Move dtype canonicalization out of core.AbstractValue subclasses.

This is a strictly mechanical change that moves abstract value canonicalization out of the core.AbstractValue subclasses and into their callers. This makes it safe to manipulate non-canonical abstract values even inside an -x32 context.

The callers to which canonicalization was added were:
a) all callers of `ConcreteArray` inside the JAX Tree.
b) all callers of `ShapedArray` and `UnshapedArray` that were found to be passing non-canonical dtypes during a global presubmit. These were identified by adding an assertion that the dtype is in fact canonical and fixing all the resulting test failures.

PiperOrigin-RevId: 414704700

Author: hawkinsp

jax/_src/abstract_arrays.py

@@ -49,18 +49,23 @@ def zeros_like_array(x):
                np.complex64, np.complex128,
                np.longlong, np.intc}
 
+def canonical_concrete_aval(val, weak_type=None):
+  return ConcreteArray(dtypes.canonicalize_dtype(np.result_type(val)), val,
+                       weak_type=weak_type)
+
 for t in array_types:
-  core.pytype_aval_mappings[t] = ConcreteArray
+  core.pytype_aval_mappings[t] = canonical_concrete_aval
   ad_util.jaxval_zeros_likers[t] = zeros_like_array
 
 core.literalable_types.update(array_types)
 
 def _zeros_like_python_scalar(t, x):
-  aval = core.ShapedArray((), dtypes.python_scalar_dtypes[t], weak_type=True)
+  dtype = dtypes.canonicalize_dtype(dtypes.python_scalar_dtypes[t])
+  aval = core.ShapedArray((), dtype, weak_type=True)
   return ad_util.zeros_like_aval(aval)
 
 def _make_concrete_python_scalar(t, x):
-  return ConcreteArray(
+  return canonical_concrete_aval(
     np.array(x, dtype=dtypes._scalar_type_to_dtype(t, x)),
     weak_type=True)
 

jax/_src/api.py

@@ -69,7 +69,7 @@
                                      local_devices, process_index,
                                      process_count, host_id, host_ids,
                                      host_count, default_backend)
-from jax.core import ConcreteArray, ShapedArray, raise_to_shaped
+from jax.core import ShapedArray, raise_to_shaped
 from jax.interpreters import partial_eval as pe
 from jax.interpreters import xla
 from jax.interpreters import pxla

jax/_src/device_array.py

@@ -23,6 +23,7 @@
 
 from jax import core
 from jax._src.config import config
+from jax._src import abstract_arrays
 from jax._src import dtypes
 from jax._src import profiler
 from jax._src.lib import xla_client as xc
@@ -306,4 +307,4 @@ class DeletedBuffer(object): pass
 device_array_types: List[type] = [xc.Buffer, _DeviceArray]
 for _device_array in device_array_types:
   core.literalable_types.add(_device_array)
-  core.pytype_aval_mappings[device_array] = core.ConcreteArray
+  core.pytype_aval_mappings[device_array] = abstract_arrays.canonical_concrete_aval

jax/_src/lax/control_flow.py

@@ -1480,7 +1480,7 @@ def scan(f, init, xs, length=None):
     return carry, stacked_y
 
   x_shapes = [masking.padded_shape_as_value(x.shape[1:]) for x in xs_flat]
-  x_dtypes = [x.dtype for x in xs_flat]
+  x_dtypes = [dtypes.canonicalize_dtype(x.dtype) for x in xs_flat]
   x_avals = tuple(_map(ShapedArray, x_shapes, x_dtypes))
 
   def _create_jaxpr(init):
@@ -2038,7 +2038,7 @@ def masked(*args):
                  for new_c, c in zip(new_carry, carry)]
     return [i + 1] + new_carry + ys
 
-  aval = ShapedArray((), dtypes.int_)
+  aval = ShapedArray((), dtypes.canonicalize_dtype(dtypes.int_))
   const_avals, carry_avals, x_avals = split_list(jaxpr.in_avals, [num_consts, num_carry])
   return _make_closed_jaxpr(masked, [aval] + const_avals + [aval] + carry_avals + x_avals)
 

jax/_src/lax/slicing.py

@@ -1240,7 +1240,8 @@ def _gather_batching_rule(batched_args, batch_dims, *, dimension_numbers,
     if core.symbolic_equal_dim(operand.shape[0], 0):
       output_shape = _gather_shape_rule(
           core.ShapedArray(operand.shape[1:], operand.dtype),
-          core.ShapedArray(indices.shape[1:], indices.dtype),
+          core.ShapedArray(indices.shape[1:],
+                           dtypes.canonicalize_dtype(indices.dtype)),
           dimension_numbers=dimension_numbers, slice_sizes=slice_sizes,
           unique_indices=unique_indices, indices_are_sorted=indices_are_sorted,
           mode=mode, fill_value=fill_value)
@@ -1456,8 +1457,8 @@ def _scatter_translation_rule(ctx, avals_in, avals_out, operand, indices,
   if mode == GatherScatterMode.CLIP:
     clip_fn = xla.lower_fun(_clamp_scatter_indices, multiple_results=False,
                             new_style=True)
-    indices, = clip_fn(ctx, avals_in, [indices_aval.update(dtype=np.int64)],
-                       operand, indices, updates, dnums=dimension_numbers)
+    indices, = clip_fn(ctx, avals_in, None, operand, indices, updates,
+                       dnums=dimension_numbers)
 
   c = ctx.builder
 
@@ -1477,8 +1478,8 @@ def _scatter_add_translation_rule(
   if mode == GatherScatterMode.CLIP:
     clip_fn = xla.lower_fun(_clamp_scatter_indices, multiple_results=False,
                             new_style=True)
-    indices, = clip_fn(ctx, avals_in, [indices_aval.update(dtype=np.int64)],
-                       operand, indices, updates, dnums=dimension_numbers)
+    indices, = clip_fn(ctx, avals_in, None, operand, indices, updates,
+                       dnums=dimension_numbers)
 
   dtype = operand_aval.dtype
   scatter_dims = _scatter_dimensions_proto(

jax/_src/lax/utils.py

@@ -60,8 +60,8 @@ def standard_abstract_eval(prim, shape_rule, dtype_rule, weak_type_rule,
   least_specialized = _max(map(type, avals),
                            key=operator.attrgetter('array_abstraction_level'))
   if least_specialized is core.ConcreteArray:
-    return core.ConcreteArray(prim.impl(*[x.val for x in avals], **kwargs),
-                         weak_type=weak_type)
+    out = prim.impl(*[x.val for x in avals], **kwargs)
+    return core.ConcreteArray(out.dtype, out, weak_type=weak_type)
   elif least_specialized is core.ShapedArray:
     return core.ShapedArray(shape_rule(*avals, **kwargs),
                             dtype_rule(*avals, **kwargs), weak_type=weak_type,
@@ -81,7 +81,7 @@ def standard_multi_result_abstract_eval(
   weak_types = weak_type_rule(*avals, **kwargs)
   if least_specialized is core.ConcreteArray:
     out_vals = prim.impl(*[x.val for x in avals], **kwargs)
-    return [core.ConcreteArray(val, weak_type=weak_type)
+    return [core.ConcreteArray(val.dtype, val, weak_type=weak_type)
             for val, weak_type in safe_zip(out_vals, weak_types)]
   elif least_specialized is core.ShapedArray:
     out_shapes = shape_rule(*avals, **kwargs)

jax/core.py

@@ -1041,7 +1041,7 @@ class UnshapedArray(AbstractValue):
   array_abstraction_level = 2
 
   def __init__(self, dtype, weak_type=False):
-    self.dtype = np.dtype(dtypes.canonicalize_dtype(dtype))
+    self.dtype = np.dtype(dtype)
     self.weak_type = weak_type
 
   def update(self, dtype=None, weak_type=None):
@@ -1183,19 +1183,20 @@ class ConcreteArray(ShapedArray):
   __slots__ = ['val']
   array_abstraction_level = 0
 
-  def __init__(self, val, weak_type=None):
-    super().__init__(np.shape(val), np.result_type(val),
-                     weak_type=dtypes.is_weakly_typed(val) if weak_type is None else weak_type)
+  def __init__(self, dtype, val, weak_type=None):
+    super().__init__(
+        np.shape(val), dtype,
+        weak_type=dtypes.is_weakly_typed(val) if weak_type is None else weak_type)
     # Note: canonicalized self.dtype doesn't necessarily match self.val
+    assert self.dtype == dtypes.canonicalize_dtype(np.result_type(val)), (val, dtype)
     self.val = val
     assert self.dtype != np.dtype('O'), val
 
-  def update(self, val=None, weak_type=None):
-    if val is None:
-      val = self.val
-    if weak_type is None:
-      weak_type = self.weak_type
-    return ConcreteArray(val, weak_type)
+  def update(self, dtype=None, val=None, weak_type=None):
+    dtype = self.dtype if dtype is None else dtype
+    val = self.val if val is None else val
+    weak_type = self.weak_type if weak_type is None else weak_type
+    return ConcreteArray(dtype, val, weak_type)
 
   def __eq__(self, other):
     if (type(self) is type(other) and self.dtype == other.dtype
@@ -1271,7 +1272,8 @@ def raise_to_shaped(aval: AbstractValue, weak_type=None):
   Bot: lambda aval, _: aval,
   UnshapedArray: lambda aval, _: aval,
   ShapedArray: lambda aval, weak_type: ShapedArray(
-      aval.shape, aval.dtype, weak_type, aval.named_shape)
+      aval.shape, dtypes.canonicalize_dtype(aval.dtype), weak_type,
+      aval.named_shape)
 }
 
 ### Operations on shapes and dimension sizes.

jax/experimental/jax2tf/call_tf.py

@@ -351,14 +351,14 @@ def is_fully_known_shape(s):
   xla_comp_parameter_shapes = xla_comp.program_shape().parameter_shapes()
   found_parameter_avals = [
       core.ShapedArray(found_xla_shape.dimensions(),
-                       found_xla_shape.numpy_dtype())
+                       dtypes.canonicalize_dtype(found_xla_shape.numpy_dtype()))
       for found_xla_shape in xla_comp_parameter_shapes
   ]
   # Add the captured_inputs to args_flat_sig_tf
   expected_args_flat_sig_tf = list(args_flat_sig_tf) + list(captured_inputs)
   expected_parameter_avals = [
       core.ShapedArray(tuple(arg_sig.shape.as_list()),
-                       arg_sig.dtype.as_numpy_dtype)
+                       dtypes.canonicalize_dtype(arg_sig.dtype.as_numpy_dtype))
       for arg_sig in expected_args_flat_sig_tf]
   if found_parameter_avals != expected_parameter_avals:
     msg = ("Compiled TensorFlow function has unexpected parameter types " +

jax/interpreters/masking.py

@@ -459,7 +459,8 @@ def __init__(self, trace, val, polymorphic_shape):
 
   @property
   def aval(self):
-    return ShapedArray(self.polymorphic_shape, self.dtype)
+    return ShapedArray(self.polymorphic_shape,
+                       dtypes.canonicalize_dtype(self.dtype))
 
   @property
   def dtype(self):

jax/interpreters/pxla.py

@@ -45,6 +45,7 @@
 from jax._src.config import config
 from jax import core
 from jax import linear_util as lu
+from jax._src import abstract_arrays
 from jax._src.abstract_arrays import array_types
 from jax.core import ConcreteArray, ShapedArray
 from jax._src import device_array
@@ -740,7 +741,7 @@ def _register_handlers_for_sharded_device_array(sda):
   shard_arg_handlers[sda] = _shard_sharded_device_array_slow_path
   xla.register_constant_handler(sda, _sharded_device_array_constant_handler)
 
-  core.pytype_aval_mappings[sda] = ConcreteArray
+  core.pytype_aval_mappings[sda] = abstract_arrays.canonical_concrete_aval
   dispatch.device_put_handlers[sda] = dispatch._device_put_array
   xla.pytype_aval_mappings[sda] = op.attrgetter("aval")
   xla.canonicalize_dtype_handlers[sda] = identity