PiperOrigin-RevId: 328206256
Change-Id: I2f8a7a14d4a246fb86cd78b02e77c734e4178514

PiperOrigin-RevId: 328206157
Change-Id: I111ed8a3bcf8266babaad7a1018fcd39495c9350

It's reported that op_level_cost_estimator crashes when input depth or output depth
is zero.

This CL checks any zero value in dim and just returns zero Costs.

PiperOrigin-RevId: 328201526
Change-Id: I52c085d3e4b9f549a3e1d2a46c5d6b34bb696540

These ops are later clustered by _xla_outside_compiled attribute value.  Unsupported ops marked for outside compilation
should be clustered in a separate pass.

PiperOrigin-RevId: 328200978
Change-Id: Ia3c30138ebb6a42d2e1064277c2b7461f2f53204

PiperOrigin-RevId: 328196830
Change-Id: I57e903bff15216cefacd54eced45752d0b574c91

PiperOrigin-RevId: 328196743
Change-Id: Ib46dfc21b4d953caba2baf565664db7f1f7c2275

PiperOrigin-RevId: 328193995
Change-Id: I189e4c8f121df487521138914e29c1f46890f717

I don't think you're supposed to create RefCounted objects on stack. Since the
destructor should only be called when the reference becomes zero, i.e. by
Unref().

PiperOrigin-RevId: 328193689
Change-Id: I6500c78fdbcf1906e2d22302c2b45f4dc2cfd504

PiperOrigin-RevId: 328193073
Change-Id: Idf41aafe4b095dd2a6ea08fced44e04b3a67566c

PiperOrigin-RevId: 328189510
Change-Id: Ic2791b36123d374fa1d3521e66b18dd7b82e5c4a

PiperOrigin-RevId: 328185599
Change-Id: I6c080b2410f92d03911ed89bc979b948633eec8c

PiperOrigin-RevId: 328182995
Change-Id: I2bfb11c5043218b9dfb042c984f81c3d6e114e7d

PiperOrigin-RevId: 328182943
Change-Id: I5e5784a9185742f2f1d114d69f23c3d1da412b3f

PiperOrigin-RevId: 328181751
Change-Id: Icc9c27218a9024e64b5a95a1cd35d83c46ab876b

PiperOrigin-RevId: 328181545
Change-Id: I4e5d52e775a060fe686ee623986b423787171e81

__array_module__ is an experimental protocol for "duck array" compatibility that
for indicating how to find a "numpy compatible" module. The hope is to make it
easier to write generic code that works across a range of array libraries.

A full example, which should work equally work for TF-NumPy as for JAX, can be
found at https://github.com/google/jax/pull/4076. More examples, and motivation for
this protocol can be found at https://numpy.org/neps/nep-0037-array-module.html.

This design has not yet been finalized in NumPy, so at present it requires using
the experimental numpy_dispatch module: https://github.com/seberg/numpy-dispatch.

Unlike NumPy's __array_ufunc__ and __array_function__ protocols, __array_module__
by design has no backwards compatibility consequences. The protocol only controls
the behavior of numpy_dispatch.get_array_module() or numpy.get_array_module() --
it does not change any existing NumPy functions.

PiperOrigin-RevId: 328181308
Change-Id: I98b648a59709ed3bc295aaf97f471a1ff7ae1f05

PiperOrigin-RevId: 328180923
Change-Id: I7c1032442a881346bda0c09412cc089fe54f7724

Force long-running model revive tests to run in separate shards by putting them in a single Test class.

PiperOrigin-RevId: 328180778
Change-Id: Id0c7d39de91b7ca87c9a7ee874c41ee29f505970

Because jax.jit is annotated on top-level functions, and that backends are initialised only after GoogleInit, we cannot have the client when calling the C++ jit.

PiperOrigin-RevId: 328177723
Change-Id: I462385e4a687461bf41c0d3e2875d4736285c42d

PiperOrigin-RevId: 328177436
Change-Id: I76f5bbbc4ce6a235e9e5cfd5df32802d207be849

This is part of an effort to remove entirely the global constructor registration
in favor of explicit registry in order to get more manageable build dependencies.

PiperOrigin-RevId: 328176806
Change-Id: I2dd02eef77d439fcaf1f79f2fd537b8021aa2e2d

PiperOrigin-RevId: 328175218
Change-Id: I3e4aca437f68bc748595eff537799b3a809c5c24

Update third_party/tensorflow/compiler/mlir/tensorflow/utils/... to not depend on the global Dialect Registry (NFC)

PiperOrigin-RevId: 328171679
Change-Id: I3a4e40a04cb14b9c3d53239b31d3d642bb97daac

PiperOrigin-RevId: 328171480
Change-Id: I1adbe658c9e3f435d4a42c2627cbbfef297e02f3

PiperOrigin-RevId: 328168585
Change-Id: I702964790826638b194eb7c5b21f98492e90c727

PiperOrigin-RevId: 328164449
Change-Id: I2421d154e6c2f702def7d4274a4df51877f436aa

This is part of a series of patches intended to make MLIR Dialects explictly
registered and remove the "always_link" global-constructor based registration.
The registration is mainly for tools that need to parse a "TF" MLIR input.

PiperOrigin-RevId: 328161577
Change-Id: I533930a804d2a22b76d0ef36b5970b80500542ea

1. Update to use v2 public API from their v1 version.
2. Use the Gfile API to read/write file content for save_model code.

PiperOrigin-RevId: 328159642
Change-Id: I38373fb16449ab8d19f15f4e22ad99d3c598266b

Remove const type from vectors.

PiperOrigin-RevId: 328157979
Change-Id: I8df58b0b23831b842c04c3243290ca61ecf7f4aa

PiperOrigin-RevId: 328157361
Change-Id: I57675a6822cdd56c662aa877077d4688aa6a4411

PiperOrigin-RevId: 328150132
Change-Id: If1468b900e95398106fcca29d0263ebb8869731e

PiperOrigin-RevId: 328149666
Change-Id: I0c5561152383f12126ab9568c0facc4c3043c6a3

Updates LLVM usage to match
[bad7d6b3735d](https://github.com/llvm/llvm-project/commit/bad7d6b3735d)

PiperOrigin-RevId: 328142991
Change-Id: I54a09bd386149cd5a649f5e0ca7bcc97028692c5

PiperOrigin-RevId: 328141067
Change-Id: I154c55c9b1cc47ec069bfc663a1d3fb0f135e67d

PiperOrigin-RevId: 328140122
Change-Id: Iacbb71fb3288057de3db6e0ab06e1cea20961223

PiperOrigin-RevId: 328137876
Change-Id: I809b345ca9c54c352959987c904d122347436aa6

PiperOrigin-RevId: 328134070
Change-Id: I3ad10cd9ee442e7f478192c3ea0e1254077c5f2e

PiperOrigin-RevId: 328133653
Change-Id: I99ceafa58d99364665175222748ed1a4ae16b0e4

PiperOrigin-RevId: 328129947
Change-Id: If3fc9bd3ccb4a271c60e48e052969c07174326e4

PiperOrigin-RevId: 328128307
Change-Id: Ia9a5e9f4c20d78deee32b738ae1002d80eb935c1