Commits

Branch selector

 master

User selector

blakehechtman

Datepicker

All time

Commit History

Commits on May 23, 2022

• [XLA] Dump and parse the entry_computation_layout in string format, proto format already serialized this information

  Show description for 21754c7
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  21754c7

  Copy full SHA for 21754c7

• [XLA] Do not check fail in proto copy from if the backend config proto and desired proto type do not match.

  Show description for 5801406
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  5801406

  Copy full SHA for 5801406

Commits on May 20, 2022

• [XLA] Dump and parse the entry_computation_layout in string format, proto format already serialized this information

  Show description for 9c67f78
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  9c67f78

  Copy full SHA for 9c67f78

Commits on May 19, 2022

• [XLA] Dump and parse the entry_computation_layout in string format, proto format already serialized this information

  Show description for 52970bd
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  52970bd

  Copy full SHA for 52970bd

Commits on May 18, 2022

• Enable SMPD with Outside Compilation for the old bridge.

  Show description for 6fc1e4d
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  6fc1e4d

  Copy full SHA for 6fc1e4d

Commits on May 10, 2022

• [XLA:SHAPE] Add a helper method for accessing minor sizes from a shape.

  Show description for 5270712
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  5270712

  Copy full SHA for 5270712

Commits on Apr 26, 2022

• [XLA:ALGEBRAIC_SIMPLIFER] Ensure that Optimization Barriers do not prevent dead

  Show description for cebcbb1
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  cebcbb1

  Copy full SHA for cebcbb1

Commits on Apr 20, 2022

• [XLA:ALGEBRAIC_SIMPLIFER] Ensure that Optimization Barriers do not prevent dead

  Show description for 59b5c7f
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  59b5c7f

  Copy full SHA for 59b5c7f

Commits on Apr 7, 2022

• [XLA] We should not remat layout constrained collectives

  Show description for 9e7e66c
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  9e7e66c

  Copy full SHA for 9e7e66c

Commits on Apr 6, 2022

• [XLA:SHAPE_UTIL] Return nullopt instead of a check failure if the input dimensions are not sorted.

  Show description for 0d5668c
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  0d5668c

  Copy full SHA for 0d5668c

Commits on Apr 2, 2022

• [XLA] Allow for indirect CSE of iota and constants before layout assignment and direct CSE of constants only after layout assignment.

  Show description for 3ff2c4e
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  3ff2c4e

  Copy full SHA for 3ff2c4e

Commits on Mar 18, 2022

• [XLA] Reduce duplication in instructions with dimensions field.

  Show description for 4157def
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4157def

  Copy full SHA for 4157def

Commits on Mar 11, 2022

• [XLA] Allow can_share_buffer to alias non-equal shapes.

  Show description for 55c8152
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  55c8152

  Copy full SHA for 55c8152

• [XLA:DATAFLOW_ANALYSIS] Allow backends to match unequal shapes for aliasing.

  Show description for 3b2ce05
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  3b2ce05

  Copy full SHA for 3b2ce05

Commits on Mar 8, 2022

• [XLA] Add IsDead method to hlo_instruction and use it. NFCI

  Show description for 8f0de74
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  8f0de74

  Copy full SHA for 8f0de74

Commits on Mar 1, 2022

• [XLA:LAYOUT_ASSIGNMENT] Remove additional references to send/recv channel

  Show description for 2735a0b
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  2735a0b

  Copy full SHA for 2735a0b

Commits on Feb 28, 2022

• [XLA:ALGEBRAIC_SIMPLIFIER] Handle Base Dilation correctly when folding a pad into a reduce window. The padding on the operand is dilated.

  Show description for 4926e28
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4926e28

  Copy full SHA for 4926e28

Commits on Feb 26, 2022

• [XLA:DYNAMIC_PADDER] Do not pad init values into non-reduced dimensions

  Show description for 9e51184
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  9e51184

  Copy full SHA for 9e51184

Commits on Feb 24, 2022

• [XLA:LAYOUT_ASSIGNMENT] Let backends constraint send/recv layouts.

  Show description for 025ee7b
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  025ee7b

  Copy full SHA for 025ee7b

Commits on Feb 17, 2022

• [XLA:ALGEBRAIC_SIMPLIFIER] Combine nested dynamic slices and nested dynamic update slices.

  Show description for e46322b
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  e46322b

  Copy full SHA for e46322b

• [XLA] Do not add a channel id to rematerialized instructions that did not have

  Show description for b32a239
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  b32a239

  Copy full SHA for b32a239

• [XLA:ALGEBRAIC_SIMPLIFIER] Merge nested dynamic slice and dynamic udpate slices into a single version of the operation.

  Show description for 4b0ff84
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4b0ff84

  Copy full SHA for 4b0ff84

Commits on Feb 14, 2022

• [XLA] Default Reshape and transpose to propagate operand metadata.

  Show description for eca37e1
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  eca37e1

  Copy full SHA for eca37e1

• [XLA] Add a more convinent way to add a derived instruction. NFCI from a

  Show description for c7074a7
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  c7074a7

  Copy full SHA for c7074a7

Commits on Feb 12, 2022

• [XLA] Expand Gathers into broadcasts if possible.

  Show description for e59dc0e
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  e59dc0e

  Copy full SHA for e59dc0e

Commits on Feb 4, 2022

• [XLA:ALL_REDUCE_SIMPLIFIER] Remove trivial AllGather and ReduceScatter

  Show description for 5717cb6
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  5717cb6

  Copy full SHA for 5717cb6

Commits on Feb 3, 2022

• [XLA] Do not re-invent common factors inside of align layouts, just use it

  Show description for bf8c4b7
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  bf8c4b7

  Copy full SHA for bf8c4b7

Commits on Jan 31, 2022

• [XLA:SPMD] Generate AllGathers without degenerate major dimensions.

  Show description for 96b8219
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  96b8219

  Copy full SHA for 96b8219

Commits on Jan 28, 2022

• [XLA] Don't cannonicalize vector or matrix all-gathers

  Show description for 13a5094
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  13a5094

  Copy full SHA for 13a5094

Commits on Jan 19, 2022

• [XLA] Extend Converted Dimensions to include partially modified dimensions

  Show description for 98fdeb7
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  98fdeb7

  Copy full SHA for 98fdeb7

Commits on Nov 17, 2021

• [XLA] Avoid excessive collisions in CSE with many scalar constants.

  Show description for 5afa725
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  5afa725

  Copy full SHA for 5afa725

Commits on Nov 12, 2021

• [XLA] add an accessor methods for the mutable operands of an HLO

  Show description for 4061b99
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4061b99

  Copy full SHA for 4061b99

Commits on Nov 1, 2021

• [XLA] Add an option to set the operand precision on any HLO and not just dot/conv

  Show description for 23984dc
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  23984dc

  Copy full SHA for 23984dc

Commits on Oct 14, 2021

• [XLA:ALGEBRAIC_SIMPLIFIER] cleanup the IsConvertPairNoOp to use the primitive util formulation.

  Show description for 08fcfae
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  08fcfae

  Copy full SHA for 08fcfae

Commits on Sep 26, 2021

• [XLA] Correctly handle the types of zeros in dot and convolution since they natively support mixed precision.

  Show description for 37cacc4
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  37cacc4

  Copy full SHA for 37cacc4

Pagination

Previous

Next