[HLSL] [DirectX] Invert the result of firstbithigh

clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h

@@ -1073,78 +1073,6 @@ float3 f16tof32(uint3);
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_f16tof32)
 float4 f16tof32(uint4);
 
-//===----------------------------------------------------------------------===//
-// firstbithigh builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn T firstbithigh(T Val)
-/// \brief Returns the location of the first set bit starting from the highest
-/// order bit and working downward, per component.
-/// \param Val the input value.
-
-#ifdef __HLSL_ENABLE_16_BIT
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(int16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(int16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(int16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(int16_t4);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(uint16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(uint16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(uint16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(uint16_t4);
-#endif
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(int);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(int2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(int3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(int4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(uint);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(uint2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(uint3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(uint4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(int64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(int64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(int64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(int64_t4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(uint64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(uint64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(uint64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(uint64_t4);
-
 //===----------------------------------------------------------------------===//
 // firstbitlow builtins
 //===----------------------------------------------------------------------===//

clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h

@@ -148,6 +148,18 @@ template <typename T> constexpr T ldexp_impl(T X, T Exp) {
   return exp2(Exp) * X;
 }
 
+template <typename K, typename T, int BitWidth>
+constexpr K firstbithigh_impl(T X) {
+  K FBH = __builtin_hlsl_elementwise_firstbithigh(X);
+#if defined(__DIRECTX__)
+  // The firstbithigh DXIL ops count bits from the wrong side, so we need to
+  // invert it for DirectX.
+  K Inversion = (BitWidth - 1) - FBH;
+  FBH = select(FBH == -1, FBH, Inversion);
+#endif
+  return FBH;
+}
+
 } // namespace __detail
 } // namespace hlsl
 

clang/lib/Headers/hlsl/hlsl_intrinsics.h

@@ -261,6 +261,67 @@ faceforward(__detail::HLSL_FIXED_VECTOR<float, L> N,
   return __detail::faceforward_impl(N, I, Ng);
 }
 
+//===----------------------------------------------------------------------===//
+// firstbithigh builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T firstbithigh(T Val)
+/// \brief Returns the location of the first set bit starting from the lowest
+/// order bit and working upward, per component.
+/// \param Val the input value.
+
+#ifdef __HLSL_ENABLE_16_BIT
+
+template <typename T>
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+const inline __detail::enable_if_t<__detail::is_same<int16_t, T>::value ||
+                                       __detail::is_same<uint16_t, T>::value,
+                                   uint> firstbithigh(T X) {
+  return __detail::firstbithigh_impl<uint, T, 16>(X);
+}
+
+template <typename T, int N>
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+const
+    inline __detail::enable_if_t<__detail::is_same<int16_t, T>::value ||
+                                     __detail::is_same<uint16_t, T>::value,
+                                 vector<uint, N>> firstbithigh(vector<T, N> X) {
+  return __detail::firstbithigh_impl<vector<uint, N>, vector<T, N>, 16>(X);
+}
+
+#endif
+
+template <typename T>
+const inline __detail::enable_if_t<
+    __detail::is_same<int, T>::value || __detail::is_same<uint, T>::value, uint>
+firstbithigh(T X) {
+  return __detail::firstbithigh_impl<uint, T, 32>(X);
+}
+
+template <typename T, int N>
+const inline __detail::enable_if_t<__detail::is_same<int, T>::value ||
+                                       __detail::is_same<uint, T>::value,
+                                   vector<uint, N>>
+firstbithigh(vector<T, N> X) {
+  return __detail::firstbithigh_impl<vector<uint, N>, vector<T, N>, 32>(X);
+}
+
+template <typename T>
+const inline __detail::enable_if_t<__detail::is_same<int64_t, T>::value ||
+                                       __detail::is_same<uint64_t, T>::value,
+                                   uint>
+firstbithigh(T X) {
+  return __detail::firstbithigh_impl<uint, T, 64>(X);
+}
+
+template <typename T, int N>
+const inline __detail::enable_if_t<__detail::is_same<int64_t, T>::value ||
+                                       __detail::is_same<uint64_t, T>::value,
+                                   vector<uint, N>>
+firstbithigh(vector<T, N> X) {
+  return __detail::firstbithigh_impl<vector<uint, N>, vector<T, N>, 64>(X);
+}
+
 //===----------------------------------------------------------------------===//
 // fmod builtins
 //===----------------------------------------------------------------------===//