[Mlir-commits] [mlir] [mlir] allow inlining complex ops (PR #77514)

[Kai Sasaki]

================
@@ -28,6 +41,7 @@ void complex::ComplexDialect::initialize() {
 #include "mlir/Dialect/Complex/IR/ComplexAttributes.cpp.inc"
       >();
   declarePromisedInterface<ComplexDialect, ConvertToLLVMPatternInterface>();
+  addInterfaces<ComplexInlinerInterface>();
----------------
Lewuathe wrote:

Could you add the test case checking the inlining correctly for complex dialects?

It looks like something like this test can cover the inlining. 

```mlir
// RUN: mlir-opt -allow-unregistered-dialect %s -inline="default-pipeline=''" | FileCheck %s

// Basic test that functions within affine operations are inlined.
func.func @double_complex(%cplx: complex<f32>) -> complex<f32> {
  %double = complex.add %cplx, %cplx : complex<f32>  
  return %double : complex<f32>
}


// CHECK-LABEL: func @inline_with_complex_ops
func.func @inline_with_complex_ops() -> complex<f32> {
  %c1 = arith.constant 1.0 : f32
  %c2 = arith.constant 2.0 : f32
  %c = complex.create %c1, %c2 : complex<f32>

  // CHECK: complex.add
  // CHECK-NOT: call
  %r = call @double_complex(%c) : (complex<f32>) -> (complex<f32>)
  return %r : complex<f32>
}

```

https://github.com/llvm/llvm-project/pull/77514