Change: 128958134

Change: 128401884

improvements for fp16
Added SpecialFunctions to the list of eigen headers TensorFlow depends on
Change: 127264575

Change: 127253427

improvements for fp16
Change: 127233960

Change: 126335170

Adds a "currentThreadIndex" method to Eigen's ThreadPoolDevice. Use it to handle per-thread buffer allocation for the tileable executor without resorting to thread_local that is not fully supported on Android.
Change: 126009029

Upgraded Eigen to the latest version that provides new scan operations. This will enable the implementation of the cumsum operation in TensorFlow
Change: 125697517

performance of the toy mnist training by 1 order of magnitude
Change: 124374286

NEW
BM_fullReduction/10        4591       4595     153149  20.8M items/s
BM_fullReduction/64        5073       5075     100000  770.0M items/s
BM_fullReduction/512       9067       9070      75263  26.9G items/s
BM_fullReduction/4k      243984     244125       2868  64.0G items/s
BM_fullReduction/5k      359125     359273       1951  64.8G items/s

OLD
BM_fullReduction/10        9085       9087      74395  10.5M items/s
BM_fullReduction/64        9478       9478      72014  412.1M items/s
BM_fullReduction/512      14643      14646      46902  16.7G items/s
BM_fullReduction/4k      260338     260384       2678  60.0G items/s
BM_fullReduction/5k      385076     385178       1818  60.5G items/s
Change: 124290852

gradients, some variants etc.).
Change: 124197406

Change: 124183870

gradients, some variants etc.).
Change: 123967787

gradients, some variants etc.).
Change: 123967117

Change: 123901292

Change: 123659102

Change: 123468144

tf.train.Example.
Change: 123445810

up to the OSS build yet, we're working on it.)
Change: 123248081

Change: 123238579

Change: 123026122

Implements an authentication mechanism based on Application Default Credentials:
https://developers.google.com/identity/protocols/application-default-credentials
https://developers.google.com/identity/protocols/OAuth2ServiceAccount
Change: 122741738

with many cpu cores

For example, the wall time for the following tutorial went down from 13m35 to 5m27:
bazel run -c opt --copt=-mavx tensorflow/examples/tutorials/word2vec/word2vec_basic
Change: 122462177

and remove the cuda_crosstool_condition build condition.  Now if_cuda is
just using_nvcc || using_gcudacc.
Change: 122291892

This is unused at the moment, but will eventually let us build CUDA code
with vanilla clang.
Change: 122289910

This has no practical effect, as CUDA builds are always with nvcc, but
it lets us modify the build config rule
//third_party/gpus/cuda:using_nvcc so it returns true, rather than
false, for CUDA builds.
Change: 122288952

Change: 122192081

by about 3 orders of magnitude as well as some partial reductions by 30% when using cuda 7.5 or above
Change: 122191448

Change: 121586635

gpus
Updated the check numerics code to make it compatible with fp16
Change: 120980302

Change: 120739269

tensorflow: switch to eigen thread pool

This is first step of switching tensorflow to the new
non-blocking thread pool in eigen.
Change: 120510292

Change: 120505517

offered by AWS
Change: 120369420

sigmoid of fp16 and introduces a condition estimator.
Change: 119907721

Change: 119850987

improvements for fp16
Change: 119771118

Change: 119768540

Change: 119458778

as well as fp16
Change: 119398881