Commits

Branch selector

 master

User selector

benoitsteiner

Datepicker

All time

Commit History

Commits on May 4, 2017

• Added missing header file to fix the MacOS builds

  Show description for ddf30ca
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  ddf30ca

  Copy full SHA for ddf30ca

Commits on May 3, 2017

• Made sure that both queue shapes and types are updated in one single pass instead of two.

  Show description for 27aaf4a
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  27aaf4a

  Copy full SHA for 27aaf4a

• Calling InferenceContext::UnknownShapes twice produces 2 shape handles for which ShapeHandle::SameHandle returns false. Therefore we need to merge the shapes handles in InferenceContext::set_input,…

  Show description for a8d720c
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  a8d720c

  Copy full SHA for a8d720c

Commits on May 2, 2017

• Reset the inputs to ShapeRefiner::RunShapeFn so that it behaves the same every time it's called.

  Show description for 498565a
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  498565a

  Copy full SHA for 498565a

• Properly track the persistent memory usage of lookup tables.

  Show description for 0c372ad
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  0c372ad

  Copy full SHA for 0c372ad

• Added a unit test to check what happens when 2 shapes with known rank but

  Show description for 8304066
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  8304066

  Copy full SHA for 8304066

• Don't try to refine the shapes for a node if its inference context wasn't

  Show description for e8eafd9
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  e8eafd9

  Copy full SHA for e8eafd9

• Enable grappler to propagate shapes through queues.

  Show description for 0287e87
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  0287e87

  Copy full SHA for 0287e87

Commits on Apr 28, 2017

• Apply the grappler graph optimizations on the Inception model: this improves the performance by almost 20% on my GPU (from 55 images/s to 62 images/s)

  Show description for 05e2aee
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  05e2aee

  Copy full SHA for 05e2aee

Commits on Apr 27, 2017

• AutoReloadVariable and VarHandleOp are variables.

  Show description for 6fede95
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  6fede95

  Copy full SHA for 6fede95

• Fix the shape function of the StackPush op

  Show description for b8e4c44
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  b8e4c44

  Copy full SHA for b8e4c44

Commits on Apr 26, 2017

• GrapplerItem is a struct and not a class

  Show description for 5925811
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  5925811

  Copy full SHA for 5925811

• Improved support for lookup tables. These tables are often initialized from large files stored on disk, and can take tens of minutes to load: increased the initialization timeout to give grappler e…

  Show description for 377eca1
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  377eca1

  Copy full SHA for 377eca1

Commits on Apr 25, 2017

• Added a python API to the meta graph optimizer

  Show description for 58fe576
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  58fe576

  Copy full SHA for 58fe576

Commits on Apr 20, 2017

• Create a per session inter op thread pool instead of relying on a global thread

  Show description for f12c967
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  f12c967

  Copy full SHA for f12c967

Commits on Apr 19, 2017

• Made sure that the cost model corresponding to the initialization is exported

  Show description for cfc2ee9
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  cfc2ee9

  Copy full SHA for cfc2ee9

• Added the ability to cancel a reader_op_kernel

  Show description for 9d4ec29
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  9d4ec29

  Copy full SHA for 9d4ec29

Commits on Apr 17, 2017

• Don't assert on unimplemented layout

  Show description for 4f95b09
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  4f95b09

  Copy full SHA for 4f95b09

• Error out when running out of GPU memory instead of triggering a fatal error.

  Show description for 698a961
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  698a961

  Copy full SHA for 698a961

Commits on Apr 14, 2017

• Added a new IsVariable function to detect op variables

  Show description for 4254fa1
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  4254fa1

  Copy full SHA for 4254fa1

• Only record the cost graph in the queue runner: this ensures that the memory

  Show description for dd3a6d3
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  dd3a6d3

  Copy full SHA for dd3a6d3

Commits on Apr 7, 2017

• Added the ability to retrieve the amount of usable gpu memory

  Show description for 0ca0b63
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  0ca0b63

  Copy full SHA for 0ca0b63

Commits on Apr 6, 2017

• Started a set of utilities to categorize op types

  Show description for 9d57702
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  9d57702

  Copy full SHA for 9d57702

• Added the memory optimizer to the meta optimizer.

  Show description for dc14caf
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  dc14caf

  Copy full SHA for dc14caf

Commits on Apr 5, 2017

• Preserve graph version during optimization

  Show description for ff247e5
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  ff247e5

  Copy full SHA for ff247e5

Commits on Apr 4, 2017

• Added a memory optimizer to grappler.

  Show description for 98f6dbe
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  98f6dbe

  Copy full SHA for 98f6dbe

Commits on Apr 3, 2017

• Simplified the testImportIntoNamescope test

  Show description for 0864bed
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  0864bed

  Copy full SHA for 0864bed

• Enabled the generation of big kernels by libxsmm (#8929)

  

  benoitsteiner

  authored and

  gunan

  committed
  7c0e6af

  Copy full SHA for 7c0e6af

Commits on Apr 1, 2017

• Don't prune nodes that are driven by control dependencies to avoid potentially

  Show description for 512c21c
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  512c21c

  Copy full SHA for 512c21c

Commits on Mar 30, 2017

• Upgraded to libxsmm 1.8

  

  benoitsteiner

  authored and

  gunan

  committed
  b5749aa

  Copy full SHA for b5749aa

• FIxed merge issues

  

  benoitsteiner

  authored and

  gunan

  committed
  bbdd4a7

  Copy full SHA for bbdd4a7

• Upgraded to the latest version of libxsmm to fix build errors

  

  benoitsteiner

  authored and

  gunan

  committed
  8ea7c8f

  Copy full SHA for 8ea7c8f

• Unregister the CUPTI callbacks if needed when the gpu tracer is deleted to prevent them from accessing freed memory.

  Show description for 1f6c19e
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  1f6c19e

  Copy full SHA for 1f6c19e

Commits on Mar 28, 2017

• Avoid pruning fetch nodes

  Show description for 164b6a8
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  164b6a8

  Copy full SHA for 164b6a8

• Turned the meta optimizer into a regular optimizer

  Show description for 151d367
  

  benoitsteiner

  authored and

  tensorflower-gardener

  committed
  151d367

  Copy full SHA for 151d367

Pagination

Previous

Next