rocm_jax/jax/_src/pallas/pallas_call.py
Peter Hawkins 7f4ef63cd8 Run pyupgrade --py310-plus.
Also apply manual fixes to import sorting and unused imports.
2024-06-26 16:10:18 -04:00

1048 lines
41 KiB
Python

# Copyright 2023 The JAX Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Module for calling pallas functions from JAX."""
from __future__ import annotations
from collections.abc import Callable, Sequence
from functools import partial, reduce
import itertools
from typing import Any
import jax
from jax import api_util
from jax import lax
from jax import tree_util
from jax._src import ad_util
from jax._src import checkify
from jax._src import config
from jax._src import core as jax_core
from jax._src import effects
from jax._src import linear_util as lu
from jax._src import state
from jax._src.interpreters import ad
from jax._src.interpreters import batching
from jax._src.interpreters import mlir
from jax._src.interpreters import partial_eval as pe
from jax._src.interpreters import xla
from jax._src.pallas import core as pallas_core
from jax._src.pallas.primitives import uninitialized_value
from jax._src.state import discharge as state_discharge
from jax._src.state import primitives as sp
from jax._src.util import (
safe_map,
safe_zip,
split_list,
tuple_insert,
weakref_lru_cache,
)
import jax.numpy as jnp
import numpy as np
map, unsafe_map = safe_map, map
zip, unsafe_zip = safe_zip, zip
Grid = pallas_core.Grid
GridSpec = pallas_core.GridSpec
BlockMapping = pallas_core.BlockMapping
GridMapping = pallas_core.GridMapping
BlockSpec = pallas_core.BlockSpec
BlockSpecTree = pallas_core.BlockSpecTree
NoBlockSpec = pallas_core.NoBlockSpec
no_block_spec = pallas_core.no_block_spec
pallas_call_p = jax_core.Primitive('pallas_call')
pallas_call_p.multiple_results = True
def _maybe_dynamic_slice(start_idx, block_shape, value, is_indexing):
if start_idx is None:
assert is_indexing is None
return value
assert is_indexing is not None
start_idx = tuple(jnp.asarray(s, dtype=jnp.int32) for s in start_idx)
output = lax.dynamic_slice(value, start_idx, slice_sizes=block_shape)
squeeze_dims = tuple(np.arange(len(is_indexing))[np.array(is_indexing,
dtype=np.bool_)])
return lax.squeeze(output, squeeze_dims)
def _maybe_dynamic_update_slice(start_idx, block_shape, value, update,
is_indexing):
if start_idx is None:
assert is_indexing is None
return update
assert is_indexing is not None
start_idx = tuple(jnp.asarray(s, dtype=jnp.int32) for s in start_idx)
broadcast_dims = tuple(i for i, b in enumerate(is_indexing)
if not b)
update = lax.broadcast_in_dim(update, block_shape, broadcast_dims)
assert update.shape == block_shape
return lax.dynamic_update_slice(value, update, start_idx)
def _pad_values_to_block_dimension(value,
block_shape):
"""Pads values so the shape evenly divides into block dimensions.
For example, if values has a shape of (33, 2, 5) with a block_shape of
(32, 2, 4), this function will pad the value of shape to (64, 2, 8).
Args:
value: Array to be padded.
block_shape: Block shapes to use for padding. If None, no padding will
be performed.
Returns:
A padded array.
"""
if block_shape is None:
return value
padded_shape = tuple(
((v - 1) // b + 1) * b for v, b in zip(value.shape, block_shape)
)
if padded_shape != value.shape:
pad_width = tuple((0, a-b) for a, b in zip(padded_shape, value.shape))
pad_value = uninitialized_value(shape=(), dtype=value.dtype)
value = jnp.pad(value, pad_width, constant_values=pad_value)
return value
def _initialize_scratch_vals(scratch_avals) -> tuple[jax.Array, ...]:
scratch_avals = (jax_core.raise_to_shaped(x) for x in scratch_avals)
return tuple(uninitialized_value(a.shape, a.dtype) for a in scratch_avals)
def _initialize_output_vals(
out_shapes, input_args, input_output_aliases) -> Sequence[jax.Array]:
oi_map = {v: k for k, v in input_output_aliases}
output_vals = []
for i, out_shape in enumerate(out_shapes):
if i in oi_map:
output_vals.append(input_args[oi_map[i]])
else:
output_vals.append(uninitialized_value(out_shape.shape, out_shape.dtype))
return output_vals
def _logical_to_interpret_mode_dtype(dtype):
"""Converts logical dtypes into JAX dtypes for interpret mode.
This function is used to convert device-specific dtypes that have no
corresponding equivalent in JAX/XLA into a type that can be executed
by the XLA interpreter (e.g. TPU semaphores -> int32).
"""
if (hasattr(dtype, "_rules") and
hasattr(dtype._rules, "pallas_interpret_element_aval")):
return dtype._rules.pallas_interpret_element_aval(dtype).dtype
return dtype
def _logical_aval_to_interpret_mode_aval(aval):
"""Logical to interpret mode aval conversion."""
if isinstance(aval, pallas_core.AbstractMemoryRef):
inner_aval = _logical_aval_to_interpret_mode_aval(aval.inner_aval)
return aval.update(inner_aval=inner_aval)
if isinstance(aval, jax_core.ShapedArray):
inner_dtype = _logical_to_interpret_mode_dtype(aval.dtype)
return jax_core.ShapedArray(aval.shape,
inner_dtype,
weak_type=aval.weak_type, named_shape=aval.named_shape)
return aval
def _get_next_indices(grid, indices):
next_indices = []
carry = True
for dim_size, index in reversed(list(zip(grid, indices))):
i = jnp.where(carry, index + 1, index)
carry = dim_size == i
next_indices.append(jnp.where(carry, 0, i))
return tuple(reversed(next_indices))
def _pallas_call_impl(*args, jaxpr, name, out_shapes,
interpret, debug: bool,
in_shapes,
input_output_aliases: tuple[tuple[int, int], ...],
grid_mapping: GridMapping,
compiler_params: Any):
dynamic_grid_args, args = split_list( # type: ignore
args, [grid_mapping.num_dynamic_grid_bounds]
)
if interpret:
# If we're in interpreter mode, we *scan* over the grid and eval the
# discharged jaxpr. This should reproduce exactly what compiling to Triton
# will do.
dynamic_grid_args_iter = iter(dynamic_grid_args)
grid = tuple(
a if a is not pallas_core.dynamic_grid_dim
else next(dynamic_grid_args_iter)
for a in grid_mapping.grid
)
assert next(dynamic_grid_args_iter, None) is None
with grid_mapping.trace_env():
discharged_jaxpr, consts = state_discharge.discharge_state(jaxpr, ())
if debug:
print(discharged_jaxpr)
out = _initialize_output_vals(out_shapes, args, input_output_aliases)
scalars, args = split_list(args, [grid_mapping.num_index_operands]) # type: ignore
# invars: [*scalar_prefetch, *inputs, *outputs, *scratch]
num_invars = len(jaxpr.invars)
num_inputs_outputs = (
num_invars
- grid_mapping.num_index_operands
- grid_mapping.num_scratch_operands
)
_, _, scratch_invars = split_list(
jaxpr.invars, [grid_mapping.num_index_operands, num_inputs_outputs]
)
scratch_avals = [v.aval for v in scratch_invars]
scratch_values = _initialize_scratch_vals(scratch_avals)
carry = []
for x, bm in zip(itertools.chain(args, out), grid_mapping.block_mappings):
if bm is not None and isinstance(bm.indexing_mode, pallas_core.Unblocked):
padding = bm.indexing_mode.padding
if padding is not None and any(p != (0, 0) for p in padding):
if input_output_aliases:
raise NotImplementedError("Padding with aliasing not supported.")
x = lax.pad(x, jnp.zeros((), x.dtype), [(*p, 0) for p in padding])
carry.append(x)
block_shapes_without_mapped_dims = [
None if block_mapping is None else block_mapping.block_shape
for block_mapping in grid_mapping.block_mappings
]
is_indexing_dim = [
None if bm is None else tuple(b is pallas_core.mapped for b in bm)
for bm in block_shapes_without_mapped_dims
]
block_shapes = [
None if (bm is None or iid is None)
else tuple(1 if i else b for i, b in zip(iid, bm))
for iid, bm in zip(is_indexing_dim, block_shapes_without_mapped_dims)
]
# Pad values to evenly divide into block dimensions.
# This allows interpret mode to catch errors on OOB memory accesses
# by poisoning values with NaN. It also fixes an inconstency with
# lax.dynamic_slice where if the slice goes out of bounds, it will instead
# move the start_index backwards so the slice will fit in memory.
carry = map(_pad_values_to_block_dimension, carry, block_shapes)
carry.extend(scratch_values)
num_inout = len(args) + len(out)
grid_start_indices = (jnp.int32(0),) * len(grid)
if grid:
num_iterations = reduce(jnp.multiply, grid)
else:
# Base case is always one iteration when grid is ()
num_iterations = 1
def cond(carry):
i, *_ = carry
return i < num_iterations
def body(carry):
i, loop_idx, *carry = carry
local_grid_env = tuple(
pallas_core.GridAxis(idx, b)
for dim, (idx, b) in enumerate(zip(loop_idx, grid))
if dim not in grid_mapping.mapped_dims
)
carry, scratch = split_list(carry, [num_inout])
with pallas_core.grid_env(local_grid_env):
start_indices = [
None if bm is None else bm.compute_start_indices(loop_idx, *scalars)
for bm in grid_mapping.block_mappings]
blocks = map(_maybe_dynamic_slice, start_indices, block_shapes, carry,
is_indexing_dim)
with pallas_core.grid_env(local_grid_env):
assert len(discharged_jaxpr.invars) == len(scalars) + len(blocks) + len(
scratch_values
), (
len(discharged_jaxpr.invars),
len(scalars),
len(blocks),
len(scratch_values),
)
blocks = jax.core.eval_jaxpr(discharged_jaxpr, consts, *scalars,
*blocks, *scratch)
blocks = blocks[grid_mapping.num_index_operands:]
blocks, out_scratch = split_list(blocks, [num_inout])
carry = map(_maybe_dynamic_update_slice, start_indices, block_shapes,
carry, blocks, is_indexing_dim)
return (i + 1, _get_next_indices(grid, loop_idx), *carry, *out_scratch)
(_, _, *carry) = lax.while_loop(
cond, body, (jnp.int32(0), grid_start_indices, *carry)
)
_, out, _ = split_list(carry, [len(args), len(out)])
assert len(grid_mapping.block_mappings) == len(args) + len(out)
out_block_mappings = grid_mapping.block_mappings[len(args):]
out_nopad = []
for o, bm in zip(out, out_block_mappings):
if bm is not None and isinstance(bm.indexing_mode, pallas_core.Unblocked):
padding = bm.indexing_mode.padding
if padding is not None and any(p != (0, 0) for p in padding):
if input_output_aliases:
raise NotImplementedError("Padding with aliasing not supported.")
pad_low, pad_high = zip(*padding)
limit_indices = [s - p for s, p in zip(o.shape, pad_high)]
o = lax.slice(o, pad_low, limit_indices)
out_nopad.append(o)
return out_nopad
return xla.apply_primitive(pallas_call_p, *args, jaxpr=jaxpr, name=name,
in_shapes=in_shapes,
out_shapes=out_shapes,
grid_mapping=grid_mapping, interpret=interpret,
debug=debug,
input_output_aliases=input_output_aliases,
compiler_params=compiler_params)
pallas_call_p.def_impl(_pallas_call_impl)
def _pallas_call_abstract_eval(*avals, out_shapes, **_):
return map(lambda x: jax_core.ShapedArray(x.shape, x.dtype), out_shapes)
pallas_call_p.def_abstract_eval(_pallas_call_abstract_eval)
def _pallas_call_jvp_rule(primals, tangents, *, jaxpr, name,
input_output_aliases: tuple[tuple[int, int], ...],
in_shapes, out_shapes, grid_mapping, debug, interpret, compiler_params: Any):
if grid_mapping.num_dynamic_grid_bounds:
raise NotImplementedError("interpret with dynamic grid bounds unsupported")
if grid_mapping.num_index_operands:
raise NotImplementedError
if input_output_aliases:
raise NotImplementedError("JVP with aliasing not supported.")
nonzero_tangents = [not isinstance(t, ad_util.Zero) for t in tangents]
tangents = [t for t in tangents if type(t) is not ad_util.Zero]
nonzero_tangents_with_outputs = nonzero_tangents + [True] * len(out_shapes)
closed_jaxpr = jax_core.ClosedJaxpr(jaxpr, ())
jvp_jaxpr_, _ = ad.jvp_jaxpr(closed_jaxpr, nonzero_tangents_with_outputs, [])
jvp_jaxpr, () = jvp_jaxpr_.jaxpr, jvp_jaxpr_.consts # TODO consts
# `pallas_call` takes in inputs and returns outputs but its jaxpr *does not*.
# `pallas_call` takes in a stateful jaxpr, meaning the jaxpr accepts input
# `Ref`s that are read from followed by output `Ref`s that are written to.
# This means that when we do `jvp_jaxpr` on the `jaxpr`, we get out a new
# jaxpr that has tangents following primals. In order for this jaxpr to be
# compatible w/ `pallas_call` (inputs then outputs), we need to shuffle around
# the jaxpr's invars.
primal_refs, primal_out_refs, tangent_refs, tangent_out_refs = split_list(
jvp_jaxpr.invars, [len(primals), len(out_shapes), len(tangents)]
)
invars = (*primal_refs, *tangent_refs, *primal_out_refs, *tangent_out_refs)
effs = []
for eff in jvp_jaxpr.effects:
if isinstance(eff, effects.JaxprInputEffect):
eff = eff.replace(
input_index=invars.index(jvp_jaxpr.invars[eff.input_index])
)
effs.append(eff)
jvp_jaxpr = jvp_jaxpr.replace(invars=invars, effects=effs)
if debug:
print(jvp_jaxpr)
in_bms, out_bms = split_list(grid_mapping.block_mappings, [len(primals)])
jvp_bms = (*in_bms, *in_bms, *out_bms, *out_bms)
out_flat = pallas_call_p.bind(
*primals,
*tangents,
jaxpr=jvp_jaxpr,
name=f"{name}_jvp",
in_shapes=(*in_shapes, *in_shapes),
out_shapes=(*out_shapes, *out_shapes),
grid_mapping=grid_mapping.replace(block_mappings=jvp_bms),
interpret=interpret,
debug=debug,
input_output_aliases=(),
compiler_params=compiler_params,
)
out_primals, out_tangents = split_list(out_flat, [len(out_flat) // 2])
return out_primals, out_tangents
ad.primitive_jvps[pallas_call_p] = _pallas_call_jvp_rule
def _batch_block_mapping(grid_mapping: GridMapping, aval: jax_core.ShapedArray,
dim: int | batching.NotMapped,
block_mapping: BlockMapping | None) -> BlockMapping:
def _block_map_function(new_idx, *args):
if block_mapping is None:
indices = [0] * len(aval.shape)
else:
indices = jax_core.eval_jaxpr(block_mapping.index_map_jaxpr.jaxpr,
block_mapping.index_map_jaxpr.consts,
*args)
if dim is not batching.not_mapped:
indices.insert(dim, new_idx)
return tuple(indices)
i32_aval = jax_core.ShapedArray((), jnp.int32)
if block_mapping is None:
idx_avals = [i32_aval] * (len(grid_mapping.grid) + 1)
else:
idx_avals = [i32_aval, *block_mapping.index_map_jaxpr.in_avals]
with grid_mapping.trace_env():
block_mapping_jaxpr, _, consts, () = pe.trace_to_jaxpr_dynamic(
lu.wrap_init(_block_map_function), idx_avals)
shape = aval.shape if block_mapping is None else block_mapping.block_shape
if dim is batching.not_mapped:
new_block_shape = shape
else:
new_block_shape = tuple_insert(shape, dim, pallas_core.mapped)
jaxpr = jax_core.ClosedJaxpr(block_mapping_jaxpr, consts)
if block_mapping is None:
return BlockMapping(
block_shape=new_block_shape,
index_map_jaxpr=jaxpr,
indexing_mode=pallas_core.blocked,
)
return block_mapping.replace(block_shape=new_block_shape,
index_map_jaxpr=jaxpr)
def _broadcast_input_output_aliases(
args: Sequence[jax.Array],
dims: Sequence[int | batching.NotMapped],
*,
input_output_aliases: tuple[tuple[int, int], ...],
axis_size: int,
) -> tuple[tuple[jax.Array, ...], tuple[int | batching.NotMapped, ...]]:
"""Broadcast input/output operands.
When we have input/output aliasing, since the output will be mapped, we need
to make sure to broadcast the input across that dimension if it is not
mapped. If the input is mapped, but on a different axis, we tranpose the input
to match the output.
"""
args_ = list(args)
dims_ = list(dims)
for input_index, _ in input_output_aliases:
dim = dims_[input_index]
dims_[input_index] = 0
if dim is batching.not_mapped:
args_[input_index] = batching.broadcast(args_[input_index], axis_size, 0)
elif dim != 0:
# TODO(cjfj): Change output batching axis instead?
args_[input_index] = jnp.moveaxis(args[input_index], dim, 0)
return tuple(args_), tuple(dims_)
def _batch_with_explicit_loop(
args: Sequence[jax.Array],
dims: Sequence[int | batching.NotMapped],
*,
jaxpr: jax_core.Jaxpr,
name: str,
in_shapes: tuple[jax.ShapeDtypeStruct, ...],
out_shapes: tuple[jax.ShapeDtypeStruct, ...],
grid_mapping: GridMapping,
input_output_aliases: tuple[tuple[int, int], ...],
debug: bool,
interpret: bool,
compiler_params: Any,
):
"""Batch the pallas_call by calling it in loop over the batch size.
This function provides a fallback implementation of batching a pallas_call
for the cases in which adding a batch dimension to the pallas grid is not
supported. This is currently the case when the batched dimension corresponds
to a dynamic axis or a scalar prefetch argument.
This implementation builds a HLO loop that dynamic_slices the inputs according
to the current iteration index and dynamic_updates an (initially empty) output
allocation.
"""
if not dims:
raise NotImplementedError("vmapping pallas_call with no arguments.")
(axis_size,) = {
arg.shape[dim]
for arg, dim in zip(args, dims)
if dim is not batching.not_mapped
}
args, dims = _broadcast_input_output_aliases(
args,
dims,
input_output_aliases=input_output_aliases,
axis_size=axis_size,
)
# The output arrays are completelly overwritten, so we can just initialize
# empty arrays.
initial_state = [
jnp.empty(
tuple_insert(out_shape.shape, 0, axis_size), dtype=out_shape.dtype
)
for out_shape in out_shapes
]
def body(batch_index: jax.Array, state: list[jax.Array]) -> list[jax.Array]:
batch_args = []
for arg, dim in zip(args, dims):
# If the argument is mapped, extract a slice of size 1 in the mapped
# dimension at the current index.
if dim is batching.not_mapped:
batch_args.append(arg)
else:
batch_args.append(
jnp.squeeze(
jax.lax.dynamic_slice_in_dim(
operand=arg,
start_index=batch_index,
slice_size=1,
axis=dim,
),
axis=dim,
)
)
batch_out = pallas_call_p.bind(
*batch_args,
jaxpr=jaxpr,
name=name,
in_shapes=in_shapes,
out_shapes=out_shapes,
grid_mapping=grid_mapping,
input_output_aliases=input_output_aliases,
debug=debug,
interpret=interpret,
compiler_params=compiler_params,
)
for i, batch_out_array in enumerate(batch_out):
state[i] = jax.lax.dynamic_update_index_in_dim(
state[i],
batch_out_array,
batch_index,
axis=0,
)
return state
result = jax.lax.fori_loop(0, axis_size, body, initial_state, unroll=False)
return result, (0,) * len(result)
def _pallas_call_batching_rule(
args,
dims,
*,
jaxpr: jax_core.Jaxpr,
name: str,
in_shapes: tuple[jax.ShapeDtypeStruct, ...],
out_shapes: tuple[jax.ShapeDtypeStruct, ...],
grid_mapping: GridMapping,
input_output_aliases: tuple[tuple[int, int], ...],
debug: bool,
interpret: bool,
compiler_params: Any,
):
def _maybe_squeeze_out_bdim(
x: jax.Array, bdim: int | batching.NotMapped
) -> jax.Array:
if bdim is batching.not_mapped:
return x
return jnp.squeeze(x, axis=bdim)
axis_size, = {x.shape[d] for x, d in zip(args, dims)
if d is not batching.not_mapped}
if axis_size == 1:
# Why are we even vmapping?
args = map(_maybe_squeeze_out_bdim, args, dims)
out = pallas_call_p.bind(
*args,
jaxpr=jaxpr,
name=name,
in_shapes=in_shapes,
out_shapes=out_shapes,
grid_mapping=grid_mapping,
input_output_aliases=input_output_aliases,
debug=debug,
interpret=interpret,
compiler_params=compiler_params,
)
return [jnp.expand_dims(x, 0) for x in out], (0,) * len(out)
# The first num_dynamic_grid_bounds arguments are size-1 arrays that store
# the size of the dynamic bounds.
dynamic_grid_args, args = split_list(
args, [grid_mapping.num_dynamic_grid_bounds]
)
dynamic_grid_dims, dims = split_list(
dims, [grid_mapping.num_dynamic_grid_bounds]
)
if all(
bdim is batching.not_mapped or arg.shape[bdim] == 1
for arg, bdim in zip(dynamic_grid_args, dynamic_grid_dims)
):
dynamic_grid_args = safe_map(
_maybe_squeeze_out_bdim, dynamic_grid_args, dynamic_grid_dims
)
elif any(bdim is not batching.not_mapped for bdim in dynamic_grid_dims):
# TODO(amagni, sharadmv): Explore possibility of batching dynamic grid
# bounds.
return _batch_with_explicit_loop(
args=dynamic_grid_args + args,
dims=dynamic_grid_dims + dims,
jaxpr=jaxpr,
name=name,
in_shapes=in_shapes,
out_shapes=out_shapes,
grid_mapping=grid_mapping,
input_output_aliases=input_output_aliases,
debug=debug,
interpret=interpret,
compiler_params=compiler_params,
)
else:
pass # No dynamic grid dimensions
del dynamic_grid_dims
if grid_mapping.num_index_operands:
scalar_args, args = split_list(args, [grid_mapping.num_index_operands])
scalar_bdims, bdims = split_list(dims, [grid_mapping.num_index_operands])
# Ordinarily, adding support for scalar prefetch in vmap would involve
# modifying the block specs in a nontrivial way. However, if we are only
# vmapping over 1-sized dimensions, we can just get rid of the dimensions
# and pretend we were never vmapping over them at all.
if all(
bdim is batching.not_mapped or arg.shape[bdim] == 1
for arg, bdim in zip(scalar_args, scalar_bdims)
):
scalar_args = safe_map(_maybe_squeeze_out_bdim, scalar_args, scalar_bdims)
scalar_bdims = [None] * len(scalar_args)
args = (*scalar_args, *args)
dims = (*scalar_bdims, *bdims)
else:
# TODO(amagni,sharadmv,apaszke): enable efficient batching over
# prefetched scalar args.
return _batch_with_explicit_loop(
args=scalar_args + args,
dims=scalar_bdims + bdims,
jaxpr=jaxpr,
name=name,
in_shapes=in_shapes,
out_shapes=out_shapes,
grid_mapping=grid_mapping,
input_output_aliases=input_output_aliases,
debug=debug,
interpret=interpret,
compiler_params=compiler_params,
)
if not dims:
raise NotImplementedError("vmapping pallas_call with no arguments.")
block_mappings = grid_mapping.block_mappings
avals = [v.aval for v in jaxpr.invars]
# How should we pick output dimensions? This actually matters because XLA
# can't optimize our pallas kernels, and this layout impacts performance. For
# now, because `vmap` doesn't really offer a way of inferring good output
# dimensions. For now, we just use 0.
# TODO(sharadmv): explore inferring better output dimensions via a heuristic
# TODO(sharadmv): explore a long term solution to output dim inference
args, dims = _broadcast_input_output_aliases(
args, dims, input_output_aliases=input_output_aliases, axis_size=axis_size
)
all_dims = list(dims) + [0] * len(out_shapes)
num_index_operands = grid_mapping.num_index_operands
num_scratch_operands = grid_mapping.num_scratch_operands
# Only add a batch dimension for the avals that actually have a grid mapping.
# This excludes scalar prefetch inputs (the first in the list) and scratch
# operands (the last in the list).
avals_to_batch = avals[num_index_operands:(len(avals) - num_scratch_operands)]
batched_block_mappings = map(
partial(_batch_block_mapping, grid_mapping),
avals_to_batch,
all_dims[num_index_operands:],
block_mappings,
)
batched_in_shapes = tuple(
jax.ShapeDtypeStruct(x.shape if dim is batching.not_mapped else
tuple_insert(x.shape, dim, axis_size),
x.dtype)
for x, dim in zip(in_shapes, dims))
batched_out_shapes = tuple(
jax.ShapeDtypeStruct(tuple_insert(x.shape, 0, axis_size), x.dtype)
for x in out_shapes)
batched_grid_mapping = grid_mapping.replace(
grid=(axis_size, *grid_mapping.grid),
block_mappings=tuple(batched_block_mappings),
mapped_dims=(0,) + tuple(a + 1 for a in grid_mapping.mapped_dims))
out = pallas_call_p.bind(
*dynamic_grid_args,
*args,
jaxpr=jaxpr,
name=f"batched_{name}",
in_shapes=batched_in_shapes,
out_shapes=batched_out_shapes,
grid_mapping=batched_grid_mapping,
input_output_aliases=input_output_aliases,
debug=debug,
interpret=interpret,
compiler_params=compiler_params,
)
return out, (0,) * len(out)
batching.primitive_batchers[pallas_call_p] = _pallas_call_batching_rule
def _hoist_consts_to_refs(jaxpr: jax_core.Jaxpr) -> jax_core.Jaxpr:
"""Hoists the constants in the given jaxpr into invars.
Args:
jaxpr: The jaxpr.
Returns:
A new jaxpr where the constants were hoisted into invars as ``Ref``s.
The invars for the constants are added *before* any existing invars.
"""
if not jaxpr.constvars:
return jaxpr # Nothing to hoist.
is_const_ref = [
isinstance(var.aval, state.AbstractRef) for var in jaxpr.constvars
]
const_avals = [
var.aval if is_ref else state.AbstractRef(var.aval)
for is_ref, var in zip(is_const_ref, jaxpr.constvars)
]
in_avals = const_avals + [var.aval for var in jaxpr.invars]
def _hoist(*consts_args):
all_consts, args = split_list(consts_args, [len(const_avals)])
# We immediately read the const values out of the `Ref`s.
all_consts = [
c if is_ref else sp.ref_get(c, ())
for is_ref, c in zip(is_const_ref, all_consts)
]
return jax_core.eval_jaxpr(jaxpr, all_consts, *args)
hoisted_jaxpr, _, consts, () = pe.trace_to_jaxpr_dynamic(
lu.wrap_init(_hoist), in_avals)
assert not consts, "All consts should have been converted to refs"
return hoisted_jaxpr
def checkify_pallas_kernel_body_jaxpr(
body_jaxpr: jax_core.ClosedJaxpr,
enabled_errors,
error: checkify.Error,
grid_mapping: GridMapping) -> tuple[
jax_core.ClosedJaxpr, tree_util.PyTreeDef, set[checkify.ErrorEffect]]:
err_vals, err_tree = tree_util.tree_flatten(error)
err_vals = map(checkify.get_shaped_aval, err_vals)
flat_err_and_in_vals = [*err_vals, *body_jaxpr.in_avals]
with pallas_core.tracing_grid_env(grid_mapping.grid, ()):
checked_jaxpr, out_tree, error_effects = checkify.jaxpr_to_checkify_jaxpr(
body_jaxpr, enabled_errors, err_tree, *flat_err_and_in_vals)
return checked_jaxpr, out_tree, error_effects
def pallas_call_checkify_rule(error: checkify.Error,
enabled_errors,
*args: jax_core.Value,
jaxpr: jax_core.Jaxpr,
interpret: bool,
input_output_aliases: tuple[tuple[int, int], ...],
grid_mapping: GridMapping,
out_shapes,
**kwargs):
# TODO(b/346651778): Support TPU/GPU checkify.
if not interpret:
raise NotImplementedError(
"Checkify for pallas_call only supports interpret mode.")
# We implement the checkify rule in 4 steps:
# 1) First, trace the kernel body to get the expected error shapes.
# 2) Checkify the kernel body to obtain a jaxpr with errors as inputs
# and outputs.
# 3) Create a new kernel which stores the errors in output memrefs instead of
# returning them, since pallas kernels do not return outputs.
# 4) Create block specs for the error state and call pallas_call with
# the new kernel.
dynamic_grid_bounds, scalars, args = split_list( # type: ignore
args, [grid_mapping.num_dynamic_grid_bounds, grid_mapping.num_index_operands]
)
num_scalars = len(scalars)
num_invars = len(jaxpr.invars)
num_inputs_outputs = (
num_invars
- grid_mapping.num_index_operands
- grid_mapping.num_scratch_operands
)
num_kernel_inputs = len(args)
num_scratch = num_invars - num_inputs_outputs
num_kernel_outputs = num_invars - num_scratch - num_kernel_inputs
# Trace the jaxpr to get an initial error value so the kernel jaxpr has all of
# the required inputs.
closed_jaxpr = pe.close_jaxpr(jaxpr)
_jaxpr, _, error_effects = checkify_pallas_kernel_body_jaxpr(
closed_jaxpr, enabled_errors, error, grid_mapping)
error = error._add_placeholder_effects(error_effects)
err_vals, err_tree = jax.tree.flatten(error)
shaped_err_avals = map(checkify.get_shaped_aval, err_vals)
# Trace the kernel jaxpr to get a checkified jaxpr. This jaxpr will have
# all enabled errors removed, but have the error as inputs and return values.
input_avals = [v.aval for v in jaxpr.invars]
num_err_vals = len(err_vals)
shaped_input_avals = tuple(jax_core.raise_to_shaped(x) for x in input_avals)
checkify_in_avals = [*shaped_err_avals,
*shaped_input_avals]
closed_kernel_jaxpr = pe.close_jaxpr(jaxpr)
with pallas_core.tracing_grid_env(grid_mapping.grid, ()):
checked_jaxpr, out_tree, _ = checkify.jaxpr_to_checkify_jaxpr(
closed_kernel_jaxpr, enabled_errors, err_tree, *checkify_in_avals)
# Create a new kernel to remove the error as an return value and instead
# write them to a memref. This is because pallas kernels are expected
# to have no return values but instead write their outputs to a ref.
def checked_kernel_fn(*args):
(scalars, _, inputs, out_error_refs, outputs, scratch
) = split_list(
args,
[num_scalars, num_err_vals,
num_kernel_inputs, num_err_vals, num_kernel_outputs])
input_error_vals = [err_ref[...] for err_ref in out_error_refs]
# We need to re-order the inputs here. A checkified jaxpr always expects
# errors before other arguments.
jaxpr_args = [*input_error_vals, *scalars, *inputs, *outputs, *scratch]
assert len(checked_jaxpr.jaxpr.invars) == len(jaxpr_args)
result_flat = jax.core.eval_jaxpr(
checked_jaxpr.jaxpr, checked_jaxpr.consts, *jaxpr_args)
output_errors, _ = split_list(result_flat, [num_err_vals])
# Store new errors back in the error refs.
for out_ref, error in zip(out_error_refs, output_errors):
out_ref[...] = error
return []
# Trace the new checked_kernel_fn with Memref inputs so that
# we can replace the old kernel jaxpr with the new checked jaxpr in
# pallas_call.
# TODO(justinfu): Place errors in scalar memory for non-interpret mode.
error_mem_space = None
error_memref_aval = [pallas_core.AbstractMemoryRef(
err_val, error_mem_space) for err_val in shaped_err_avals]
shaped_scalar_avals, input_aval, output_aval, scratch_aval = split_list(
shaped_input_avals, [num_scalars, num_kernel_inputs, num_kernel_outputs])
retrace_in_avals = [*shaped_scalar_avals, *error_memref_aval, *input_aval,
*error_memref_aval, *output_aval, *scratch_aval]
jaxpr_flat_avals, jaxpr_in_tree = tree_util.tree_flatten(retrace_in_avals)
wrapped_kernel_with_err, out_tree_thunk = api_util.flatten_fun_nokwargs(
lu.wrap_init(checked_kernel_fn), jaxpr_in_tree)
debug = pe.debug_info(
checked_kernel_fn, jaxpr_in_tree, out_tree_thunk, False, "checkify_pallas")
with pallas_core.tracing_grid_env(grid_mapping.grid, ()):
final_jaxpr, _, _, () = pe.trace_to_jaxpr_dynamic(
wrapped_kernel_with_err, jaxpr_flat_avals, debug)
# Prepare pallas_call inputs. We need to create new block specs
# for the new error inputs and outputs.
scalar_avals = map(checkify.get_shaped_aval, scalars)
error_block_specs = [no_block_spec] * num_err_vals
grid_avals = [
jax_core.ShapedArray((), jnp.dtype("int32"))] * len(grid_mapping.grid)
# TODO(justinfu): Place these in device-specific scalar memory.
scalar_ref_avals = [
pallas_core.AbstractMemoryRef(
jax_core.ShapedArray(aval.shape, aval.dtype), None)
for aval in scalar_avals]
grid_tree = tree_util.tree_structure(((*grid_avals, *scalar_avals), {}))
error_block_mappings = map(
partial(
pallas_core._convert_block_spec_to_block_mapping,
(*grid_avals, *scalar_ref_avals),
in_tree=grid_tree,
grid=grid_mapping.grid,
mapped_dims=grid_mapping.mapped_dims),
error_block_specs, error_memref_aval)
input_block_mappings, output_block_mappings = split_list(
grid_mapping.block_mappings, [num_kernel_inputs,])
grid_mapping_with_error = grid_mapping.replace(
block_mappings=(*error_block_mappings, *input_block_mappings,
*error_block_mappings, *output_block_mappings)
)
error_out_shapes = tuple(
jax.ShapeDtypeStruct(e.shape, e.dtype) for e in shaped_err_avals)
# Bump all input_output_aliases by num_err_vals to make room for error
# TODO(justinfu): Don't bump scalars here.
input_output_aliases = tuple(
(i+num_err_vals, o+num_err_vals) for (i, o) in input_output_aliases)
input_output_aliases_with_error = tuple(
(i+num_scalars, i) for i in range(num_err_vals)) + input_output_aliases
new_vals_in = [*scalars, *err_vals, *args]
result = pallas_call_p.bind(*dynamic_grid_bounds, *new_vals_in,
jaxpr=final_jaxpr,
interpret=interpret,
grid_mapping=grid_mapping_with_error,
input_output_aliases=input_output_aliases_with_error,
out_shapes=error_out_shapes + out_shapes,
**kwargs)
errors, results = split_list(result, [num_err_vals])
new_error, _ = jax.tree.unflatten(out_tree, errors)
return new_error, results
checkify.error_checks[pallas_call_p] = pallas_call_checkify_rule
@weakref_lru_cache
def _trace_to_jaxpr(fun: Callable, grid_spec: GridSpec, flat_in_avals,
flat_out_avals, in_tree, out_tree, interpret: bool):
avals, grid_mapping = grid_spec.get_grid_mapping(flat_in_avals, in_tree,
flat_out_avals, out_tree)
if interpret:
avals = jax.tree_util.tree_map(_logical_aval_to_interpret_mode_aval, avals)
jaxpr_flat_avals, jaxpr_in_tree = tree_util.tree_flatten(avals)
wrapped_fun, out_tree_thunk = api_util.flatten_fun_nokwargs(
lu.wrap_init(fun), jaxpr_in_tree)
debug = pe.debug_info(fun, jaxpr_in_tree, out_tree_thunk, False, "pallas_call")
with pallas_core.tracing_grid_env(grid_mapping.grid, ()):
jaxpr, _, consts, () = pe.trace_to_jaxpr_dynamic(wrapped_fun,
jaxpr_flat_avals, debug)
if consts:
jaxpr = _hoist_consts_to_refs(jaxpr)
# Pad ``block_mappings`` to account for the hoisted constants.
grid_mapping = grid_mapping.replace(
block_mappings=(*grid_mapping.block_mappings, *[None] * len(consts)),
num_constant_operands=len(consts),
)
return grid_mapping, jaxpr, consts, out_tree_thunk()
def _extract_function_name(f: Callable, name: str | None) -> str:
if name is None:
name = f.__name__ if hasattr(f, "__name__") and f.__name__ else "func"
return name
_PALLAS_USE_MOSAIC_GPU = config.bool_flag(
"jax_pallas_use_mosaic_gpu",
default=config.bool_env("JAX_PALLAS_USE_MOSAIC_GPU", False),
help=(
"If True, lower Pallas kernels to the experimental Mosaic GPU"
" dialect, instead of Trition IR."
),
)
def _unsupported_lowering_error(platform: str) -> Exception:
return ValueError(
f"Cannot lower pallas_call on platform: {platform}. To use Pallas on GPU,"
" install jaxlib GPU 0.4.24 or newer. To use Pallas on TPU, install"
" jaxlib TPU and libtpu. See"
" https://jax.readthedocs.io/en/latest/installation.html."
)
def _pallas_call_lowering(
ctx: mlir.LoweringRuleContext, *in_nodes, interpret: bool, **params
):
if interpret:
# If we are in interpret mode, we don't care what platform we are on.
impl = partial(_pallas_call_impl, **params, interpret=True)
return mlir.lower_fun(impl, multiple_results=True)(ctx, *in_nodes)
def cpu_lowering(ctx: mlir.LoweringRuleContext,
*in_nodes: mlir.ir.Value | Sequence[mlir.ir.Value],
**params):
raise ValueError("Only interpret mode is supported on CPU backend.")
def tpu_lowering(ctx: mlir.LoweringRuleContext,
*in_nodes: mlir.ir.Value | Sequence[mlir.ir.Value],
**params):
try:
from jax._src.pallas.mosaic import pallas_call_registration
except ImportError:
raise _unsupported_lowering_error("tpu")
else:
return pallas_call_registration.pallas_call_tpu_lowering_rule(
ctx, *in_nodes, **params
)
def gpu_lowering(ctx: mlir.LoweringRuleContext,
*in_nodes: mlir.ir.Value | Sequence[mlir.ir.Value],
**params):
try:
if _PALLAS_USE_MOSAIC_GPU.value:
from jax._src.pallas.mosaic_gpu import pallas_call_registration
else:
from jax._src.pallas.triton import pallas_call_registration # type: ignore
except ImportError:
raise _unsupported_lowering_error("gpu")
else:
return pallas_call_registration.pallas_call_lowering(
ctx, *in_nodes, **params
)
return mlir.lower_per_platform(ctx, "pallas_call",
dict(cpu=cpu_lowering,
tpu=tpu_lowering,
cuda=gpu_lowering,
rocm=gpu_lowering),
None, # default_rule
effects.no_effects,
*in_nodes,
interpret=interpret,
**params)
mlir.register_lowering(pallas_call_p, _pallas_call_lowering)
def pallas_call(
f: Callable[..., None],
out_shape: Any,
*,
grid_spec: GridSpec | None = None,
debug: bool = False,
grid: Grid | None = None,
in_specs: BlockSpecTree = no_block_spec,
out_specs: BlockSpecTree = no_block_spec,
input_output_aliases: dict[int, int] = {},
interpret: bool = False,
name: str | None = None,
compiler_params: dict[str, Any] | None = None,
) -> Callable[..., Any]:
name = _extract_function_name(f, name)
if compiler_params is None:
compiler_params = {}
if grid is not None and grid_spec is not None:
raise ValueError("Cannot specify both grid and grid_spec at the same time.")
if grid_spec is None:
grid_spec = GridSpec(grid, in_specs, out_specs)
grid_spec, dynamic_grid_bounds = grid_spec.unzip_dynamic_grid_bounds()
if isinstance(out_shape, list):
out_shape = tuple(out_shape)
flat_out_shapes, out_tree = tree_util.tree_flatten(out_shape)
flat_out_shapes = [jax.ShapeDtypeStruct(x.shape, x.dtype)
for x in flat_out_shapes]
@jax.jit
def wrapped(*args):
flat_args, in_tree = tree_util.tree_flatten(args)
flat_in_avals = tuple(jax_core.raise_to_shaped(jax_core.get_aval(a))
for a in flat_args)
flat_out_avals = tuple(jax_core.ShapedArray(v.shape, v.dtype)
for v in flat_out_shapes)
grid_mapping, jaxpr, consts, _ = _trace_to_jaxpr(
f, grid_spec, flat_in_avals, flat_out_avals, in_tree,
out_tree, interpret=interpret)
out_flat = pallas_call_p.bind(
*dynamic_grid_bounds, *consts, *flat_args,
jaxpr=jaxpr, name=name,
in_shapes=tuple(jax.ShapeDtypeStruct(a.shape, a.dtype)
for a in flat_args),
out_shapes=tuple(flat_out_shapes), debug=debug,
interpret=interpret,
grid_mapping=grid_mapping,
input_output_aliases=tuple(input_output_aliases.items()),
compiler_params=compiler_params)
out = tree_util.tree_unflatten(out_tree, out_flat)
return out
return wrapped