Since jax2tf.convert is called recursively for the purpose of
serializing the vjp function, we must ensure that if the primal
function is a pjit with shardings then the vjp function must also
be converted as a pjit.
Without this fix the serialization with gradients of a pjit function
will fail the an error that there are shardings but not pjit at
the top-level.
This is the first step in a revision to how we handle the debug info pertaining
to staged functions' parameter names and result pytree paths. To limit
complexity, this first step adds machinery required to make our MLIR lowerings'
parameter and result names work, but it does *not* yet unify it with existing
arg-name machinery used at tracing time (in partial_eval.py, e.g.
partial_eval.DebugInfo etc). That unification will come in a follow up commits.
(I wrote the unified version first, then broke it down into this sequence of
commits.)
Another thing that will arrive in follow-up commits is pmap support (handling
static_broadcasted_argnames). This PR doesn't include support for pmap because
pmap's final style implementation requires slightly different machinery than
jit/pjit's initial style implementation. Indeed this PR removes the previous
support for pmap arg/result info, and skips the corresponding tests, because
the previous support didn't handle pmap's static_broadcasted_argnums (and I
think it could even lead to silently incorrect annotations when pmap was not at
the top-level, though I didn't work out an example case to be sure that was
possible).
This commit includes the changes from PR #15079, so that PR should be merged first.
Here's the _why_ of this change:
* The pre-existing solution (from PRs #14702, #14764, and #14813) did not
handle static_argnums or static_argnames correctly. Instead it would fail,
resulting in debug info being dropped from the jaxpr and ultimately the MLIR
computation (but no Exception raised). We need to handle
static_argnums/argnames because while the corresponding parameters remain on
the Python callable signature, they are excluded from the args/kwargs
pytrees; the previous solution didn't account for that divergence.
* The best way to handle static_argnums/argnames is to work out this debug info
when we still have the original args/kwargs in hand, i.e. much earlier than
the previous mechanism. We then just have to pass this debug info to the
right places. Indeed we often already had to work out some debug-related
information at these call sites (e.g. whether the function is being staged
out for jit, or scan, or whatever), so after this change we're working out
all the debug info at the same time.
* A side benefit is that now to get this debug info we no longer need to
unflatten user pytree defs with dummy objects (to reconstruct dummy
args/kwargs trees so that we can call inspect.signature(fun).bind), since we
just use the original args/kwargs instead. Since some user pytree node types
are not fully polymorphic in their element types (e.g. their __init__ methods
sometimes contained assertions about their elements' shapes, expecting them
to be arrays), that means the new mechanism is fundamentally more compatible
with custom pytree node types.
More concretely, effecting those high-level changes led to:
* replacing the previous `core.DebugInfo` with a class `core.JaxprDebugInfo`,
which in addition to the more precise name has fields like
`arg_names: Tuple[Optional[str], ...]` and
`result_paths: Tuple[Optional[str], ...]`, rather than
`in_tree: Optional[PyTreeDef]`, reflecting the fact that we work out the
actual debug info more eagerly than before and we don't need pytrees for
dummy-unflattening;
* introducing the new `partial_eval.TracingDebugInfo` class representing the
debug info about inputs which we have available at tracing time; in a
follow-up PR, we'll adapt partial_eval.py to use this new class and we'll
delete `partial_eval.DebugInfo` and its corresponding helper methods (not
done in this commit just to reduce complexity of each change);
* moving the old `core.DebugInfo`, which before #14702 lived in
partial_eval.py, back to partial_eval.py pending cleanup (deletion) of that
partial_eval.py debug info code;
* making specific jaxpr-processing functions produce an appropriately updated
`core.JaxprDebugInfo` object for their output (e.g. `pe.dce_jaxpr` prunes
elements from the `arg_names` field), maintaining now-checked invariants like
a Jaxpr's `debug_info` should have the same number of argument names as the
jaxpr has invars (the jaxpr-processing functions updated here are enough for
top-level jit jaxprs to have debug info attached, handling the original
intended use case of jit(f).lower, but not e.g. grad-of-jit cases, which can
be handled later by updating `ad.jvp_jaxpr` and the like to produce updated
debug info on their outputs);
* add some tests for static_argnums/static_argnames.
Phew! Can't wait to land those follow-ups too :P
* Remove {in|out}_positional_semantics from pjit_p.bind
* Remove `in_is_global` from lower_sharding_computation
* Remove local_to_global and global_to_local
* Clean up some arguments of sharded_lowering since they are not needed
PiperOrigin-RevId: 517469390
* Define use_cpp_class and use_cpp_method decorators as no-ops for type checking.
* Remove the use of abc.ABC when defining the Sharding type. This triggers a pytype bug: the easiest fix seems to be to skip the use of the ABC.
* Write use_cpp_class decorator differently on ArrayImpl to work around pytype bug.
* Fix a few new type errors.
PiperOrigin-RevId: 516631428
By defining the Sharding base class in its own module, we can pull it out into a separate Bazel submodule, which will help pytype inference when defining Array.
PiperOrigin-RevId: 516223009
This work is an effort to reduce cyclic dependencies in JAX internals.
Move the _global_to_local and _local_to_global methods out of Mesh and into pxla as free functions. This removes the need for jax._src.mesh to depend on things like avals.
PiperOrigin-RevId: 515667671
In a previous CL we introduced cross-lowering support without any
changes in JAX core, but at the expense of some overly complex code
in jax2tf, along with overriding a JAX core function. Plus, those
changes were not enough to handle some xmap and pmap cases.
Here we introduce a `_experimental_lowering_platform: Optional[str]` parameter
to the `.lower()` methods and then we thread the `lowering_platform`
all the way to the calls to `mlir.lower_jaxpr_to_module2`. That's it.
Note that this parameter to `.lower()` is experimental and not supposed
to be used outside jax2tf. It may also gobble user kwargs.
Limit jax._src.lib to shims around jaxlib and nothing else.
The goal of this change is to avoid a dependency cycle between the rest of jax and jax._src.lib in a Bazel build. This allows the types for jax._src.lib to be inferred by pytype in isolation without referring to the rest of JAX.
PiperOrigin-RevId: 512922397
Make jax.experimental.global_device_array a shim around jax._src.global_device_array.
Change in preparation for deprecating global device arrays.
PiperOrigin-RevId: 510261140
Before:
```
ValueError: Devices of all `Array` inputs and outputs should be the same. Got array device ids [0] on platform CPU and another array's device ids [0, 1, 2, 3] on platform CPU
```
After:
```
ValueError: Received incompatible devices for jitted computation. Got argument inp of ArrayPjitTest.test_jit_with_sharding_constraint_committed_inp_error.<locals>.sharded_inp with bfloat16[8,2] and device ids [0] on platform CPU and with_sharding_constraint or nested pjit or shard_map with device ids [0, 1, 2, 3] on platform CPU at jax/tests/pjit_test.py:2509 (sharded_inp)
```
PiperOrigin-RevId: 508746961
