[flang] Allow assumed-shape element pass to dummy arg with ignore_tkr

[tblah]

This is allowed by gfortran and ifort with ![GCC|DEC]$ ATTRIBUTES NO_ARG_CHECK

[tblah]

I'm not sure if ignore_tkr(r) is the right specifier for this, maybe there should be a new one? e.g. ignore_tkr(s)

[llvmbot]

@llvm/pr-subscribers-flang-semantics

Author: Tom Eccles (tblah)

Changes

This is allowed by gfortran and ifort with ![GCC|DEC]$ ATTRIBUTES NO_ARG_CHECK

---

Full diff: https://github.com/llvm/llvm-project/pull/78196.diff

2 Files Affected:

• (modified) flang/lib/Semantics/check-call.cpp (+2-1)
• (added) flang/test/Semantics/ignore_tkr03.f90 (+18)

diff --git a/flang/lib/Semantics/check-call.cpp b/flang/lib/Semantics/check-call.cpp
index ec8f99ca6bf48e..ba8903089f836d 100644
--- a/flang/lib/Semantics/check-call.cpp
+++ b/flang/lib/Semantics/check-call.cpp
@@ -535,7 +535,8 @@ static void CheckExplicitDataArg(const characteristics::DummyDataObject &dummy,
             messages.Say(
                 "Element of pointer array may not be associated with a %s array"_err_en_US,
                 dummyName);
-          } else if (IsAssumedShape(*actualLastSymbol)) {
+          } else if (IsAssumedShape(*actualLastSymbol) &&
+              !dummy.ignoreTKR.test(common::IgnoreTKR::Rank)) {
             basicError = true;
             messages.Say(
                 "Element of assumed-shape array may not be associated with a %s array"_err_en_US,
diff --git a/flang/test/Semantics/ignore_tkr03.f90 b/flang/test/Semantics/ignore_tkr03.f90
new file mode 100644
index 00000000000000..25182d879e1097
--- /dev/null
+++ b/flang/test/Semantics/ignore_tkr03.f90
@@ -0,0 +1,18 @@
+! RUN: %flang_fc1 -fdebug-unparse %s 2>&1 | FileCheck %s
+module library
+contains
+  subroutine lib_sub(buf)
+!dir$ ignore_tkr(r) buf
+    real :: buf(1:*)
+  end subroutine
+end module
+
+module user
+  use library
+contains
+  subroutine sub(var)
+    real :: var(:,:,:)
+! CHECK: CALL lib_sub
+    call lib_sub(var(1, 2, 3))
+  end subroutine
+end module

[klausler]

I don't think that 'R' is the right letter to use here. It's not about rank, it's about storage sequence association with a potentially non-contiguous object. Besides assumed-shape arrays, you probably also want to allow non-CONTIGUOUS pointers.

[tblah]

Thanks for the review. I repurposed the (previously legacy) ignore_tkr(c).

[klausler]

Thanks for the review. I repurposed the (previously legacy) ignore_tkr(c).

That might conflict with legacy codes still being compiled by nvfortran. I'll check.

[klausler]

Thanks for the review. I repurposed the (previously legacy) ignore_tkr(c).

That might conflict with legacy codes still being compiled by nvfortran. I'll check.

The legacy meaning of C is your interpretation here, so I think it's okay.

[tblah]

@klausler thanks for checking my use of C.

As C is okay, is there anything blocking merging this patch?

[klausler]

Ship it with my thanks.