[HLSL] add extra scalar vector overloads for clamp

clang/lib/Headers/CMakeLists.txt

@@ -87,6 +87,7 @@ set(hlsl_h
 set(hlsl_subdir_files
   hlsl/hlsl_basic_types.h
   hlsl/hlsl_alias_intrinsics.h
+  hlsl/hlsl_compat_overloads.h
   hlsl/hlsl_intrinsic_helpers.h
   hlsl/hlsl_intrinsics.h
   hlsl/hlsl_detail.h

clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h

@@ -9,6 +9,8 @@
 #ifndef _HLSL_HLSL_ALIAS_INTRINSICS_H_
 #define _HLSL_HLSL_ALIAS_INTRINSICS_H_
 
+#include "hlsl_compat_overloads.h"
+
 namespace hlsl {
 
 // Note: Functions in this file are sorted alphabetically, then grouped by base

clang/lib/Headers/hlsl/hlsl_compat_overloads.h

@@ -0,0 +1,60 @@
+//===--- hlsl_compat_overloads.h - Additional HLSL overload definitions for
+// intrinsics --===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _HLSL_COMPAT_OVERLOADS_H_
+#define _HLSl_COMPAT_OVERLOADS_H_
+
+namespace hlsl {
+
+// Note: Functions in this file are sorted alphabetically, then grouped by base
+// element type, and the element types are sorted by size, then singed integer,
+// unsigned integer and floating point. Keeping this ordering consistent will
+// help keep this file manageable as it grows.
+
+//===----------------------------------------------------------------------===//
+// clamp builtins overloads
+//===----------------------------------------------------------------------===//
+
+template <typename T, typename R, typename U, uint N>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value, vector<T, N>>
+clamp(vector<T, N> p0, vector<R, N> p1, U p2) {
+  return __builtin_hlsl_elementwise_clamp(p0, (vector<T, N>)p1,
+                                          (vector<T, N>)p2);
+}
+template <typename T, typename R, typename U, uint N>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value, vector<T, N>>
+clamp(vector<T, N> p0, U p1, vector<R, N> p2) {
+  return __builtin_hlsl_elementwise_clamp(p0, (vector<T, N>)p1,
+                                          (vector<T, N>)p2);
+}
+template <typename T, typename U, typename V, uint N>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value &&
+                                    __detail::is_arithmetic<V>::Value,
+                                vector<T, N>>
+clamp(vector<T, N> p0, U p1, V p2) {
+  return __builtin_hlsl_elementwise_clamp(p0, (vector<T, N>)p1,
+                                          (vector<T, N>)p2);
+}
+template <typename T, typename R, typename S, uint N>
+constexpr vector<T, N> clamp(vector<T, N> p0, vector<R, N> p1,
+                             vector<S, N> p2) {
+  return __builtin_hlsl_elementwise_clamp(p0, (vector<T, N>)p1,
+                                          (vector<T, N>)p2);
+}
+template <typename U, typename V, typename W>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value &&
+                                    __detail::is_arithmetic<V>::Value &&
+                                    __detail::is_arithmetic<W>::Value,
+                                U>
+clamp(U p0, V p1, W p2) {
+  return __builtin_hlsl_elementwise_clamp(p0, (U)p1, (U)p2);
+}
+
+} // namespace hlsl
+#endif // _HLSL_COMPAT_OVERLOADS_H_

clang/lib/Sema/SemaHLSL.cpp

@@ -2041,6 +2041,22 @@ static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
   return false;
 }
 
+static bool CheckAllArgsHaveSameType(Sema *S, CallExpr *TheCall) {
+  QualType ArgTy0 = TheCall->getArg(0)->getType();
+
+  for (unsigned I = 1, N = TheCall->getNumArgs(); I < N; ++I) {
+    if (!S->getASTContext().hasSameUnqualifiedType(
+            ArgTy0, TheCall->getArg(I)->getType())) {
+      S->Diag(TheCall->getBeginLoc(), diag::err_vec_builtin_incompatible_vector)
+          << TheCall->getDirectCallee() << /*useAllTerminology*/ true
+          << SourceRange(TheCall->getArg(0)->getBeginLoc(),
+                         TheCall->getArg(N - 1)->getEndLoc());
+      return true;
+    }
+  }
+  return false;
+}
+
 static bool CheckArgTypeMatches(Sema *S, Expr *Arg, QualType ExpectedType) {
   QualType ArgType = Arg->getType();
   if (!S->getASTContext().hasSameUnqualifiedType(ArgType, ExpectedType)) {
@@ -2392,7 +2408,8 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case Builtin::BI__builtin_hlsl_elementwise_clamp: {
     if (SemaRef.checkArgCount(TheCall, 3))
       return true;
-    if (CheckVectorElementCallArgs(&SemaRef, TheCall))
+    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0) ||
+        CheckAllArgsHaveSameType(&SemaRef, TheCall))
       return true;
     if (SemaRef.BuiltinElementwiseTernaryMath(
             TheCall, /*CheckForFloatArgs*/

clang/test/CodeGenHLSL/builtins/clamp.hlsl

@@ -28,6 +28,9 @@ int16_t3 test_clamp_short3(int16_t3 p0, int16_t3 p1) { return clamp(p0, p1,p1);
 // NATIVE_HALF: define [[FNATTRS]] <4 x i16> @_Z17test_clamp_short4
 // NATIVE_HALF: call <4 x i16> @llvm.[[TARGET]].sclamp.v4i16
 int16_t4 test_clamp_short4(int16_t4 p0, int16_t4 p1) { return clamp(p0, p1,p1); }
+// NATIVE_HALF: define [[FNATTRS]] <4 x i16> {{.*}}test_clamp_short4_mismatch
+// NATIVE_HALF: call <4 x i16> @llvm.[[TARGET]].sclamp.v4i16
+int16_t4 test_clamp_short4_mismatch(int16_t4 p0, int16_t p1) { return clamp(p0, p0,p1); }
 
 // NATIVE_HALF: define [[FNATTRS]] i16 @_Z17test_clamp_ushort
 // NATIVE_HALF: call i16 @llvm.[[TARGET]].uclamp.i16(
@@ -41,6 +44,9 @@ uint16_t3 test_clamp_ushort3(uint16_t3 p0, uint16_t3 p1) { return clamp(p0, p1,p
 // NATIVE_HALF: define [[FNATTRS]] <4 x i16> @_Z18test_clamp_ushort4
 // NATIVE_HALF: call <4 x i16> @llvm.[[TARGET]].uclamp.v4i16
 uint16_t4 test_clamp_ushort4(uint16_t4 p0, uint16_t4 p1) { return clamp(p0, p1,p1); }
+// NATIVE_HALF: define [[FNATTRS]] <4 x i16> {{.*}}test_clamp_ushort4_mismatch
+// NATIVE_HALF: call <4 x i16> @llvm.[[TARGET]].uclamp.v4i16
+uint16_t4 test_clamp_ushort4_mismatch(uint16_t4 p0, uint16_t p1) { return clamp(p0, p0,p1); }
 #endif
 
 // CHECK: define [[FNATTRS]] i32 @_Z14test_clamp_int
@@ -55,6 +61,9 @@ int3 test_clamp_int3(int3 p0, int3 p1) { return clamp(p0, p1,p1); }
 // CHECK: define [[FNATTRS]] <4 x i32> @_Z15test_clamp_int4
 // CHECK: call <4 x i32> @llvm.[[TARGET]].sclamp.v4i32
 int4 test_clamp_int4(int4 p0, int4 p1) { return clamp(p0, p1,p1); }
+// CHECK: define [[FNATTRS]] <4 x i32> {{.*}}test_clamp_int4_mismatch
+// CHECK: call <4 x i32> @llvm.[[TARGET]].sclamp.v4i32
+int4 test_clamp_int4_mismatch(int4 p0, int p1) { return clamp(p0, p0,p1); }
 
 // CHECK: define [[FNATTRS]] i32 @_Z15test_clamp_uint
 // CHECK: call i32 @llvm.[[TARGET]].uclamp.i32(
@@ -68,6 +77,9 @@ uint3 test_clamp_uint3(uint3 p0, uint3 p1) { return clamp(p0, p1,p1); }
 // CHECK: define [[FNATTRS]] <4 x i32> @_Z16test_clamp_uint4
 // CHECK: call <4 x i32> @llvm.[[TARGET]].uclamp.v4i32
 uint4 test_clamp_uint4(uint4 p0, uint4 p1) { return clamp(p0, p1,p1); }
+// CHECK: define [[FNATTRS]] <4 x i32> {{.*}}test_clamp_uint4_mismatch
+// CHECK: call <4 x i32> @llvm.[[TARGET]].uclamp.v4i32
+uint4 test_clamp_uint4_mismatch(uint4 p0, uint p1) { return clamp(p0, p0,p1); }
 
 // CHECK: define [[FNATTRS]] i64 @_Z15test_clamp_long
 // CHECK: call i64 @llvm.[[TARGET]].sclamp.i64(
@@ -81,6 +93,9 @@ int64_t3 test_clamp_long3(int64_t3 p0, int64_t3 p1) { return clamp(p0, p1,p1); }
 // CHECK: define [[FNATTRS]] <4 x i64> @_Z16test_clamp_long4
 // CHECK: call <4 x i64> @llvm.[[TARGET]].sclamp.v4i64
 int64_t4 test_clamp_long4(int64_t4 p0, int64_t4 p1) { return clamp(p0, p1,p1); }
+// CHECK: define [[FNATTRS]] <4 x i64> {{.*}}test_clamp_long4_mismatch
+// CHECK: call <4 x i64> @llvm.[[TARGET]].sclamp.v4i64
+int64_t4 test_clamp_long4_mismatch(int64_t4 p0, int64_t4 p1) { return clamp(p0, p0,p1); }
 
 // CHECK: define [[FNATTRS]] i64 @_Z16test_clamp_ulong
 // CHECK: call i64 @llvm.[[TARGET]].uclamp.i64(
@@ -94,6 +109,9 @@ uint64_t3 test_clamp_ulong3(uint64_t3 p0, uint64_t3 p1) { return clamp(p0, p1,p1
 // CHECK: define [[FNATTRS]] <4 x i64> @_Z17test_clamp_ulong4
 // CHECK: call <4 x i64> @llvm.[[TARGET]].uclamp.v4i64
 uint64_t4 test_clamp_ulong4(uint64_t4 p0, uint64_t4 p1) { return clamp(p0, p1,p1); }
+// CHECK: define [[FNATTRS]] <4 x i64> {{.*}}test_clamp_ulong4_mismatch
+// CHECK: call <4 x i64> @llvm.[[TARGET]].uclamp.v4i64
+uint64_t4 test_clamp_ulong4_mismatch(uint64_t4 p0, uint64_t4 p1) { return clamp(p0, p0,p1); }
 
 // NATIVE_HALF: define [[FNATTRS]] [[FFNATTRS]] half @_Z15test_clamp_half
 // NATIVE_HALF: call reassoc nnan ninf nsz arcp afn half @llvm.[[TARGET]].nclamp.f16(
@@ -115,6 +133,11 @@ half3 test_clamp_half3(half3 p0, half3 p1) { return clamp(p0, p1,p1); }
 // NO_HALF: define [[FNATTRS]] [[FFNATTRS]] <4 x float> @_Z16test_clamp_half4
 // NO_HALF: call reassoc nnan ninf nsz arcp afn <4 x float> @llvm.[[TARGET]].nclamp.v4f32(
 half4 test_clamp_half4(half4 p0, half4 p1) { return clamp(p0, p1,p1); }
+// NATIVE_HALF: define [[FNATTRS]] [[FFNATTRS]] <4 x half> {{.*}}test_clamp_half4_mismatch
+// NATIVE_HALF: call reassoc nnan ninf nsz arcp afn <4 x half> @llvm.[[TARGET]].nclamp.v4f16
+// NO_HALF: define [[FNATTRS]] [[FFNATTRS]] <4 x float> {{.*}}test_clamp_half4_mismatch
+// NO_HALF: call reassoc nnan ninf nsz arcp afn <4 x float> @llvm.[[TARGET]].nclamp.v4f32(
+half4 test_clamp_half4_mismatch(half4 p0, half p1) { return clamp(p0, p0,p1); }
 
 // CHECK: define [[FNATTRS]] [[FFNATTRS]] float @_Z16test_clamp_float
 // CHECK: call reassoc nnan ninf nsz arcp afn float @llvm.[[TARGET]].nclamp.f32(
@@ -128,6 +151,9 @@ float3 test_clamp_float3(float3 p0, float3 p1) { return clamp(p0, p1,p1); }
 // CHECK: define [[FNATTRS]] [[FFNATTRS]] <4 x float> @_Z17test_clamp_float4
 // CHECK: call reassoc nnan ninf nsz arcp afn <4 x float> @llvm.[[TARGET]].nclamp.v4f32
 float4 test_clamp_float4(float4 p0, float4 p1) { return clamp(p0, p1,p1); }
+// CHECK: define [[FNATTRS]] [[FFNATTRS]] <4 x float> {{.*}}test_clamp_float4_mismatch
+// CHECK: call reassoc nnan ninf nsz arcp afn <4 x float> @llvm.[[TARGET]].nclamp.v4f32
+float4 test_clamp_float4_mismatch(float4 p0, float p1) { return clamp(p0, p0,p1); }
 
 // CHECK: define [[FNATTRS]] [[FFNATTRS]] double @_Z17test_clamp_double
 // CHECK: call reassoc nnan ninf nsz arcp afn double @llvm.[[TARGET]].nclamp.f64(
@@ -141,3 +167,9 @@ double3 test_clamp_double3(double3 p0, double3 p1) { return clamp(p0, p1,p1); }
 // CHECK: define [[FNATTRS]] [[FFNATTRS]] <4 x double> @_Z18test_clamp_double4
 // CHECK: call reassoc nnan ninf nsz arcp afn <4 x double> @llvm.[[TARGET]].nclamp.v4f64
 double4 test_clamp_double4(double4 p0, double4 p1) { return clamp(p0, p1,p1); }
+// CHECK: define [[FNATTRS]] [[FFNATTRS]] <4 x double> {{.*}}test_clamp_double4_mismatch
+// CHECK: call reassoc nnan ninf nsz arcp afn <4 x double> @llvm.[[TARGET]].nclamp.v4f64
+double4 test_clamp_double4_mismatch(double4 p0, double p1) { return clamp(p0, p0,p1); }
+// CHECK: define [[FNATTRS]] [[FFNATTRS]] <4 x double> {{.*}}test_clamp_double4_mismatch2
+// CHECK: call reassoc nnan ninf nsz arcp afn <4 x double> @llvm.[[TARGET]].nclamp.v4f64
+double4 test_clamp_double4_mismatch2(double4 p0, double p1) { return clamp(p0, p1,p0); }

clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl

@@ -1,4 +1,4 @@
-// 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
+// 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=note
 
 float2 test_no_second_arg(float2 p0) {
   return __builtin_hlsl_elementwise_clamp(p0);
@@ -23,51 +23,54 @@ float2 test_clamp_no_second_arg(float2 p0) {
 float2 test_clamp_vector_size_mismatch(float3 p0, float2 p1) {
   return clamp(p0, p0, p1);
   // expected-warning@-1 {{implicit conversion truncates vector: 'float3' (aka 'vector<float, 3>') to 'vector<float, 2>' (vector of 2 'float' values)}}
+  // expected-warning@-2 {{implicit conversion truncates vector: 'float3' (aka 'vector<float, 3>') to 'vector<float, 2>' (vector of 2 'float' values)}}
 }
 
 float2 test_clamp_builtin_vector_size_mismatch(float3 p0, float2 p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
   // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
+// allowed by the overloads in hlsl_compat_overloads.h
+// support for this overload might be removed in a future version of hlsl
 float test_clamp_scalar_mismatch(float p0, half p1) {
   return clamp(p1, p0, p1);
-  // expected-error@-1 {{call to 'clamp' is ambiguous}}
 }
 
+// allowed by the overloads in hlsl_compat_overloads.h
+// support for this overload might be removed in a future version of hlsl
 float2 test_clamp_element_type_mismatch(half2 p0, float2 p1) {
   return clamp(p1, p0, p1);
-  // expected-error@-1 {{call to 'clamp' is ambiguous}}
 }
 
 float2 test_builtin_clamp_float2_splat(float p0, float2 p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
-  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must be vectors}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float3 test_builtin_clamp_float3_splat(float p0, float3 p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
-  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must be vectors}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float4 test_builtin_clamp_float4_splat(float p0, float4 p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
-  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must be vectors}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float2 test_clamp_float2_int_splat(float2 p0, int p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
-  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must be vectors}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float3 test_clamp_float3_int_splat(float3 p0, int p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
-  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must be vectors}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float2 test_builtin_clamp_int_vect_to_float_vec_promotion(int2 p0, float p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p1, p1);
-  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must be vectors}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float test_builtin_clamp_bool_type_promotion(bool p0) {
@@ -77,15 +80,15 @@ float test_builtin_clamp_bool_type_promotion(bool p0) {
 
 float builtin_bool_to_float_type_promotion(float p0, bool p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p0, p1);
-  // expected-error@-1 {{3rd argument must be a floating point type (was 'bool')}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float builtin_bool_to_float_type_promotion2(bool p0, float p1) {
   return __builtin_hlsl_elementwise_clamp(p1, p0, p1);
-  // expected-error@-1 {{2nd argument must be a floating point type (was 'bool')}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }
 
 float builtin_clamp_int_to_float_promotion(float p0, int p1) {
   return __builtin_hlsl_elementwise_clamp(p0, p0, p1);
-  // expected-error@-1 {{3rd argument must be a floating point type (was 'int')}}
+  // expected-error@-1 {{all arguments to '__builtin_hlsl_elementwise_clamp' must have the same type}}
 }