Skip to content

Commit

Permalink
[clang][HLSL] Add sign intrinsic part 4 (llvm#108396)
Browse files Browse the repository at this point in the history 
- Add handling for unsigned integers to hlsl_elementwise_sign
- Use `select` instead of adding dx and spirv intrinsics for unsigned
integers as [discussed previously
](llvm#101988 (comment))

fixes llvm#70078

### Related PRs
- llvm#101987
- llvm#101988
- llvm#101989

cc @farzonl @pow2clk @bob80905 @bogner @llvm-beanz
  • Loading branch information
@tgymnich @DanielCChen
tgymnich authored and DanielCChen committed Oct 16, 2024
1 parent 1ba5f11 commit 6121292
Show file tree
Hide file tree
Showing 4 changed files with 108 additions and 7 deletions.
15 changes: 11 additions & 4 deletions clang/lib/CodeGen/CGBuiltin.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18876,18 +18876,25 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
return EmitRuntimeCall(Intrinsic::getDeclaration(&CGM.getModule(), ID));
}
case Builtin::BI__builtin_hlsl_elementwise_sign: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
auto *Arg0 = E->getArg(0);
Value *Op0 = EmitScalarExpr(Arg0);
llvm::Type *Xty = Op0->getType();
llvm::Type *retType = llvm::Type::getInt32Ty(this->getLLVMContext());
if (Xty->isVectorTy()) {
auto *XVecTy = E->getArg(0)->getType()->getAs<VectorType>();
auto *XVecTy = Arg0->getType()->getAs<VectorType>();
retType = llvm::VectorType::get(
retType, ElementCount::getFixed(XVecTy->getNumElements()));
}
assert((E->getArg(0)->getType()->hasFloatingRepresentation() ||
E->getArg(0)->getType()->hasSignedIntegerRepresentation()) &&
assert((Arg0->getType()->hasFloatingRepresentation() ||
Arg0->getType()->hasIntegerRepresentation()) &&
"sign operand must have a float or int representation");

if (Arg0->getType()->hasUnsignedIntegerRepresentation()) {
Value *Cmp = Builder.CreateICmpEQ(Op0, ConstantInt::get(Xty, 0));
return Builder.CreateSelect(Cmp, ConstantInt::get(retType, 0),
ConstantInt::get(retType, 1), "hlsl.sign");
}

return Builder.CreateIntrinsic(
retType, CGM.getHLSLRuntime().getSignIntrinsic(),
ArrayRef<Value *>{Op0}, nullptr, "hlsl.sign");
Expand Down
31 changes: 31 additions & 0 deletions clang/lib/Headers/hlsl/hlsl_intrinsics.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2088,6 +2088,19 @@ int3 sign(int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(int16_t4);

_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int sign(uint16_t);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int2 sign(uint16_t2);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(uint16_t4);
#endif

_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
Expand All @@ -2112,6 +2125,15 @@ int3 sign(int3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(int4);

_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int sign(uint);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int2 sign(uint2);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(uint3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(uint4);

_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int sign(float);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
Expand All @@ -2130,6 +2152,15 @@ int3 sign(int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(int64_t4);

_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int sign(uint64_t);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int2 sign(uint64_t2);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int3 sign(uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int4 sign(uint64_t4);

_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
int sign(double);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_sign)
Expand Down
6 changes: 3 additions & 3 deletions clang/lib/Sema/SemaHLSL.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1708,9 +1708,9 @@ static bool CheckNoDoubleVectors(Sema *S, CallExpr *TheCall) {
return CheckArgsTypesAreCorrect(S, TheCall, S->Context.FloatTy,
checkDoubleVector);
}
static bool CheckFloatingOrSignedIntRepresentation(Sema *S, CallExpr *TheCall) {
static bool CheckFloatingOrIntRepresentation(Sema *S, CallExpr *TheCall) {
auto checkAllSignedTypes = [](clang::QualType PassedType) -> bool {
return !PassedType->hasSignedIntegerRepresentation() &&
return !PassedType->hasIntegerRepresentation() &&
!PassedType->hasFloatingRepresentation();
};
return CheckArgsTypesAreCorrect(S, TheCall, S->Context.IntTy,
Expand Down Expand Up @@ -1966,7 +1966,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
break;
}
case Builtin::BI__builtin_hlsl_elementwise_sign: {
if (CheckFloatingOrSignedIntRepresentation(&SemaRef, TheCall))
if (CheckFloatingOrIntRepresentation(&SemaRef, TheCall))
return true;
if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
return true;
Expand Down
63 changes: 63 additions & 0 deletions clang/test/CodeGenHLSL/builtins/sign.hlsl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -112,6 +112,27 @@ int3 test_sign_int16_t3(int16_t3 p0) { return sign(p0); }
// NATIVE_HALF: %hlsl.sign = call <4 x i32> @llvm.[[TARGET]].sign.v4i16(
// NATIVE_HALF: ret <4 x i32> %hlsl.sign
int4 test_sign_int16_t4(int16_t4 p0) { return sign(p0); }


// NATIVE_HALF: define [[FNATTRS]] i32 @
// NATIVE_HALF: [[CMP:%.*]] = icmp eq i16 [[ARG:%.*]], 0
// NATIVE_HALF: %hlsl.sign = select i1 [[CMP]], i32 0, i32 1
int test_sign_uint16_t(uint16_t p0) { return sign(p0); }

// NATIVE_HALF: define [[FNATTRS]] <2 x i32> @
// NATIVE_HALF: [[CMP:%.*]] = icmp eq <2 x i16> [[ARG:%.*]], zeroinitializer
// NATIVE_HALF: %hlsl.sign = select <2 x i1> [[CMP]], <2 x i32> zeroinitializer, <2 x i32> <i32 1, i32 1>
int2 test_sign_uint16_t2(uint16_t2 p0) { return sign(p0); }

// NATIVE_HALF: define [[FNATTRS]] <3 x i32> @
// NATIVE_HALF: [[CMP:%.*]] = icmp eq <3 x i16> [[ARG:%.*]], zeroinitializer
// NATIVE_HALF: %hlsl.sign = select <3 x i1> [[CMP]], <3 x i32> zeroinitializer, <3 x i32> <i32 1, i32 1, i32 1>
int3 test_sign_uint16_t3(uint16_t3 p0) { return sign(p0); }

// NATIVE_HALF: define [[FNATTRS]] <4 x i32> @
// NATIVE_HALF: [[CMP:%.*]] = icmp eq <4 x i16> [[ARG:%.*]], zeroinitializer
// NATIVE_HALF: %hlsl.sign = select <4 x i1> [[CMP]], <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
int4 test_sign_uint16_t4(uint16_t4 p0) { return sign(p0); }
#endif // __HLSL_ENABLE_16_BIT


Expand All @@ -136,6 +157,27 @@ int3 test_sign_int3(int3 p0) { return sign(p0); }
int4 test_sign_int4(int4 p0) { return sign(p0); }


// CHECK: define [[FNATTRS]] i32 @
// CHECK: [[CMP:%.*]] = icmp eq i32 [[ARG:%.*]], 0
// CHECK: %hlsl.sign = select i1 [[CMP]], i32 0, i32 1
int test_sign_uint(uint p0) { return sign(p0); }

// CHECK: define [[FNATTRS]] <2 x i32> @
// CHECK: [[CMP:%.*]] = icmp eq <2 x i32> [[ARG:%.*]], zeroinitializer
// CHECK: %hlsl.sign = select <2 x i1> [[CMP]], <2 x i32> zeroinitializer, <2 x i32> <i32 1, i32 1>
int2 test_sign_uint2(uint2 p0) { return sign(p0); }

// CHECK: define [[FNATTRS]] <3 x i32> @
// CHECK: [[CMP:%.*]] = icmp eq <3 x i32> [[ARG:%.*]], zeroinitializer
// CHECK: %hlsl.sign = select <3 x i1> [[CMP]], <3 x i32> zeroinitializer, <3 x i32> <i32 1, i32 1, i32 1>
int3 test_sign_uint3(uint3 p0) { return sign(p0); }

// CHECK: define [[FNATTRS]] <4 x i32> @
// CHECK: [[CMP:%.*]] = icmp eq <4 x i32> [[ARG:%.*]], zeroinitializer
// CHECK: %hlsl.sign = select <4 x i1> [[CMP]], <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
int4 test_sign_uint4(uint4 p0) { return sign(p0); }


// CHECK: define [[FNATTRS]] i32 @
// CHECK: %hlsl.sign = call i32 @llvm.[[TARGET]].sign.i64(
// CHECK: ret i32 %hlsl.sign
Expand All @@ -155,3 +197,24 @@ int3 test_sign_int64_t3(int64_t3 p0) { return sign(p0); }
// CHECK: %hlsl.sign = call <4 x i32> @llvm.[[TARGET]].sign.v4i64(
// CHECK: ret <4 x i32> %hlsl.sign
int4 test_sign_int64_t4(int64_t4 p0) { return sign(p0); }


// CHECK: define [[FNATTRS]] i32 @
// CHECK: [[CMP:%.*]] = icmp eq i64 [[ARG:%.*]], 0
// CHECK: %hlsl.sign = select i1 [[CMP]], i32 0, i32 1
int test_sign_int64_t(uint64_t p0) { return sign(p0); }

// CHECK: define [[FNATTRS]] <2 x i32> @
// CHECK: [[CMP:%.*]] = icmp eq <2 x i64> [[ARG:%.*]], zeroinitializer
// CHECK: %hlsl.sign = select <2 x i1> [[CMP]], <2 x i32> zeroinitializer, <2 x i32> <i32 1, i32 1>
int2 test_sign_int64_t2(uint64_t2 p0) { return sign(p0); }

// CHECK: define [[FNATTRS]] <3 x i32> @
// CHECK: [[CMP:%.*]] = icmp eq <3 x i64> [[ARG:%.*]], zeroinitializer
// CHECK: %hlsl.sign = select <3 x i1> [[CMP]], <3 x i32> zeroinitializer, <3 x i32> <i32 1, i32 1, i32 1>
int3 test_sign_int64_t3(uint64_t3 p0) { return sign(p0); }

// CHECK: define [[FNATTRS]] <4 x i32> @
// CHECK: [[CMP:%.*]] = icmp eq <4 x i64> [[ARG:%.*]], zeroinitializer
// CHECK: %hlsl.sign = select <4 x i1> [[CMP]], <4 x i32> zeroinitializer, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
int4 test_sign_int64_t4(uint64_t4 p0) { return sign(p0); }

0 comments on commit 6121292

Please sign in to comment.