[clang] [llvm] [AArch64][SVE] Fix definition of bfloat fcvt intrinsics. (PR #110281)

[David Sherwood via cfe-commits]

================
@@ -4072,6 +4078,30 @@ static Value *upgradeX86IntrinsicCall(StringRef Name, CallBase *CI, Function *F,
   return Rep;
 }
 
+static Value *upgradeAArch64IntrinsicCall(StringRef Name, CallBase *CI,
+                                          Function *F, IRBuilder<> &Builder) {
+  Intrinsic::ID NewID =
+      StringSwitch<Intrinsic::ID>(Name)
+          .Case("sve.fcvt.bf16f32", Intrinsic::aarch64_sve_fcvt_bf16f32_v2)
+          .Case("sve.fcvtnt.bf16f32", Intrinsic::aarch64_sve_fcvtnt_bf16f32_v2)
+          .Default(Intrinsic::not_intrinsic);
+  if (NewID == Intrinsic::not_intrinsic)
+    llvm_unreachable("Unhandled Intrinsic!");
+
+  SmallVector<Value *, 3> Args(CI->args());
+
+  // The original intrinsics incorrectly used a predicate based on the smallest
+  // element type rather than the largest.
+  Type *PredTy = ScalableVectorType::get(Builder.getInt1Ty(), 4);
+  Args[1] = Builder.CreateIntrinsic(Intrinsic::aarch64_sve_convert_to_svbool,
----------------
david-arm wrote:

Given we won't get any IR verification of the old intrinsic, is it worth asserting or reporting an error if the predicate isn't <vscale x 8 x i1> here?

https://github.com/llvm/llvm-project/pull/110281