aboutsummaryrefslogtreecommitdiffhomepage
path: root/tensorflow/compiler/xla/service/hlo_rematerialization.cc
Commit message (Collapse)AuthorAge
* [XLA] Migrate from gtl::FlatSet to absl::flat_hash_setGravatar Benjamin Kramer2018-10-01
| | | | PiperOrigin-RevId: 215324035
* [XLA] Migrate from gtl::FlatMap to absl::flat_hash_mapGravatar Benjamin Kramer2018-10-01
| | | | PiperOrigin-RevId: 215272497
* Add interface for HLO passes which run on HloModuleGroup.Gravatar Mark Heffernan2018-09-19
| | | | | | Derive HloModulePass and HloModuleGroupPass from HloPassInterface which run module-scoped and module-group-scoped respectively. Replace all existing uses of HloPassInterface with HloModulePass because all existing passes are module-scoped. Also rewrite HloPassPipeline to support both module-scoped and module-group-scoped passes. PiperOrigin-RevId: 213629604
* Make scheduling and rematerialization HLO passes.Gravatar Mark Heffernan2018-09-08
| | | | | | | | Now that HloSchedule is a field on the HLO module, scheduling can be done as an HLO pass. Similarly, rematerialization which requires a schedule can also be a pass which just gets the schedule from the module. Also as a clean up, hoist calls to CopyInsertion out of rematerialization. PiperOrigin-RevId: 212119795
* Rollforward of cl/211656888 after fixing failing unit test.Gravatar Mark Heffernan2018-09-05
| | | | | | | | | | | *** Original change description *** Add HloSchedule class representing a sequential order of an HloModule. Currently we represent a sequential schedule of a module using a SequentialHloOrdering::HloModuleSequence which is a type alias of a bare map from HloComputation* to std::vector<HloInstruction*>. This CL replaces this with a proper class which results in better encap... *** PiperOrigin-RevId: 211726890
* BEGIN_PUBLICGravatar Mark Heffernan2018-09-05
| | | | | | Automated rollback of commit 7fa693209fe238478739b3982f652a7e35be91f3 PiperOrigin-RevId: 211681957
* Add HloSchedule class representing a sequential order of an HloModule.Gravatar Mark Heffernan2018-09-05
| | | | | | | | Currently we represent a sequential schedule of a module using a SequentialHloOrdering::HloModuleSequence which is a type alias of a bare map from HloComputation* to std::vector<HloInstruction*>. This CL replaces this with a proper class which results in better encapsulation of code which deals with schedules and better enforcement of invariants. This CL also fixes a corner-case bug in dataflow analysis, where values of instructions which are live out of the computation erroneously did not interfere with the values of instructions scheduled after the root instruction. PiperOrigin-RevId: 211656888
* [XLA] Rename all (Mutable)ArraySlice to absl::Span.Gravatar Tim Shen2018-08-30
| | | | PiperOrigin-RevId: 210998142
* [XLA] Switch to absl::StrFormat.Gravatar Justin Lebar2018-08-27
| | | | | | | | Unlike Printf, StrFormat does not require type-length qualifiers, e.g %z, %ll. Nor does it require that you call c_str() to print strings. So these are fixed up here as well. PiperOrigin-RevId: 210435915
* Run AddSpecialCaseCopies in HloRematerialization.Gravatar Mark Heffernan2018-08-24
| | | | | | RemoveUnnecessaryCopies which runs in rematerialization to take advantage of scheduling can sometimes remove copies which are needed to non-interference reasons. This requires running AddSpecialCaseCopies to add them back in. Furthermore, the schedule needs to be updated to account for the changes to the module, so add an UpdateSchedule function which can patch up a schedule in light a limited set of transformations to the module (addition and deletion of instructions). PiperOrigin-RevId: 210186375
* [XLA] Use "absl::" rather than "::absl".Gravatar Justin Lebar2018-08-23
| | | | | | Also move 'using' statements into namespaces. PiperOrigin-RevId: 210055083
* [XLA] Stop including str_util.h.Gravatar Justin Lebar2018-08-23
| | | | PiperOrigin-RevId: 210049592
* [XLA] Switch from tensorflow::str_util::Join to absl::StrJoin.Gravatar Justin Lebar2018-08-23
| | | | PiperOrigin-RevId: 210018843
* [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
* Merged commit includes the following changes:Gravatar Yifei Feng2018-08-21
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | 209663919 by yifeif<yifeif@google.com>: Internal change. -- 209663914 by amitpatankar<amitpatankar@google.com>: Fix the topk_op_test for numpy>1.15. -- 209660476 by jdduke<jdduke@google.com>: Fix model lifetime for TensorFlow Lite C# bindings Ensure the model's existence for the duration of the interpreter, as per API requirements. -- 209655960 by scottzhu<scottzhu@google.com>: Unify RNN Cell interface between TF and Keras. -- 209655731 by A. Unique TensorFlower<gardener@tensorflow.org>: Added tests for PredictionOps and PartitionExamplesOps -- 209655291 by nolivia<nolivia@google.com>: adding rate class so that we can save global_step/sec using tf.contrib.summary. The function takes the rate in relation to any tensors provided that the numerator and denominator are broadcastable and have dtypes that can be cast to float64 -- 209654655 by kramerb<kramerb@google.com>: [XLA] Switch from tensorflow::gtl::InlinedVector to absl::InlinedVector This one comes with extra goodies like a move constructor. -- 209653851 by A. Unique TensorFlower<gardener@tensorflow.org>: Internal build specification change -- PiperOrigin-RevId: 209663919
* [TF:XLA] Enable better fusion-operand sharing in copy elision.Gravatar A. Unique TensorFlower2018-07-20
| | | | | | | | Make use of the back-end specific fusion-operand sharing functions in copy insertion. This allows potentially better copy insertion/elision. PiperOrigin-RevId: 205481101
* Automated g4 rollback of changelist 203171335Gravatar A. Unique TensorFlower2018-07-03
| | | | PiperOrigin-RevId: 203211687
* [TF:XLA] Run copy insertion with HLO scheduling and rematerialization.Gravatar A. Unique TensorFlower2018-07-03
| | | | | | This allows copy elision to use the same backend-specific HLO dataflow analysis and potentially elide more copies than when running copy insertion/elision on its own. PiperOrigin-RevId: 203171335
* [TF:XLA] Copy elision does not need to know about existing copies.Gravatar A. Unique TensorFlower2018-06-28
| | | | | | | | It already detects layout-changing copies and those are already left unchanged by copy elision. Special case copies are also skipped because they are tagged separately (SetCopyElisionAllowed) PiperOrigin-RevId: 202574858
* Automated g4 rollback of changelist 200309129Gravatar A. Unique TensorFlower2018-06-13
| | | | PiperOrigin-RevId: 200472722
* Automated g4 rollback of changelist 200292049Gravatar A. Unique TensorFlower2018-06-12
| | | | PiperOrigin-RevId: 200309129
* PUBLIC: [TF:XLA] Eliminate more copies after HLO scheduling.Gravatar A. Unique TensorFlower2018-06-12
| | | | | | | | | | | | After scheduling HLOs it is very beneficial to try more copy elision: The sequential ordering from the schedule is stricter than the data-dependency ordering used during copy insertion. Also, allow more operands to share a buffer with their user. In particular, the user has to be element-wise only wrt to the specified operand, and not wrt to all operands. These two changes allow more copies to be eliminated. PiperOrigin-RevId: 200292049
* [TF:XLA] Move methods MinimumMemoryFor... from hlo_scheduling to heap_simulator.Gravatar Dimitris Vardoulakis2018-06-12
| | | | | | | These methods have nothing to do with scheduling. Also, rename methods CreateMemoryMinimizingSequence in hlo_scheduling. PiperOrigin-RevId: 200254100
* Cache the rematerializable status.Gravatar A. Unique TensorFlower2018-06-06
| | | | PiperOrigin-RevId: 199567935
* Move DoesNotUseOperandBuffer and CanShareOperandBufferWithUser fromGravatar Jeremy Lau2018-05-16
| | | | | | liveness_util to methods on TuplePointsToAnalysis and HloDataflowAnalysis. PiperOrigin-RevId: 196903216
* BufferValue is a new base class for LogicalBuffer and HloValue. This makes itGravatar Jeremy Lau2018-05-02
| | | | | | | easier to migrate from TuplePointsToAnalysis/LogicalBuffer to HloDataflowAnalysis/HloValue. No functional changes. PiperOrigin-RevId: 195179676
* [XLA] Make HLO memory schedulers pluggable. Introduce a typedef ↵Gravatar Peter Hawkins2018-03-20
| | | | | | MemorySchedulerAlgorithm which is a function instead of an enum to allow experimentation with non-standard schedulers. Refactoring only; no functional changes to the scheduling itself. PiperOrigin-RevId: 189830685
* Automated g4 rollback of changelist 185623948Gravatar A. Unique TensorFlower2018-02-16
| | | | PiperOrigin-RevId: 186038783
* Automated g4 rollback of changelist 185598764Gravatar A. Unique TensorFlower2018-02-13
| | | | PiperOrigin-RevId: 185623948
* Wire in support for XLA kConditional instruction.Gravatar A. Unique TensorFlower2018-02-13
| | | | PiperOrigin-RevId: 185598764
* [XLA] Add a flag to control the HLO scheduling algorithm choice.Gravatar Chris Leary2017-12-15
| | | | | | | List scheduling is more easily rematerialized sometimes, this gives the ability to force list scheduling via the API. PiperOrigin-RevId: 179246142
* [XLA] Humanize some print-outs.Gravatar A. Unique TensorFlower2017-12-05
| | | | PiperOrigin-RevId: 178040190
* Check per HLO instruction only at vlog=1 in non-opt build.Gravatar Jacques Pienaar2017-11-28
| | | | PiperOrigin-RevId: 177229069
* [XLA] Rematerialization and fusion need to consider nested computations to ↵Gravatar Blake Hechtman2017-11-16
| | | | | | | | determine if rematerializablity or Fusabiltiy. PiperOrigin-RevId: 176064783
* Change for asynchronous Send and Recv by splitting Send into {Send, SendDone}Gravatar HyoukJoong Lee2017-11-10
| | | | | | | and Recv into {Recv, RecvDone}. See operation_semantics.md for the updated semantics. PiperOrigin-RevId: 175216012
* [XLA] Make HloModule::computations() return raw pointers.Gravatar Justin Lebar2017-09-29
| | | | | | | | | | | | | | | | | | Like HloComputation::instructions(), HloModule::computations() used to return a list of unique_ptrs. But this is an implementation detail that shouldn't be leaked into the public API. This patch also adds HloModule::MakeNonFusionComputations(), because many of the callers of computations() went on to filter out all the fusion computations. It would be possible to implement MakeNonFusionComputations() "in place" using a filtering iterator, but I don't think it's necessary -- we never have *that* many computations, and since many callers go on to copy the list of non-fusion computations, making it unconditionally a copy is simpler and avoids a footgun. PiperOrigin-RevId: 170529051
* [XLA] Make HloComputation::instructions() return a view of HloInstruction*s.Gravatar Justin Lebar2017-09-28
| | | | | | | | Currently it returns a view of unique_ptr<HloInstruction>s. But the fact that these are unique_ptrs is an implementation detail, and it's ugly to leak it everywhere. PiperOrigin-RevId: 170445375
* Remove overly-conservative CHECK in rematerialization. The CHECK verifiedGravatar Mark Heffernan2017-09-13
| | | | | | | that the memory use tracker used in remat matched expectations, but the expectation was not always correct in the case of dead values. PiperOrigin-RevId: 168604525
* [XLA] Allow caller to inspect rematerialization before/after sizes.Gravatar Chris Leary2017-09-05
| | | | PiperOrigin-RevId: 167609460
* [XLA] Sanity check the list of called computations for fusion nodesGravatar David Majnemer2017-08-31
| | | | | | | called_compuatations for a fusion node should only include the fusion computation that it calls. PiperOrigin-RevId: 167149669
* Reduce XLA compile time by ~7% for a convolutional image model:Gravatar A. Unique TensorFlower2017-08-18
| | | | | | | | | | | | | | | | | | | | | | * Added CompactPointerSet<T>, which is optimized for set size <= 1. * Changed expensive CHECKs to DCHECKS in buffer_assignment.cc * Reserve space in DFS state array before starting DFS. * Use unsigned arithmetic in DFS state maintenance. * HloInstruction: - Moved frequently used fields to start for better cache locality. - Use InlinedVector instead of vector for operand array. - Use InlinedVector instead of vector for DFS stack. * Pre-compute "is array" and "is tuple" for LogicalBuffer. * PointsToSet: - Combine two ShapeTrees into one. - Use CompactPointerSet instead of std::set to hold sources. - Use CompactPointerSet instead of std::set to hold flattened buffers. * ShapeTree: use unique_ptr instead of optional for shape storage (reduces size and destruction overhead). * Add proper const qualifiers to some FlatSet iterator methods. Co-author=jeff PiperOrigin-RevId: 165759117
* [Rematerialization] Reconsider to remat operations with control dependenciesGravatar A. Unique TensorFlower2017-08-17
| | | | | | We added a conservartive logic to not rematerialize operations with control dependencies since the rematerialized operations could result in undesired ordering. However, we now realize that when we remat an operation, we also copy the dependencies of them, which guarantees the rematerialized operation has the same constraint as the original operation. PiperOrigin-RevId: 165654629
* Improve performance of compilation by ~8% by speeding up theGravatar A. Unique TensorFlower2017-08-01
| | | | | | | | | | | | | hlo rematerialization pass. Changes: . Wrap each HloInstruction* inside an Item structure that keeps associated data. This allows us to get rid of a bunch of hash tables indexed by HloInstruction*. * Switch to an intrusive linked list (instead of std::list) so that we can avoid a hash table that maps to std::list::iterator. * Use inlined vector in a few places. PiperOrigin-RevId: 163848365
* Fix the constructor parameter in InstructionList to be pass-by-reference.Gravatar A. Unique TensorFlower2017-07-28
| | | | PiperOrigin-RevId: 163535344
* [XLA] Finish normalizing fusion computations into standard computationsGravatar A. Unique TensorFlower2017-07-26
| | | | PiperOrigin-RevId: 163210327
* [XLA:HLO] Move sequence functions from hlo_ordering.h to hlo_scheduling.h.Gravatar A. Unique TensorFlower2017-06-21
| | | | | | | | | | | | | | | | | | | | | | | | This is required for upcoming changes to convert the sequence creation functions (and HeapSimulator and BufferAssignment) over to using the new Hlo{Dataflow,Alias}Analysis. It's required because otherwise there's a dependency cycle: Hlo{Dataflow,Alias}Analysis depends on HloOrdering CreateMemoryMinimizingSequence will depend on Hlo{Dataflow,Alias}Analysis There's already a cycle here, if both HloOrdering and CreateMemoryMinimizingSequence are in the same file. Also note that: MinimumMemoryForSequence depends on HeapSimulator HeapSimulator will depend on Hlo{Dataflow,Alias}Analysis Hlo{Dataflow,Alias}Analysis depends on HloOrdering Splitting out the sequence functions resolves the cycle. Refactoring only; no functional changes. PiperOrigin-RevId: 159731836
* Adjust rematerialization cost function to only be the inverse of its benefit ↵Gravatar Blake Hechtman2017-06-16
| | | | | | (total_memory/saved_memory). PiperOrigin-RevId: 159290105
* [XLA] Simplify Shape traversal visitors.Gravatar Mark Heffernan2017-06-06
| | | | | | Simplify shape traversal visitors in ShapeUtil and ShapeTree. Add a non-Status form because most uses of the traversal methods do not use it, and remove is_leaf parameter from ShapeTree.ForEach* as it is not frequently used. PiperOrigin-RevId: 158201574
* Improve the error message for live set memory check.Gravatar Jianwei Xie2017-05-29
| | | | PiperOrigin-RevId: 157415647
* Merged commit includes the following changes:Gravatar A. Unique TensorFlower2017-05-10
| | | | | | | | | | | | | | | 155425029 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- 155424167 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- PiperOrigin-RevId: 155425029
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-27 21:39:04 +0000