[clang] 2d47a0b - Add step builtins and step HLSL function to DirectX and SPIR-V backend (#106471)
via cfe-commits
cfe-commits at lists.llvm.org
Thu Sep 12 10:26:43 PDT 2024
Author: Joshua Batista
Date: 2024-09-12T10:26:38-07:00
New Revision: 2d47a0baba6f18b5cdf14c007e01b79e37d15c72
URL: https://github.com/llvm/llvm-project/commit/2d47a0baba6f18b5cdf14c007e01b79e37d15c72
DIFF: https://github.com/llvm/llvm-project/commit/2d47a0baba6f18b5cdf14c007e01b79e37d15c72.diff
LOG: Add step builtins and step HLSL function to DirectX and SPIR-V backend (#106471)
This PR adds the step intrinsic and an HLSL function that uses it.
The SPIRV backend is also implemented.
Used https://github.com/llvm/llvm-project/pull/102683 as a reference.
Fixes https://github.com/llvm/llvm-project/issues/99157
Added:
clang/test/CodeGenHLSL/builtins/step.hlsl
clang/test/SemaHLSL/BuiltIns/step-errors.hlsl
llvm/test/CodeGen/DirectX/step.ll
llvm/test/CodeGen/SPIRV/hlsl-intrinsics/step.ll
Modified:
clang/include/clang/Basic/Builtins.td
clang/lib/CodeGen/CGBuiltin.cpp
clang/lib/CodeGen/CGHLSLRuntime.h
clang/lib/Headers/hlsl/hlsl_intrinsics.h
clang/lib/Sema/SemaHLSL.cpp
llvm/include/llvm/IR/IntrinsicsDirectX.td
llvm/include/llvm/IR/IntrinsicsSPIRV.td
llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
Removed:
################################################################################
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 3dc04f68b3172a..6cf03d27055cd9 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4763,6 +4763,7 @@ def HLSLSaturate : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+
def HLSLSelect : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_select"];
let Attributes = [NoThrow, Const];
@@ -4775,6 +4776,12 @@ def HLSLSign : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLStep: LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_step"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 9950c06a0b9a6b..27abeba92999b3 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18861,6 +18861,16 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
return SelectVal;
}
+ case Builtin::BI__builtin_hlsl_step: {
+ Value *Op0 = EmitScalarExpr(E->getArg(0));
+ Value *Op1 = EmitScalarExpr(E->getArg(1));
+ assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
+ E->getArg(1)->getType()->hasFloatingRepresentation() &&
+ "step operands must have a float representation");
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/Op0->getType(), CGM.getHLSLRuntime().getStepIntrinsic(),
+ ArrayRef<Value *>{Op0, Op1}, nullptr, "hlsl.step");
+ }
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
return EmitRuntimeCall(CGM.CreateRuntimeFunction(
llvm::FunctionType::get(IntTy, {}, false), "__hlsl_wave_get_lane_index",
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 6e226808fcbaf5..a8aabca7348ffb 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -81,6 +81,7 @@ class CGHLSLRuntime {
GENERATE_HLSL_INTRINSIC_FUNCTION(Rsqrt, rsqrt)
GENERATE_HLSL_INTRINSIC_FUNCTION(Saturate, saturate)
GENERATE_HLSL_INTRINSIC_FUNCTION(Sign, sign)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(Step, step)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 7a1edd93984de7..d08dcd350d558b 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -1717,6 +1717,39 @@ float3 sqrt(float3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_sqrt)
float4 sqrt(float4);
+//===----------------------------------------------------------------------===//
+// step builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T step(T x, T y)
+/// \brief Returns 1 if the x parameter is greater than or equal to the y
+/// parameter; otherwise, 0. vector. \param x [in] The first floating-point
+/// value to compare. \param y [in] The first floating-point value to compare.
+///
+/// Step is based on the following formula: (x >= y) ? 1 : 0
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+half step(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+half2 step(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+half3 step(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+half4 step(half4, half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+float step(float, float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+float2 step(float2, float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+float3 step(float3, float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_step)
+float4 step(float4, float4);
+
//===----------------------------------------------------------------------===//
// tan builtins
//===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 4e44813fe515ce..527718c8e7e324 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1747,6 +1747,18 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
SetElementTypeAsReturnType(&SemaRef, TheCall, getASTContext().IntTy);
break;
}
+ case Builtin::BI__builtin_hlsl_step: {
+ if (SemaRef.checkArgCount(TheCall, 2))
+ return true;
+ if (CheckFloatOrHalfRepresentations(&SemaRef, TheCall))
+ return true;
+
+ ExprResult A = TheCall->getArg(0);
+ QualType ArgTyA = A.get()->getType();
+ // return type is the same as the input type
+ TheCall->setType(ArgTyA);
+ break;
+ }
// Note these are llvm builtins that we want to catch invalid intrinsic
// generation. Normal handling of these builitns will occur elsewhere.
case Builtin::BI__builtin_elementwise_bitreverse: {
diff --git a/clang/test/CodeGenHLSL/builtins/step.hlsl b/clang/test/CodeGenHLSL/builtins/step.hlsl
new file mode 100644
index 00000000000000..442f4930ca579c
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/step.hlsl
@@ -0,0 +1,84 @@
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -fnative-half-type \
+// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
+// RUN: --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DFNATTRS=noundef -DTARGET=dx
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -emit-llvm -disable-llvm-passes \
+// RUN: -o - | FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DFNATTRS=noundef -DTARGET=dx
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \
+// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
+// RUN: --check-prefixes=CHECK,NATIVE_HALF \
+// RUN: -DFNATTRS="spir_func noundef" -DTARGET=spv
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \
+// RUN: -o - | FileCheck %s --check-prefixes=CHECK,NO_HALF \
+// RUN: -DFNATTRS="spir_func noundef" -DTARGET=spv
+
+// NATIVE_HALF: define [[FNATTRS]] half @
+// NATIVE_HALF: call half @llvm.[[TARGET]].step.f16(half
+// NO_HALF: call float @llvm.[[TARGET]].step.f32(float
+// NATIVE_HALF: ret half
+// NO_HALF: ret float
+half test_step_half(half p0, half p1)
+{
+ return step(p0, p1);
+}
+// NATIVE_HALF: define [[FNATTRS]] <2 x half> @
+// NATIVE_HALF: call <2 x half> @llvm.[[TARGET]].step.v2f16(<2 x half>
+// NO_HALF: call <2 x float> @llvm.[[TARGET]].step.v2f32(<2 x float>
+// NATIVE_HALF: ret <2 x half> %hlsl.step
+// NO_HALF: ret <2 x float> %hlsl.step
+half2 test_step_half2(half2 p0, half2 p1)
+{
+ return step(p0, p1);
+}
+// NATIVE_HALF: define [[FNATTRS]] <3 x half> @
+// NATIVE_HALF: call <3 x half> @llvm.[[TARGET]].step.v3f16(<3 x half>
+// NO_HALF: call <3 x float> @llvm.[[TARGET]].step.v3f32(<3 x float>
+// NATIVE_HALF: ret <3 x half> %hlsl.step
+// NO_HALF: ret <3 x float> %hlsl.step
+half3 test_step_half3(half3 p0, half3 p1)
+{
+ return step(p0, p1);
+}
+// NATIVE_HALF: define [[FNATTRS]] <4 x half> @
+// NATIVE_HALF: call <4 x half> @llvm.[[TARGET]].step.v4f16(<4 x half>
+// NO_HALF: call <4 x float> @llvm.[[TARGET]].step.v4f32(<4 x float>
+// NATIVE_HALF: ret <4 x half> %hlsl.step
+// NO_HALF: ret <4 x float> %hlsl.step
+half4 test_step_half4(half4 p0, half4 p1)
+{
+ return step(p0, p1);
+}
+
+// CHECK: define [[FNATTRS]] float @
+// CHECK: call float @llvm.[[TARGET]].step.f32(float
+// CHECK: ret float
+float test_step_float(float p0, float p1)
+{
+ return step(p0, p1);
+}
+// CHECK: define [[FNATTRS]] <2 x float> @
+// CHECK: %hlsl.step = call <2 x float> @llvm.[[TARGET]].step.v2f32(
+// CHECK: ret <2 x float> %hlsl.step
+float2 test_step_float2(float2 p0, float2 p1)
+{
+ return step(p0, p1);
+}
+// CHECK: define [[FNATTRS]] <3 x float> @
+// CHECK: %hlsl.step = call <3 x float> @llvm.[[TARGET]].step.v3f32(
+// CHECK: ret <3 x float> %hlsl.step
+float3 test_step_float3(float3 p0, float3 p1)
+{
+ return step(p0, p1);
+}
+// CHECK: define [[FNATTRS]] <4 x float> @
+// CHECK: %hlsl.step = call <4 x float> @llvm.[[TARGET]].step.v4f32(
+// CHECK: ret <4 x float> %hlsl.step
+float4 test_step_float4(float4 p0, float4 p1)
+{
+ return step(p0, p1);
+}
diff --git a/clang/test/SemaHLSL/BuiltIns/step-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/step-errors.hlsl
new file mode 100644
index 00000000000000..823585201ca62d
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/step-errors.hlsl
@@ -0,0 +1,31 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -disable-llvm-passes -verify -verify-ignore-unexpected
+
+void test_too_few_arg()
+{
+ return __builtin_hlsl_step();
+ // expected-error at -1 {{too few arguments to function call, expected 2, have 0}}
+}
+
+void test_too_many_arg(float2 p0)
+{
+ return __builtin_hlsl_step(p0, p0, p0);
+ // expected-error at -1 {{too many arguments to function call, expected 2, have 3}}
+}
+
+bool builtin_bool_to_float_type_promotion(bool p1)
+{
+ return __builtin_hlsl_step(p1, p1);
+ // expected-error at -1 {passing 'bool' to parameter of incompatible type 'float'}}
+}
+
+bool builtin_step_int_to_float_promotion(int p1)
+{
+ return __builtin_hlsl_step(p1, p1);
+ // expected-error at -1 {{passing 'int' to parameter of incompatible type 'float'}}
+}
+
+bool2 builtin_step_int2_to_float2_promotion(int2 p1)
+{
+ return __builtin_hlsl_step(p1, p1);
+ // expected-error at -1 {{passing 'int2' (aka 'vector<int, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(float)))) float' (vector of 2 'float' values)}}
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index f1017bdd512496..97c69638fbb0ef 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -87,7 +87,7 @@ def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLV
def int_dx_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty]>;
def int_dx_rcp : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
def int_dx_rsqrt : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
-
def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty]>;
+def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>]>;
}
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 766fc0d99d2dbb..a4c01952927175 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -67,6 +67,7 @@ let TargetPrefix = "spv" in {
def int_spv_normalize : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty]>;
def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty]>;
def int_spv_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
+ def int_spv_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [LLVMMatchType<0>, llvm_anyfloat_ty]>;
def int_spv_fdot :
DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
[llvm_anyfloat_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
diff --git a/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp b/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
index 72fa9891bfd8e6..b46800773498de 100644
--- a/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
+++ b/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
@@ -50,6 +50,7 @@ static bool isIntrinsicExpansion(Function &F) {
case Intrinsic::dx_sdot:
case Intrinsic::dx_udot:
case Intrinsic::dx_sign:
+ case Intrinsic::dx_step:
return true;
}
return false;
@@ -322,6 +323,28 @@ static Value *expandPowIntrinsic(CallInst *Orig) {
return Exp2Call;
}
+static Value *expandStepIntrinsic(CallInst *Orig) {
+
+ Value *X = Orig->getOperand(0);
+ Value *Y = Orig->getOperand(1);
+ Type *Ty = X->getType();
+ IRBuilder<> Builder(Orig);
+
+ Constant *One = ConstantFP::get(Ty->getScalarType(), 1.0);
+ Constant *Zero = ConstantFP::get(Ty->getScalarType(), 0.0);
+ Value *Cond = Builder.CreateFCmpOLT(Y, X);
+
+ if (Ty != Ty->getScalarType()) {
+ auto *XVec = dyn_cast<FixedVectorType>(Ty);
+ One = ConstantVector::getSplat(
+ ElementCount::getFixed(XVec->getNumElements()), One);
+ Zero = ConstantVector::getSplat(
+ ElementCount::getFixed(XVec->getNumElements()), Zero);
+ }
+
+ return Builder.CreateSelect(Cond, Zero, One);
+}
+
static Intrinsic::ID getMaxForClamp(Type *ElemTy,
Intrinsic::ID ClampIntrinsic) {
if (ClampIntrinsic == Intrinsic::dx_uclamp)
@@ -433,8 +456,9 @@ static bool expandIntrinsic(Function &F, CallInst *Orig) {
case Intrinsic::dx_sign:
Result = expandSignIntrinsic(Orig);
break;
+ case Intrinsic::dx_step:
+ Result = expandStepIntrinsic(Orig);
}
-
if (Result) {
Orig->replaceAllUsesWith(Result);
Orig->eraseFromParent();
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 831d7f76ac14c9..b526c9f29f1e6a 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -263,6 +263,9 @@ class SPIRVInstructionSelector : public InstructionSelector {
bool selectSpvThreadId(Register ResVReg, const SPIRVType *ResType,
MachineInstr &I) const;
+ bool selectStep(Register ResVReg, const SPIRVType *ResType,
+ MachineInstr &I) const;
+
bool selectUnmergeValues(MachineInstr &I) const;
Register buildI32Constant(uint32_t Val, MachineInstr &I,
@@ -1710,6 +1713,25 @@ bool SPIRVInstructionSelector::selectSign(Register ResVReg,
return Result;
}
+bool SPIRVInstructionSelector::selectStep(Register ResVReg,
+ const SPIRVType *ResType,
+ MachineInstr &I) const {
+
+ assert(I.getNumOperands() == 4);
+ assert(I.getOperand(2).isReg());
+ assert(I.getOperand(3).isReg());
+ MachineBasicBlock &BB = *I.getParent();
+
+ return BuildMI(BB, I, I.getDebugLoc(), TII.get(SPIRV::OpExtInst))
+ .addDef(ResVReg)
+ .addUse(GR.getSPIRVTypeID(ResType))
+ .addImm(static_cast<uint32_t>(SPIRV::InstructionSet::GLSL_std_450))
+ .addImm(GL::Step)
+ .addUse(I.getOperand(2).getReg())
+ .addUse(I.getOperand(3).getReg())
+ .constrainAllUses(TII, TRI, RBI);
+}
+
bool SPIRVInstructionSelector::selectBitreverse(Register ResVReg,
const SPIRVType *ResType,
MachineInstr &I) const {
@@ -2468,6 +2490,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
.addUse(GR.getSPIRVTypeID(ResType))
.addUse(GR.getOrCreateConstInt(3, I, IntTy, TII));
}
+ case Intrinsic::spv_step:
+ return selectStep(ResVReg, ResType, I);
default: {
std::string DiagMsg;
raw_string_ostream OS(DiagMsg);
diff --git a/llvm/test/CodeGen/DirectX/step.ll b/llvm/test/CodeGen/DirectX/step.ll
new file mode 100644
index 00000000000000..0393c1533cb939
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/step.ll
@@ -0,0 +1,78 @@
+; RUN: opt -S -dxil-intrinsic-expansion < %s | FileCheck %s --check-prefix=CHECK
+; RUN: opt -S -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s --check-prefix=CHECK
+
+; Make sure dxil operation function calls for step are generated for half/float.
+
+declare half @llvm.dx.step.f16(half, half)
+declare <2 x half> @llvm.dx.step.v2f16(<2 x half>, <2 x half>)
+declare <3 x half> @llvm.dx.step.v3f16(<3 x half>, <3 x half>)
+declare <4 x half> @llvm.dx.step.v4f16(<4 x half>, <4 x half>)
+
+declare float @llvm.dx.step.f32(float, float)
+declare <2 x float> @llvm.dx.step.v2f32(<2 x float>, <2 x float>)
+declare <3 x float> @llvm.dx.step.v3f32(<3 x float>, <3 x float>)
+declare <4 x float> @llvm.dx.step.v4f32(<4 x float>, <4 x float>)
+
+define noundef half @test_step_half(half noundef %p0, half noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt half %p1, %p0
+ ; CHECK: %1 = select i1 %0, half 0xH0000, half 0xH3C00
+ %hlsl.step = call half @llvm.dx.step.f16(half %p0, half %p1)
+ ret half %hlsl.step
+}
+
+define noundef <2 x half> @test_step_half2(<2 x half> noundef %p0, <2 x half> noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt <2 x half> %p1, %p0
+ ; CHECK: %1 = select <2 x i1> %0, <2 x half> zeroinitializer, <2 x half> <half 0xH3C00, half 0xH3C00>
+ %hlsl.step = call <2 x half> @llvm.dx.step.v2f16(<2 x half> %p0, <2 x half> %p1)
+ ret <2 x half> %hlsl.step
+}
+
+define noundef <3 x half> @test_step_half3(<3 x half> noundef %p0, <3 x half> noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt <3 x half> %p1, %p0
+ ; CHECK: %1 = select <3 x i1> %0, <3 x half> zeroinitializer, <3 x half> <half 0xH3C00, half 0xH3C00, half 0xH3C00>
+ %hlsl.step = call <3 x half> @llvm.dx.step.v3f16(<3 x half> %p0, <3 x half> %p1)
+ ret <3 x half> %hlsl.step
+}
+
+define noundef <4 x half> @test_step_half4(<4 x half> noundef %p0, <4 x half> noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt <4 x half> %p1, %p0
+ ; CHECK: %1 = select <4 x i1> %0, <4 x half> zeroinitializer, <4 x half> <half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00>
+ %hlsl.step = call <4 x half> @llvm.dx.step.v4f16(<4 x half> %p0, <4 x half> %p1)
+ ret <4 x half> %hlsl.step
+}
+
+define noundef float @test_step_float(float noundef %p0, float noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt float %p1, %p0
+ ; CHECK: %1 = select i1 %0, float 0.000000e+00, float 1.000000e+00
+ %hlsl.step = call float @llvm.dx.step.f32(float %p0, float %p1)
+ ret float %hlsl.step
+}
+
+define noundef <2 x float> @test_step_float2(<2 x float> noundef %p0, <2 x float> noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt <2 x float> %p1, %p0
+ ; CHECK: %1 = select <2 x i1> %0, <2 x float> zeroinitializer, <2 x float> <float 1.000000e+00, float 1.000000e+00>
+ %hlsl.step = call <2 x float> @llvm.dx.step.v2f32(<2 x float> %p0, <2 x float> %p1)
+ ret <2 x float> %hlsl.step
+}
+
+define noundef <3 x float> @test_step_float3(<3 x float> noundef %p0, <3 x float> noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt <3 x float> %p1, %p0
+ ; CHECK: %1 = select <3 x i1> %0, <3 x float> zeroinitializer, <3 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
+ %hlsl.step = call <3 x float> @llvm.dx.step.v3f32(<3 x float> %p0, <3 x float> %p1)
+ ret <3 x float> %hlsl.step
+}
+
+define noundef <4 x float> @test_step_float4(<4 x float> noundef %p0, <4 x float> noundef %p1) {
+entry:
+ ; CHECK: %0 = fcmp olt <4 x float> %p1, %p0
+ ; CHECK: %1 = select <4 x i1> %0, <4 x float> zeroinitializer, <4 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
+ %hlsl.step = call <4 x float> @llvm.dx.step.v4f32(<4 x float> %p0, <4 x float> %p1)
+ ret <4 x float> %hlsl.step
+}
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/step.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/step.ll
new file mode 100644
index 00000000000000..bb50d8c790f8ad
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/step.ll
@@ -0,0 +1,33 @@
+; RUN: llc -O0 -mtriple=spirv-unknown-unknown %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o - -filetype=obj | spirv-val %}
+
+; Make sure SPIRV operation function calls for step are lowered correctly.
+
+; CHECK-DAG: %[[#op_ext_glsl:]] = OpExtInstImport "GLSL.std.450"
+; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32
+; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16
+; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4
+; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4
+
+define noundef <4 x half> @step_half4(<4 x half> noundef %a, <4 x half> noundef %b) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_16]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]]
+ ; CHECK: %[[#]] = OpExtInst %[[#vec4_float_16]] %[[#op_ext_glsl]] Step %[[#arg0]] %[[#arg1]]
+ %hlsl.step = call <4 x half> @llvm.spv.step.v4f16(<4 x half> %a, <4 x half> %b)
+ ret <4 x half> %hlsl.step
+}
+
+define noundef <4 x float> @step_float4(<4 x float> noundef %a, <4 x float> noundef %b) {
+entry:
+ ; CHECK: %[[#]] = OpFunction %[[#vec4_float_32]] None %[[#]]
+ ; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_32]]
+ ; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_32]]
+ ; CHECK: %[[#]] = OpExtInst %[[#vec4_float_32]] %[[#op_ext_glsl]] Step %[[#arg0]] %[[#arg1]]
+ %hlsl.step = call <4 x float> @llvm.spv.step.v4f32(<4 x float> %a, <4 x float> %b)
+ ret <4 x float> %hlsl.step
+}
+
+declare <4 x half> @llvm.spv.step.v4f16(<4 x half>, <4 x half>)
+declare <4 x float> @llvm.spv.step.v4f32(<4 x float>, <4 x float>)
More information about the cfe-commits
mailing list