[flang-commits] [flang] [flang][FIR] handle argument attributes in fir.call (PR #126711)

[Slava Zakharin via flang-commits]

================
@@ -2853,3 +2853,21 @@ gpu.module @cuda_device_mod {
 // CHECK: llvm.func @malloc(i64) -> !llvm.ptr
 // CHECK: llvm.call @malloc
 // CHECK: lvm.call @free
+
+// -----
+
+func.func private @somefunc(i32, !fir.ref<i64>)
+
+// CHECK-LABEL: @test_call_arg_attrs_direct
+func.func @test_call_arg_attrs_direct(%arg0: i32, %arg1: !fir.ref<i64>) {
+  // CHECK: llvm.call @somefunc(%{{.*}}, %{{.*}}) : (i32, !llvm.ptr {llvm.byval = i64}) -> ()
+  fir.call @somefunc(%arg0, %arg1) : (i32, !fir.ref<i64> {llvm.byval = i64}) -> ()
+  return
+}
+
+// CHECK-LABEL: @test_call_arg_attrs_indirect
+func.func @test_call_arg_attrs_indirect(%arg0: i16, %arg1: (i16)-> i16) -> i16 {
+  // CHECK: llvm.call %arg1(%{{.*}}) : !llvm.ptr, (i16 {llvm.signext}) -> (i16 {llvm.signext})
+  %0 = fir.call %arg1(%arg0) : (i16 {llvm.noundef, llvm.signext}) -> (i16 {llvm.signext})
----------------
vzakhari wrote:

Is it okay that `noundef` is dropped?

https://github.com/llvm/llvm-project/pull/126711