1 | /* Copyright (C) 2013-2017 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_DETAIL_INSN_CONV_SHRINK_TO_INT32_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_CONV_SHRINK_TO_INT32_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/core/permute4.h> |
17 | #include <simdpp/core/unzip_lo.h> |
18 | #include <simdpp/core/zip_lo.h> |
19 | |
20 | namespace simdpp { |
21 | namespace SIMDPP_ARCH_NAMESPACE { |
22 | namespace detail { |
23 | namespace insn { |
24 | |
25 | |
26 | SIMDPP_INL uint32<4> i_to_uint32(const uint64<4>& a) |
27 | { |
28 | #if SIMDPP_USE_NULL |
29 | uint32<4> r; |
30 | for (unsigned i = 0; i < 4; i++) { |
31 | r.el(i) = uint32_t(a.vec(i/2).el(i%2)); |
32 | } |
33 | return r; |
34 | #elif SIMDPP_USE_AVX512VL |
35 | return _mm256_cvtepi64_epi32(a.native()); |
36 | #elif SIMDPP_USE_AVX2 |
37 | uint64<4> a64; |
38 | a64 = permute4<0,2,0,2>((uint32<8>) a); |
39 | a64 = permute4<0,2,0,2>(a64); |
40 | return _mm256_castsi256_si128(a64.native()); |
41 | #elif SIMDPP_USE_NEON64 |
42 | uint32x2_t low = vmovn_u64(a.vec(0).native()); |
43 | return vmovn_high_u64(low, a.vec(1).native()); |
44 | #elif SIMDPP_USE_NEON |
45 | uint32x2_t low = vmovn_u64(a.vec(0).native()); |
46 | uint32x2_t high = vmovn_u64(a.vec(1).native()); |
47 | return vcombine_u32(low, high); |
48 | #elif SIMDPP_USE_VSX_207 |
49 | return vec_pack(a.vec(0).native(), a.vec(1).native()); |
50 | #elif SIMDPP_USE_ALTIVEC && SIMDPP_BIG_ENDIAN |
51 | uint32<4> r1, r2; |
52 | r1 = a.vec(0); |
53 | r2 = a.vec(1); |
54 | return unzip4_hi(r1, r2); |
55 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_MSA || (SIMDPP_USE_ALTIVEC && SIMDPP_LITTLE_ENDIAN) |
56 | uint32<4> r1, r2; |
57 | r1 = a.vec(0); |
58 | r2 = a.vec(1); |
59 | return unzip4_lo(r1, r2); |
60 | #endif |
61 | } |
62 | |
63 | #if SIMDPP_USE_AVX2 |
64 | SIMDPP_INL uint32<8> i_to_uint32(const uint64<8>& a) |
65 | { |
66 | #if SIMDPP_USE_AVX512F |
67 | return _mm512_cvtepi64_epi32(a.native()); |
68 | #else |
69 | uint64<4> b0, b1; |
70 | b0 = permute4<0,2,0,2>((uint32<8>) a.vec(0)); |
71 | b1 = permute4<0,2,0,2>((uint32<8>) a.vec(1)); |
72 | b0 = zip2_lo(b0, b1); |
73 | b0 = permute4<0,2,1,3>(b0); |
74 | return (uint32<8>) b0; |
75 | #endif |
76 | } |
77 | #endif |
78 | |
79 | #if SIMDPP_USE_AVX512F |
80 | SIMDPP_INL uint32<16> i_to_uint32(const uint64<16>& a) |
81 | { |
82 | uint32<8> r1 = _mm512_cvtepi64_epi32(a.vec(0).native()); |
83 | uint32<8> r2 = _mm512_cvtepi64_epi32(a.vec(1).native()); |
84 | return combine(r1, r2); |
85 | } |
86 | #endif |
87 | |
88 | template<unsigned N> SIMDPP_INL |
89 | uint32<N> i_to_uint32(const uint64<N>& a) |
90 | { |
91 | SIMDPP_VEC_ARRAY_IMPL_CONV_EXTRACT(uint32<N>, i_to_uint32, a) |
92 | } |
93 | |
94 | |
95 | } // namespace insn |
96 | } // namespace detail |
97 | } // namespace SIMDPP_ARCH_NAMESPACE |
98 | } // namespace simdpp |
99 | |
100 | #endif |
101 | |
102 | |
103 | |