https://github.com/spall updated
https://github.com/llvm/llvm-project/pull/129939
>From 23debaf2084f953e60847b8f0814c5d1ee27c726 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Tue, 4 Mar 2025 09:53:56 -0800
Subject: [PATCH 01/12] extra scalar vector overloads for clamp
---
.../lib/Headers/hlsl/hlsl_alias_intrinsics.h | 99 ++++++++++++-------
clang/test/CodeGenHLSL/builtins/clamp.hlsl | 32 ++++++
.../test/SemaHLSL/BuiltIns/clamp-errors.hlsl | 2 +-
3 files changed, 96 insertions(+), 37 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
index 89dfeb475488e..35246e222387e 100644
--- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
@@ -35,26 +35,44 @@ namespace hlsl {
#define _HLSL_16BIT_AVAILABILITY_STAGE(environment, version, stage)
#endif
-#define GEN_VEC_SCALAR_OVERLOADS(FUNC_NAME, BASE_TYPE, AVAIL)
\
- GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, BASE_TYPE##2, AVAIL)
\
- GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, BASE_TYPE##3, AVAIL)
\
- GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, BASE_TYPE##4, AVAIL)
-
-#define GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, VECTOR_TYPE, AVAIL)
\
- IF_TRUE_##AVAIL(
\
- _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_TYPE
\
- FUNC_NAME(VECTOR_TYPE p0, BASE_TYPE p1) {
\
- return __builtin_elementwise_##FUNC_NAME(p0, (VECTOR_TYPE)p1);
\
+#define _HLSL_CAT(a,b) a##b
+#define _HLSL_VEC_SCALAR_OVERLOADS(NAME, BASE_T, AVAIL) \
+ _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, _HLSL_CAT(_HLSL_NUM_ARGS_,NAME))
+
+#define _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, NUM_ARGS) \
+ _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T,2),
AVAIL) \
+ _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T,3),
AVAIL) \
+ _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T,4),
AVAIL)
+
+#define _HLSL_BOTH_OVERLOADS_2(NAME, BASE_T, VECTOR_T, AVAIL) \
+ _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)( \
+ _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T \
+ NAME(VECTOR_T p0, BASE_T p1) { \
+ return _HLSL_CAT(__builtin_elementwise_,NAME)(p0, (VECTOR_T)p1); \
}
\
- IF_TRUE_##AVAIL(
\
- _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_TYPE
\
- FUNC_NAME(BASE_TYPE p0, VECTOR_TYPE p1) {
\
- return __builtin_elementwise_##FUNC_NAME((VECTOR_TYPE)p0, p1);
\
+ _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)( \
+ _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T \
+ NAME(BASE_T p0, VECTOR_T p1) { \
+ return _HLSL_CAT(__builtin_elementwise_,NAME)((VECTOR_T)p0, p1); \
}
-#define IF_TRUE_0(EXPR)
-#define IF_TRUE_1(EXPR) EXPR
+#define _HLSL_BOTH_OVERLOADS_3(NAME, BASE_T, VECTOR_T, AVAIL) \
+ _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)(_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2))
\
+ constexpr VECTOR_T NAME(VECTOR_T p0, VECTOR_T p1, BASE_T p2) { \
+ return _HLSL_CAT(__builtin_hlsl_elementwise_,NAME)(p0, p1, (VECTOR_T)p2); \
+ } \
+ _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)(_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) \
+ constexpr VECTOR_T NAME(VECTOR_T p0, BASE_T p1, VECTOR_T p2) { \
+ return _HLSL_CAT(__builtin_hlsl_elementwise_,NAME)(p0, (VECTOR_T)p1, p2); \
+ }
+
+#define _HLSL_IF_TRUE_0(EXPR)
+#define _HLSL_IF_TRUE_1(EXPR) EXPR
+#define _HLSL_NUM_ARGS_min 2
+#define _HLSL_NUM_ARGS_max 2
+#define _HLSL_NUM_ARGS_clamp 3
+
//===----------------------------------------------------------------------===//
// abs builtins
//===----------------------------------------------------------------------===//
@@ -603,7 +621,8 @@ half3 clamp(half3, half3, half3);
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
half4 clamp(half4, half4, half4);
-
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, half, 1)
+
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
@@ -617,7 +636,8 @@ int16_t3 clamp(int16_t3, int16_t3, int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int16_t4 clamp(int16_t4, int16_t4, int16_t4);
-
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, int16_t, 1)
+
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint16_t clamp(uint16_t, uint16_t, uint16_t);
@@ -630,6 +650,7 @@ uint16_t3 clamp(uint16_t3, uint16_t3, uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint16_t4 clamp(uint16_t4, uint16_t4, uint16_t4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, uint16_t, 1)
#endif
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
@@ -640,6 +661,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int3 clamp(int3, int3, int3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int4 clamp(int4, int4, int4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, int, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint clamp(uint, uint, uint);
@@ -649,6 +671,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint3 clamp(uint3, uint3, uint3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint4 clamp(uint4, uint4, uint4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, uint, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int64_t clamp(int64_t, int64_t, int64_t);
@@ -658,6 +681,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int64_t3 clamp(int64_t3, int64_t3, int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int64_t4 clamp(int64_t4, int64_t4, int64_t4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, int64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint64_t clamp(uint64_t, uint64_t, uint64_t);
@@ -667,6 +691,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint64_t3 clamp(uint64_t3, uint64_t3, uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint64_t4 clamp(uint64_t4, uint64_t4, uint64_t4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, uint64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
float clamp(float, float, float);
@@ -676,6 +701,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
float3 clamp(float3, float3, float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
float4 clamp(float4, float4, float4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, float, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
double clamp(double, double, double);
@@ -685,6 +711,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
double3 clamp(double3, double3, double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
double4 clamp(double4, double4, double4);
+_HLSL_VEC_SCALAR_OVERLOADS(clamp, double, 0)
//===----------------------------------------------------------------------===//
// clip builtins
@@ -1597,7 +1624,7 @@ half3 max(half3, half3);
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
half4 max(half4, half4);
-GEN_VEC_SCALAR_OVERLOADS(max, half, 1)
+_HLSL_VEC_SCALAR_OVERLOADS(max, half, 1)
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
@@ -1612,7 +1639,7 @@ int16_t3 max(int16_t3, int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int16_t4 max(int16_t4, int16_t4);
-GEN_VEC_SCALAR_OVERLOADS(max, int16_t, 1)
+_HLSL_VEC_SCALAR_OVERLOADS(max, int16_t, 1)
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
@@ -1626,7 +1653,7 @@ uint16_t3 max(uint16_t3, uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint16_t4 max(uint16_t4, uint16_t4);
-GEN_VEC_SCALAR_OVERLOADS(max, uint16_t, 1)
+_HLSL_VEC_SCALAR_OVERLOADS(max, uint16_t, 1)
#endif
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
@@ -1637,7 +1664,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int3 max(int3, int3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int4 max(int4, int4);
-GEN_VEC_SCALAR_OVERLOADS(max, int, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(max, int, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint max(uint, uint);
@@ -1647,7 +1674,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint3 max(uint3, uint3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint4 max(uint4, uint4);
-GEN_VEC_SCALAR_OVERLOADS(max, uint, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(max, uint, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int64_t max(int64_t, int64_t);
@@ -1657,7 +1684,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int64_t3 max(int64_t3, int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int64_t4 max(int64_t4, int64_t4);
-GEN_VEC_SCALAR_OVERLOADS(max, int64_t, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(max, int64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint64_t max(uint64_t, uint64_t);
@@ -1667,7 +1694,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint64_t3 max(uint64_t3, uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint64_t4 max(uint64_t4, uint64_t4);
-GEN_VEC_SCALAR_OVERLOADS(max, uint64_t, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(max, uint64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
float max(float, float);
@@ -1677,7 +1704,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
float3 max(float3, float3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
float4 max(float4, float4);
-GEN_VEC_SCALAR_OVERLOADS(max, float, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(max, float, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
double max(double, double);
@@ -1687,7 +1714,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
double3 max(double3, double3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
double4 max(double4, double4);
-GEN_VEC_SCALAR_OVERLOADS(max, double, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(max, double, 0)
//===----------------------------------------------------------------------===//
// min builtins
@@ -1710,7 +1737,7 @@ half3 min(half3, half3);
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
half4 min(half4, half4);
-GEN_VEC_SCALAR_OVERLOADS(min, half, 1)
+_HLSL_VEC_SCALAR_OVERLOADS(min, half, 1)
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
@@ -1725,7 +1752,7 @@ int16_t3 min(int16_t3, int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int16_t4 min(int16_t4, int16_t4);
-GEN_VEC_SCALAR_OVERLOADS(min, int16_t, 1)
+_HLSL_VEC_SCALAR_OVERLOADS(min, int16_t, 1)
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
@@ -1739,7 +1766,7 @@ uint16_t3 min(uint16_t3, uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint16_t4 min(uint16_t4, uint16_t4);
-GEN_VEC_SCALAR_OVERLOADS(min, uint16_t, 1)
+_HLSL_VEC_SCALAR_OVERLOADS(min, uint16_t, 1)
#endif
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
@@ -1750,7 +1777,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int3 min(int3, int3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int4 min(int4, int4);
-GEN_VEC_SCALAR_OVERLOADS(min, int, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(min, int, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint min(uint, uint);
@@ -1760,7 +1787,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint3 min(uint3, uint3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint4 min(uint4, uint4);
-GEN_VEC_SCALAR_OVERLOADS(min, uint, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(min, uint, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
float min(float, float);
@@ -1770,7 +1797,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
float3 min(float3, float3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
float4 min(float4, float4);
-GEN_VEC_SCALAR_OVERLOADS(min, float, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(min, float, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int64_t min(int64_t, int64_t);
@@ -1780,7 +1807,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int64_t3 min(int64_t3, int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int64_t4 min(int64_t4, int64_t4);
-GEN_VEC_SCALAR_OVERLOADS(min, int64_t, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(min, int64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint64_t min(uint64_t, uint64_t);
@@ -1790,7 +1817,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint64_t3 min(uint64_t3, uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint64_t4 min(uint64_t4, uint64_t4);
-GEN_VEC_SCALAR_OVERLOADS(min, uint64_t, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(min, uint64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
double min(double, double);
@@ -1800,7 +1827,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
double3 min(double3, double3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
double4 min(double4, double4);
-GEN_VEC_SCALAR_OVERLOADS(min, double, 0)
+_HLSL_VEC_SCALAR_OVERLOADS(min, double, 0)
//===----------------------------------------------------------------------===//
// normalize builtins
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index d01c2a45c43c8..51454d6479708 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/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); }
diff --git a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
index 036f04cdac0b5..a1850d47b105d 100644
--- a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
@@ -22,7 +22,7 @@ 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-error@-1 {{call to 'clamp' is ambiguous}}
}
float2 test_clamp_builtin_vector_size_mismatch(float3 p0, float2 p1) {
>From 44a41ccbcaf7c4874c44802c1cceb5df12ecab26 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Wed, 5 Mar 2025 13:21:48 -0800
Subject: [PATCH 02/12] make clang format happy
---
.../lib/Headers/hlsl/hlsl_alias_intrinsics.h | 63 ++++++++++---------
1 file changed, 34 insertions(+), 29 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
index 35246e222387e..af306c9e6428e 100644
--- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
@@ -35,35 +35,40 @@ namespace hlsl {
#define _HLSL_16BIT_AVAILABILITY_STAGE(environment, version, stage)
#endif
-#define _HLSL_CAT(a,b) a##b
-#define _HLSL_VEC_SCALAR_OVERLOADS(NAME, BASE_T, AVAIL) \
- _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, _HLSL_CAT(_HLSL_NUM_ARGS_,NAME))
-
-#define _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, NUM_ARGS) \
- _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T,2),
AVAIL) \
- _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T,3),
AVAIL) \
- _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T,4),
AVAIL)
-
-#define _HLSL_BOTH_OVERLOADS_2(NAME, BASE_T, VECTOR_T, AVAIL) \
- _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)( \
- _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T \
- NAME(VECTOR_T p0, BASE_T p1) { \
- return _HLSL_CAT(__builtin_elementwise_,NAME)(p0, (VECTOR_T)p1); \
+#define _HLSL_CAT(a, b) a##b
+#define _HLSL_VEC_SCALAR_OVERLOADS(NAME, BASE_T, AVAIL)
\
+ _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, _HLSL_CAT(_HLSL_NUM_ARGS_, NAME))
+
+#define _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, NUM_ARGS)
\
+ _HLSL_CAT(_HLSL_BOTH_OVERLOADS_, NUM_ARGS)
\
+ (NAME, BASE_T, _HLSL_CAT(BASE_T, 2), AVAIL)
\
+ _HLSL_CAT(_HLSL_BOTH_OVERLOADS_, NUM_ARGS)(NAME, BASE_T,
\
+ _HLSL_CAT(BASE_T, 3), AVAIL)
\
+ _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,
\
+ NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T, 4), AVAIL)
+
+#define _HLSL_BOTH_OVERLOADS_2(NAME, BASE_T, VECTOR_T, AVAIL)
\
+ _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
+ (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
+ VECTOR_T p0, BASE_T p1) {
\
+ return _HLSL_CAT(__builtin_elementwise_, NAME)(p0, (VECTOR_T)p1);
\
}
\
- _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)( \
- _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T \
- NAME(BASE_T p0, VECTOR_T p1) { \
- return _HLSL_CAT(__builtin_elementwise_,NAME)((VECTOR_T)p0, p1); \
+ _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
+ (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
+ BASE_T p0, VECTOR_T p1) {
\
+ return _HLSL_CAT(__builtin_elementwise_, NAME)((VECTOR_T)p0, p1);
\
}
-#define _HLSL_BOTH_OVERLOADS_3(NAME, BASE_T, VECTOR_T, AVAIL) \
- _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)(_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2))
\
- constexpr VECTOR_T NAME(VECTOR_T p0, VECTOR_T p1, BASE_T p2) { \
- return _HLSL_CAT(__builtin_hlsl_elementwise_,NAME)(p0, p1, (VECTOR_T)p2); \
- } \
- _HLSL_CAT(_HLSL_IF_TRUE_,AVAIL)(_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) \
- constexpr VECTOR_T NAME(VECTOR_T p0, BASE_T p1, VECTOR_T p2) { \
- return _HLSL_CAT(__builtin_hlsl_elementwise_,NAME)(p0, (VECTOR_T)p1, p2); \
+#define _HLSL_BOTH_OVERLOADS_3(NAME, BASE_T, VECTOR_T, AVAIL)
\
+ _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
+ (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
+ VECTOR_T p0, VECTOR_T p1, BASE_T p2) {
\
+ return _HLSL_CAT(__builtin_hlsl_elementwise_, NAME)(p0, p1, (VECTOR_T)p2);
\
+ }
\
+ _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
+ (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
+ VECTOR_T p0, BASE_T p1, VECTOR_T p2) {
\
+ return _HLSL_CAT(__builtin_hlsl_elementwise_, NAME)(p0, (VECTOR_T)p1, p2);
\
}
#define _HLSL_IF_TRUE_0(EXPR)
@@ -72,7 +77,7 @@ namespace hlsl {
#define _HLSL_NUM_ARGS_min 2
#define _HLSL_NUM_ARGS_max 2
#define _HLSL_NUM_ARGS_clamp 3
-
+
//===----------------------------------------------------------------------===//
// abs builtins
//===----------------------------------------------------------------------===//
@@ -622,7 +627,7 @@ _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
half4 clamp(half4, half4, half4);
_HLSL_VEC_SCALAR_OVERLOADS(clamp, half, 1)
-
+
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
@@ -637,7 +642,7 @@ _HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int16_t4 clamp(int16_t4, int16_t4, int16_t4);
_HLSL_VEC_SCALAR_OVERLOADS(clamp, int16_t, 1)
-
+
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint16_t clamp(uint16_t, uint16_t, uint16_t);
>From ac260e6d3a04531bf863ac90b6901762e440ea8c Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Thu, 6 Mar 2025 14:09:52 -0800
Subject: [PATCH 03/12] implement clamp overloads with templates instead of
macros
---
clang/lib/Headers/CMakeLists.txt | 1 +
.../lib/Headers/hlsl/hlsl_alias_intrinsics.h | 100 ++++++------------
.../lib/Headers/hlsl/hlsl_compat_overloads.h | 45 ++++++++
clang/lib/Sema/SemaHLSL.cpp | 10 +-
4 files changed, 90 insertions(+), 66 deletions(-)
create mode 100644 clang/lib/Headers/hlsl/hlsl_compat_overloads.h
diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index d26de236998ca..acf49e40c447e 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/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
diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
index af306c9e6428e..a6b30c74c6ae9 100644
--- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
+++ b/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
@@ -35,48 +37,25 @@ namespace hlsl {
#define _HLSL_16BIT_AVAILABILITY_STAGE(environment, version, stage)
#endif
-#define _HLSL_CAT(a, b) a##b
-#define _HLSL_VEC_SCALAR_OVERLOADS(NAME, BASE_T, AVAIL)
\
- _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, _HLSL_CAT(_HLSL_NUM_ARGS_, NAME))
-
-#define _HLSL_ALL_OVERLOADS(NAME, BASE_T, AVAIL, NUM_ARGS)
\
- _HLSL_CAT(_HLSL_BOTH_OVERLOADS_, NUM_ARGS)
\
- (NAME, BASE_T, _HLSL_CAT(BASE_T, 2), AVAIL)
\
- _HLSL_CAT(_HLSL_BOTH_OVERLOADS_, NUM_ARGS)(NAME, BASE_T,
\
- _HLSL_CAT(BASE_T, 3), AVAIL)
\
- _HLSL_CAT(_HLSL_BOTH_OVERLOADS_,
\
- NUM_ARGS)(NAME, BASE_T, _HLSL_CAT(BASE_T, 4), AVAIL)
-
-#define _HLSL_BOTH_OVERLOADS_2(NAME, BASE_T, VECTOR_T, AVAIL)
\
- _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
- (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
- VECTOR_T p0, BASE_T p1) {
\
- return _HLSL_CAT(__builtin_elementwise_, NAME)(p0, (VECTOR_T)p1);
\
- }
\
- _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
- (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
- BASE_T p0, VECTOR_T p1) {
\
- return _HLSL_CAT(__builtin_elementwise_, NAME)((VECTOR_T)p0, p1);
\
- }
+#define GEN_VEC_SCALAR_OVERLOADS(FUNC_NAME, BASE_TYPE, AVAIL)
\
+ GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, BASE_TYPE##2, AVAIL)
\
+ GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, BASE_TYPE##3, AVAIL)
\
+ GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, BASE_TYPE##4, AVAIL)
-#define _HLSL_BOTH_OVERLOADS_3(NAME, BASE_T, VECTOR_T, AVAIL)
\
- _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
- (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
- VECTOR_T p0, VECTOR_T p1, BASE_T p2) {
\
- return _HLSL_CAT(__builtin_hlsl_elementwise_, NAME)(p0, p1, (VECTOR_T)p2);
\
+#define GEN_BOTH_OVERLOADS(FUNC_NAME, BASE_TYPE, VECTOR_TYPE, AVAIL)
\
+ IF_TRUE_##AVAIL(
\
+ _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_TYPE
\
+ FUNC_NAME(VECTOR_TYPE p0, BASE_TYPE p1) {
\
+ return __builtin_elementwise_##FUNC_NAME(p0, (VECTOR_TYPE)p1);
\
}
\
- _HLSL_CAT(_HLSL_IF_TRUE_, AVAIL)
\
- (_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_T NAME(
\
- VECTOR_T p0, BASE_T p1, VECTOR_T p2) {
\
- return _HLSL_CAT(__builtin_hlsl_elementwise_, NAME)(p0, (VECTOR_T)p1, p2);
\
+ IF_TRUE_##AVAIL(
\
+ _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)) constexpr VECTOR_TYPE
\
+ FUNC_NAME(BASE_TYPE p0, VECTOR_TYPE p1) {
\
+ return __builtin_elementwise_##FUNC_NAME((VECTOR_TYPE)p0, p1);
\
}
-#define _HLSL_IF_TRUE_0(EXPR)
-#define _HLSL_IF_TRUE_1(EXPR) EXPR
-
-#define _HLSL_NUM_ARGS_min 2
-#define _HLSL_NUM_ARGS_max 2
-#define _HLSL_NUM_ARGS_clamp 3
+#define IF_TRUE_0(EXPR)
+#define IF_TRUE_1(EXPR) EXPR
//===----------------------------------------------------------------------===//
// abs builtins
@@ -626,7 +605,6 @@ half3 clamp(half3, half3, half3);
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
half4 clamp(half4, half4, half4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, half, 1)
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
@@ -641,7 +619,6 @@ int16_t3 clamp(int16_t3, int16_t3, int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int16_t4 clamp(int16_t4, int16_t4, int16_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, int16_t, 1)
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
@@ -655,7 +632,6 @@ uint16_t3 clamp(uint16_t3, uint16_t3, uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint16_t4 clamp(uint16_t4, uint16_t4, uint16_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, uint16_t, 1)
#endif
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
@@ -666,7 +642,6 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int3 clamp(int3, int3, int3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int4 clamp(int4, int4, int4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, int, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint clamp(uint, uint, uint);
@@ -676,7 +651,6 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint3 clamp(uint3, uint3, uint3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint4 clamp(uint4, uint4, uint4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, uint, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int64_t clamp(int64_t, int64_t, int64_t);
@@ -686,7 +660,6 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int64_t3 clamp(int64_t3, int64_t3, int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
int64_t4 clamp(int64_t4, int64_t4, int64_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, int64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint64_t clamp(uint64_t, uint64_t, uint64_t);
@@ -696,7 +669,6 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint64_t3 clamp(uint64_t3, uint64_t3, uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
uint64_t4 clamp(uint64_t4, uint64_t4, uint64_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, uint64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
float clamp(float, float, float);
@@ -706,7 +678,6 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
float3 clamp(float3, float3, float3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
float4 clamp(float4, float4, float4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, float, 0)
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
double clamp(double, double, double);
@@ -716,7 +687,6 @@ _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
double3 clamp(double3, double3, double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_clamp)
double4 clamp(double4, double4, double4);
-_HLSL_VEC_SCALAR_OVERLOADS(clamp, double, 0)
//===----------------------------------------------------------------------===//
// clip builtins
@@ -1629,7 +1599,7 @@ half3 max(half3, half3);
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
half4 max(half4, half4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, half, 1)
+GEN_VEC_SCALAR_OVERLOADS(max, half, 1)
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
@@ -1644,7 +1614,7 @@ int16_t3 max(int16_t3, int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int16_t4 max(int16_t4, int16_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, int16_t, 1)
+GEN_VEC_SCALAR_OVERLOADS(max, int16_t, 1)
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
@@ -1658,7 +1628,7 @@ uint16_t3 max(uint16_t3, uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint16_t4 max(uint16_t4, uint16_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, uint16_t, 1)
+GEN_VEC_SCALAR_OVERLOADS(max, uint16_t, 1)
#endif
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
@@ -1669,7 +1639,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int3 max(int3, int3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int4 max(int4, int4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, int, 0)
+GEN_VEC_SCALAR_OVERLOADS(max, int, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint max(uint, uint);
@@ -1679,7 +1649,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint3 max(uint3, uint3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint4 max(uint4, uint4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, uint, 0)
+GEN_VEC_SCALAR_OVERLOADS(max, uint, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int64_t max(int64_t, int64_t);
@@ -1689,7 +1659,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int64_t3 max(int64_t3, int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
int64_t4 max(int64_t4, int64_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, int64_t, 0)
+GEN_VEC_SCALAR_OVERLOADS(max, int64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint64_t max(uint64_t, uint64_t);
@@ -1699,7 +1669,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint64_t3 max(uint64_t3, uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
uint64_t4 max(uint64_t4, uint64_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, uint64_t, 0)
+GEN_VEC_SCALAR_OVERLOADS(max, uint64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
float max(float, float);
@@ -1709,7 +1679,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
float3 max(float3, float3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
float4 max(float4, float4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, float, 0)
+GEN_VEC_SCALAR_OVERLOADS(max, float, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
double max(double, double);
@@ -1719,7 +1689,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
double3 max(double3, double3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_max)
double4 max(double4, double4);
-_HLSL_VEC_SCALAR_OVERLOADS(max, double, 0)
+GEN_VEC_SCALAR_OVERLOADS(max, double, 0)
//===----------------------------------------------------------------------===//
// min builtins
@@ -1742,7 +1712,7 @@ half3 min(half3, half3);
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
half4 min(half4, half4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, half, 1)
+GEN_VEC_SCALAR_OVERLOADS(min, half, 1)
#ifdef __HLSL_ENABLE_16_BIT
_HLSL_AVAILABILITY(shadermodel, 6.2)
@@ -1757,7 +1727,7 @@ int16_t3 min(int16_t3, int16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int16_t4 min(int16_t4, int16_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, int16_t, 1)
+GEN_VEC_SCALAR_OVERLOADS(min, int16_t, 1)
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
@@ -1771,7 +1741,7 @@ uint16_t3 min(uint16_t3, uint16_t3);
_HLSL_AVAILABILITY(shadermodel, 6.2)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint16_t4 min(uint16_t4, uint16_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, uint16_t, 1)
+GEN_VEC_SCALAR_OVERLOADS(min, uint16_t, 1)
#endif
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
@@ -1782,7 +1752,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int3 min(int3, int3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int4 min(int4, int4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, int, 0)
+GEN_VEC_SCALAR_OVERLOADS(min, int, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint min(uint, uint);
@@ -1792,7 +1762,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint3 min(uint3, uint3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint4 min(uint4, uint4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, uint, 0)
+GEN_VEC_SCALAR_OVERLOADS(min, uint, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
float min(float, float);
@@ -1802,7 +1772,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
float3 min(float3, float3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
float4 min(float4, float4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, float, 0)
+GEN_VEC_SCALAR_OVERLOADS(min, float, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int64_t min(int64_t, int64_t);
@@ -1812,7 +1782,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int64_t3 min(int64_t3, int64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
int64_t4 min(int64_t4, int64_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, int64_t, 0)
+GEN_VEC_SCALAR_OVERLOADS(min, int64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint64_t min(uint64_t, uint64_t);
@@ -1822,7 +1792,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint64_t3 min(uint64_t3, uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
uint64_t4 min(uint64_t4, uint64_t4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, uint64_t, 0)
+GEN_VEC_SCALAR_OVERLOADS(min, uint64_t, 0)
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
double min(double, double);
@@ -1832,7 +1802,7 @@ _HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
double3 min(double3, double3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_min)
double4 min(double4, double4);
-_HLSL_VEC_SCALAR_OVERLOADS(min, double, 0)
+GEN_VEC_SCALAR_OVERLOADS(min, double, 0)
//===----------------------------------------------------------------------===//
// normalize builtins
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
new file mode 100644
index 0000000000000..a6b2b1ea667ac
--- /dev/null
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -0,0 +1,45 @@
+//===--- 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_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 2e6a333f3d768..f26f22dc87741 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2392,8 +2392,16 @@ 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) ||
+ !(SemaRef.Context.hasSameUnqualifiedType(TheCall->getArg(0)->getType(),
+
TheCall->getArg(1)->getType()) &&
+ SemaRef.Context.hasSameUnqualifiedType(TheCall->getArg(0)->getType(),
+
TheCall->getArg(2)->getType()))) {
+ SemaRef.Diag(TheCall->getBeginLoc(),
diag::err_typecheck_call_different_arg_types)
+ << TheCall->getArg(0)->getType() << TheCall->getArg(1)->getType()
+ << TheCall->getBeginLoc() << TheCall->getBeginLoc();
return true;
+ }
if (SemaRef.BuiltinElementwiseTernaryMath(
TheCall, /*CheckForFloatArgs*/
TheCall->getArg(0)->getType()->hasFloatingRepresentation()))
>From 3311cbd2b0a7dc6f766f1923e5f561cf5cc67806 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Mon, 10 Mar 2025 11:49:26 -0700
Subject: [PATCH 04/12] update semantic checking for
__builtin_hlsl_elementwise_clamp; update test with new error message
---
clang/lib/Sema/SemaHLSL.cpp | 23 ++++++++++-----
.../test/SemaHLSL/BuiltIns/clamp-errors.hlsl | 29 ++++++++++---------
2 files changed, 31 insertions(+), 21 deletions(-)
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index f26f22dc87741..c0ed8f84d61ca 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2041,6 +2041,20 @@ 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)) {
@@ -2393,15 +2407,8 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
if (SemaRef.checkArgCount(TheCall, 3))
return true;
if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0) ||
- !(SemaRef.Context.hasSameUnqualifiedType(TheCall->getArg(0)->getType(),
-
TheCall->getArg(1)->getType()) &&
- SemaRef.Context.hasSameUnqualifiedType(TheCall->getArg(0)->getType(),
-
TheCall->getArg(2)->getType()))) {
- SemaRef.Diag(TheCall->getBeginLoc(),
diag::err_typecheck_call_different_arg_types)
- << TheCall->getArg(0)->getType() << TheCall->getArg(1)->getType()
- << TheCall->getBeginLoc() << TheCall->getBeginLoc();
+ CheckAllArgsHaveSameType(&SemaRef, TheCall))
return true;
- }
if (SemaRef.BuiltinElementwiseTernaryMath(
TheCall, /*CheckForFloatArgs*/
TheCall->getArg(0)->getType()->hasFloatingRepresentation()))
diff --git a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
index a1850d47b105d..cf53a1126dbaf 100644
--- a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
+++ b/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);
@@ -22,7 +22,8 @@ float2 test_clamp_no_second_arg(float2 p0) {
float2 test_clamp_vector_size_mismatch(float3 p0, float2 p1) {
return clamp(p0, p0, p1);
- // expected-error@-1 {{call to 'clamp' is ambiguous}}
+ // 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) {
@@ -30,44 +31,46 @@ float2 test_clamp_builtin_vector_size_mismatch(float3 p0,
float2 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}}
}
>From fec28cf6856b542e8f20e300794fae01840ed320 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Mon, 10 Mar 2025 11:54:59 -0700
Subject: [PATCH 05/12] make clang format happy
---
.../lib/Headers/hlsl/hlsl_compat_overloads.h | 49 ++++++++++++-------
clang/lib/Sema/SemaHLSL.cpp | 10 ++--
2 files changed, 38 insertions(+), 21 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index a6b2b1ea667ac..2e769ec02e1df 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -1,4 +1,5 @@
-//===--- hlsl_compat_overloads.h - Additional HLSL overload definitions for
intrinsics --===//
+//===--- 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.
@@ -20,26 +21,40 @@ namespace hlsl {
// 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, 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 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 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 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);
+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_
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index c0ed8f84d61ca..564eed7bd2989 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2045,10 +2045,12 @@ 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())) {
+ 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());
+ << TheCall->getDirectCallee() << /*useAllTerminology*/ true
+ << SourceRange(TheCall->getArg(0)->getBeginLoc(),
+ TheCall->getArg(N - 1)->getEndLoc());
return true;
}
}
@@ -2407,7 +2409,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned
BuiltinID, CallExpr *TheCall) {
if (SemaRef.checkArgCount(TheCall, 3))
return true;
if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0) ||
- CheckAllArgsHaveSameType(&SemaRef, TheCall))
+ CheckAllArgsHaveSameType(&SemaRef, TheCall))
return true;
if (SemaRef.BuiltinElementwiseTernaryMath(
TheCall, /*CheckForFloatArgs*/
>From 101beb2f016131d669ea676359c7f57c33e0d917 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Mon, 10 Mar 2025 12:23:25 -0700
Subject: [PATCH 06/12] make clang format happy again
---
clang/lib/Headers/hlsl/hlsl_compat_overloads.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index 2e769ec02e1df..673180f7a5802 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -1,5 +1,5 @@
//===--- hlsl_compat_overloads.h - Additional HLSL overload definitions for
-//intrinsics --===//
+// intrinsics --===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM
Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
>From 69dab196e89f908972229a5f440ca6cc088b7eeb Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Tue, 11 Mar 2025 12:42:13 -0700
Subject: [PATCH 07/12] self review + respond to pr comments
---
clang/lib/Headers/hlsl.h | 1 +
.../lib/Headers/hlsl/hlsl_alias_intrinsics.h | 2 --
clang/lib/Sema/SemaHLSL.cpp | 1 +
clang/test/CodeGenHLSL/builtins/clamp.hlsl | 20 +++++++++++++++++++
.../test/SemaHLSL/BuiltIns/clamp-errors.hlsl | 20 +++++++++++++++++++
5 files changed, 42 insertions(+), 2 deletions(-)
diff --git a/clang/lib/Headers/hlsl.h b/clang/lib/Headers/hlsl.h
index 4be68eb84a34a..e4c43ab5ee539 100644
--- a/clang/lib/Headers/hlsl.h
+++ b/clang/lib/Headers/hlsl.h
@@ -23,6 +23,7 @@
// HLSL standard library function declarations/definitions.
#include "hlsl/hlsl_alias_intrinsics.h"
#include "hlsl/hlsl_intrinsics.h"
+#include "hlsl/hlsl_compat_overloads.h"
#if defined(__clang__)
#pragma clang diagnostic pop
diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
index a6b30c74c6ae9..89dfeb475488e 100644
--- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
@@ -9,8 +9,6 @@
#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
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 564eed7bd2989..b55b8f9e43efa 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2042,6 +2042,7 @@ static bool CheckVectorElementCallArgs(Sema *S, CallExpr
*TheCall) {
}
static bool CheckAllArgsHaveSameType(Sema *S, CallExpr *TheCall) {
+ assert(TheCall->getNumArgs() > 1);
QualType ArgTy0 = TheCall->getArg(0)->getType();
for (unsigned I = 1, N = TheCall->getNumArgs(); I < N; ++I) {
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index 51454d6479708..a3483898a0052 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
@@ -173,3 +173,23 @@ double4 test_clamp_double4_mismatch(double4 p0, double p1)
{ return clamp(p0, p0
// 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); }
+
+// CHECK: define [[FNATTRS]] <2 x i32> {{.*}}_overloads1
+// CHECK: call <2 x i32> @llvm.[[TARGET]].sclamp.v2i32
+int2 test_overloads1(int2 p0, float2 p1, uint p2) { return clamp(p0, p1, p2); }
+
+// CHECK: define [[FNATTRS]] [[FFNATTRS]] <2 x float> {{.*}}test_overloads2
+// CHECK: call reassoc nnan ninf nsz arcp afn <2 x float>
@llvm.[[TARGET]].nclamp.v2f32
+float2 test_overloads2(float2 p0, uint p1, int2 p2) { return clamp(p0, p1,
p2); }
+
+// CHECK: define [[FNATTRS]] <3 x i32> {{.*}}test_overloads3
+// CHECK: call <3 x i32> @llvm.[[TARGET]].uclamp.v3i32
+uint3 test_overloads3(uint3 p0, int p1, float p2) { return clamp(p0, p1, p2); }
+
+// CHECK: define [[FNATTRS]] [[FFNATTRS]] <4 x double> {{.*}}test_overloads4
+// CHECK: call reassoc nnan ninf nsz arcp afn <4 x double>
@llvm.[[TARGET]].nclamp.v4f64
+double4 test_overloads4(double4 p0, float4 p1, int4 p2) { return clamp(p0, p1,
p2); }
+
+// CHECK: define [[FNATTRS]] i32 {{.*}}test_overloads5
+// CHECK: call i32 @llvm.[[TARGET]].sclamp.i32(
+int test_overloads5(int p0, uint p1, double p2) { return clamp(p0, p1, p2); }
diff --git a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
index cf53a1126dbaf..58002047c9454 100644
--- a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
@@ -20,6 +20,26 @@ float2 test_clamp_no_second_arg(float2 p0) {
// expected-error@-1 {{no matching function for call to 'clamp'}}
}
+float test_scalar_first_arg(float p0, float2 p1) {
+ return clamp(p0, p1, p1);
+ // expected-error@-1 {{call to 'clamp' is ambiguous}}
+}
+
+float test_scalar_first_arg2(float p0, float2 p1) {
+ return clamp(p0, p0, p1);
+ // expected-error@-1 {{call to 'clamp' is ambiguous}}
+}
+
+float2 test_scalar_first_arg3(float p0, float2 p1) {
+ return clamp(p0, p0, p1);
+ // expected-error@-1 {{call to 'clamp' is ambiguous}}
+}
+
+float3 test_thing(float3 p0, float2 p1) {
+ return clamp(p0, p0, p1);
+ // expected-error@-1 {{cannot initialize return object of type 'float3' (aka
'vector<float, 3>') with an rvalue of type 'vector<float, 2>' (vector of 2
'float' values)}}
+}
+
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)}}
>From 47c90c4be48c0c54dbfa803cb15fada50061dfa5 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Tue, 11 Mar 2025 15:13:18 -0700
Subject: [PATCH 08/12] add vector size checks to clamp templates
---
clang/lib/Headers/hlsl/hlsl_compat_overloads.h | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index 673180f7a5802..15747b4ad0668 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -22,27 +22,27 @@ namespace hlsl {
//===----------------------------------------------------------------------===//
template <typename T, typename R, typename U, uint N>
-constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value, vector<T,
N>>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value && (N > 1 &&
N <= 4), 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>>
+constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value && (N > 1 &&
N <= 4), 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,
+ __detail::is_arithmetic<V>::Value && (N > 1 && N <= 4),
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,
+constexpr __detail::enable_if_t<(N > 1 && N <= 4), 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);
>From dd0765a5d99256df3d3ee7cfacd9e9dbc0e1cdf6 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Tue, 11 Mar 2025 15:24:32 -0700
Subject: [PATCH 09/12] make clang format happy
---
clang/lib/Headers/hlsl.h | 2 +-
clang/lib/Headers/hlsl/hlsl_compat_overloads.h | 13 ++++++++-----
2 files changed, 9 insertions(+), 6 deletions(-)
diff --git a/clang/lib/Headers/hlsl.h b/clang/lib/Headers/hlsl.h
index e4c43ab5ee539..d233f6092ffcd 100644
--- a/clang/lib/Headers/hlsl.h
+++ b/clang/lib/Headers/hlsl.h
@@ -22,8 +22,8 @@
// HLSL standard library function declarations/definitions.
#include "hlsl/hlsl_alias_intrinsics.h"
-#include "hlsl/hlsl_intrinsics.h"
#include "hlsl/hlsl_compat_overloads.h"
+#include "hlsl/hlsl_intrinsics.h"
#if defined(__clang__)
#pragma clang diagnostic pop
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index 15747b4ad0668..80008f108ba20 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -22,28 +22,31 @@ namespace hlsl {
//===----------------------------------------------------------------------===//
template <typename T, typename R, typename U, uint N>
-constexpr __detail::enable_if_t<__detail::is_arithmetic<U>::Value && (N > 1 &&
N <= 4), vector<T, N>>
+constexpr __detail::enable_if_t<
+ __detail::is_arithmetic<U>::Value && (N > 1 && N <= 4), 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 && (N > 1 &&
N <= 4), vector<T, N>>
+constexpr __detail::enable_if_t<
+ __detail::is_arithmetic<U>::Value && (N > 1 && N <= 4), 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 && (N > 1 && N <= 4),
+ __detail::is_arithmetic<V>::Value &&
+ (N > 1 && N <= 4),
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 __detail::enable_if_t<(N > 1 && N <= 4), vector<T, N>>
clamp(vector<T, N> p0, vector<R, N> p1,
- vector<S, N> p2) {
+constexpr __detail::enable_if_t<(N > 1 && N <= 4), 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);
}
>From 819cae525644f74353b72a49399880c5a4f26f94 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Tue, 11 Mar 2025 15:47:22 -0700
Subject: [PATCH 10/12] test for float5 error
---
clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
index 58002047c9454..745516932442e 100644
--- a/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/clamp-errors.hlsl
@@ -40,6 +40,14 @@ float3 test_thing(float3 p0, float2 p1) {
// expected-error@-1 {{cannot initialize return object of type 'float3' (aka
'vector<float, 3>') with an rvalue of type 'vector<float, 2>' (vector of 2
'float' values)}}
}
+typedef float float5 __attribute__((ext_vector_type(5)));
+
+// check vectors of wrong size are rejected
+float5 vec_too_big(float5 p0) {
+ return clamp(p0, p0, p0);
+ // expected-error@-1 {{call to 'clamp' is ambiguous}}
+}
+
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)}}
>From 8e01d99b985649deeb3febc56fd45c081b046d80 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Thu, 13 Mar 2025 11:06:00 -0700
Subject: [PATCH 11/12] fix two tests
---
clang/test/CodeGenHLSL/builtins/clamp.hlsl | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
index a3483898a0052..f04b6096fa220 100644
--- a/clang/test/CodeGenHLSL/builtins/clamp.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/clamp.hlsl
@@ -95,7 +95,7 @@ int64_t3 test_clamp_long3(int64_t3 p0, int64_t3 p1) { return
clamp(p0, p1,p1); }
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); }
+int64_t4 test_clamp_long4_mismatch(int64_t4 p0, int64_t p1) { return clamp(p0,
p0,p1); }
// CHECK: define [[FNATTRS]] i64 @_Z16test_clamp_ulong
// CHECK: call i64 @llvm.[[TARGET]].uclamp.i64(
@@ -111,7 +111,7 @@ uint64_t3 test_clamp_ulong3(uint64_t3 p0, uint64_t3 p1) {
return clamp(p0, p1,p1
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); }
+uint64_t4 test_clamp_ulong4_mismatch(uint64_t4 p0, uint64_t 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(
>From 4ed05077627897c64eacc8b78db1a1272c64d21b Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahsp...@microsoft.com>
Date: Fri, 14 Mar 2025 17:03:05 -0700
Subject: [PATCH 12/12] change clamp templates so they call clamp instead of
__builtin_hlsl_elementwise_clamp, to ensure clamp can't be called on 16 bit
types in a shader model before 6.2. Add tests to show this
---
.../lib/Headers/hlsl/hlsl_compat_overloads.h | 14 ++++------
.../SemaHLSL/BuiltIns/clamp-errors-16bit.hlsl | 27 +++++++++++++++++++
2 files changed, 32 insertions(+), 9 deletions(-)
create mode 100644 clang/test/SemaHLSL/BuiltIns/clamp-errors-16bit.hlsl
diff --git a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
index 80008f108ba20..3a939c02464e8 100644
--- a/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
+++ b/clang/lib/Headers/hlsl/hlsl_compat_overloads.h
@@ -25,15 +25,13 @@ template <typename T, typename R, typename U, uint N>
constexpr __detail::enable_if_t<
__detail::is_arithmetic<U>::Value && (N > 1 && N <= 4), 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);
+ return 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 && (N > 1 && N <= 4), 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);
+ return 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 &&
@@ -41,14 +39,12 @@ constexpr
__detail::enable_if_t<__detail::is_arithmetic<U>::Value &&
(N > 1 && N <= 4),
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);
+ return clamp(p0, (vector<T, N>)p1, (vector<T, N>)p2);
}
template <typename T, typename R, typename S, uint N>
constexpr __detail::enable_if_t<(N > 1 && N <= 4), 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);
+ return 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 &&
@@ -56,7 +52,7 @@ constexpr
__detail::enable_if_t<__detail::is_arithmetic<U>::Value &&
__detail::is_arithmetic<W>::Value,
U>
clamp(U p0, V p1, W p2) {
- return __builtin_hlsl_elementwise_clamp(p0, (U)p1, (U)p2);
+ return clamp(p0, (U)p1, (U)p2);
}
} // namespace hlsl
diff --git a/clang/test/SemaHLSL/BuiltIns/clamp-errors-16bit.hlsl
b/clang/test/SemaHLSL/BuiltIns/clamp-errors-16bit.hlsl
new file mode 100644
index 0000000000000..976dbaaa065b2
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/clamp-errors-16bit.hlsl
@@ -0,0 +1,27 @@
+// RUN: not %clang_dxc -enable-16bit-types -T cs_6_0 -HV 202x %s 2>&1 |
FileCheck %s
+
+// check we error on 16 bit type if shader model is too old
+// CHECK: '-enable-16bit-types' option requires target HLSL Version >= 2018
and shader model >= 6.2, but HLSL Version is 'hlsl202x' and shader model is
'6.0'
+int16_t test_int16_t_error(int16_t p0, int p1) {
+ return clamp(p0, p0, p1);
+}
+
+int16_t3 test_int16_t3_error(int16_t3 p0, int3 p1) {
+ return clamp(p0, p0, p1);
+}
+
+half test_half_error(half p0, int p1) {
+ return clamp(p0, p1, p1);
+}
+
+half3 test_half3_error(half3 p0, int3 p1) {
+ return clamp(p0, p0, p1);
+}
+
+uint16_t test_uint16_t_error(uint16_t p0, int p1) {
+ return clamp(p0, p0, p1);
+}
+
+uint16_t3 test_uint16_t3_error(uint16_t3 p0, int3 p1) {
+ return clamp(p0, p1, p1);
+}
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
