[Clang] Add elementwise abs builtin.

[fhahn]

Summary

This patch implements __builtin_elementwise_abs as specified in
D111529.

[fhahn]

polish tests a bit

[craig.topper]

clang/lib/CodeGen/CGBuiltin.cpp

3109

Did we discuss that this is different than __builtin_abs which is undefined for INT_MIN?

[craig.topper]

clang/lib/CodeGen/CGBuiltin.cpp

3109

I don't think we add nsw to vector signed integer subtract, so I guess using false here is consistent.

[fhahn]

Pass is_int_min_poison=true to llvm.abs.* to reflect latest update to the specification to make taking abs of the most negative integer undefined.

[fhahn]

clang/lib/CodeGen/CGBuiltin.cpp

3109

Did we discuss that this is different than __builtin_abs which is undefined for INT_MIN?

So far it was not explicitly called out in the definition. I think we should explicitly make it undefined, so I updated the spec in D111529 and passed true here.

[scanon]

What's the rationale for making abs undefined on the minimum value? AFAIK every actual simd implementation defines the result and they agree (and even if one didn't, it would be pretty easy to get the "right" result. Introducing UB here just seems like punishing users for no reason.

[fhahn]

In D111986#3072946, @scanon wrote:

What's the rationale for making abs undefined on the minimum value? AFAIK every actual simd implementation defines the result and they agree (and even if one didn't, it would be pretty easy to get the "right" result. Introducing UB here just seems like punishing users for no reason.

Thanks! As making it UB does not really help with lowering I went back to allowing INT_MIN and updated the spec to explicitly spell out the behavior.

[scanon]

Two minor questions, but also LGTM as is.

clang/lib/Sema/SemaChecking.cpp

16667

Given that I expect this particular test to occur fairly frequently, maybe worth abstracting into some sort of get-elementwise-type operation.

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

25

For the purposes of C++ templates it might be nice to allow abs on unsigned (as the identity function). I don't have strong feelings though, and a library wrapping the builtins can do this themselves.

[fhahn]

Use checkMathBuiltinElementType helper added in D111985.

[fhahn]

clang/lib/Sema/SemaChecking.cpp

16667

moved out to a helper in the previous patch in the series, D111985

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

25

yeah that would be an option. I don't have any strong feelings either

[fhahn]

Rebased after recent changes in D111985: changes the return type of SemaBuiltinElementwiseMathOneArg to bool and drop the CallResult argument. Also added tests with const, typedef'ed and variables with address space

[aaron.ballman]

clang/include/clang/Basic/DiagnosticSemaKinds.td

8757

I feel like we must already have a diagnostic that covers this case...

clang/lib/Sema/SemaChecking.cpp

16664

Should this type undergo the usual promotions?

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

25

I was wondering about this myself, but also don't have strong opinions.

[fhahn]

Rebased and updated to use err_builtin_invalid_arg_type.

clang/include/clang/Basic/DiagnosticSemaKinds.td

8757

Updated to use err_builtin_invalid_arg_type.

clang/lib/Sema/SemaChecking.cpp

16664

I'm not sure, but given that we only have a single argument then wouldn't it be sufficient to avoid promotion? I don't think promotion to wider types would impact the results of the provided builtins.

[aaron.ballman]

clang/lib/Sema/SemaChecking.cpp

16664

You set the type of the call to be the type of the argument, which means passing in a const int will result in a const int that's observable and probably unexpected. e.g. this will fail,

const int a = -12;
static_assert(!std::is_const_v<decltype(__builtin_elementwise_abs(a))>);

(This can come up with overload resolution or in template specializations.)

[fhahn]

Apply UsualUnaryConversions to input argument.

[fhahn]

clang/lib/Sema/SemaChecking.cpp

16664

Ah right. I originally was planning on just using getUnqualifiedType, but for consistency it seems better to just apply the usual unary conversions. I also added the test above and a codegen tests with short.

[fhahn]

add comment to arg of Diag message

[aaron.ballman]

LGTM, thank you!

clang/include/clang/Basic/Builtins.def

 BUILTIN(__builtin_debugtrap, "v", "n")
 BUILTIN(__builtin_unreachable, "v", "nr")
 BUILTIN(__builtin_shufflevector, "v."   , "nct")
 BUILTIN(__builtin_convertvector, "v."   , "nct")
 BUILTIN(__builtin_alloca, "v*z"   , "Fn")
 BUILTIN(__builtin_alloca_with_align, "v*zIz", "Fn")
 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_matrix_transpose, "v.", "nFt")
 BUILTIN(__builtin_matrix_column_major_load, "v.", "nFt")
 BUILTIN(__builtin_matrix_column_major_store, "v.", "nFt")
 
 // "Overloaded" Atomic operator builtins.  These are overloaded to support data

clang/include/clang/Basic/DiagnosticSemaKinds.td

 let CategoryName = "Semantic Issue" in {
 
 def err_elementwise_math_arg_types_mismatch : Error <
   "argument types do not match, %0 != %1">;
 
 def err_elementwise_math_invalid_arg_type: Error <
   "argument type %0 is not supported">;
 
+def err_elementwise_math_invalid_arg_type_2: Error <
+  "argument must have a %0 type, but had a %1 type">;
+
 def err_invalid_conversion_between_matrixes : Error<
   "conversion between matrix types%diff{ $ and $|}0,1 of different size is not allowed">;
 
 def err_invalid_conversion_between_matrix_and_type : Error<
   "conversion between matrix type %0 and incompatible type %1 is not allowed">;
 
 def err_invalid_conversion_between_vectors : Error<
   "invalid conversion between vector type%diff{ $ and $|}0,1 of different "

clang/include/clang/Sema/Sema.h

bool SemaBuiltinARMSpecialReg(unsigned BuiltinID, CallExpr *TheCall,
                                 int ArgNum, unsigned ExpectedFieldNum,
                                 bool AllowName);
   bool SemaBuiltinARMMemoryTaggingCall(unsigned BuiltinID, CallExpr *TheCall);
   bool SemaBuiltinPPCMMACall(CallExpr *TheCall, unsigned BuiltinID,
                              const char *TypeDesc);
 
   bool CheckPPCMMAType(QualType Type, SourceLocation TypeLoc);
 
-  ExprResult SemaBuiltinElementwiseMath(CallExpr *TheCall,
+  ExprResult SemaBuiltinElementwiseMathOneArg(CallExpr *TheCall,
+                                              ExprResult CallResult);
+  ExprResult SemaBuiltinElementwiseMathTwoArgs(CallExpr *TheCall,
                                                ExprResult CallResult);
 
   // Matrix builtin handling.
   ExprResult SemaBuiltinMatrixTranspose(CallExpr *TheCall,
                                         ExprResult CallResult);
   ExprResult SemaBuiltinMatrixColumnMajorLoad(CallExpr *TheCall,
                                               ExprResult CallResult);
   ExprResult SemaBuiltinMatrixColumnMajorStore(CallExpr *TheCall,
                                                ExprResult CallResult);

clang/lib/CodeGen/CGBuiltin.cpp

Value *Sub =
         Builder.CreateSub(llvm::ConstantInt::get(IntTy, ExpMask), AbsV);
     // V = sign bit (Sub) <=> V = (Sub < 0)
     V = Builder.CreateLShr(Sub, llvm::ConstantInt::get(IntTy, bitsize - 1));
     if (bitsize > 32)
       V = Builder.CreateTrunc(V, ConvertType(E->getType()));
     return RValue::get(V);
   }
 
+  case Builtin::BI__builtin_elementwise_abs: {
+    Value *Op0 = EmitScalarExpr(E->getArg(0));
+    Value *Result;
+    if (Op0->getType()->isIntOrIntVectorTy())
+      Result = Builder.CreateBinaryIntrinsic(
+          llvm::Intrinsic::abs, Op0, Builder.getFalse(), nullptr, "elt.abs");
+    else
+      Result = Builder.CreateUnaryIntrinsic(llvm::Intrinsic::fabs, Op0, nullptr,
+                                            "elt.abs");
+    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();

clang/lib/Sema/SemaChecking.cpp

if (!TheCall->getArg(0)->isValueDependent() &&
       Diag(TheCall->getBeginLoc(), diag::warn_frame_address)
           << ((BuiltinID == Builtin::BI__builtin_return_address)
                   ? "__builtin_return_address"
                   : "__builtin_frame_address")
           << TheCall->getSourceRange();
     break;
   }
 
+  case Builtin::BI__builtin_elementwise_abs:
+    return SemaBuiltinElementwiseMathOneArg(TheCall, TheCallResult);
+
   case Builtin::BI__builtin_elementwise_min:
   case Builtin::BI__builtin_elementwise_max:
-    return SemaBuiltinElementwiseMath(TheCall, TheCallResult);
+    return SemaBuiltinElementwiseMathTwoArgs(TheCall, TheCallResult);
 
   case Builtin::BI__builtin_matrix_transpose:
     return SemaBuiltinMatrixTranspose(TheCall, TheCallResult);
   case Builtin::BI__builtin_matrix_column_major_load:
     return SemaBuiltinMatrixColumnMajorLoad(TheCall, TheCallResult);
 
   case Builtin::BI__builtin_matrix_column_major_store:
     return SemaBuiltinMatrixColumnMajorStore(TheCall, TheCallResult);

 void Sema::CheckAddressOfPackedMember(Expr *rhs) {
   using namespace std::placeholders;
 
   RefersToMemberWithReducedAlignment(
       rhs, std::bind(&Sema::AddPotentialMisalignedMembers, std::ref(*this), _1,
                      _2, _3, _4));
 }
 
-ExprResult Sema::SemaBuiltinElementwiseMath(CallExpr *TheCall,
+ExprResult Sema::SemaBuiltinElementwiseMathOneArg(CallExpr *TheCall,
+                                                  ExprResult CallResult) {
+  if (checkArgCount(*this, TheCall, 1))
+    return ExprError();
+
+  Expr *A = TheCall->getArg(0);
+  QualType TyA = A->getType();
+
+  if (!TyA->getAs<VectorType>() && !ConstantMatrixType::isValidElementType(TyA))
+    return Diag(A->getBeginLoc(), diag::err_elementwise_math_invalid_arg_type)
+           << TyA;
+
+  QualType EltTy = TyA;
+  if (auto *VecTy = EltTy->getAs<VectorType>())
+    EltTy = VecTy->getElementType();
+  if (EltTy->isUnsignedIntegerType())
+    return Diag(A->getBeginLoc(), diag::err_elementwise_math_invalid_arg_type_2)
+           << "signed integer or floating point"
+           << "unsigned integer";
+
+  TheCall->setType(TyA);
+  return CallResult;
+}
+
+ExprResult Sema::SemaBuiltinElementwiseMathTwoArgs(CallExpr *TheCall,
                                                    ExprResult CallResult) {
   if (checkArgCount(*this, TheCall, 2))
     return ExprError();
 
   Expr *A = TheCall->getArg(0);
   Expr *B = TheCall->getArg(1);
   QualType TyA = A->getType();
   QualType TyB = B->getType();
 

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

 // RUN: %clang_cc1 -disable-noundef-analysis -triple x86_64-apple-darwin %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s
 
 typedef float float4 __attribute__((ext_vector_type(4)));
 typedef short int si8 __attribute__((ext_vector_type(8)));
 typedef unsigned int u4 __attribute__((ext_vector_type(4)));
 
+void builtin_abs(float f1, float f2, double d1, double d2, float4 vf1, float4 vf2, si8 vi1, si8 vi2, long long int i1, long long int i2) {
+  // CHECK-LABEL: builtin_abs(
+  // CHECK:      [[F1:%.+]] = load float, float* %f1.addr, align 4
+  // CHECK-NEXT:  call float @llvm.fabs.f32(float [[F1]])
+  f2 = __builtin_elementwise_abs(f1);
+
+  // CHECK:      [[D1:%.+]] = load double, double* %d1.addr, align 8
+  // CHECK-NEXT: call double @llvm.fabs.f64(double [[D1]])
+  d2 = __builtin_elementwise_abs(d1);
+
+  // CHECK:      [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call <4 x float> @llvm.fabs.v4f32(<4 x float> [[VF1]])
+  vf2 = __builtin_elementwise_abs(vf1);
+
+  // CHECK:      [[I1:%.+]] = load i64, i64* %i1.addr, align 8
+  // CHECK-NEXT: call i64 @llvm.abs.i64(i64 [[I1]], i1 false)
+  i2 = __builtin_elementwise_abs(i1);
+
+  // CHECK:      [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call <8 x i16> @llvm.abs.v8i16(<8 x i16> [[VI1]], i1 false)
+  vi2 = __builtin_elementwise_abs(vi1);
+}
+
 void builtin_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) {
   // CHECK-LABEL: define void @builtin_max(
+
   // CHECK:      [[F1:%.+]] = load float, float* %f1.addr, align 4
   // CHECK-NEXT: [[F2:%.+]] = load float, float* %f2.addr, align 4
   // CHECK-NEXT:  call float @llvm.maxnum.f32(float %0, float %1)
   f1 = __builtin_elementwise_max(f1, f2);
 
   // CHECK:      [[D1:%.+]] = load double, double* %d1.addr, align 8
   // CHECK-NEXT: [[D2:%.+]] = load double, double* %d2.addr, align 8
   // CHECK-NEXT: call double @llvm.maxnum.f64(double [[D1]], double [[D2]])

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

 // RUN: %clang_cc1 %s -pedantic -verify -triple=x86_64-apple-darwin9
 
 typedef float float4 __attribute__((ext_vector_type(4)));
 typedef int int3 __attribute__((ext_vector_type(3)));
+typedef unsigned unsigned4 __attribute__((ext_vector_type(4)));
 
 struct Foo {
   char *p;
 };
 
+void builtin_elementwise_abs(int i, double d, float4 v, int3 iv, unsigned u, unsigned4 uv) {
+  struct Foo s = __builtin_elementwise_abs(i);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_elementwise_abs();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_elementwise_abs(i, i);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_elementwise_abs(v);
+  // expected-error@-1 {{assigning to 'int' from incompatible type 'float4' (vector of 4 'float' values)}}
+
+  u = __builtin_elementwise_abs(u);
+  // expected-error@-1 {{argument must have a signed integer or floating point type, but had a unsigned integer type}}
+
+  uv = __builtin_elementwise_abs(uv);
+  // expected-error@-1 {{argument must have a signed integer or floating point type, but had a unsigned integer type}}
+}
+
 void builtin_elementwise_max(int i, double d, float4 v, int3 iv) {
   i = __builtin_elementwise_max(i, d);
   // expected-error@-1 {{argument types do not match, 'int' != 'double'}}
 
   struct Foo s = __builtin_elementwise_max(i, i);
   // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
 
   i = __builtin_elementwise_max(i);