[CIR][CodeGen] Replace LLVMIntrinsicCallOp with ACosOp in __builtin_elementwise_acos (#1507)

Closes: #1374

Replaces LLVMIntrinsicCallOp with ACosOp in __builtin_elementwise_acos.

Author: ayokunle321

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -4738,6 +4738,7 @@ class UnaryFPToFPBuiltinOp<string mnemonic, string llvmOpName>
 }
 
 def CeilOp : UnaryFPToFPBuiltinOp<"ceil", "FCeilOp">;
+def ACosOp : UnaryFPToFPBuiltinOp<"acos", "ACosOp">;
 def CosOp : UnaryFPToFPBuiltinOp<"cos", "CosOp">;
 def ExpOp : UnaryFPToFPBuiltinOp<"exp", "ExpOp">;
 def Exp2Op : UnaryFPToFPBuiltinOp<"exp2", "Exp2Op">;

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -1401,7 +1401,7 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
     return RValue::get(result);
   }
   case Builtin::BI__builtin_elementwise_acos: {
-    return emitBuiltinWithOneOverloadedType<1>(E, "acos");
+    return emitUnaryFPBuiltin<cir::ACosOp>(*this, *E);
   }
   case Builtin::BI__builtin_elementwise_asin:
     llvm_unreachable("BI__builtin_elementwise_asin NYI");

clang/lib/CIR/CodeGen/CIRGenFunction.h

@@ -1451,8 +1451,8 @@ class CIRGenFunction : public CIRGenTypeCache {
                          const clang::CallExpr *E, ReturnValueSlot ReturnValue);
   RValue emitRotate(const CallExpr *E, bool IsRotateRight);
   template <uint32_t N>
-  RValue emitBuiltinWithOneOverloadedType(const CallExpr *E,
-                                          llvm::StringRef Name) {
+  [[maybe_unused]] RValue
+  emitBuiltinWithOneOverloadedType(const CallExpr *E, llvm::StringRef Name) {
     static_assert(N, "expect non-empty argument");
     mlir::Type cirTy = convertType(E->getArg(0)->getType());
     SmallVector<mlir::Value, N> args;

clang/test/CIR/CodeGen/builtins-elementwise.c

@@ -41,19 +41,19 @@ void test_builtin_elementwise_acos(float f, double d, vfloat4 vf4,
                                    vdouble4  vd4) {
   // CIR-LABEL: test_builtin_elementwise_acos
   // LLVM-LABEL: test_builtin_elementwise_acos
-  // CIR: {{%.*}} = cir.llvm.intrinsic "acos" {{%.*}} : (!cir.float) -> !cir.float
+  // CIR: {{%.*}} = cir.acos {{%.*}} : !cir.float
   // LLVM: {{%.*}} = call float @llvm.acos.f32(float {{%.*}})
   f = __builtin_elementwise_acos(f);
 
-  // CIR: {{%.*}} = cir.llvm.intrinsic "acos" {{%.*}} : (!cir.double) -> !cir.double
+  // CIR: {{%.*}} = cir.acos {{%.*}} : !cir.double
   // LLVM: {{%.*}} = call double @llvm.acos.f64(double {{%.*}})
   d = __builtin_elementwise_acos(d);
 
-  // CIR: {{%.*}} = cir.llvm.intrinsic "acos" {{%.*}} : (!cir.vector<!cir.float x 4>) -> !cir.vector<!cir.float x 4>
+  // CIR: {{%.*}} = cir.acos {{%.*}} : !cir.vector<!cir.float x 4>
   // LLVM: {{%.*}} = call <4 x float> @llvm.acos.v4f32(<4 x float> {{%.*}})
   vf4 = __builtin_elementwise_acos(vf4);
 
-  // CIR: {{%.*}} = cir.llvm.intrinsic "acos" {{%.*}} : (!cir.vector<!cir.double x 4>) -> !cir.vector<!cir.double x 4>
+  // CIR: {{%.*}} = cir.acos {{%.*}} : !cir.vector<!cir.double x 4>
   // LLVM: {{%.*}} = call <4 x double> @llvm.acos.v4f64(<4 x double> {{%.*}})
   vd4 = __builtin_elementwise_acos(vd4);
 }