| 1 | /* Copyright (C) 2012 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_NEON_MEMORY_STORE_H |
| 9 | #define LIBSIMDPP_NEON_MEMORY_STORE_H |
| 10 | #if SIMDPP_USE_NEON |
| 11 | |
| 12 | #ifndef LIBSIMDPP_SIMD_H |
| 13 | #error "This file must be included through simd.h" |
| 14 | #endif |
| 15 | |
| 16 | #include <simdpp/types.h> |
| 17 | |
| 18 | namespace simdpp { |
| 19 | namespace SIMDPP_ARCH_NAMESPACE { |
| 20 | namespace detail { |
| 21 | namespace neon { |
| 22 | |
| 23 | /** Stores @a n elements of a 128-bit vector starting from the @a pos-th one. |
| 24 | @a p must be aligned to 16 bytes. |
| 25 | |
| 26 | @a n must be a power of 2. @a pos must be a multiple of @a n. The behavior is |
| 27 | undefined if @a pos+n exceeds the number of elements in the specified vector. |
| 28 | */ |
| 29 | template<unsigned pos, unsigned n> SIMDPP_INL |
| 30 | void store_lane(char* p, const uint8x16& a) |
| 31 | { |
| 32 | static_assert(n == 1 || n == 2 || n == 4 || n == 8, "Size not supported" ); |
| 33 | static_assert(pos%n == 0, "pos must be a multiple of n" ); |
| 34 | static_assert(pos+n <= 16, "Index out of bounds" ); |
| 35 | switch (n) { |
| 36 | case 1: |
| 37 | vst1q_lane_u8(reinterpret_cast<uint8_t*>(p), a.native(), pos); |
| 38 | return; |
| 39 | case 2: |
| 40 | vst1q_lane_u16(reinterpret_cast<uint16_t*>(p), vreinterpretq_u16_u8(a.native()), pos/2); |
| 41 | return; |
| 42 | case 4: |
| 43 | vst1q_lane_u32(reinterpret_cast<uint32_t*>(p), vreinterpretq_u32_u8(a.native()), pos/4); |
| 44 | return; |
| 45 | case 8: |
| 46 | vst1q_lane_u64(reinterpret_cast<uint64_t*>(p), vreinterpretq_u64_u8(a.native()), pos/8); |
| 47 | return; |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | template<unsigned pos, unsigned n> SIMDPP_INL |
| 52 | void store_lane(char* p, const uint16x8& a) |
| 53 | { |
| 54 | static_assert(n == 1 || n == 2 || n == 4, "Size not supported" ); |
| 55 | static_assert(pos%n == 0, "pos must be a multiple of n" ); |
| 56 | static_assert(pos+n <= 8, "Index out of bounds" ); |
| 57 | switch (n) { |
| 58 | case 1: |
| 59 | vst1q_lane_u16(reinterpret_cast<uint16_t*>(p), a.native(), pos); |
| 60 | return; |
| 61 | case 2: |
| 62 | vst1q_lane_u32(reinterpret_cast<uint32_t*>(p), |
| 63 | vreinterpretq_u32_u16(a.native()), pos/2); |
| 64 | return; |
| 65 | case 4: |
| 66 | vst1q_lane_u64(reinterpret_cast<uint64_t*>(p), |
| 67 | vreinterpretq_u64_u16(a.native()), pos/4); |
| 68 | return; |
| 69 | } |
| 70 | } |
| 71 | |
| 72 | template<unsigned pos, unsigned n> SIMDPP_INL |
| 73 | void store_lane(char* p, const uint32x4& a) |
| 74 | { |
| 75 | static_assert(n == 1 || n == 2, "Size not supported" ); |
| 76 | static_assert(pos%n == 0, "pos must be a multiple of n" ); |
| 77 | static_assert(pos+n <= 4, "Index out of bounds" ); |
| 78 | switch (n) { |
| 79 | case 1: |
| 80 | vst1q_lane_u32(reinterpret_cast<uint32_t*>(p), a.native(), pos); |
| 81 | return; |
| 82 | case 2: |
| 83 | vst1q_lane_u64(reinterpret_cast<uint64_t*>(p), |
| 84 | vreinterpretq_u64_u32(a.native()), pos/2); |
| 85 | return; |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | template<unsigned pos, unsigned n> SIMDPP_INL |
| 90 | void store_lane(char* p, const uint64x2& a) |
| 91 | { |
| 92 | static_assert(n == 1, "Size not supported" ); |
| 93 | static_assert(pos%n == 0, "pos must be a multiple of n" ); |
| 94 | static_assert(pos+n <= 2, "Index out of bounds" ); |
| 95 | vst1q_lane_u64(reinterpret_cast<uint64_t*>(p), a.native(), pos); |
| 96 | } |
| 97 | |
| 98 | template<unsigned pos, unsigned n> SIMDPP_INL |
| 99 | void store_lane(char* p, const float32x4& a) |
| 100 | { |
| 101 | store_lane<pos,n>(p, uint32x4(a)); |
| 102 | } |
| 103 | |
| 104 | } // namespace neon |
| 105 | } // namespace detail |
| 106 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 107 | } // namespace simdpp |
| 108 | |
| 109 | #endif |
| 110 | #endif |
| 111 | |