Patch the windows branch with CMake support#4043
Closed
recheliu wants to merge 7 commits intoBVLC:windowsfrom
recheliu:windows
Closed
Patch the windows branch with CMake support#4043recheliu wants to merge 7 commits intoBVLC:windowsfrom recheliu:windows
recheliu wants to merge 7 commits intoBVLC:windowsfrom
recheliu:windows
Conversation
…cmake. Modified CMakeLists.txt Modified cmake/Targets.cmake Modified cmake/Templates/caffe_config.h.in Modified examples/cpp_classification/classification.cpp Modified examples/mnist/convert_mnist_data.cpp Modified include/caffe/layer.hpp Modified include/caffe/solver.hpp Modified include/caffe/test/test_gradient_check_util.hpp Modified include/caffe/util/db.hpp Modified include/caffe/util/io.hpp Modified python/CMakeLists.txt Modified src/caffe/common.cpp Modified src/caffe/layer_factory.cpp Modified src/caffe/layers/contrastive_loss_layer.cpp Modified src/caffe/layers/data_layer.cpp Modified src/caffe/layers/scale_layer.cpp Modified src/caffe/parallel.cpp Modified src/caffe/solver.cpp Modified src/caffe/test/CMakeLists.txt Modified src/caffe/test/test_contrastive_loss_layer.cpp Modified src/caffe/util/hdf5.cpp Modified src/caffe/util/signal_handler.cpp Modified tools/CMakeLists.txt 1. [MOD] Automatically patched. The diff was created & patched via the following command (64d9e1acbefe1660efa22d9bd9630b877d93a0ae is the commit in our internal repo): $ git diff -w 4b8fa07 64d9e1acbefe1660efa22d9bd9630b877d93a0ae > CaffeOnWindows.patch # Copy CaffeOnWindows.patch to the current source folder. $ patch -p1 < CaffeOnWindows.patch Modified cmake/Dependencies.cmake Modified cmake/External/gflags.cmake Modified cmake/Modules/FindLevelDB.cmake Modified include/caffe/common.hpp 1. [MOD] Manually fixed the parts that cannot be patched. Add cmake/Msvc.cmake 1. [1ST] Check in the cmake rule for Visual Studio. The rule will scan for registered classes and add code to force them to be linked. Add include/caffe/msvc/caffe_msvc.hpp 1. [1ST] Check in the misc setting for compatibility on Windows. Add include/caffe/msvc/caffe_msvc_io.hpp 1. [1ST] Check in the io routines that mimic the postix one with Windows API. Add include/caffe/msvc/caffe_msvc_registry.hpp 1. [1ST] Check in the code that define the rule to force the linking of registered classes. Add src/caffe/msvc/caffe_classes.cpp 1. [1ST] Check in classes that might be not registered if certain preprocessor is disabled.
…cmake. Modified cmake/Dependencies.cmake Modified cmake/External/gflags.cmake Modified cmake/Modules/FindLevelDB.cmake Modified include/caffe/common.hpp 1. [CLEAN] Remove local test labels.
…cmake. Modified cmake/Dependencies.cmake Modified cmake/External/gflags.cmake 1. [MOD] Correct the indents.
Modified CMakeLists.txt Modified src/caffe/test/CMakeLists.txt 1. [MOD] Move the definition of CMAKE_BUILD from unit test to the root CMakeLists.txt. Modified include/caffe/msvc/caffe_msvc_registry.hpp 1. [MOD] Only include the CMake-generated caffe_classes.hpp when using CMake.
Modified CMakeLists.txt Modified include/caffe/msvc/caffe_msvc_registry.hpp Modified src/caffe/test/CMakeLists.txt 1. [CLEAN] Remove local test labels.
Modified include/caffe/util/io.hpp 1. [MOD] Manually rollback io.hpp.
Contributor
Merged
Author
|
Hi @willyd, Thanks for your reference. I compared your PR with mine. While our fixes to CMake are pretty similar, the main differences are listed below:
On the other hand, I saw that you almost finished your PR. The Travis CI failures are mainly related to styles. AppVeyor failure is that the file solver_factory.cpp is not compiled. If you can fixed them, I think that we can just use your PR. |
Contributor
I would say that another major difference is that the forced symbol linking is handled differently too. Your PR parses the sources files, while mine lists the symbols from the generated caffe.lib. I don't actually require Ninja, but I haven't tested the Visual Studio generator so it might not work but it should not be too hard to support it as well.
Note quite finished, it still needs more testing and proper CI support but where getting close you are right.
Just updated the PR with a fix for this. Any contribution is welcome. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This pull request will patch Caffe so its source code and CMake file can work on both Linux and Windows. We have tested the same files on both Windows and Linux as seen below.
This patch does not address the building of 3rd party libraries though.
The followings are the tests we conducted for this pull request:
Build via CMake on Windows. CPU-only:
Build via CMake on Windows. CPU & GPU:
Build via the original windows/Caffe.sln:
Build via CMake on Linux. CPU & GPU:
Note that the LMDB for mnist/cifar10 databases were generated on Linux.
The following is our test environment;
[Windows]
Windows 7 x64 with Visual Studio 2013.
CUDA: 7.0 for CMake and 7.5 for the original .sln file.
boost: 1.55
CUDNN: 3.0
BLAS: OpenBLAS.
Python: WinPython 2.7.10.3.
[Linux]
Ubuntu 14.04.
CUDA: 7.0.
CUDNN: 3.0
BLAS: Atlas.
Python: 2.7.