[MLIR][Python] Support Python-defined passes in MLIR #156000

PragmaTwice · 2025-08-29T09:57:12Z

STATUS: This PR is ready-for-review but quite experimental :)

It tries to close #155996.

This PR exports a class mlir.passmanager.Pass for Python-side to use for defining new MLIR passes.

This is a simple example of a Python-defined pass.

from mlir.passmanager import PassManager
from mlir.passes import Pass

class DemoPass(Pass):
  def __init__(self):
    super().__init__("DemoPass")

  def run(op):
    # do something with op
    pass

pm = PassManager('any', ctx)
pm.add(DemoPass())
pm.run(..)

TODO list:

tests for this change
interop with PDL rewriting ([mlir] Add PDL C & Python usage #94714)
support to clone passes
use Python-native ref-count for lifetime

PragmaTwice · 2025-08-29T16:25:02Z

mlir/include/mlir-c/Rewrite.h

+MLIR_CAPI_EXPORTED MlirLogicalResult mlirApplyPatternsAndFoldGreedilyForOp(
+    MlirOperation op, MlirFrozenRewritePatternSet patterns,
+    MlirGreedyRewriteDriverConfig);


Not sure which name is suitable.. So the current name is just a placeholder. And modifying mlirApplyPatternsAndFoldGreedily seems a breaking change.

It is a breaking change, but not sure how widely and also C API is really best effort stable (meaning, we try not to break it). I probably should have included a Module suffix on the original, and then this could have gone without. How many usages in the wild can you find of this call?

(Nit: ForOp suffix makes me think it's related to ForOp).

Hmmm, I perform a simple check on public repos via GitHub search (https://github.com/search?q=mlirApplyPatternsAndFoldGreedily+&type=code) and there seems to be some use cases of this API so I'm not sure if we should change it or not (maybe leave to maintainers for this decision : ).

Currently I'm going to rename it to mlirApplyPatternsAndFoldGreedilyWithOp to avoid such confusion with ForOp as you mentioned : )

Can you please extract this change in an independent PR? This is largely unrelated (your test python pass could just do another kind of rewrite right now).

Renamed in ca80408.

Can you please extract this change in an independent PR? This is largely unrelated (your test python pass could just do another kind of rewrite right now).

Ahh got it.

I think I can do it via op/block/region/walk/erase APIs directly, but it maybe a little ugly since the normal rewriter API is not ported to python yet AFAIK?

github-actions · 2025-08-29T16:25:38Z

✅ With the latest revision this PR passed the Python code formatter.

github-actions · 2025-08-29T16:25:38Z

✅ With the latest revision this PR passed the C/C++ code formatter.

PragmaTwice · 2025-08-29T16:26:20Z

mlir/lib/Bindings/Python/Pass.cpp

+        mlirTypeIDCreate(this), mlirStringRefCreate(name.data(), name.length()),
+        mlirStringRefCreate(argument.data(), argument.length()),
+        mlirStringRefCreate(description.data(), description.length()),
+        mlirStringRefCreate(opName.data(), opName.size()), 0, nullptr,


dependent dialects are not yet supported.

PragmaTwice · 2025-08-29T17:05:52Z

mlir/lib/Bindings/Python/Pass.cpp

+
 #include "mlir-c/Bindings/Python/Interop.h" // This is expected after nanobind.
+#include "nanobind/trampoline.h"


The empty line is to workaround clang-format: otherwise it will report that "mlir-c/Bindings/Python/Interop.h" should be put before "mlir/Bindings/Python/Nanobind.h" : )

you should do

// clang-format off #include "mlir/Bindings/Python/Nanobind.h" #include "mlir-c/Bindings/Python/Interop.h" // ON WINDOWS This is expected after nanobind. // clang-format on

(the ON WINDOWS isn't currently there but it should be because I discovered recently that it's in fact only windows that this matters for).

Ahhh thank you for your suggestion. It is done in c8c2fae.

llvmbot · 2025-08-29T17:10:44Z

@llvm/pr-subscribers-mlir

Author: Twice (PragmaTwice)

Changes

STATUS: This PR is work-in-progress now :)

It tries to close #155996.

This PR exports a class mlir.passmanager.Pass for Python-side to use for defining new MLIR passes.

This is a simple example of a Python-defined pass.

from mlir.passmanager import Pass, PassManager

class DemoPass(Pass):
  def run(op):
    # do something with op
    pass

pm = PassManager('any', ctx)
pm.add(DemoPass())
pm.run(..)

TODO list:

tests for this change
interop with PDL rewriting
support to clone passes
use Python-native ref-count for lifetime

Full diff: https://github.com/llvm/llvm-project/pull/156000.diff

7 Files Affected:

(modified) mlir/include/mlir-c/Rewrite.h (+4)
(modified) mlir/lib/Bindings/Python/MainModule.cpp (+1)
(modified) mlir/lib/Bindings/Python/Pass.cpp (+103)
(modified) mlir/lib/Bindings/Python/Pass.h (+1)
(modified) mlir/lib/Bindings/Python/Rewrite.cpp (+21-10)
(modified) mlir/lib/CAPI/Transforms/Rewrite.cpp (+7)
(added) mlir/test/python/pass.py (+79)

diff --git a/mlir/include/mlir-c/Rewrite.h b/mlir/include/mlir-c/Rewrite.h
index 61d3446317550..21ae236d6f73f 100644
--- a/mlir/include/mlir-c/Rewrite.h
+++ b/mlir/include/mlir-c/Rewrite.h
@@ -301,6 +301,10 @@ mlirFreezeRewritePattern(MlirRewritePatternSet op);
 MLIR_CAPI_EXPORTED void
 mlirFrozenRewritePatternSetDestroy(MlirFrozenRewritePatternSet op);
 
+MLIR_CAPI_EXPORTED MlirLogicalResult mlirApplyPatternsAndFoldGreedilyForOp(
+    MlirOperation op, MlirFrozenRewritePatternSet patterns,
+    MlirGreedyRewriteDriverConfig);
+
 MLIR_CAPI_EXPORTED MlirLogicalResult mlirApplyPatternsAndFoldGreedily(
     MlirModule op, MlirFrozenRewritePatternSet patterns,
     MlirGreedyRewriteDriverConfig);
diff --git a/mlir/lib/Bindings/Python/MainModule.cpp b/mlir/lib/Bindings/Python/MainModule.cpp
index 278847e7ac7f5..590e862a8d358 100644
--- a/mlir/lib/Bindings/Python/MainModule.cpp
+++ b/mlir/lib/Bindings/Python/MainModule.cpp
@@ -139,4 +139,5 @@ NB_MODULE(_mlir, m) {
   auto passModule =
       m.def_submodule("passmanager", "MLIR Pass Management Bindings");
   populatePassManagerSubmodule(passModule);
+  populatePassSubmodule(passModule);
 }
diff --git a/mlir/lib/Bindings/Python/Pass.cpp b/mlir/lib/Bindings/Python/Pass.cpp
index 1030dea7f364c..3fd27fed34587 100644
--- a/mlir/lib/Bindings/Python/Pass.cpp
+++ b/mlir/lib/Bindings/Python/Pass.cpp
@@ -11,7 +11,9 @@
 #include "IRModule.h"
 #include "mlir-c/Pass.h"
 #include "mlir/Bindings/Python/Nanobind.h"
+
 #include "mlir-c/Bindings/Python/Interop.h" // This is expected after nanobind.
+#include "nanobind/trampoline.h"
 
 namespace nb = nanobind;
 using namespace nb::literals;
@@ -20,6 +22,81 @@ using namespace mlir::python;
 
 namespace {
 
+// A base class for defining passes in Python
+// Users are expected to subclass this and implement the `run` method, e.g.
+// ```
+// class MyPass(mlir.passmanager.Pass):
+//   def __init__(self):
+//     super().__init__("MyPass", ..)
+//     # other init stuff..
+//   def run(self, operation):
+//     # do something with operation..
+//     pass
+// ```
+class PyPassBase {
+public:
+  PyPassBase(std::string name, std::string argument, std::string description,
+             std::string opName)
+      : name(std::move(name)), argument(std::move(argument)),
+        description(std::move(description)), opName(std::move(opName)) {
+    callbacks.construct = [](void *obj) {};
+    callbacks.destruct = [](void *obj) {
+      nb::handle(static_cast<PyObject *>(obj)).dec_ref();
+    };
+    callbacks.run = [](MlirOperation op, MlirExternalPass, void *obj) {
+      auto handle = nb::handle(static_cast<PyObject *>(obj));
+      nb::cast<PyPassBase *>(handle)->run(op);
+    };
+    callbacks.clone = [](void *obj) -> void * {
+      nb::object copy = nb::module_::import_("copy");
+      nb::object deepcopy = copy.attr("deepcopy");
+      return deepcopy(obj).release().ptr();
+    };
+    callbacks.initialize = nullptr;
+  }
+
+  // this method should be overridden by subclasses in Python.
+  virtual void run(MlirOperation op) = 0;
+
+  virtual ~PyPassBase() = default;
+
+  // Make an MlirPass instance on-the-fly that wraps this object.
+  // Note that passmanager will take the ownership of the returned
+  // object and release it when appropriate.
+  MlirPass make() {
+    auto *obj = nb::find(this).release().ptr();
+    return mlirCreateExternalPass(
+        mlirTypeIDCreate(this), mlirStringRefCreate(name.data(), name.length()),
+        mlirStringRefCreate(argument.data(), argument.length()),
+        mlirStringRefCreate(description.data(), description.length()),
+        mlirStringRefCreate(opName.data(), opName.size()), 0, nullptr,
+        callbacks, obj);
+  }
+
+  const std::string &getName() const { return name; }
+  const std::string &getArgument() const { return argument; }
+  const std::string &getDescription() const { return description; }
+  const std::string &getOpName() const { return opName; }
+
+private:
+  MlirExternalPassCallbacks callbacks;
+
+  std::string name;
+  std::string argument;
+  std::string description;
+  std::string opName;
+};
+
+// A trampoline class upon PyPassBase.
+// Refer to
+// https://nanobind.readthedocs.io/en/latest/classes.html#overriding-virtual-functions-in-python
+class PyPass : PyPassBase {
+public:
+  NB_TRAMPOLINE(PyPassBase, 1);
+
+  void run(MlirOperation op) override { NB_OVERRIDE_PURE(run, op); }
+};
+
 /// Owning Wrapper around a PassManager.
 class PyPassManager {
 public:
@@ -52,6 +129,26 @@ class PyPassManager {
 
 } // namespace
 
+void mlir::python::populatePassSubmodule(nanobind::module_ &m) {
+  //----------------------------------------------------------------------------
+  // Mapping of the Python-defined Pass interface
+  //----------------------------------------------------------------------------
+  nb::class_<PyPassBase, PyPass>(m, "Pass")
+      .def(nb::init<std::string, std::string, std::string, std::string>(),
+           "name"_a, nb::kw_only(), "argument"_a = "", "description"_a = "",
+           "op_name"_a = "", "Create a new Pass.")
+      .def("run", &PyPassBase::run, "operation"_a,
+           "Run the pass on the provided operation.")
+      .def_prop_ro("name",
+                   [](const PyPassBase &self) { return self.getName(); })
+      .def_prop_ro("argument",
+                   [](const PyPassBase &self) { return self.getArgument(); })
+      .def_prop_ro("description",
+                   [](const PyPassBase &self) { return self.getDescription(); })
+      .def_prop_ro("op_name",
+                   [](const PyPassBase &self) { return self.getOpName(); });
+}
+
 /// Create the `mlir.passmanager` here.
 void mlir::python::populatePassManagerSubmodule(nb::module_ &m) {
   //----------------------------------------------------------------------------
@@ -157,6 +254,12 @@ void mlir::python::populatePassManagerSubmodule(nb::module_ &m) {
           "pipeline"_a,
           "Add textual pipeline elements to the pass manager. Throws a "
           "ValueError if the pipeline can't be parsed.")
+      .def(
+          "add",
+          [](PyPassManager &passManager, PyPassBase &pass) {
+            mlirPassManagerAddOwnedPass(passManager.get(), pass.make());
+          },
+          "pass"_a, "Add a python-defined pass to the pass manager.")
       .def(
           "run",
           [](PyPassManager &passManager, PyOperationBase &op,
diff --git a/mlir/lib/Bindings/Python/Pass.h b/mlir/lib/Bindings/Python/Pass.h
index bc40943521829..ba3fbb707fed7 100644
--- a/mlir/lib/Bindings/Python/Pass.h
+++ b/mlir/lib/Bindings/Python/Pass.h
@@ -15,6 +15,7 @@ namespace mlir {
 namespace python {
 
 void populatePassManagerSubmodule(nanobind::module_ &m);
+void populatePassSubmodule(nanobind::module_ &m);
 
 } // namespace python
 } // namespace mlir
diff --git a/mlir/lib/Bindings/Python/Rewrite.cpp b/mlir/lib/Bindings/Python/Rewrite.cpp
index 0373f9c7affe9..675bd685ec2db 100644
--- a/mlir/lib/Bindings/Python/Rewrite.cpp
+++ b/mlir/lib/Bindings/Python/Rewrite.cpp
@@ -99,14 +99,25 @@ void mlir::python::populateRewriteSubmodule(nb::module_ &m) {
       .def(MLIR_PYTHON_CAPI_FACTORY_ATTR,
            &PyFrozenRewritePatternSet::createFromCapsule);
   m.def(
-      "apply_patterns_and_fold_greedily",
-      [](MlirModule module, MlirFrozenRewritePatternSet set) {
-        auto status = mlirApplyPatternsAndFoldGreedily(module, set, {});
-        if (mlirLogicalResultIsFailure(status))
-          // FIXME: Not sure this is the right error to throw here.
-          throw nb::value_error("pattern application failed to converge");
-      },
-      "module"_a, "set"_a,
-      "Applys the given patterns to the given module greedily while folding "
-      "results.");
+       "apply_patterns_and_fold_greedily",
+       [](MlirModule module, MlirFrozenRewritePatternSet set) {
+         auto status = mlirApplyPatternsAndFoldGreedily(module, set, {});
+         if (mlirLogicalResultIsFailure(status))
+           // FIXME: Not sure this is the right error to throw here.
+           throw nb::value_error("pattern application failed to converge");
+       },
+       "module"_a, "set"_a,
+       "Applys the given patterns to the given module greedily while folding "
+       "results.")
+      .def(
+          "apply_patterns_and_fold_greedily_for_op",
+          [](MlirOperation op, MlirFrozenRewritePatternSet set) {
+            auto status = mlirApplyPatternsAndFoldGreedilyForOp(op, set, {});
+            if (mlirLogicalResultIsFailure(status))
+              // FIXME: Not sure this is the right error to throw here.
+              throw nb::value_error("pattern application failed to converge");
+          },
+          "op"_a, "set"_a,
+          "Applys the given patterns to the given op greedily while folding "
+          "results.");
 }
diff --git a/mlir/lib/CAPI/Transforms/Rewrite.cpp b/mlir/lib/CAPI/Transforms/Rewrite.cpp
index a4df97f7beace..d606445cfad31 100644
--- a/mlir/lib/CAPI/Transforms/Rewrite.cpp
+++ b/mlir/lib/CAPI/Transforms/Rewrite.cpp
@@ -294,6 +294,13 @@ mlirApplyPatternsAndFoldGreedily(MlirModule op,
   return wrap(mlir::applyPatternsGreedily(unwrap(op), *unwrap(patterns)));
 }
 
+MlirLogicalResult
+mlirApplyPatternsAndFoldGreedilyForOp(MlirOperation op,
+                                      MlirFrozenRewritePatternSet patterns,
+                                      MlirGreedyRewriteDriverConfig) {
+  return wrap(mlir::applyPatternsGreedily(unwrap(op), *unwrap(patterns)));
+}
+
 //===----------------------------------------------------------------------===//
 /// PDLPatternModule API
 //===----------------------------------------------------------------------===//
diff --git a/mlir/test/python/pass.py b/mlir/test/python/pass.py
new file mode 100644
index 0000000000000..1c338e4bc6b49
--- /dev/null
+++ b/mlir/test/python/pass.py
@@ -0,0 +1,79 @@
+# RUN: %PYTHON %s 2>&1 | FileCheck %s
+
+import gc, sys
+from mlir.ir import *
+from mlir.passmanager import *
+from mlir.dialects.builtin import ModuleOp
+from mlir.dialects import pdl
+from mlir.rewrite import *
+
+
+def log(*args):
+    print(*args, file=sys.stderr)
+    sys.stderr.flush()
+
+
+def run(f):
+    log("\nTEST:", f.__name__)
+    f()
+    gc.collect()
+    assert Context._get_live_count() == 0
+
+
+def make_pdl_module():
+    with Location.unknown():
+        pdl_module = Module.create()
+        with InsertionPoint(pdl_module.body):
+            # Change all arith.addi with index types to arith.muli.
+            @pdl.pattern(benefit=1, sym_name="addi_to_mul")
+            def pat():
+                # Match arith.addi with index types.
+                index_type = pdl.TypeOp(IndexType.get())
+                operand0 = pdl.OperandOp(index_type)
+                operand1 = pdl.OperandOp(index_type)
+                op0 = pdl.OperationOp(
+                    name="arith.addi", args=[operand0, operand1], types=[index_type]
+                )
+
+                # Replace the matched op with arith.muli.
+                @pdl.rewrite()
+                def rew():
+                    newOp = pdl.OperationOp(
+                        name="arith.muli", args=[operand0, operand1], types=[index_type]
+                    )
+                    pdl.ReplaceOp(op0, with_op=newOp)
+
+        return pdl_module
+
+
+# CHECK-LABEL: TEST: testCustomPass
+@run
+def testCustomPass():
+    with Context():
+        pdl_module = make_pdl_module()
+
+        class CustomPass(Pass):
+            def __init__(self):
+                super().__init__("CustomPass", op_name="builtin.module")
+
+            def run(self, m):
+                frozen = PDLModule(pdl_module).freeze()
+                apply_patterns_and_fold_greedily_for_op(m, frozen)
+
+        module = ModuleOp.parse(
+            r"""
+            module {
+              func.func @add(%a: index, %b: index) -> index {
+                %sum = arith.addi %a, %b : index
+                return %sum : index
+              }
+            }
+        """
+        )
+
+        # CHECK-LABEL: Dump After CustomPass
+        # CHECK: arith.muli
+        pm = PassManager("any")
+        pm.enable_ir_printing()
+        pm.add(CustomPass())
+        pm.run(module)

PragmaTwice · 2025-09-01T02:20:03Z

mlir/lib/Bindings/Python/Pass.cpp

+      nb::object deepcopy = copy.attr("deepcopy");
+      return deepcopy(obj).release().ptr();
+    };
+    callbacks.initialize = nullptr;


initialize is not ported to python side yet.

PragmaTwice · 2025-09-01T03:16:34Z

mlir/lib/Bindings/Python/MainModule.cpp

  auto passModule =
      m.def_submodule("passmanager", "MLIR Pass Management Bindings");
  populatePassManagerSubmodule(passModule);
+  populatePassSubmodule(passModule);


It seems that pass is a keyword in python so that name like mlir.pass.Pass doesn't work. 🤔

Ah yes, pass vs pass :-) I don't have a good suggestion given how established that is MLIR/compilers and Python side ... MlirPass or PyPass or pydefpass would be most obvious.

Currently the full name of the base class is mlir.passmanager.Pass. Is it good enough or we'd better to rename it with another module name?

Given that PEP8 recommends appending a underscore in case of names of arguments / attributes clashing with reserved keywords, I feel mlir.pass_.Pass is an option as it is not too surprising for me personally.

I think putting a trailing underscore in a namespace is a little awkward. How about mlir.passes? or mlir.passinfra? or I dunno something like that that I'm failing to come up with right now...

Got it. For now, I will rename it to mlir.passes.Pass. Let me know if anyone think of a better name : )

Done in 01e68c5.

PragmaTwice · 2025-09-01T07:12:18Z

mlir/lib/Bindings/Python/Pass.cpp

+  // Note that passmanager will take the ownership of the returned
+  // object and release it when appropriate.
+  MlirPass make() {
+    auto *obj = nb::find(this).release().ptr();


About lifetime of the python object: here we increase the ref count by nb::find (and by release() we avoid decrease the count here), and when the ExternalPass is destructed, callback.destructor is called so that dec_ref() is called for this object.

…dFoldGreedilyWithOp

makslevental

I need to take another pass (not pun intended) but here are some initial non-nit comments.

makslevental · 2025-09-02T02:32:27Z

mlir/lib/Bindings/Python/MainModule.cpp

  auto passModule =
      m.def_submodule("passmanager", "MLIR Pass Management Bindings");
  populatePassManagerSubmodule(passModule);
+  populatePassSubmodule(passModule);


I think putting a trailing underscore in a namespace is a little awkward. How about mlir.passes? or mlir.passinfra? or I dunno something like that that I'm failing to come up with right now...

makslevental · 2025-09-02T02:37:00Z

mlir/lib/Bindings/Python/Pass.cpp

+
 #include "mlir-c/Bindings/Python/Interop.h" // This is expected after nanobind.
+#include "nanobind/trampoline.h"


you should do

// clang-format off #include "mlir/Bindings/Python/Nanobind.h" #include "mlir-c/Bindings/Python/Interop.h" // ON WINDOWS This is expected after nanobind. // clang-format on

(the ON WINDOWS isn't currently there but it should be because I discovered recently that it's in fact only windows that this matters for).

makslevental · 2025-09-02T02:42:41Z

mlir/lib/Bindings/Python/Pass.cpp

+  //----------------------------------------------------------------------------
+  // Mapping of the Python-defined Pass interface
+  //----------------------------------------------------------------------------
+  nb::class_<PyPassBase, PyPass>(m, "Pass")


isn't this backwards? cf <PyOperation, PyOperationBase>

Ahh this is a quite interesting case that I just did some research on : )

Refer to the nanobind documentation (https://nanobind.readthedocs.io/en/latest/api_core.html#_CPPv4I0DpEN8nanobind6class_E):

template<typename T, typename ...Ts>
class class_ : public object

The variable length parameter Ts is optional and can be used to specify the base class of T and/or an alias needed to realize trampoline classes.

So in the case you mentioned (<PyOperation, PyOperationBase>), the parameter Ts (PyOperationBase) is a base class of T (PyOperation); and for the case here (<PyPassBase, PyPass>), the Ts (PyPass) is a trampoline class of T (PyPassBase). So I think both of them is correct here.

For example, in the documentation of trampoline classes (https://nanobind.readthedocs.io/en/latest/classes.html#overriding-virtual-functions-in-python), we can see such an instance of nb::class_:

nb::class_<Dog, PyDog /* <-- trampoline */>(m, "Dog")

And here PyDog is a derived class of Dog but also a trampoline class of Dog : )

PragmaTwice · 2025-09-02T09:38:28Z

mlir/lib/Bindings/Python/Pass.cpp

+  // Make an MlirPass instance on-the-fly that wraps this object.
+  // Note that passmanager will take the ownership of the returned
+  // object and release it when appropriate.
+  MlirPass make() {


The reason that in this design I add a new class PyPassBase and lazily construct the MlirPass while adding into the pass manager instead of just wrapping MlirPass as a python object is that: the lifetime of MlirPass is a bit tricky.

If we don't add it into the pass manager, the owership of MlirPass should be in our hand (and we are responsible to release it finally), and after adding into a pass manager, the ownership is transfered into the manager, and delete myPassPtr will be called when the manager is destructed on the C++ side. So, the ownership state of MlirPass is changed between and after adding to pass manager.

By lazily constructing the MlirPass only when adding to a pass manager, we can avoid to expose the MlirPass object into the python world so that we don't need to care about how to interoperate between c++ delete passPtr and python object ref-count. And also we can ensure that the lifetime of the MlirPass is always handled by the pass manager.

This approach might seem a bit roundabout but works. I’m also open to trying other methods if there are any : ) (Maybe I can check how the Region/Block binding APIs work since the situation is similiar? Not sure yet.) WDYT?

PragmaTwice added 3 commits August 29, 2025 17:43

[MLIR][Python] Support Python-defined passes in MLIR

8386c87

add ctor with args for Pass

cb82621

fix lifetime issue

7556ca2

makslevental assigned makslevental and unassigned makslevental Aug 29, 2025

add test case

d5055a7

PragmaTwice commented Aug 29, 2025

View reviewed changes

PragmaTwice added 4 commits August 30, 2025 00:31

fix header

6d2f472

format

1a98ae8

format

751fe84

format

7966ddd

PragmaTwice commented Aug 29, 2025

View reviewed changes

remove useless comment

6a9ec66

PragmaTwice marked this pull request as ready for review August 29, 2025 17:10

PragmaTwice requested review from makslevental, rolfmorel, ftynse and stellaraccident as code owners August 29, 2025 17:10

llvmbot added the mlir label Aug 29, 2025

PragmaTwice added 2 commits August 30, 2025 13:31

improve test

2965a9e

Merge branch 'main' into mlir-python-pass

1dde449

PragmaTwice commented Sep 1, 2025

View reviewed changes

rename mlirApplyPatternsAndFoldGreedilyForOp with mlirApplyPatternsAn…

ca80408

…dFoldGreedilyWithOp

makslevental reviewed Sep 2, 2025

View reviewed changes

PragmaTwice added 2 commits September 2, 2025 14:26

add clang-format annotation for header ordering

c8c2fae

rename mlir.passmanager.Pass to mlir.passes.Pass

01e68c5

PragmaTwice commented Sep 2, 2025

View reviewed changes


		#include "mlir-c/Bindings/Python/Interop.h" // This is expected after nanobind.
		#include "nanobind/trampoline.h"

[MLIR][Python] Support Python-defined passes in MLIR #156000

Are you sure you want to change the base?

[MLIR][Python] Support Python-defined passes in MLIR #156000

Uh oh!

Conversation

PragmaTwice commented Aug 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PragmaTwice Sep 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Aug 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Aug 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

llvmbot commented Aug 29, 2025

Uh oh!

PragmaTwice Sep 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rolfmorel Sep 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PragmaTwice Sep 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

makslevental left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PragmaTwice Sep 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

PragmaTwice commented Aug 29, 2025 •

edited

Loading

PragmaTwice Sep 1, 2025 •

edited

Loading

github-actions bot commented Aug 29, 2025 •

edited

Loading

github-actions bot commented Aug 29, 2025 •

edited

Loading

PragmaTwice Sep 1, 2025 •

edited

Loading

rolfmorel Sep 1, 2025 •

edited

Loading

PragmaTwice Sep 1, 2025 •

edited

Loading

PragmaTwice Sep 2, 2025 •

edited

Loading