[clang][bytecode] Implement __builtin_reduce_or (#118292)

Author: tbaederr

clang/lib/AST/ByteCode/InterpBuiltin.cpp

@@ -104,6 +104,8 @@ template <typename T>
 static void pushInteger(InterpState &S, T Val, QualType QT) {
   if constexpr (std::is_same_v<T, APInt>)
     pushInteger(S, APSInt(Val, !std::is_signed_v<T>), QT);
+  else if constexpr (std::is_same_v<T, APSInt>)
+    pushInteger(S, Val, QT);
   else
     pushInteger(S,
                 APSInt(APInt(sizeof(T) * 8, static_cast<uint64_t>(Val),
@@ -1726,11 +1728,13 @@ static bool interp__builtin_vector_reduce(InterpState &S, CodePtr OpPC,
 
       } else if (ID == Builtin::BI__builtin_reduce_and) {
         (void)T::bitAnd(Result, Elem, BitWidth, &Result);
+      } else if (ID == Builtin::BI__builtin_reduce_or) {
+        (void)T::bitOr(Result, Elem, BitWidth, &Result);
       } else {
         llvm_unreachable("Unhandled vector reduce builtin");
       }
     }
-    pushInteger(S, Result, Call->getType());
+    pushInteger(S, Result.toAPSInt(), Call->getType());
   });
 
   return true;
@@ -2210,6 +2214,7 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
   case Builtin::BI__builtin_reduce_add:
   case Builtin::BI__builtin_reduce_mul:
   case Builtin::BI__builtin_reduce_and:
+  case Builtin::BI__builtin_reduce_or:
     if (!interp__builtin_vector_reduce(S, OpPC, Frame, F, Call))
       return false;
     break;

clang/test/AST/ByteCode/builtin-functions.cpp

@@ -1080,6 +1080,22 @@ namespace ReduceAnd {
 #endif
 }
 
+namespace ReduceOr {
+  static_assert(__builtin_reduce_or((vector4char){}) == 0);
+  static_assert(__builtin_reduce_or((vector4char){(char)0x11, (char)0x22, (char)0x44, (char)0x88}) == (char)0xFF);
+  static_assert(__builtin_reduce_or((vector4short){(short)0x1111, (short)0x2222, (short)0x4444, (short)0x8888}) == (short)0xFFFF);
+  static_assert(__builtin_reduce_or((vector4int){(int)0x11111111, (int)0x22222222, (int)0x44444444, (int)0x88888888}) == (int)0xFFFFFFFF);
+#if __INT_WIDTH__ == 32
+  static_assert(__builtin_reduce_or((vector4long){(long long)0x1111111111111111L, (long long)0x2222222222222222L, (long long)0x4444444444444444L, (long long)0x8888888888888888L}) == (long long)0xFFFFFFFFFFFFFFFFL);
+  static_assert(__builtin_reduce_or((vector4char){(char)0, (char)0x22, (char)0x44, (char)0x88}) == ~0x11);
+  static_assert(__builtin_reduce_or((vector4short){(short)0x1111, (short)0, (short)0x4444, (short)0x8888}) == ~0x2222);
+  static_assert(__builtin_reduce_or((vector4int){(int)0x11111111, (int)0x22222222, (int)0, (int)0x88888888}) == ~0x44444444);
+  static_assert(__builtin_reduce_or((vector4long){(long long)0x1111111111111111L, (long long)0x2222222222222222L, (long long)0x4444444444444444L, (long long)0}) == ~0x8888888888888888L);
+  static_assert(__builtin_reduce_or((vector4uint){0x11111111U, 0x22222222U, 0x44444444U, 0x88888888U}) == 0xFFFFFFFFU);
+  static_assert(__builtin_reduce_or((vector4ulong){0x1111111111111111UL, 0x2222222222222222UL, 0x4444444444444444UL, 0x8888888888888888UL}) == 0xFFFFFFFFFFFFFFFFL);
+#endif
+}
+
 namespace BuiltinMemcpy {
   constexpr int simple() {
     int a = 12;