aboutsummaryrefslogtreecommitdiffhomepage
path: root/tensorflow/compiler/xla/service/algebraic_simplifier.cc
Commit message (Collapse)AuthorAge
* [XLA] Make kConvolution, kDot HLO attributes mandatoryGravatar David Majnemer2018-09-04
| | | | | | | | 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
* [XLA] Simplify effective scalar iota to zeroGravatar David Majnemer2018-09-02
| | | | | | Happened to observe this come up in a linear algebra workload. PiperOrigin-RevId: 211290278
* Change headers to directly include absl::Span, and clean up the buildGravatar Tim Shen2018-08-30
| | | | | | dependencies as well. PiperOrigin-RevId: 211038094
* [XLA] Merge kPad into kConvolution's window where possible.Gravatar Justin Lebar2018-08-30
| | | | | | | This allows us to use e.g. cudnn's padding, instead of materializing a kPad instruction. PiperOrigin-RevId: 211028379
* [XLA] Rename all (Mutable)ArraySlice to absl::Span.Gravatar Tim Shen2018-08-30
| | | | PiperOrigin-RevId: 210998142
* [XLA] Switch to using kIota from TFGravatar David Majnemer2018-08-29
| | | | | | | | | | We were using a broadcast of a constant instead of the kIota HLO. To make switching to kIota practical, we need to do a few things first: - Don't constant fold kIota. - Don't hoist kIota from loops without good cause. PiperOrigin-RevId: 210825834
* [XLA] Add support for algebraic simplifications involving kIotaGravatar David Majnemer2018-08-28
| | | | PiperOrigin-RevId: 210634966
* [XLA] Use absl string types and functions instead of the TF versions.Gravatar Justin Lebar2018-08-23
| | | | | | | Unfortunately this has to be one big patch, because e.g. absl::StrCat doesn't accept a TF StringPiece, but as soon as we switch to absl::string_view, we have to switch away from all of the TF functions. PiperOrigin-RevId: 209957896
* [XLA] Expose a way to control dot/conv precisionGravatar David Majnemer2018-08-22
| | | | | | | | | | This adds a field to the proto so that we may serialize it. On TPUs, we can simulate higher precision by splitting a float32 number into several bfloat16 numbers such that their sum closely approximates the original number. A tensor contraction operation like convolution or a dot product can be computed by forming several partial products which approximate the correct answer to a closer margin. PiperOrigin-RevId: 209720948
* [XLA] gtl::optional->absl::optionalGravatar Yunxing Dai2018-08-21
| | | | PiperOrigin-RevId: 209686671
* [XLA] Use absl::make_unique instead of xla::MakeUnique.Gravatar Justin Lebar2018-08-20
| | | | | | Same for WrapUnique. PiperOrigin-RevId: 209531124
* [XLA] Switch to absl versions of the c_foo functions.Gravatar Justin Lebar2018-08-20
| | | | PiperOrigin-RevId: 209502513
* Automated rollback of commit 4a41f50648929197954d892559587cb76458d306Gravatar A. Unique TensorFlower2018-08-17
| | | | PiperOrigin-RevId: 209248552
* [XLA] Switch to absl versions of the c_foo functions.Gravatar Justin Lebar2018-08-17
| | | | PiperOrigin-RevId: 209247783
* [XLA] Optimize Transpose and Reshape of single-use Rng.Gravatar Blake Hechtman2018-08-14
| | | | PiperOrigin-RevId: 208746565
* [XLA] The algebraic simplifier can't handle variadic reduce at this point.Gravatar Michael Kuperstein2018-08-09
| | | | PiperOrigin-RevId: 208092411
* Simplify sort in Algebraic simplifier if no sorting is needed.Gravatar Adrian Kuegel2018-07-31
| | | | PiperOrigin-RevId: 206782323
* [XLA] Simplify trivial dynamic-slices regardless of the start indices.Gravatar Michael Kuperstein2018-07-25
| | | | | | With clamping semantics, if the shapes match, the indices are always clamped to zero. PiperOrigin-RevId: 206013929
* [XLA] Simplify slice(slice()) to a single slice and simplify Reduce(Concat())Gravatar Blake Hechtman2018-07-23
| | | | | | to Accumulate(Reudce(),Reduce(),...Reduce()) PiperOrigin-RevId: 205740411
* [XLA] Simplify A*0 and 0*A to 0 for integral types.Gravatar Peter Hawkins2018-07-16
| | | | PiperOrigin-RevId: 204797049
* [TF:XLA] Split literal_util into {literal, literal_util}.Gravatar Kay Zhu2018-07-03
| | | | | | | | | 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
* [XLA] Add key-value version of Sort HLO.Gravatar Michael Kuperstein2018-06-29
| | | | | | This is only currently implemented in the evaluator backend, and even that implementation is partial - the key and value type must match. PiperOrigin-RevId: 202673122
* [XLA] Add test case for TOKEN constants. Make the test case pass.Gravatar Peter Hawkins2018-06-27
| | | | PiperOrigin-RevId: 202401460
* Automated g4 rollback of changelist 201765455Gravatar Blake Hechtman2018-06-26
| | | | PiperOrigin-RevId: 202152026
* Automated g4 rollback of changelist 201765455Gravatar A. Unique TensorFlower2018-06-22
| | | | PiperOrigin-RevId: 201783076
* [XLA] Disallow implicit scalar broadcast.Gravatar Blake Hechtman2018-06-22
| | | | PiperOrigin-RevId: 201765455
* [XLA] Teach algebraic simplifier to convert a copy instruction to a bitcastGravatar Bixia Zheng2018-06-22
| | | | | | | | | | | | instruction. Extend ReshapeIsBitcast to handle copy instructions. This allows algebraic simplifier, for example, to replace a copy from shape f16[1,1,128,128]{3,2,1,0} to shape f16[1,1,128,128]{1,0,3,2} to bitcast for the CPU and GPU backends. Add a test case. PiperOrigin-RevId: 201699400
* Fix assumptions that a Shape must be a tuple or an array.Gravatar Mark Heffernan2018-06-13
| | | | | | | | | | | | A TOKEN primitive type was added with cl/199215963 and XLA also has an OPAQUE primitive type. However, in many places in XLA we assume either a tuple or array. This CL fixes many of those instances, but some may remain. Identified instances were discovered by searching for IsTuple or IsArray so the set of fixes is not exhaustive. Also opportunistically addressed a couple potential points of confusion in the ShapeUtil interface: (1) Rename ShapeUtil::HasZeroElements to ShapeUtil::IsZeroElementArray. The point of confusion here is that tuples can also have zero elements and HasZeroElements would check fail on tuple shapes. Method no longer check fails if the given shape is not an array. (2) ShapeUtil::IsNil now returns true only for empty tuples. Previously it also returned true for zero-element array types which was confusing because ShapeUtil::MakeNil creates an empty tuple. PiperOrigin-RevId: 200452672
* [XLA] Fold consecutive reduces.Gravatar Blake Hechtman2018-06-11
| | | | PiperOrigin-RevId: 200086761
* [XLA] Add comments for the Reduce->Reshape simplifier pass.Gravatar Kay Zhu2018-06-01
| | | | | | Also forcing reduction order for init to be on lhs for ReduceWindow->Map pass. PiperOrigin-RevId: 198953817
* [XLA] Fix batchnorm rewriter to not use implicit broadcasts. AlgebraicGravatar Blake Hechtman2018-05-31
| | | | | | simplifier reshape change is now covered by ReshapeMover. PiperOrigin-RevId: 198802494
* [XLA] Remove maps with a single instructionGravatar David Majnemer2018-05-24
| | | | | | | These maps aren't really pulling their weight, fold them to the instruction that they compute. PiperOrigin-RevId: 197967117
* [XLA] Cache computations when creating reduces in algebraic simplifier or ↵Gravatar Benjamin Kramer2018-05-15
| | | | | | | | | | | | | | batchnorm expander Otherwise we create a lot of identical small computations. This shouldn't have an effect except for cluttering the HLO, but turns out HloCSE doesn't look inside of the computation of reduces, effectively never eliminating reduces that were produced via this code path. While there clean up some YAGNI, this only worked for F32 anyways, so just hardcode it. PiperOrigin-RevId: 196689316
* [XLA] First step in adding Literal slice classes, to improve interface safetyGravatar Kay Zhu2018-05-09
| | | | | | | | | | | | | | | | | | | | | | | | | | | and prepare for enabling more efficient interfacing from Tensor to Literal to reduce host to device latency. More specically: * Introducing a new LiteralBase abstract base class that contains all immutable methods of from the old Literal class. * Introducing a subclass LiteralSlice to replace original LiteralView class. LiteralSlice class is read-only and does not own Shape nor any buffer through the Pieces. Change a number of callers to use LiteralSlice directly. * Change Literal class to explicitly own the underlying Shape as well as owning the underlying buffer via Piece. * Conversion from Literal to LiteralSlice is now done via an implicit conversion constructor instead of inheritance. * Decouple ShapeTree from Literal classes. * Use copy-and-swap for assignment constructors. * Other minor cleanups. PiperOrigin-RevId: 196016576
* Re-land: Optimize dot(DynamicSlice(ConstA), ConstantB) by memoizing ↵Gravatar Alina Sbirlea2018-05-08
| | | | | | | | | | | | | | dot(ConstA, ConstB) Make transformation when ConstA and ConstB are 2D, and DynamicSlice is slicing a full row, column respectively. Handle: dot(DynamicSlice(Index, ConstA), ConstB) => DynamicSlice(Index, dot*(ConstA, ConstB)); and dot(ConstA, DynamicSlice(Index, ConstB)) => DynamicSlice(Index, dot*(ConstA, ConstB)); Reason to roll forward: Previous issue of getting out of memory errors when generating LLVM constants was resolved by CSE-ing constants before allocation. PiperOrigin-RevId: 195853680
* Automated g4 rollback of changelist 192180356Gravatar Dimitris Vardoulakis2018-04-18
| | | | PiperOrigin-RevId: 193427566
* [XLA] Document and enforce reduction order of init valueGravatar Nick Desaulniers2018-04-16
| | | | | | All existing backends apply the init_value on the lhs, except for the evaluator. This causes problems for reductions which apply an identity function to a reduce or reduce window. PiperOrigin-RevId: 193091323
* [XLA] Use pattern matcher in algebraic simplifierGravatar A. Unique TensorFlower2018-04-13
| | | | PiperOrigin-RevId: 192862841
* Add opcode for new instruction that broadcasts degenerate dimensions.Gravatar Dimitris Vardoulakis2018-04-09
| | | | | | | Implicit broadcasts can be translated to the new instruction instead of a reshape-and-broadcast. Follow-up CLs will add support in UserComputation and the various backends. PiperOrigin-RevId: 192180356
* [XLA] Fold reduce-window(convert(pad(X))) into reduce-window(convert(X))Gravatar David Majnemer2018-03-27
| | | | | | | | | | | ReduceWindow operations are done in higher precision to avoid accumulation error. Convert operations can find their way between a ReduceWindow and a Pad which can prevent a Pad from combining with a ReduceWindow. Fix this by looking past the Convert while also checking that the Convert'd Pad's init value is identical to the reduce-window value. PiperOrigin-RevId: 190686175
* Misc typo fixes in the XLA sources and docs.Gravatar Dimitris Vardoulakis2018-03-25
| | | | PiperOrigin-RevId: 190322644
* Automated g4 rollback of changelist 190139303Gravatar Dimitris Vardoulakis2018-03-25
| | | | PiperOrigin-RevId: 190242291
* Automated g4 rollback of changelist 190127730Gravatar Mark Heffernan2018-03-22
| | | | PiperOrigin-RevId: 190139303
* Merge consecutive broadcast HLO instructions.Gravatar Dimitris Vardoulakis2018-03-22
| | | | | | As an optimization, replace consecutive broadcast instructions with a single equivalent broadcast in algebraic simplification. PiperOrigin-RevId: 190127730
* Rename CreateXyzHlo utilities to MakeXyzHlo as discussed on cr/188968478; NFCGravatar Sanjoy Das2018-03-15
| | | | | | | The rationale here is that MakeXyzHlo is less likely to be confused with HloInstruction::CreateXyz and we already have a convention of using a "Make" prefix for ergonomic factory functions. PiperOrigin-RevId: 189259036
* Add some simple HLO creation utilities to auto-infer result shapesGravatar Sanjoy Das2018-03-08
| | | | | | I need something like this for my Gather HLO->HLO lowering pass. PiperOrigin-RevId: 188365102
* [XLA] Optimize away DynamicUpdateSlice with update parameter with a dimensionGravatar Nick Desaulniers2018-03-01
| | | | | | | | of zero. A zero sized update has no effect. PiperOrigin-RevId: 187510099
* [XLA] Remove bitcast-converts between same shape.Gravatar Jacques Pienaar2018-02-25
| | | | PiperOrigin-RevId: 186929931
* [XLA] Convert large constants of the same value into broadcasts.Gravatar Blake Hechtman2018-02-21
| | | | PiperOrigin-RevId: 186539902
* Automated g4 rollback of changelist 185891869Gravatar Alina Sbirlea2018-02-15
| | | | PiperOrigin-RevId: 185944719
generated by cgit v1.2.3 (git 2.39.1) at 2024-09-29 06:16:59 +0000