2021-04-22 17:32:10 +02:00
|
|
|
//===- IRModule.cpp - IR pybind module ------------------------------------===//
|
|
|
|
|
//
|
|
|
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
|
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
|
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
|
|
|
//
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
|
|
#include "IRModule.h"
|
|
|
|
|
|
2023-11-03 13:28:20 -05:00
|
|
|
#include <optional>
|
|
|
|
|
#include <vector>
|
|
|
|
|
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
#include "Globals.h"
|
|
|
|
|
#include "NanobindUtils.h"
|
2025-08-12 17:59:59 -04:00
|
|
|
#include "mlir-c/Bindings/Python/Interop.h"
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
#include "mlir-c/Support.h"
|
2024-12-20 23:32:32 -05:00
|
|
|
#include "mlir/Bindings/Python/Nanobind.h"
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
|
|
|
|
|
namespace nb = nanobind;
|
2021-04-22 17:32:10 +02:00
|
|
|
using namespace mlir;
|
|
|
|
|
using namespace mlir::python;
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------------
|
|
|
|
|
// PyGlobals
|
|
|
|
|
// -----------------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
PyGlobals *PyGlobals::instance = nullptr;
|
|
|
|
|
|
|
|
|
|
PyGlobals::PyGlobals() {
|
|
|
|
|
assert(!instance && "PyGlobals already constructed");
|
|
|
|
|
instance = this;
|
2021-09-03 00:37:00 +00:00
|
|
|
// The default search path include {mlir.}dialects, where {mlir.} is the
|
|
|
|
|
// package prefix configured at compile time.
|
2021-12-22 00:19:53 +00:00
|
|
|
dialectSearchPrefixes.emplace_back(MAKE_MLIR_PYTHON_QUALNAME("dialects"));
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
PyGlobals::~PyGlobals() { instance = nullptr; }
|
|
|
|
|
|
2023-11-03 13:28:20 -05:00
|
|
|
bool PyGlobals::loadDialectModule(llvm::StringRef dialectNamespace) {
|
2025-01-13 12:00:31 +01:00
|
|
|
{
|
|
|
|
|
nb::ft_lock_guard lock(mutex);
|
|
|
|
|
if (loadedDialectModules.contains(dialectNamespace))
|
|
|
|
|
return true;
|
|
|
|
|
}
|
2021-04-22 17:32:10 +02:00
|
|
|
// Since re-entrancy is possible, make a copy of the search prefixes.
|
|
|
|
|
std::vector<std::string> localSearchPrefixes = dialectSearchPrefixes;
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object loaded = nb::none();
|
2021-04-22 17:32:10 +02:00
|
|
|
for (std::string moduleName : localSearchPrefixes) {
|
|
|
|
|
moduleName.push_back('.');
|
|
|
|
|
moduleName.append(dialectNamespace.data(), dialectNamespace.size());
|
|
|
|
|
|
|
|
|
|
try {
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
loaded = nb::module_::import_(moduleName.c_str());
|
|
|
|
|
} catch (nb::python_error &e) {
|
2021-04-22 17:32:10 +02:00
|
|
|
if (e.matches(PyExc_ModuleNotFoundError)) {
|
|
|
|
|
continue;
|
|
|
|
|
}
|
2021-12-20 19:45:05 +00:00
|
|
|
throw;
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
2023-11-03 13:28:20 -05:00
|
|
|
if (loaded.is_none())
|
|
|
|
|
return false;
|
2021-04-22 17:32:10 +02:00
|
|
|
// Note: Iterator cannot be shared from prior to loading, since re-entrancy
|
|
|
|
|
// may have occurred, which may do anything.
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
2023-11-03 13:28:20 -05:00
|
|
|
loadedDialectModules.insert(dialectNamespace);
|
|
|
|
|
return true;
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
|
|
|
|
|
2022-12-21 16:22:39 -08:00
|
|
|
void PyGlobals::registerAttributeBuilder(const std::string &attributeKind,
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::callable pyFunc, bool replace) {
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object &found = attributeBuilderMap[attributeKind];
|
2023-11-03 13:28:20 -05:00
|
|
|
if (found && !replace) {
|
2022-12-21 16:22:39 -08:00
|
|
|
throw std::runtime_error((llvm::Twine("Attribute builder for '") +
|
[mlir][python bindings] generate all the enums
This PR implements python enum bindings for *all* the enums - this includes `I*Attrs` (including positional/bit) and `Dialect/EnumAttr`.
There are a few parts to this:
1. CMake: a small addition to `declare_mlir_dialect_python_bindings` and `declare_mlir_dialect_extension_python_bindings` to generate the enum, a boolean arg `GEN_ENUM_BINDINGS` to make it opt-in (even though it works for basically all of the dialects), and an optional `GEN_ENUM_BINDINGS_TD_FILE` for handling corner cases.
2. EnumPythonBindingGen.cpp: there are two weedy aspects here that took investigation:
1. If an enum attribute is not a `Dialect/EnumAttr` then the `EnumAttrInfo` record is canonical, as far as both the cases of the enum **and the `AttrDefName`**. On the otherhand, if an enum is a `Dialect/EnumAttr` then the `EnumAttr` record has the correct `AttrDefName` ("load bearing", i.e., populates `ods.ir.AttributeBuilder('<NAME>')`) but its `enum` field contains the cases, which is an instance of `EnumAttrInfo`. The solution is to generate an one enum class for both `Dialect/EnumAttr` and "independent" `EnumAttrInfo` but to make that class interopable with two builder registrations that both do the right thing (see next sub-bullet).
2. Because we don't have a good connection to cpp `EnumAttr`, i.e., only the `enum class` getters are exposed (like `DimensionAttr::get(Dimension value)`), we have to resort to parsing e.g., `Attribute.parse(f'#gpu<dim {x}>')`. This means that the set of supported `assemblyFormat`s (for the enum) is fixed at compile of MLIR (currently 2, the only 2 I saw). There might be some things that could be done here but they would require quite a bit more C API work to support generically (e.g., casting ints to enum cases and binding all the getters or going generically through the `symbolize*` methods, like `symbolizeDimension(uint32_t)` or `symbolizeDimension(StringRef)`).
A few small changes:
1. In addition, since this patch registers default builders for attributes where people might've had their own builders already written, I added a `replace` param to `AttributeBuilder.insert` (`False` by default).
2. `makePythonEnumCaseName` can't handle all the different ways in which people write their enum cases, e.g., `llvm.CConv.Intel_OCL_BI`, which gets turned into `INTEL_O_C_L_B_I` (because `llvm::convertToSnakeFromCamelCase` doesn't look for runs of caps). So I dropped it. On the otherhand regularization does need to done because some enums have `None` as a case (and others might have other python keywords).
3. I turned on `llvm` dialect generation here in order to test `nvvm.WGMMAScaleIn`, which is an enum with [[ https://github.com/llvm/llvm-project/blob/d7e26b56207cbd8995296c5bb7c11ce676b649da/mlir/include/mlir/IR/EnumAttr.td#L22-L25 | no explicit discriminator ]] for the `neg` case.
Note, dialects that didn't get a `GEN_ENUM_BINDINGS` don't have any enums to generate.
Let me know if I should add more tests (the three trivial ones I added exercise both the supported `assemblyFormat`s and `replace=True`).
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D157934
2023-08-23 13:27:08 -05:00
|
|
|
attributeKind +
|
|
|
|
|
"' is already registered with func: " +
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::cast<std::string>(nb::str(found)))
|
2022-12-21 16:22:39 -08:00
|
|
|
.str());
|
|
|
|
|
}
|
|
|
|
|
found = std::move(pyFunc);
|
|
|
|
|
}
|
|
|
|
|
|
2023-05-26 10:23:17 -05:00
|
|
|
void PyGlobals::registerTypeCaster(MlirTypeID mlirTypeID,
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::callable typeCaster, bool replace) {
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object &found = typeCasterMap[mlirTypeID];
|
2023-11-03 13:28:20 -05:00
|
|
|
if (found && !replace)
|
|
|
|
|
throw std::runtime_error("Type caster is already registered with caster: " +
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::cast<std::string>(nb::str(found)));
|
2023-05-26 10:23:17 -05:00
|
|
|
found = std::move(typeCaster);
|
|
|
|
|
}
|
|
|
|
|
|
[mlir][python] value casting (#69644)
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
https://github.com/llvm/llvm-project/commit/bfb1ba752655bf09b35c486f6cc9817dbedfb1bb
and
https://github.com/llvm/llvm-project/commit/9566ee280607d91fa2e5eca730a6765ac84dfd0f.
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`.
2023-11-07 10:49:41 -06:00
|
|
|
void PyGlobals::registerValueCaster(MlirTypeID mlirTypeID,
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::callable valueCaster, bool replace) {
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object &found = valueCasterMap[mlirTypeID];
|
[mlir][python] value casting (#69644)
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
https://github.com/llvm/llvm-project/commit/bfb1ba752655bf09b35c486f6cc9817dbedfb1bb
and
https://github.com/llvm/llvm-project/commit/9566ee280607d91fa2e5eca730a6765ac84dfd0f.
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`.
2023-11-07 10:49:41 -06:00
|
|
|
if (found && !replace)
|
|
|
|
|
throw std::runtime_error("Value caster is already registered: " +
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::cast<std::string>(nb::repr(found)));
|
[mlir][python] value casting (#69644)
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
https://github.com/llvm/llvm-project/commit/bfb1ba752655bf09b35c486f6cc9817dbedfb1bb
and
https://github.com/llvm/llvm-project/commit/9566ee280607d91fa2e5eca730a6765ac84dfd0f.
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`.
2023-11-07 10:49:41 -06:00
|
|
|
found = std::move(valueCaster);
|
|
|
|
|
}
|
|
|
|
|
|
2021-04-22 17:32:10 +02:00
|
|
|
void PyGlobals::registerDialectImpl(const std::string &dialectNamespace,
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object pyClass) {
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object &found = dialectClassMap[dialectNamespace];
|
2021-04-22 17:32:10 +02:00
|
|
|
if (found) {
|
2023-05-22 17:30:12 -05:00
|
|
|
throw std::runtime_error((llvm::Twine("Dialect namespace '") +
|
|
|
|
|
dialectNamespace + "' is already registered.")
|
|
|
|
|
.str());
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
|
|
|
|
found = std::move(pyClass);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void PyGlobals::registerOperationImpl(const std::string &operationName,
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object pyClass, bool replace) {
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
nb::object &found = operationClassMap[operationName];
|
[mlir][python] remove mixins (#68853)
This PR replaces the mixin `OpView` extension mechanism with the
standard inheritance mechanism.
Why? Firstly, mixins are not very pythonic (inheritance is usually used
for this), a little convoluted, and too "tight" (can only be used in the
immediately adjacent `_ext.py`). Secondly, it (mixins) are now blocking
are correct implementation of "value builders" (see
[here](https://github.com/llvm/llvm-project/pull/68764)) where the
problem becomes how to choose the correct base class that the value
builder should call.
This PR looks big/complicated but appearances are deceiving; 4 things
were needed to make this work:
1. Drop `skipDefaultBuilders` in
`OpPythonBindingGen::emitDefaultOpBuilders`
2. Former mixin extension classes are converted to inherit from the
generated `OpView` instead of being "mixins"
a. extension classes that simply were calling into an already generated
`super().__init__` continue to do so
b. (almost all) extension classes that were calling `self.build_generic`
because of a lack of default builder being generated can now also just
call `super().__init__`
3. To handle the [lone single
use-case](https://sourcegraph.com/search?q=context%3Aglobal+select_opview_mixin&patternType=standard&sm=1&groupBy=repo)
of `select_opview_mixin`, namely
[linalg](https://github.com/llvm/llvm-project/blob/main/mlir/python/mlir/dialects/_linalg_ops_ext.py#L38),
only a small change was necessary in `opdsl/lang/emitter.py` (thanks to
the emission/generation of default builders/`__init__`s)
4. since the `extend_opview_class` decorator is removed, we need a way
to register extension classes as the desired `OpView` that `op.opview`
conjures into existence; so we do the standard thing and just enable
replacing the existing registered `OpView` i.e.,
`register_operation(_Dialect, replace=True)`.
Note, the upgrade path for the common case is to change an extension to
inherit from the generated builder and decorate it with
`register_operation(_Dialect, replace=True)`. In the slightly more
complicated case where `super().__init(self.build_generic(...))` is
called in the extension's `__init__`, this needs to be updated to call
`__init__` in `OpView`, i.e., the grandparent (see updated docs).
Note, also `<DIALECT>_ext.py` files/modules will no longer be automatically loaded.
Note, the PR has 3 base commits that look funny but this was done for
the purpose of tracking the line history of moving the
`<DIALECT>_ops_ext.py` class into `<DIALECT>.py` and updating (commit
labeled "fix").
2023-10-19 16:20:14 -05:00
|
|
|
if (found && !replace) {
|
2023-05-22 17:30:12 -05:00
|
|
|
throw std::runtime_error((llvm::Twine("Operation '") + operationName +
|
|
|
|
|
"' is already registered.")
|
|
|
|
|
.str());
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
|
|
|
|
found = std::move(pyClass);
|
|
|
|
|
}
|
|
|
|
|
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
std::optional<nb::callable>
|
2022-12-21 16:22:39 -08:00
|
|
|
PyGlobals::lookupAttributeBuilder(const std::string &attributeKind) {
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
2022-12-21 16:22:39 -08:00
|
|
|
const auto foundIt = attributeBuilderMap.find(attributeKind);
|
|
|
|
|
if (foundIt != attributeBuilderMap.end()) {
|
2023-11-03 13:28:20 -05:00
|
|
|
assert(foundIt->second && "attribute builder is defined");
|
2022-12-21 16:22:39 -08:00
|
|
|
return foundIt->second;
|
|
|
|
|
}
|
|
|
|
|
return std::nullopt;
|
|
|
|
|
}
|
|
|
|
|
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
std::optional<nb::callable> PyGlobals::lookupTypeCaster(MlirTypeID mlirTypeID,
|
2023-05-26 10:23:17 -05:00
|
|
|
MlirDialect dialect) {
|
2023-11-20 19:54:55 -06:00
|
|
|
// Try to load dialect module.
|
|
|
|
|
(void)loadDialectModule(unwrap(mlirDialectGetNamespace(dialect)));
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
2023-11-03 13:28:20 -05:00
|
|
|
const auto foundIt = typeCasterMap.find(mlirTypeID);
|
|
|
|
|
if (foundIt != typeCasterMap.end()) {
|
|
|
|
|
assert(foundIt->second && "type caster is defined");
|
|
|
|
|
return foundIt->second;
|
2023-05-26 10:23:17 -05:00
|
|
|
}
|
2023-11-03 13:28:20 -05:00
|
|
|
return std::nullopt;
|
2023-05-26 10:23:17 -05:00
|
|
|
}
|
|
|
|
|
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
std::optional<nb::callable> PyGlobals::lookupValueCaster(MlirTypeID mlirTypeID,
|
[mlir][python] value casting (#69644)
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
https://github.com/llvm/llvm-project/commit/bfb1ba752655bf09b35c486f6cc9817dbedfb1bb
and
https://github.com/llvm/llvm-project/commit/9566ee280607d91fa2e5eca730a6765ac84dfd0f.
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`.
2023-11-07 10:49:41 -06:00
|
|
|
MlirDialect dialect) {
|
2023-11-20 19:54:55 -06:00
|
|
|
// Try to load dialect module.
|
|
|
|
|
(void)loadDialectModule(unwrap(mlirDialectGetNamespace(dialect)));
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
[mlir][python] value casting (#69644)
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
https://github.com/llvm/llvm-project/commit/bfb1ba752655bf09b35c486f6cc9817dbedfb1bb
and
https://github.com/llvm/llvm-project/commit/9566ee280607d91fa2e5eca730a6765ac84dfd0f.
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`.
2023-11-07 10:49:41 -06:00
|
|
|
const auto foundIt = valueCasterMap.find(mlirTypeID);
|
|
|
|
|
if (foundIt != valueCasterMap.end()) {
|
|
|
|
|
assert(foundIt->second && "value caster is defined");
|
|
|
|
|
return foundIt->second;
|
|
|
|
|
}
|
|
|
|
|
return std::nullopt;
|
|
|
|
|
}
|
|
|
|
|
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
std::optional<nb::object>
|
2021-04-22 17:32:10 +02:00
|
|
|
PyGlobals::lookupDialectClass(const std::string &dialectNamespace) {
|
2023-11-03 13:28:20 -05:00
|
|
|
// Make sure dialect module is loaded.
|
|
|
|
|
if (!loadDialectModule(dialectNamespace))
|
|
|
|
|
return std::nullopt;
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
2021-04-22 17:32:10 +02:00
|
|
|
const auto foundIt = dialectClassMap.find(dialectNamespace);
|
|
|
|
|
if (foundIt != dialectClassMap.end()) {
|
2023-11-03 13:28:20 -05:00
|
|
|
assert(foundIt->second && "dialect class is defined");
|
2021-04-22 17:32:10 +02:00
|
|
|
return foundIt->second;
|
|
|
|
|
}
|
2023-11-03 13:28:20 -05:00
|
|
|
// Not found and loading did not yield a registration.
|
2022-12-06 00:03:44 -08:00
|
|
|
return std::nullopt;
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
|
|
|
|
|
[mlir python] Port Python core code to nanobind. (#120473)
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
2024-12-18 21:55:42 -05:00
|
|
|
std::optional<nb::object>
|
2023-01-22 23:31:18 -05:00
|
|
|
PyGlobals::lookupOperationClass(llvm::StringRef operationName) {
|
2023-11-03 13:28:20 -05:00
|
|
|
// Make sure dialect module is loaded.
|
2021-04-22 17:32:10 +02:00
|
|
|
auto split = operationName.split('.');
|
|
|
|
|
llvm::StringRef dialectNamespace = split.first;
|
2023-11-03 13:28:20 -05:00
|
|
|
if (!loadDialectModule(dialectNamespace))
|
2022-12-06 00:03:44 -08:00
|
|
|
return std::nullopt;
|
2021-04-22 17:32:10 +02:00
|
|
|
|
2025-01-13 12:00:31 +01:00
|
|
|
nb::ft_lock_guard lock(mutex);
|
2023-11-03 13:28:20 -05:00
|
|
|
auto foundIt = operationClassMap.find(operationName);
|
|
|
|
|
if (foundIt != operationClassMap.end()) {
|
|
|
|
|
assert(foundIt->second && "OpView is defined");
|
|
|
|
|
return foundIt->second;
|
|
|
|
|
}
|
|
|
|
|
// Not found and loading did not yield a registration.
|
|
|
|
|
return std::nullopt;
|
2021-04-22 17:32:10 +02:00
|
|
|
}
|
2025-08-12 17:59:59 -04:00
|
|
|
|
|
|
|
|
bool PyGlobals::TracebackLoc::locTracebacksEnabled() {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
return locTracebackEnabled_;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void PyGlobals::TracebackLoc::setLocTracebacksEnabled(bool value) {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
locTracebackEnabled_ = value;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
size_t PyGlobals::TracebackLoc::locTracebackFramesLimit() {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
return locTracebackFramesLimit_;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void PyGlobals::TracebackLoc::setLocTracebackFramesLimit(size_t value) {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
locTracebackFramesLimit_ = std::min(value, kMaxFrames);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void PyGlobals::TracebackLoc::registerTracebackFileInclusion(
|
|
|
|
|
const std::string &file) {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
auto reg = "^" + llvm::Regex::escape(file);
|
|
|
|
|
if (userTracebackIncludeFiles.insert(reg).second)
|
|
|
|
|
rebuildUserTracebackIncludeRegex = true;
|
|
|
|
|
if (userTracebackExcludeFiles.count(reg)) {
|
|
|
|
|
if (userTracebackExcludeFiles.erase(reg))
|
|
|
|
|
rebuildUserTracebackExcludeRegex = true;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void PyGlobals::TracebackLoc::registerTracebackFileExclusion(
|
|
|
|
|
const std::string &file) {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
auto reg = "^" + llvm::Regex::escape(file);
|
|
|
|
|
if (userTracebackExcludeFiles.insert(reg).second)
|
|
|
|
|
rebuildUserTracebackExcludeRegex = true;
|
|
|
|
|
if (userTracebackIncludeFiles.count(reg)) {
|
|
|
|
|
if (userTracebackIncludeFiles.erase(reg))
|
|
|
|
|
rebuildUserTracebackIncludeRegex = true;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bool PyGlobals::TracebackLoc::isUserTracebackFilename(
|
|
|
|
|
const llvm::StringRef file) {
|
|
|
|
|
nanobind::ft_lock_guard lock(mutex);
|
|
|
|
|
if (rebuildUserTracebackIncludeRegex) {
|
|
|
|
|
userTracebackIncludeRegex.assign(
|
|
|
|
|
llvm::join(userTracebackIncludeFiles, "|"));
|
|
|
|
|
rebuildUserTracebackIncludeRegex = false;
|
|
|
|
|
isUserTracebackFilenameCache.clear();
|
|
|
|
|
}
|
|
|
|
|
if (rebuildUserTracebackExcludeRegex) {
|
|
|
|
|
userTracebackExcludeRegex.assign(
|
|
|
|
|
llvm::join(userTracebackExcludeFiles, "|"));
|
|
|
|
|
rebuildUserTracebackExcludeRegex = false;
|
|
|
|
|
isUserTracebackFilenameCache.clear();
|
|
|
|
|
}
|
|
|
|
|
if (!isUserTracebackFilenameCache.contains(file)) {
|
|
|
|
|
std::string fileStr = file.str();
|
|
|
|
|
bool include = std::regex_search(fileStr, userTracebackIncludeRegex);
|
|
|
|
|
bool exclude = std::regex_search(fileStr, userTracebackExcludeRegex);
|
|
|
|
|
isUserTracebackFilenameCache[file] = include || !exclude;
|
|
|
|
|
}
|
|
|
|
|
return isUserTracebackFilenameCache[file];
|
|
|
|
|
}
|