Tianqi Chen
275114b327
## Summary - Make `PrimExpr` a typed C++ view over `Expr` values whose `ExprNode::ty` is `PrimType`, instead of using a separate runtime node class as the proof of primitive-ness. - Use the shared `ir::Call` node for Relax, TIRX, and primitive-valued calls, while keeping primitive-only APIs explicit at their semantic boundaries. - Keep Python on the general `Expr` surface for primitive-typed values so `isinstance` behavior does not imply a nominal primitive-expression subclass. ## Design Rationale The main advantage of this change is that common expression nodes such as `Call` can be unified without specializing each one to `PrimType`. A single `ir::Call` can represent a Relax tensor call, a Relax scalar call, or a primitive-valued intrinsic call; the result type stored in `ExprNode::ty` determines whether that particular value can be viewed as `PrimExpr`. This keeps the IR node hierarchy focused on expression structure rather than result-type categories. Nodes that are intrinsically primitive, such as integer and floating-point literals or TIRX primitive operators, still have strongly typed C++ APIs and data structures. General nodes whose result type may vary, such as `Call`, remain general `Expr` nodes and are narrowed to `PrimExpr` only where primitive-only semantics are required. The PR also keeps the compatibility surface practical: C++ primitive-only APIs continue to accept `PrimExpr`, Python exposes a compatibility predicate for checking the primitive typed category, and visitors/printers use one natural `Call` path rather than duplicating Relax and primitive call handling. Missing expression types are represented explicitly with `Type::Missing()` so constructors can leave type inference to later analysis without relying on nullable `Type` values.
107 lines
3.5 KiB
Python
107 lines
3.5 KiB
Python
# Licensed to the Apache Software Foundation (ASF) under one
|
|
# or more contributor license agreements. See the NOTICE file
|
|
# distributed with this work for additional information
|
|
# regarding copyright ownership. The ASF licenses this file
|
|
# to you 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
|
|
#
|
|
# http://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.
|
|
# pylint: disable=redefined-builtin,missing-docstring, invalid-name, unused-import, redefined-outer-name
|
|
# ruff: noqa: F401
|
|
|
|
from typing import Any, Optional, Union
|
|
|
|
from tvm.ir import Range
|
|
from tvm.relax import TensorType
|
|
from tvm.relax.distributed import DeviceMesh, DTensorType, Placement, device_mesh
|
|
from tvm.relax.script.builder.distributed import (
|
|
annotate_sharding,
|
|
call_tir,
|
|
call_tir_local_view,
|
|
const,
|
|
redistribute,
|
|
redistribute_replica_to_shard,
|
|
)
|
|
from tvm.script.ir_builder import IRBuilder
|
|
from tvm.script.ir_builder.ir import IRModuleFrame
|
|
from tvm.tirx import Expr
|
|
|
|
from .entry import TensorProxy, TypeProxy
|
|
|
|
############################### R.DTensor ###############################
|
|
|
|
|
|
class DTensorProxy(TypeProxy):
|
|
tensor_ty_proxy: TensorProxy
|
|
device_mesh: DeviceMesh
|
|
placement: Placement
|
|
|
|
def __init__(
|
|
self,
|
|
tensor_ty_proxy: TensorProxy,
|
|
device_mesh: DeviceMesh,
|
|
placement: Placement,
|
|
) -> None:
|
|
self.device_mesh = device_mesh
|
|
self.placement = placement
|
|
self.tensor_ty_proxy = tensor_ty_proxy
|
|
super().__init__()
|
|
|
|
def get_symbolic_vars(self) -> set[str]:
|
|
return self.tensor_ty_proxy.get_symbolic_vars()
|
|
|
|
def as_ty(self, dict_globals: dict[str, Any] | None = None) -> DTensorType:
|
|
return DTensorType(
|
|
self.tensor_ty_proxy.as_ty(dict_globals),
|
|
self.device_mesh,
|
|
self.placement,
|
|
)
|
|
|
|
|
|
def DTensor(
|
|
shape: list[Expr | str] | None = None,
|
|
dtype: str | None = None,
|
|
device_mesh: DeviceMesh | str = DeviceMesh([], Range(0, 1)),
|
|
placement: Placement | str = "",
|
|
*,
|
|
ndim: int = -1,
|
|
) -> DTensorProxy:
|
|
# scalar tensor case
|
|
if shape is not None and len(shape) == 0:
|
|
shape = []
|
|
if isinstance(shape, str) and dtype is None:
|
|
dtype = shape
|
|
shape = None
|
|
|
|
if shape is not None and not isinstance(shape, tuple | list):
|
|
raise ValueError(f"shape must be a list or tuple, but got: {shape}")
|
|
if isinstance(device_mesh, str):
|
|
if not IRBuilder.is_in_scope():
|
|
return (
|
|
DTensorProxy(
|
|
TensorProxy(shape, dtype, None, ndim), DeviceMesh([], Range(0, 1)), ""
|
|
),
|
|
)
|
|
name, index = device_mesh.split("[")
|
|
index = int(index[:-1])
|
|
frames = IRBuilder.current().frames
|
|
for f in frames:
|
|
if isinstance(f, IRModuleFrame):
|
|
device_mesh = f.global_infos[name][index]
|
|
break
|
|
assert isinstance(device_mesh, DeviceMesh)
|
|
if isinstance(placement, str):
|
|
placement = Placement.from_text(placement)
|
|
return DTensorProxy(TensorProxy(shape, dtype, None, ndim), device_mesh, placement)
|
|
|
|
|
|
__all__ = ["DTensor", "device_mesh"]
|