[clang] [CIR] Add custom assembly format for alloca op to fix flag parsing (PR #198962)

Henrich Lauko via cfe-commits cfe-commits at lists.llvm.org
Mon Jun 1 22:56:57 PDT 2026


================
@@ -653,13 +631,13 @@ def CIR_AllocaOp : CIR_Op<"alloca", [
   }];
 
   let assemblyFormat = [{
-    $allocaType `,` qualified(type($addr)) `,`
+    $name
+    `align` `(` $alignment `)`
+    oilist( `init`              $init
+          | `const`             $constant
+          | `cleanup_dest_slot` $cleanup_dest_slot)
     ($dynAllocSize^ `:` type($dynAllocSize) `,`)?
-    `[` $name
-       (`,` `init` $init^)?
-       (`,` `const` $constant^)?
-       (`,` `cleanup_dest_slot` $cleanup_dest_slot^)?
-    `]`
+    $allocaType `->` qualified(type($addr))
----------------
xlauko wrote:

> i tried a few things to drop the type of `dynAllocSize`:
> 
> * removing `type($dynAllocSize)` from the format gives: `error: type of operand #0, named 'dynAllocSize', is not buildable and a buildable type cannot be inferred`
> * changing to `Optional<UI64>` gives: `error: 'cir.alloca' op operand #0 must be 64-bit unsigned integer, but got '!cir.int<u, 64'`  since CIR uses `!cir.int<u, 64` which is different from the MLIR builtin `i64`
> * changing to `Optional<CIR_IntType>` gives the same buildable type error
> 
> i also confirmed that in practice it's always `!u64i` from looking at the generated CIR. am i missing something here? if not i think we'd need to keep the type in the format:
> 
> `%1 = cir.alloca "n" align(4) init size(%sz : !u64i) : !cir.ptr<!s32i`

You can then use `CIR_UInt64`, for type to be buildable it needs ne be buildable just from context. See `CIR_UInt` in `CIRTypeConstraints.td`.

https://github.com/llvm/llvm-project/pull/198962


More information about the cfe-commits mailing list