0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-11-17 23:40:57 +01:00

ircd::simd: Split lane_cast into header.

This commit is contained in:
Jason Volk 2020-06-28 16:37:00 -07:00
parent 7254d3c945
commit 86aa96fa91
2 changed files with 42 additions and 23 deletions

View file

@ -0,0 +1,40 @@
// The Construct
//
// Copyright (C) The Construct Developers, Authors & Contributors
// Copyright (C) 2016-2020 Jason Volk <jason@zemos.net>
//
// Permission to use, copy, modify, and/or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright notice and this permission notice is present in all copies. The
// full license for this software is available in the LICENSE file.
#pragma once
#define HAVE_IRCD_SIMD_LANE_CAST_H
namespace ircd::simd
{
template<class R,
class T>
R lane_cast(const T &);
}
/// Convert each lane from a smaller type to a larger type
template<class R,
class T>
inline R
ircd::simd::lane_cast(const T &in)
{
static_assert
(
lanes<R>() == lanes<T>(), "Types must have matching number of lanes."
);
if constexpr(__has_builtin(__builtin_convertvector))
return __builtin_convertvector(in, R);
R ret;
for(size_t i(0); i < lanes<R>(); ++i)
ret[i] = in[i];
return ret;
}

View file

@ -13,6 +13,7 @@
#include "type.h"
#include "traits.h"
#include "lane_cast.h"
#include "print.h"
namespace ircd::simd
@ -24,35 +25,13 @@ namespace ircd::simd
template<int bits> u128x1 shr(const u128x1 &a) noexcept;
template<int bits> u256x1 shl(const u256x1 &a) noexcept;
template<int bits> u256x1 shr(const u256x1 &a) noexcept;
template<class R, class T> R lane_cast(const T &);
}
namespace ircd
{
using simd::shl;
using simd::shr;
}
/// Convert each lane from a smaller type to a larger type
template<class R,
class T>
inline R
ircd::simd::lane_cast(const T &in)
{
static_assert
(
lanes<R>() == lanes<T>(), "Types must have matching number of lanes."
);
if constexpr(__has_builtin(__builtin_convertvector))
return __builtin_convertvector(in, R);
R ret;
for(size_t i(0); i < lanes<R>(); ++i)
ret[i] = in[i];
return ret;
using simd::lane_cast;
}
#ifdef HAVE_X86INTRIN_H