Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Feb 22, 2017

• Fix an assertion issue in executor.cc when running on 32bit CPU

  Show description for 4c47723
  tensorflower-gardener

  committed
  4c47723

  Copy full SHA for 4c47723

Commits on Feb 20, 2017

• A couple of performance oriented changes:

  Show description for 55d1978
  

  jeffreyadean

  authored and

  tensorflower-gardener

  committed
  55d1978

  Copy full SHA for 55d1978

Commits on Feb 14, 2017

• Fix code that ignores tensorflow::Status.

  Show description for bc225bf
  

  hawkinsp

  authored and

  tensorflower-gardener

  committed
  bc225bf

  Copy full SHA for bc225bf

Commits on Feb 10, 2017

• Record persistent memory in tracking allocator and cost model.

  Show description for f128935
  

  Yuefeng Zhou

  authored and

  tensorflower-gardener

  committed
  f128935

  Copy full SHA for f128935

Commits on Feb 6, 2017

• Streamlined representation of PendingCounts data structure.

  Show description for de94a13
  

  jeffreyadean

  authored and

  tensorflower-gardener

  committed
  de94a13

  Copy full SHA for de94a13

Commits on Feb 4, 2017

• This implemented Jeff's idea to avoid most of the refcount related atomic operations when propagating tensors to the input slots of its consumers.

  Show description for 8cfb9ac
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  8cfb9ac

  Copy full SHA for 8cfb9ac

Commits on Feb 2, 2017

• Improve memory system behavior of core graph execution module.

  Show description for 9f827c6
  

  jeffreyadean

  authored and

  tensorflower-gardener

  committed
  9f827c6

  Copy full SHA for 9f827c6

• Add GraphView internal helper class, which is initialized from the Graph

  Show description for 6ced722
  

  jeffreyadean

  authored and

  tensorflower-gardener

  committed
  6ced722

  Copy full SHA for 6ced722

Commits on Jan 31, 2017

• 1. Always run the cancellation callbacks in the event of any execution error. Without the fix, async ops can deadlock when error occurs. This may fix some of the deadlock problems we have seen with…

  Show description for 26460c8
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  26460c8

  Copy full SHA for 26460c8

Commits on Dec 15, 2016

• Make the context correct for the gradient of mixed nesting of conds and loops.

  Show description for 6764c34
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  6764c34

  Copy full SHA for 6764c34

Commits on Nov 28, 2016

• Uses containers for per_step resources instead of a separate resource_mgr.

  Show description for b571f2e
  tensorflower-gardener

  committed
  b571f2e

  Copy full SHA for b571f2e

Commits on Nov 22, 2016

• Do not always call device->Sync().

  Show description for 67f8aa2
  tensorflower-gardener

  committed
  67f8aa2

  Copy full SHA for 67f8aa2

Commits on Nov 18, 2016

• Small tidy-up. Fix the comment on the combination of CSE and control flow.

  Show description for bcdb8eb
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  bcdb8eb

  Copy full SHA for bcdb8eb

Commits on Nov 15, 2016

• Generalized ExecutorImpl::InferAllocAttr to handle more types of accelerators.

  Show description for fb4c9bb
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  fb4c9bb

  Copy full SHA for fb4c9bb

Commits on Oct 31, 2016

• Use gtl::FlatMap and gtl::FlatSet in several places instead of

  Show description for 12629a0
  tensorflower-gardener

  committed
  12629a0

  Copy full SHA for 12629a0

Commits on Oct 26, 2016

• Introduced per-loop PendingCount and Entry[]. This could significantly reduce the cost of iteration creation if the partition graph is large but the loop body is small, as explained in this old todo:

  Show description for de7aaad
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  de7aaad

  Copy full SHA for de7aaad

Commits on Oct 19, 2016

• Move Device::Sync() from the execution of SINK node to ExecutorState::Finish().

  Show description for c2c59fc
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  c2c59fc

  Copy full SHA for c2c59fc

Commits on Oct 12, 2016

• Small changes on CL/135819904 based on ZF's review comments.

  Show description for 9c88695
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  9c88695

  Copy full SHA for 9c88695

Commits on Oct 11, 2016

• Reworked the locking in the executor. This is mainly to improve the performance of some use cases that have many lightweight nested loops on a single device. See idunning@'s comment below.

  Show description for 1e3bdbc
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  1e3bdbc

  Copy full SHA for 1e3bdbc

Commits on Oct 7, 2016

• A few micro-optimizations / cleanup in tensorflow/core.

  Show description for 09326f0
  tensorflower-gardener

  committed
  09326f0

  Copy full SHA for 09326f0

Commits on Oct 5, 2016

• A small cleanup: Removed a useless cleanup call.

  Show description for edf957b
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  edf957b

  Copy full SHA for edf957b

Commits on Sep 30, 2016

• Some small refactoring in preparation for a deeper change of locking in executor.

  Show description for 7cbf945
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  7cbf945

  Copy full SHA for 7cbf945

Commits on Sep 26, 2016

• Skip stats collection for ops in untaken branch of cond.

  Show description for b9fdf5b
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  b9fdf5b

  Copy full SHA for b9fdf5b

Commits on Sep 22, 2016

• Use hardware stats to build the cost model for operations that run on GPU

  Show description for ca69f69
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  ca69f69

  Copy full SHA for ca69f69

Commits on Aug 19, 2016

• Fix crash bug in debug printout.

  Show description for 0beb771
  tensorflower-gardener

  committed
  0beb771

  Copy full SHA for 0beb771

Commits on Aug 18, 2016

• Plumb the per-step resource manager through to the executor used when executing functions.

  Show description for ace4f5b
  tensorflower-gardener

  committed
  ace4f5b

  Copy full SHA for ace4f5b

Commits on Aug 16, 2016

• Removes dead code.

  Show description for 2c5718c
  tensorflower-gardener

  committed
  2c5718c

  Copy full SHA for 2c5718c

Commits on Aug 10, 2016

• Fixed a gradient bug when a while loop is executed on an untaken branch of a conditional. Also fixed the doc about Merge op.

  Show description for a8c3076
  

  yuanbyu

  authored and

  tensorflower-gardener

  committed
  a8c3076

  Copy full SHA for a8c3076

Commits on Aug 8, 2016

• Make it possible to initialize the step_resource_mgr in the Executor before any operators run.

  Show description for c8d5be8
  tensorflower-gardener

  committed
  c8d5be8

  Copy full SHA for c8d5be8

• Reduce some allocations on grpc code paths.

  Show description for 8f6a17f
  

  Suharsh Sivakumar

  authored and

  tensorflower-gardener

  committed
  8f6a17f

  Copy full SHA for 8f6a17f

Commits on Jul 26, 2016

• Deleted dead code

  Show description for 4d82547
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  4d82547

  Copy full SHA for 4d82547

Commits on Jul 18, 2016

• Reduce overhead of running kernels:

  Show description for b8ccc5a
  tensorflower-gardener

  committed
  b8ccc5a

  Copy full SHA for b8ccc5a

Commits on Jun 27, 2016

• A series of changes to significantly reduce the number of allocations

  Show description for a120b0b
  tensorflower-gardener

  committed
  a120b0b

  Copy full SHA for a120b0b

Commits on Jun 26, 2016

• TF debugger core callback and DebugGateway

  Show description for 0868ce6
  

  caisq

  authored and

  tensorflower-gardener

  committed
  0868ce6

  Copy full SHA for 0868ce6

Commits on Jun 2, 2016

• Update copyright for 3p/tf/core.

  Show description for c8b59c0
  

  A. Unique TensorFlower

  authored and

  tensorflower-gardener

  committed
  c8b59c0

  Copy full SHA for c8b59c0