@cpp-linter/clang-tidy-linux-arm64 22.1.8 → 23.1.4

package/lib/clang/{22 → 23}/include/gpuintrin.h

@@ -32,30 +32,6 @@ _Pragma("push_macro(\"bool\")");
 #define bool _Bool
 #endif
 
-_Pragma("omp begin declare target device_type(nohost)");
-_Pragma("omp begin declare variant match(device = {kind(gpu)})");
-
-// Forward declare a few functions for the implementation header.
-
-// Returns a bitmask marking all lanes that have the same value of __x.
-_DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_any_u32_impl(uint64_t __lane_mask, uint32_t __x);
-
-// Returns a bitmask marking all lanes that have the same value of __x.
-_DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_any_u64_impl(uint64_t __lane_mask, uint64_t __x);
-
-// Returns the current lane mask if every lane contains __x.
-_DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_all_u32_impl(uint64_t __lane_mask, uint32_t __x);
-
-// Returns the current lane mask if every lane contains __x.
-_DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_all_u64_impl(uint64_t __lane_mask, uint64_t __x);
-
-_Pragma("omp end declare variant");
-_Pragma("omp end declare target");
-
 #if defined(__NVPTX__)
 #include <nvptxintrin.h>
 #elif defined(__AMDGPU__)
@@ -69,6 +45,9 @@ _Pragma("omp end declare target");
 _Pragma("omp begin declare target device_type(nohost)");
 _Pragma("omp begin declare variant match(device = {kind(gpu)})");
 
+// Attribute to declare a function as a kernel.
+#define __gpu_kernel __attribute__((device_kernel, visibility("protected")))
+
 #define __GPU_X_DIM 0
 #define __GPU_Y_DIM 1
 #define __GPU_Z_DIM 2
@@ -144,11 +123,10 @@ __gpu_is_first_in_lane(uint64_t __lane_mask) {
 // Copies the value from the first active thread to the rest.
 _DEFAULT_FN_ATTRS static __inline__ uint64_t
 __gpu_read_first_lane_u64(uint64_t __lane_mask, uint64_t __x) {
-  uint32_t __hi = (uint32_t)(__x >> 32ull);
-  uint32_t __lo = (uint32_t)(__x & 0xFFFFFFFFull);
-  return ((uint64_t)__gpu_read_first_lane_u32(__lane_mask, __hi) << 32ull) |
-         ((uint64_t)__gpu_read_first_lane_u32(__lane_mask, __lo) &
-          0xFFFFFFFFull);
+  uint32_t __hi = (uint32_t)(__x >> 32);
+  uint32_t __lo = (uint32_t)(__x & 0xFFFFFFFF);
+  return ((uint64_t)__gpu_read_first_lane_u32(__lane_mask, __hi) << 32) |
+         ((uint64_t)__gpu_read_first_lane_u32(__lane_mask, __lo) & 0xFFFFFFFF);
 }
 
 // Gets the first floating point value from the active lanes.
@@ -171,11 +149,10 @@ __gpu_read_first_lane_f64(uint64_t __lane_mask, double __x) {
 _DEFAULT_FN_ATTRS static __inline__ uint64_t
 __gpu_shuffle_idx_u64(uint64_t __lane_mask, uint32_t __idx, uint64_t __x,
                       uint32_t __width) {
-  uint32_t __hi = (uint32_t)(__x >> 32ull);
+  uint32_t __hi = (uint32_t)(__x >> 32);
   uint32_t __lo = (uint32_t)(__x & 0xFFFFFFFF);
   uint32_t __mask = (uint32_t)__lane_mask;
-  return ((uint64_t)__gpu_shuffle_idx_u32(__mask, __idx, __hi, __width)
-          << 32ull) |
+  return ((uint64_t)__gpu_shuffle_idx_u32(__mask, __idx, __hi, __width) << 32) |
          ((uint64_t)__gpu_shuffle_idx_u32(__mask, __idx, __lo, __width));
 }
 
@@ -198,125 +175,166 @@ __gpu_shuffle_idx_f64(uint64_t __lane_mask, uint32_t __idx, double __x,
                             __builtin_bit_cast(uint64_t, __x), __width));
 }
 
-// Gets the accumulator scan of the threads in the warp or wavefront.
-#define __DO_LANE_SCAN(__type, __bitmask_type, __suffix)                       \
-  _DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_lane_scan_##__suffix(     \
-      uint64_t __lane_mask, uint32_t __x) {                                    \
-    uint64_t __first = __lane_mask >> __builtin_ctzll(__lane_mask);            \
-    bool __divergent = __gpu_read_first_lane_##__suffix(                       \
-        __lane_mask, __first & (__first + 1));                                 \
-    if (__divergent) {                                                         \
-      __type __accum = 0;                                                      \
-      for (uint64_t __mask = __lane_mask; __mask; __mask &= __mask - 1) {      \
-        __type __index = __builtin_ctzll(__mask);                              \
-        __type __tmp = __gpu_shuffle_idx_##__suffix(__lane_mask, __index, __x, \
-                                                    __gpu_num_lanes());        \
-        __x = __gpu_lane_id() == __index ? __accum + __tmp : __x;              \
-        __accum += __tmp;                                                      \
-      }                                                                        \
-    } else {                                                                   \
-      for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) {     \
-        uint32_t __index = __gpu_lane_id() - __step;                           \
-        __bitmask_type bitmask = __gpu_lane_id() >= __step;                    \
-        __x += __builtin_bit_cast(                                             \
-            __type,                                                            \
-            -bitmask & __builtin_bit_cast(__bitmask_type,                      \
-                                          __gpu_shuffle_idx_##__suffix(        \
-                                              __lane_mask, __index, __x,       \
-                                              __gpu_num_lanes())));            \
-      }                                                                        \
+// Implements scan and reduction operations across a GPU warp or wavefront.
+//
+// Both scans work by iterating log2(N) steps. The bitmask tracks the currently
+// unprocessed lanes, above or below the current lane in the case of a suffix or
+// prefix scan. Each iteration we shuffle in the unprocessed neighbors and then
+// clear the bits that this operation handled.
+#define __DO_LANE_OPS(__type, __op, __identity, __prefix, __suffix)            \
+  _DEFAULT_FN_ATTRS static __inline__ __type                                   \
+  __gpu_suffix_scan_##__prefix##_##__suffix(uint64_t __lane_mask,              \
+                                            __type __x) {                      \
+    uint64_t __above = __lane_mask & -(UINT64_C(2) << __gpu_lane_id());        \
+    for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) {       \
+      uint32_t __src = __builtin_ctzg(__above, (int)sizeof(__above) * 8);      \
+      __type __result = __gpu_shuffle_idx_##__suffix(__lane_mask, __src, __x,  \
+                                                     __gpu_num_lanes());       \
+      __x = __op(__x, __above ? __result : (__type)__identity);                \
+      for (uint32_t __i = 0; __i < __step; ++__i)                              \
+        __above &= __above - 1;                                                \
     }                                                                          \
     return __x;                                                                \
-  }
-__DO_LANE_SCAN(uint32_t, uint32_t, u32); // uint32_t __gpu_lane_scan_u32(m, x)
-__DO_LANE_SCAN(uint64_t, uint64_t, u64); // uint64_t __gpu_lane_scan_u64(m, x)
-__DO_LANE_SCAN(float, uint32_t, f32);    // float __gpu_lane_scan_f32(m, x)
-__DO_LANE_SCAN(double, uint64_t, f64);   // double __gpu_lane_scan_f64(m, x)
-#undef __DO_LANE_SCAN
-
-// Gets the sum of all lanes inside the warp or wavefront.
-#define __DO_LANE_SUM(__type, __suffix)                                        \
-  _DEFAULT_FN_ATTRS static __inline__ __type __gpu_lane_sum_##__suffix(        \
-      uint64_t __lane_mask, __type __x) {                                      \
-    uint64_t __first = __lane_mask >> __builtin_ctzll(__lane_mask);            \
-    bool __divergent = __gpu_read_first_lane_##__suffix(                       \
-        __lane_mask, __first & (__first + 1));                                 \
-    if (__divergent) {                                                         \
-      return __gpu_shuffle_idx_##__suffix(                                     \
-          __lane_mask, 63 - __builtin_clzll(__lane_mask),                      \
-          __gpu_lane_scan_##__suffix(__lane_mask, __x), __gpu_num_lanes());    \
-    } else {                                                                   \
-      for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) {     \
-        uint32_t __index = __step + __gpu_lane_id();                           \
-        __x += __gpu_shuffle_idx_##__suffix(__lane_mask, __index, __x,         \
-                                            __gpu_num_lanes());                \
-      }                                                                        \
-      return __gpu_read_first_lane_##__suffix(__lane_mask, __x);               \
+  }                                                                            \
+                                                                               \
+  _DEFAULT_FN_ATTRS static __inline__ __type                                   \
+  __gpu_prefix_scan_##__prefix##_##__suffix(uint64_t __lane_mask,              \
+                                            __type __x) {                      \
+    uint64_t __below = __lane_mask & ((UINT64_C(1) << __gpu_lane_id()) - 1);   \
+    for (uint32_t __step = 1; __step < __gpu_num_lanes(); __step *= 2) {       \
+      uint32_t __src = 63 - __builtin_clzg(__below, (int)sizeof(__below) * 8); \
+      __type __result = __gpu_shuffle_idx_##__suffix(__lane_mask, __src, __x,  \
+                                                     __gpu_num_lanes());       \
+      __x = __op(__x, __below ? __result : (__type)__identity);                \
+      for (uint32_t __i = 0; __i < __step; ++__i)                              \
+        __below ^=                                                             \
+            (UINT64_C(1) << (63 - __builtin_clzg(__below, 0))) & __below;      \
     }                                                                          \
+    return __x;                                                                \
+  }                                                                            \
+                                                                               \
+  _DEFAULT_FN_ATTRS static __inline__ __type                                   \
+  __gpu_lane_##__prefix##_##__suffix(uint64_t __lane_mask, __type __x) {       \
+    return __gpu_read_first_lane_##__suffix(                                   \
+        __lane_mask,                                                           \
+        __gpu_suffix_scan_##__prefix##_##__suffix(__lane_mask, __x));          \
   }
-__DO_LANE_SUM(uint32_t, u32); // uint32_t __gpu_lane_sum_u32(m, x)
-__DO_LANE_SUM(uint64_t, u64); // uint64_t __gpu_lane_sum_u64(m, x)
-__DO_LANE_SUM(float, f32);    // float __gpu_lane_sum_f32(m, x)
-__DO_LANE_SUM(double, f64);   // double __gpu_lane_sum_f64(m, x)
-#undef __DO_LANE_SUM
+
+#define __GPU_OP(__x, __y) ((__x) + (__y))
+__DO_LANE_OPS(uint32_t, __GPU_OP, 0, add, u32);
+__DO_LANE_OPS(uint64_t, __GPU_OP, 0, add, u64);
+__DO_LANE_OPS(float, __GPU_OP, 0, add, f32);
+__DO_LANE_OPS(double, __GPU_OP, 0, add, f64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) ((__x) & (__y))
+__DO_LANE_OPS(uint32_t, __GPU_OP, UINT32_MAX, and, u32);
+__DO_LANE_OPS(uint64_t, __GPU_OP, UINT64_MAX, and, u64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) ((__x) | (__y))
+__DO_LANE_OPS(uint32_t, __GPU_OP, 0, or, u32);
+__DO_LANE_OPS(uint64_t, __GPU_OP, 0, or, u64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) ((__x) ^ (__y))
+__DO_LANE_OPS(uint32_t, __GPU_OP, 0, xor, u32);
+__DO_LANE_OPS(uint64_t, __GPU_OP, 0, xor, u64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) ((__x) < (__y) ? (__x) : (__y))
+__DO_LANE_OPS(uint32_t, __GPU_OP, UINT32_MAX, min, u32);
+__DO_LANE_OPS(uint64_t, __GPU_OP, UINT64_MAX, min, u64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) ((__x) > (__y) ? (__x) : (__y))
+__DO_LANE_OPS(uint32_t, __GPU_OP, 0, max, u32);
+__DO_LANE_OPS(uint64_t, __GPU_OP, 0, max, u64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) __builtin_elementwise_minnum((__x), (__y))
+__DO_LANE_OPS(float, __GPU_OP, __builtin_inff(), minnum, f32);
+__DO_LANE_OPS(double, __GPU_OP, __builtin_inf(), minnum, f64);
+#undef __GPU_OP
+
+#define __GPU_OP(__x, __y) __builtin_elementwise_maxnum((__x), (__y))
+__DO_LANE_OPS(float, __GPU_OP, -__builtin_inff(), maxnum, f32);
+__DO_LANE_OPS(double, __GPU_OP, -__builtin_inf(), maxnum, f64);
+#undef __GPU_OP
+
+#undef __DO_LANE_OPS
 
 // Returns a bitmask marking all lanes that have the same value of __x.
+#ifndef __gpu_match_any_u32_impl
 _DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_any_u32_impl(uint64_t __lane_mask, uint32_t __x) {
+__gpu_match_any_u32(uint64_t __lane_mask, uint32_t __x) {
   uint64_t __match_mask = 0;
 
   bool __done = 0;
   for (uint64_t __active_mask = __lane_mask; __active_mask;
        __active_mask = __gpu_ballot(__lane_mask, !__done)) {
     if (!__done) {
-      uint32_t __first = __gpu_read_first_lane_u32(__active_mask, __x);
+      uint32_t __first = __gpu_shuffle_idx_u32(
+          __active_mask, __builtin_ctzg(__active_mask), __x, __gpu_num_lanes());
+      uint64_t __ballot = __gpu_ballot(__active_mask, __first == __x);
       if (__first == __x) {
-        __match_mask = __gpu_lane_mask();
+        __match_mask = __ballot;
         __done = 1;
       }
     }
   }
-  __gpu_sync_lane(__lane_mask);
   return __match_mask;
 }
+#endif
+#undef __gpu_match_any_u32_impl
 
 // Returns a bitmask marking all lanes that have the same value of __x.
+#ifndef __gpu_match_any_u64_impl
 _DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_any_u64_impl(uint64_t __lane_mask, uint64_t __x) {
+__gpu_match_any_u64(uint64_t __lane_mask, uint64_t __x) {
   uint64_t __match_mask = 0;
 
   bool __done = 0;
   for (uint64_t __active_mask = __lane_mask; __active_mask;
        __active_mask = __gpu_ballot(__lane_mask, !__done)) {
     if (!__done) {
-      uint64_t __first = __gpu_read_first_lane_u64(__active_mask, __x);
+      uint64_t __first = __gpu_shuffle_idx_u64(
+          __active_mask, __builtin_ctzg(__active_mask), __x, __gpu_num_lanes());
+      uint64_t __ballot = __gpu_ballot(__active_mask, __first == __x);
       if (__first == __x) {
-        __match_mask = __gpu_lane_mask();
+        __match_mask = __ballot;
         __done = 1;
       }
     }
   }
-  __gpu_sync_lane(__lane_mask);
   return __match_mask;
 }
+#endif
+#undef __gpu_match_any_u64_impl
 
 // Returns the current lane mask if every lane contains __x.
+#ifndef __gpu_match_all_u32_impl
 _DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_all_u32_impl(uint64_t __lane_mask, uint32_t __x) {
-  uint32_t __first = __gpu_read_first_lane_u32(__lane_mask, __x);
+__gpu_match_all_u32(uint64_t __lane_mask, uint32_t __x) {
+  uint32_t __first = __gpu_shuffle_idx_u32(
+      __lane_mask, __builtin_ctzg(__lane_mask), __x, __gpu_num_lanes());
   uint64_t __ballot = __gpu_ballot(__lane_mask, __x == __first);
-  __gpu_sync_lane(__lane_mask);
-  return __ballot == __gpu_lane_mask() ? __gpu_lane_mask() : 0ull;
+  return __ballot == __lane_mask ? __lane_mask : UINT64_C(0);
 }
+#endif
+#undef __gpu_match_all_u32_impl
 
 // Returns the current lane mask if every lane contains __x.
+#ifndef __gpu_match_all_u64_impl
 _DEFAULT_FN_ATTRS static __inline__ uint64_t
-__gpu_match_all_u64_impl(uint64_t __lane_mask, uint64_t __x) {
-  uint64_t __first = __gpu_read_first_lane_u64(__lane_mask, __x);
+__gpu_match_all_u64(uint64_t __lane_mask, uint64_t __x) {
+  uint64_t __first = __gpu_shuffle_idx_u64(
+      __lane_mask, __builtin_ctzg(__lane_mask), __x, __gpu_num_lanes());
   uint64_t __ballot = __gpu_ballot(__lane_mask, __x == __first);
-  __gpu_sync_lane(__lane_mask);
-  return __ballot == __gpu_lane_mask() ? __gpu_lane_mask() : 0ull;
+  return __ballot == __lane_mask ? __lane_mask : UINT64_C(0);
 }
+#endif
+#undef __gpu_match_all_u64_impl
 
 _Pragma("omp end declare variant");
 _Pragma("omp end declare target");

package/lib/clang/23/include/hlsl/hlsl_alias_intrinsics.h

@@ -0,0 +1,222 @@
+//===--- hlsl_alias_intrinsics.h - HLSL alias 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_HLSL_ALIAS_INTRINSICS_H_
+#define _HLSL_HLSL_ALIAS_INTRINSICS_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.
+
+#define _HLSL_BUILTIN_ALIAS(builtin)                                           \
+  __attribute__((clang_builtin_alias(builtin)))
+#define _HLSL_AVAILABILITY(platform, version)                                  \
+  __attribute__((availability(platform, introduced = version)))
+#define _HLSL_AVAILABILITY_STAGE(platform, version, stage)                     \
+  __attribute__((                                                              \
+      availability(platform, introduced = version, environment = stage)))
+
+#ifdef __HLSL_ENABLE_16_BIT
+#define _HLSL_16BIT_AVAILABILITY(platform, version, ...)                       \
+  __attribute__((availability(platform, introduced = version)))
+#define _HLSL_16BIT_AVAILABILITY_STAGE(platform, version, stage)               \
+  __attribute__((                                                              \
+      availability(platform, introduced = version, environment = stage)))
+#define _HLSL_16BIT_AVAILABILITY_SHADERMODEL_DEFAULT()                         \
+  _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+#else
+#define _HLSL_16BIT_AVAILABILITY(platform, version, ...)                       \
+  __VA_OPT__(_HLSL_AVAILABILITY(platform, __VA_ARGS__))
+#define _HLSL_16BIT_AVAILABILITY_STAGE(environment, version, stage)
+#define _HLSL_16BIT_AVAILABILITY_SHADERMODEL_DEFAULT()
+#endif
+
+// Generated by clang-tblgen from HLSLIntrinsics.td (alias intrinsics).
+#include "hlsl_alias_intrinsics_gen.inc"
+
+//===----------------------------------------------------------------------===//
+// mul builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn R mul(X x, Y y)
+/// \brief Multiplies x and y using matrix math.
+/// \param x [in] The first input value. If x is a vector, it is treated as a
+///   row vector.
+/// \param y [in] The second input value. If y is a vector, it is treated as a
+///   column vector.
+///
+/// The inner dimension x-columns and y-rows must be equal. The result has the
+/// dimension x-rows x y-columns. When both x and y are vectors, the result is
+/// a dot product (scalar). Scalar operands are multiplied element-wise.
+///
+/// This function supports 9 overloaded forms:
+///   1. scalar * scalar -> scalar
+///   2. scalar * vector -> vector
+///   3. scalar * matrix -> matrix
+///   4. vector * scalar -> vector
+///   5. vector * vector -> scalar (dot product)
+///   6. vector * matrix -> vector
+///   7. matrix * scalar -> matrix
+///   8. matrix * vector -> vector
+///   9. matrix * matrix -> matrix
+
+// Cases 1, 2, 3, 4, 5, and 7 of mul are defined in hlsl_intrinsics.h as
+// header-only implementations because they are elementwise operations and dot
+// products easily expressed in HLSL.
+
+// Cases 6, 8, and 9 are defined below to alias the mul builtin so that they can
+// be lowered to the llvm.matrix.multiply intrinsic which is not exposed
+// directly to HLSL.
+
+// Case 6: vector * matrix -> vector
+template <int R, int C>
+_HLSL_16BIT_AVAILABILITY_SHADERMODEL_DEFAULT()
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_mul) vector<half, C> mul(vector<half, R>,
+                                                            matrix<half, R, C>);
+
+template <typename T, int R, int C>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_mul)
+vector<T, C> mul(vector<T, R>, matrix<T, R, C>);
+
+// Case 8: matrix * vector -> vector
+template <int R, int C>
+_HLSL_16BIT_AVAILABILITY_SHADERMODEL_DEFAULT()
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_mul) vector<half, R> mul(matrix<half, R, C>,
+                                                            vector<half, C>);
+
+template <typename T, int R, int C>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_mul)
+vector<T, R> mul(matrix<T, R, C>, vector<T, C>);
+
+// Case 9: matrix * matrix -> matrix
+template <int R, int K, int C>
+_HLSL_16BIT_AVAILABILITY_SHADERMODEL_DEFAULT()
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_mul) matrix<half, R, C> mul(
+    matrix<half, R, K>, matrix<half, K, C>);
+
+template <typename T, int R, int K, int C>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_mul)
+matrix<T, R, C> mul(matrix<T, R, K>, matrix<T, K, C>);
+
+//===----------------------------------------------------------------------===//
+// transpose builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn matrix<T, C, R> transpose(matrix<T, R, C> x)
+/// \brief Returns the transpose of the input matrix.
+/// \param x [in] The input matrix.
+
+template <int R, int C>
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_transpose) matrix<half, C, R> transpose(
+    matrix<half, R, C>);
+
+template <typename T, int R, int C>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_transpose)
+matrix<T, C, R> transpose(matrix<T, R, C>);
+
+//===----------------------------------------------------------------------===//
+// select builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T select(bool Cond, T TrueVal, T FalseVal)
+/// \brief ternary operator.
+/// \param Cond The Condition input value.
+/// \param TrueVal The Value returned if Cond is true.
+/// \param FalseVal The Value returned if Cond is false.
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+T select(bool, T, T);
+
+/// \fn vector<T,Sz> select(vector<bool,Sz> Conds, vector<T,Sz> TrueVals,
+///                         vector<T,Sz> FalseVals)
+/// \brief ternary operator for vectors. All vectors must be the same size.
+/// \param Conds The Condition input values.
+/// \param TrueVals The vector values are chosen from when conditions are true.
+/// \param FalseVals The vector values are chosen from when conditions are
+/// false.
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 2> select(vector<bool, 2>, vector<T, 2>, vector<T, 2>);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 3> select(vector<bool, 3>, vector<T, 3>, vector<T, 3>);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 4> select(vector<bool, 4>, vector<T, 4>, vector<T, 4>);
+
+/// \fn vector<T,Sz> select(vector<bool,Sz> Conds, T TrueVal,
+///                         vector<T,Sz> FalseVals)
+/// \brief ternary operator for vectors. All vectors must be the same size.
+/// \param Conds The Condition input values.
+/// \param TrueVal The scalar value to splat from when conditions are true.
+/// \param FalseVals The vector values are chosen from when conditions are
+/// false.
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 2> select(vector<bool, 2>, T, vector<T, 2>);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 3> select(vector<bool, 3>, T, vector<T, 3>);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 4> select(vector<bool, 4>, T, vector<T, 4>);
+
+/// \fn vector<T,Sz> select(vector<bool,Sz> Conds, vector<T,Sz> TrueVals,
+///                         T FalseVal)
+/// \brief ternary operator for vectors. All vectors must be the same size.
+/// \param Conds The Condition input values.
+/// \param TrueVals The vector values are chosen from when conditions are true.
+/// \param FalseVal The scalar value to splat from when conditions are false.
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 2> select(vector<bool, 2>, vector<T, 2>, T);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 3> select(vector<bool, 3>, vector<T, 3>, T);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+vector<T, 4> select(vector<bool, 4>, vector<T, 4>, T);
+
+/// \fn vector<T,Sz> select(vector<bool,Sz> Conds, vector<T,Sz> TrueVals,
+///                         T FalseVal)
+/// \brief ternary operator for vectors. All vectors must be the same size.
+/// \param Conds The Condition input values.
+/// \param TrueVal The scalar value to splat from when conditions are true.
+/// \param FalseVal The scalar value to splat from when conditions are false.
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+__detail::enable_if_t<__detail::is_arithmetic<T>::Value, vector<T, 2>> select(
+    vector<bool, 2>, T, T);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+__detail::enable_if_t<__detail::is_arithmetic<T>::Value, vector<T, 3>> select(
+    vector<bool, 3>, T, T);
+
+template <typename T>
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_select)
+__detail::enable_if_t<__detail::is_arithmetic<T>::Value, vector<T, 4>> select(
+    vector<bool, 4>, T, T);
+
+} // namespace hlsl
+#endif //_HLSL_HLSL_ALIAS_INTRINSICS_H_