[libcxx-commits] [libcxx] [libc++] Use new builtins in __algorithm/simd_utils.h (PR #178158)

[Nikolas Klauser via libcxx-commits]

https://github.com/philnik777 created https://github.com/llvm/llvm-project/pull/178158

This patch makes use of new builtins provided by clang inside `__algorithm/simd_utils.h`. Specifically, `__builtin_masked_load` is now used to implement `__load_vector` on pointers and `__count{l,r}_zero` are now called directly on vectors. The first builtin generates better code than the loop we currently have, resulting in Clang being able to better estimate the cost of the code, which can result in better inlining decisions. The second change generalizes `__find_first_set` to work on vectors of arbitrary length and improves compile times a bit.


>From 79bf0cf3a5b3f1ebfd32163c0f9cc3979ff92c2a Mon Sep 17 00:00:00 2001
From: Nikolas Klauser <nikolasklauser at berlin.de>
Date: Tue, 27 Jan 2026 11:52:25 +0100
Subject: [PATCH] [libc++] Use new builtins in __algorithm/simd_utils.h

---
 libcxx/include/__algorithm/simd_utils.h | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/libcxx/include/__algorithm/simd_utils.h b/libcxx/include/__algorithm/simd_utils.h
index f73c9ea4b6ea7..5925189bcc6e0 100644
--- a/libcxx/include/__algorithm/simd_utils.h
+++ b/libcxx/include/__algorithm/simd_utils.h
@@ -114,6 +114,13 @@ template <class _VecT, class _Iter>
   }(make_index_sequence<__simd_vector_size_v<_VecT>>{});
 }
 
+#  if __has_builtin(__builtin_masked_load)
+template <class _VecT, class _Tp>
+[[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI _VecT __load_vector(_Tp* __ptr) noexcept {
+  return __builtin_masked_load(__simd_vector<bool, __simd_vector_size_v<_VecT>>(true), __ptr);
+}
+#  endif
+
 // Load the first _Np elements, zero the rest
 _LIBCPP_DIAGNOSTIC_PUSH
 _LIBCPP_CLANG_DIAGNOSTIC_IGNORED("-Wpsabi")
@@ -153,16 +160,22 @@ template <class _Tp, size_t _Np>
 template <class _Tp, size_t _Np>
 [[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI size_t __find_first_set(__simd_vector<_Tp, _Np> __vec) noexcept {
   using __mask_vec = __simd_vector<bool, _Np>;
-
+#  if defined(_LIBCPP_CLANG_VER) && _LIBCPP_CLANG_VER >= 2201
+#    if defined(_LIBCPP_BIG_ENDIAN)
+  return std::__countl_zero(__builtin_convertvector(__vec, __mask_vec));
+#    else
+  return std::__countr_zero(__builtin_convertvector(__vec, __mask_vec));
+#    endif
+#  else
   // This has MSan disabled du to https://llvm.org/PR85876
   auto __impl = [&]<class _MaskT>(_MaskT) _LIBCPP_NO_SANITIZE("memory") noexcept {
-#  if defined(_LIBCPP_BIG_ENDIAN)
+#    if defined(_LIBCPP_BIG_ENDIAN)
     return std::min<size_t>(
         _Np, std::__countl_zero(__builtin_bit_cast(_MaskT, __builtin_convertvector(__vec, __mask_vec))));
-#  else
+#    else
     return std::min<size_t>(
         _Np, std::__countr_zero(__builtin_bit_cast(_MaskT, __builtin_convertvector(__vec, __mask_vec))));
-#  endif
+#    endif
   };
 
   if constexpr (sizeof(__mask_vec) == sizeof(uint8_t)) {
@@ -177,6 +190,7 @@ template <class _Tp, size_t _Np>
     static_assert(sizeof(__mask_vec) == 0, "unexpected required size for mask integer type");
     return 0;
   }
+#  endif
 }
 
 template <class _Tp, size_t _Np>