gcc
compilers
Deleted: disable-werror4.patch
===================================================================
--- disable-werror4.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ disable-werror4.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,15 +0,0 @@
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index b9addcf005b..27d2c761f4e 100644
---- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -797,10 +797,8 @@ if(NOT MSVC)
- # Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459
- string(APPEND CMAKE_CXX_FLAGS " -Wall")
- string(APPEND CMAKE_CXX_FLAGS " -Wextra")
-- append_cxx_flag_if_supported("-Werror=return-type" CMAKE_CXX_FLAGS)
- append_cxx_flag_if_supported("-Werror=non-virtual-dtor" CMAKE_CXX_FLAGS)
- append_cxx_flag_if_supported("-Werror=braced-scalar-init" CMAKE_CXX_FLAGS)
-- append_cxx_flag_if_supported("-Werror=range-loop-construct" CMAKE_CXX_FLAGS)
- append_cxx_flag_if_supported("-Werror=bool-operation" CMAKE_CXX_FLAGS)
- append_cxx_flag_if_supported("-Wnarrowing" CMAKE_CXX_FLAGS)
- append_cxx_flag_if_supported("-Wno-missing-field-initializers"
CMAKE_CXX_FLAGS)
Copied: python-pytorch/repos/community-staging-x86_64/disable-werror4.patch
(from rev 1444460, python-pytorch/trunk/disable-werror4.patch)
===================================================================
--- disable-werror4.patch (rev 0)
+++ disable-werror4.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,15 @@
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index b9addcf005b..27d2c761f4e 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -797,10 +797,8 @@ if(NOT MSVC)
+ # Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459
+ string(APPEND CMAKE_CXX_FLAGS " -Wall")
+ string(APPEND CMAKE_CXX_FLAGS " -Wextra")
+- append_cxx_flag_if_supported("-Werror=return-type" CMAKE_CXX_FLAGS)
+ append_cxx_flag_if_supported("-Werror=non-virtual-dtor" CMAKE_CXX_FLAGS)
+ append_cxx_flag_if_supported("-Werror=braced-scalar-init" CMAKE_CXX_FLAGS)
+- append_cxx_flag_if_supported("-Werror=range-loop-construct" CMAKE_CXX_FLAGS)
+ append_cxx_flag_if_supported("-Werror=bool-operation" CMAKE_CXX_FLAGS)
+ append_cxx_flag_if_supported("-Wnarrowing" CMAKE_CXX_FLAGS)
+ append_cxx_flag_if_supported("-Wno-missing-field-initializers"
CMAKE_CXX_FLAGS)
Deleted: ffmpeg4.4.patch
===================================================================
--- ffmpeg4.4.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ ffmpeg4.4.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,51 +0,0 @@
-diff --git a/cmake/Modules/FindFFmpeg.cmake b/cmake/Modules/FindFFmpeg.cmake
-index 04437562ee..dc8fe4aa5c 100644
---- a/cmake/Modules/FindFFmpeg.cmake
-+++ b/cmake/Modules/FindFFmpeg.cmake
-@@ -14,34 +14,40 @@ else (FFMPEG_LIBRARIES AND FFMPEG_INCLUDE_DIR)
-
- find_path(FFMPEG_AVCODEC_INCLUDE_DIR
- NAMES libavcodec/avcodec.h
-- PATHS ${_FFMPEG_AVCODEC_INCLUDE_DIRS} /usr/include /usr/local/include
/opt/local/include /sw/include
-+ PATHS /usr/include/ffmpeg4.4 /usr/local/include /opt/local/include
/sw/include
- PATH_SUFFIXES ffmpeg libav
-+ NO_DEFAULT_PATH
- )
-
- find_library(FFMPEG_LIBAVCODEC
- NAMES avcodec
-- PATHS ${_FFMPEG_AVCODEC_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
-+ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
-+ NO_DEFAULT_PATH
- )
-
- find_library(FFMPEG_LIBAVFORMAT
- NAMES avformat
-- PATHS ${_FFMPEG_AVFORMAT_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
-+ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
-+ NO_DEFAULT_PATH
- )
-
- find_library(FFMPEG_LIBAVUTIL
- NAMES avutil
-- PATHS ${_FFMPEG_AVUTIL_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
-+ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
-+ NO_DEFAULT_PATH
- )
-
-
- find_library(FFMPEG_LIBSWSCALE
- NAMES swscale
-- PATHS ${_FFMPEG_SWSCALE_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
-+ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
-+ NO_DEFAULT_PATH
- )
-
- find_library(FFMPEG_LIBSWRESAMPLE
- NAMES swresample
-- PATHS ${_FFMPEG_SWSCALE_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
-+ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
-+ NO_DEFAULT_PATH
- )
-
- if (FFMPEG_LIBAVCODEC AND FFMPEG_LIBAVFORMAT)
Copied: python-pytorch/repos/community-staging-x86_64/ffmpeg4.4.patch (from rev
1444460, python-pytorch/trunk/ffmpeg4.4.patch)
===================================================================
--- ffmpeg4.4.patch (rev 0)
+++ ffmpeg4.4.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,51 @@
+diff --git a/cmake/Modules/FindFFmpeg.cmake b/cmake/Modules/FindFFmpeg.cmake
+index 04437562ee..dc8fe4aa5c 100644
+--- a/cmake/Modules/FindFFmpeg.cmake
++++ b/cmake/Modules/FindFFmpeg.cmake
+@@ -14,34 +14,40 @@ else (FFMPEG_LIBRARIES AND FFMPEG_INCLUDE_DIR)
+
+ find_path(FFMPEG_AVCODEC_INCLUDE_DIR
+ NAMES libavcodec/avcodec.h
+- PATHS ${_FFMPEG_AVCODEC_INCLUDE_DIRS} /usr/include /usr/local/include
/opt/local/include /sw/include
++ PATHS /usr/include/ffmpeg4.4 /usr/local/include /opt/local/include
/sw/include
+ PATH_SUFFIXES ffmpeg libav
++ NO_DEFAULT_PATH
+ )
+
+ find_library(FFMPEG_LIBAVCODEC
+ NAMES avcodec
+- PATHS ${_FFMPEG_AVCODEC_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
++ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
++ NO_DEFAULT_PATH
+ )
+
+ find_library(FFMPEG_LIBAVFORMAT
+ NAMES avformat
+- PATHS ${_FFMPEG_AVFORMAT_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
++ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
++ NO_DEFAULT_PATH
+ )
+
+ find_library(FFMPEG_LIBAVUTIL
+ NAMES avutil
+- PATHS ${_FFMPEG_AVUTIL_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
++ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
++ NO_DEFAULT_PATH
+ )
+
+
+ find_library(FFMPEG_LIBSWSCALE
+ NAMES swscale
+- PATHS ${_FFMPEG_SWSCALE_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
++ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
++ NO_DEFAULT_PATH
+ )
+
+ find_library(FFMPEG_LIBSWRESAMPLE
+ NAMES swresample
+- PATHS ${_FFMPEG_SWSCALE_LIBRARY_DIRS} /usr/lib /usr/local/lib
/opt/local/lib /sw/lib
++ PATHS /usr/lib/ffmpeg4.4 /usr/local/lib /opt/local/lib /sw/lib
++ NO_DEFAULT_PATH
+ )
+
+ if (FFMPEG_LIBAVCODEC AND FFMPEG_LIBAVFORMAT)
Deleted: fix-building-for-torchvision.patch
===================================================================
--- fix-building-for-torchvision.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ fix-building-for-torchvision.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,25 +0,0 @@
-From 011495d8045c44527fbd7796ce860618120ae127 Mon Sep 17 00:00:00 2001
-From: Butui Hu <[email protected]>
-Date: Fri, 30 Apr 2021 11:36:30 +0800
-Subject: [PATCH] fix building torchvision
-
----
- aten/src/ATen/core/op_registration/op_allowlist.h | 2 +-
- 1 file changed, 1 insertion(+), 1 deletion(-)
-
-diff --git a/aten/src/ATen/core/op_registration/op_allowlist.h
b/aten/src/ATen/core/op_registration/op_allowlist.h
-index f93462bb2cf..12903d1cc09 100644
---- a/aten/src/ATen/core/op_registration/op_allowlist.h
-+++ b/aten/src/ATen/core/op_registration/op_allowlist.h
-@@ -59,7 +59,7 @@ constexpr bool op_allowlist_contains(string_view allowlist,
string_view item) {
- // Returns true iff the given op name is on the allowlist
- // and should be registered
- constexpr bool op_allowlist_check(string_view op_name) {
-- assert(op_name.find("::") != string_view::npos);
-+// assert(op_name.find("::") != string_view::npos);
- // Use assert() instead of throw() due to a gcc bug. See:
- // https://stackoverflow.com/questions/34280729/throw-in-constexpr-function
- // https://github.com/fmtlib/fmt/issues/682
---
-2.31.1
-
Copied:
python-pytorch/repos/community-staging-x86_64/fix-building-for-torchvision.patch
(from rev 1444460, python-pytorch/trunk/fix-building-for-torchvision.patch)
===================================================================
--- fix-building-for-torchvision.patch (rev 0)
+++ fix-building-for-torchvision.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,25 @@
+From 011495d8045c44527fbd7796ce860618120ae127 Mon Sep 17 00:00:00 2001
+From: Butui Hu <[email protected]>
+Date: Fri, 30 Apr 2021 11:36:30 +0800
+Subject: [PATCH] fix building torchvision
+
+---
+ aten/src/ATen/core/op_registration/op_allowlist.h | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/aten/src/ATen/core/op_registration/op_allowlist.h
b/aten/src/ATen/core/op_registration/op_allowlist.h
+index f93462bb2cf..12903d1cc09 100644
+--- a/aten/src/ATen/core/op_registration/op_allowlist.h
++++ b/aten/src/ATen/core/op_registration/op_allowlist.h
+@@ -59,7 +59,7 @@ constexpr bool op_allowlist_contains(string_view allowlist,
string_view item) {
+ // Returns true iff the given op name is on the allowlist
+ // and should be registered
+ constexpr bool op_allowlist_check(string_view op_name) {
+- assert(op_name.find("::") != string_view::npos);
++// assert(op_name.find("::") != string_view::npos);
+ // Use assert() instead of throw() due to a gcc bug. See:
+ // https://stackoverflow.com/questions/34280729/throw-in-constexpr-function
+ // https://github.com/fmtlib/fmt/issues/682
+--
+2.31.1
+
Deleted: fix_include_system.patch
===================================================================
--- fix_include_system.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ fix_include_system.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,13 +0,0 @@
-diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
-index ca673033e1..c79ce8d37b 100644
---- a/torch/utils/cpp_extension.py
-+++ b/torch/utils/cpp_extension.py
-@@ -1760,7 +1760,7 @@ def _write_ninja_file_to_build_library(path,
- common_cflags.append(f'-DPYBIND11_{pname}=\\"{pval}\\"')
-
- common_cflags += [f'-I{include}' for include in user_includes]
-- common_cflags += [f'-isystem {include}' for include in system_includes]
-+ common_cflags += [f'-I{include}' for include in system_includes]
-
- common_cflags += ['-D_GLIBCXX_USE_CXX11_ABI=' +
str(int(torch._C._GLIBCXX_USE_CXX11_ABI))]
-
Copied: python-pytorch/repos/community-staging-x86_64/fix_include_system.patch
(from rev 1444460, python-pytorch/trunk/fix_include_system.patch)
===================================================================
--- fix_include_system.patch (rev 0)
+++ fix_include_system.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,13 @@
+diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
+index ca673033e1..c79ce8d37b 100644
+--- a/torch/utils/cpp_extension.py
++++ b/torch/utils/cpp_extension.py
+@@ -1760,7 +1760,7 @@ def _write_ninja_file_to_build_library(path,
+ common_cflags.append(f'-DPYBIND11_{pname}=\\"{pval}\\"')
+
+ common_cflags += [f'-I{include}' for include in user_includes]
+- common_cflags += [f'-isystem {include}' for include in system_includes]
++ common_cflags += [f'-I{include}' for include in system_includes]
+
+ common_cflags += ['-D_GLIBCXX_USE_CXX11_ABI=' +
str(int(torch._C._GLIBCXX_USE_CXX11_ABI))]
+
Deleted: rocblas-batched.patch
===================================================================
--- rocblas-batched.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ rocblas-batched.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,36 +0,0 @@
-diff --git a/caffe2/utils/math_gpu.cu b/caffe2/utils/math_gpu.cu
-index 2906d0acd9..33610c65f7 100644
---- a/caffe2/utils/math_gpu.cu
-+++ b/caffe2/utils/math_gpu.cu
-@@ -838,6 +838,24 @@ CAFFE2_CUDA_EXPORT void GemmBatched<at::Half,
CUDAContext>(
- at::Half** C,
- CUDAContext* context,
- TensorProto::DataType math_type) {
-+#if defined(USE_ROCM)
-+ // loop over matrices in the batch
-+ for (int i = 0; i < batch_size; ++i) {
-+ Gemm<at::Half, CUDAContext>(
-+ trans_A,
-+ trans_B,
-+ M,
-+ N,
-+ K,
-+ alpha,
-+ A[i],
-+ B[i],
-+ beta,
-+ C[i],
-+ context,
-+ math_type);
-+ }
-+#else
- // Note that cublas follows fortran order, so the order is different from
- // the cblas convention.
- const int lda = (trans_A == CblasNoTrans) ? K : M;
-@@ -912,6 +930,7 @@ CAFFE2_CUDA_EXPORT void GemmBatched<at::Half, CUDAContext>(
- } else {
- CAFFE_THROW("Unsupported math type");
- }
-+#endif
- }
-
Copied: python-pytorch/repos/community-staging-x86_64/rocblas-batched.patch
(from rev 1444460, python-pytorch/trunk/rocblas-batched.patch)
===================================================================
--- rocblas-batched.patch (rev 0)
+++ rocblas-batched.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,36 @@
+diff --git a/caffe2/utils/math_gpu.cu b/caffe2/utils/math_gpu.cu
+index 2906d0acd9..33610c65f7 100644
+--- a/caffe2/utils/math_gpu.cu
++++ b/caffe2/utils/math_gpu.cu
+@@ -838,6 +838,24 @@ CAFFE2_CUDA_EXPORT void GemmBatched<at::Half,
CUDAContext>(
+ at::Half** C,
+ CUDAContext* context,
+ TensorProto::DataType math_type) {
++#if defined(USE_ROCM)
++ // loop over matrices in the batch
++ for (int i = 0; i < batch_size; ++i) {
++ Gemm<at::Half, CUDAContext>(
++ trans_A,
++ trans_B,
++ M,
++ N,
++ K,
++ alpha,
++ A[i],
++ B[i],
++ beta,
++ C[i],
++ context,
++ math_type);
++ }
++#else
+ // Note that cublas follows fortran order, so the order is different from
+ // the cblas convention.
+ const int lda = (trans_A == CblasNoTrans) ? K : M;
+@@ -912,6 +930,7 @@ CAFFE2_CUDA_EXPORT void GemmBatched<at::Half, CUDAContext>(
+ } else {
+ CAFFE_THROW("Unsupported math type");
+ }
++#endif
+ }
+
Deleted: test.py
===================================================================
--- test.py 2023-04-10 16:10:27 UTC (rev 1444460)
+++ test.py 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,7 +0,0 @@
-#!/usr/bin/env python
-
-import torch
-
-d = torch.device('cuda')
-a = torch.rand(1, 2).to(d)
-print(a + 0)
Copied: python-pytorch/repos/community-staging-x86_64/test.py (from rev
1444460, python-pytorch/trunk/test.py)
===================================================================
--- test.py (rev 0)
+++ test.py 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,7 @@
+#!/usr/bin/env python
+
+import torch
+
+d = torch.device('cuda')
+a = torch.rand(1, 2).to(d)
+print(a + 0)
Deleted: use-system-libuv.patch
===================================================================
--- use-system-libuv.patch 2023-04-10 16:10:27 UTC (rev 1444460)
+++ use-system-libuv.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -1,13 +0,0 @@
-diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
-index 06464e799a..93410bc210 100644
---- a/cmake/Dependencies.cmake
-+++ b/cmake/Dependencies.cmake
-@@ -1346,7 +1346,7 @@ if(USE_DISTRIBUTED AND USE_TENSORPIPE)
- set(TP_USE_CUDA ON CACHE BOOL "" FORCE)
- set(TP_ENABLE_CUDA_IPC ON CACHE BOOL "" FORCE)
- endif()
-- set(TP_BUILD_LIBUV ON CACHE BOOL "" FORCE)
-+ set(TP_BUILD_LIBUV OFF CACHE BOOL "" FORCE)
- set(TP_STATIC_OR_SHARED STATIC CACHE STRING "" FORCE)
-
- add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/tensorpipe)
Copied: python-pytorch/repos/community-staging-x86_64/use-system-libuv.patch
(from rev 1444460, python-pytorch/trunk/use-system-libuv.patch)
===================================================================
--- use-system-libuv.patch (rev 0)
+++ use-system-libuv.patch 2023-04-10 16:11:07 UTC (rev 1444461)
@@ -0,0 +1,13 @@
+diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
+index 06464e799a..93410bc210 100644
+--- a/cmake/Dependencies.cmake
++++ b/cmake/Dependencies.cmake
+@@ -1346,7 +1346,7 @@ if(USE_DISTRIBUTED AND USE_TENSORPIPE)
+ set(TP_USE_CUDA ON CACHE BOOL "" FORCE)
+ set(TP_ENABLE_CUDA_IPC ON CACHE BOOL "" FORCE)
+ endif()
+- set(TP_BUILD_LIBUV ON CACHE BOOL "" FORCE)
++ set(TP_BUILD_LIBUV OFF CACHE BOOL "" FORCE)
+ set(TP_STATIC_OR_SHARED STATIC CACHE STRING "" FORCE)
+
+ add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/tensorpipe)
