aboutsummaryrefslogtreecommitdiffhomepage
Commit message (Collapse)AuthorAge
* Clean up some redundant and unused build settings.Gravatar Michael Case2017-11-10
| | | | | | | | --copts are passed to both c++ and c (so is redundent with --cxxopts). Configs passed to "bazel build" are inherited by "bazel run" and "bazel test". Also removed some unused configs. PiperOrigin-RevId: 175326697
* Removed StringPiece::set and StringPiece::clear, as they have no ↵Gravatar A. Unique TensorFlower2017-11-10
| | | | | | | | | | | | | absl::string_view equivalents. This will allow for a more convenient transition to absl::string_view. Calls to set StringPiece::set and StringPiece::clear were replaced with the StringPiece constructor as follows: string_piece_foo.set(data, size) => string_piece_foo = StringPiece(data, size) string_piece_foo.clear() => string_piece_foo = StringPiece() PiperOrigin-RevId: 175326576
* Remove the backward incompatible shape checks.Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175324895
* [CMake] Add missing framework dependency to `tf_c_python_api`.Gravatar Derek Murray2017-11-10
| | | | | | | This was occasionally causing flaky failures due to the module being compiled before the proto_text headers had been generated. PiperOrigin-RevId: 175322168
* [XLA:GPU] Dump PTX when --xla_dump_ir is set.Gravatar Justin Lebar2017-11-10
| | | | | | | | | | I know, I know, PTX isn't really IR. (Or is it?? It's not machine code...) In any case, if you pass this flag and get PTX, you're unlikely to be disappointed. PiperOrigin-RevId: 175320353
* ...comment format only change...Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175319401
* Make Operation.node_def fetch from C API when enabledGravatar Skye Wanderman-Milne2017-11-10
| | | | | | Also fixes previously-introduced memory management bugs in graph_def_versions and op_def. PiperOrigin-RevId: 175314829
* Skip generating input / output properties for _Send and _Recv ops if those opsGravatar A. Unique TensorFlower2017-11-10
| | | | | | are not created from VirtualScheduler. PiperOrigin-RevId: 175314193
* Include memory size info in error message.Gravatar James Qin2017-11-10
| | | | PiperOrigin-RevId: 175311543
* Fixes API difference between eager and graph cond.Gravatar Alexandre Passos2017-11-10
| | | | PiperOrigin-RevId: 175307853
* Internal Change.Gravatar Andrew Selle2017-11-10
| | | | PiperOrigin-RevId: 175307445
* Adds properties to LossFunction to access inputs separated by minibatch.Gravatar A. Unique TensorFlower2017-11-10
| | | | | | | | | Currently, information about separate minibatches registered by `LossFunction`s is private, and only the concatenation of all minibatch inputs is exposed through the `inputs` property. This change adds `input_minibatches` and `num_registered_minibatches` to `LossFunction` to expose this information. PiperOrigin-RevId: 175306297
* Add suffix to newly created Mul op in the optimizer to avoid the name collisionGravatar A. Unique TensorFlower2017-11-10
| | | | | | when two Conv2D objects depend on the same Const. PiperOrigin-RevId: 175305425
* [XLA] Include kConvolution in the dot-add liveness optimizationGravatar David Majnemer2017-11-10
| | | | PiperOrigin-RevId: 175304705
* [TF:XLA] Add helper to append buffers to ShapedBuffer.Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175304150
* Correct an IndexedSlices-related test in replicate_model_fn_test.py.Gravatar Igor Saprykin2017-11-10
| | | | | | I attempted to exercise compute_sum_on_device(IndexedSlices) via the DNNCLassifier test per reviewer's suggestion. This changelist is way to do it correctly. I verified that it indeed triggers the required codepath by adding logging and removing compute_sum_on_device(IndexedSlices) support. PiperOrigin-RevId: 175303333
* Add proper element_shape values to tf.nn.dynamic_rnn TensorArrays.Gravatar Eugene Brevdo2017-11-10
| | | | PiperOrigin-RevId: 175302425
* n/a (internal change only)Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175297329
* [XLA] Make TuplePointsToAnalysis and LogicalBufferAnalysis track nested ↵Gravatar A. Unique TensorFlower2017-11-10
| | | | | | fusion instructions. PiperOrigin-RevId: 175295981
* Extend the Array class with more functionality.Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175277161
* Correct comment in K-FAC's layer_collectionGravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175275184
* Java/OS X: Workaround for how the framework library is packaged in the .jar byGravatar Asim Shankar2017-11-10
| | | | | | | | the release process. See #13872 PiperOrigin-RevId: 175261983
* [XLA:CPU] Make one of the tile dimensions in the LLVM IR GEMV tunable.Gravatar Sanjoy Das2017-11-10
| | | | | | | | | | | | The tiling dimension corresponding to the number of vector registers in the tile can be changed easily. Expose this value as a backend specific flag so that we can experiment with it to find a good default value. This CL also fixes a bug exposed by a variable tiling factor in the row major GEMV implementation. This wasn't caught before because having tile_rows == tile_cols hides the bug. PiperOrigin-RevId: 175258553
* Get rid of IteratorBase::is_exhausted flag since it is not possible to rely ↵Gravatar Saurabh Saxena2017-11-10
| | | | | | | | | | | | | on it unless we lock each call to GetNext which is not preferable. Each iterator now handles saving/restoring exhausted state. As a guideline, we always reset the input_impl(s) when they get exhausted. This can be used as an indicator of exhausted-ness for non-terminal iterators. Also reduces memory overhead. Each iterator should also handle calls to GetNextInternal when it is exhausted. Fixed this for some datasets. Also fix a bug in dataset_serialization_test_base. We were not saving a checkpoint after exhausting the iterator so verify_exhausted_iterator was not really testing restoring an exhausted iterator. PiperOrigin-RevId: 175253023
* Add bfloat support to XLA.Gravatar Yunxing Dai2017-11-10
| | | | | | This is necessary in providing bfloat support in GPU backend. RELNOTES: bfloat support is now added to XLA infra. PiperOrigin-RevId: 175252067
* [XLA] Don't deemphasize nodes inside of subcomputations in dumped XLA graphs.Gravatar Justin Lebar2017-11-10
| | | | | | | | | | Nodes inside of subcomputations (e.g. fusion computations) are always printed by the HLO graph dumper. Before this change, the dumper was not fully aware of this fact, leading it to mark as "deemphasized" (i.e. draw as gray with a dashed outline) nodes that had no business of being deemphasized. PiperOrigin-RevId: 175247474
* Hlo parser: support padding.Gravatar A. Unique TensorFlower2017-11-10
| | | | | | Also, give PaddingConfig its own ToString format. PiperOrigin-RevId: 175239832
* [StreamExecutor] LOG(ERROR) the driver version when cudnnCreate fails.Gravatar Justin Lebar2017-11-10
| | | | | | | | | Previously we LOG(INFO)'ed the driver version, which meant it wouldn't be printed unless you passed --logtostderr. But this information is pretty important, especially since cudnnCreate failing is likely to be a fatal error. PiperOrigin-RevId: 175235628
* Improved the reporting of dimensionsGravatar Benoit Steiner2017-11-10
| | | | PiperOrigin-RevId: 175232587
* De-flakify and re-enable tests in replicate_model_fn_test.py.Gravatar Igor Saprykin2017-11-10
| | | | | | | | "Reduce metric variables" operation is a single operation across all metric variables, which means it is across all eval metrics. Previously, an update op for every eval metric was conditioned on a copy of overall "reduce metric variables" op. The latter was meant to be idempotent and thus the end result was supposed to be correct. However, "reduce metric variables" op consists of a number of variable assignments and thus is not atomic. If execution of two "reduce metric variables" ops interleaves, then the end result might come out to be incorrect. This caused flakiness in replicate_model_fn_test.py. To fix the problem, there is now a single copy of the "reduce metric variables" and every eval metric is associated with that single instance. PiperOrigin-RevId: 175232016
* Support more instructions in Hlo parser: Broadcast, Concatenate, Map, ↵Gravatar A. Unique TensorFlower2017-11-10
| | | | | | Reduce, SelectAndScatter, Reverse, Slice, DynamicSlice, DynamicUpdateSlice, Transpose, BatchNormTraining, BatchNormInference, BatchNormGrad. PiperOrigin-RevId: 175231463
* Increase tolerance in flaky multinomial test.Gravatar Gunhan Gulsoy2017-11-10
| | | | PiperOrigin-RevId: 175230217
* Always push updated nodes to the queue for possible further optimization.Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175229944
* Added some additional documentation to the swish() functionGravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175228315
* Materialize constants in more cases.Gravatar Benoit Steiner2017-11-10
| | | | PiperOrigin-RevId: 175228264
* Tensor template argument to gradienttape was unnecessary.Gravatar Alexandre Passos2017-11-10
| | | | PiperOrigin-RevId: 175225805
* Fix typo in tensorflow/python/framework/function.pyGravatar Yifei Feng2017-11-10
| | | | PiperOrigin-RevId: 175219920
* Make LLVMCompilerTest less stateful.Gravatar Sanjoy Das2017-11-10
| | | | | | | | | | | | Instead of assigning the pre and post optimization to a singleton xla::Compiler object, prefer creating a short-lived CpuCompiler or a GpuCompiler instance on the stack. Without this change, adding a second test case on the (Cpu|Gpu)Compiler in the same process triggers a use-after-free. (Btw, LLVMCompiler should really be spelled LlvmCompiler per Google C++ style, I'll do that rename shortly). PiperOrigin-RevId: 175218617
* [TF:XLA] Clean up unused XLA options and functions.Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175217850
* Adds streaming_dynamic_auc to Tensorflow contrib metrics. This metric ↵Gravatar A. Unique TensorFlower2017-11-10
| | | | | | differs from streaming_auc because it uses every prediction as a threshold rather than linearly spaced fixed thresholds. PiperOrigin-RevId: 175217002
* 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
* Internal change.Gravatar Anna R2017-11-10
| | | | PiperOrigin-RevId: 175213336
* Add tf.nn.softmax_cross_entropy_with_logits_v2 which enables backprop wrt ↵Gravatar Eugene Brevdo2017-11-10
| | | | | | | | | | the labels. Clarify current backprop behavior. Original bugfix by Alexandre Passos. PiperOrigin-RevId: 175211803
* Expose padded_batch_and_drop_remainderGravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175210678
* Limit internal fragmentation in BFCAllocator to 128mb per allocation.Gravatar Justin Lebar2017-11-10
| | | | | | | | | | Previously, if you had a very large allocation, it would round up to the next power of 2, and then, if this didn't fit in your GPU's available memory, eat all remaining memory in the device. Now we waste at most 128mb of memory in a large alloc. PiperOrigin-RevId: 175209995
* Instances per second in the eager microbenchmarks.Gravatar Alexandre Passos2017-11-10
| | | | PiperOrigin-RevId: 175207829
* Fix bug reported in b/69059093, by skipping rewrites that we can determine ↵Gravatar A. Unique TensorFlower2017-11-10
| | | | | | | | have already been applied. Make sure rewrites are idempotent by running the optimizer twice in unit tests. PiperOrigin-RevId: 175206742
* eager: README title tweak.Gravatar Asim Shankar2017-11-10
| | | | PiperOrigin-RevId: 175205782
* Fix GANEstimator docstring.Gravatar A. Unique TensorFlower2017-11-10
| | | | PiperOrigin-RevId: 175204075
* Modify quantization to support add ops that occur after all quantizable ↵Gravatar A. Unique TensorFlower2017-11-10
| | | | | | types, not just Conv2D. PiperOrigin-RevId: 175204002
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-29 08:56:21 +0000