[flang-commits] [flang] [flang][codegen] Update FIR codegen to use mlir.llvm opaque pointers (PR #69692)

[via flang-commits]

================
@@ -11,93 +11,93 @@ func.func private @foo0(%arg0: !fir.array<10x12xi64>)
 // CHECK-SAME: !llvm.array<12 x array<10 x i64>>
 func.func private @foo1(%arg0: !fir.array<?xf32>)
 // CHECK-LABEL: foo1
-// CHECK-SAME: !llvm.ptr<f32>
+// CHECK-SAME: f32
----------------
jeanPerier wrote:

This test makes no sense from an IR point view, it is only meant to test type conversions.

One should not create an SSA value with a fir.array value with non constant extents without wrapping it into a memory reference. The only point of this test was/is to test fir.array type converter alone. We are checking that we are building the "most complete" llvm value type from an fir.array (we want to keep inner constant dimensions into llvm as much as possible).

I would happily wrap that into a fir.ptr to make legal IR, but then I would only be testing llvm.ptr, which is less interesting.

https://github.com/llvm/llvm-project/pull/69692