[clang] [clang][AMDGPU] precommit test for ballot on Windows (PR #73920)

[Valery Pykhtin via cfe-commits]

================
@@ -0,0 +1,27 @@
+// REQUIRES: amdgpu-registered-target
+// XFAIL: *
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -aux-triple x86_64-pc-windows-msvc -target-cpu gfx900 -x hip -emit-llvm -fcuda-is-device -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -aux-triple x86_64-pc-windows-msvc -target-cpu gfx900 -x hip -S -fcuda-is-device -o - %s | FileCheck %s --check-prefix=GFX9
+
+// Unlike OpenCL, HIP depends on the C++ interpration of "unsigned long", which
+// is 64 bits long on Linux and 32 bits long on Windows. The return type of the
+// ballot intrinsic needs to be a 64 bit integer on both platforms. This test
+// cross-compiles to Windows to confirm that the return type is indeed 64 bits
+// on Windows.
+
+// FIXME: The Clang declaration of the wave-64 builtin uses "UL" as the return
+// type, which is interpreted as a 32-bit unsigned integer on Windows. This
+// emits an incorrect LLVM declaration with i32 return type instead of i64. The
+// clang declaration needs to be fixed to use "WU" instead.
+
+// CHECK-LABEL: @_Z3fooi
+// CHECK: call i64 @llvm.amdgcn.ballot.i64
+
+// GFX9-LABEL: _Z3fooi:
+// GFX9: v_cmp_ne_u32_e64
----------------
vpykhtin wrote:

Is it really needed to test codegen here?

Nice find, BTW.

https://github.com/llvm/llvm-project/pull/73920