These were temporary forwarding targets that are no longer needed; use //jaxlib/cpu:cpu_kernels and //jaxlib/cuda:cuda_gpu_kernels instead.
PiperOrigin-RevId: 738085234
This is an exact port of the current Python implementation to C++ for speed.
I am being careful not to change the topological order we return in any way in this change, although we may do so in a future change.
PiperOrigin-RevId: 737014989
For the CUDA and ROCM plugins, we only support exact matches between the plugin and jaxlib version, and bad things can happen if we try and load mismatched versions. This change issues a warning and skips importing a plugin when there is a version mismatch.
There are a handful of other places where plugins are imported throughout the JAX codebase (e.g. in lax_numpy, mosaic_gpu, and in the plugins themselves). In a follow up it would be good to add version checking there too, but let's start with just these ones.
PiperOrigin-RevId: 731808733
This change improves the stability and backward compatibility of Pallas Triton
calls, because unlike PTX, the Triton dialect has no stability guarantees
and does change in practice.
See #25196.
A few notes
* Pallas Triton no longer delegates compilation to PTX to XLA:GPU. Instead,
compilation is done via a new PjRt extension, which uses its own compilation
pipeline mirrored after the one in the Triton Python bindings.
* The implementation of the old custom call used by Pallas Triton is
deprecated and will be removed after 6 months as per
[compatibility guarantees] [*]
[*]: https://jax.readthedocs.io/en/latest/export/export.html#compatibility-guarantees
PiperOrigin-RevId: 722773884
This change does not yet do the work necessary to make any tests pass with threading enabled, which will come in future changes.
This approach is broadly inspired by a6d205dd4c/testtools/testsuite.py (L113) and by unittest-ft.
We add a custom TestResult class that batches up any test result actions and applies them under a lock. We also add a custom TestSuite class that runs individual test cases in parallel using a thread-pool.
We need a reader-writer lock to implement a `@jtu.thread_hostile_test` decorator, which we do by adding bindings around absl::Mutex to jaxlib.
PiperOrigin-RevId: 713312937
The `jaxlib/cuda_plugin_extension.cc` and `jaxlib/rocm_plugin_extension.cc` files were nearly identical so this change consolidates the shared implementation into a single target.
PiperOrigin-RevId: 704785926
There were two helper functions for implementing FFI calls that were included directly alongside jaxlib's CPU kernels that will be useful for the GPU kernels as well. This moves those functions into ffi_helpers so that they are accessible from there too.
PiperOrigin-RevId: 658002501
The OpenXLA project is working on an open source, MLIR, named-axis based propagation (and in the future SP<D partitioning) system that will be dialect agnostic (would work for any dialect - MHLO, StableHLO, YourDialect). We plan on having frontends like JAX and PyTorch target this when using XLA and wanting SPMD propagation/partitioning. See www.github.com/openxla/shardy for more info.
Currently Shardy is implemented inside the XLA compiler, requiring us to round-trip between StableHLO and HLO with `mhlo.sharding`s. But we will eventually make Shardy the first pass in the XLA pipeline while it's still working on StableHLO. Partitioning (the system that adds the collectives like all-gathers/all-reduces) will still be the GSPMD Partitioner, but next year the Shardy partitioner will be developed, allowing for propagation and partitioning to be completely in MLIR and the first pass in the pipeline. So then we'd have:
1. Traced jaxpr
2. Jaxpr -> StableHLO
3. StableHLO with Shardy propagation
4. StableHLO with Shardy partitioning
5. StableHLO -> HLO
6. XLA optimizations
The following test:
```py
def test_sdy_lowering(self):
mesh = jtu.create_global_mesh((4, 2), ('x', 'y'))
np_inp = np.arange(16).reshape(8, 2)
s = jax.sharding.NamedSharding(mesh, P('x', 'y'))
arr = jax.device_put(np_inp, s)
@partial(jax.jit, out_shardings=s)
def f(x):
return x * 2
print(f.lower(arr).as_text())
```
outputs:
```
module @jit_f attributes {mhlo.num_partitions = 8 : i32, mhlo.num_replicas = 1 : i32} {
sdy.mesh @mesh = <"x"=4, "y"=2>
func.func public @main(%arg0: tensor<8x2xi64> {mhlo.layout_mode = "{1,0}", sdy.sharding = #sdy.sharding<@mesh, [{"x"}, {"y"}]>}) -> (tensor<8x2xi64> {jax.result_info = "", mhlo.layout_mode = "default", sdy.sharding = #sdy.sharding<@mesh, [{"x"}, {"y"}]>}) {
%c = stablehlo.constant dense<2> : tensor<i64>
%0 = stablehlo.broadcast_in_dim %c, dims = [] : (tensor<i64>) -> tensor<8x2xi64>
%1 = stablehlo.multiply %arg0, %0 : tensor<8x2xi64>
return %1 : tensor<8x2xi64>
}
}
```
Shardy will be hidden behind the `jax_use_shardy_partitioner` flag initially before becoming enabled by default in the future.
PiperOrigin-RevId: 655127611
Some of the macros that were used in jaxlib's FFI calls to LAPACK turned out to
be useful for other FFI calls. This change consolidates these macros in the
ffi_helper header.
PiperOrigin-RevId: 651166306
This change does a few things (arguably too many):
1. The key change here is that it fixes the handler registration in `jaxlib/gpu/gpu_kernels.cc` for the two handlers that use the XLA FFI API. A previous attempt at this change caused downstream issues because of duplicate registrations, but we were able to fix that directly in XLA.
2. A second related change is to declare and define the XLA FFI handlers consistently using the `XLA_FFI_DECLARE_HANDLER_SYMBOL` and `XLA_FFI_DEFINE_HANDLER_SYMBOL` macros. We need to use these macros instead of the `XLA_FFI_DEFINE_HANDLER` version which produces a lambda, so that when XLA checks the address of the handler during registration it is consistent. Without this change, the downstream tests would continue to fail.
3. The final change is to consolidate the `cholesky_update_kernel` and `lu_pivot_kernels` implementations into a common `linalg_kernels` target. This makes the implementation of the `_linalg` nanobind module consistent with the other targets within `jaxlib/gpu`, and (I think!) makes the details easier to follow. This last change is less urgent, but it was what I set out to do so that's why I'm suggesting them all together, but I can split this in two if that would be preferred.
PiperOrigin-RevId: 651107659
This re-enables the tests removed in https://github.com/google/jax/pull/21563
and adds support for exposing the XLA FFI headers in the
`jax.extend.ffi.include_dir` directory during a bazel build. While it's
unlikely that these will be useful for most bazel users, it is good to provide
a consistent interface with the wheel build and to be able to test this feature.
PiperOrigin-RevId: 640194961
This lets us avoid bundling a whole another copy of LLVM with JAX packages
and so we can finally start building Mosaic GPU by default.
PiperOrigin-RevId: 638569750
JAX has stopped generating code that uses directly
the DUCC FFT custom calls.
The 6 months backwards compatibility window has also expired.
PiperOrigin-RevId: 638132572
The runfiles of the original targets were lost when the symlinked files were used.
This change is needed for future Hermetic CUDA implementation. Bazel will download CUDA distributives in cache, and CUDA executables and libraries will be added in the runfiles of the targets. When `xla_extension` is simlinked, the content of the runfiles is lost. With `genrule` the content of the runfiles is preserved.
PiperOrigin-RevId: 632508121
The nanobind switch for the GPU callback code means that we are now using the NumPy APIs rather than pybind11's clone of them. It is important to initialize the NumPy APIs before using them in each module.
PiperOrigin-RevId: 613036056
Jax isn't using this, and in fact our code to build this wasn't including the C++ parts, so it was broken anyway. Remove it until someone actually needs it for something.
PiperOrigin-RevId: 587323808
- Add a py extension to call the custom call C API.
- Change the implementation of register_custom_call_target to store handlers for the custom call targets and delays the registration until the handler for a xla platform is registered.
- Change register_plugin to load PJRT plugin when register_pluin is called (instead of when a client is created), and let it return the PJRT_Api* loaded.
- Delay calling discover_pjrt_plugins() and register_pjrt_plugin_factories_from_env() until the first time backends() is called.
PiperOrigin-RevId: 568265745
nanobind has a number of advantages (https://nanobind.readthedocs.io/en/latest/why.html), notably speed of compilation and dispatch, but the main reason to do this for these bindings is because nanobind can target the Python Stable ABI starting with Python 3.12. This means that we will not need to ship per-Python version CUDA plugins starting with Python 3.12.
PiperOrigin-RevId: 559898790
