[clang] [CIR][X86] Implement convert_half builtins (PR #171615)

[Bruno Cardoso Lopes via cfe-commits]

================
@@ -0,0 +1,98 @@
+// RUN: %clang_cc1 -fclangir -triple x86_64-unknown-linux-gnu -target-feature +avx512fp16 -target-feature +avx512bf16 -emit-llvm %s -o - | FileCheck %s --check-prefix=CIR
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -target-feature +avx512fp16 -target-feature +avx512bf16 -emit-llvm %s -o - | FileCheck %s --check-prefix=OGCG
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -target-feature +avx512fp16 -target-feature +avx512bf16 -emit-llvm %s -o - | FileCheck %s --check-prefix=LLVM
+
+// REQUIRES: avx512fp16
+// REQUIRES: avx512bf16
----------------
bcardosolopes wrote:

Are you sure this is not just silently passing the test? I don't see this being used in clang for any test involving `avx512fp16`. It's also used twice, which is unnecessary. Can you please remove both?

https://github.com/llvm/llvm-project/pull/171615