Skip to content
Permalink
Branch: master
Commits on Sep 17, 2019
  1. compat: Update forward compatibility horizon to 2019-09-17

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269521565
  2. Update GraphDef version to 161.

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269521367
  3. Update ops-related pbtxt files.

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269473605
  4. Merge pull request #32263 from nouiz:utils_pr_window_utils

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269464303
  5. iOS Metal GPU delegate: relu tests added.

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269458898
  6. Fix typo in GetProtoTextFunctionCode comments.

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269458212
  7. Autogenerate (de)serialization for Extended Instruction Sets

    tensorflower-gardener committed Sep 17, 2019
    A generic mechanism for (de)serialization of extended instruction sets
    is added with this CL. To facilitate this, a new class
    "SPV_ExtendedInstSetOp" is added which is a base class for all
    operations corresponding to extended instruction sets. The methods to
    (de)serialization such ops as well as its dispatch is generated
    automatically.
    
    The behavior controlled by autogenSerialization and hasOpcode is also
    slightly modified to enable this. They are now decoupled.
    1) Setting hasOpcode=1 means the operation has a corresponding
       opcode in SPIR-V binary format, and its dispatch for
       (de)serialization is automatically generated.
    2) Setting autogenSerialization=1 generates the function for
       (de)serialization automatically.
    So now it is possible to have hasOpcode=0 and autogenSerialization=1
    (for example SPV_ExtendedInstSetOp).
    
    Since the dispatch functions is also auto-generated, the input file
    needs to contain all operations. To this effect, SPIRVGLSLOps.td is
    included into SPIRVOps.td. This makes the previously added
    SPIRVGLSLOps.h and SPIRVGLSLOps.cpp unnecessary, and are deleted.
    
    The SPIRVUtilsGen.cpp is also changed to make better use of
    formatv,making the code more readable.
    
    PiperOrigin-RevId: 269456263
  8. Added new experimental layout IMAGE_BUFFER.

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269455466
  9. Yet another gen_file fix to support '>' namespace separators in op na…

    tensorflower-gardener committed Sep 17, 2019
    …mes. (In this case when an op has multiple named outputs)
    
    PiperOrigin-RevId: 269455111
  10. iOS Metal GPU delegate: padding test added.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269449368
  11. Merge pull request #31713 from frreiss:issue-data-lmdb-doc

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269448211
  12. Merge pull request #32509 from Intel-tensorflow:fixing-Mkl_QuantizeV2_Op

    tensorflower-gardener committed Sep 17, 2019
    PiperOrigin-RevId: 269447608
  13. Update ops-related pbtxt files.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269446893
  14. Factor out Variant wrapper for shared ptrs.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269446584
Commits on Sep 16, 2019
  1. Save unnecessary lookup in NodeMap constructor.

    tensorflower-gardener committed Sep 16, 2019
    Run on XXXX (72 X 2991 MHz CPUs); 2019-09-16T15:05:55.08629839-07:00
    CPU: Intel Skylake Xeon with HyperThreading (36 cores) dL1:32KB dL2:1024KB dL3:24MB
    Benchmark                          Base (ns)  New (ns) Improvement
    ------------------------------------------------------------------
    BM_NodeMapConstruct/8                   1504      1324    +12.0%
    BM_NodeMapConstruct/64                 20784     17478    +15.9%
    BM_NodeMapConstruct/512               213301    178139    +16.5%
    BM_NodeMapConstruct/4k               2200710   1871703    +15.0%
    BM_NodeMapConstruct/32k             25063213  22232692    +11.3%
    BM_NodeMapConstruct/256k            370569289  326468834    +11.9%
    BM_NodeMapConstruct/1M              1832415652  1596115918    +12.9%
    PiperOrigin-RevId: 269444504
  2. Implemented "Greedy by size planner" for that produces more optimal m…

    tensorflower-gardener committed Sep 16, 2019
    …emory allocation.
    
    - in Mobilenet V1 has 35% has memory reduction
    - in Mobilenet V2 has 2% reduction
    Majority of all other tested models demonstrated ~10-15% improvement.
    
    PiperOrigin-RevId: 269423980
  3. Added possibility to use different storage types for the same model.

    tensorflower-gardener committed Sep 16, 2019
    Using TensorStorageType::SINGLE_TEXTURE_2D for input/output tensors when it possible (channels < 4)
    
    PiperOrigin-RevId: 269415203
  4. Implement new Gpu delegate to combine OpenGL and OpenCL based acceler…

    tensorflower-gardener committed Sep 16, 2019
    …ation. This delegate replaces existing GPU delegate defined in gl_delegate.
    
    It will not support hardware buffers as input and output. For such advanced usecase one should use GPU api directly.
    
    PiperOrigin-RevId: 269414244
  5. iOS Metal GPU delegate: reshape tests added.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269413768
  6. Introduce optional reallocation graph for parallel operations executi…

    tensorflower-gardener committed Sep 16, 2019
    …on and support it in greedy in order memory allocation algorithm.
    
    PiperOrigin-RevId: 269412894
  7. Added a lowering operation from tf.select to xla_hlo.select.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269410436
  8. Automated rollback of commit 4bd8a42

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269410078
  9. Use DeleteSubrange when deleting nodes from the graph in the new Grap…

    tensorflower-gardener committed Sep 16, 2019
    …hView.
    
    PiperOrigin-RevId: 269403726
  10. Fix gRPC deadline setting in CallWithRetry

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269390248
  11. Merge pull request #32463 from HarikrishnanBalagopal:patch-1

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269386850
  12. Merge pull request #32498 from gmiodice:pool_int8_test

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269386702
  13. [tfdbg] Add mnist and fib examples for tensorflow v2

    tensorflower-gardener committed Sep 16, 2019
    Both examples replicate the same behavior of the tensorflow
    v1 examples.
    
    PiperOrigin-RevId: 269385509
  14. iOS Metal GPU backend: upsample operation tests added.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269376277
  15. internal change

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269366745
  16. TFL iOS delegate build static binary framework.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269361936
  17. Merge pull request #32456 from Mycaster:add_log_info

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269357134
  18. Added method for more flexible checks.

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269349505
  19. [tfdbg] Fix logic issues in tf v1 mnist example

    tensorflower-gardener committed Sep 16, 2019
    The mnist example had been breaking because the placeholders had
    an incorrectly defined shape, and the model was being trained on
    the incorrect dataset.
    
    PiperOrigin-RevId: 269345214
  20. [tfdbg] Move tensorflow v1 debugger examples to separate folder

    tensorflower-gardener committed Sep 16, 2019
    PiperOrigin-RevId: 269338589
  21. Overhaul the SDBM expression kind hierarchy

    tensorflower-gardener committed Sep 16, 2019
    Swap the allowed nesting of sum and diff expressions: now a diff expression can
    contain a sum expression, but only on the left hand side.  A difference of two
    expressions sum must be canonicalized by grouping their constant terms in a
    single expression.  This change of sturcture became possible thanks to the
    introduction of the "direct" super-kind.  It is necessary to enable support of
    sum expressions on the left hand side of the stripe expression.
    
    SDBM expressions are now grouped into the following structure
    - expression
      - varying
        - direct
          - sum <- (term, constant)
          - term
            - symbol
            - dimension
            - stripe <- (term, constant)
        - negation <- (direct)
        - difference <- (direct, term)
      - constant
    The notation <- (...) denotes the types of subexpressions a compound
    expression can combine.
    
    PiperOrigin-RevId: 269337222
Older
You can’t perform that action at this time.