[XLA] Use max size of buffer in heap simulator and buffer assignment and make size DCHECKs into CHECKs

PiperOrigin-RevId: 564421004

PiperOrigin-RevId: 564420754

PiperOrigin-RevId: 564415721

Sometimes, there's an exorbitant number of edges between two same notes A and B
(e.g. passing parameters through). After ~64 it becomes unreadable, so better
to skip others.

PiperOrigin-RevId: 564379502

different flag default.

PiperOrigin-RevId: 564374361

Updates LLVM usage to match
[e592c2dcf5b7](https://github.com/llvm/llvm-project/commit/e592c2dcf5b7)

PiperOrigin-RevId: 564346651

[Rolling forward]

PiperOrigin-RevId: 564345870

This makes it harder to accidentally use -1.

PiperOrigin-RevId: 564334287

This reduces massively the memory allocated by XNNPACK for highly partitioned subgraphs

PiperOrigin-RevId: 564313169

PiperOrigin-RevId: 564309076

PiperOrigin-RevId: 564309075

When we use LOG(INFO), the messages appear on screen in the OSS build - this is usually not what we want.

PiperOrigin-RevId: 564305557

The change has been made in Bazel in 2020, so it should be safe to rename in TF as well.

PiperOrigin-RevId: 564301792

Extend conversion function from GoogleEdgeTpuSettings in protobuf format to GoogleEdgeTpuSettings in flatbuffer format.

PiperOrigin-RevId: 564301432

- Remove Experimental in the message name
- Add NO_QUANTIZE quantization scheme
- Change freeze_all_variables to bool type
- Allow users to specify both op type and node name in the quantization unit

PiperOrigin-RevId: 564231195

Explicit forbid dynamic shape aliasing due to potential runtime-compiler dynamic shape size calculation mismatching

PiperOrigin-RevId: 564183652

PiperOrigin-RevId: 564133181

PiperOrigin-RevId: 564133178

These are the minimal support ops required for initializing and running tests.

PiperOrigin-RevId: 564096777

PiperOrigin-RevId: 563976196

PiperOrigin-RevId: 563968239

PiperOrigin-RevId: 563968227

PiperOrigin-RevId: 563925797

PiperOrigin-RevId: 563919380

These utilities and MLIR types are required to enable PAX
model checkpointing involving the new TF int4 type.

PiperOrigin-RevId: 563915180

PiperOrigin-RevId: 563892946

Updates LLVM usage to match
[f8a1c8b7c1d0](https://github.com/llvm/llvm-project/commit/f8a1c8b7c1d0)

PiperOrigin-RevId: 563871269

PiperOrigin-RevId: 563858734

This is required to prevent a circular dependency in https://github.com/tensorflow/tensorflow/pull/60161

PiperOrigin-RevId: 563836979

PiperOrigin-RevId: 563836927

These types are used by JAX/XLA.  Expanding coverage in TF to match.

PiperOrigin-RevId: 563831916

Update CMake build rules to adopt new TSL path

PiperOrigin-RevId: 563829912

PiperOrigin-RevId: 563829150

This does not need to be disabled anymore now that TensorFlow IO has Apple Silicon wheels available.

See https://pypi.org/project/tensorflow-io-gcs-filesystem/0.34.0/#files

PiperOrigin-RevId: 563826666

PiperOrigin-RevId: 563822172

The current toolchain doesn't support it.

PiperOrigin-RevId: 563796297

PiperOrigin-RevId: 563787003

This method is no longer used by JAX, and its existence is misleading (what it returns does not correspond to any particular default device assignment that JAX uses). Remove it.

PiperOrigin-RevId: 563784500

PiperOrigin-RevId: 563784409

PiperOrigin-RevId: 563776249