Branch 158278922 #10495
Closed
Branch 158278922 #10495
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This is so we can get JavaScript protobufs. This CL also improves the web_aspect and makes some peculiar Closure Compiler errors go away relating to externs. PiperOrigin-RevId: 158061198
… and avoiding unnecessary casts. Benchmark with AVX+FMA enabled: Run on <redacted> (12 X 3492 MHz CPUs); 2017-06-05T12:54:07.881672447-07:00 CPU: Intel Haswell with HyperThreading (6 cores) dL1:32KB dL2:256KB dL3:15MB Benchmark Base (ns) New (ns) Improvement ------------------------------------------------------------------ BM_Multinomial_cpu_1_10000_4 250817 172953 +31.0% BM_Multinomial_cpu_1_10000_128 273834 187552 +31.5% BM_Multinomial_cpu_1_10000_10000 1174175 1130778 +3.7% BM_Multinomial_cpu_1_100000_4 2040741 1276761 +37.4% BM_Multinomial_cpu_32_10000_4 10221765 4498666 +56.0% BM_Multinomial_cpu_32_10000_128 10638159 4994754 +53.0% BM_Multinomial_cpu_32_100000_4 100790019 44193314 +56.2% BM_Multinomial_cpu_128_100000_1 431269640 182506078 +57.7% PiperOrigin-RevId: 158061480
…convolution for images smaller than 16x16. PiperOrigin-RevId: 158061669
PiperOrigin-RevId: 158062553
* Don't fold enter/exit nodes since that can interact badly with frames * Create proper control dependencies on switch nodes PiperOrigin-RevId: 158066691
This is the first step to replacing TuplePointsToAnalysis with a global, module-scoped analysis. This dataflow analysis identifies all values and their defs and uses in the XLA graph. The analysis is currently unused. Follow up CLs will add buffer alias analysis using this dataflow analysis, and incrementally switch the transformation passes (for example, CopyInsertion) to use these new module-scoped analyses. PiperOrigin-RevId: 158067910
PiperOrigin-RevId: 158073112
PiperOrigin-RevId: 158075939
PiperOrigin-RevId: 158076005
PiperOrigin-RevId: 158078511
… based on the input shapes of the op. In some cases (E.g: shape), knowing the shapes of the input is all that is necessary to infer the content of the output tensor. This improves shape inference. PiperOrigin-RevId: 158079306
PiperOrigin-RevId: 158082288
PiperOrigin-RevId: 158083952
…ined. Added tests for dnn part of dnn-linear-combined estimator. PiperOrigin-RevId: 158084898
PiperOrigin-RevId: 158086284
temp_workaround_http_archive still requires it. This change silences the spurious message. PiperOrigin-RevId: 158089834
… convolution for images smaller than 16x16. PiperOrigin-RevId: 158111294
PiperOrigin-RevId: 158120864
…on that is located inside of the org_tensorflow repo (previously it *had* to be a remote repo declared in workspace file). PiperOrigin-RevId: 158138601
…structures. This changes the behavior of constructors like `tf.contrib.data.Dataset.from_tensors()` when passed a list. Previously, the `nest` utility would recurse into each element of such a list and create a separate Dataset component. Now the list will be converted to a tensor, allowing code like: ```python dataset = tf.contrib.data.Dataset.from_tensor_slices(([1, 2, 3], [4, 5, 6])) ``` ...to define a dataset with two components (each of shape `()`). This change also adds support for dictionaries as nested structures, which simplifies integration with dictionary-returning ops like `tf.parse_example()`. Fixes tensorflow#10151. RELNOTES: Breaking change to `tf.contrib.data.Dataset` APIs that expect a nested structure. Lists are now converted to tf.Tensor implicitly. You may need to change uses of lists to tuples in existing code. In addition, dicts are now supported as a nested structure. PiperOrigin-RevId: 158139467
…try computation. PiperOrigin-RevId: 158140349
PiperOrigin-RevId: 158141564
PiperOrigin-RevId: 158141962
PiperOrigin-RevId: 158142102
…"return" and "static". PiperOrigin-RevId: 158143418
to avoid asan error about assigning inf to int64 (this comes in from a divide-by-0). PiperOrigin-RevId: 158155488
- Add a completed_sweeps variable to keep track of sweeps that have been completed during training. - Add a StopAtSweepHook, which can request a stop after completing a specified number of sweeps. PiperOrigin-RevId: 158156347
This implements a simple tone generator, with sine waves, square waves, and triangle waves, plus two simple combinations of sine waves. The step value is used to control the frequency. PiperOrigin-RevId: 158160889
PiperOrigin-RevId: 158161411
PiperOrigin-RevId: 158163537
…le them from being folded into a constant by graph optimizer. PiperOrigin-RevId: 158182282
PiperOrigin-RevId: 158186454
…fier. PiperOrigin-RevId: 158190192
PiperOrigin-RevId: 158191361
PiperOrigin-RevId: 158192152
Make reference edges orange. Remove animations from tooltips in the graph documentation. Previously, arrowheads were only added to reference edges (because we assumed users knew about the convention that arrowless edges flow upwards). That decision nicely reduces clutter. However, recently, some internal and external folks have expressed confusion, and so I want to try adding arrowheads to all data flow edges. And make the reference edges starkly different. See tensorflow#10428 PiperOrigin-RevId: 158195388
…ombined. Added tests for linear part of dnn-linear-combined estimator. PiperOrigin-RevId: 158200827
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
PiperOrigin-RevId: 158205361
…, +/- Inf). This helps the user identify problematic ops. Also moved the debugger data logic within tf-graph-info into a new tf-graph-debugger-data-card component. PiperOrigin-RevId: 158208679
To do so, NodeState now handles different output ports of a node (in case a node has multiple outputs). Also, VirtualScheduler code is cleaned up with more comments. PiperOrigin-RevId: 158209068
PiperOrigin-RevId: 158212897
Also fix indentation in configure. PiperOrigin-RevId: 158232959
PiperOrigin-RevId: 158265934
… on it. PiperOrigin-RevId: 158267803
PiperOrigin-RevId: 158268933
PiperOrigin-RevId: 158274845
ClientLibraryTestBase will now parse command-line flags for debug options automatically, permitting subclasses to override certain options by using mutable_debug_options. main() still has to call AppendDebugOptionsFlags() explicitly before running the TF flag parser. In the mean-time, this CL leaves flag handling to the current "legacy" approach. However, this is part of a larger plan to move *all* debugging flags for XLA into the DebugOptions message and expose them as flags from a single place. The other flags (which are not controlling debugging options) will have to be propagated more explicitly. PiperOrigin-RevId: 158276294
Add an HloLocation abstraction to dataflow analysis which indicates where (in the output of what instruction and at which index) an HloValue may appear. Previously only uses were stored with an HLO value where a use is an edge in the HLO graph (instruction, operand number and ShapeIndex). Also, change the handling of tuple-shaped kSelect instructions when ssa_form is true. Previously a phi value would be created. With this change the the value set instead contains the union of it's inputs identical to the ssa_form=false case. PiperOrigin-RevId: 158276598
PiperOrigin-RevId: 158278026
PiperOrigin-RevId: 158278512
controls the "tab name" of the summary that is displayed. This solution keeps using name_scope to keep names unique, but then prefixes the tag with the family name if provided. PiperOrigin-RevId: 158278922
copybara-service bot
pushed a commit
that referenced
this pull request
Apr 27, 2024
Imported from GitHub PR openxla/xla#10495 This patch makes sure that the host offloader will not introduce layout mismatches after removing the offloading custom calls by constraining the layout assignment to assign the same layout for the custom call's input and output. Copybara import of the project: -- ce6aeac3ac5445ab014d36c92a181dbe7afc35e7 by Jaroslav Sevcik <jsevcik@nvidia.com>: Enforce same input/output layout for offloading ops Merging this change closes #10495 FUTURE_COPYBARA_INTEGRATE_REVIEW=openxla/xla#10495 from jaro-sevcik:host-offload-call-layout-fix ce6aeac3ac5445ab014d36c92a181dbe7afc35e7 PiperOrigin-RevId: 628707325
copybara-service bot
pushed a commit
that referenced
this pull request
Apr 29, 2024
Imported from GitHub PR openxla/xla#10495 This patch makes sure that the host offloader will not introduce layout mismatches after removing the offloading custom calls by constraining the layout assignment to assign the same layout for the custom call's input and output. Copybara import of the project: -- ce6aeac3ac5445ab014d36c92a181dbe7afc35e7 by Jaroslav Sevcik <jsevcik@nvidia.com>: Enforce same input/output layout for offloading ops Merging this change closes #10495 FUTURE_COPYBARA_INTEGRATE_REVIEW=openxla/xla#10495 from jaro-sevcik:host-offload-call-layout-fix ce6aeac3ac5445ab014d36c92a181dbe7afc35e7 PiperOrigin-RevId: 628707325
copybara-service bot
pushed a commit
that referenced
this pull request
Apr 29, 2024
Imported from GitHub PR openxla/xla#10495 This patch makes sure that the host offloader will not introduce layout mismatches after removing the offloading custom calls by constraining the layout assignment to assign the same layout for the custom call's input and output. Copybara import of the project: -- ce6aeac3ac5445ab014d36c92a181dbe7afc35e7 by Jaroslav Sevcik <jsevcik@nvidia.com>: Enforce same input/output layout for offloading ops Merging this change closes #10495 PiperOrigin-RevId: 628970016
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.