search

conda-forge / pytorch-cpu-feedstock

A conda-smithy repository for pytorch-cpu.
BSD 3-Clause "New" or "Revised" License
19 stars 50 forks source link

Windows builds #32

Open h-vetinari opened 3 years ago

h-vetinari commented 3 years ago

There's CI for windows, but it fails. The only available builds are for pytorch-cpu, v1.1.0 more than 1.5 years ago, nothing for the newer pytorch output yet.

Would be cool, as I'm trying to add builds for windows for https://github.com/conda-forge/allennlp-feedstock/pull/12

h-vetinari commented 3 years ago

Just saw the work in #27, but even with cuda_compiler_version=="None", the azure machine runs out of memory...

hmaarrfk commented 3 years ago

i think we need somebody to build this locally then upload it. but licensing issues have stopped us from doing that with windows.

h-vetinari commented 3 years ago

How do licencing issues come into this? 🤔 (that wouldn't equally apply e.g. to https://github.com/conda-forge/qt-feedstock/)

hmaarrfk commented 3 years ago

i'm not too sure. I think a ping to to the right person with a large windows machine might go a long way.

@isuruf any ideas?

isuruf commented 3 years ago

Build fails with,

Building CXX object caffe2\torch\CMakeFiles\torch_python.dir\csrc\Device.cpp.obj
FAILED: caffe2/torch/CMakeFiles/torch_python.dir/csrc/Device.cpp.obj
"C:\PROGRA~2\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.28.29333\bin\Hostx64\x64\cl.exe"  /nologo /TP -DBUILDING_TESTS -DFMT_HEADER_ONLY=1 -DMINIZ_DISABLE_ZIP_READER_CRC32_CHECKS -DONNXIFI_ENABLE_EXT=1 -DONNX_ML=1 -DONNX_NAMESPACE=onnx_torch -DTHP_BUILD_MAIN_LIB -DUSE_C10D -DUSE_C10D_GLOO -DUSE_DISTRIBUTED -DUSE_EXTERNAL_MZCRC -DUSE_NUMPY -DWIN32_LEAN_AND_MEAN -D_CRT_SECURE_NO_DEPRECATE=1 -Dtorch_python_EXPORTS -Iaten\src -I..\aten\src -I. -I..\ -I..\cmake\..\third_party\benchmark\include -Icaffe2\contrib\aten -I..\third_party\onnx -Ithird_party\onnx -I..\third_party\foxi -Ithird_party\foxi -I..\torch\.. -I..\torch\..\aten\src -I..\torch\..\aten\src\TH -Icaffe2\aten\src -Ithird_party -I..\torch\..\third_party\valgrind-headers -I..\torch\..\third_party\gloo -I..\torch\..\third_party\onnx -I..\torch\csrc -I..\torch\csrc\api\include -I..\torch\lib -I..\torch\lib\libshm_windows -I..\torch\csrc\api -I..\c10\.. -Ithird_party\ideep\mkl-dnn\include -I..\third_party\ideep\mkl-dnn\src\..\include -I..\third_party\fmt\include -I..\torch\lib\c10d\.. -I..\torch\lib\c10d\..\..\.. -Ithird_party\gloo -I..\cmake\..\third_party\gloo -I..\cmake\..\third_party\googletest\googlemock\include -I..\cmake\..\third_party\googletest\googletest\include -I..\third_party\protobuf\src -I..\third_party -I..\cmake\..\third_party\eigen -IC:\bld\pytorch-recipe_1609392400304\_h_env\include -IC:\bld\pytorch-recipe_1609392400304\_h_env\lib\site-packages\numpy\core\include -I..\cmake\..\third_party\pybind11\include -I..\third_party\ideep\mkl-dnn\include -I..\third_party\ideep\include /DWIN32 /D_WINDOWS /GR /EHsc /w /bigobj -openmp:experimental -DNDEBUG -DUSE_FBGEMM -DUSE_VULKAN_WRAPPER -DHAVE_AVX_CPU_DEFINITION -DHAVE_AVX2_CPU_DEFINITION /MD /O2 /Ob2 /DNDEBUG /w /bigobj -DNDEBUG -DCAFFE2_USE_GLOO -DUSE_GCC_GET_CPUID -DUSE_AVX -DUSE_AVX2 -DTH_HAVE_THREAD /EHsc /DNOMINMAX /wd4267 /wd4251 /wd4522 /wd4838 /wd4305 /wd4244 /wd4190 /wd4101 /wd4996 /wd4275 /bigobj -std:c++14 /showIncludes /Focaffe2\torch\CMakeFiles\torch_python.dir\csrc\Device.cpp.obj /Fdcaffe2\torch\CMakeFiles\torch_python.dir\ /FS -c ..\torch\csrc\Device.cpp
C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.28.29333\include\complex(354): error C2039: 'copysign': is not a member of '`global namespace''

cc @seemethere

h-vetinari commented 9 months ago

Would the new cirun setup support windows, or is it linux-only?

pgunn commented 4 months ago

Any news on this? I'm hoping to migrate a 3-OS package I maintain from Tensorflow to Pytorch to deal with Google's more-or-less abandonment of tensorflow on windows, and not having Pytorch builds will be a showstopper.

baszalmstra commented 3 months ago

Any news on this? I'm hoping to migrate a 3-OS package I maintain from Tensorflow to Pytorch to deal with Google's more-or-less abandonment of tensorflow on windows, and not having Pytorch builds will be a showstopper.

Im still working on this in: https://github.com/conda-forge/pytorch-cpu-feedstock/issues/32 Hopefully not long now!

pgunn commented 2 months ago

@baszalmstra Hello, sorry to keep pestering you on this; how's this going?

baszalmstra commented 2 months ago

@pgunn The "only" thing that needs to happen is use protobuf from conda-forge instead of using the vendored one.

I have not gotten around to continue this yet. Hope to get back to it soon!

pgunn commented 2 months ago

Ha, not surprising to see Google vendoring stuff (I never worked there, but in another big tech company I used to work at we had an annoying tendency to vendor things). It creates a lot of tangles.

pgunn commented 1 month ago

@baszalmstra Hello, Sorry to bug you again on this, but with a mix of anaconda starting to become all-elbows with enforcing its new license for the defaults channel and https://github.com/conda-forge/conda-forge-ci-setup-feedstock/pull/339 this has become urgent.

Is there any chance you might have time to finish this anytime soon? I regret that I lack the context or time to be able to hop in and help.

pgunn commented 1 month ago

(if you have a tip jar or something I'd be happy to toss something in, in appreciation for the work you've done on this)

hmaarrfk commented 1 month ago

(if you have a tip jar or something I'd be happy to toss something in, in appreciation for the work you've done on this)

I just opened https://github.com/conda-forge/pytorch-cpu-feedstock/issues/268 to track better funding for this.

baszalmstra commented 1 month ago

Hey @pgunn ! I apologize, I haven't been able to find the time.

Let me get a local build up and running today so I can at least report the issue Im facing.

pgunn commented 2 weeks ago

@baszalmstra How's this going?

pgunn commented 3 days ago

Any news on this?

baszalmstra commented 3 days ago

Best to follow the PR here: https://github.com/conda-forge/pytorch-cpu-feedstock/pull/231