| 1 |
// Generic simd conversions -*- C++ -*- |
| 2 |
|
| 3 |
// Copyright (C) 2020-2021 Free Software Foundation, Inc. |
| 4 |
// |
| 5 |
// This file is part of the GNU ISO C++ Library. This library is free |
| 6 |
// software; you can redistribute it and/or modify it under the |
| 7 |
// terms of the GNU General Public License as published by the |
| 8 |
// Free Software Foundation; either version 3, or (at your option) |
| 9 |
// any later version. |
| 10 |
|
| 11 |
// This library is distributed in the hope that it will be useful, |
| 12 |
// but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 |
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 |
// GNU General Public License for more details. |
| 15 |
|
| 16 |
// Under Section 7 of GPL version 3, you are granted additional |
| 17 |
// permissions described in the GCC Runtime Library Exception, version |
| 18 |
// 3.1, as published by the Free Software Foundation. |
| 19 |
|
| 20 |
// You should have received a copy of the GNU General Public License and |
| 21 |
// a copy of the GCC Runtime Library Exception along with this program; |
| 22 |
// see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 23 |
// <http://www.gnu.org/licenses/>. |
| 24 |
|
| 25 |
#ifndef _GLIBCXX_EXPERIMENTAL_SIMD_CONVERTER_H_ |
| 26 |
#define _GLIBCXX_EXPERIMENTAL_SIMD_CONVERTER_H_ |
| 27 |
|
| 28 |
#if __cplusplus >= 201703L |
| 29 |
|
| 30 |
_GLIBCXX_SIMD_BEGIN_NAMESPACE |
| 31 |
// _SimdConverter scalar -> scalar {{{ |
| 32 |
template <typename _From, typename _To> |
| 33 |
struct _SimdConverter<_From, simd_abi::scalar, _To, simd_abi::scalar, |
| 34 |
enable_if_t<!is_same_v<_From, _To>>> |
| 35 |
{ |
| 36 |
_GLIBCXX_SIMD_INTRINSIC constexpr _To operator()(_From __a) const noexcept |
| 37 |
{ return static_cast<_To>(__a); } |
| 38 |
}; |
| 39 |
|
| 40 |
// }}} |
| 41 |
// _SimdConverter scalar -> "native" {{{ |
| 42 |
template <typename _From, typename _To, typename _Abi> |
| 43 |
struct _SimdConverter<_From, simd_abi::scalar, _To, _Abi, |
| 44 |
enable_if_t<!is_same_v<_Abi, simd_abi::scalar>>> |
| 45 |
{ |
| 46 |
using _Ret = typename _Abi::template __traits<_To>::_SimdMember; |
| 47 |
|
| 48 |
template <typename... _More> |
| 49 |
_GLIBCXX_SIMD_INTRINSIC constexpr _Ret |
| 50 |
operator()(_From __a, _More... __more) const noexcept |
| 51 |
{ |
| 52 |
static_assert(sizeof...(_More) + 1 == _Abi::template _S_size<_To>); |
| 53 |
static_assert(conjunction_v<is_same<_From, _More>...>); |
| 54 |
return __make_vector<_To>(__a, __more...); |
| 55 |
} |
| 56 |
}; |
| 57 |
|
| 58 |
// }}} |
| 59 |
// _SimdConverter "native 1" -> "native 2" {{{ |
| 60 |
template <typename _From, typename _To, typename _AFrom, typename _ATo> |
| 61 |
struct _SimdConverter< |
| 62 |
_From, _AFrom, _To, _ATo, |
| 63 |
enable_if_t<!disjunction_v< |
| 64 |
__is_fixed_size_abi<_AFrom>, __is_fixed_size_abi<_ATo>, |
| 65 |
is_same<_AFrom, simd_abi::scalar>, is_same<_ATo, simd_abi::scalar>, |
| 66 |
conjunction<is_same<_From, _To>, is_same<_AFrom, _ATo>>>>> |
| 67 |
{ |
| 68 |
using _Arg = typename _AFrom::template __traits<_From>::_SimdMember; |
| 69 |
using _Ret = typename _ATo::template __traits<_To>::_SimdMember; |
| 70 |
using _V = __vector_type_t<_To, simd_size_v<_To, _ATo>>; |
| 71 |
|
| 72 |
template <typename... _More> |
| 73 |
_GLIBCXX_SIMD_INTRINSIC constexpr _Ret |
| 74 |
operator()(_Arg __a, _More... __more) const noexcept |
| 75 |
{ return __vector_convert<_V>(__a, __more...); } |
| 76 |
}; |
| 77 |
|
| 78 |
// }}} |
| 79 |
// _SimdConverter scalar -> fixed_size<1> {{{1 |
| 80 |
template <typename _From, typename _To> |
| 81 |
struct _SimdConverter<_From, simd_abi::scalar, _To, simd_abi::fixed_size<1>, |
| 82 |
void> |
| 83 |
{ |
| 84 |
_GLIBCXX_SIMD_INTRINSIC constexpr _SimdTuple<_To, simd_abi::scalar> |
| 85 |
operator()(_From __x) const noexcept |
| 86 |
{ return {static_cast<_To>(__x)}; } |
| 87 |
}; |
| 88 |
|
| 89 |
// _SimdConverter fixed_size<1> -> scalar {{{1 |
| 90 |
template <typename _From, typename _To> |
| 91 |
struct _SimdConverter<_From, simd_abi::fixed_size<1>, _To, simd_abi::scalar, |
| 92 |
void> |
| 93 |
{ |
| 94 |
_GLIBCXX_SIMD_INTRINSIC constexpr _To |
| 95 |
operator()(_SimdTuple<_From, simd_abi::scalar> __x) const noexcept |
| 96 |
{ return {static_cast<_To>(__x.first)}; } |
| 97 |
}; |
| 98 |
|
| 99 |
// _SimdConverter fixed_size<_Np> -> fixed_size<_Np> {{{1 |
| 100 |
template <typename _From, typename _To, int _Np> |
| 101 |
struct _SimdConverter<_From, simd_abi::fixed_size<_Np>, _To, |
| 102 |
simd_abi::fixed_size<_Np>, |
| 103 |
enable_if_t<!is_same_v<_From, _To>>> |
| 104 |
{ |
| 105 |
using _Ret = __fixed_size_storage_t<_To, _Np>; |
| 106 |
using _Arg = __fixed_size_storage_t<_From, _Np>; |
| 107 |
|
| 108 |
_GLIBCXX_SIMD_INTRINSIC constexpr _Ret |
| 109 |
operator()(const _Arg& __x) const noexcept |
| 110 |
{ |
| 111 |
if constexpr (is_same_v<_From, _To>) |
| 112 |
return __x; |
| 113 |
|
| 114 |
// special case (optimize) int signedness casts |
| 115 |
else if constexpr (sizeof(_From) == sizeof(_To) |
| 116 |
&& is_integral_v<_From> && is_integral_v<_To>) |
| 117 |
return __bit_cast<_Ret>(__x); |
| 118 |
|
| 119 |
// special case if all ABI tags in _Ret are scalar |
| 120 |
else if constexpr (__is_scalar_abi<typename _Ret::_FirstAbi>()) |
| 121 |
{ |
| 122 |
return __call_with_subscripts( |
| 123 |
__x, make_index_sequence<_Np>(), |
| 124 |
[](auto... __values) constexpr->_Ret { |
| 125 |
return __make_simd_tuple<_To, decltype((void) __values, |
| 126 |
simd_abi::scalar())...>( |
| 127 |
static_cast<_To>(__values)...); |
| 128 |
}); |
| 129 |
} |
| 130 |
|
| 131 |
// from one vector to one vector |
| 132 |
else if constexpr (_Arg::_S_first_size == _Ret::_S_first_size) |
| 133 |
{ |
| 134 |
_SimdConverter<_From, typename _Arg::_FirstAbi, _To, |
| 135 |
typename _Ret::_FirstAbi> |
| 136 |
__native_cvt; |
| 137 |
if constexpr (_Arg::_S_tuple_size == 1) |
| 138 |
return {__native_cvt(__x.first)}; |
| 139 |
else |
| 140 |
{ |
| 141 |
constexpr size_t _NRemain = _Np - _Arg::_S_first_size; |
| 142 |
_SimdConverter<_From, simd_abi::fixed_size<_NRemain>, _To, |
| 143 |
simd_abi::fixed_size<_NRemain>> |
| 144 |
__remainder_cvt; |
| 145 |
return {__native_cvt(__x.first), __remainder_cvt(__x.second)}; |
| 146 |
} |
| 147 |
} |
| 148 |
|
| 149 |
// from one vector to multiple vectors |
| 150 |
else if constexpr (_Arg::_S_first_size > _Ret::_S_first_size) |
| 151 |
{ |
| 152 |
const auto __multiple_return_chunks |
| 153 |
= __convert_all<__vector_type_t<_To, _Ret::_S_first_size>>( |
| 154 |
__x.first); |
| 155 |
constexpr auto __converted = __multiple_return_chunks.size() |
| 156 |
* _Ret::_FirstAbi::template _S_size<_To>; |
| 157 |
constexpr auto __remaining = _Np - __converted; |
| 158 |
if constexpr (_Arg::_S_tuple_size == 1 && __remaining == 0) |
| 159 |
return __to_simd_tuple<_To, _Np>(__multiple_return_chunks); |
| 160 |
else if constexpr (_Arg::_S_tuple_size == 1) |
| 161 |
{ // e.g. <int, 3> -> <double, 2, 1> or <short, 7> -> <double, 4, 2, |
| 162 |
// 1> |
| 163 |
using _RetRem |
| 164 |
= __remove_cvref_t<decltype(__simd_tuple_pop_front<__converted>( |
| 165 |
_Ret()))>; |
| 166 |
const auto __return_chunks2 |
| 167 |
= __convert_all<__vector_type_t<_To, _RetRem::_S_first_size>, 0, |
| 168 |
__converted>(__x.first); |
| 169 |
constexpr auto __converted2 |
| 170 |
= __converted |
| 171 |
+ __return_chunks2.size() * _RetRem::_S_first_size; |
| 172 |
if constexpr (__converted2 == _Np) |
| 173 |
return __to_simd_tuple<_To, _Np>(__multiple_return_chunks, |
| 174 |
__return_chunks2); |
| 175 |
else |
| 176 |
{ |
| 177 |
using _RetRem2 = __remove_cvref_t< |
| 178 |
decltype(__simd_tuple_pop_front<__return_chunks2.size() |
| 179 |
* _RetRem::_S_first_size>( |
| 180 |
_RetRem()))>; |
| 181 |
const auto __return_chunks3 = __convert_all< |
| 182 |
__vector_type_t<_To, _RetRem2::_S_first_size>, 0, |
| 183 |
__converted2>(__x.first); |
| 184 |
constexpr auto __converted3 |
| 185 |
= __converted2 |
| 186 |
+ __return_chunks3.size() * _RetRem2::_S_first_size; |
| 187 |
if constexpr (__converted3 == _Np) |
| 188 |
return __to_simd_tuple<_To, _Np>(__multiple_return_chunks, |
| 189 |
__return_chunks2, |
| 190 |
__return_chunks3); |
| 191 |
else |
| 192 |
{ |
| 193 |
using _RetRem3 |
| 194 |
= __remove_cvref_t<decltype(__simd_tuple_pop_front< |
| 195 |
__return_chunks3.size() |
| 196 |
* _RetRem2::_S_first_size>( |
| 197 |
_RetRem2()))>; |
| 198 |
const auto __return_chunks4 = __convert_all< |
| 199 |
__vector_type_t<_To, _RetRem3::_S_first_size>, 0, |
| 200 |
__converted3>(__x.first); |
| 201 |
constexpr auto __converted4 |
| 202 |
= __converted3 |
| 203 |
+ __return_chunks4.size() * _RetRem3::_S_first_size; |
| 204 |
if constexpr (__converted4 == _Np) |
| 205 |
return __to_simd_tuple<_To, _Np>( |
| 206 |
__multiple_return_chunks, __return_chunks2, |
| 207 |
__return_chunks3, __return_chunks4); |
| 208 |
else |
| 209 |
__assert_unreachable<_To>(); |
| 210 |
} |
| 211 |
} |
| 212 |
} |
| 213 |
else |
| 214 |
{ |
| 215 |
constexpr size_t _NRemain = _Np - _Arg::_S_first_size; |
| 216 |
_SimdConverter<_From, simd_abi::fixed_size<_NRemain>, _To, |
| 217 |
simd_abi::fixed_size<_NRemain>> |
| 218 |
__remainder_cvt; |
| 219 |
return __simd_tuple_concat( |
| 220 |
__to_simd_tuple<_To, _Arg::_S_first_size>( |
| 221 |
__multiple_return_chunks), |
| 222 |
__remainder_cvt(__x.second)); |
| 223 |
} |
| 224 |
} |
| 225 |
|
| 226 |
// from multiple vectors to one vector |
| 227 |
// _Arg::_S_first_size < _Ret::_S_first_size |
| 228 |
// a) heterogeneous input at the end of the tuple (possible with partial |
| 229 |
// native registers in _Ret) |
| 230 |
else if constexpr (_Ret::_S_tuple_size == 1 |
| 231 |
&& _Np % _Arg::_S_first_size != 0) |
| 232 |
{ |
| 233 |
static_assert(_Ret::_FirstAbi::template _S_is_partial<_To>); |
| 234 |
return _Ret{__generate_from_n_evaluations< |
| 235 |
_Np, typename _VectorTraits<typename _Ret::_FirstType>::type>( |
| 236 |
[&](auto __i) { return static_cast<_To>(__x[__i]); })}; |
| 237 |
} |
| 238 |
else |
| 239 |
{ |
| 240 |
static_assert(_Arg::_S_tuple_size > 1); |
| 241 |
constexpr auto __n |
| 242 |
= __div_roundup(_Ret::_S_first_size, _Arg::_S_first_size); |
| 243 |
return __call_with_n_evaluations<__n>( |
| 244 |
[&__x](auto... __uncvted) { |
| 245 |
// assuming _Arg Abi tags for all __i are _Arg::_FirstAbi |
| 246 |
_SimdConverter<_From, typename _Arg::_FirstAbi, _To, |
| 247 |
typename _Ret::_FirstAbi> |
| 248 |
__native_cvt; |
| 249 |
if constexpr (_Ret::_S_tuple_size == 1) |
| 250 |
return _Ret{__native_cvt(__uncvted...)}; |
| 251 |
else |
| 252 |
return _Ret{ |
| 253 |
__native_cvt(__uncvted...), |
| 254 |
_SimdConverter< |
| 255 |
_From, simd_abi::fixed_size<_Np - _Ret::_S_first_size>, _To, |
| 256 |
simd_abi::fixed_size<_Np - _Ret::_S_first_size>>()( |
| 257 |
__simd_tuple_pop_front<_Ret::_S_first_size>(__x))}; |
| 258 |
}, |
| 259 |
[&__x](auto __i) { return __get_tuple_at<__i>(__x); }); |
| 260 |
} |
| 261 |
} |
| 262 |
}; |
| 263 |
|
| 264 |
// _SimdConverter "native" -> fixed_size<_Np> {{{1 |
| 265 |
// i.e. 1 register to ? registers |
| 266 |
template <typename _From, typename _Ap, typename _To, int _Np> |
| 267 |
struct _SimdConverter<_From, _Ap, _To, simd_abi::fixed_size<_Np>, |
| 268 |
enable_if_t<!__is_fixed_size_abi_v<_Ap>>> |
| 269 |
{ |
| 270 |
static_assert( |
| 271 |
_Np == simd_size_v<_From, _Ap>, |
| 272 |
"_SimdConverter to fixed_size only works for equal element counts"); |
| 273 |
|
| 274 |
using _Ret = __fixed_size_storage_t<_To, _Np>; |
| 275 |
|
| 276 |
_GLIBCXX_SIMD_INTRINSIC constexpr _Ret |
| 277 |
operator()(typename _SimdTraits<_From, _Ap>::_SimdMember __x) const noexcept |
| 278 |
{ |
| 279 |
if constexpr (_Ret::_S_tuple_size == 1) |
| 280 |
return {__vector_convert<typename _Ret::_FirstType::_BuiltinType>(__x)}; |
| 281 |
else |
| 282 |
{ |
| 283 |
using _FixedNp = simd_abi::fixed_size<_Np>; |
| 284 |
_SimdConverter<_From, _FixedNp, _To, _FixedNp> __fixed_cvt; |
| 285 |
using _FromFixedStorage = __fixed_size_storage_t<_From, _Np>; |
| 286 |
if constexpr (_FromFixedStorage::_S_tuple_size == 1) |
| 287 |
return __fixed_cvt(_FromFixedStorage{__x}); |
| 288 |
else if constexpr (_FromFixedStorage::_S_tuple_size == 2) |
| 289 |
{ |
| 290 |
_FromFixedStorage __tmp; |
| 291 |
static_assert(sizeof(__tmp) <= sizeof(__x)); |
| 292 |
__builtin_memcpy(&__tmp.first, &__x, sizeof(__tmp.first)); |
| 293 |
__builtin_memcpy(&__tmp.second.first, |
| 294 |
reinterpret_cast<const char*>(&__x) |
| 295 |
+ sizeof(__tmp.first), |
| 296 |
sizeof(__tmp.second.first)); |
| 297 |
return __fixed_cvt(__tmp); |
| 298 |
} |
| 299 |
else |
| 300 |
__assert_unreachable<_From>(); |
| 301 |
} |
| 302 |
} |
| 303 |
}; |
| 304 |
|
| 305 |
// _SimdConverter fixed_size<_Np> -> "native" {{{1 |
| 306 |
// i.e. ? register to 1 registers |
| 307 |
template <typename _From, int _Np, typename _To, typename _Ap> |
| 308 |
struct _SimdConverter<_From, simd_abi::fixed_size<_Np>, _To, _Ap, |
| 309 |
enable_if_t<!__is_fixed_size_abi_v<_Ap>>> |
| 310 |
{ |
| 311 |
static_assert( |
| 312 |
_Np == simd_size_v<_To, _Ap>, |
| 313 |
"_SimdConverter to fixed_size only works for equal element counts"); |
| 314 |
|
| 315 |
using _Arg = __fixed_size_storage_t<_From, _Np>; |
| 316 |
|
| 317 |
_GLIBCXX_SIMD_INTRINSIC constexpr |
| 318 |
typename _SimdTraits<_To, _Ap>::_SimdMember |
| 319 |
operator()(_Arg __x) const noexcept |
| 320 |
{ |
| 321 |
if constexpr (_Arg::_S_tuple_size == 1) |
| 322 |
return __vector_convert<__vector_type_t<_To, _Np>>(__x.first); |
| 323 |
else if constexpr (_Arg::_S_is_homogeneous) |
| 324 |
return __call_with_n_evaluations<_Arg::_S_tuple_size>( |
| 325 |
[](auto... __members) { |
| 326 |
if constexpr ((is_convertible_v<decltype(__members), _To> && ...)) |
| 327 |
return __vector_type_t<_To, _Np>{static_cast<_To>(__members)...}; |
| 328 |
else |
| 329 |
return __vector_convert<__vector_type_t<_To, _Np>>(__members...); |
| 330 |
}, |
| 331 |
[&](auto __i) { return __get_tuple_at<__i>(__x); }); |
| 332 |
else if constexpr (__fixed_size_storage_t<_To, _Np>::_S_tuple_size == 1) |
| 333 |
{ |
| 334 |
_SimdConverter<_From, simd_abi::fixed_size<_Np>, _To, |
| 335 |
simd_abi::fixed_size<_Np>> |
| 336 |
__fixed_cvt; |
| 337 |
return __fixed_cvt(__x).first; |
| 338 |
} |
| 339 |
else |
| 340 |
{ |
| 341 |
const _SimdWrapper<_From, _Np> __xv |
| 342 |
= __generate_from_n_evaluations<_Np, __vector_type_t<_From, _Np>>( |
| 343 |
[&](auto __i) { return __x[__i]; }); |
| 344 |
return __vector_convert<__vector_type_t<_To, _Np>>(__xv); |
| 345 |
} |
| 346 |
} |
| 347 |
}; |
| 348 |
|
| 349 |
// }}}1 |
| 350 |
_GLIBCXX_SIMD_END_NAMESPACE |
| 351 |
#endif // __cplusplus >= 201703L |
| 352 |
#endif // _GLIBCXX_EXPERIMENTAL_SIMD_CONVERTER_H_ |
| 353 |
|
| 354 |
// vim: foldmethod=marker sw=2 noet ts=8 sts=2 tw=80 |
