Commit message (Collapse) | Author | Age | |
---|---|---|---|
* | [Grappler] Add RemoveStackStridedSliceSameAxis optimizer. | Eugene Brevdo | 2018-10-10 |
| | | | | | | | | | | | | | | | | // Replace operations of the form: // x = stack((a_0, a_1, ..., a_{n-1}), axis=k)[:,...,i,...] // with // a_i // when the strided slice index `i` is applied in the k'th axis. // // Similarly, replace operations of the form: // x = stack((a_0, a_1, ..., a_{n-1}), axis=k)[:,...,i:i+1,...] // with // expand_dims(a_i, axis=k) // PiperOrigin-RevId: 216535346 | ||
* | Automated rollback of commit 950cf87104bfee28e2165fe368f66337b8a1336d | A. Unique TensorFlower | 2018-10-10 |
| | | | | PiperOrigin-RevId: 216500702 | ||
* | Add support for modeling fast memory close to the processor/gpu | A. Unique TensorFlower | 2018-10-09 |
| | | | | PiperOrigin-RevId: 216453979 | ||
* | [tf.data vectorization] Add vectorizer for `Add` op | Rachel Lim | 2018-10-09 |
| | | | | PiperOrigin-RevId: 216424512 | ||
* | [tf.data] NUMA-aware MapAndBatch dataset. | Brennan Saeta | 2018-10-09 |
| | | | | PiperOrigin-RevId: 216395709 | ||
* | [tf.data vectorization] Handle captured inputs in MapVectorization optimization | Rachel Lim | 2018-10-09 |
| | | | | PiperOrigin-RevId: 216381943 | ||
* | Change LOG(WARNING) to VLOG(1) in utils | Peter Ma | 2018-10-09 |
| | | | | PiperOrigin-RevId: 216369081 | ||
* | Automated rollback of commit 5f308cb408eb46ec9af0546be6b9ae1d5166b185 | A. Unique TensorFlower | 2018-10-08 |
| | | | | PiperOrigin-RevId: 216309111 | ||
* | Refactor CalculateOutputSize() from VirtualScheduler protected member ↵ | Peter Ma | 2018-10-08 |
| | | | | | | function to utils; Refactor EstimateSize() from memory_optimizer.cc to utils; some small changes for readability improvement PiperOrigin-RevId: 216307257 | ||
* | Automated rollback of commit 07df147ab20c4a5329148e5fb5f7f6b187cb73a4 | Reed Wanderman-Milne | 2018-10-08 |
| | | | | PiperOrigin-RevId: 216299809 | ||
* | Consolidate device parameter arguments into a shared DeviceInfo struct | A. Unique TensorFlower | 2018-10-08 |
| | | | | PiperOrigin-RevId: 216280197 | ||
* | Add timeout mechanism to Grappler meta optimizer. This is only a best-effort ↵ | A. Unique TensorFlower | 2018-10-08 |
| | | | | | | mechanism, since the meta optimizer only checks if it has been cancelled before running each sub-optimizer. We can add cancellation to each sub-optimizer if necessary. PiperOrigin-RevId: 216234262 | ||
* | Enable PinToHostOptimizer. | A. Unique TensorFlower | 2018-10-08 |
| | | | | PiperOrigin-RevId: 216201732 | ||
* | Optimize PinToHostOptimizer by adding cache, also add PinToHostOptimizer to ↵ | A. Unique TensorFlower | 2018-10-08 |
| | | | | | | | | | benchmarks. original runtime: 4.83492736816 secs w/ cache runtime: 2.19033999443 secs PiperOrigin-RevId: 216195286 | ||
* | [tf.data vectorization] Feed inputs to vectorizers with notion of stackedness | Rachel Lim | 2018-10-05 |
| | | | | PiperOrigin-RevId: 215989259 | ||
* | Fix bug in Grappler constant folding: The logic detecting full reductions ↵ | A. Unique TensorFlower | 2018-10-04 |
| | | | | | | | | was flawed. Added better test coverage. Also added a extra test for a related symbolic shape inference operation that I first suspected to be broken. PiperOrigin-RevId: 215812753 | ||
* | [tf.data] Add a notion of `captured args` to MapDefun | Rachel Lim | 2018-10-04 |
| | | | | PiperOrigin-RevId: 215788485 | ||
* | Add ability to vectorize nodes that do not derive from function arguments. ↵ | Rachel Lim | 2018-10-04 |
| | | | | | | (This indirectly handles "Const" outputs automagically, since they are always unstacked.) PiperOrigin-RevId: 215749824 | ||
* | PinToHostOptimizer: Refactored code. Update blacklist. Added recursive ↵ | A. Unique TensorFlower | 2018-10-03 |
| | | | | | | lookback for Identity op. This fixes many performance regressions. PiperOrigin-RevId: 215662393 | ||
* | [tf.data] Add utility to deduplicate graph node names (after vectorization) | Rachel Lim | 2018-10-03 |
| | | | | PiperOrigin-RevId: 215595078 | ||
* | Support shape_invariants in while_v2. Note that this arg is temporary and ↵ | Saurabh Saxena | 2018-10-02 |
| | | | | | | | | may be replaced by automatic shape inference in TF 2.0 (or before). Add a output_shapes attr to While op to allow output shapes to be different from the incoming loop_vars. PiperOrigin-RevId: 215446737 | ||
* | Automated rollback of commit cb98ceba9cff8c10ee3c7e89dc8925c88b28118e | A. Unique TensorFlower | 2018-10-01 |
| | | | | PiperOrigin-RevId: 215254762 | ||
* | Add allowed optimizations to GrapplerItem. | Eugene Zhulenev | 2018-10-01 |
| | | | | | | (1) Skip UnaryOpComposition rewrite if the optimized graph needs to have a gradient registered for all nodes. PiperOrigin-RevId: 215188461 | ||
* | Disable PinToHostOptimizer for NoOp. | A. Unique TensorFlower | 2018-09-29 |
| | | | | PiperOrigin-RevId: 215079134 | ||
* | Bunch of micro move optimizations | Piotr Padlewski | 2018-09-28 |
| | | | | PiperOrigin-RevId: 215018984 | ||
* | Add a rewrite_config option to disable meta_optimizer. | A. Unique TensorFlower | 2018-09-28 |
| | | | | PiperOrigin-RevId: 215014737 | ||
* | [tf.data] Use Graph instead of GraphDef/FunctionDef for vectorization transforms | Rachel Lim | 2018-09-28 |
| | | | | PiperOrigin-RevId: 215011835 | ||
* | Optimize ParseNodeNameAsStringPiece and related functions, since they are ↵ | A. Unique TensorFlower | 2018-09-27 |
| | | | | | | the most costly functions in Grappler. PiperOrigin-RevId: 214853009 | ||
* | Fix support for custom optimizers in explicit schedule | A. Unique TensorFlower | 2018-09-27 |
| | | | | PiperOrigin-RevId: 214794973 | ||
* | Add support for explicit fetches when creating grappler items | A. Unique TensorFlower | 2018-09-27 |
| | | | | PiperOrigin-RevId: 214732243 | ||
* | Misc. micro-optimizations in Grappler optimizers. | A. Unique TensorFlower | 2018-09-26 |
| | | | | | | Make shape inference lazy in optimizers that may not trigger. PiperOrigin-RevId: 214669034 | ||
* | [tf.data] Small utils cleanup to expose generic function | Rachel Lim | 2018-09-26 |
| | | | | PiperOrigin-RevId: 214659488 | ||
* | Hoisting RandomUniform out of functions | Piotr Padlewski | 2018-09-26 |
| | | | | | | | This patch introduces optimization that hoists RandomUniform out of map functions. By doing it, we make function stateless, which is crucial for parallelization and vectorization. PiperOrigin-RevId: 214623178 | ||
* | Bug fix for OpOutputPortIdToArgId, include type_list_attr. | A. Unique TensorFlower | 2018-09-25 |
| | | | | PiperOrigin-RevId: 214505566 | ||
* | Fix a bug in debug_stripper. | Jingyue Wu | 2018-09-25 |
| | | | | | | AsControlDependency accepts a node name not a tensor name. PiperOrigin-RevId: 214451885 | ||
* | Swap Const ops back to GPU greedily. | A. Unique TensorFlower | 2018-09-25 |
| | | | | PiperOrigin-RevId: 214415906 | ||
* | Use less memory by only storing pointers to ops that feed inplace ops. | A. Unique TensorFlower | 2018-09-25 |
| | | | | | | Handle empty strings in NodePositionIfSameNode. PiperOrigin-RevId: 214393567 | ||
* | Disable PinToHostOptimizer for any TPU graphs. | A. Unique TensorFlower | 2018-09-24 |
| | | | | PiperOrigin-RevId: 214338297 | ||
* | Speed up DedupComputation in arithmetic optimizer. | A. Unique TensorFlower | 2018-09-24 |
| | | | | PiperOrigin-RevId: 214338100 | ||
* | Turn on PinToHostOptimizer by default. | A. Unique TensorFlower | 2018-09-24 |
| | | | | PiperOrigin-RevId: 214275960 | ||
* | Fix noop elimination optimization. | Piotr Padlewski | 2018-09-23 |
| | | | | | | | Fix for b/116169724 Only remove noops if they refer to const nodes. PiperOrigin-RevId: 214199200 | ||
* | Add blacklist ops to PinToHostOptimizer. Fix test. | A. Unique TensorFlower | 2018-09-23 |
| | | | | PiperOrigin-RevId: 214195020 | ||
* | Merge pull request #22453 from samikama:custom_optimizer_ordering | TensorFlower Gardener | 2018-09-22 |
|\ | | | | | | | PiperOrigin-RevId: 214132703 | ||
* | | Add PinToHostOptimizer to grappler: force small ops to happen on CPU (instead of | A. Unique TensorFlower | 2018-09-22 |
| | | | | | | | | | | | | GPU). This avoids many unnecessary CPU<->GPU memcpy and syncs. PiperOrigin-RevId: 214108484 | ||
* | | Don't crash on Pack nodes with no axis argument set. | A. Unique TensorFlower | 2018-09-21 |
| | | | | | | | | PiperOrigin-RevId: 214035048 | ||
* | | [tf.data] Add a ConverterRegistry for vectorization converters | Rachel Lim | 2018-09-21 |
| | | | | | | | | PiperOrigin-RevId: 214027910 | ||
| * | Minor style fix. | drpngx | 2018-09-21 |
| | | |||
| * | Add possibility to include default optimizers in custom optimizer list | Sami Kama | 2018-09-21 |
|/ | |||
* | Merge pull request #22402 from kitstar:master | TensorFlower Gardener | 2018-09-20 |
|\ | | | | | | | PiperOrigin-RevId: 213912651 | ||
* | | [tf.data] Some vectorization cleanup | Rachel Lim | 2018-09-20 |
| | | | | | | | | PiperOrigin-RevId: 213886813 |