[Clang] Add elementwise saturated add/sub builtins

[RKSimon]

Summary

This patch implements __builtin_elementwise_add_sat and __builtin_elementwise_sub_sat builtins

These map to the add/sub saturated math intrinsics described here:
https://llvm.org/docs/LangRef.html#saturation-arithmetic-intrinsics

With this in place we should then be able to replace the x86 SSE adds/subs intrinsics with these generic variants - it looks like other targets should be able to use these as well (arm/aarch64/webassembly all have similar examples in cgbuiltin).

[craig.topper]

clang/docs/LanguageExtensions.rst

549

Not sure if I'm reading this right due to the columns, but is "unsigned" missing after the "signed or"

clang/include/clang/Basic/Builtins.def

655

I don't think these are in alphabetical order. ceil is after min. Looks more like they were grouped by similar operations.

clang/lib/Sema/SemaChecking.cpp

2255

clang-format

[RKSimon]

address feedback

[RKSimon]

ping?

[fhahn]

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

122

It might be good to have tests where one argument is signed and the other unsigned as well. Those appear to be missing for other builtins as well unfortunately.

[RKSimon]

rebase and add signed/unsigned integer mismatch tests

[aaron.ballman]

clang/docs/LanguageExtensions.rst

549

This reads strangely to me as well. "..., clamped to the range of signed or integer types unsigned values representable by.."

clang/lib/CodeGen/CGBuiltin.cpp

3157–3184

Almost all of this logic is shared (except for picking the intrinsic), should it be combined?

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

122

I think it'd be good to have some codegen tests for _BitInt to make sure the behavior is correct.

[RKSimon]

rebase, simplify description in documentation and add _BitInt test coverage

[aaron.ballman]

Aside from some nits, this LGTM.

clang/docs/LanguageExtensions.rst

549

This still seems unaddressed.

clang/lib/CodeGen/CGBuiltin.cpp

3157–3184

Same with this one.

[RKSimon]

clang/docs/LanguageExtensions.rst

549

I rephrased it - what did you have in mind?

[aaron.ballman]

clang/docs/LanguageExtensions.rst

549

I think what's been throwing me off is "range of integer types signed or unsigned values", but I don't think signed or unsigned really matters here, so I tried to reword it a bit. Does this work for you?

return the sum of x and y, clamped to the range of representable values for the integer type.

(Similar below for difference).

clang/docs/LanguageExtensions.rst

  T __builtin_elementwise_roundeven(T x)    round x to the nearest integer value in floating point format,   floating point types
                                            rounding halfway cases to even (that is, to the nearest value
                                            that is an even integer), regardless of the current rounding
                                            direction.
  T__builtin_elementwise_trunc(T x)         return the integral value nearest to but no larger in            floating point types
                                            magnitude than x
  T __builtin_elementwise_max(T x, T y)     return x or y, whichever is larger                               integer and floating point types
  T __builtin_elementwise_min(T x, T y)     return x or y, whichever is smaller                              integer and floating point types
+ T __builtin_elementwise_add_sat(T x, T y) return the sum of x and y, clamped to the range of               integer types
+                                           representable values for the signed/unsigned integer type.
+ T __builtin_elementwise_sub_sat(T x, T y) return the difference of x and y, clamped to the range of        integer types
+                                           representable values for the signed/unsigned integer type..
 ========================================= ================================================================ =========================================
 
 
 *Reduction Builtins*
 
 Each builtin returns a scalar equivalent to applying the specified
 operation(x, y) as recursive even-odd pairwise reduction to all vector
 elements. ``operation(x, y)`` is repeatedly applied to each non-overlapping

clang/include/clang/Basic/Builtins.def

 BUILTIN(__builtin_call_with_static_chain, "v.", "nt")
 
 BUILTIN(__builtin_elementwise_abs, "v.", "nct")
 BUILTIN(__builtin_elementwise_max, "v.", "nct")
 BUILTIN(__builtin_elementwise_min, "v.", "nct")
 BUILTIN(__builtin_elementwise_ceil, "v.", "nct")
 BUILTIN(__builtin_elementwise_floor, "v.", "nct")
 BUILTIN(__builtin_elementwise_roundeven, "v.", "nct")
 BUILTIN(__builtin_elementwise_trunc, "v.", "nct")
+BUILTIN(__builtin_elementwise_add_sat, "v.", "nct")
+BUILTIN(__builtin_elementwise_sub_sat, "v.", "nct")
 BUILTIN(__builtin_reduce_max, "v.", "nct")
 BUILTIN(__builtin_reduce_min, "v.", "nct")
 BUILTIN(__builtin_reduce_xor, "v.", "nct")
 BUILTIN(__builtin_reduce_or, "v.", "nct")
 BUILTIN(__builtin_reduce_and, "v.", "nct")
 
 BUILTIN(__builtin_matrix_transpose, "v.", "nFt")
 BUILTIN(__builtin_matrix_column_major_load, "v.", "nFt")

clang/lib/CodeGen/CGBuiltin.cpp

return RValue::get(
         emitUnaryBuiltin(*this, E, llvm::Intrinsic::floor, "elt.floor"));
   case Builtin::BI__builtin_elementwise_roundeven:
     return RValue::get(emitUnaryBuiltin(*this, E, llvm::Intrinsic::roundeven,
                                         "elt.roundeven"));
   case Builtin::BI__builtin_elementwise_trunc:
     return RValue::get(
         emitUnaryBuiltin(*this, E, llvm::Intrinsic::trunc, "elt.trunc"));
 
+  case Builtin::BI__builtin_elementwise_add_sat:
+  case Builtin::BI__builtin_elementwise_sub_sat: {
+    Value *Op0 = EmitScalarExpr(E->getArg(0));
+    Value *Op1 = EmitScalarExpr(E->getArg(1));
+    Value *Result;
+    assert(Op0->getType()->isIntOrIntVectorTy() && "integer type expected");
+    QualType Ty = E->getArg(0)->getType();
+    if (auto *VecTy = Ty->getAs<VectorType>())
+      Ty = VecTy->getElementType();
+    bool IsSigned = Ty->isSignedIntegerType();
+    unsigned Opc;
+    if (BuiltinIDIfNoAsmLabel == Builtin::BI__builtin_elementwise_add_sat)
+      Opc = IsSigned ? llvm::Intrinsic::sadd_sat : llvm::Intrinsic::uadd_sat;
+    else
+      Opc = IsSigned ? llvm::Intrinsic::ssub_sat : llvm::Intrinsic::usub_sat;
+    Result = Builder.CreateBinaryIntrinsic(Opc, Op0, Op1, nullptr, "elt.sat");
+    return RValue::get(Result);
+  }
+
   case Builtin::BI__builtin_elementwise_max: {
     Value *Op0 = EmitScalarExpr(E->getArg(0));
     Value *Op1 = EmitScalarExpr(E->getArg(1));
     Value *Result;
     if (Op0->getType()->isIntOrIntVectorTy()) {
       QualType Ty = E->getArg(0)->getType();
       if (auto *VecTy = Ty->getAs<VectorType>())
         Ty = VecTy->getElementType();
       Result = Builder.CreateBinaryIntrinsic(Ty->isSignedIntegerType()
                                                  ? llvm::Intrinsic::smax
                                                  : llvm::Intrinsic::umax,
                                              Op0, Op1, nullptr, "elt.max");
     } else
       Result = Builder.CreateMaxNum(Op0, Op1, "elt.max");
     return RValue::get(Result);
   }
   case Builtin::BI__builtin_elementwise_min: {

clang/lib/Sema/SemaChecking.cpp

if (!EltTy->isFloatingType()) {
            diag::err_builtin_invalid_arg_type)
           << 1 << /* float ty*/ 5 << ArgTy;
 
       return ExprError();
     }
     break;
   }
 
+  // These builtins restrict the element type to integer
+  // types only.
+  case Builtin::BI__builtin_elementwise_add_sat:
+  case Builtin::BI__builtin_elementwise_sub_sat: {
+    if (SemaBuiltinElementwiseMath(TheCall))
+      return ExprError();
+
+    const Expr *Arg = TheCall->getArg(0);
+    QualType ArgTy = Arg->getType();
+    QualType EltTy = ArgTy;
+
+    if (auto *VecTy = EltTy->getAs<VectorType>())
+      EltTy = VecTy->getElementType();
+
+    if (!EltTy->isIntegerType()) {
+      Diag(Arg->getBeginLoc(), diag::err_builtin_invalid_arg_type)
+          << 1 << /* integer ty */ 6 << ArgTy;
+      return ExprError();
+    }
+    break;
+  }
+
   case Builtin::BI__builtin_elementwise_min:
   case Builtin::BI__builtin_elementwise_max:
     if (SemaBuiltinElementwiseMath(TheCall))
       return ExprError();
     break;
   case Builtin::BI__builtin_reduce_max:
   case Builtin::BI__builtin_reduce_min: {
     if (PrepareBuiltinReduceMathOneArgCall(TheCall))

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

void test_builtin_elementwise_abs(float f1, float f2, double d1, double d2,
 
   // CHECK:      [[SI:%.+]] = load i16, i16* %si.addr, align 2
   // CHECK-NEXT: [[SI_EXT:%.+]] = sext i16 [[SI]] to i32
   // CHECK-NEXT: [[RES:%.+]] = call i32 @llvm.abs.i32(i32 [[SI_EXT]], i1 false)
   // CHECK-NEXT: = trunc i32 [[RES]] to i16
   si = __builtin_elementwise_abs(si);
 }
 
+void test_builtin_elementwise_add_sat(float f1, float f2, double d1, double d2,
+                                      float4 vf1, float4 vf2, long long int i1,
+                                      long long int i2, si8 vi1, si8 vi2,
+                                      unsigned u1, unsigned u2, u4 vu1, u4 vu2,
+                                      _BitInt(31) bi1, _BitInt(31) bi2,
+                                      unsigned _BitInt(55) bu1, unsigned _BitInt(55) bu2) {
+  // CHECK:      [[I1:%.+]] = load i64, i64* %i1.addr, align 8
+  // CHECK-NEXT: [[I2:%.+]] = load i64, i64* %i2.addr, align 8
+  // CHECK-NEXT: call i64 @llvm.sadd.sat.i64(i64 [[I1]], i64 [[I2]])
+  i1 = __builtin_elementwise_add_sat(i1, i2);
+
+  // CHECK:      [[I1:%.+]] = load i64, i64* %i1.addr, align 8
+  // CHECK-NEXT: call i64 @llvm.sadd.sat.i64(i64 [[I1]], i64 10)
+  i1 = __builtin_elementwise_add_sat(i1, 10);
+
+  // CHECK:      [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: [[VI2:%.+]] = load <8 x i16>, <8 x i16>* %vi2.addr, align 16
+  // CHECK-NEXT: call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[VI1]], <8 x i16> [[VI2]])
+  vi1 = __builtin_elementwise_add_sat(vi1, vi2);
+
+  // CHECK:      [[U1:%.+]] = load i32, i32* %u1.addr, align 4
+  // CHECK-NEXT: [[U2:%.+]] = load i32, i32* %u2.addr, align 4
+  // CHECK-NEXT: call i32 @llvm.uadd.sat.i32(i32 [[U1]], i32 [[U2]])
+  u1 = __builtin_elementwise_add_sat(u1, u2);
+
+  // CHECK:      [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: [[VU2:%.+]] = load <4 x i32>, <4 x i32>* %vu2.addr, align 16
+  // CHECK-NEXT: call <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32> [[VU1]], <4 x i32> [[VU2]])
+  vu1 = __builtin_elementwise_add_sat(vu1, vu2);
+
+  // CHECK:      [[BI1:%.+]] = load i31, i31* %bi1.addr, align 4
+  // CHECK-NEXT: [[BI2:%.+]] = load i31, i31* %bi2.addr, align 4
+  // CHECK-NEXT: call i31 @llvm.sadd.sat.i31(i31 [[BI1]], i31 [[BI2]])
+  bi1 = __builtin_elementwise_add_sat(bi1, bi2);
+
+  // CHECK:      [[BU1:%.+]] = load i55, i55* %bu1.addr, align 8
+  // CHECK-NEXT: [[BU2:%.+]] = load i55, i55* %bu2.addr, align 8
+  // CHECK-NEXT: call i55 @llvm.uadd.sat.i55(i55 [[BU1]], i55 [[BU2]])
+  bu1 = __builtin_elementwise_add_sat(bu1, bu2);
+
+  // CHECK:      [[IAS1:%.+]] = load i32, i32 addrspace(1)* @int_as_one, align 4
+  // CHECK-NEXT: [[B:%.+]] = load i32, i32* @b, align 4
+  // CHECK-NEXT: call i32 @llvm.sadd.sat.i32(i32 [[IAS1]], i32 [[B]])
+  int_as_one = __builtin_elementwise_add_sat(int_as_one, b);
+
+  // CHECK: call i32 @llvm.sadd.sat.i32(i32 1, i32 97)
+  i1 = __builtin_elementwise_add_sat(1, 'a');
+}
+
+void test_builtin_elementwise_sub_sat(float f1, float f2, double d1, double d2,
+                                      float4 vf1, float4 vf2, long long int i1,
+                                      long long int i2, si8 vi1, si8 vi2,
+                                      unsigned u1, unsigned u2, u4 vu1, u4 vu2,
+                                      _BitInt(31) bi1, _BitInt(31) bi2,
+                                      unsigned _BitInt(55) bu1, unsigned _BitInt(55) bu2) {
+  // CHECK:      [[I1:%.+]] = load i64, i64* %i1.addr, align 8
+  // CHECK-NEXT: [[I2:%.+]] = load i64, i64* %i2.addr, align 8
+  // CHECK-NEXT: call i64 @llvm.ssub.sat.i64(i64 [[I1]], i64 [[I2]])
+  i1 = __builtin_elementwise_sub_sat(i1, i2);
+
+  // CHECK:      [[I1:%.+]] = load i64, i64* %i1.addr, align 8
+  // CHECK-NEXT: call i64 @llvm.ssub.sat.i64(i64 [[I1]], i64 10)
+  i1 = __builtin_elementwise_sub_sat(i1, 10);
+
+  // CHECK:      [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: [[VI2:%.+]] = load <8 x i16>, <8 x i16>* %vi2.addr, align 16
+  // CHECK-NEXT: call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> [[VI1]], <8 x i16> [[VI2]])
+  vi1 = __builtin_elementwise_sub_sat(vi1, vi2);
+
+  // CHECK:      [[U1:%.+]] = load i32, i32* %u1.addr, align 4
+  // CHECK-NEXT: [[U2:%.+]] = load i32, i32* %u2.addr, align 4
+  // CHECK-NEXT: call i32 @llvm.usub.sat.i32(i32 [[U1]], i32 [[U2]])
+  u1 = __builtin_elementwise_sub_sat(u1, u2);
+
+  // CHECK:      [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: [[VU2:%.+]] = load <4 x i32>, <4 x i32>* %vu2.addr, align 16
+  // CHECK-NEXT: call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> [[VU1]], <4 x i32> [[VU2]])
+  vu1 = __builtin_elementwise_sub_sat(vu1, vu2);
+
+  // CHECK:      [[BI1:%.+]] = load i31, i31* %bi1.addr, align 4
+  // CHECK-NEXT: [[BI2:%.+]] = load i31, i31* %bi2.addr, align 4
+  // CHECK-NEXT: call i31 @llvm.ssub.sat.i31(i31 [[BI1]], i31 [[BI2]])
+  bi1 = __builtin_elementwise_sub_sat(bi1, bi2);
+
+  // CHECK:      [[BU1:%.+]] = load i55, i55* %bu1.addr, align 8
+  // CHECK-NEXT: [[BU2:%.+]] = load i55, i55* %bu2.addr, align 8
+  // CHECK-NEXT: call i55 @llvm.usub.sat.i55(i55 [[BU1]], i55 [[BU2]])
+  bu1 = __builtin_elementwise_sub_sat(bu1, bu2);
+
+  // CHECK:      [[IAS1:%.+]] = load i32, i32 addrspace(1)* @int_as_one, align 4
+  // CHECK-NEXT: [[B:%.+]] = load i32, i32* @b, align 4
+  // CHECK-NEXT: call i32 @llvm.ssub.sat.i32(i32 [[IAS1]], i32 [[B]])
+  int_as_one = __builtin_elementwise_sub_sat(int_as_one, b);
+
+  // CHECK: call i32 @llvm.ssub.sat.i32(i32 1, i32 97)
+  i1 = __builtin_elementwise_sub_sat(1, 'a');
+}
+
 void test_builtin_elementwise_max(float f1, float f2, double d1, double d2,
                                   float4 vf1, float4 vf2, long long int i1,
                                   long long int i2, si8 vi1, si8 vi2,
                                   unsigned u1, unsigned u2, u4 vu1, u4 vu2,
                                   _BitInt(31) bi1, _BitInt(31) bi2,
                                   unsigned _BitInt(55) bu1, unsigned _BitInt(55) bu2) {
   // CHECK-LABEL: define void @test_builtin_elementwise_max(
   // CHECK:      [[F1:%.+]] = load float, float* %f1.addr, align 4

clang/test/Sema/builtins-elementwise-math.c

void test_builtin_elementwise_abs(int i, double d, float4 v, int3 iv, unsigned u, unsigned4 uv) {
 
   u = __builtin_elementwise_abs(u);
   // expected-error@-1 {{1st argument must be a signed integer or floating point type (was 'unsigned int')}}
 
   uv = __builtin_elementwise_abs(uv);
   // expected-error@-1 {{1st argument must be a signed integer or floating point type (was 'unsigned4' (vector of 4 'unsigned int' values))}}
 }
 
+void test_builtin_elementwise_add_sat(int i, short s, double d, float4 v, int3 iv, unsigned3 uv, int *p) {
+  i = __builtin_elementwise_add_sat(p, d);
+  // expected-error@-1 {{arguments are of different types ('int *' vs 'double')}}
+
+  struct Foo foo = __builtin_elementwise_add_sat(i, i);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_elementwise_add_sat(i);
+  // expected-error@-1 {{too few arguments to function call, expected 2, have 1}}
+
+  i = __builtin_elementwise_add_sat();
+  // expected-error@-1 {{too few arguments to function call, expected 2, have 0}}
+
+  i = __builtin_elementwise_add_sat(i, i, i);
+  // expected-error@-1 {{too many arguments to function call, expected 2, have 3}}
+
+  i = __builtin_elementwise_add_sat(v, iv);
+  // expected-error@-1 {{arguments are of different types ('float4' (vector of 4 'float' values) vs 'int3' (vector of 3 'int' values))}}
+
+  i = __builtin_elementwise_add_sat(uv, iv);
+  // expected-error@-1 {{arguments are of different types ('unsigned3' (vector of 3 'unsigned int' values) vs 'int3' (vector of 3 'int' values))}}
+
+  v = __builtin_elementwise_add_sat(v, v);
+  // expected-error@-1 {{1st argument must be a vector of integers (was 'float4' (vector of 4 'float' values))}}
+
+  s = __builtin_elementwise_add_sat(i, s);
+
+  enum e { one,
+           two };
+  i = __builtin_elementwise_add_sat(one, two);
+
+  enum f { three };
+  enum f x = __builtin_elementwise_add_sat(one, three);
+
+  _BitInt(32) ext; // expected-warning {{'_BitInt' in C17 and earlier is a Clang extension}}
+  ext = __builtin_elementwise_add_sat(ext, ext);
+
+  const int ci;
+  i = __builtin_elementwise_add_sat(ci, i);
+  i = __builtin_elementwise_add_sat(i, ci);
+  i = __builtin_elementwise_add_sat(ci, ci);
+
+  i = __builtin_elementwise_add_sat(i, int_as_one); // ok (attributes don't match)?
+  i = __builtin_elementwise_add_sat(i, b);          // ok (sugar doesn't match)?
+
+  int A[10];
+  A = __builtin_elementwise_add_sat(A, A);
+  // expected-error@-1 {{1st argument must be a vector, integer or floating point type (was 'int *')}}
+
+  int(ii);
+  int j;
+  j = __builtin_elementwise_add_sat(i, j);
+
+  _Complex float c1, c2;
+  c1 = __builtin_elementwise_add_sat(c1, c2);
+  // expected-error@-1 {{1st argument must be a vector, integer or floating point type (was '_Complex float')}}
+}
+
+void test_builtin_elementwise_sub_sat(int i, short s, double d, float4 v, int3 iv, unsigned3 uv, int *p) {
+  i = __builtin_elementwise_sub_sat(p, d);
+  // expected-error@-1 {{arguments are of different types ('int *' vs 'double')}}
+
+  struct Foo foo = __builtin_elementwise_sub_sat(i, i);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_elementwise_sub_sat(i);
+  // expected-error@-1 {{too few arguments to function call, expected 2, have 1}}
+
+  i = __builtin_elementwise_sub_sat();
+  // expected-error@-1 {{too few arguments to function call, expected 2, have 0}}
+
+  i = __builtin_elementwise_sub_sat(i, i, i);
+  // expected-error@-1 {{too many arguments to function call, expected 2, have 3}}
+
+  i = __builtin_elementwise_sub_sat(v, iv);
+  // expected-error@-1 {{arguments are of different types ('float4' (vector of 4 'float' values) vs 'int3' (vector of 3 'int' values))}}
+
+  i = __builtin_elementwise_sub_sat(uv, iv);
+  // expected-error@-1 {{arguments are of different types ('unsigned3' (vector of 3 'unsigned int' values) vs 'int3' (vector of 3 'int' values))}}
+
+  v = __builtin_elementwise_sub_sat(v, v);
+  // expected-error@-1 {{1st argument must be a vector of integers (was 'float4' (vector of 4 'float' values))}}
+
+  s = __builtin_elementwise_sub_sat(i, s);
+
+  enum e { one,
+           two };
+  i = __builtin_elementwise_sub_sat(one, two);
+
+  enum f { three };
+  enum f x = __builtin_elementwise_sub_sat(one, three);
+
+  _BitInt(32) ext; // expected-warning {{'_BitInt' in C17 and earlier is a Clang extension}}
+  ext = __builtin_elementwise_sub_sat(ext, ext);
+
+  const int ci;
+  i = __builtin_elementwise_sub_sat(ci, i);
+  i = __builtin_elementwise_sub_sat(i, ci);
+  i = __builtin_elementwise_sub_sat(ci, ci);
+
+  i = __builtin_elementwise_sub_sat(i, int_as_one); // ok (attributes don't match)?
+  i = __builtin_elementwise_sub_sat(i, b);          // ok (sugar doesn't match)?
+
+  int A[10];
+  A = __builtin_elementwise_sub_sat(A, A);
+  // expected-error@-1 {{1st argument must be a vector, integer or floating point type (was 'int *')}}
+
+  int(ii);
+  int j;
+  j = __builtin_elementwise_sub_sat(i, j);
+
+  _Complex float c1, c2;
+  c1 = __builtin_elementwise_sub_sat(c1, c2);
+  // expected-error@-1 {{1st argument must be a vector, integer or floating point type (was '_Complex float')}}
+}
+
 void test_builtin_elementwise_max(int i, short s, double d, float4 v, int3 iv, unsigned3 uv, int *p) {
   i = __builtin_elementwise_max(p, d);
   // expected-error@-1 {{arguments are of different types ('int *' vs 'double')}}
 
   struct Foo foo = __builtin_elementwise_max(i, i);
   // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
 
   i = __builtin_elementwise_max(i);

clang/test/SemaCXX/builtins-elementwise-math.cpp

 
 void test_builtin_elementwise_abs() {
   const int a = 2;
   int b = 1;
   static_assert(!is_const<decltype(__builtin_elementwise_abs(a))>::value);
   static_assert(!is_const<decltype(__builtin_elementwise_abs(b))>::value);
 }
 
+void test_builtin_elementwise_add_sat() {
+  const int a = 2;
+  int b = 1;
+  static_assert(!is_const<decltype(__builtin_elementwise_add_sat(a, b))>::value);
+  static_assert(!is_const<decltype(__builtin_elementwise_add_sat(b, a))>::value);
+  static_assert(!is_const<decltype(__builtin_elementwise_add_sat(a, a))>::value);
+}
+
+void test_builtin_elementwise_sub_sat() {
+  const int a = 2;
+  int b = 1;
+  static_assert(!is_const<decltype(__builtin_elementwise_sub_sat(a, b))>::value);
+  static_assert(!is_const<decltype(__builtin_elementwise_sub_sat(b, a))>::value);
+  static_assert(!is_const<decltype(__builtin_elementwise_sub_sat(a, a))>::value);
+}
+
 void test_builtin_elementwise_max() {
   const int a = 2;
   int b = 1;
   static_assert(!is_const<decltype(__builtin_elementwise_max(a, b))>::value);
   static_assert(!is_const<decltype(__builtin_elementwise_max(b, a))>::value);
   static_assert(!is_const<decltype(__builtin_elementwise_max(a, a))>::value);
 }