Update MLIR's default SM to `sm_75`. This matches the behavior of
offline compilation tools in the CUDA Toolkit (`nvcc`, `ptxas`, ...) and
follows suit with 9fc5fd0ad689eed94f65b1d6d10f9c5642935e68.
Additionally, `sm_75` is the oldest GPU variant compatible with the
widest range of recent major CUDA Toolkit versions (11/12/13).
# 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 commit moves the code responsible for adding newlines and tracking
indent, so that it can be used not only for operation printers, but also
for attribute and type printers.
It could be useful for nested attributes, where proper formatting with
newlines and indents would benefit the readability of the IR. Currently,
everything is printed on one line, which makes it difficult to read if
the attribute is more verbose and there are multiple levels of nesting.
Co-authored-by: Andruszkiewicz, Jacenty <andruszkiewicz.jacenty@intel.com>
# Problem:
There are two build system bugs on MacOS in the case where one intends
to use multiple bindings packages simultaneously (same Python
interpreter session):
1. The nanobind modules are built with
[`-Wl,-flat_namespace`](8518d2c405/llvm/cmake/modules/HandleLLVMOptions.cmake (L268))
thereby leading to ambiguous symbols across multiple whatever dylibs;
2. Intra-library symbol resolution (within the C API aggregate dylib)
fails to resolve symbols correctly unless things are built with
`-DCMAKE_C_VISIBILITY_PRESET=hidden -DCMAKE_CXX_VISIBILITY_PRESET=hidden
-DCMAKE_VISIBILITY_INLINES_HIDDEN=ON`.
# Repro:
On a Mac (with this patch applied):
1. Build without `twolevel_namespace` and without hidden vis properties
and run `LIT_FILTER=test.toy ninja check-mlir` (assuming you have
`-DLLVM_BUILD_EXAMPLES=ON -DLLVM_INCLUDE_EXAMPLES=ON`) and you will see:
```
LLVM ERROR: can't create Attribute 'mlir::StringAttr' because storage
uniquer isn't initialized: the dialect was likely not loaded, or the
attribute wasn't added with addAttributes<...>() in the
Dialect::initialize() method.
```
2. Build with `twolevel_namespace` but not hidden vis and run the same
lit test and you will see:
```
LLVM ERROR: Attempting to attach an interface to an unregistered
operation builtin.unrealized_conversion_cast.
```
# Fix
We only do a partial fix here (adding `twolevel_namespace` to Python
bindings modules) because a full fix requires adding visibility
attributes to all object files. I added docs discussing this.
# Why is this not happening on Linux
Using `DYLD_PRINT_BINDINGS=1` I observe that for the checked-in/updated
test (without the fix) `libMLIRPythonCAPI` resolves many of its symbols
to `libStandalonePythonCAPI`:
```
dyld[98449]: looking for weak-def symbol '__ZN4mlir6TypeID3getINS_13AffineMapAttrEEES0_v':
dyld[98449]: found __ZN4mlir6TypeID3getINS_13AffineMapAttrEEES0_v in map, using impl from /Users/maksimlevental/dev_projects/llvm-project/cmake-build-debug/tools/mlir/test/Examples/standalone/python_packages/standalone/mlir_standalone/_mlir_libs/libStandalonePythonCAPI.dylib
dyld[98449]: <libMLIRPythonCAPI.dylib/bind#22> -> 0x11348fa9c <libStandalonePythonCAPI.dylib/__ZN4mlir6TypeID3getINS_13AffineMapAttrEEES0_v>)
dyld[98449]: looking for weak-def symbol '__ZN4mlir6TypeID3getINS_9ArrayAttrEEES0_v':
dyld[98449]: found __ZN4mlir6TypeID3getINS_9ArrayAttrEEES0_v in map, using impl from /Users/maksimlevental/dev_projects/llvm-project/cmake-build-debug/tools/mlir/test/Examples/standalone/python_packages/standalone/mlir_standalone/_mlir_libs/libStandalonePythonCAPI.dylib
dyld[98449]: <libMLIRPythonCAPI.dylib/bind#23> -> 0x11348f990 <libStandalonePythonCAPI.dylib/__ZN4mlir6TypeID3getINS_9ArrayAttrEEES0_v>)
dyld[98449]: looking for weak-def symbol '__ZN4mlir6TypeID3getINS_14DictionaryAttrEEES0_v':
dyld[98449]: found __ZN4mlir6TypeID3getINS_14DictionaryAttrEEES0_v in map, using impl from /Users/maksimlevental/dev_projects/llvm-project/cmake-build-debug/tools/mlir/test/Examples/standalone/python_packages/standalone/mlir_standalone/_mlir_libs/libStandalonePythonCAPI.dylib
dyld[98449]: <libMLIRPythonCAPI.dylib/bind#24> -> 0x11348eec0 <libStandalonePythonCAPI.dylib/__ZN4mlir6TypeID3getINS_14DictionaryAttrEEES0_v>)
```
Turns out this is "expected" behavior:
> It appears on macOS, when a static library is compiled without
-fvisibility=hidden, its C++ template instantiations could lead to
leftover weak symbols that are resolved and bound at runtime
https://joyeecheung.github.io/blog/2025/01/11/executable-loading-and-startup-performance-on-macos/🤷
Mention that there is no formal definition of the canonical form. Also
add more examples for users to understand what kind of transformations
the community has agreed upon in the past.
---------
Co-authored-by: Mehdi Amini <joker.eph@gmail.com>
Introduce DefaultValuedEnumAttr, which similarly to DefaultValuedAttr
decorates an enum attribute to have a default value from a specific enum
case when not present. The default is constructed as the fully-qualified
enum case symbol.
In comparison to DefaultValuedAttr, this allows using a TableGen
EnumCase
variable instead of a raw string.
Noticed the link was dead when reading the docs. According to the
wayback machine, it was a link to "Revisiting Out-of-SSA Translation for
Correctness, Code Quality, and Efficiency" by Boissinot et al., so
providing a more permanent link.
Currently, Dialect Interfaces can't be defined in ODS. This PR adds the
support for dialect interfaces. It follows the same approach with other
interfaces and extends on top of `Interface` class defined in
`mlir/TableGen/Interfaces.h`.
Given the following input:
```tablegen
#ifndef MY_INTERFACES
#define MY_INTERFACES
include "mlir/IR/Interfaces.td"
def DialectInlinerInterface : DialectInterface<"DialectInlinerInterface"> {
let description = [{
Define a base inlining interface class to allow for dialects to opt-in to the inliner.
}];
let cppNamespace = "::mlir";
let methods = [
InterfaceMethod<
/*desc=*/ [{
Returns true if the given region 'src' can be inlined into the region
'dest' that is attached to an operation registered to the current dialect.
'valueMapping' contains any remapped values from within the 'src' region.
This can be used to examine what values will replace entry arguments into
the 'src' region, for example.
}],
/*returnType=*/ "bool",
/*methodName=*/ "isLegalToInline",
/*args=*/ (ins "::mlir::Region *":$dest, "::mlir::Region *":$src, "::mlir::IRMapping &":$valueMapping),
/*methodBody=*/ [{
return true;
}]
>
];
}
#endif
```
It will generate the following code:
```cpp
/*===- TableGen'erated file -------------------------------------*- C++ -*-===*\
|* *|
|* Dialect Interface Declarations *|
|* *|
|* Automatically generated file, do not edit! *|
|* *|
\*===----------------------------------------------------------------------===*/
namespace mlir {
/// Define a base inlining interface class to allow for dialects to opt-in to the inliner.
class DialectInlinerInterface : public ::mlir::DialectInterface::Base<DialectInlinerInterface> {
public:
/// Returns true if the given region 'src' can be inlined into the region
/// 'dest' that is attached to an operation registered to the current dialect.
/// 'valueMapping' contains any remapped values from within the 'src' region.
/// This can be used to examine what values will replace entry arguments into
/// the 'src' region, for example.
virtual bool isLegalToInline(::mlir::Region * dest, ::mlir::Region * src, ::mlir::IRMapping & valueMapping) const;
protected:
DialectInlinerInterface(::mlir::Dialect *dialect) : Base(dialect) {}
};
} // namespace mlir
bool ::mlir::DialectInlinerInterface::isLegalToInline(::mlir::Region * dest, ::mlir::Region * src, ::mlir::IRMapping & valueMapping) const {
return true;
}
```
Disallow implicit casting, which is surprising, and, IME, usually
indicative of copy-paste errors.
Because the initial value must be a scalar, I don't expect this to
affect any data movement.
This patch updates the NVVM Dialect docs to:
* include information on the type of pointers for the memory spaces.
* include high-level information on mbarrier objects.
Signed-off-by: Durgadoss R <durgadossr@nvidia.com>
Same as with pass def & decl. This doesn't change anything with registry
and the big flag kept (e.g., GEN_PASS_REGISTRATION behaves like
GEN_PASS_DECL and so too for sub ones).
…consistent
Since the bindings now use nanobind, I changed the code examples and
mentions in the documentation prose to mention nanobind concepts and
symbols wherever applicable.
I also made the spelling of "Python" consistent by choosing the
uppercase name everywhere that's not an executable name, part of a URL,
or directory name.
----------------
Note that I left mentions of `PybindAdaptors.h` in because of
https://github.com/llvm/llvm-project/pull/162309.
Are there any thoughts about adding a virtual environment setup guide
using [uv](https://docs.astral.sh/uv/)? It has gotten pretty popular,
and is much faster than a "vanilla" Python pip install. It can also
bootstrap an interpreter not present on the user's machine, for example
a free-threaded Python build, with the `-p` flag to the `uv venv`
virtual environment creation command.
Add builders on the Python side that match builders in the C++ side, add tests for launching GPU kernels and regions, and correct some small documentation mistakes. This reflects the API decisions already made in the func dialect's Python bindings and makes use of the GPU dialect's bindings work more similar to C++ interface.
Add documentation for the no-rollback conversion driver. Also improve
the documentation of the old rollback driver. In particular: which
modifications are performed immediately and which are delayed.
The MLIR Python bindings now support defining new passes, new rewrite
patterns (through either `RewritePatternSet` or `PDLModule`), as well as
new dialects using the IRDL bindings. Adding a dedicated section to
document these features would make it easier for users to discover and
understand the full capabilities of the Python bindings.
Python-defined passes have been merged into the main branch for some
time now. I believe adding a corresponding section in the documentation
will help more users learn about this feature and understand how to use
it.
This PR adds such a section to the docs of Python bindings, summarizing
the feature and providing an example.
This patch fixes a documentation typo in the MLIR Toy tutorial (Ch6).
The tutorial currently refers to `examples/toy/Ch6/toy.cpp`, but the
correct
file name is `examples/toy/Ch6/toyc.cpp`.
The current implementation is overly conservative and disable all
possible caching as soon as a context-aware conversion is present.
However the context-aware conversion only affects subsequent converters,
we can cache the previous ones.
This isn't NFC because if fixed a bug where we use to unconditionally
cache when using the `convertType(Type t, ...` API, while now all APIs
are aware of context-aware conversions.
Adds a check that the bytecode buffer is aligned to any section
alignment requirements. Without this check, if the source buffer is not
sufficiently aligned, we may return early when aligning the data
pointer. In that case, we may end up trying to read successive sections
from an incorrect offset, giving the appearance of invalid bytecode.
This requirement is documented in the bytecode unit tests, but is not
otherwise documented in the code or bytecode reference.
Historical context: `PyMlirContext::liveOperations` was an optimization
meant to cut down on the number of Python object allocations and
(partially) a mechanism for updating validity of ops after
transformation. E.g. during walking/transforming the AST. See original
patch [here](https://reviews.llvm.org/D87958).
Inspired by a
[renewed](https://github.com/llvm/llvm-project/pull/139721#issuecomment-3217131918)
interest in https://github.com/llvm/llvm-project/pull/139721 (which has
become a little stale...)
<p align="center">
<img width="504" height="375" alt="image"
src="https://github.com/user-attachments/assets/0daad562-d3d1-4876-8d01-5dba382ab186"
/>
</p>
In the previous go-around
(https://github.com/llvm/llvm-project/pull/92631) there were two issues
which have been resolved
1. ops that were "fetched" under a root op which has been transformed
are no longer reported as invalid. We simply "[formally
forbid](https://github.com/llvm/llvm-project/pull/92631#issuecomment-2119397018)"
this;
2. `Module._CAPICreate(module_capsule)` must now be followed by a
`module._clear_mlir_module()` to prevent double-freeing of the actual
`ModuleOp` object (i.e. calling the dtor on the
`OwningOpRef<ModuleOp>`):
```python
module = ...
module_dup = Module._CAPICreate(module._CAPIPtr)
module._clear_mlir_module()
```
- **the alternative choice** here is to remove the `Module._CAPICreate`
API altogether and replace it with something like `Module._move(module)`
which will do both `Module._CAPICreate` and `module._clear_mlir_module`.
Note, the other approach I explored last year was a [weakref
system](https://github.com/llvm/llvm-project/pull/97340) for
`mlir::Operation` which would effectively hoist this `liveOperations`
thing into MLIR core. Possibly doable but I now believe it's a bad idea.
The other potentially breaking change is `is`, which checks object
equality rather than value equality, will now report `False` because we
are always allocating `new` Python objects (ie that's the whole point of
this change). Users wanting to check equality for `Operation` and
`Module` should use `==`.
This commit adds support for context-aware type conversions: type
conversion rules that can return different types depending on the IR.
There is no change for existing (context-unaware) type conversion rules:
```c++
// Example: Conversion any integer type to f32.
converter.addConversion([](IntegerType t) {
return Float32Type::get(t.getContext());
}
```
There is now an additional overload to register context-aware type
conversion rules:
```c++
// Example: Type conversion rule for integers, depending on the context:
// Get the defining op of `v`, read its "increment" attribute and return an
// integer with a bitwidth that is increased by "increment".
converter.addConversion([](Value v) -> std::optional<Type> {
auto intType = dyn_cast<IntegerType>(v.getType());
if (!intType)
return std::nullopt;
Operation *op = v.getDefiningOp();
if (!op)
return std::nullopt;
auto incrementAttr = op->getAttrOfType<IntegerAttr>("increment");
if (!incrementAttr)
return std::nullopt;
return IntegerType::get(v.getContext(),
intType.getWidth() + incrementAttr.getInt());
});
```
For performance reasons, the type converter caches the result of type
conversions. This is no longer possible when there context-aware type
conversions because each conversion could compute a different type
depending on the context. There is no performance degradation when there
are only context-unaware type conversions.
Note: This commit just adds context-aware type conversions to the
dialect conversion framework. There are many existing patterns that
still call `converter.convertType(someValue.getType())`. These should be
gradually updated in subsequent commits to call
`converter.convertType(someValue)`.
Co-authored-by: Markus Böck <markus.boeck02@gmail.com>
Retry landing https://github.com/llvm/llvm-project/pull/153373
## Major changes from previous attempt
- remove the test in CAPI because no existing tests in CAPI deal with
sanitizer exemptions
- update `mlir/docs/Dialects/GPU.md` to reflect the new behavior: load
GPU binary in global ctors, instead of loading them at call site.
- skip the test on Aarch64 since we have an issue with initialization there
---------
Co-authored-by: Mehdi Amini <joker.eph@gmail.com>
