[mlir][emitC] Add support to emitter for `classop`, `fieldop` and `getfieldop` #145605

Jaddyen · 2025-06-24T21:58:01Z

Add support to the emitter for ClassOp, FieldOp and GetFieldOp.
These ops were introduced in #141158

llvmbot · 2025-06-24T22:12:54Z

@llvm/pr-subscribers-mlir-emitc

@llvm/pr-subscribers-mlir

Author: Jaden Angella (Jaddyen)

Changes

Extends the emitter to accurately emit classes from classops, fieldops and getfieldops

Full diff: https://github.com/llvm/llvm-project/pull/145605.diff

2 Files Affected:

(modified) mlir/lib/Target/Cpp/TranslateToCpp.cpp (+57-1)
(added) mlir/test/mlir-translate/emitc_classops.mlir (+36)

diff --git a/mlir/lib/Target/Cpp/TranslateToCpp.cpp b/mlir/lib/Target/Cpp/TranslateToCpp.cpp
index 067a0470b14e4..9cd30eae8d36d 100644
--- a/mlir/lib/Target/Cpp/TranslateToCpp.cpp
+++ b/mlir/lib/Target/Cpp/TranslateToCpp.cpp
@@ -997,6 +997,61 @@ static LogicalResult printOperation(CppEmitter &emitter, ModuleOp moduleOp) {
   return success();
 }
 
+static LogicalResult printOperation(CppEmitter &emitter, ClassOp classOp) {
+  CppEmitter::Scope classScope(emitter);
+  raw_indented_ostream &os = emitter.ostream();
+  os << "class " << classOp.getSymName() << " final {\n";
+  os << "public:\n\n";
+
+  os.indent();
+  os << "const std::map<std::string, char*> _buffer_map {\n";
+  for (Operation &op : classOp) {
+    if (auto fieldOp = dyn_cast<FieldOp>(op))
+      os << "  { \"" << fieldOp.getSymName() << "\", reinterpret_cast<char*>(&"
+         << fieldOp.getAttrs() << ") },\n";
+  }
+  os << "};\n";
+
+  os << "char* getBufferForName(const std::string& name) const {\n";
+  os << "  auto it = _buffer_map.find(name);\n";
+  os << "  return (it == _buffer_map.end()) ? nullptr : it->second;\n";
+  os << "}\n\n";
+  for (Operation &op : classOp) {
+    if (failed(emitter.emitOperation(op, /*trailingSemicolon=*/false)))
+      return failure();
+  }
+
+  os.unindent();
+  os << "};";
+  return success();
+}
+
+static LogicalResult printOperation(CppEmitter &emitter, FieldOp fieldOp) {
+  raw_ostream &os = emitter.ostream();
+  if (failed(emitter.emitType(fieldOp->getLoc(), fieldOp.getType())))
+    return failure();
+  os << " " << fieldOp.getSymName() << ";";
+  return success();
+}
+
+static LogicalResult printOperation(CppEmitter &emitter,
+                                    GetFieldOp getFieldOp) {
+  raw_indented_ostream &os = emitter.ostream();
+  Location loc = getFieldOp->getLoc();
+
+  if (getFieldOp->getNumResults() > 0) {
+    Value result = getFieldOp->getResult(0);
+    if (failed(emitter.emitType(loc, result.getType())))
+      return failure();
+    os << " ";
+    if (failed(emitter.emitOperand(result)))
+      return failure();
+    os << " = ";
+  }
+  os << getFieldOp.getFieldName().str();
+  return success();
+}
+
 static LogicalResult printOperation(CppEmitter &emitter, FileOp file) {
   if (!emitter.shouldEmitFile(file))
     return success();
@@ -1612,7 +1667,8 @@ LogicalResult CppEmitter::emitOperation(Operation &op, bool trailingSemicolon) {
                 emitc::LoadOp, emitc::LogicalAndOp, emitc::LogicalNotOp,
                 emitc::LogicalOrOp, emitc::MulOp, emitc::RemOp, emitc::ReturnOp,
                 emitc::SubOp, emitc::SwitchOp, emitc::UnaryMinusOp,
-                emitc::UnaryPlusOp, emitc::VariableOp, emitc::VerbatimOp>(
+                emitc::UnaryPlusOp, emitc::VariableOp, emitc::VerbatimOp,
+                emitc::ClassOp, emitc::FieldOp, emitc::GetFieldOp>(
 
               [&](auto op) { return printOperation(*this, op); })
           // Func ops.
diff --git a/mlir/test/mlir-translate/emitc_classops.mlir b/mlir/test/mlir-translate/emitc_classops.mlir
new file mode 100644
index 0000000000000..3232954f1149a
--- /dev/null
+++ b/mlir/test/mlir-translate/emitc_classops.mlir
@@ -0,0 +1,36 @@
+// RUN: mlir-translate --mlir-to-cpp %s | FileCheck %s
+
+emitc.class @modelClass {
+    emitc.field @input_tensor : !emitc.array<1xf32> 
+    emitc.field @some_feature : !emitc.array<1xf32>  {emitc.opaque = ["some_feature"]}
+    emitc.func @execute() {
+        %0 = "emitc.constant"() <{value = 0 : index}> : () -> !emitc.size_t
+        %1 = get_field @input_tensor : !emitc.array<1xf32>
+        %2 = get_field @some_feature : !emitc.array<1xf32>
+        %3 = subscript %1[%0] : (!emitc.array<1xf32>, !emitc.size_t) -> !emitc.lvalue<f32>
+        return
+    }
+}
+
+// CHECK: class modelClass final {
+// CHECK-NEXT: public:
+// CHECK-EMPTY:
+// CHECK-NEXT:  const std::map<std::string, char*> _buffer_map {
+// CHECK-NEXT:    { "input_tensor", reinterpret_cast<char*>(&None) },
+// CHECK-NEXT:    { "some_feature", reinterpret_cast<char*>(&{emitc.opaque = ["some_feature"]}) }, 
+// CHECK-NEXT:  };
+// CHECK-NEXT:  char* getBufferForName(const std::string& name) const {
+// CHECK-NEXT:     auto it = _buffer_map.find(name);
+// CHECK-NEXT:     return (it == _buffer_map.end()) ? nullptr : it->second;
+// CHECK-NEXT:  }
+// CHECK-EMPTY:
+// CHECK-NEXT:  float[1] input_tensor;
+// CHECK-NEXT:  float[1] some_feature;
+// CHECK-NEXT:  void execute() {
+// CHECK-NEXT:    size_t v1 = 0;
+// CHECK-NEXT:    float[1] v2 = input_tensor;
+// CHECK-NEXT:    float[1] v3 = some_feature;
+// CHECK-NEXT:    return;
+// CHECK-NEXT:  }
+// CHECK-EMPTY:
+// CHECK-NEXT: };

mtrofin · 2025-06-24T22:23:32Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

The reflection bits should be predicated on there being those attributes.

mtrofin · 2025-06-24T22:24:20Z

mlir/test/mlir-translate/emitc_classops.mlir

add a second test where the fields don't have the attribute - everything should work, except no reflection support.

mtrofin · 2025-06-24T22:25:25Z

mlir/test/mlir-translate/emitc_classops.mlir

emitc.name_hint not emitc.opaque, no?

jpienaar · 2025-06-25T05:34:30Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

This does not seem specific to ClassOp but some usage of it. So these should be introduced before translation.

ack, will do in different step.

jpienaar · 2025-06-25T05:35:48Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

Lets keep this list sorted.

jpienaar · 2025-06-25T05:36:13Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

I think this may be specific to the MLGO use-case. Perhaps that could be something that could be encoded in the ClassOp? Then its just a matter of lowering/refining the IR correctly ahead of time, like for other constructs.

yes, we can specify in the ClassOp so this print allows for more use cases.

Yes, this could either be an attribute of type ::mlir::UnitAttr or implemented similar to the specifiers attribute (type ::mlir::ArrayAttr) of the emitc::FuncOp.

jpienaar · 2025-06-25T05:37:19Z

mlir/test/mlir-translate/emitc_classops.mlir

These don't look correct.

thanks for pointing this out! will handle this when working on buffer_map op in next stage

ilovepi · 2025-06-25T16:12:38Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

I think this may be specific to the MLGO use-case. Perhaps that could be something that could be encoded in the ClassOp? Then its just a matter of lowering/refining the IR correctly ahead of time, like for other constructs.

ilovepi · 2025-06-25T16:46:49Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

Should these lines be indented manually? It's certainly easier to read, but you could end up w/ inconsistent indentation, right?

true! it would be nicer to indent using the os. since i am deferring the emission of the buffermap to a stage where we have a buffermapop, i will handle this at that point.

jpienaar · 2025-06-26T17:24:09Z

mlir/lib/Target/Cpp/TranslateToCpp.cpp

Isn't GetFieldOp defined to have 1 result? (

llvm-project/mlir/include/mlir/Dialect/EmitC/IR/EmitC.td

Line 1678 in 5c310d1

let results = (outs AnyTypeOf<[EmitC_ArrayType, EmitC_LValueType]>:$result);

).

Can also do here getFieldOp.getResult() rather than using the generic accessor on Operation.

yes it is!
thanks for the pointer.
ive addressed this now.

add trnalsation for classops, fieldops and getfieldops

0ab534f

Jaddyen changed the title ~~[emitC] Edit emitter to emit classop, 'fieldop and getfieldop` correctly~~ [emitC] Edit emitter to emit classop, fieldop and getfieldop correctly Jun 24, 2025

Jaddyen marked this pull request as ready for review June 24, 2025 22:12

llvmbot added mlir mlir:emitc labels Jun 24, 2025

mtrofin requested review from ilovepi, jpienaar, marbre, mtrofin and simon-camp June 24, 2025 22:17

mtrofin reviewed Jun 24, 2025

View reviewed changes

Jaddyen changed the title ~~[emitC] Edit emitter to emit classop, fieldop and getfieldop correctly~~ [mlir][emitC] Add support to emitter for classop, fieldop and getfieldop Jun 24, 2025

mtrofin reviewed Jun 24, 2025

View reviewed changes

jpienaar requested changes Jun 25, 2025

View reviewed changes

Added arg to test

0d0aeb3

ilovepi reviewed Jun 25, 2025

View reviewed changes

Simplify the classop translation

812d478

ajaden-codes force-pushed the emit-class branch from ed132db to 812d478 Compare June 25, 2025 20:49

Remove unnecessary include

d1e4a7e

mtrofin approved these changes Jun 26, 2025

View reviewed changes

jpienaar approved these changes Jun 26, 2025

View reviewed changes

Corrected minor getFieldOp emission err

1180658

Jaddyen force-pushed the emit-class branch from 07f697a to 1180658 Compare June 26, 2025 19:42

ilovepi approved these changes Jun 26, 2025

View reviewed changes

mtrofin merged commit 1dfdd1e into llvm:main Jun 26, 2025
7 checks passed

Jaddyen deleted the emit-class branch June 26, 2025 21:11

[mlir][emitC] Add support to emitter for classop, fieldop and getfieldop #145605

[mlir][emitC] Add support to emitter for classop, fieldop and getfieldop #145605

Uh oh!

Conversation

Jaddyen commented Jun 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jun 24, 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!

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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Jaddyen Jun 25, 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!

Jaddyen Jun 25, 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!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

[mlir][emitC] Add support to emitter for `classop`, `fieldop` and `getfieldop` #145605

[mlir][emitC] Add support to emitter for `classop`, `fieldop` and `getfieldop` #145605

Jaddyen commented Jun 24, 2025 •

edited

Loading

llvmbot commented Jun 24, 2025 •

edited

Loading

Jaddyen Jun 25, 2025 •

edited

Loading

Jaddyen Jun 25, 2025 •

edited

Loading