1 | /* Copyright (C) 2013-2014 Povilas Kanapickas <povilas@radix.lt> |
2 | |
3 | Distributed under the Boost Software License, Version 1.0. |
4 | (See accompanying file LICENSE_1_0.txt or copy at |
5 | http://www.boost.org/LICENSE_1_0.txt) |
6 | */ |
7 | |
8 | #ifndef LIBSIMDPP_SIMDPP_CORE_LOAD_PACKED4_H |
9 | #define LIBSIMDPP_SIMDPP_CORE_LOAD_PACKED4_H |
10 | |
11 | #ifndef LIBSIMDPP_SIMD_H |
12 | #error "This file must be included through simd.h" |
13 | #endif |
14 | |
15 | #include <simdpp/types.h> |
16 | #include <simdpp/detail/insn/load_packed4.h> |
17 | |
18 | namespace simdpp { |
19 | namespace SIMDPP_ARCH_NAMESPACE { |
20 | |
21 | |
22 | /** Loads values packed in quartets, de-interleaves them and stores the result |
23 | into four vectors. |
24 | |
25 | @code |
26 | a = [ *(p), *(p+4), *(p+8), ... , *(p+M*4-4) ] |
27 | b = [ *(p+1), *(p+5), *(p+9), ... , *(p+M*4-3) ] |
28 | c = [ *(p+2), *(p+6), *(p+10), ... , *(p+M*4-2) ] |
29 | d = [ *(p+3), *(p+7), *(p+11), ... , *(p+M*4-1) ] |
30 | @endcode |
31 | |
32 | Here M is the number of elements in the vector |
33 | |
34 | @a p must be aligned to the vector size in bytes |
35 | */ |
36 | template<unsigned N, class V, class T> SIMDPP_INL |
37 | void load_packed4(any_vec<N,V>& a, any_vec<N,V>& b, |
38 | any_vec<N,V>& c, any_vec<N,V>& d, |
39 | const T* p) |
40 | { |
41 | static_assert(!is_mask<V>::value, "Mask types can not be loaded" ); |
42 | typename detail::get_expr_nosign<V>::type ra, rb, rc, rd; |
43 | detail::insn::i_load_packed4(ra, rb, rc, rd, reinterpret_cast<const char*>(p)); |
44 | a.wrapped() = ra; |
45 | b.wrapped() = rb; |
46 | c.wrapped() = rc; |
47 | d.wrapped() = rd; |
48 | } |
49 | |
50 | |
51 | } // namespace SIMDPP_ARCH_NAMESPACE |
52 | } // namespace simdpp |
53 | |
54 | #endif |
55 | |
56 | |