| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
|
| |
* Fix build error, where nccl requires -lrt link option
* Remove config_setting defines in nccl.BUILD and curl.BUILD
* Fix typo in curl.BUILD
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* Adding FreeBSD compatibility (non-breaking changes)
* Adding FreeBSD compatibility (BREAKING changes)
- FreeBSD has libdl integrated into libc so -ldl is not available
* Adding FreeBSD compatibility (BREAKING changes)
- FreeBSD has libdl integrated into libc so -ldl is not available
* - Define FreeBSD config
- Clear -ldl linkopts for FreeBSD only
* Clear -ldl linkopts for FreeBSD only
* Changing to local condition just as previously done with darwin
* Removing one indentation for syntax validation
* Removing one indentation here too
* - Changing new config name to lowercase
- Fixing buildifier warning
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I'm working on fixing https://github.com/bazelbuild/bazel/issues/1681, which
will change the way Bazel organizes the execution root. This has two
implications for tensorflow:
* The protobuf library needs to be updated to a version that works with
old/new versions of Bazel (older versions won't be able to find
external protos).
* The curl library needs to use ../curl as an include prefix in new
versions of Bazel and external/curl in old versions.
This patch should be forward/backward compatible.
|
| |
|
| |
|
|\ |
|
| |
| |
| |
| | |
Change: 147051664
|
| |
| |
| |
| | |
Change: 146970526
|
| |
| |
| |
| | |
Change: 146918929
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
running configure. (#6921)
* Added Intel MKL graph optimization code and the ability to enable MKL when running configure.
* Fixed inclusion of MKL into build files.
* Made changes to MKL graph optimization files based on review comments.
* Adding support for MKL matmul op
* Revert "Adding support for MKL matmul op"
This reverts commit 20d6cdd865dbe790bd03765e82b709bd750d8823.
* Changes based on Vijay Vasudevan's comments
* Made format-related fixes to some build files.
* Setting the execute bit (755) for tensorflow/tools/pip_package/build_pip_package.sh
|
|\| |
|
| |
| |
| |
| |
| |
| |
| | |
Fixes the problem of XLA being unable to find libdevice files if not executed
through bazel. For example,
http://stackoverflow.com/questions/41729019/notfounderror-running-tensorflow-xla-example-libdevice-compute-35-10-bc/41800414#41800414
Change: 146329331
|
| |
| |
| |
| | |
Change: 146316196
|
|\ \
| | |
| | | |
Upgraded to libxsmm 1.7
|
| | |
| | |
| | |
| | | |
* Fixed libxsmm_config_arguments: Fixed the incorrect value supposed to trigger auto-prefetch. Fixed the 0-threshold, which is now accounted for in LIBXSMM (by just populating the default threshold). The problem arised from the assumption "threshold: fallback to BLAS if n*m*k above this", which is wrong (the threshold populates an upper bound until which JIT code is generated). The previous configuration perhaps caused all sorts of issues due to other values derived from the 0-threshold. Note, explicitly JIT'ting code is/was never subject to a threshold.
|
|\ \ \
| | |/
| |/| |
|
| |/
|/|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Update _find_cuda_define() and _cudnn_version() to use
_find_cudnn_header_dir() to locate the cudnn.h header. The latter
already knows how to look in multiple locations for the header.
This works on POWER Ubuntu when cuDNN is installed via tarball or
deb package and CUDNN_INSTALL_PATH is set to /usr/local/cuda-8.0.
In those cases, cudnn.h is ultimately found at:
$ sudo tar -C /usr/local -xzvf cudnn-8.0-linux-ppc64le-v5.1.tgz
/usr/local/cuda-8.0/targets/ppc64le-linux/include/cudnn.h
-or-
$ sudo dpkg -i libcudnn5*deb
/usr/include/cudnn.h (is soft link to /etc/alternatives/libcudnn
is soft link to /usr/include/powerpc64le-linux-gnu/cudnn_v5.h)
|
| |
| |
| |
| | |
Change: 145804953
|
| |
| |
| |
| |
| | |
The DSO loader intermediate layer is no longer required, so it is removed here.
Change: 145589281
|
| |
| |
| |
| |
| |
| |
| |
| | |
The updated version contains a fix that makes hashing consistent when CPU
endian changes.
Fixed #6823
Change: 145563886
|
| |
| |
| |
| |
| | |
server.
Change: 145475050
|
|\| |
|
| |
| |
| |
| | |
Change: 145363673
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
tensorflow.
This change is motivated by the need to use middleware to provide authentication internally. It also makes the TensorBoard backend more embeddable/re-usable.
* Adds a dependency on werkzeug for convenience in setting up WSGI apps.
* Replaces "TensorBoardHandler" with "TensorBoardWSGIApp". Now each route is a WSGI application. Generally uses convenient werkzeug decorators for making these applications.
* Removes handler.py, server.py; these are merged into application.py
* The Respond method in http.py now returns a Werkzeug Response object.
* Modify projector plugin to also be a WSGI application.
* Add a few error messages for helpfulness.
* Remove some dead code along the way. It has to do with sampling, but we never incorporated it anywhere.
Change: 145183192
|
|\|
| |
| |
| |
| |
| | |
Fixed conflicts in by taking the latest from internal:
tensorflow/core/kernels/sparse_matmul_op.cc
tensorflow/workspace.bzl
|
| |
| |
| |
| |
| | |
This is in preparation for making TensorBoard a WSGI app.
Change: 144917110
|
|\ \
| | |
| | |
| | |
| | | |
Conflicts:
tensorflow/workspace.bzl
|
| | |
| | |
| | |
| | |
| | |
| | | |
when TensorFlow is linked as a submodule.
Change: 144764356
|
| |\|
| | |
| | |
| | |
| | | |
Conflict tensorflow/contrib/layers/python/layers/layers.py:
Preserved both indentation on "decay" change and doc fix on "center".
|
| | |
| | |
| | |
| | | |
Change: 144396000
|
| | |
| | |
| | |
| | |
| | |
| | | |
submodule. Undo change when this use-case is supported in native.http_archive in Bazel.
Change: 144390772
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
and other places where we call malloc/free.
- Only enabled on Linux for now.
- Added as a ./configure option defaulting to enabled.
Change: 144266237
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
third_party/eigen3/unsupported/Eigen/CXX/FixedPoint and
third_party/eigen3/unsupported/Eigen/CXX/Tensor use different paths to
reach into unsupported/Eigen/CXX/src/Tensor. The former's paths cause
the build errors documented in GitHub issue #6558. This commit removes
the includes from third_party/eigen3/unsupported/Eigen/CXX/FixedPoint
pointing into third_party/eigen3/unsupported/Eigen/CXX/FixedPoint,
because they do not appear to be necessary.
|
| |\| |
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
XLA is a compiler-based linear algebra execution engine that targets CPUs, GPUs and custom accelerators.
XLA is still experimental; we are releasing it early to get the community involved.
Change: 143990941
|
| |\| |
|
| | |
| | |
| | |
| | |
| | | |
cuda_default_copts() and moved CUDA-specific options there.
Change: 143811735
|
| | |
| | |
| | |
| | | |
Change: 143710639
|
| | |
| | |
| | |
| | | |
Change: 143689873
|
| | |
| | |
| | |
| | | |
Change: 143412147
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Currently, if a user specifies the full cuda or cudnn versions (with
major, minor, and patch numbers), cuda_configure will fail because we
only detect the version numbers we use (major and minor for cuda and
major only for cudnn). This change detects full versions and matches the
user-provided versions against the full versions. If the user provides
only partial version numbers (e.g. without patch version), then the
match will still succeed if the provided numbers match the detected
numbers.
|
| |\| |
|
| | |
| | |
| | |
| | | |
Change: 143204614
|
| | |
| | |
| | |
| | |
| | | |
This frees up space on the TensorFlow GitHub home page!
Change: 143161497
|
| | |
| | |
| | |
| | | |
Change: 143131671
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Google Cloud Platform support is now available for everyone by default.
Fixes #5349
Change: 143077241
|
|/| |
| | |
| | | |
Write rules to construct headers for jpeg-turbo using existing templates.
|
| | | |
|
| | | |
|