[clang] [CIR][AArch64] Upstream narrowing-subtraction NEON builtins (PR #207115)

[Vicky Nguyen via cfe-commits]

================
@@ -3,6 +3,9 @@
 // RUN:                   %clang_cc1_cg_arm64_neon           -emit-llvm %s -disable-O0-optnone | opt -S -passes=mem2reg,sroa | FileCheck %s --check-prefixes=LLVM
 // RUN: %if cir-enabled %{%clang_cc1_cg_arm64_neon -fclangir -emit-llvm %s -disable-O0-optnone | opt -S -passes=mem2reg,sroa | FileCheck %s --check-prefixes=LLVM %}
 // RUN: %if cir-enabled %{%clang_cc1_cg_arm64_neon -fclangir -emit-cir  %s -disable-O0-optnone |                               FileCheck %s --check-prefixes=CIR %}
+// Narrowing-subtraction checks use instcombine to fold no-op bitcasts (LLVM-IC prefix).
+// RUN:                   %clang_cc1_cg_arm64_neon           -emit-llvm %s -disable-O0-optnone | opt -S -passes=mem2reg,sroa,instcombine | FileCheck %s --check-prefixes=LLVM-IC
+// RUN: %if cir-enabled %{%clang_cc1_cg_arm64_neon -fclangir -emit-llvm %s -disable-O0-optnone | opt -S -passes=mem2reg,sroa,instcombine | FileCheck %s --check-prefixes=LLVM-IC %}
----------------
iamvickynguyen wrote:

I'll create a follow-up PR that applies `instcombine` globally and fix other checks.

https://github.com/llvm/llvm-project/pull/207115