PiperOrigin-RevId: 564864652

For an 8x8 uint32_t transpose, we had:
- 4x `vinsertf128 ymm, ymm, xmm`
- 4x `vperm2f128`.

These are very expensive instructions because they cross the 128-bit lane boundary.

Now, we have `8x vinsertf128` but crucially, the inserted operand now comes from memory. This is important because modern X86 HW can easily broadcast on load which means that `vinsertf128` turns into a blend instead of a shuffle.

We use the same trick for handling matrices which are smaller than the vector width to accelerate the transpose. We still require a cross lane step but we cut down all the other shuffles in half compared to SSE2.

While we are here, don't claim to support kernels which don't exist. This makes the transpose system choose unoptimized implementations.

PiperOrigin-RevId: 564860657

This check can be removed since tf2xla can run ops with non-const input even if CompileTimeConstant attribute is set with the help of valueinference.

PiperOrigin-RevId: 564851049

PiperOrigin-RevId: 564845580

Updates LLVM usage to match
[c1796be93fe5](https://github.com/llvm/llvm-project/commit/c1796be93fe5)

PiperOrigin-RevId: 564842806

PiperOrigin-RevId: 564840225

PiperOrigin-RevId: 564825677

Without this additional bazelrc file TSL will be built with the system compiler which happens to be GCC.

I also had to disable a warning that was raised by clang. It puzzles me a bit that this is not needed for the Tensorflow build which definitely uses Clang.

PiperOrigin-RevId: 564822820

I've seen this file take over 5 minutes to build. Shard it by type.

PiperOrigin-RevId: 564820851

PiperOrigin-RevId: 564813085

This is fixing a UB issue which occurs with newer version of Clang (17+).
The fix is also upstreamed through https://github.com/NVIDIA/nccl/pull/916.

In addition I'm changing the handling of `enqueue.cc` which needs to be compiled
in cuda mode under clang. The previous solution with just passing in the `-x cuda` option fails with CUDA 12+.

I'm also correcting the version number that we set in the patch - not sure if this version is reported in some logs, but if it is, it should be correct.

PiperOrigin-RevId: 564811002

Add CopyToMemorySpace to the PjRtBuffer API. This CL does not implement any instance of the method, but adds the ability to do so in followup CLs.

PiperOrigin-RevId: 564807735

PiperOrigin-RevId: 564807274

PiperOrigin-RevId: 564804399

Remove `uint64` -> `double` -> `uint64` conversions from `NowMicros()` and their users.

PiperOrigin-RevId: 564802323

This broke some tests outside of tflite and darwinn. Temporarily adding back these patterns until the test issues are resolved.

PiperOrigin-RevId: 564800351

PiperOrigin-RevId: 564799226

At the moment, StableHLO does not have the spec and the interpreter for `broadcast`. `broadcast` is a special case of `broadcast_in_dim`, and this pass makes `broadcast -> broadcast_in_dim` conversion.

PiperOrigin-RevId: 564796578

PiperOrigin-RevId: 564789520

PiperOrigin-RevId: 564767887

This was causing a missing symbol error for certain build targets.

PiperOrigin-RevId: 564766163

Imported from GitHub PR https://github.com/openxla/xla/pull/5512

Some build tests were failing due to unresolved symbols so add in their dependencies

This is a redo of https://github.com/tensorflow/tensorflow/pull/61489 which was not merged in time before the code was moved.
Copybara import of the project:

--
a1f80543ccb7b8cccad2e4ce4f33f18b68c3baf1 by Andrew Goodbody <andrew.goodbody@linaro.org>:

Fix link errors in build tests

Some build tests were failing due to unresolved symbols
so add in their dependencies

Merging this change closes #5512

PiperOrigin-RevId: 564758391

PiperOrigin-RevId: 564749546

PiperOrigin-RevId: 564748979

PiperOrigin-RevId: 564739804

PiperOrigin-RevId: 564721615

PiperOrigin-RevId: 564719695

use_tfrt=True has been the default for over a year, and the flag currently does nothing.
PiperOrigin-RevId: 564712316

It's going away in https://github.com/llvm/llvm-project/commit/5378117327317ac08eb25e52519f482af7571445

PiperOrigin-RevId: 564711879

PiperOrigin-RevId: 564693541

PiperOrigin-RevId: 564693001

Without this additional bazelrc file TSL will be built with the system compiler which happens to be GCC.

I also had to disable a warning that was raised by clang. It puzzles me a bit that this is not needed for the Tensorflow build which definitely uses Clang.

PiperOrigin-RevId: 564688122

PiperOrigin-RevId: 564686943

PiperOrigin-RevId: 564685838

Also reduce the number of arguments to most lambdas.

PiperOrigin-RevId: 564654639

Updates LLVM usage to match
[5bab59de4463](https://github.com/llvm/llvm-project/commit/5bab59de4463)

PiperOrigin-RevId: 564649900

PiperOrigin-RevId: 564649508

PiperOrigin-RevId: 564649507

PiperOrigin-RevId: 564648984

The following changes are included:
- Starting from TF2.13v, TensorFlow uses Clang as compiler for Linux. Thus, we update the toolchain in release_cpu_linux and release_gpu_linux configs.
- Preserve the old linux build options in the unsupported_cpu_linux and unsupported_gpu_linux configs. If your project fails to build with Clang, you can use these unsupported flags to replace the release flags in your build command. However, please note that the old toolchain is no longer officially supported by TensorFlow and the unsupported configs will be removed soon. We strongly recommend that you migrate to Clang as your compiler for TensorFlow Linux builds. Instructions are available in the official documentation: https://www.tensorflow.org/install/source#install_clang_recommended_linux_only. Another good alternative is to use our Docker containers to build and test TensorFlow: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/tools/tf_sig_build_dockerfiles.
- Add official linker options and container environment settings to release linux configs.
- Deduplicate build options in cpu.bazelrc and gpu.bazelrc.
- Delete outdated CI jobs.

PiperOrigin-RevId: 564620926