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
Previously, division was only supported in certain situation, and this
led to errors, e.g., when using strides. Now we generalize the polynomials
to also include "floordiv(E, E)" and "mod(E, E)" as atoms, in addition
to dimension variables. A symbolic dimension is now a sum of products
of atoms. (We also changed the documentation to use symbolic dimension
instead of dimension polynomials).
In presence of dynamic shapes the ThreeFry2x32Descriptor will contain the
value n=-1, and the actual desired output length will be passed as
an additional operand. If the shape is static then the length will be
passed as part of the descriptor.
PiperOrigin-RevId: 497945778
... in preparation for paring down `jax.core`'s exported symbols.
Also includes a few import fixups along the way, and a TODO comment to avoid an
import cycle in `_src/dtypes.py`.
PiperOrigin-RevId: 496024782
This CL renames occurrences of "mhlo" in: 1) names, 2) tests, 3) prose in order
to prepare for the upcoming migration.
Unchanged occurrences:
1) Public API that contains "mhlo", e.g. XlaLowering.mhlo and the "mhlo"
argument value in Lowering.as_text and Lowering.compiler_ir.
2) Documentation (changelog, JEPs, IR examples, etc).
3) One rare situation where prose says "StableHLO" and "MHLO" in one sentence,
so both are necessary to disambiguate.
PiperOrigin-RevId: 495771153
The immediate motivation for this is to support the lowering
to StableHLO for programs with polymorphic shapes. This requires
mixing of dynamic shapes with opaque types.
The general strategy is to push the actual selection of the MHLO ops
down into mlir module (e.g., mlir.slice_op, mlir.broadcast_in_dim)
so that we have one place where we pick whether we use the Dynamic
or static ops. These routines can also handle the opaque type.
This will result in a recursive
call to, e.g., mlir.slice_op, but the inner call will be using
the physical avals, which should not be opaque anymore.
While making this change I was confused by the fact that the
custom KeyTyRules in prng.py have lowerings that return multiple
MHLO ops. See https://github.com/google/jax/pull/11768#issuecomment-1342349102
and I changed the rules to return a single op.
.
jax2tf already supports many cases of shape polymorphism, e.g., those
where the shapes of all intermediates can be expressed as polynomials
in the dimension variables in the input. We want to achieve the same
same coverage, or more, while using StableHLO as the lowering format,
rather than tf.Graph.
For native serialization we will support two lowering implementations:
* one is using the growing support in JAX for dynamic shapes,
of which shape polymorphism is a special case.
This implementation is enabled with the --jax_dynamic_shapes flag.
At the moment, the JAX dynamic shapes support is still
incomplete and over 300 jax2tf shape polymorphism tests fail.
* a new one (added) here in which we form a Jaxpr using abstract
values that express dimension sizes as dimension polynomials
(as for the standard jax2tf). Then we lower the Jaxpr to StableHLO.
This implementation is enabled when --jax_dynamic_shapes is off.
With this implementation only 50 jax2tf tests fail (to be fixed
separately).
The key contribution here is to enable lowering a Jaxpr that contains
dimension polynomials in some of the intermediate shapes. Many lowering rules
already have some partial support for Jaxprs where the shapes contain
`Var`s. To the extent possible, we try to write lowering rules that should
cover both cases of dynamic shapes: Var or polynomials in shapes.
The lowering convention is that at top level we collect the sorted list
of dimension variable names in the inputs, and we store it in ModuleContext.dim_vars.
All IR functions will take N additional prefix arguments of int32 type
containing the values of the dimension variables. This is stored as
a list of `ir.Value` in `LoweringContext.dim_var_values`.
Note that the Jaxprs are not changed to have extra Vars for the dimension
variable values. An alternative implementation could work by transforming
the Jaxpr to replace dimension polynomials into Vars.
The key code pattern used in the lowering rule is::
if not core.is_constant_shape(shape): # Handles both Var, and polynomials
shape = mlir.eval_dynamic_shape(ctx, shape)
return mhlo.DynamicXXX(..., shape)
else:
return mhlo.XXX(..., shape)
with `mlir.eval_dynamic_shape` handling both cases::
def eval_dynamic_shape(ctx, shape):
if config.jax_dynamic_shapes:
# Using Var
return ... subst using ctx.axis_size_env ...
else:
# Using polynomials
return ... subst using ctx.module_context.dim_vars and ctx.dim_var_values
In order to support the above some lowering functions need to take a
LoweringContext parameter, e.g., mlir.broadcast_mhlo.
I expect that the changes here will improve the --jax_dynamic_shapes coverage
as well.
Namely, make it so that `split(key, n)[i]` equals `fold_in(key, i)`
for any key and for `0 <= i < n`.
This change affects the observed random bits for a fixed key (indirectly
through splits and folds), so here we guard it behind
`jax.config.jax_threefry_partitionable`. It's not described very well
by the flag name, but it makes for a simple way to bundle together
several random-bit-altering changes as part of the same upgrade cycle.
