| Commit message (Collapse) | Author | Age |
|
|
|
|
|
| |
into the number of shards used. This is a variant of threadpool::parallelFor
PiperOrigin-RevId: 213920649
|
|
|
|
| |
PiperOrigin-RevId: 201236564
|
|
|
|
|
|
| |
fork-join parallelism (e.g., Eigen/ThreadpoolDevice or Shard).
PiperOrigin-RevId: 200756626
|
|
|
|
| |
PiperOrigin-RevId: 198582954
|
|
|
|
|
| |
Get rid of EIGEN_USE_NONBLOCKING_THREAD_POOL symbol.
Change: 137961673
|
|
|
|
| |
Change: 123900938
|
|
|
|
|
|
| |
max_parallelism is less than number of workers in the pool.
Change: 123139558
|
|
|
|
|
|
|
| |
tensorflow::ThreadPool::ParallelFor, such that limits on number of tasks provided to work sharder can be respected.
Adds a rudimentary unit test for ParallelFor.
Change: 122695936
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Provide finer-grained Shard parallelization for the new non-blocking thread pool.
This significantly resembles the parallel for algorithm in eigen executors:
we choose a good block size based on amount of work and parallel efficiency,
and then use recursive division in halves.
Benchmark Time(ns): old new diff CPU(ns): old new diff
==========================================================================================
cpu_RandomUniform/1M 647541 301220 -53.48% 9576553 10553619 +10.20%
cpu_RandomUniform/2M 1116118 495724 -55.58% 18285896 19635580 +7.38%
cpu_RandomUniform/8M 2691384 1671594 -37.89% 67830397 72105713 +6.30%
cpu_RandomNormal/1M 2126780 1269039 -40.33% 46887528 53197040 +13.46%
cpu_RandomNormal/2M 3529118 2350399 -33.40% 94337705 104481933 +10.75%
cpu_RandomNormal/8M 12429704 8984079 -27.72% 383278086 410900286 +7.21%
cpu_TruncatedNormal/1M 2513508 1504161 -40.16% 59181937 66096798 +11.68%
cpu_TruncatedNormal/2M 4012258 2890855 -27.95% 122164300 129760843 +6.22%
cpu_TruncatedNormal/8M 17628696 11159204 -36.70% 465946492 513345503 +10.17%
TESTED:
- passed opensource_build
http://ci.tensorflow.org/view/Internal/job/tensorflow-cl-presubmit-multijob/281/
Change: 121971279
|
|
|
|
| |
Change: 117520810
|
|
|
|
|
| |
After this we can replace port.h with types.h.
Change: 112727463
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Changes:
* error message that refers to removed `DefaultSession` method.
* -Wnull-conversion warnings
* the "_start_time" attr for recvs when the flag "--brain_enable_scheduling_for_recvs" is set.
* typo in tutorial data download progress message.
* a typo ("however their installing"=>"however installing").
* typo, rename "TensorFlow Mechanics" to "How To" to be consistent with the website.
* a typo ("subtact"=>"subtract").
* protobuf examples in comments in tensorflow::Example.proto.
* formula formatting in MNIST beginner tutorial
* negative fraction-of-queue-full stats
* protobuf inclusion path so that Android demo will build under Blaze.
* small typo (moderatly > moderately)
* Session.run() to check that tensor arguments come from the session's graph.
* another six import
* seq2seq typo in bazel command
Base CL: 108349164
|
|
TensorFlow is an open source software library for numerical computation
using data flow graphs.
Base CL: 107276108
|