[sharding_in_types] Enforce AxisTypes to always exist if set_mesh i…

…s used.

Also support `Auto` mode fully or mixed in with `User` mode. This works by overriding the sharding of `Auto` axes in the PartitionSpec with `Unconstrained` in `ShapedArray` constructor. The `ShapedArray` constructor is the central place where we can make such substitutions.

During lowering of shardings with auto axes, we mark the auto dims are `unspecifed_dims`. We don't mark all dims as unspecified because that would enable XLA to shard them even further which is not what we want if some of the dims are user sharded.

PiperOrigin-RevId: 704911253

jax/BUILD

@@ -455,6 +455,7 @@ pytype_strict_library(
         ":dtypes",
         ":effects",
         ":mesh",
+        ":partition_spec",
         ":pretty_printer",
         ":source_info_util",
         ":traceback_util",
@@ -558,6 +559,7 @@ pytype_strict_library(
         ":layout",
         ":op_shardings",
         ":partial_eval",
+        ":partition_spec",
         ":path",
         ":pickle_util",
         ":sharding",

jax/_src/core.py

@@ -39,6 +39,7 @@
 from jax._src import effects
 from jax._src import compute_on
 from jax._src import mesh as mesh_lib
+from jax._src.partition_spec import PartitionSpec as P, UnconstrainedSingleton
 from jax._src.errors import (
     ConcretizationTypeError, TracerArrayConversionError, TracerBoolConversionError,
     TracerIntegerConversionError, UnexpectedTracerError)
@@ -1599,13 +1600,30 @@ def _invalid_shape_error(shape: Shape, context: str=""):
 
   return TypeError(msg)
 
+# TODO(yashkatariya): Only works with User/Auto. Generalize it to work with
+# Collective too.
+def _maybe_modify_sharding(sharding):
+  if mesh_lib.AxisTypes.Auto not in sharding.mesh.axis_types:
+    return sharding
+
+  new_spec = []
+  for s in sharding.spec:
+    if s is None or isinstance(s, UnconstrainedSingleton):
+      new_spec.append(s)
+    else:
+      temp_s = s[0] if isinstance(s, tuple) else s
+      new_spec.append(
+          P.UNCONSTRAINED
+          if sharding.mesh._name_to_type[temp_s] == mesh_lib.AxisTypes.Auto else s)
+  return sharding.with_spec(new_spec)
+
 
 def get_sharding(sharding, ndim):
-  from jax._src.sharding_impls import NamedSharding, PartitionSpec as P  # type: ignore
+  from jax._src.sharding_impls import NamedSharding  # type: ignore
 
   if sharding is not None:
     assert len(sharding.spec) == ndim
-    return sharding
+    return _maybe_modify_sharding(sharding)
 
   context_mesh = mesh_lib.get_abstract_mesh()
   # TODO(yashkatariya): Error out and ask users to set the context mesh in their
@@ -1675,9 +1693,7 @@ def str_short(self, short_dtypes=False):
     dt_str = dt_str.replace('void', 'float0')
     if hasattr(self, 'sharding') and self.sharding is not None:
       shapestr = _get_shape_sharding_str(self.shape, self.sharding.spec)
-      axis_types = self.sharding.mesh.axis_types
-      axt = _get_axis_type_str(axis_types) if axis_types is not None else ''
-      return f'{dt_str}[{shapestr}]{axt}'
+      return f'{dt_str}[{shapestr}]'
     else:
       shapestr = ','.join(map(str, self.shape))
       return f'{dt_str}[{shapestr}]'
@@ -1689,26 +1705,13 @@ def _len(self, ignored_tracer):
       raise TypeError("len() of unsized object") from err  # same as numpy error
 
 
-def _get_axis_type_str(axis_types):
-  from jax._src.mesh import AxisTypes  # type: ignore
-
-  out = []
-  for t, axes in axis_types.items():
-    a = f"({','.join(a for a in axes)})" if isinstance(axes, tuple) else axes
-    if t == AxisTypes.Collective:
-      out.append(f"C:{a}")
-    elif t == AxisTypes.User:
-      out.append(f"U:{a}")
-    else:
-      assert t == AxisTypes.Auto
-      out.append(f"A:{a}")
-  return f"{{{', '.join(out)}}}"
-
 def _get_shape_sharding_str(shape, spec):
   out = []
   for s1, s2 in zip(shape, spec):
     if s2 is None:
       out.append(f"{s1}")
+    elif isinstance(s2, UnconstrainedSingleton):
+      out.append(f"{s1}")
     elif isinstance(s2, tuple):
       ss = ','.join(s for s in s2)
       out.append(f"{s1}@({ss})")

jax/_src/interpreters/mlir.py

@@ -50,6 +50,7 @@
 from jax._src.layout import AutoLayout, DeviceLocalLayout
 from jax._src.sharding import Sharding as JSharding
 from jax._src.sharding_impls import AUTO
+from jax._src.partition_spec import UnconstrainedSingleton
 from jax._src.lib import xla_client as xc
 from jax._src.lib import xla_extension
 from jax._src.lib.mlir import dialects, ir, passmanager
@@ -2524,12 +2525,19 @@ def lower_sharding_under_shit(ctx, op, aval, sharding_proto=None):
   # Don't emit a wsc under full manual mode to avoid increasing HLO size.
   if aval.sharding.mesh._are_all_axes_collective:
     return op
+  if aval.sharding.mesh._are_all_axes_auto:
+    return op
+  # TODO(yashkatariya): If all the axes in pspec are AUTO or collective,
+  # `return op` early and avoid bloating HLO size.
   proto = (aval.sharding._to_xla_hlo_sharding(aval.ndim).to_proto()
            if sharding_proto is None else sharding_proto)
-  # TODO(yashkatariya): Enable this
-  # unspecified_dims = (set(range(aval.ndim))
-  #                     if aval.sharding.mesh._any_axis_collective else None)
-  return wrap_with_sharding_op(ctx, op, aval, proto)
+  unspecified_dims = None
+  if aval.sharding.mesh._any_axis_collective:
+    unspecified_dims = set(range(aval.ndim))
+  elif aval.sharding.mesh._any_axis_auto:
+    unspecified_dims = {i for i, s in enumerate(aval.sharding.spec)
+                        if isinstance(s, UnconstrainedSingleton)}
+  return wrap_with_sharding_op(ctx, op, aval, proto, unspecified_dims)
 
 
 def set_sharding(op, sharding: xc.OpSharding | sharding_impls.SdyArraySharding):

jax/_src/interpreters/pxla.py

@@ -63,7 +63,7 @@
 from jax._src.lib import xla_client as xc
 from jax._src.lib.mlir import ir
 from jax._src.lib.mlir.dialects import hlo
-from jax._src.partition_spec import PartitionSpec
+from jax._src.partition_spec import PartitionSpec, UnconstrainedSingleton
 from jax._src.sharding import Sharding as JSharding
 from jax._src.sharding_impls import (
     ArrayMapping, ArrayMappingOrAutoOrUnspecified, AUTO, UNSPECIFIED,
@@ -2123,11 +2123,13 @@ def _concretize_abstract_shardings(shardings, avals, device_assignment):
   @lru_cache(maxsize=128)
   def _abstract_to_concrete_mesh(abstract_mesh):
     return mesh_lib.Mesh(
-        np_dev.reshape(abstract_mesh.axis_sizes), abstract_mesh.axis_names)
+        np_dev.reshape(abstract_mesh.axis_sizes), abstract_mesh.axis_names,
+        axis_types=abstract_mesh.axis_types)
 
   out = []
   for s, a in zip(shardings, avals):
-    if isinstance(s, UnspecifiedValue) and a.sharding is not None:
+    if (isinstance(s, UnspecifiedValue) and a.sharding is not None and
+        all(not isinstance(s, UnconstrainedSingleton) for s in a.sharding.spec)):
       out.append(NamedSharding(_abstract_to_concrete_mesh(a.sharding.mesh),
                                a.sharding.spec))
     else:

jax/_src/mesh.py

@@ -124,6 +124,7 @@ def axis_names_to_types(axis_types) -> dict[str, AxisTypes]:
 
 _mesh_object_dict = {}  # type: ignore
 
+MeshAxisType = dict[AxisTypes, str | tuple[str, ...]]
 
 class Mesh(contextlib.ContextDecorator):
   """Declare the hardware resources available in the scope of this manager.
@@ -178,11 +179,11 @@ class Mesh(contextlib.ContextDecorator):
 
   devices: np.ndarray
   axis_names: tuple[MeshAxisName, ...]
-  axis_types: dict[AxisTypes, str | tuple[str, ...]] | None
+  axis_types: MeshAxisType | None
 
   def __new__(cls, devices: np.ndarray | Sequence[xc.Device],
-              axis_names: str | Sequence[MeshAxisName],
-              axis_types: dict[AxisTypes, str | tuple[str, ...]] | None = None):
+              axis_names: str | Sequence[MeshAxisName], *,
+              axis_types: MeshAxisType | None = None):
     if not isinstance(devices, np.ndarray):
       devices = np.array(devices)
     if isinstance(axis_names, str):
@@ -216,7 +217,8 @@ def __new__(cls, devices: np.ndarray | Sequence[xc.Device],
     return self
 
   def __reduce__(self):
-    return (type(self), (self.devices, self.axis_names, self.axis_types))
+    return (type(self), (self.devices, self.axis_names),
+            {'axis_types': self.axis_types})
 
   def __eq__(self, other):
     if not isinstance(other, Mesh):
@@ -348,7 +350,7 @@ def local_devices(self):
 
   @functools.cached_property
   def abstract_mesh(self):
-    return AbstractMesh(self.shape_tuple, self.axis_types)
+    return AbstractMesh(self.shape_tuple, axis_types=self.axis_types)
 
 
 EMPTY_ENV = ResourceEnv(Mesh(np.empty((), dtype=object), ()))
@@ -373,8 +375,8 @@ class AbstractMesh:
   details.
   """
 
-  def __init__(self, shape_tuple: tuple[tuple[str, int], ...],
-               axis_types: dict[AxisTypes, str | tuple[str, ...]] | None = None):
+  def __init__(self, shape_tuple: tuple[tuple[str, int], ...], *,
+               axis_types: MeshAxisType | None = None):
     self.shape_tuple = shape_tuple
     self.axis_types = axis_types
     if self.shape_tuple:
@@ -434,6 +436,24 @@ def _are_all_axes_collective(self) -> bool:
       return False
     return all(t == AxisTypes.Collective for t in self.axis_types.keys())
 
+  @functools.cached_property
+  def _are_all_axes_auto(self) -> bool:
+    if self.axis_types is None:
+      return False
+    return all(t == AxisTypes.Auto for t in self.axis_types.keys())
+
+  @functools.cached_property
+  def _any_axis_collective(self) -> bool:
+    if self.axis_types is None:
+      return False
+    return any(t == AxisTypes.Collective for t in self.axis_types.keys())
+
+  @functools.cached_property
+  def _any_axis_auto(self) -> bool:
+    if self.axis_types is None:
+      return False
+    return any(t == AxisTypes.Auto for t in self.axis_types.keys())
+
   @property
   def devices(self):
     _raise_value_error("devices")
@@ -474,6 +494,8 @@ def _raise_value_error(name):
 
 @contextlib.contextmanager
 def set_abstract_mesh(mesh: AbstractMesh):
+  if mesh is not None and mesh.axis_types is None:
+    raise RuntimeError('Please set the AxisTypes of Mesh.')
   prev_val = jax_config.abstract_mesh_context_manager.swap_local(mesh)
   try:
     yield

jax/_src/partition_spec.py

@@ -14,7 +14,7 @@
 
 from __future__ import annotations
 
-class _UnconstrainedPartitionSingleton:
+class UnconstrainedSingleton:
 
   def __repr__(self):
     return "UNCONSTRAINED"
@@ -23,7 +23,7 @@ def __repr__(self):
 # Unconstrained sentinel value for PartitionSpec, representing a dimension for
 # which the user wants XLA to assign the best partitioning.
 # TODO(yashkatariya): May rename to AUTO.
-_UNCONSTRAINED_PARTITION = _UnconstrainedPartitionSingleton()
+_UNCONSTRAINED_PARTITION = UnconstrainedSingleton()
 
 
 class PartitionSpec(tuple):

jax/_src/sharding_impls.py

@@ -67,6 +67,19 @@ def _check_mesh_resource_axis(mesh, parsed_pspec, _manual_axes):
               f"is also found in manual_axes: {_manual_axes}.") from None
 
 
+@util.cache(max_size=128, trace_context_in_key=False)
+def _check_axis_type_consistency(mesh, parsed_pspec):
+  if mesh.axis_types is None:
+    return
+  for p in parsed_pspec:
+    if p is not None:
+      if not all(mesh._name_to_type[p[0]] == mesh._name_to_type[r] for r in p):
+        raise ValueError(
+            'AxisTypes should be the same in a tuple subset of PartitionSpec:'
+            f' {parsed_pspec.get_partition_spec()}. Got subset {p} with axis'
+            f' types: ({", ".join(str(mesh._name_to_type[r]) for r in p)})')
+
+
 def hashed_index(x) -> int:
   # This works for both `pjit` indices and `pmap` indices (which might
   # have an integer instead of a slice).
@@ -1084,6 +1097,7 @@ def preprocess(mesh, spec, parsed_pspec, _manual_axes=frozenset()):
         PartitionSpec() if spec is None else spec,
         "NamedSharding spec", allow_unconstrained_dims=True)
   _check_mesh_resource_axis(mesh, parsed_pspec, _manual_axes)
+  _check_axis_type_consistency(mesh, parsed_pspec)
   return parsed_pspec
 
 
@@ -1673,7 +1687,8 @@ def _gspmd_to_named_sharding_via_mesh(
 
 
 def make_mesh(axis_shapes: Sequence[int], axis_names: Sequence[str],
-              *, devices: Sequence[xc.Device] | None = None) -> mesh_lib.Mesh:
+              *, devices: Sequence[xc.Device] | None = None,
+              axis_types: mesh_lib.MeshAxisType | None = None) -> mesh_lib.Mesh:
   """Creates an efficient mesh with the shape and axis names specified.
 
   This function attempts to automatically compute a good mapping from a set of
@@ -1735,4 +1750,4 @@ def make_mesh(axis_shapes: Sequence[int], axis_names: Sequence[str],
   mesh_devices = mesh_utils.create_device_mesh(
       new_axis_shapes, devices,
       allow_split_physical_axes=allow_split_physical_axes)
-  return mesh_lib.Mesh(mesh_devices, axis_names)
+  return mesh_lib.Mesh(mesh_devices, axis_names, axis_types=axis_types)

jax/_src/test_util.py

@@ -1443,26 +1443,28 @@ def with_and_without_mesh(f):
       ('Mesh', (('x', 2),), (('i', 'x'),))
     ))(with_mesh_from_kwargs(f))
 
-def with_user_mesh(sizes, names):
+def with_user_mesh(sizes, names, axis_types=None):
+  axis_types = ({mesh_lib.AxisTypes.User: names}
+                if axis_types is None else axis_types)
   def decorator(fn):
     def mesh_fn(*args, **kwargs):
-      mesh = create_mesh(sizes, names)
+      mesh = create_mesh(sizes, names, axis_types=axis_types)
       with mesh_lib.set_mesh(mesh):
         return fn(*args, **kwargs, mesh=mesh)
     return mesh_fn
   return decorator
 
 
-def create_mesh(mesh_shape, axis_names, iota_order=False):
+def create_mesh(mesh_shape, axis_names, iota_order=False, axis_types=None):
   size = math.prod(mesh_shape)
   if len(jax.devices()) < size:
     raise unittest.SkipTest(f"Test requires {size} global devices.")
   if iota_order:
     devices = sorted(jax.devices(), key=lambda d: d.id)
     mesh_devices = np.array(devices[:size]).reshape(mesh_shape)
-    return jax.sharding.Mesh(mesh_devices, axis_names)
+    return jax.sharding.Mesh(mesh_devices, axis_names, axis_types=axis_types)
   else:
-    return jax.make_mesh(mesh_shape, axis_names)
+    return jax.make_mesh(mesh_shape, axis_names, axis_types=axis_types)
 
 class _cached_property:
   null = object()

jax/experimental/shard_map.py

@@ -46,7 +46,8 @@
 from jax._src import traceback_util
 from jax._src import util
 from jax._src.core import Tracer
-from jax._src.mesh import AbstractMesh, Mesh, AxisTypes, set_abstract_mesh
+from jax._src.mesh import (AbstractMesh, Mesh, AxisTypes, set_abstract_mesh,
+                           get_abstract_mesh)
 from jax._src.api import _shared_code_pmap, _prepare_pmap
 from jax._src.lax import (lax, parallel as lax_parallel, slicing,
                           windowed_reductions, convolution, fft, linalg,
@@ -536,7 +537,7 @@ def _shard_shaped_array(mesh: Mesh, names: AxisNames, aval: core.AbstractValue
                     for i, sz in enumerate(aval.shape))
   if config.sharding_in_types.value:
     new_mesh = AbstractMesh(
-        mesh.shape_tuple, {AxisTypes.Collective: mesh.axis_names})
+        mesh.shape_tuple, axis_types={AxisTypes.Collective: mesh.axis_names})
     new_sharding = NamedSharding(new_mesh, P(*[None] * aval.ndim))
   else:
     new_sharding = None
@@ -548,11 +549,9 @@ def _unshard_shaped_array(mesh: Mesh, names: AxisNames,
   assert isinstance(aval, core.ShapedArray)
   new_shape = tuple(sz * prod(mesh.shape[n] for n in names.get(i, ()))
                     for i, sz in enumerate(aval.shape))
-  # TODO(yashkatariya): Reset the mesh properly based on the input avals if the
-  # mesh of shard_map specifies collective axes.
   if config.sharding_in_types.value:
     spec = _names_to_pspec(names)._normalized_spec(aval.ndim)
-    new_sharding = NamedSharding(AbstractMesh(mesh.shape_tuple), spec)
+    new_sharding = NamedSharding(get_abstract_mesh(), spec)
   else:
     new_sharding = None
   return aval.update(shape=new_shape, sharding=new_sharding)