[HLSL] Implement support for HLSL intrinsic - saturate (#104619)

bharadwajy · web-flow · commit 6a38e19c92ed · 2024-08-20T12:46:33.000-04:00
Implement support for HLSL intrinsic saturate.
Implement DXIL codegen for the intrinsic saturate by lowering it to DXIL
Op dx.saturate.
Implement SPIRV codegen by transforming saturate(x) to clamp(x, 0.0f,
1.0f).

Add tests for DXIL and SPIRV CodeGen.
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
@@ -4745,6 +4745,12 @@ def HLSLRSqrt : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLSaturate : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_elementwise_saturate"];
+  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
@@ -18667,6 +18667,15 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
         /*ReturnType=*/Op0->getType(), CGM.getHLSLRuntime().getRsqrtIntrinsic(),
         ArrayRef<Value *>{Op0}, nullptr, "hlsl.rsqrt");
   }
+  case Builtin::BI__builtin_hlsl_elementwise_saturate: {
+    Value *Op0 = EmitScalarExpr(E->getArg(0));
+    assert(E->getArg(0)->getType()->hasFloatingRepresentation() &&
+           "saturate operand must have a float representation");
+    return Builder.CreateIntrinsic(
+        /*ReturnType=*/Op0->getType(),
+        CGM.getHLSLRuntime().getSaturateIntrinsic(), ArrayRef<Value *>{Op0},
+        nullptr, "hlsl.saturate");
+  }
   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
@@ -79,6 +79,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(Lerp, lerp)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Normalize, normalize)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Rsqrt, rsqrt)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Saturate, saturate)
   GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
 
   //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -916,7 +916,7 @@ float4 lerp(float4, float4, float4);
 /// \brief Returns the length of the specified floating-point vector.
 /// \param x [in] The vector of floats, or a scalar float.
 ///
-/// Length is based on the following formula: sqrt(x[0]^2 + x[1]^2 + �).
+/// Length is based on the following formula: sqrt(x[0]^2 + x[1]^2 + ...).
 
 _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_length)
@@ -1564,6 +1564,45 @@ float3 round(float3);
 _HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 float4 round(float4);
 
+//===----------------------------------------------------------------------===//
+// saturate builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T saturate(T Val)
+/// \brief Returns input value, \a Val, clamped within the range of 0.0f
+/// to 1.0f. \param Val The input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+half saturate(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+half2 saturate(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+half3 saturate(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+half4 saturate(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+float saturate(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+float2 saturate(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+float3 saturate(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+float4 saturate(float4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+double saturate(double);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+double2 saturate(double2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+double3 saturate(double3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_saturate)
+double4 saturate(double4);
+
 //===----------------------------------------------------------------------===//
 // sin builtins
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
@@ -356,7 +356,7 @@ static bool isLegalTypeForHLSLSV_DispatchThreadID(QualType T) {
   return true;
 }
 
-void SemaHLSL::handleSV_DispatchThreadIDAttr(Decl *D, const ParsedAttr &AL) {  
+void SemaHLSL::handleSV_DispatchThreadIDAttr(Decl *D, const ParsedAttr &AL) {
   auto *VD = cast<ValueDecl>(D);
   if (!isLegalTypeForHLSLSV_DispatchThreadID(VD->getType())) {
     Diag(AL.getLoc(), diag::err_hlsl_attr_invalid_type)
@@ -1045,6 +1045,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     break;
   }
+  case Builtin::BI__builtin_hlsl_elementwise_saturate:
   case Builtin::BI__builtin_hlsl_elementwise_rcp: {
     if (CheckAllArgsHaveFloatRepresentation(&SemaRef, TheCall))
       return true;
diff --git a/clang/test/CodeGenHLSL/builtins/saturate.hlsl b/clang/test/CodeGenHLSL/builtins/saturate.hlsl
@@ -0,0 +1,95 @@
+// RUN: %clang_cc1 -std=hlsl2021 -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: %clang_cc1 -std=hlsl2021 -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: %clang_cc1 -std=hlsl2021 -finclude-default-header -x hlsl -triple \
+// RUN:   spirv-unknown-vulkan-library %s -fnative-half-type \
+// RUN:   -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
+// RUN:   --check-prefixes=SPIRV,SPIRV_HALF
+// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -x hlsl -triple \
+// RUN:   spirv-unknown-vulkan-library %s \
+// RUN:   -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
+// RUN:   --check-prefixes=SPIRV,SPIRV_NO_HALF
+
+// NATIVE_HALF: define noundef half @
+// NATIVE_HALF: call half @llvm.dx.saturate.f16(
+// NO_HALF: define noundef float @"?test_saturate_half
+// NO_HALF: call float @llvm.dx.saturate.f32(
+// SPIRV_HALF: define spir_func noundef half @_Z18test_saturate_halfDh(half
+// SPIRV_HALF: call half @llvm.spv.saturate.f16(half
+// SPIRV_NO_HALF: define spir_func noundef float @_Z18test_saturate_halfDh(float
+// SPIRV_NO_HALF: call float @llvm.spv.saturate.f32(float
+half test_saturate_half(half p0) { return saturate(p0); }
+// NATIVE_HALF: define noundef <2 x half> @
+// NATIVE_HALF: call <2 x half> @llvm.dx.saturate.v2f16
+// NO_HALF: define noundef <2 x float> @"?test_saturate_half2
+// NO_HALF: call <2 x float> @llvm.dx.saturate.v2f32(
+// SPIRV_HALF: define spir_func noundef <2 x half> @_Z19test_saturate_half2Dv2_Dh(
+// SPIRV_HALF: call <2 x half> @llvm.spv.saturate.v2f16(<2 x half>
+// SPIRV_NO_HALF: define spir_func noundef <2 x float> @_Z19test_saturate_half2Dv2_Dh(<2 x float>
+// SPIRV_NO_HALF: call <2 x float> @llvm.spv.saturate.v2f32(<2 x float>
+half2 test_saturate_half2(half2 p0) { return saturate(p0); }
+// NATIVE_HALF: define noundef <3 x half> @
+// NATIVE_HALF: call <3 x half> @llvm.dx.saturate.v3f16
+// NO_HALF: define noundef <3 x float> @"?test_saturate_half3
+// NO_HALF: call <3 x float> @llvm.dx.saturate.v3f32(
+// SPIRV_HALF: define spir_func noundef <3 x half> @_Z19test_saturate_half3Dv3_Dh(
+// SPIRV_HALF: call <3 x half> @llvm.spv.saturate.v3f16(<3 x half>
+// SPIRV_NO_HALF: define spir_func noundef <3 x float> @_Z19test_saturate_half3Dv3_Dh(<3 x float>
+// SPIRV_NO_HALF: call <3 x float> @llvm.spv.saturate.v3f32(<3 x float>
+half3 test_saturate_half3(half3 p0) { return saturate(p0); }
+// NATIVE_HALF: define noundef <4 x half> @
+// NATIVE_HALF: call <4 x half> @llvm.dx.saturate.v4f16
+// NO_HALF: define noundef <4 x float> @"?test_saturate_half4
+// NO_HALF: call <4 x float> @llvm.dx.saturate.v4f32(
+// SPIRV_HALF: define spir_func noundef <4 x half> @_Z19test_saturate_half4Dv4_Dh(
+// SPIRV_HALF: call <4 x half> @llvm.spv.saturate.v4f16(<4 x half>
+// SPIRV_NO_HALF: define spir_func noundef <4 x float> @_Z19test_saturate_half4Dv4_Dh(<4 x float>
+// SPIRV_NO_HALF: call <4 x float> @llvm.spv.saturate.v4f32(<4 x float>
+half4 test_saturate_half4(half4 p0) { return saturate(p0); }
+
+// CHECK: define noundef float @"?test_saturate_float
+// CHECK: call float @llvm.dx.saturate.f32(
+// SPIRV: define spir_func noundef float @_Z19test_saturate_floatf(float
+// SPIRV: call float @llvm.spv.saturate.f32(float
+float test_saturate_float(float p0) { return saturate(p0); }
+// CHECK: define noundef <2 x float> @"?test_saturate_float2
+// CHECK: call <2 x float> @llvm.dx.saturate.v2f32
+// SPIRV: define spir_func noundef <2 x float> @_Z20test_saturate_float2Dv2_f(<2 x float>
+// SPIRV: call <2 x float> @llvm.spv.saturate.v2f32(<2 x float>
+float2 test_saturate_float2(float2 p0) { return saturate(p0); }
+// CHECK: define noundef <3 x float> @"?test_saturate_float3
+// CHECK: call <3 x float> @llvm.dx.saturate.v3f32
+// SPIRV: define spir_func noundef <3 x float> @_Z20test_saturate_float3Dv3_f(<3 x float>
+// SPIRV: call <3 x float> @llvm.spv.saturate.v3f32(<3 x float>
+float3 test_saturate_float3(float3 p0) { return saturate(p0); }
+// CHECK: define noundef <4 x float> @"?test_saturate_float4
+// CHECK: call <4 x float> @llvm.dx.saturate.v4f32
+// SPIRV: define spir_func noundef <4 x float> @_Z20test_saturate_float4Dv4_f(<4 x float>
+// SPIRV: call <4 x float> @llvm.spv.saturate.v4f32(<4 x float>
+float4 test_saturate_float4(float4 p0) { return saturate(p0); }
+
+// CHECK: define noundef double @
+// CHECK: call double @llvm.dx.saturate.f64(
+// SPIRV: define spir_func noundef double @_Z20test_saturate_doubled(double
+// SPIRV: call double @llvm.spv.saturate.f64(double
+double test_saturate_double(double p0) { return saturate(p0); }
+// CHECK: define noundef <2 x double> @
+// CHECK: call <2 x double> @llvm.dx.saturate.v2f64
+// SPIRV: define spir_func noundef <2 x double> @_Z21test_saturate_double2Dv2_d(<2 x double>
+// SPIRV: call <2 x double> @llvm.spv.saturate.v2f64(<2 x double>
+double2 test_saturate_double2(double2 p0) { return saturate(p0); }
+// CHECK: define noundef <3 x double> @
+// CHECK: call <3 x double> @llvm.dx.saturate.v3f64
+// SPIRV: define spir_func noundef <3 x double> @_Z21test_saturate_double3Dv3_d(<3 x double>
+// SPIRV: call <3 x double> @llvm.spv.saturate.v3f64(<3 x double>
+double3 test_saturate_double3(double3 p0) { return saturate(p0); }
+// CHECK: define noundef <4 x double> @
+// CHECK: call <4 x double> @llvm.dx.saturate.v4f64
+// SPIRV: define spir_func noundef <4 x double> @_Z21test_saturate_double4Dv4_d(<4 x double>
+// SPIRV: call <4 x double> @llvm.spv.saturate.v4f64(<4 x double>
+double4 test_saturate_double4(double4 p0) { return saturate(p0); }
diff --git a/clang/test/SemaHLSL/BuiltIns/saturate-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/saturate-errors.hlsl
@@ -0,0 +1,31 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify -verify-ignore-unexpected -Werror
+
+float2 test_no_arg() {
+  return saturate();
+  // expected-error@-1 {{no matching function for call to 'saturate'}}
+}
+
+float2 test_too_many_arg(float2 p0) {
+  return saturate(p0, p0, p0, p0);
+  // expected-error@-1 {{no matching function for call to 'saturate'}}
+}
+
+float2 test_saturate_vector_size_mismatch(float3 p0) {
+  return saturate(p0);
+  // expected-error@-1 {{implicit conversion truncates vector: 'float3' (aka 'vector<float, 3>') to 'vector<float, 2>'}}
+}
+
+float2 test_saturate_float2_int_splat(int p0) {
+  return saturate(p0);
+  // expected-error@-1 {{call to 'saturate' is ambiguous}}
+}
+
+float2 test_saturate_int_vect_to_float_vec_promotion(int2 p0) {
+  return saturate(p0);
+  // expected-error@-1 {{call to 'saturate' is ambiguous}}
+}
+
+float test_saturate_bool_type_promotion(bool p0) {
+  return saturate(p0);
+  // expected-error@-1 {{call to 'saturate' is ambiguous}}
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -34,6 +34,7 @@ def int_dx_all : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
 def int_dx_any : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
 def int_dx_clamp : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
 def int_dx_uclamp : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>; 
+def int_dx_saturate : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
 
 def int_dx_dot2 : 
     Intrinsic<[LLVMVectorElementType<0>], 
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -61,9 +61,10 @@ let TargetPrefix = "spv" in {
   def int_spv_all : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
   def int_spv_any : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
   def int_spv_frac : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty]>;
-  def int_spv_lerp : Intrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>,LLVMMatchType<0>], 
+  def int_spv_lerp : Intrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>,LLVMMatchType<0>],
     [IntrNoMem, IntrWillReturn] >;
   def int_spv_length : DefaultAttrsIntrinsic<[LLVMVectorElementType<0>], [llvm_anyfloat_ty]>;
   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>]>;
 }
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
@@ -330,6 +330,16 @@ def Abs :  DXILOp<6, unary> {
   let attributes = [Attributes<DXIL1_0, [ReadNone]>];
 }
 
+def Saturate :  DXILOp<7, unary> {
+  let Doc = "Clamps a single or double precision floating point value to [0.0f...1.0f].";
+  let LLVMIntrinsic = int_dx_saturate;
+  let arguments = [overloadTy];
+  let result = overloadTy;
+  let overloads = [Overloads<DXIL1_0, [halfTy, floatTy, doubleTy]>];
+  let stages = [Stages<DXIL1_0, [all_stages]>];
+  let attributes = [Attributes<DXIL1_0, [ReadNone]>];
+}
+
 def IsInf :  DXILOp<9, isSpecialFloat> {
   let Doc = "Determines if the specified value is infinite.";
   let LLVMIntrinsic = int_dx_isinf;
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
diff --git a/llvm/test/CodeGen/DirectX/saturate.ll b/llvm/test/CodeGen/DirectX/saturate.ll
diff --git a/llvm/test/CodeGen/DirectX/saturate_errors.ll b/llvm/test/CodeGen/DirectX/saturate_errors.ll
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/saturate.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/saturate.ll
