| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
|
|
|
|
| |
* redundant string conversion
* using decl 'Eq' is unused
* using decl 'HasSubstr' is unused
* redundant StrCat calls
* please use StrAppend instead of StrCat when appending to an existing string (4 times)
* parameters of type 'absl::Span<...>' should be taken by value (23 times)
PiperOrigin-RevId: 212439742
|
|
|
|
| |
PiperOrigin-RevId: 212313258
|
|
|
|
|
|
| |
The "Proto" suffix adds little clarity but makes a long type name even longer.
PiperOrigin-RevId: 211693871
|
|
|
|
|
|
|
|
| |
HLO transformations would forget to propagate the feature depth attribute.
Making these attributes mandatory, while slightly less convenient for tests,
makes HLO transformations more robust.
PiperOrigin-RevId: 211490160
|
|
|
|
| |
PiperOrigin-RevId: 210998142
|
|
|
|
|
|
| |
Same for WrapUnique.
PiperOrigin-RevId: 209531124
|
|
|
|
| |
PiperOrigin-RevId: 206111380
|
|
|
|
|
|
|
|
|
| |
Currently Literal classes sits in literal_util.{h,cc} instead of literal.{h,cc}.
It also contains helper functions that are better fit to be their own separate
class/namespace. This change starts this process by moving most static factory
methods to LiteralUtil namespace.
PiperOrigin-RevId: 203217065
|
|
|
|
|
|
|
| |
- Require a module config when creating an HloModule.
- All tests using HloTestBase create a module using CreateNewModule.
PiperOrigin-RevId: 194684585
|
|
|
|
| |
PiperOrigin-RevId: 193998684
|
|
|
|
|
|
|
|
| |
- Add test in select and scatter to test we break ties by lexicographical order.
- Fix reference util to break ties by lexicographical order.
- Fix evaluator to break ties by lexicographical order.
RELNOTES: select and scatter in reference util and evaluator now uses lexicographical order to break ties.
PiperOrigin-RevId: 189067946
|
|
|
|
|
|
| |
allow arbitary padding.
PiperOrigin-RevId: 188966155
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Extend the stream interface ThenBlasGemmWithAlgorithm to support F16 matrix
multiplication with computation type FP32.
Extend the stream executor interface DoBlasGemmWithAlgorithm to support F16
GEMM with computation type FP32.
Extend the CPU IR emitter to handle F16 Dot instruction, and add F16 matrix
multiplication implementation to the CPU runtime.
Extend the GPU backend to handle FP16 GEMM Thunk.
Replicate the existing matrix multiplication test cases in
matrix_ops_simple_test and dot_operation_test for FP16.
RELNOTES:
PiperOrigin-RevId: 187369731
|
|
|
|
|
|
|
| |
* Adds rank-3 test harness for ReduceWindow, along with some test cases.
* Minor style correction.
PiperOrigin-RevId: 182256143
|
|
|
|
|
|
| |
Also fix out-of-bounds read in ReferenceUtil::Reduce4DTo1D.
PiperOrigin-RevId: 179888724
|
|
|
|
|
|
|
|
|
| |
xla/tests:map_map_test which tests this change.
Additionally:
- templatize Evaluate* methods to specialize on both std::unique_ptr<Literal> and const Literal* type of input literal arguments.
- add ResetVisitState to DfsHloVisitor such that a visitor instance can traverse the same HLO graph more than once.
PiperOrigin-RevId: 179263540
|
|
|
|
|
|
| |
padding. Modifies reference util reduce-window 1D implementation to accept general padding.
PiperOrigin-RevId: 177243527
|
|
|
|
|
|
| |
This lets us reason about input spatial dimensions as distinct from output spatial dimensions. By doing this, it opens up more opportunities for assigning more interesting, different, layouts for the activations and the output.
PiperOrigin-RevId: 177117140
|
|
|
|
| |
PiperOrigin-RevId: 177109696
|
|
|
|
|
|
| |
the parent field is currently null.
PiperOrigin-RevId: 163983198
|
|
|
|
|
|
|
| |
Also Speed up HloEvaluator's HandleConvolution in non-opt build, by moving calls
to HloInstruction::shape() out of the inner loop.
PiperOrigin-RevId: 163416183
|
|
|
|
| |
PiperOrigin-RevId: 163170549
|
|
|
|
| |
PiperOrigin-RevId: 163164566
|
|
|
|
| |
PiperOrigin-RevId: 162453537
|
|
|
|
|
|
| |
By marking std::function parameters as const& where applicable; this also makes it more consistent with the other utils that use std::function parameters.
PiperOrigin-RevId: 162365501
|
|
|
|
|
|
| |
Add an R1 reduction to the reference util.
PiperOrigin-RevId: 160659719
|
|
|
|
|
|
| |
RELNOTES: n/a
PiperOrigin-RevId: 160307959
|
|
|
|
|
|
| |
Fix broken XLA build.
PiperOrigin-RevId: 160284588
|
|
|
|
|
|
|
|
|
| |
not all inputs are 0, leading to an all-0 output, which masks most possible bugs.
We do not check-fail on 0-sized dimensions as tests for these special cases
exist.
PiperOrigin-RevId: 160274593
|
|
|
|
|
|
| |
from function parameters when moving reduces the amount of copying.
PiperOrigin-RevId: 157867333
|
|
|
|
| |
PiperOrigin-RevId: 156919832
|
|
|
|
| |
Change: 154603866
|
|
|
|
|
|
| |
- CommaSeparatedString and VectorString added to xla_util.h
- ReferenceUtil can now do more general Pad ops.
Change: 153782516
|
|
|
|
| |
Change: 147072336
|
|
|
|
| |
Change: 146889081
|
|
|
|
|
|
|
|
| |
ReferenceUtil::ConvArray4DGeneralDimensionsDilated.
It's trivial to avoid half of the idivs in this function; they're just
loop induction variables.
Change: 146809277
|
|
|
|
|
|
|
|
|
| |
Avoid 64-bit divides where possible. Per Agner Fog's manuals, 64-bit
idiv is at least 4x slower than 32-bit idiv. perf confirms that the
idivs are on the critical path.
http://www.agner.org/optimize/instruction_tables.pdf
Change: 146806085
|
|
|
|
|
|
|
|
|
|
|
| |
Define semantics of negative padding in the Pad instruction to be identical to padding inside of convolution operation ConvWithGeneralPadding. Also make negative padding work in the backends.
Specific changes:
(1) Add transformation to algebraic simplifier which replaces negative padding with slices.
(2) fix ReferenceUtil to properly handle negative padding and interior padding.
(3) Add negative padding explanation to operation semantics g3doc.
(4) Extend LayoutsInShapesEqual and CopyLayoutBetweenShapes to work with shapes which are not exactly compatible but have the same rank and tuple structure.
Change: 145355127
|
|
|
|
| |
Change: 144951784
|
|
XLA is a compiler-based linear algebra execution engine that targets CPUs, GPUs and custom accelerators.
XLA is still experimental; we are releasing it early to get the community involved.
Change: 143990941
|