[SAOC 2026] MLIR support Weekly Update #6

[Asadbek]

Hello everyone,

This week, I spent some time refactoring my `inlineMLIR` changes 
[1] to reduce the binary size overhead caused by registering all 
MLIR dialects, including the ones we don't actually use.

I finally got basic integer binary operations lowering from D to 
the MLIR `arith`[2] and `func`[3] dialects! For example, this D 
code:

```d
int add(int a, int b) { return a + b; }
int mul(int a, int b) { return a * b; }
int div(int a, int b) { return a / b; }
int min(int a, int b) { return a - b; }
```

now lowers directly to:

```
module {
   func.func @_D4test3addFiiZi(%arg0: i32, %arg1: i32) -> i32 {
     %0 = arith.addi %arg0, %arg1 : i32
     return %0 : i32
   }
   func.func @_D4test3mulFiiZi(%arg0: i32, %arg1: i32) -> i32 {
     %0 = arith.muli %arg0, %arg1 : i32
     return %0 : i32
   }
   func.func @_D4test3divFiiZi(%arg0: i32, %arg1: i32) -> i32 {
     %0 = arith.divsi %arg0, %arg1 : i32
     return %0 : i32
   }
   func.func @_D4test3minFiiZi(%arg0: i32, %arg1: i32) -> i32 {
     %0 = arith.subi %arg0, %arg1 : i32
     return %0 : i32
   }
}
```

Next, I will continue expanding support for more binary ops, 
including floating-point types. Nicholas also suggested adding 
control flow constructs soon to better test the MLIR integration, 
so I plan to start lowering into the `scf`[4] (Structured Control 
Flow) dialect in the coming days. Once that is working, I want to 
start experimenting with D-specific optimizations and constructs 
in MLIR, which is the main goal of the project.

Feedback and questions are welcome!

Best,
Asadbek

[1] https://github.com/ldc-developers/ldc/pull/5307
[2] https://mlir.llvm.org/docs/Dialects/ArithOps/
[3] https://mlir.llvm.org/docs/Dialects/Func/
[4] https://mlir.llvm.org/docs/Dialects/SCFDialect/