This PR adds "downcasting" of `ir.Value` to either `BlockArgument` or
`OpResult` (and then potentially further down if a user-registered
"value caster" exists). Also this PR changes `__str__` to return the
correct thing (`OpResult(...)` or `BlockArgument(...)` instead of
generic `Value(...)`).
# What
This PR adds a shared library `MLIRPythonSupport` which contains all of
the CRTP classes ike `PyConcreteValue`, `PyConcreteType`,
`PyConcreteAttribute`, as well as other useful code like `Defaulting*`
and etc enabling their reuse in downstream projects. Downstream projects
can now do
```c++
struct PyTestType : mlir::python::MLIR_BINDINGS_PYTHON_DOMAIN::PyConcreteType<PyTestType> {
...
};
class PyTestAttr : public mlir::python::MLIR_BINDINGS_PYTHON_DOMAIN::PyConcreteAttribute<PyTestAttr> {
...
}
NB_MODULE(_mlirPythonTestNanobind, m) {
PyTestType::bind(m);
PyTestAttr::bind(m);
}
```
instead of using the discordant alternative
`mlir_type_subclass`/`mlir_attr_subclass` (same goes for
`PyConcreteValue`/`mlir_value_subclass`).
# Why
This PR is mostly code motion (along with CMake) but before I describe
the changes I want to state the goals/benefits:
1. Currently upstream "core" extensions and "dialect" extensions ([all
of the `Dialect*` extensions
here](d7c734b5a1/mlir/lib/Bindings/Python))
are a two-tier system;
**a**. [core
extensions](https://github.com/llvm/llvm-project/blob/main/mlir/lib/Bindings/Python/IRTypes.cpp#L361)
enjoy first class support as far as type inference[^3], type stub
generation, and ease of implementation, while dialect extensions [have
poorer support](https://reviews.llvm.org/D150927), incorrect type stub
generation much more tedious (boilerplate) implementation;
**b**. Crucially, this two-tiered system is reflected in the fact that
**the two sets of types/attributes are not in the same Python object
hierarchy**. To wit: `isinstance(..., Type)` and `isinstance(...,
Attribute)` are not supported for the dialect extensions[^2];
**c**. Since these types are not exposed in public headers, downstream
users (dialect extensions or not) cannot write functions that overload
on e.g. `PyFloat8*Type` - that's quite a [useful
feature](fdbee98df8/cpp_ext/TorchOps.cpp (L29-L69))!
2. The dialect extensions incur a sizeable performance penalty relative
to the core extensions in that every single trip across the wire (either
`python->cpp` or `cpp->python`) requires work in addition to nanobind's
own casting/construction pipeline;
**a**. When going from `python->cpp`, [we extract the capsule object
from the Python
object](https://github.com/llvm/llvm-project/blob/main/mlir/include/mlir/Bindings/Python/NanobindAdaptors.h#L219C24-L219C46)
and then extract from the capsule the `Mlir*` opaque struct/ptr. This
side isn't so onerous;
**b**. When going from `cpp->python` we call long-hand call Python
`import` APIs and construct the Python object using `_CAPICreate`. Note,
there at least 2 `attr` calls incurred in addition to `_CAPICreate`;
this is already much more [efficiently handled by nanobind
itself](4ba51fcf79/src/nb_internals.h (L381-L382))!
3. This division blocks various features: in some configurations[^1] we
trigger a circular import bug because "dialect" types and attributes
perform an [import of the root `_mlir`
module](bd9651bf78/mlir/include/mlir/Bindings/Python/NanobindAdaptors.h (L585))
when they are created (the types themselves, not even instances of those
types). This blocks type stub generation for dialect extensions (i.e.,
the reason we currently only generate type stubs for `_mlir`).
# How
Prior this was not done/possible because of "ODR" issues but I have
resolved those issues; the basic idea for how we solve this is "move
things we want to share into shared libraries":
1. Move IRCore (stuff like `PyConcreteValue`, `PyConcreteType`,
`PyConcreteAttribute`) into `MLIRPythonSupport`;
- Note, we move the rest of the things in `IRModule.h` (renamed to
`IRCore.h`) because `PyConcreteValue`, `PyConcreteType`,
`PyConcreteAttribute` depend on them. This makes for a bigger PR than
one would hope for but ultimately I think we should give people access
to these classes to use as they see fit (specifically inherit from, but
also liberally use in bindings signatures instead of the opaque `Mlir*`
struct wrappers).
2. Put all of this code into a nested namespace
`MLIR_BINDINGS_PYTHON_DOMAIN` which is determined by a compile time
define (and tied to `MLIR_BINDINGS_PYTHON_NB_DOMAIN`). This is necessary
in order to prevent conflicts on both symbol name **and** typeid
(necessary for nanobind to not double register binded types) between
multiple bindings libraries (e.g., `torch-mlir`, and `jax`). Note
[nanobind doesn't support `module_local` like
pybind11](https://nanobind.readthedocs.io/en/latest/porting.html#removed-features).
It does support `NB_DOMAIN` but that is not sufficient for
disambiguating typeids across projects (to wit: we currently define
`NB_DOMAIN` and it was still necessary to move everything to a nested
namespace);
3. Build the [nanobind library itself as a shared
object](https://github.com/wjakob/nanobind/blob/master/cmake/nanobind-config.cmake#L127)
(and link it to both the extensions and `MLIRPythonSupport`).
4. CMake to make this work, in-tree, out-of-tree, downstream, upstream,
etc.
# Testing
Three tests are added here
1. `PythonTestModuleNanobind` is ported to use
`PyConcreteType<PyTestType>` instead of `mlir_type_subclass` and
`PyConcreteAttribute<PyTestAttr>` instead of `mlir_atrr_subclass`,
verifying this works for non-core extensions in-tree;
2. `StandaloneExtensionNanobind` is ported to use `struct PyCustomType :
mlir::python::MLIR_BINDINGS_PYTHON_DOMAIN::PyConcreteType<PyCustomType>`
instead of `mlir_type_subclass` verifying this works for non-core
extensions out-of-tree;
3. `StandaloneExtensionNanobind`'s `smoketest` is extended to also load
another bindings package (namely `mlir`) verifying
`MLIR_BINDINGS_PYTHON_DOMAIN` successfully disambiguates symbols and
typeids.
I have also tested this downstream:
https://github.com/llvm/eudsl/pull/287 as well run the following builder
bots:
mlir-nvidia-gcc7:
https://lab.llvm.org/buildbot/#/buildrequests/6654424?redirect_to_build=true
I have also tested against IREE:
https://github.com/iree-org/iree/pull/21916
# Integration
It is highly recommended to set the CMake var
`MLIR_BINDINGS_PYTHON_NB_DOMAIN` (which will also determine
`MLIR_BINDINGS_PYTHON_DOMAIN`) to something unique for each downstream.
This can also be passed explicitly to `add_mlir_python_modules` if your
project builds multiple bindings packages. I added a `WARNING` to this
effect in `AddMLIRPython.cmake`.
[^3]: Python values being typed correctly when exiting from cpp;
[^1]: Specifically when the modules are imported using `importlib`,
which occurs with nanobind's
[stubgen](https://github.com/wjakob/nanobind/blob/master/src/stubgen.py#L965);
[^2]: The workaround we implemented was a class method for the dialect
bindings called `Class.isinstance(...)`;
This makes it similar to `mlir::TypedValue` in the MLIR C++ API and
allows users to be more specific about the values they produce or
accept.
Co-authored-by: Maksim Levental <maksim.levental@gmail.com>
Currently the type hints on the returns of the "value builders" are
`ir.Value`, `Sequence[ir.Value]`, and `ir.Operation`, none of which are
correct. The correct possibilities are `ir.OpResult`, `ir.OpResultList`,
the OpView class itself (e.g., `AttrSizedResultsOp`) or the union of the
3 (for variadic results). This PR fixes those hints.
Currently in MLIR python bindings, operations with inferable result
types (e.g. with `InferTypeOpInterface` or `SameOperandsAndResultType`)
will generate such builder functions:
```python
def my_op(arg1, arg2 .. argN, *, loc=None, ip=None):
... # result types will be inferred automatically
```
However, in some cases we may want to provide the result types
explicitly. For example, the implementation of interface method
`inferResultTypes(..)` can return a failure and then we cannot build the
op in that way. Also, in the C++ side we have multiple `build` methods
for both explicitly specify the result types and automatically inferring
them.
In this PR, we change the signature of this builder function to:
```python
def my_op(arg1, arg2 .. argN, *, results=None, loc=None, ip=None):
... # result types will be inferred automatically if results is None
```
If the `results` is not provided, it will be inferred automatically,
otherwise the provided result types will be utilized. Also, `__init__`
methods of the generated op classes are changed correspondingly. Note
that for operations without inferable result types, the signature remain
unchanged, i.e. `def my_op(res1 .. resN, arg1 .. argN, *, loc=None,
ip=None)`.
---
Previously I have considered an approach like `my_op(arg, *, res1=None,
res2=None, loc=None, ip=None)`, but I quickly realized it had some
issues. For example, if the user only provides some of the arguments—say
`my_op(v1, res1=i32)`—this could lead to problems. Moreover, we don’t
seem to have a mechanism for inferring only part of result types. A
unified `results` parameter seems to be more simple and straightforward.
This is mentioned as a "must" in
https://nanobind.readthedocs.io/en/latest/porting.html#type-casters when
implementing type casters.
While most of the existing `from_cpp` methods were already marked
noexcept, many of the `from_python` methods were not. This commit adds
the missing noexcept declarations to all type casters found in
`NanobindAdaptors.h`.
---------
Co-authored-by: Maksim Levental <maksim.levental@gmail.com>
This PR allows out-of-tree dialects to write Python dialect modules
using nanobind instead of pybind11.
It may make sense to migrate in-tree dialects and some of the ODS Python
infrastructure to nanobind, but that is a topic for a future change.
This PR makes the following changes:
* adds nanobind to the CMake and Bazel build systems. We also add
robin_map to the Bazel build, which is a dependency of nanobind.
* adds a PYTHON_BINDING_LIBRARY option to various CMake functions, such
as declare_mlir_python_extension, allowing users to select a Python
binding library.
* creates a fork of mlir/include/mlir/Bindings/Python/PybindAdaptors.h
named NanobindAdaptors.h. This plays the same role, using nanobind
instead of pybind11.
* splits CollectDiagnosticsToStringScope out of PybindAdaptors.h and
into a new header mlir/include/mlir/Bindings/Python/Diagnostics.h, since
it is code that is no way related to pybind11 or for that matter,
Python.
* changed the standalone Python extension example to have both pybind11
and nanobind variants.
* changed mlir/python/mlir/dialects/python_test.py to have both pybind11
and nanobind variants.
Notes:
* A slightly unfortunate thing that I needed to do in the CMake
integration was to use FindPython in addition to FindPython3, since
nanobind's CMake integration expects the Python_ names for variables.
Perhaps there's a better way to do this.
As reported in https://github.com/llvm/llvm-project/issues/101132, this
fixes two bugs:
1. When accessing variadic operands inside an operation, it must be
accessed as `self.operation.operands` instead of `operation.operands`
2. The implementation of the `equally_sized_accessor` function is doing
wrong arithmetics when calculating the resulting index and group sizes.
I have added a test for the `equally_sized_accessor` function, which did
not have a test previously.
This PR adds "value casting", i.e., a mechanism to wrap `ir.Value` in a
proxy class that overloads dunders such as `__add__`, `__sub__`, and
`__mul__` for fun and great profit.
This is thematically similar to
bfb1ba7526
and
9566ee2806.
The example in the test demonstrates the value of the feature (no pun
intended):
```python
@register_value_caster(F16Type.static_typeid)
@register_value_caster(F32Type.static_typeid)
@register_value_caster(F64Type.static_typeid)
@register_value_caster(IntegerType.static_typeid)
class ArithValue(Value):
__add__ = partialmethod(_binary_op, op="add")
__sub__ = partialmethod(_binary_op, op="sub")
__mul__ = partialmethod(_binary_op, op="mul")
a = arith.constant(value=FloatAttr.get(f16_t, 42.42))
b = a + a
# CHECK: ArithValue(%0 = arith.addf %cst, %cst : f16)
print(b)
a = arith.constant(value=FloatAttr.get(f32_t, 42.42))
b = a - a
# CHECK: ArithValue(%1 = arith.subf %cst_0, %cst_0 : f32)
print(b)
a = arith.constant(value=FloatAttr.get(f64_t, 42.42))
b = a * a
# CHECK: ArithValue(%2 = arith.mulf %cst_1, %cst_1 : f64)
print(b)
```
**EDIT**: this now goes through the bindings and thus supports automatic
casting of `OpResult` (including as an element of `OpResultList`),
`BlockArgument` (including as an element of `BlockArgumentList`), as
well as `Value`.
<img
src="https://github.com/llvm/llvm-project/assets/5657668/443852b6-ac25-45bb-a38b-5dfbda09d5a7"
height="400" />
<p></p>
So turns out that none of the `replace=True` things actually work
because of the map caches (except for
`register_attribute_builder(replace=True)`, which doesn't use such a
cache). This was hidden by a series of unfortunate events:
1. `register_type_caster` failure was hidden because it was the same
`TestIntegerRankedTensorType` being replaced with itself (d'oh).
2. `register_operation` failure was hidden behind the "order of events"
in the lifecycle of typical extension import/use. Since extensions are
loaded/registered almost immediately after generated builders are
registered, there is no opportunity for the `operationClassMapCache` to
be populated (through e.g., `module.body.operations[2]` or
`module.body.operations[2].opview` or something). Of course as soon as
you as actually do "late-bind/late-register" the extension, you see it's
not successfully replacing the stale one in `operationClassMapCache`.
I'll take this opportunity to propose we ditch the caches all together.
I've been cargo-culting them but I really don't understand how they
work. There's this comment above `operationClassMapCache`
```cpp
/// Cache of operation name to external operation class object. This is
/// maintained on lookup as a shadow of operationClassMap in order for repeat
/// lookups of the classes to only incur the cost of one hashtable lookup.
llvm::StringMap<pybind11::object> operationClassMapCache;
```
But I don't understand how that's true given that the canonical thing
`operationClassMap` is already a map:
```cpp
/// Map of full operation name to external operation class object.
llvm::StringMap<pybind11::object> operationClassMap;
```
Maybe it wasn't always the case? Anyway things work now but it seems
like an unnecessary layer of complexity for not much gain? But maybe I'm
wrong.
This patch adds attribute builders for all buildable attributes from the
builtin dialect that did not previously have any. These builders can be
used to construct attributes of a particular type identified by a string
from a Python argument without knowing the details of how to pass that
Python argument to the attribute constructor. This is used, for example,
in the generated code of the Python bindings of ops.
The list of "all" attributes was produced with:
(
grep -h "ods_ir.AttrBuilder.get" $(find ../build/ -name "*_ops_gen.py") \
| cut -f2 -d"'"
git grep -ho "^def [a-zA-Z0-9_]*" -- include/mlir/IR/CommonAttrConstraints.td \
| cut -f2 -d" "
) | sort -u
Then, I only retained those that had an occurence in
`mlir/include/mlir/IR`. In particular, this drops many dialect-specific
attributes; registering those builders is something that those dialects
should do. Finally, I removed those attrbiutes that had a match in
`mlir/python/mlir/ir.py` already and implemented the remaining ones. The
only ones that still miss a builder now are the following:
* Represent more than one possible attribute type:
- `Any.*Attr` (9x)
- `IntNonNegative`
- `IntPositive`
- `IsNullAttr`
- `ElementsAttr`
* I am not sure what "constant attributes" are:
- `ConstBoolAttrFalse`
- `ConstBoolAttrTrue`
- `ConstUnitAttr`
* `Location` not exposed by Python bindings:
- `LocationArrayAttr`
- `LocationAttr`
* `get` function not implemented in Python bindings:
- `StringElementsAttr`
This patch also fixes a compilation problem with
`I64SmallVectorArrayAttr`.
Reviewed By: makslevental, rkayaith
Differential Revision: https://reviews.llvm.org/D159403
Right now `inferTypeOpInterface.inferReturnTypes` fails because there's a cast in there to `py::sequence` which throws a `TypeError` when it tries to cast the `None`s. Note `None`s are inserted into `operands` for omitted operands passed to the generated builder:
```
operands.append(_get_op_result_or_value(start) if start is not None else None)
operands.append(_get_op_result_or_value(stop) if stop is not None else None)
operands.append(_get_op_result_or_value(step) if step is not None else None)
```
Note also that skipping appending to the list operands doesn't work either because [[ 27c37327da/mlir/lib/Bindings/Python/IRCore.cpp (L1585) | build generic ]] checks against the number of operand segments expected.
Currently the only way around is to handroll through `ir.Operation.create`.
Reviewed By: rkayaith
Differential Revision: https://reviews.llvm.org/D151409
depends on D150839
This diff uses `MlirTypeID` to register `TypeCaster`s (i.e., `[](PyType pyType) -> DerivedTy { return pyType; }`) for all concrete types (i.e., `PyConcrete<...>`) that are then queried for (by `MlirTypeID`) and called in `struct type_caster<MlirType>::cast`. The result is that anywhere an `MlirType mlirType` is returned from a python binding, that `mlirType` is automatically cast to the correct concrete type. For example:
```
c0 = arith.ConstantOp(f32, 0.0)
# CHECK: F32Type(f32)
print(repr(c0.result.type))
unranked_tensor_type = UnrankedTensorType.get(f32)
unranked_tensor = tensor.FromElementsOp(unranked_tensor_type, [c0]).result
# CHECK: UnrankedTensorType
print(type(unranked_tensor.type).__name__)
# CHECK: UnrankedTensorType(tensor<*xf32>)
print(repr(unranked_tensor.type))
```
This functionality immediately extends to typed attributes (i.e., `attr.type`).
The diff also implements similar functionality for `mlir_type_subclass`es but in a slightly different way - for such types (which have no cpp corresponding `class` or `struct`) the user must provide a type caster in python (similar to how `AttrBuilder` works) or in cpp as a `py::cpp_function`.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D150927
This is an ongoing series of commits that are reformatting our
Python code.
Reformatting is done with `black`.
If you end up having problems merging this commit because you
have made changes to a python file, the best way to handle that
is to run git checkout --ours <yourfile> and then reformat it
with black.
If you run into any problems, post to discourse about it and
we will try to help.
RFC Thread below:
https://discourse.llvm.org/t/rfc-document-and-standardize-python-code-style
Differential Revision: https://reviews.llvm.org/D150782
This diff adds python bindings for `MlirTypeID`. It paves the way for returning accurately typed `Type`s from python APIs (see D150927) and then further along building type "conscious" `Value` APIs (see D150413).
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D150839
Add more attribute builders, such as "F32Attr", "F64Attr" and "F64ArrayAttr", which are useful to create operations by python bindings. For example, tosa.clamp in _tosa_ops_gen.py need 'F32Attr'.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D150757
Add C and python bindings for InferShapedTypeOpInterface
and ShapedTypeComponents. This allows users to invoke
InferShapedTypeOpInterface for ops that implement it.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D149494
This diff modifies `mlir-tblgen` to generate Python Operation class `__init__()`
functions that use Python keyword-only arguments.
Previously, all `__init__()` function arguments were positional. Python code to
create MLIR Operations was required to provide values for ALL builder arguments,
including optional arguments (attributes and operands). Callers that did not
provide, for example, an optional attribute would be forced to provide `None`
as an argument for EACH optional attribute. Proposed changes in this diff use
`tblgen` record information (as provided by ODS) to generate keyword arguments
for:
- optional operands
- optional attributes (which includes unit attributes)
- default-valued attributes
These `__init__()` function keyword arguments have default `None` values (i.e.
the argument form is `optionalAttr=None`), allowing callers to create Operations
more easily.
Note that since optional arguments become keyword-only arguments (since they are
placed after the bare `*` argument), this diff will require ALL optional
operands and attributes to be provided using explicit keyword syntax. This may,
in the short term, break any out-of-tree Python code that provided values via
positional arguments. However, in the long term, it seems that requiring
keywords for optional arguments will be more robust to operation changes that
add arguments.
Tests were modified to reflect the updated Operation builder calling convention.
This diff partially addresses the requests made in the github issue below.
https://github.com/llvm/llvm-project/issues/54932
Reviewed By: stellaraccident, mikeurbach
Differential Revision: https://reviews.llvm.org/D124717
When attempting to cast a pybind11 handle to an MLIR C API object through
capsules, the binding code would attempt to directly access the "_CAPIPtr"
attribute on the object, leading to a rather obscure AttributeError when the
attribute was missing, e.g., on non-MLIR types. Check for its presence and
throw a TypeError instead.
Depends On D117646
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D117658
The constructor function was being defined without indicating its "__init__"
name, which made it interpret it as a regular fuction rather than a
constructor. When overload resolution failed, Pybind would attempt to print the
arguments actually passed to the function, including "self", which is not
initialized since the constructor couldn't be called. This would result in
"__repr__" being called with "self" referencing an uninitialized MLIR C API
object, which in turn would cause undefined behavior when attempting to print
in C++. Even if the correct name is provided, the mechanism used by
PybindAdaptors.h to bind constructors directly as "__init__" functions taking
"self" is deprecated by Pybind. The new mechanism does not seem to have access
to a fully-constructed "self" object (i.e., the constructor in C++ takes a
`pybind11::detail::value_and_holder` that cannot be forwarded back to Python).
Instead, redefine "__new__" to perform the required checks (there are no
additional initialization needed for attributes and types as they are all
wrappers around a C++ pointer). "__new__" can call its equivalent on a
superclass without needing "self".
Bump pybind11 dependency to 3.8.0, which is the first version that allows one
to redefine "__new__".
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D117646
Previously, in case there was only one `Optional` operand/result within
the list, we would always return `None` from the accessor, e.g., for a
single optional result we would generate:
```
return self.operation.results[0] if len(self.operation.results) > 1 else None
```
But what we really want is to return `None` only if the length of
`results` is smaller than the total number of element groups (i.e.,
the optional operand/result is in fact missing).
This commit also renames a few local variables in the generator to make
the distinction between `isVariadic()` and `isVariableLength()` a bit
more clear.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D113855
In several cases, operation result types can be unambiguously inferred from
operands and attributes at operation construction time. Stop requiring the user
to provide these types as arguments in the ODS-generated constructors in Python
bindings. In particular, handle the SameOperandAndResultTypes and
FirstAttrDerivedResultType traits as well as InferTypeOpInterface using the
recently added interface support. This is a significant usability improvement
for IR construction, similar to what C++ ODS provides.
Depends On D111656
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D111811
Introduce the initial support for operation interfaces in C API and Python
bindings. Interfaces are a key component of MLIR's extensibility and should be
available in bindings to make use of full potential of MLIR.
This initial implementation exposes InferTypeOpInterface all the way to the
Python bindings since it can be later used to simplify the operation
construction methods by inferring their return types instead of requiring the
user to do so. The general infrastructure for binding interfaces is defined and
InferTypeOpInterface can be used as an example for binding other interfaces.
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D111656
* NFC but has some fixes for CMake glitches discovered along the way (things not cleaning properly, co-mingled depends).
* Includes previously unsubmitted fix in D98681 and a TODO to fix it more appropriately in a smaller followup.
Differential Revision: https://reviews.llvm.org/D101493
