Help the compiler vectorize ranges::iota (#4647)

Co-authored-by: Stephan T. Lavavej <[email protected]>

benchmarks/src/iota.cpp

@@ -7,14 +7,24 @@
 #include <numeric>
 #include <vector>
 
-template <class T>
+enum class Alg {
+ Std,
+ Rng,
+};
+
+template <class T, Alg Algorithm>
 void bm(benchmark::State& state) {
  const auto size = static_cast<std::size_t>(state.range(0));
 
  std::vector<T> a(size);
 
  for (auto _ : state) {
- std::iota(a.begin(), a.end(), T{22});
+ if constexpr (Algorithm == Alg::Std) {
+ std::iota(a.begin(), a.end(), T{22});
+ } else if constexpr (Algorithm == Alg::Rng) {
+ std::ranges::iota(a, T{22});
+ }
+
  benchmark::DoNotOptimize(a);
  }
 }
@@ -23,7 +33,10 @@ void common_args(auto bm) {
  bm->Arg(7)->Arg(18)->Arg(43)->Arg(131)->Arg(315)->Arg(1212);
 }
 
-BENCHMARK(bm<std::uint32_t>)->Apply(common_args);
-BENCHMARK(bm<std::uint64_t>)->Apply(common_args);
+BENCHMARK(bm<std::uint32_t, Alg::Std>)->Apply(common_args);
+BENCHMARK(bm<std::uint64_t, Alg::Std>)->Apply(common_args);
+
+BENCHMARK(bm<std::uint32_t, Alg::Rng>)->Apply(common_args);
+BENCHMARK(bm<std::uint64_t, Alg::Rng>)->Apply(common_args);
 
 BENCHMARK_MAIN();

stl/inc/numeric

@@ -567,6 +567,24 @@ namespace ranges {
  _STL_INTERNAL_STATIC_ASSERT(weakly_incrementable<_Ty>);
  _STL_INTERNAL_STATIC_ASSERT(indirectly_writable<_It, const _Ty&>);
 
+ if constexpr (_Iterator_is_contiguous<_It> && sized_sentinel_for<_Se, _It> && is_integral_v<_Ty>
+ && sizeof(_Ty) >= 4) {
+ // TRANSITION, DevCom-10593477: help the compiler vectorize
+ const auto _Ptr = _To_address(_First);
+ const auto _Size = static_cast<size_t>(_Last - _First);
+
+ if (_STD _In_range<_Ty>(_Size)) {
+ const auto _Size_typed = static_cast<_Ty>(_Size);
+ for (_Ty _Ix = 0; _Ix != _Size_typed; ++_Ix) {
+ const _Ty _Const_val = _Val + _Ix;
+ _Ptr[_Ix] = _Const_val;
+ }
+
+ _Val += _Size_typed;
+ return _First + _Size;
+ }
+ }
+
  const _Ty& _Const_val = _Val;
  for (; _First != _Last; ++_First, (void) ++_Val) {
  *_First = _Const_val;