| 1 | /* Copyright (C) 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_SHUFFLE2x2_H |
| 9 | #define LIBSIMDPP_SIMDPP_CORE_SHUFFLE2x2_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/shuffle4x2.h> |
| 17 | |
| 18 | namespace simdpp { |
| 19 | namespace SIMDPP_ARCH_NAMESPACE { |
| 20 | |
| 21 | /** Selects 32-bit values from two vectors. |
| 22 | The selector values must be in range [0; 3]. |
| 23 | |
| 24 | @code |
| 25 | For each 64-bit segment: |
| 26 | ab = [ a..b ] |
| 27 | r0 = ab[s0] |
| 28 | r1 = ab[s1] |
| 29 | @endcode |
| 30 | */ |
| 31 | template<unsigned s0, unsigned s1, unsigned N, |
| 32 | class V1, class V2> SIMDPP_INL |
| 33 | typename detail::get_expr2_nomask<V1, V2>::empty |
| 34 | shuffle2x2(const any_vec32<N,V1>& a, const any_vec32<N,V2>& b) |
| 35 | { |
| 36 | #if SIMDPP_USE_NULL || SIMDPP_USE_SSE2 || SIMDPP_USE_NEON |
| 37 | static_assert(s0 < 4 && s1 < 4, "Selector out of range" ); |
| 38 | typename detail::get_expr2_nomask_nosign<V1,V2,void>::type a0 = a.wrapped().eval(), |
| 39 | b0 = b.wrapped().eval(); |
| 40 | return detail::insn::i_shuffle2x2<s0,s1>(a0, b0); |
| 41 | #else |
| 42 | return SIMDPP_NOT_IMPLEMENTED_TEMPLATE2(V1, a, b); |
| 43 | #endif |
| 44 | } |
| 45 | |
| 46 | /** Selects 64-bit values from two vectors. |
| 47 | The selector values must be in range [0; 3]. |
| 48 | |
| 49 | @code |
| 50 | For each 128-bit segment: |
| 51 | ab = [ a..b ] |
| 52 | r0 = ab[s0] |
| 53 | r1 = ab[s1] |
| 54 | @endcode |
| 55 | */ |
| 56 | template<unsigned s0, unsigned s1, unsigned N, |
| 57 | class V1, class V2> SIMDPP_INL |
| 58 | typename detail::get_expr2_nomask<V1, V2>::empty |
| 59 | shuffle2x2(const any_vec64<N,V1>& a, const any_vec64<N,V2>& b) |
| 60 | { |
| 61 | #if SIMDPP_USE_NULL || SIMDPP_USE_SSE2 || SIMDPP_USE_NEON |
| 62 | static_assert(s0 < 4 && s1 < 4, "Selector out of range" ); |
| 63 | typename detail::get_expr2_nomask_nosign<V1,V2,void>::type a0 = a.wrapped().eval(), |
| 64 | b0 = b.wrapped().eval(); |
| 65 | return detail::insn::i_shuffle2x2<s0,s1>(a0, b0); |
| 66 | #else |
| 67 | return SIMDPP_NOT_IMPLEMENTED_TEMPLATE2(V1, a, b); |
| 68 | #endif |
| 69 | } |
| 70 | |
| 71 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 72 | } // namespace simdpp |
| 73 | |
| 74 | #endif |
| 75 | |
| 76 | |