CMT_INLINE vec<T, N> operator()(cinput_t, size_t index, vec_t<T, N>) const
{
static_assert(is_poweroftwo(N), "N must be a power of two");
vec<T, N> result;
static_cast<const expression_vtable<T, N>*>(vtable)->get(instance, index, result);
return result;
}
#include "function.hpp"
#include <algorithm>
#include <utility>
namespace kfr
{
namespace internal
{
template <typename T, typename ReduceFn>
CMT_INLINE T horizontal_impl(const vec<T, 1>& value, ReduceFn&&)
{
return T(value[0]);
}
template <typename T, size_t N, typename ReduceFn, KFR_ENABLE_IF(N > 1 && is_poweroftwo(N))>
CMT_INLINE T horizontal_impl(const vec<T, N>& value, ReduceFn&& reduce)
{
return horizontal_impl(reduce(low(value), high(value)), std::forward<ReduceFn>(reduce));
}
template <typename T, size_t N, typename ReduceFn, KFR_ENABLE_IF(N > 1 && !is_poweroftwo(N))>
CMT_INLINE T horizontal_impl(const vec<T, N>& value, ReduceFn&& reduce)
{
const T initial = reduce(initialvalue<T>());
return horizontal_impl(widen<next_poweroftwo(N)>(value, initial), std::forward<ReduceFn>(reduce));
}
}
template <typename T, size_t N, typename ReduceFn>
CMT_INLINE T horizontal(const vec<T, N>& value, ReduceFn&& reduce)
{
template <size_t N, typename T>
KFR_INTRINSIC simd<T, N> simd_read(const T* src) CMT_NOEXCEPT
{
return reinterpret_cast<typename simd_storage<T, N, false>::const_pointer>(src)->value;
}
template <size_t N, bool A = false, typename T, KFR_ENABLE_IF(is_poweroftwo(N))>
KFR_INTRINSIC vec<T, N> read(cunaligned_t, csize_t<N>, const T* src) CMT_NOEXCEPT
{
// Clang requires a separate function returning vector (simd).
// Direct returning vec causes aligned read instruction
return simd_read<N>(src);
}
template <size_t N, bool A = false, typename T, KFR_ENABLE_IF(!is_poweroftwo(N)), typename = void>
KFR_INTRINSIC vec<T, N> read(cunaligned_t, csize_t<N>, const T* src) CMT_NOEXCEPT
{
constexpr size_t first = prev_poweroftwo(N);
return concat(read(cunaligned, csize<first>, src), read(cunaligned, csize<N - first>, src + first));
}
template <bool A = false, size_t N, typename T, KFR_ENABLE_IF(is_poweroftwo(N))>
KFR_INTRINSIC void write(cunaligned_t, T* dest, const vec<T, N>& x) CMT_NOEXCEPT
{
reinterpret_cast<typename simd_storage<T, N, A>::pointer>(dest)->value = x.v;
}
template <bool A = false, size_t N, typename T, KFR_ENABLE_IF(!is_poweroftwo(N)),
size_t Nlow = prev_poweroftwo(N - 1)>
KFR_INTRINSIC void write(cunaligned_t, T* dest, const vec<T, N>& x) CMT_NOEXCEPT
