divisible_by_cardinal

Detailed Description

States that a range holds a whole number of SIMD registers.

Variables
constexpr auto	eve::algo::divisible_by_width = ::rbr::flag( divisible_by_width_tag{} )
	an trait to tell that the input data is strictly divisible by width.

Variable Documentation

divisible_by_width

auto eve::algo::divisible_by_width = ::rbr::flag( divisible_by_width_tag{} )

inlineconstexpr

an trait to tell that the input data is strictly divisible by width.

namespace eve::algo
{
  inline constexpr auto divisible_by_cardinal = eve_implementation_defined;
}

States that the range holds a whole number of SIMD registers, so the loop needs no tail handling. Alignment takes precedence: a loop that aligns its accesses reads a partial first register, and the tail handling returns. The trait therefore takes effect on an unaligned pointer, or with eve::algo::no_aligning.

Note

this trait is deduced automatically if both begin and end of the range are aligned_ptr with alignment >= width. If you are using this, consider aligning your data too.

std::vector<float> data(1024);
auto positive = [](auto x) { return x > 0.f; };
 
// 1024 is a multiple of every cardinal: no tail, the accesses left unaligned.
auto found = find_if[divisible_by_cardinal][no_aligning](data, positive);

See also

no_aligning