| 1 |
use crate::soft::{x2, x4}; |
| 2 |
use crate::types::*; |
| 3 |
use crate::vec128_storage; |
| 4 |
use crate::x86_64::Avx2Machine; |
| 5 |
use crate::x86_64::SseMachine as Machine86; |
| 6 |
use crate::x86_64::{NoS3, NoS4, YesS3, YesS4}; |
| 7 |
use core::arch::x86_64::*; |
| 8 |
use core::marker::PhantomData; |
| 9 |
use core::ops::{ |
| 10 |
Add, AddAssign, BitAnd, BitAndAssign, BitOr, BitOrAssign, BitXor, BitXorAssign, Not, |
| 11 |
}; |
| 12 |
use zerocopy::transmute; |
| 13 |
|
| 14 |
macro_rules! impl_binop { |
| 15 |
($vec:ident, $trait:ident, $fn:ident, $impl_fn:ident) => { |
| 16 |
impl<S3, S4, NI> $trait for $vec<S3, S4, NI> { |
| 17 |
type Output = Self; |
| 18 |
#[inline(always)] |
| 19 |
fn $fn(self, rhs: Self) -> Self::Output { |
| 20 |
Self::new(unsafe { $impl_fn(self.x, rhs.x) }) |
| 21 |
} |
| 22 |
} |
| 23 |
}; |
| 24 |
} |
| 25 |
|
| 26 |
macro_rules! impl_binop_assign { |
| 27 |
($vec:ident, $trait:ident, $fn_assign:ident, $fn:ident) => { |
| 28 |
impl<S3, S4, NI> $trait for $vec<S3, S4, NI> |
| 29 |
where |
| 30 |
$vec<S3, S4, NI>: Copy, |
| 31 |
{ |
| 32 |
#[inline(always)] |
| 33 |
fn $fn_assign(&mut self, rhs: Self) { |
| 34 |
*self = self.$fn(rhs); |
| 35 |
} |
| 36 |
} |
| 37 |
}; |
| 38 |
} |
| 39 |
|
| 40 |
macro_rules! def_vec { |
| 41 |
($vec:ident, $word:ident) => { |
| 42 |
zerocopy::cryptocorrosion_derive_traits! { |
| 43 |
#[repr(transparent)] |
| 44 |
#[allow(non_camel_case_types)] |
| 45 |
#[derive(Copy, Clone)] |
| 46 |
pub struct $vec<S3, S4, NI> { |
| 47 |
x: __m128i, |
| 48 |
s3: PhantomData<S3>, |
| 49 |
s4: PhantomData<S4>, |
| 50 |
ni: PhantomData<NI>, |
| 51 |
} |
| 52 |
} |
| 53 |
|
| 54 |
impl<S3, S4, NI> Store<vec128_storage> for $vec<S3, S4, NI> { |
| 55 |
#[inline(always)] |
| 56 |
unsafe fn unpack(x: vec128_storage) -> Self { |
| 57 |
Self::new(x.sse2) |
| 58 |
} |
| 59 |
} |
| 60 |
impl<S3, S4, NI> From<$vec<S3, S4, NI>> for vec128_storage { |
| 61 |
#[inline(always)] |
| 62 |
fn from(x: $vec<S3, S4, NI>) -> Self { |
| 63 |
vec128_storage { sse2: x.x } |
| 64 |
} |
| 65 |
} |
| 66 |
impl<S3, S4, NI> $vec<S3, S4, NI> { |
| 67 |
#[inline(always)] |
| 68 |
fn new(x: __m128i) -> Self { |
| 69 |
$vec { |
| 70 |
x, |
| 71 |
s3: PhantomData, |
| 72 |
s4: PhantomData, |
| 73 |
ni: PhantomData, |
| 74 |
} |
| 75 |
} |
| 76 |
} |
| 77 |
|
| 78 |
impl<S3, S4, NI> StoreBytes for $vec<S3, S4, NI> |
| 79 |
where |
| 80 |
Self: BSwap, |
| 81 |
{ |
| 82 |
#[inline(always)] |
| 83 |
unsafe fn unsafe_read_le(input: &[u8]) -> Self { |
| 84 |
assert_eq!(input.len(), 16); |
| 85 |
Self::new(_mm_loadu_si128(input.as_ptr() as *const _)) |
| 86 |
} |
| 87 |
#[inline(always)] |
| 88 |
unsafe fn unsafe_read_be(input: &[u8]) -> Self { |
| 89 |
assert_eq!(input.len(), 16); |
| 90 |
Self::new(_mm_loadu_si128(input.as_ptr() as *const _)).bswap() |
| 91 |
} |
| 92 |
#[inline(always)] |
| 93 |
fn write_le(self, out: &mut [u8]) { |
| 94 |
assert_eq!(out.len(), 16); |
| 95 |
unsafe { _mm_storeu_si128(out.as_mut_ptr() as *mut _, self.x) } |
| 96 |
} |
| 97 |
#[inline(always)] |
| 98 |
fn write_be(self, out: &mut [u8]) { |
| 99 |
assert_eq!(out.len(), 16); |
| 100 |
let x = self.bswap().x; |
| 101 |
unsafe { |
| 102 |
_mm_storeu_si128(out.as_mut_ptr() as *mut _, x); |
| 103 |
} |
| 104 |
} |
| 105 |
} |
| 106 |
|
| 107 |
impl<S3, S4, NI> Default for $vec<S3, S4, NI> { |
| 108 |
#[inline(always)] |
| 109 |
fn default() -> Self { |
| 110 |
Self::new(unsafe { _mm_setzero_si128() }) |
| 111 |
} |
| 112 |
} |
| 113 |
|
| 114 |
impl<S3, S4, NI> Not for $vec<S3, S4, NI> { |
| 115 |
type Output = Self; |
| 116 |
#[inline(always)] |
| 117 |
fn not(self) -> Self::Output { |
| 118 |
unsafe { |
| 119 |
let ff = _mm_set1_epi64x(-1i64); |
| 120 |
self ^ Self::new(ff) |
| 121 |
} |
| 122 |
} |
| 123 |
} |
| 124 |
|
| 125 |
impl<S3: Copy, S4: Copy, NI: Copy> BitOps0 for $vec<S3, S4, NI> {} |
| 126 |
impl_binop!($vec, BitAnd, bitand, _mm_and_si128); |
| 127 |
impl_binop!($vec, BitOr, bitor, _mm_or_si128); |
| 128 |
impl_binop!($vec, BitXor, bitxor, _mm_xor_si128); |
| 129 |
impl_binop_assign!($vec, BitAndAssign, bitand_assign, bitand); |
| 130 |
impl_binop_assign!($vec, BitOrAssign, bitor_assign, bitor); |
| 131 |
impl_binop_assign!($vec, BitXorAssign, bitxor_assign, bitxor); |
| 132 |
impl<S3: Copy, S4: Copy, NI: Copy> AndNot for $vec<S3, S4, NI> { |
| 133 |
type Output = Self; |
| 134 |
#[inline(always)] |
| 135 |
fn andnot(self, rhs: Self) -> Self { |
| 136 |
Self::new(unsafe { _mm_andnot_si128(self.x, rhs.x) }) |
| 137 |
} |
| 138 |
} |
| 139 |
}; |
| 140 |
} |
| 141 |
|
| 142 |
macro_rules! impl_bitops32 { |
| 143 |
($vec:ident) => { |
| 144 |
impl<S3: Copy, S4: Copy, NI: Copy> BitOps32 for $vec<S3, S4, NI> where |
| 145 |
$vec<S3, S4, NI>: RotateEachWord32 |
| 146 |
{ |
| 147 |
} |
| 148 |
}; |
| 149 |
} |
| 150 |
|
| 151 |
macro_rules! impl_bitops64 { |
| 152 |
($vec:ident) => { |
| 153 |
impl_bitops32!($vec); |
| 154 |
impl<S3: Copy, S4: Copy, NI: Copy> BitOps64 for $vec<S3, S4, NI> where |
| 155 |
$vec<S3, S4, NI>: RotateEachWord64 + RotateEachWord32 |
| 156 |
{ |
| 157 |
} |
| 158 |
}; |
| 159 |
} |
| 160 |
|
| 161 |
macro_rules! impl_bitops128 { |
| 162 |
($vec:ident) => { |
| 163 |
impl_bitops64!($vec); |
| 164 |
impl<S3: Copy, S4: Copy, NI: Copy> BitOps128 for $vec<S3, S4, NI> where |
| 165 |
$vec<S3, S4, NI>: RotateEachWord128 |
| 166 |
{ |
| 167 |
} |
| 168 |
}; |
| 169 |
} |
| 170 |
|
| 171 |
macro_rules! rotr_32_s3 { |
| 172 |
($name:ident, $k0:expr, $k1:expr) => { |
| 173 |
#[inline(always)] |
| 174 |
fn $name(self) -> Self { |
| 175 |
Self::new(unsafe { _mm_shuffle_epi8(self.x, _mm_set_epi64x($k0, $k1)) }) |
| 176 |
} |
| 177 |
}; |
| 178 |
} |
| 179 |
macro_rules! rotr_32 { |
| 180 |
($name:ident, $i:expr) => { |
| 181 |
#[inline(always)] |
| 182 |
fn $name(self) -> Self { |
| 183 |
Self::new(unsafe { |
| 184 |
_mm_or_si128( |
| 185 |
_mm_srli_epi32(self.x, $i as i32), |
| 186 |
_mm_slli_epi32(self.x, 32 - $i as i32), |
| 187 |
) |
| 188 |
}) |
| 189 |
} |
| 190 |
}; |
| 191 |
} |
| 192 |
impl<S4: Copy, NI: Copy> RotateEachWord32 for u32x4_sse2<YesS3, S4, NI> { |
| 193 |
rotr_32!(rotate_each_word_right7, 7); |
| 194 |
rotr_32_s3!( |
| 195 |
rotate_each_word_right8, |
| 196 |
0x0c0f_0e0d_080b_0a09, |
| 197 |
0x0407_0605_0003_0201 |
| 198 |
); |
| 199 |
rotr_32!(rotate_each_word_right11, 11); |
| 200 |
rotr_32!(rotate_each_word_right12, 12); |
| 201 |
rotr_32_s3!( |
| 202 |
rotate_each_word_right16, |
| 203 |
0x0d0c_0f0e_0908_0b0a, |
| 204 |
0x0504_0706_0100_0302 |
| 205 |
); |
| 206 |
rotr_32!(rotate_each_word_right20, 20); |
| 207 |
rotr_32_s3!( |
| 208 |
rotate_each_word_right24, |
| 209 |
0x0e0d_0c0f_0a09_080b, |
| 210 |
0x0605_0407_0201_0003 |
| 211 |
); |
| 212 |
rotr_32!(rotate_each_word_right25, 25); |
| 213 |
} |
| 214 |
impl<S4: Copy, NI: Copy> RotateEachWord32 for u32x4_sse2<NoS3, S4, NI> { |
| 215 |
rotr_32!(rotate_each_word_right7, 7); |
| 216 |
rotr_32!(rotate_each_word_right8, 8); |
| 217 |
rotr_32!(rotate_each_word_right11, 11); |
| 218 |
rotr_32!(rotate_each_word_right12, 12); |
| 219 |
#[inline(always)] |
| 220 |
fn rotate_each_word_right16(self) -> Self { |
| 221 |
Self::new(swap16_s2(self.x)) |
| 222 |
} |
| 223 |
rotr_32!(rotate_each_word_right20, 20); |
| 224 |
rotr_32!(rotate_each_word_right24, 24); |
| 225 |
rotr_32!(rotate_each_word_right25, 25); |
| 226 |
} |
| 227 |
|
| 228 |
macro_rules! rotr_64_s3 { |
| 229 |
($name:ident, $k0:expr, $k1:expr) => { |
| 230 |
#[inline(always)] |
| 231 |
fn $name(self) -> Self { |
| 232 |
Self::new(unsafe { _mm_shuffle_epi8(self.x, _mm_set_epi64x($k0, $k1)) }) |
| 233 |
} |
| 234 |
}; |
| 235 |
} |
| 236 |
macro_rules! rotr_64 { |
| 237 |
($name:ident, $i:expr) => { |
| 238 |
#[inline(always)] |
| 239 |
fn $name(self) -> Self { |
| 240 |
Self::new(unsafe { |
| 241 |
_mm_or_si128( |
| 242 |
_mm_srli_epi64(self.x, $i as i32), |
| 243 |
_mm_slli_epi64(self.x, 64 - $i as i32), |
| 244 |
) |
| 245 |
}) |
| 246 |
} |
| 247 |
}; |
| 248 |
} |
| 249 |
impl<S4: Copy, NI: Copy> RotateEachWord32 for u64x2_sse2<YesS3, S4, NI> { |
| 250 |
rotr_64!(rotate_each_word_right7, 7); |
| 251 |
rotr_64_s3!( |
| 252 |
rotate_each_word_right8, |
| 253 |
0x080f_0e0d_0c0b_0a09, |
| 254 |
0x0007_0605_0403_0201 |
| 255 |
); |
| 256 |
rotr_64!(rotate_each_word_right11, 11); |
| 257 |
rotr_64!(rotate_each_word_right12, 12); |
| 258 |
rotr_64_s3!( |
| 259 |
rotate_each_word_right16, |
| 260 |
0x0908_0f0e_0d0c_0b0a, |
| 261 |
0x0100_0706_0504_0302 |
| 262 |
); |
| 263 |
rotr_64!(rotate_each_word_right20, 20); |
| 264 |
rotr_64_s3!( |
| 265 |
rotate_each_word_right24, |
| 266 |
0x0a09_080f_0e0d_0c0b, |
| 267 |
0x0201_0007_0605_0403 |
| 268 |
); |
| 269 |
rotr_64!(rotate_each_word_right25, 25); |
| 270 |
} |
| 271 |
impl<S4: Copy, NI: Copy> RotateEachWord32 for u64x2_sse2<NoS3, S4, NI> { |
| 272 |
rotr_64!(rotate_each_word_right7, 7); |
| 273 |
rotr_64!(rotate_each_word_right8, 8); |
| 274 |
rotr_64!(rotate_each_word_right11, 11); |
| 275 |
rotr_64!(rotate_each_word_right12, 12); |
| 276 |
#[inline(always)] |
| 277 |
fn rotate_each_word_right16(self) -> Self { |
| 278 |
Self::new(swap16_s2(self.x)) |
| 279 |
} |
| 280 |
rotr_64!(rotate_each_word_right20, 20); |
| 281 |
rotr_64!(rotate_each_word_right24, 24); |
| 282 |
rotr_64!(rotate_each_word_right25, 25); |
| 283 |
} |
| 284 |
impl<S3: Copy, S4: Copy, NI: Copy> RotateEachWord64 for u64x2_sse2<S3, S4, NI> { |
| 285 |
#[inline(always)] |
| 286 |
fn rotate_each_word_right32(self) -> Self { |
| 287 |
Self::new(unsafe { _mm_shuffle_epi32(self.x, 0b10110001) }) |
| 288 |
} |
| 289 |
} |
| 290 |
|
| 291 |
macro_rules! rotr_128 { |
| 292 |
($name:ident, $i:expr) => { |
| 293 |
#[inline(always)] |
| 294 |
fn $name(self) -> Self { |
| 295 |
Self::new(unsafe { |
| 296 |
_mm_or_si128( |
| 297 |
_mm_srli_si128(self.x, $i as i32), |
| 298 |
_mm_slli_si128(self.x, 128 - $i as i32), |
| 299 |
) |
| 300 |
}) |
| 301 |
} |
| 302 |
}; |
| 303 |
} |
| 304 |
// TODO: completely unoptimized |
| 305 |
impl<S3: Copy, S4: Copy, NI: Copy> RotateEachWord32 for u128x1_sse2<S3, S4, NI> { |
| 306 |
rotr_128!(rotate_each_word_right7, 7); |
| 307 |
rotr_128!(rotate_each_word_right8, 8); |
| 308 |
rotr_128!(rotate_each_word_right11, 11); |
| 309 |
rotr_128!(rotate_each_word_right12, 12); |
| 310 |
rotr_128!(rotate_each_word_right16, 16); |
| 311 |
rotr_128!(rotate_each_word_right20, 20); |
| 312 |
rotr_128!(rotate_each_word_right24, 24); |
| 313 |
rotr_128!(rotate_each_word_right25, 25); |
| 314 |
} |
| 315 |
// TODO: completely unoptimized |
| 316 |
impl<S3: Copy, S4: Copy, NI: Copy> RotateEachWord64 for u128x1_sse2<S3, S4, NI> { |
| 317 |
rotr_128!(rotate_each_word_right32, 32); |
| 318 |
} |
| 319 |
impl<S3: Copy, S4: Copy, NI: Copy> RotateEachWord128 for u128x1_sse2<S3, S4, NI> {} |
| 320 |
|
| 321 |
def_vec!(u32x4_sse2, u32); |
| 322 |
def_vec!(u64x2_sse2, u64); |
| 323 |
def_vec!(u128x1_sse2, u128); |
| 324 |
|
| 325 |
impl<S3, NI> MultiLane<[u32; 4]> for u32x4_sse2<S3, YesS4, NI> { |
| 326 |
#[inline(always)] |
| 327 |
fn to_lanes(self) -> [u32; 4] { |
| 328 |
unsafe { |
| 329 |
let x = _mm_cvtsi128_si64(self.x) as u64; |
| 330 |
let y = _mm_extract_epi64(self.x, 1) as u64; |
| 331 |
[x as u32, (x >> 32) as u32, y as u32, (y >> 32) as u32] |
| 332 |
} |
| 333 |
} |
| 334 |
#[inline(always)] |
| 335 |
fn from_lanes(xs: [u32; 4]) -> Self { |
| 336 |
unsafe { |
| 337 |
let mut x = _mm_cvtsi64_si128((xs[0] as u64 | ((xs[1] as u64) << 32)) as i64); |
| 338 |
x = _mm_insert_epi64(x, (xs[2] as u64 | ((xs[3] as u64) << 32)) as i64, 1); |
| 339 |
Self::new(x) |
| 340 |
} |
| 341 |
} |
| 342 |
} |
| 343 |
impl<S3, NI> MultiLane<[u32; 4]> for u32x4_sse2<S3, NoS4, NI> { |
| 344 |
#[inline(always)] |
| 345 |
fn to_lanes(self) -> [u32; 4] { |
| 346 |
unsafe { |
| 347 |
let x = _mm_cvtsi128_si64(self.x) as u64; |
| 348 |
let y = _mm_cvtsi128_si64(_mm_shuffle_epi32(self.x, 0b11101110)) as u64; |
| 349 |
[x as u32, (x >> 32) as u32, y as u32, (y >> 32) as u32] |
| 350 |
} |
| 351 |
} |
| 352 |
#[inline(always)] |
| 353 |
fn from_lanes(xs: [u32; 4]) -> Self { |
| 354 |
unsafe { |
| 355 |
let x = (xs[0] as u64 | ((xs[1] as u64) << 32)) as i64; |
| 356 |
let y = (xs[2] as u64 | ((xs[3] as u64) << 32)) as i64; |
| 357 |
let x = _mm_cvtsi64_si128(x); |
| 358 |
let y = _mm_slli_si128(_mm_cvtsi64_si128(y), 8); |
| 359 |
Self::new(_mm_or_si128(x, y)) |
| 360 |
} |
| 361 |
} |
| 362 |
} |
| 363 |
impl<S3, NI> MultiLane<[u64; 2]> for u64x2_sse2<S3, YesS4, NI> { |
| 364 |
#[inline(always)] |
| 365 |
fn to_lanes(self) -> [u64; 2] { |
| 366 |
unsafe { |
| 367 |
[ |
| 368 |
_mm_cvtsi128_si64(self.x) as u64, |
| 369 |
_mm_extract_epi64(self.x, 1) as u64, |
| 370 |
] |
| 371 |
} |
| 372 |
} |
| 373 |
#[inline(always)] |
| 374 |
fn from_lanes(xs: [u64; 2]) -> Self { |
| 375 |
unsafe { |
| 376 |
let mut x = _mm_cvtsi64_si128(xs[0] as i64); |
| 377 |
x = _mm_insert_epi64(x, xs[1] as i64, 1); |
| 378 |
Self::new(x) |
| 379 |
} |
| 380 |
} |
| 381 |
} |
| 382 |
impl<S3, NI> MultiLane<[u64; 2]> for u64x2_sse2<S3, NoS4, NI> { |
| 383 |
#[inline(always)] |
| 384 |
fn to_lanes(self) -> [u64; 2] { |
| 385 |
unsafe { |
| 386 |
[ |
| 387 |
_mm_cvtsi128_si64(self.x) as u64, |
| 388 |
_mm_cvtsi128_si64(_mm_srli_si128(self.x, 8)) as u64, |
| 389 |
] |
| 390 |
} |
| 391 |
} |
| 392 |
#[inline(always)] |
| 393 |
fn from_lanes(xs: [u64; 2]) -> Self { |
| 394 |
unsafe { |
| 395 |
let x = _mm_cvtsi64_si128(xs[0] as i64); |
| 396 |
let y = _mm_slli_si128(_mm_cvtsi64_si128(xs[1] as i64), 8); |
| 397 |
Self::new(_mm_or_si128(x, y)) |
| 398 |
} |
| 399 |
} |
| 400 |
} |
| 401 |
impl<S3, S4, NI> MultiLane<[u128; 1]> for u128x1_sse2<S3, S4, NI> { |
| 402 |
#[inline(always)] |
| 403 |
fn to_lanes(self) -> [u128; 1] { |
| 404 |
unimplemented!() |
| 405 |
} |
| 406 |
#[inline(always)] |
| 407 |
fn from_lanes(xs: [u128; 1]) -> Self { |
| 408 |
unimplemented!("{:?}", xs) |
| 409 |
} |
| 410 |
} |
| 411 |
|
| 412 |
impl<S3, S4, NI> MultiLane<[u64; 4]> for u64x4_sse2<S3, S4, NI> |
| 413 |
where |
| 414 |
u64x2_sse2<S3, S4, NI>: MultiLane<[u64; 2]> + Copy, |
| 415 |
{ |
| 416 |
#[inline(always)] |
| 417 |
fn to_lanes(self) -> [u64; 4] { |
| 418 |
let (a, b) = (self.0[0].to_lanes(), self.0[1].to_lanes()); |
| 419 |
[a[0], a[1], b[0], b[1]] |
| 420 |
} |
| 421 |
#[inline(always)] |
| 422 |
fn from_lanes(xs: [u64; 4]) -> Self { |
| 423 |
let (a, b) = ( |
| 424 |
u64x2_sse2::from_lanes([xs[0], xs[1]]), |
| 425 |
u64x2_sse2::from_lanes([xs[2], xs[3]]), |
| 426 |
); |
| 427 |
x2::new([a, b]) |
| 428 |
} |
| 429 |
} |
| 430 |
|
| 431 |
macro_rules! impl_into { |
| 432 |
($from:ident, $to:ident) => { |
| 433 |
impl<S3, S4, NI> From<$from<S3, S4, NI>> for $to<S3, S4, NI> { |
| 434 |
#[inline(always)] |
| 435 |
fn from(x: $from<S3, S4, NI>) -> Self { |
| 436 |
$to::new(x.x) |
| 437 |
} |
| 438 |
} |
| 439 |
}; |
| 440 |
} |
| 441 |
|
| 442 |
impl_into!(u128x1_sse2, u32x4_sse2); |
| 443 |
impl_into!(u128x1_sse2, u64x2_sse2); |
| 444 |
|
| 445 |
impl_bitops32!(u32x4_sse2); |
| 446 |
impl_bitops64!(u64x2_sse2); |
| 447 |
impl_bitops128!(u128x1_sse2); |
| 448 |
|
| 449 |
impl<S3: Copy, S4: Copy, NI: Copy> ArithOps for u32x4_sse2<S3, S4, NI> where |
| 450 |
u32x4_sse2<S3, S4, NI>: BSwap |
| 451 |
{ |
| 452 |
} |
| 453 |
impl<S3: Copy, S4: Copy, NI: Copy> ArithOps for u64x2_sse2<S3, S4, NI> where |
| 454 |
u64x2_sse2<S3, S4, NI>: BSwap |
| 455 |
{ |
| 456 |
} |
| 457 |
impl_binop!(u32x4_sse2, Add, add, _mm_add_epi32); |
| 458 |
impl_binop!(u64x2_sse2, Add, add, _mm_add_epi64); |
| 459 |
impl_binop_assign!(u32x4_sse2, AddAssign, add_assign, add); |
| 460 |
impl_binop_assign!(u64x2_sse2, AddAssign, add_assign, add); |
| 461 |
|
| 462 |
impl<S3: Copy, S4: Copy, NI: Copy> u32x4<Machine86<S3, S4, NI>> for u32x4_sse2<S3, S4, NI> |
| 463 |
where |
| 464 |
u32x4_sse2<S3, S4, NI>: RotateEachWord32 + BSwap + MultiLane<[u32; 4]> + Vec4<u32>, |
| 465 |
Machine86<S3, S4, NI>: Machine, |
| 466 |
{ |
| 467 |
} |
| 468 |
impl<S3: Copy, S4: Copy, NI: Copy> u64x2<Machine86<S3, S4, NI>> for u64x2_sse2<S3, S4, NI> |
| 469 |
where |
| 470 |
u64x2_sse2<S3, S4, NI>: |
| 471 |
RotateEachWord64 + RotateEachWord32 + BSwap + MultiLane<[u64; 2]> + Vec2<u64>, |
| 472 |
Machine86<S3, S4, NI>: Machine, |
| 473 |
{ |
| 474 |
} |
| 475 |
impl<S3: Copy, S4: Copy, NI: Copy> u128x1<Machine86<S3, S4, NI>> for u128x1_sse2<S3, S4, NI> |
| 476 |
where |
| 477 |
u128x1_sse2<S3, S4, NI>: Swap64 + RotateEachWord64 + RotateEachWord32 + BSwap, |
| 478 |
Machine86<S3, S4, NI>: Machine, |
| 479 |
u128x1_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u32x4>, |
| 480 |
u128x1_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u64x2>, |
| 481 |
{ |
| 482 |
} |
| 483 |
|
| 484 |
impl<NI: Copy> u32x4<Avx2Machine<NI>> for u32x4_sse2<YesS3, YesS4, NI> |
| 485 |
where |
| 486 |
u32x4_sse2<YesS3, YesS4, NI>: RotateEachWord32 + BSwap + MultiLane<[u32; 4]> + Vec4<u32>, |
| 487 |
Machine86<YesS3, YesS4, NI>: Machine, |
| 488 |
{ |
| 489 |
} |
| 490 |
impl<NI: Copy> u64x2<Avx2Machine<NI>> for u64x2_sse2<YesS3, YesS4, NI> |
| 491 |
where |
| 492 |
u64x2_sse2<YesS3, YesS4, NI>: |
| 493 |
RotateEachWord64 + RotateEachWord32 + BSwap + MultiLane<[u64; 2]> + Vec2<u64>, |
| 494 |
Machine86<YesS3, YesS4, NI>: Machine, |
| 495 |
{ |
| 496 |
} |
| 497 |
impl<NI: Copy> u128x1<Avx2Machine<NI>> for u128x1_sse2<YesS3, YesS4, NI> |
| 498 |
where |
| 499 |
u128x1_sse2<YesS3, YesS4, NI>: Swap64 + RotateEachWord64 + RotateEachWord32 + BSwap, |
| 500 |
Machine86<YesS3, YesS4, NI>: Machine, |
| 501 |
u128x1_sse2<YesS3, YesS4, NI>: Into<<Machine86<YesS3, YesS4, NI> as Machine>::u32x4>, |
| 502 |
u128x1_sse2<YesS3, YesS4, NI>: Into<<Machine86<YesS3, YesS4, NI> as Machine>::u64x2>, |
| 503 |
{ |
| 504 |
} |
| 505 |
|
| 506 |
impl<S3, S4, NI> UnsafeFrom<[u32; 4]> for u32x4_sse2<S3, S4, NI> { |
| 507 |
#[inline(always)] |
| 508 |
unsafe fn unsafe_from(xs: [u32; 4]) -> Self { |
| 509 |
Self::new(_mm_set_epi32( |
| 510 |
xs[3] as i32, |
| 511 |
xs[2] as i32, |
| 512 |
xs[1] as i32, |
| 513 |
xs[0] as i32, |
| 514 |
)) |
| 515 |
} |
| 516 |
} |
| 517 |
|
| 518 |
impl<S3, NI> Vec4<u32> for u32x4_sse2<S3, YesS4, NI> |
| 519 |
where |
| 520 |
Self: MultiLane<[u32; 4]>, |
| 521 |
{ |
| 522 |
#[inline(always)] |
| 523 |
fn extract(self, i: u32) -> u32 { |
| 524 |
self.to_lanes()[i as usize] |
| 525 |
} |
| 526 |
#[inline(always)] |
| 527 |
fn insert(self, v: u32, i: u32) -> Self { |
| 528 |
Self::new(unsafe { |
| 529 |
match i { |
| 530 |
0 => _mm_insert_epi32(self.x, v as i32, 0), |
| 531 |
1 => _mm_insert_epi32(self.x, v as i32, 1), |
| 532 |
2 => _mm_insert_epi32(self.x, v as i32, 2), |
| 533 |
3 => _mm_insert_epi32(self.x, v as i32, 3), |
| 534 |
_ => unreachable!(), |
| 535 |
} |
| 536 |
}) |
| 537 |
} |
| 538 |
} |
| 539 |
impl<S3, NI> Vec4<u32> for u32x4_sse2<S3, NoS4, NI> |
| 540 |
where |
| 541 |
Self: MultiLane<[u32; 4]>, |
| 542 |
{ |
| 543 |
#[inline(always)] |
| 544 |
fn extract(self, i: u32) -> u32 { |
| 545 |
self.to_lanes()[i as usize] |
| 546 |
} |
| 547 |
#[inline(always)] |
| 548 |
fn insert(self, v: u32, i: u32) -> Self { |
| 549 |
Self::new(unsafe { |
| 550 |
match i { |
| 551 |
0 => { |
| 552 |
let x = _mm_andnot_si128(_mm_cvtsi32_si128(-1), self.x); |
| 553 |
_mm_or_si128(x, _mm_cvtsi32_si128(v as i32)) |
| 554 |
} |
| 555 |
1 => { |
| 556 |
let mut x = _mm_shuffle_epi32(self.x, 0b0111_1000); |
| 557 |
x = _mm_slli_si128(x, 4); |
| 558 |
x = _mm_or_si128(x, _mm_cvtsi32_si128(v as i32)); |
| 559 |
_mm_shuffle_epi32(x, 0b1110_0001) |
| 560 |
} |
| 561 |
2 => { |
| 562 |
let mut x = _mm_shuffle_epi32(self.x, 0b1011_0100); |
| 563 |
x = _mm_slli_si128(x, 4); |
| 564 |
x = _mm_or_si128(x, _mm_cvtsi32_si128(v as i32)); |
| 565 |
_mm_shuffle_epi32(x, 0b1100_1001) |
| 566 |
} |
| 567 |
3 => { |
| 568 |
let mut x = _mm_slli_si128(self.x, 4); |
| 569 |
x = _mm_or_si128(x, _mm_cvtsi32_si128(v as i32)); |
| 570 |
_mm_shuffle_epi32(x, 0b0011_1001) |
| 571 |
} |
| 572 |
_ => unreachable!(), |
| 573 |
} |
| 574 |
}) |
| 575 |
} |
| 576 |
} |
| 577 |
|
| 578 |
impl<S3, S4, NI> LaneWords4 for u32x4_sse2<S3, S4, NI> { |
| 579 |
#[inline(always)] |
| 580 |
fn shuffle_lane_words2301(self) -> Self { |
| 581 |
self.shuffle2301() |
| 582 |
} |
| 583 |
#[inline(always)] |
| 584 |
fn shuffle_lane_words1230(self) -> Self { |
| 585 |
self.shuffle1230() |
| 586 |
} |
| 587 |
#[inline(always)] |
| 588 |
fn shuffle_lane_words3012(self) -> Self { |
| 589 |
self.shuffle3012() |
| 590 |
} |
| 591 |
} |
| 592 |
|
| 593 |
impl<S3, S4, NI> Words4 for u32x4_sse2<S3, S4, NI> { |
| 594 |
#[inline(always)] |
| 595 |
fn shuffle2301(self) -> Self { |
| 596 |
Self::new(unsafe { _mm_shuffle_epi32(self.x, 0b0100_1110) }) |
| 597 |
} |
| 598 |
#[inline(always)] |
| 599 |
fn shuffle1230(self) -> Self { |
| 600 |
Self::new(unsafe { _mm_shuffle_epi32(self.x, 0b1001_0011) }) |
| 601 |
} |
| 602 |
#[inline(always)] |
| 603 |
fn shuffle3012(self) -> Self { |
| 604 |
Self::new(unsafe { _mm_shuffle_epi32(self.x, 0b0011_1001) }) |
| 605 |
} |
| 606 |
} |
| 607 |
|
| 608 |
impl<S4, NI> Words4 for u64x4_sse2<YesS3, S4, NI> { |
| 609 |
#[inline(always)] |
| 610 |
fn shuffle2301(self) -> Self { |
| 611 |
x2::new([u64x2_sse2::new(self.0[1].x), u64x2_sse2::new(self.0[0].x)]) |
| 612 |
} |
| 613 |
#[inline(always)] |
| 614 |
fn shuffle3012(self) -> Self { |
| 615 |
unsafe { |
| 616 |
x2::new([ |
| 617 |
u64x2_sse2::new(_mm_alignr_epi8(self.0[1].x, self.0[0].x, 8)), |
| 618 |
u64x2_sse2::new(_mm_alignr_epi8(self.0[0].x, self.0[1].x, 8)), |
| 619 |
]) |
| 620 |
} |
| 621 |
} |
| 622 |
#[inline(always)] |
| 623 |
fn shuffle1230(self) -> Self { |
| 624 |
unsafe { |
| 625 |
x2::new([ |
| 626 |
u64x2_sse2::new(_mm_alignr_epi8(self.0[0].x, self.0[1].x, 8)), |
| 627 |
u64x2_sse2::new(_mm_alignr_epi8(self.0[1].x, self.0[0].x, 8)), |
| 628 |
]) |
| 629 |
} |
| 630 |
} |
| 631 |
} |
| 632 |
impl<S4, NI> Words4 for u64x4_sse2<NoS3, S4, NI> { |
| 633 |
#[inline(always)] |
| 634 |
fn shuffle2301(self) -> Self { |
| 635 |
x2::new([u64x2_sse2::new(self.0[1].x), u64x2_sse2::new(self.0[0].x)]) |
| 636 |
} |
| 637 |
#[inline(always)] |
| 638 |
fn shuffle3012(self) -> Self { |
| 639 |
unsafe { |
| 640 |
let a = _mm_srli_si128(self.0[0].x, 8); |
| 641 |
let b = _mm_slli_si128(self.0[0].x, 8); |
| 642 |
let c = _mm_srli_si128(self.0[1].x, 8); |
| 643 |
let d = _mm_slli_si128(self.0[1].x, 8); |
| 644 |
let da = _mm_or_si128(d, a); |
| 645 |
let bc = _mm_or_si128(b, c); |
| 646 |
x2::new([u64x2_sse2::new(da), u64x2_sse2::new(bc)]) |
| 647 |
} |
| 648 |
} |
| 649 |
#[inline(always)] |
| 650 |
fn shuffle1230(self) -> Self { |
| 651 |
unsafe { |
| 652 |
let a = _mm_srli_si128(self.0[0].x, 8); |
| 653 |
let b = _mm_slli_si128(self.0[0].x, 8); |
| 654 |
let c = _mm_srli_si128(self.0[1].x, 8); |
| 655 |
let d = _mm_slli_si128(self.0[1].x, 8); |
| 656 |
let da = _mm_or_si128(d, a); |
| 657 |
let bc = _mm_or_si128(b, c); |
| 658 |
x2::new([u64x2_sse2::new(bc), u64x2_sse2::new(da)]) |
| 659 |
} |
| 660 |
} |
| 661 |
} |
| 662 |
|
| 663 |
impl<S3, S4, NI> UnsafeFrom<[u64; 2]> for u64x2_sse2<S3, S4, NI> { |
| 664 |
#[inline(always)] |
| 665 |
unsafe fn unsafe_from(xs: [u64; 2]) -> Self { |
| 666 |
Self::new(_mm_set_epi64x(xs[1] as i64, xs[0] as i64)) |
| 667 |
} |
| 668 |
} |
| 669 |
|
| 670 |
impl<S3, NI> Vec2<u64> for u64x2_sse2<S3, YesS4, NI> { |
| 671 |
#[inline(always)] |
| 672 |
fn extract(self, i: u32) -> u64 { |
| 673 |
unsafe { |
| 674 |
match i { |
| 675 |
0 => _mm_cvtsi128_si64(self.x) as u64, |
| 676 |
1 => _mm_extract_epi64(self.x, 1) as u64, |
| 677 |
_ => unreachable!(), |
| 678 |
} |
| 679 |
} |
| 680 |
} |
| 681 |
#[inline(always)] |
| 682 |
fn insert(self, x: u64, i: u32) -> Self { |
| 683 |
Self::new(unsafe { |
| 684 |
match i { |
| 685 |
0 => _mm_insert_epi64(self.x, x as i64, 0), |
| 686 |
1 => _mm_insert_epi64(self.x, x as i64, 1), |
| 687 |
_ => unreachable!(), |
| 688 |
} |
| 689 |
}) |
| 690 |
} |
| 691 |
} |
| 692 |
impl<S3, NI> Vec2<u64> for u64x2_sse2<S3, NoS4, NI> { |
| 693 |
#[inline(always)] |
| 694 |
fn extract(self, i: u32) -> u64 { |
| 695 |
unsafe { |
| 696 |
match i { |
| 697 |
0 => _mm_cvtsi128_si64(self.x) as u64, |
| 698 |
1 => _mm_cvtsi128_si64(_mm_shuffle_epi32(self.x, 0b11101110)) as u64, |
| 699 |
_ => unreachable!(), |
| 700 |
} |
| 701 |
} |
| 702 |
} |
| 703 |
#[inline(always)] |
| 704 |
fn insert(self, x: u64, i: u32) -> Self { |
| 705 |
Self::new(unsafe { |
| 706 |
match i { |
| 707 |
0 => _mm_or_si128( |
| 708 |
_mm_andnot_si128(_mm_cvtsi64_si128(-1), self.x), |
| 709 |
_mm_cvtsi64_si128(x as i64), |
| 710 |
), |
| 711 |
1 => _mm_or_si128( |
| 712 |
_mm_move_epi64(self.x), |
| 713 |
_mm_slli_si128(_mm_cvtsi64_si128(x as i64), 8), |
| 714 |
), |
| 715 |
_ => unreachable!(), |
| 716 |
} |
| 717 |
}) |
| 718 |
} |
| 719 |
} |
| 720 |
|
| 721 |
impl<S4, NI> BSwap for u32x4_sse2<YesS3, S4, NI> { |
| 722 |
#[inline(always)] |
| 723 |
fn bswap(self) -> Self { |
| 724 |
Self::new(unsafe { |
| 725 |
let k = _mm_set_epi64x(0x0c0d_0e0f_0809_0a0b, 0x0405_0607_0001_0203); |
| 726 |
_mm_shuffle_epi8(self.x, k) |
| 727 |
}) |
| 728 |
} |
| 729 |
} |
| 730 |
#[inline(always)] |
| 731 |
fn bswap32_s2(x: __m128i) -> __m128i { |
| 732 |
unsafe { |
| 733 |
let mut y = _mm_unpacklo_epi8(x, _mm_setzero_si128()); |
| 734 |
y = _mm_shufflehi_epi16(y, 0b0001_1011); |
| 735 |
y = _mm_shufflelo_epi16(y, 0b0001_1011); |
| 736 |
let mut z = _mm_unpackhi_epi8(x, _mm_setzero_si128()); |
| 737 |
z = _mm_shufflehi_epi16(z, 0b0001_1011); |
| 738 |
z = _mm_shufflelo_epi16(z, 0b0001_1011); |
| 739 |
_mm_packus_epi16(y, z) |
| 740 |
} |
| 741 |
} |
| 742 |
impl<S4, NI> BSwap for u32x4_sse2<NoS3, S4, NI> { |
| 743 |
#[inline(always)] |
| 744 |
fn bswap(self) -> Self { |
| 745 |
Self::new(bswap32_s2(self.x)) |
| 746 |
} |
| 747 |
} |
| 748 |
|
| 749 |
impl<S4, NI> BSwap for u64x2_sse2<YesS3, S4, NI> { |
| 750 |
#[inline(always)] |
| 751 |
fn bswap(self) -> Self { |
| 752 |
Self::new(unsafe { |
| 753 |
let k = _mm_set_epi64x(0x0809_0a0b_0c0d_0e0f, 0x0001_0203_0405_0607); |
| 754 |
_mm_shuffle_epi8(self.x, k) |
| 755 |
}) |
| 756 |
} |
| 757 |
} |
| 758 |
impl<S4, NI> BSwap for u64x2_sse2<NoS3, S4, NI> { |
| 759 |
#[inline(always)] |
| 760 |
fn bswap(self) -> Self { |
| 761 |
Self::new(unsafe { bswap32_s2(_mm_shuffle_epi32(self.x, 0b1011_0001)) }) |
| 762 |
} |
| 763 |
} |
| 764 |
|
| 765 |
impl<S4, NI> BSwap for u128x1_sse2<YesS3, S4, NI> { |
| 766 |
#[inline(always)] |
| 767 |
fn bswap(self) -> Self { |
| 768 |
Self::new(unsafe { |
| 769 |
let k = _mm_set_epi64x(0x0f0e_0d0c_0b0a_0908, 0x0706_0504_0302_0100); |
| 770 |
_mm_shuffle_epi8(self.x, k) |
| 771 |
}) |
| 772 |
} |
| 773 |
} |
| 774 |
impl<S4, NI> BSwap for u128x1_sse2<NoS3, S4, NI> { |
| 775 |
#[inline(always)] |
| 776 |
fn bswap(self) -> Self { |
| 777 |
unimplemented!() |
| 778 |
} |
| 779 |
} |
| 780 |
|
| 781 |
macro_rules! swapi { |
| 782 |
($x:expr, $i:expr, $k:expr) => { |
| 783 |
unsafe { |
| 784 |
const K: u8 = $k; |
| 785 |
let k = _mm_set1_epi8(K as i8); |
| 786 |
u128x1_sse2::new(_mm_or_si128( |
| 787 |
_mm_srli_epi16(_mm_and_si128($x.x, k), $i), |
| 788 |
_mm_and_si128(_mm_slli_epi16($x.x, $i), k), |
| 789 |
)) |
| 790 |
} |
| 791 |
}; |
| 792 |
} |
| 793 |
#[inline(always)] |
| 794 |
fn swap16_s2(x: __m128i) -> __m128i { |
| 795 |
unsafe { _mm_shufflehi_epi16(_mm_shufflelo_epi16(x, 0b1011_0001), 0b1011_0001) } |
| 796 |
} |
| 797 |
impl<S4, NI> Swap64 for u128x1_sse2<YesS3, S4, NI> { |
| 798 |
#[inline(always)] |
| 799 |
fn swap1(self) -> Self { |
| 800 |
swapi!(self, 1, 0xaa) |
| 801 |
} |
| 802 |
#[inline(always)] |
| 803 |
fn swap2(self) -> Self { |
| 804 |
swapi!(self, 2, 0xcc) |
| 805 |
} |
| 806 |
#[inline(always)] |
| 807 |
fn swap4(self) -> Self { |
| 808 |
swapi!(self, 4, 0xf0) |
| 809 |
} |
| 810 |
#[inline(always)] |
| 811 |
fn swap8(self) -> Self { |
| 812 |
u128x1_sse2::new(unsafe { |
| 813 |
let k = _mm_set_epi64x(0x0e0f_0c0d_0a0b_0809, 0x0607_0405_0203_0001); |
| 814 |
_mm_shuffle_epi8(self.x, k) |
| 815 |
}) |
| 816 |
} |
| 817 |
#[inline(always)] |
| 818 |
fn swap16(self) -> Self { |
| 819 |
u128x1_sse2::new(unsafe { |
| 820 |
let k = _mm_set_epi64x(0x0d0c_0f0e_0908_0b0a, 0x0504_0706_0100_0302); |
| 821 |
_mm_shuffle_epi8(self.x, k) |
| 822 |
}) |
| 823 |
} |
| 824 |
#[inline(always)] |
| 825 |
fn swap32(self) -> Self { |
| 826 |
u128x1_sse2::new(unsafe { _mm_shuffle_epi32(self.x, 0b1011_0001) }) |
| 827 |
} |
| 828 |
#[inline(always)] |
| 829 |
fn swap64(self) -> Self { |
| 830 |
u128x1_sse2::new(unsafe { _mm_shuffle_epi32(self.x, 0b0100_1110) }) |
| 831 |
} |
| 832 |
} |
| 833 |
impl<S4, NI> Swap64 for u128x1_sse2<NoS3, S4, NI> { |
| 834 |
#[inline(always)] |
| 835 |
fn swap1(self) -> Self { |
| 836 |
swapi!(self, 1, 0xaa) |
| 837 |
} |
| 838 |
#[inline(always)] |
| 839 |
fn swap2(self) -> Self { |
| 840 |
swapi!(self, 2, 0xcc) |
| 841 |
} |
| 842 |
#[inline(always)] |
| 843 |
fn swap4(self) -> Self { |
| 844 |
swapi!(self, 4, 0xf0) |
| 845 |
} |
| 846 |
#[inline(always)] |
| 847 |
fn swap8(self) -> Self { |
| 848 |
u128x1_sse2::new(unsafe { |
| 849 |
_mm_or_si128(_mm_slli_epi16(self.x, 8), _mm_srli_epi16(self.x, 8)) |
| 850 |
}) |
| 851 |
} |
| 852 |
#[inline(always)] |
| 853 |
fn swap16(self) -> Self { |
| 854 |
u128x1_sse2::new(swap16_s2(self.x)) |
| 855 |
} |
| 856 |
#[inline(always)] |
| 857 |
fn swap32(self) -> Self { |
| 858 |
u128x1_sse2::new(unsafe { _mm_shuffle_epi32(self.x, 0b1011_0001) }) |
| 859 |
} |
| 860 |
#[inline(always)] |
| 861 |
fn swap64(self) -> Self { |
| 862 |
u128x1_sse2::new(unsafe { _mm_shuffle_epi32(self.x, 0b0100_1110) }) |
| 863 |
} |
| 864 |
} |
| 865 |
|
| 866 |
#[derive(Copy, Clone)] |
| 867 |
pub struct G0; |
| 868 |
#[derive(Copy, Clone)] |
| 869 |
pub struct G1; |
| 870 |
|
| 871 |
#[allow(non_camel_case_types)] |
| 872 |
pub type u32x4x2_sse2<S3, S4, NI> = x2<u32x4_sse2<S3, S4, NI>, G0>; |
| 873 |
#[allow(non_camel_case_types)] |
| 874 |
pub type u64x2x2_sse2<S3, S4, NI> = x2<u64x2_sse2<S3, S4, NI>, G0>; |
| 875 |
#[allow(non_camel_case_types)] |
| 876 |
pub type u64x4_sse2<S3, S4, NI> = x2<u64x2_sse2<S3, S4, NI>, G1>; |
| 877 |
#[allow(non_camel_case_types)] |
| 878 |
pub type u128x2_sse2<S3, S4, NI> = x2<u128x1_sse2<S3, S4, NI>, G0>; |
| 879 |
|
| 880 |
#[allow(non_camel_case_types)] |
| 881 |
pub type u32x4x4_sse2<S3, S4, NI> = x4<u32x4_sse2<S3, S4, NI>>; |
| 882 |
#[allow(non_camel_case_types)] |
| 883 |
pub type u64x2x4_sse2<S3, S4, NI> = x4<u64x2_sse2<S3, S4, NI>>; |
| 884 |
#[allow(non_camel_case_types)] |
| 885 |
pub type u128x4_sse2<S3, S4, NI> = x4<u128x1_sse2<S3, S4, NI>>; |
| 886 |
|
| 887 |
impl<S3, S4, NI> Vector<[u32; 16]> for u32x4x4_sse2<S3, S4, NI> { |
| 888 |
#[inline(always)] |
| 889 |
fn to_scalars(self) -> [u32; 16] { |
| 890 |
transmute!(self) |
| 891 |
} |
| 892 |
} |
| 893 |
|
| 894 |
impl<S3: Copy, S4: Copy, NI: Copy> u32x4x2<Machine86<S3, S4, NI>> for u32x4x2_sse2<S3, S4, NI> |
| 895 |
where |
| 896 |
u32x4_sse2<S3, S4, NI>: RotateEachWord32 + BSwap, |
| 897 |
Machine86<S3, S4, NI>: Machine, |
| 898 |
u32x4x2_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u32x4; 2]>, |
| 899 |
u32x4x2_sse2<S3, S4, NI>: Vec2<<Machine86<S3, S4, NI> as Machine>::u32x4>, |
| 900 |
{ |
| 901 |
} |
| 902 |
impl<S3: Copy, S4: Copy, NI: Copy> u64x2x2<Machine86<S3, S4, NI>> for u64x2x2_sse2<S3, S4, NI> |
| 903 |
where |
| 904 |
u64x2_sse2<S3, S4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap, |
| 905 |
Machine86<S3, S4, NI>: Machine, |
| 906 |
u64x2x2_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u64x2; 2]>, |
| 907 |
u64x2x2_sse2<S3, S4, NI>: Vec2<<Machine86<S3, S4, NI> as Machine>::u64x2>, |
| 908 |
{ |
| 909 |
} |
| 910 |
impl<S3: Copy, S4: Copy, NI: Copy> u64x4<Machine86<S3, S4, NI>> for u64x4_sse2<S3, S4, NI> |
| 911 |
where |
| 912 |
u64x2_sse2<S3, S4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap, |
| 913 |
Machine86<S3, S4, NI>: Machine, |
| 914 |
u64x4_sse2<S3, S4, NI>: MultiLane<[u64; 4]> + Vec4<u64> + Words4, |
| 915 |
{ |
| 916 |
} |
| 917 |
impl<S3: Copy, S4: Copy, NI: Copy> u128x2<Machine86<S3, S4, NI>> for u128x2_sse2<S3, S4, NI> |
| 918 |
where |
| 919 |
u128x1_sse2<S3, S4, NI>: Swap64 + BSwap, |
| 920 |
Machine86<S3, S4, NI>: Machine, |
| 921 |
u128x2_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u128x1; 2]>, |
| 922 |
u128x2_sse2<S3, S4, NI>: Vec2<<Machine86<S3, S4, NI> as Machine>::u128x1>, |
| 923 |
u128x2_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u32x4x2>, |
| 924 |
u128x2_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u64x2x2>, |
| 925 |
u128x2_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u64x4>, |
| 926 |
{ |
| 927 |
} |
| 928 |
|
| 929 |
impl<NI: Copy> u32x4x2<Avx2Machine<NI>> for u32x4x2_sse2<YesS3, YesS4, NI> |
| 930 |
where |
| 931 |
u32x4_sse2<YesS3, YesS4, NI>: RotateEachWord32 + BSwap, |
| 932 |
Avx2Machine<NI>: Machine, |
| 933 |
u32x4x2_sse2<YesS3, YesS4, NI>: MultiLane<[<Avx2Machine<NI> as Machine>::u32x4; 2]>, |
| 934 |
u32x4x2_sse2<YesS3, YesS4, NI>: Vec2<<Avx2Machine<NI> as Machine>::u32x4>, |
| 935 |
{ |
| 936 |
} |
| 937 |
impl<NI: Copy> u64x2x2<Avx2Machine<NI>> for u64x2x2_sse2<YesS3, YesS4, NI> |
| 938 |
where |
| 939 |
u64x2_sse2<YesS3, YesS4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap, |
| 940 |
Avx2Machine<NI>: Machine, |
| 941 |
u64x2x2_sse2<YesS3, YesS4, NI>: MultiLane<[<Avx2Machine<NI> as Machine>::u64x2; 2]>, |
| 942 |
u64x2x2_sse2<YesS3, YesS4, NI>: Vec2<<Avx2Machine<NI> as Machine>::u64x2>, |
| 943 |
{ |
| 944 |
} |
| 945 |
impl<NI: Copy> u64x4<Avx2Machine<NI>> for u64x4_sse2<YesS3, YesS4, NI> |
| 946 |
where |
| 947 |
u64x2_sse2<YesS3, YesS4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap, |
| 948 |
Avx2Machine<NI>: Machine, |
| 949 |
u64x4_sse2<YesS3, YesS4, NI>: MultiLane<[u64; 4]> + Vec4<u64> + Words4, |
| 950 |
{ |
| 951 |
} |
| 952 |
impl<NI: Copy> u128x2<Avx2Machine<NI>> for u128x2_sse2<YesS3, YesS4, NI> |
| 953 |
where |
| 954 |
u128x1_sse2<YesS3, YesS4, NI>: Swap64 + BSwap, |
| 955 |
Avx2Machine<NI>: Machine, |
| 956 |
u128x2_sse2<YesS3, YesS4, NI>: MultiLane<[<Avx2Machine<NI> as Machine>::u128x1; 2]>, |
| 957 |
u128x2_sse2<YesS3, YesS4, NI>: Vec2<<Avx2Machine<NI> as Machine>::u128x1>, |
| 958 |
u128x2_sse2<YesS3, YesS4, NI>: Into<<Avx2Machine<NI> as Machine>::u32x4x2>, |
| 959 |
u128x2_sse2<YesS3, YesS4, NI>: Into<<Avx2Machine<NI> as Machine>::u64x2x2>, |
| 960 |
u128x2_sse2<YesS3, YesS4, NI>: Into<<Avx2Machine<NI> as Machine>::u64x4>, |
| 961 |
{ |
| 962 |
} |
| 963 |
|
| 964 |
impl<S3, S4, NI> Vec4<u64> for u64x4_sse2<S3, S4, NI> |
| 965 |
where |
| 966 |
u64x2_sse2<S3, S4, NI>: Copy + Vec2<u64>, |
| 967 |
{ |
| 968 |
#[inline(always)] |
| 969 |
fn extract(self, i: u32) -> u64 { |
| 970 |
match i { |
| 971 |
0 => self.0[0].extract(0), |
| 972 |
1 => self.0[0].extract(1), |
| 973 |
2 => self.0[1].extract(0), |
| 974 |
3 => self.0[1].extract(1), |
| 975 |
_ => panic!(), |
| 976 |
} |
| 977 |
} |
| 978 |
#[inline(always)] |
| 979 |
fn insert(mut self, w: u64, i: u32) -> Self { |
| 980 |
match i { |
| 981 |
0 => self.0[0] = self.0[0].insert(w, 0), |
| 982 |
1 => self.0[0] = self.0[0].insert(w, 1), |
| 983 |
2 => self.0[1] = self.0[1].insert(w, 0), |
| 984 |
3 => self.0[1] = self.0[1].insert(w, 1), |
| 985 |
_ => panic!(), |
| 986 |
}; |
| 987 |
self |
| 988 |
} |
| 989 |
} |
| 990 |
|
| 991 |
impl<S3: Copy, S4: Copy, NI: Copy> u32x4x4<Machine86<S3, S4, NI>> for u32x4x4_sse2<S3, S4, NI> |
| 992 |
where |
| 993 |
u32x4_sse2<S3, S4, NI>: RotateEachWord32 + BSwap, |
| 994 |
Machine86<S3, S4, NI>: Machine, |
| 995 |
u32x4x4_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u32x4; 4]>, |
| 996 |
u32x4x4_sse2<S3, S4, NI>: Vec4<<Machine86<S3, S4, NI> as Machine>::u32x4>, |
| 997 |
u32x4x4_sse2<S3, S4, NI>: Vec4Ext<<Machine86<S3, S4, NI> as Machine>::u32x4>, |
| 998 |
u32x4x4_sse2<S3, S4, NI>: Vector<[u32; 16]>, |
| 999 |
{ |
| 1000 |
} |
| 1001 |
impl<S3: Copy, S4: Copy, NI: Copy> u64x2x4<Machine86<S3, S4, NI>> for u64x2x4_sse2<S3, S4, NI> |
| 1002 |
where |
| 1003 |
u64x2_sse2<S3, S4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap, |
| 1004 |
Machine86<S3, S4, NI>: Machine, |
| 1005 |
u64x2x4_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u64x2; 4]>, |
| 1006 |
u64x2x4_sse2<S3, S4, NI>: Vec4<<Machine86<S3, S4, NI> as Machine>::u64x2>, |
| 1007 |
{ |
| 1008 |
} |
| 1009 |
impl<S3: Copy, S4: Copy, NI: Copy> u128x4<Machine86<S3, S4, NI>> for u128x4_sse2<S3, S4, NI> |
| 1010 |
where |
| 1011 |
u128x1_sse2<S3, S4, NI>: Swap64 + BSwap, |
| 1012 |
Machine86<S3, S4, NI>: Machine, |
| 1013 |
u128x4_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u128x1; 4]>, |
| 1014 |
u128x4_sse2<S3, S4, NI>: Vec4<<Machine86<S3, S4, NI> as Machine>::u128x1>, |
| 1015 |
u128x4_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u32x4x4>, |
| 1016 |
u128x4_sse2<S3, S4, NI>: Into<<Machine86<S3, S4, NI> as Machine>::u64x2x4>, |
| 1017 |
{ |
| 1018 |
} |
| 1019 |
|
| 1020 |
impl<NI: Copy> u64x2x4<Avx2Machine<NI>> for u64x2x4_sse2<YesS3, YesS4, NI> |
| 1021 |
where |
| 1022 |
u64x2_sse2<YesS3, YesS4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap, |
| 1023 |
Avx2Machine<NI>: Machine, |
| 1024 |
u64x2x4_sse2<YesS3, YesS4, NI>: MultiLane<[<Avx2Machine<NI> as Machine>::u64x2; 4]>, |
| 1025 |
u64x2x4_sse2<YesS3, YesS4, NI>: Vec4<<Avx2Machine<NI> as Machine>::u64x2>, |
| 1026 |
{ |
| 1027 |
} |
| 1028 |
impl<NI: Copy> u128x4<Avx2Machine<NI>> for u128x4_sse2<YesS3, YesS4, NI> |
| 1029 |
where |
| 1030 |
u128x1_sse2<YesS3, YesS4, NI>: Swap64 + BSwap, |
| 1031 |
Avx2Machine<NI>: Machine, |
| 1032 |
u128x4_sse2<YesS3, YesS4, NI>: MultiLane<[<Avx2Machine<NI> as Machine>::u128x1; 4]>, |
| 1033 |
u128x4_sse2<YesS3, YesS4, NI>: Vec4<<Avx2Machine<NI> as Machine>::u128x1>, |
| 1034 |
u128x4_sse2<YesS3, YesS4, NI>: Into<<Avx2Machine<NI> as Machine>::u32x4x4>, |
| 1035 |
u128x4_sse2<YesS3, YesS4, NI>: Into<<Avx2Machine<NI> as Machine>::u64x2x4>, |
| 1036 |
{ |
| 1037 |
} |
| 1038 |
|
| 1039 |
macro_rules! impl_into_x { |
| 1040 |
($from:ident, $to:ident) => { |
| 1041 |
impl<S3: Copy, S4: Copy, NI: Copy, Gf, Gt> From<x2<$from<S3, S4, NI>, Gf>> |
| 1042 |
for x2<$to<S3, S4, NI>, Gt> |
| 1043 |
{ |
| 1044 |
#[inline(always)] |
| 1045 |
fn from(x: x2<$from<S3, S4, NI>, Gf>) -> Self { |
| 1046 |
x2::new([$to::from(x.0[0]), $to::from(x.0[1])]) |
| 1047 |
} |
| 1048 |
} |
| 1049 |
impl<S3: Copy, S4: Copy, NI: Copy> From<x4<$from<S3, S4, NI>>> for x4<$to<S3, S4, NI>> { |
| 1050 |
#[inline(always)] |
| 1051 |
fn from(x: x4<$from<S3, S4, NI>>) -> Self { |
| 1052 |
x4::new([ |
| 1053 |
$to::from(x.0[0]), |
| 1054 |
$to::from(x.0[1]), |
| 1055 |
$to::from(x.0[2]), |
| 1056 |
$to::from(x.0[3]), |
| 1057 |
]) |
| 1058 |
} |
| 1059 |
} |
| 1060 |
}; |
| 1061 |
} |
| 1062 |
impl_into_x!(u128x1_sse2, u64x2_sse2); |
| 1063 |
impl_into_x!(u128x1_sse2, u32x4_sse2); |
| 1064 |
|
| 1065 |
///// Debugging |
| 1066 |
|
| 1067 |
use core::fmt::{Debug, Formatter, Result}; |
| 1068 |
|
| 1069 |
impl<W: PartialEq, G> PartialEq for x2<W, G> { |
| 1070 |
#[inline(always)] |
| 1071 |
fn eq(&self, rhs: &Self) -> bool { |
| 1072 |
self.0[0] == rhs.0[0] && self.0[1] == rhs.0[1] |
| 1073 |
} |
| 1074 |
} |
| 1075 |
|
| 1076 |
#[allow(unused)] |
| 1077 |
#[inline(always)] |
| 1078 |
unsafe fn eq128_s4(x: __m128i, y: __m128i) -> bool { |
| 1079 |
let q = _mm_shuffle_epi32(_mm_cmpeq_epi64(x, y), 0b1100_0110); |
| 1080 |
_mm_cvtsi128_si64(q) == -1 |
| 1081 |
} |
| 1082 |
|
| 1083 |
#[inline(always)] |
| 1084 |
unsafe fn eq128_s2(x: __m128i, y: __m128i) -> bool { |
| 1085 |
let q = _mm_cmpeq_epi32(x, y); |
| 1086 |
let p = _mm_cvtsi128_si64(_mm_srli_si128(q, 8)); |
| 1087 |
let q = _mm_cvtsi128_si64(q); |
| 1088 |
(p & q) == -1 |
| 1089 |
} |
| 1090 |
|
| 1091 |
impl<S3, S4, NI> PartialEq for u32x4_sse2<S3, S4, NI> { |
| 1092 |
#[inline(always)] |
| 1093 |
fn eq(&self, rhs: &Self) -> bool { |
| 1094 |
unsafe { eq128_s2(self.x, rhs.x) } |
| 1095 |
} |
| 1096 |
} |
| 1097 |
impl<S3, S4, NI> Debug for u32x4_sse2<S3, S4, NI> |
| 1098 |
where |
| 1099 |
Self: Copy + MultiLane<[u32; 4]>, |
| 1100 |
{ |
| 1101 |
#[cold] |
| 1102 |
fn fmt(&self, fmt: &mut Formatter) -> Result { |
| 1103 |
fmt.write_fmt(format_args!("{:08x?}", &self.to_lanes())) |
| 1104 |
} |
| 1105 |
} |
| 1106 |
|
| 1107 |
impl<S3, S4, NI> PartialEq for u64x2_sse2<S3, S4, NI> { |
| 1108 |
#[inline(always)] |
| 1109 |
fn eq(&self, rhs: &Self) -> bool { |
| 1110 |
unsafe { eq128_s2(self.x, rhs.x) } |
| 1111 |
} |
| 1112 |
} |
| 1113 |
impl<S3, S4, NI> Debug for u64x2_sse2<S3, S4, NI> |
| 1114 |
where |
| 1115 |
Self: Copy + MultiLane<[u64; 2]>, |
| 1116 |
{ |
| 1117 |
#[cold] |
| 1118 |
fn fmt(&self, fmt: &mut Formatter) -> Result { |
| 1119 |
fmt.write_fmt(format_args!("{:016x?}", &self.to_lanes())) |
| 1120 |
} |
| 1121 |
} |
| 1122 |
|
| 1123 |
impl<S3, S4, NI> Debug for u64x4_sse2<S3, S4, NI> |
| 1124 |
where |
| 1125 |
u64x2_sse2<S3, S4, NI>: Copy + MultiLane<[u64; 2]>, |
| 1126 |
{ |
| 1127 |
#[cold] |
| 1128 |
fn fmt(&self, fmt: &mut Formatter) -> Result { |
| 1129 |
let (a, b) = (self.0[0].to_lanes(), self.0[1].to_lanes()); |
| 1130 |
fmt.write_fmt(format_args!("{:016x?}", &[a[0], a[1], b[0], b[1]])) |
| 1131 |
} |
| 1132 |
} |
| 1133 |
|
| 1134 |
#[cfg(test)] |
| 1135 |
#[cfg(target_arch = "x86_64")] |
| 1136 |
mod test { |
| 1137 |
use super::*; |
| 1138 |
use crate::x86_64::{SSE2, SSE41, SSSE3}; |
| 1139 |
use crate::Machine; |
| 1140 |
|
| 1141 |
#[test] |
| 1142 |
#[cfg_attr(not(target_feature = "ssse3"), ignore)] |
| 1143 |
fn test_bswap32_s2_vs_s3() { |
| 1144 |
let xs = [0x0f0e_0d0c, 0x0b0a_0908, 0x0706_0504, 0x0302_0100]; |
| 1145 |
let ys = [0x0c0d_0e0f, 0x0809_0a0b, 0x0405_0607, 0x0001_0203]; |
| 1146 |
|
| 1147 |
let s2 = unsafe { SSE2::instance() }; |
| 1148 |
let s3 = unsafe { SSSE3::instance() }; |
| 1149 |
|
| 1150 |
let x_s2 = { |
| 1151 |
let x_s2: <SSE2 as Machine>::u32x4 = s2.vec(xs); |
| 1152 |
x_s2.bswap() |
| 1153 |
}; |
| 1154 |
|
| 1155 |
let x_s3 = { |
| 1156 |
let x_s3: <SSSE3 as Machine>::u32x4 = s3.vec(xs); |
| 1157 |
x_s3.bswap() |
| 1158 |
}; |
| 1159 |
|
| 1160 |
assert_eq!(x_s2, transmute!(x_s3)); |
| 1161 |
assert_eq!(x_s2, s2.vec(ys)); |
| 1162 |
} |
| 1163 |
|
| 1164 |
#[test] |
| 1165 |
#[cfg_attr(not(target_feature = "ssse3"), ignore)] |
| 1166 |
fn test_bswap64_s2_vs_s3() { |
| 1167 |
let xs = [0x0f0e_0d0c_0b0a_0908, 0x0706_0504_0302_0100]; |
| 1168 |
let ys = [0x0809_0a0b_0c0d_0e0f, 0x0001_0203_0405_0607]; |
| 1169 |
|
| 1170 |
let s2 = unsafe { SSE2::instance() }; |
| 1171 |
let s3 = unsafe { SSSE3::instance() }; |
| 1172 |
|
| 1173 |
let x_s2 = { |
| 1174 |
let x_s2: <SSE2 as Machine>::u64x2 = s2.vec(xs); |
| 1175 |
x_s2.bswap() |
| 1176 |
}; |
| 1177 |
|
| 1178 |
let x_s3 = { |
| 1179 |
let x_s3: <SSSE3 as Machine>::u64x2 = s3.vec(xs); |
| 1180 |
x_s3.bswap() |
| 1181 |
}; |
| 1182 |
|
| 1183 |
assert_eq!(x_s2, s2.vec(ys)); |
| 1184 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1185 |
} |
| 1186 |
|
| 1187 |
#[test] |
| 1188 |
#[cfg_attr(not(target_feature = "ssse3"), ignore)] |
| 1189 |
fn test_shuffle32_s2_vs_s3() { |
| 1190 |
let xs = [0x0, 0x1, 0x2, 0x3]; |
| 1191 |
let ys = [0x2, 0x3, 0x0, 0x1]; |
| 1192 |
let zs = [0x1, 0x2, 0x3, 0x0]; |
| 1193 |
|
| 1194 |
let s2 = unsafe { SSE2::instance() }; |
| 1195 |
let s3 = unsafe { SSSE3::instance() }; |
| 1196 |
|
| 1197 |
let x_s2 = { |
| 1198 |
let x_s2: <SSE2 as Machine>::u32x4 = s2.vec(xs); |
| 1199 |
x_s2.shuffle2301() |
| 1200 |
}; |
| 1201 |
let x_s3 = { |
| 1202 |
let x_s3: <SSSE3 as Machine>::u32x4 = s3.vec(xs); |
| 1203 |
x_s3.shuffle2301() |
| 1204 |
}; |
| 1205 |
assert_eq!(x_s2, s2.vec(ys)); |
| 1206 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1207 |
|
| 1208 |
let x_s2 = { |
| 1209 |
let x_s2: <SSE2 as Machine>::u32x4 = s2.vec(xs); |
| 1210 |
x_s2.shuffle3012() |
| 1211 |
}; |
| 1212 |
let x_s3 = { |
| 1213 |
let x_s3: <SSSE3 as Machine>::u32x4 = s3.vec(xs); |
| 1214 |
x_s3.shuffle3012() |
| 1215 |
}; |
| 1216 |
assert_eq!(x_s2, s2.vec(zs)); |
| 1217 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1218 |
|
| 1219 |
let x_s2 = x_s2.shuffle1230(); |
| 1220 |
let x_s3 = x_s3.shuffle1230(); |
| 1221 |
assert_eq!(x_s2, s2.vec(xs)); |
| 1222 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1223 |
} |
| 1224 |
|
| 1225 |
#[test] |
| 1226 |
#[cfg_attr(not(target_feature = "ssse3"), ignore)] |
| 1227 |
fn test_shuffle64_s2_vs_s3() { |
| 1228 |
let xs = [0x0, 0x1, 0x2, 0x3]; |
| 1229 |
let ys = [0x2, 0x3, 0x0, 0x1]; |
| 1230 |
let zs = [0x1, 0x2, 0x3, 0x0]; |
| 1231 |
|
| 1232 |
let s2 = unsafe { SSE2::instance() }; |
| 1233 |
let s3 = unsafe { SSSE3::instance() }; |
| 1234 |
|
| 1235 |
let x_s2 = { |
| 1236 |
let x_s2: <SSE2 as Machine>::u64x4 = s2.vec(xs); |
| 1237 |
x_s2.shuffle2301() |
| 1238 |
}; |
| 1239 |
let x_s3 = { |
| 1240 |
let x_s3: <SSSE3 as Machine>::u64x4 = s3.vec(xs); |
| 1241 |
x_s3.shuffle2301() |
| 1242 |
}; |
| 1243 |
assert_eq!(x_s2, s2.vec(ys)); |
| 1244 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1245 |
|
| 1246 |
let x_s2 = { |
| 1247 |
let x_s2: <SSE2 as Machine>::u64x4 = s2.vec(xs); |
| 1248 |
x_s2.shuffle3012() |
| 1249 |
}; |
| 1250 |
let x_s3 = { |
| 1251 |
let x_s3: <SSSE3 as Machine>::u64x4 = s3.vec(xs); |
| 1252 |
x_s3.shuffle3012() |
| 1253 |
}; |
| 1254 |
assert_eq!(x_s2, s2.vec(zs)); |
| 1255 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1256 |
|
| 1257 |
let x_s2 = x_s2.shuffle1230(); |
| 1258 |
let x_s3 = x_s3.shuffle1230(); |
| 1259 |
assert_eq!(x_s2, s2.vec(xs)); |
| 1260 |
assert_eq!(x_s3, transmute!(x_s3)); |
| 1261 |
} |
| 1262 |
|
| 1263 |
#[cfg_attr(not(all(target_feature = "ssse3", target_feature = "sse4.1")), ignore)] |
| 1264 |
#[test] |
| 1265 |
fn test_lanes_u32x4() { |
| 1266 |
let xs = [0x1, 0x2, 0x3, 0x4]; |
| 1267 |
|
| 1268 |
let s2 = unsafe { SSE2::instance() }; |
| 1269 |
let s3 = unsafe { SSSE3::instance() }; |
| 1270 |
let s4 = unsafe { SSE41::instance() }; |
| 1271 |
|
| 1272 |
{ |
| 1273 |
let x_s2: <SSE2 as Machine>::u32x4 = s2.vec(xs); |
| 1274 |
let y_s2 = <SSE2 as Machine>::u32x4::from_lanes(xs); |
| 1275 |
assert_eq!(x_s2, y_s2); |
| 1276 |
assert_eq!(xs, y_s2.to_lanes()); |
| 1277 |
} |
| 1278 |
|
| 1279 |
{ |
| 1280 |
let x_s3: <SSSE3 as Machine>::u32x4 = s3.vec(xs); |
| 1281 |
let y_s3 = <SSSE3 as Machine>::u32x4::from_lanes(xs); |
| 1282 |
assert_eq!(x_s3, y_s3); |
| 1283 |
assert_eq!(xs, y_s3.to_lanes()); |
| 1284 |
} |
| 1285 |
|
| 1286 |
{ |
| 1287 |
let x_s4: <SSE41 as Machine>::u32x4 = s4.vec(xs); |
| 1288 |
let y_s4 = <SSE41 as Machine>::u32x4::from_lanes(xs); |
| 1289 |
assert_eq!(x_s4, y_s4); |
| 1290 |
assert_eq!(xs, y_s4.to_lanes()); |
| 1291 |
} |
| 1292 |
} |
| 1293 |
|
| 1294 |
#[test] |
| 1295 |
#[cfg_attr(not(all(target_feature = "ssse3", target_feature = "sse4.1")), ignore)] |
| 1296 |
fn test_lanes_u64x2() { |
| 1297 |
let xs = [0x1, 0x2]; |
| 1298 |
|
| 1299 |
let s2 = unsafe { SSE2::instance() }; |
| 1300 |
let s3 = unsafe { SSSE3::instance() }; |
| 1301 |
let s4 = unsafe { SSE41::instance() }; |
| 1302 |
|
| 1303 |
{ |
| 1304 |
let x_s2: <SSE2 as Machine>::u64x2 = s2.vec(xs); |
| 1305 |
let y_s2 = <SSE2 as Machine>::u64x2::from_lanes(xs); |
| 1306 |
assert_eq!(x_s2, y_s2); |
| 1307 |
assert_eq!(xs, y_s2.to_lanes()); |
| 1308 |
} |
| 1309 |
|
| 1310 |
{ |
| 1311 |
let x_s3: <SSSE3 as Machine>::u64x2 = s3.vec(xs); |
| 1312 |
let y_s3 = <SSSE3 as Machine>::u64x2::from_lanes(xs); |
| 1313 |
assert_eq!(x_s3, y_s3); |
| 1314 |
assert_eq!(xs, y_s3.to_lanes()); |
| 1315 |
} |
| 1316 |
|
| 1317 |
{ |
| 1318 |
let x_s4: <SSE41 as Machine>::u64x2 = s4.vec(xs); |
| 1319 |
let y_s4 = <SSE41 as Machine>::u64x2::from_lanes(xs); |
| 1320 |
assert_eq!(x_s4, y_s4); |
| 1321 |
assert_eq!(xs, y_s4.to_lanes()); |
| 1322 |
} |
| 1323 |
} |
| 1324 |
|
| 1325 |
#[test] |
| 1326 |
fn test_vec4_u32x4_s2() { |
| 1327 |
let xs = [1, 2, 3, 4]; |
| 1328 |
let s2 = unsafe { SSE2::instance() }; |
| 1329 |
let x_s2: <SSE2 as Machine>::u32x4 = s2.vec(xs); |
| 1330 |
assert_eq!(x_s2.extract(0), 1); |
| 1331 |
assert_eq!(x_s2.extract(1), 2); |
| 1332 |
assert_eq!(x_s2.extract(2), 3); |
| 1333 |
assert_eq!(x_s2.extract(3), 4); |
| 1334 |
assert_eq!(x_s2.insert(0xf, 0), s2.vec([0xf, 2, 3, 4])); |
| 1335 |
assert_eq!(x_s2.insert(0xf, 1), s2.vec([1, 0xf, 3, 4])); |
| 1336 |
assert_eq!(x_s2.insert(0xf, 2), s2.vec([1, 2, 0xf, 4])); |
| 1337 |
assert_eq!(x_s2.insert(0xf, 3), s2.vec([1, 2, 3, 0xf])); |
| 1338 |
} |
| 1339 |
|
| 1340 |
#[test] |
| 1341 |
#[cfg_attr(not(all(target_feature = "ssse3", target_feature = "sse4.1")), ignore)] |
| 1342 |
fn test_vec4_u32x4_s4() { |
| 1343 |
let xs = [1, 2, 3, 4]; |
| 1344 |
let s4 = unsafe { SSE41::instance() }; |
| 1345 |
let x_s4: <SSE41 as Machine>::u32x4 = s4.vec(xs); |
| 1346 |
assert_eq!(x_s4.extract(0), 1); |
| 1347 |
assert_eq!(x_s4.extract(1), 2); |
| 1348 |
assert_eq!(x_s4.extract(2), 3); |
| 1349 |
assert_eq!(x_s4.extract(3), 4); |
| 1350 |
assert_eq!(x_s4.insert(0xf, 0), s4.vec([0xf, 2, 3, 4])); |
| 1351 |
assert_eq!(x_s4.insert(0xf, 1), s4.vec([1, 0xf, 3, 4])); |
| 1352 |
assert_eq!(x_s4.insert(0xf, 2), s4.vec([1, 2, 0xf, 4])); |
| 1353 |
assert_eq!(x_s4.insert(0xf, 3), s4.vec([1, 2, 3, 0xf])); |
| 1354 |
} |
| 1355 |
|
| 1356 |
#[test] |
| 1357 |
fn test_vec2_u64x2_s2() { |
| 1358 |
let xs = [0x1, 0x2]; |
| 1359 |
let s2 = unsafe { SSE2::instance() }; |
| 1360 |
let x_s2: <SSE2 as Machine>::u64x2 = s2.vec(xs); |
| 1361 |
assert_eq!(x_s2.extract(0), 1); |
| 1362 |
assert_eq!(x_s2.extract(1), 2); |
| 1363 |
assert_eq!(x_s2.insert(0xf, 0), s2.vec([0xf, 2])); |
| 1364 |
assert_eq!(x_s2.insert(0xf, 1), s2.vec([1, 0xf])); |
| 1365 |
} |
| 1366 |
|
| 1367 |
#[test] |
| 1368 |
#[cfg_attr(not(all(target_feature = "ssse3", target_feature = "sse4.1")), ignore)] |
| 1369 |
fn test_vec4_u64x2_s4() { |
| 1370 |
let xs = [0x1, 0x2]; |
| 1371 |
let s4 = unsafe { SSE41::instance() }; |
| 1372 |
let x_s4: <SSE41 as Machine>::u64x2 = s4.vec(xs); |
| 1373 |
assert_eq!(x_s4.extract(0), 1); |
| 1374 |
assert_eq!(x_s4.extract(1), 2); |
| 1375 |
assert_eq!(x_s4.insert(0xf, 0), s4.vec([0xf, 2])); |
| 1376 |
assert_eq!(x_s4.insert(0xf, 1), s4.vec([1, 0xf])); |
| 1377 |
} |
| 1378 |
} |
| 1379 |
|
| 1380 |
pub mod avx2 { |
| 1381 |
#![allow(non_camel_case_types)] |
| 1382 |
use crate::soft::{x2, x4}; |
| 1383 |
use crate::types::*; |
| 1384 |
use crate::x86_64::sse2::{u128x1_sse2, u32x4_sse2, G0}; |
| 1385 |
use crate::x86_64::{vec256_storage, vec512_storage, Avx2Machine, YesS3, YesS4}; |
| 1386 |
use core::arch::x86_64::*; |
| 1387 |
use core::marker::PhantomData; |
| 1388 |
use core::ops::*; |
| 1389 |
use zerocopy::transmute; |
| 1390 |
|
| 1391 |
zerocopy::cryptocorrosion_derive_traits! { |
| 1392 |
#[repr(transparent)] |
| 1393 |
#[derive(Copy, Clone)] |
| 1394 |
pub struct u32x4x2_avx2<NI> { |
| 1395 |
x: __m256i, |
| 1396 |
ni: PhantomData<NI>, |
| 1397 |
} |
| 1398 |
} |
| 1399 |
|
| 1400 |
impl<NI> u32x4x2_avx2<NI> { |
| 1401 |
#[inline(always)] |
| 1402 |
fn new(x: __m256i) -> Self { |
| 1403 |
Self { x, ni: PhantomData } |
| 1404 |
} |
| 1405 |
} |
| 1406 |
|
| 1407 |
impl<NI> u32x4x2<Avx2Machine<NI>> for u32x4x2_avx2<NI> where NI: Copy {} |
| 1408 |
impl<NI> Store<vec256_storage> for u32x4x2_avx2<NI> { |
| 1409 |
#[inline(always)] |
| 1410 |
unsafe fn unpack(p: vec256_storage) -> Self { |
| 1411 |
Self::new(p.avx) |
| 1412 |
} |
| 1413 |
} |
| 1414 |
impl<NI> StoreBytes for u32x4x2_avx2<NI> { |
| 1415 |
#[inline(always)] |
| 1416 |
unsafe fn unsafe_read_le(input: &[u8]) -> Self { |
| 1417 |
assert_eq!(input.len(), 32); |
| 1418 |
Self::new(_mm256_loadu_si256(input.as_ptr() as *const _)) |
| 1419 |
} |
| 1420 |
#[inline(always)] |
| 1421 |
unsafe fn unsafe_read_be(input: &[u8]) -> Self { |
| 1422 |
Self::unsafe_read_le(input).bswap() |
| 1423 |
} |
| 1424 |
#[inline(always)] |
| 1425 |
fn write_le(self, out: &mut [u8]) { |
| 1426 |
unsafe { |
| 1427 |
assert_eq!(out.len(), 32); |
| 1428 |
_mm256_storeu_si256(out.as_mut_ptr() as *mut _, self.x) |
| 1429 |
} |
| 1430 |
} |
| 1431 |
#[inline(always)] |
| 1432 |
fn write_be(self, out: &mut [u8]) { |
| 1433 |
self.bswap().write_le(out) |
| 1434 |
} |
| 1435 |
} |
| 1436 |
impl<NI> MultiLane<[u32x4_sse2<YesS3, YesS4, NI>; 2]> for u32x4x2_avx2<NI> { |
| 1437 |
#[inline(always)] |
| 1438 |
fn to_lanes(self) -> [u32x4_sse2<YesS3, YesS4, NI>; 2] { |
| 1439 |
unsafe { |
| 1440 |
[ |
| 1441 |
u32x4_sse2::new(_mm256_extracti128_si256(self.x, 0)), |
| 1442 |
u32x4_sse2::new(_mm256_extracti128_si256(self.x, 1)), |
| 1443 |
] |
| 1444 |
} |
| 1445 |
} |
| 1446 |
#[inline(always)] |
| 1447 |
fn from_lanes(x: [u32x4_sse2<YesS3, YesS4, NI>; 2]) -> Self { |
| 1448 |
Self::new(unsafe { _mm256_setr_m128i(x[0].x, x[1].x) }) |
| 1449 |
} |
| 1450 |
} |
| 1451 |
impl<NI> Vec2<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x2_avx2<NI> { |
| 1452 |
#[inline(always)] |
| 1453 |
fn extract(self, i: u32) -> u32x4_sse2<YesS3, YesS4, NI> { |
| 1454 |
unsafe { |
| 1455 |
match i { |
| 1456 |
0 => u32x4_sse2::new(_mm256_extracti128_si256(self.x, 0)), |
| 1457 |
1 => u32x4_sse2::new(_mm256_extracti128_si256(self.x, 1)), |
| 1458 |
_ => panic!(), |
| 1459 |
} |
| 1460 |
} |
| 1461 |
} |
| 1462 |
#[inline(always)] |
| 1463 |
fn insert(self, w: u32x4_sse2<YesS3, YesS4, NI>, i: u32) -> Self { |
| 1464 |
Self::new(unsafe { |
| 1465 |
match i { |
| 1466 |
0 => _mm256_inserti128_si256(self.x, w.x, 0), |
| 1467 |
1 => _mm256_inserti128_si256(self.x, w.x, 1), |
| 1468 |
_ => panic!(), |
| 1469 |
} |
| 1470 |
}) |
| 1471 |
} |
| 1472 |
} |
| 1473 |
impl<NI> BitOps32 for u32x4x2_avx2<NI> where NI: Copy {} |
| 1474 |
impl<NI> ArithOps for u32x4x2_avx2<NI> where NI: Copy {} |
| 1475 |
macro_rules! shuf_lane_bytes { |
| 1476 |
($name:ident, $k0:expr, $k1:expr) => { |
| 1477 |
#[inline(always)] |
| 1478 |
fn $name(self) -> Self { |
| 1479 |
Self::new(unsafe { |
| 1480 |
_mm256_shuffle_epi8(self.x, _mm256_set_epi64x($k0, $k1, $k0, $k1)) |
| 1481 |
}) |
| 1482 |
} |
| 1483 |
}; |
| 1484 |
} |
| 1485 |
macro_rules! rotr_32 { |
| 1486 |
($name:ident, $i:expr) => { |
| 1487 |
#[inline(always)] |
| 1488 |
fn $name(self) -> Self { |
| 1489 |
Self::new(unsafe { |
| 1490 |
_mm256_or_si256( |
| 1491 |
_mm256_srli_epi32(self.x, $i as i32), |
| 1492 |
_mm256_slli_epi32(self.x, 32 - $i as i32), |
| 1493 |
) |
| 1494 |
}) |
| 1495 |
} |
| 1496 |
}; |
| 1497 |
} |
| 1498 |
impl<NI: Copy> RotateEachWord32 for u32x4x2_avx2<NI> { |
| 1499 |
rotr_32!(rotate_each_word_right7, 7); |
| 1500 |
shuf_lane_bytes!( |
| 1501 |
rotate_each_word_right8, |
| 1502 |
0x0c0f_0e0d_080b_0a09, |
| 1503 |
0x0407_0605_0003_0201 |
| 1504 |
); |
| 1505 |
rotr_32!(rotate_each_word_right11, 11); |
| 1506 |
rotr_32!(rotate_each_word_right12, 12); |
| 1507 |
shuf_lane_bytes!( |
| 1508 |
rotate_each_word_right16, |
| 1509 |
0x0d0c_0f0e_0908_0b0a, |
| 1510 |
0x0504_0706_0100_0302 |
| 1511 |
); |
| 1512 |
rotr_32!(rotate_each_word_right20, 20); |
| 1513 |
shuf_lane_bytes!( |
| 1514 |
rotate_each_word_right24, |
| 1515 |
0x0e0d_0c0f_0a09_080b, |
| 1516 |
0x0605_0407_0201_0003 |
| 1517 |
); |
| 1518 |
rotr_32!(rotate_each_word_right25, 25); |
| 1519 |
} |
| 1520 |
impl<NI> BitOps0 for u32x4x2_avx2<NI> where NI: Copy {} |
| 1521 |
impl<NI> From<u32x4x2_avx2<NI>> for vec256_storage { |
| 1522 |
#[inline(always)] |
| 1523 |
fn from(x: u32x4x2_avx2<NI>) -> Self { |
| 1524 |
Self { avx: x.x } |
| 1525 |
} |
| 1526 |
} |
| 1527 |
|
| 1528 |
macro_rules! impl_assign { |
| 1529 |
($vec:ident, $Assign:ident, $assign_fn:ident, $bin_fn:ident) => { |
| 1530 |
impl<NI> $Assign for $vec<NI> |
| 1531 |
where |
| 1532 |
NI: Copy, |
| 1533 |
{ |
| 1534 |
#[inline(always)] |
| 1535 |
fn $assign_fn(&mut self, rhs: Self) { |
| 1536 |
*self = self.$bin_fn(rhs); |
| 1537 |
} |
| 1538 |
} |
| 1539 |
}; |
| 1540 |
} |
| 1541 |
impl_assign!(u32x4x2_avx2, BitXorAssign, bitxor_assign, bitxor); |
| 1542 |
impl_assign!(u32x4x2_avx2, BitOrAssign, bitor_assign, bitor); |
| 1543 |
impl_assign!(u32x4x2_avx2, BitAndAssign, bitand_assign, bitand); |
| 1544 |
impl_assign!(u32x4x2_avx2, AddAssign, add_assign, add); |
| 1545 |
|
| 1546 |
macro_rules! impl_bitop { |
| 1547 |
($vec:ident, $Op:ident, $op_fn:ident, $impl_fn:ident) => { |
| 1548 |
impl<NI> $Op for $vec<NI> { |
| 1549 |
type Output = Self; |
| 1550 |
#[inline(always)] |
| 1551 |
fn $op_fn(self, rhs: Self) -> Self::Output { |
| 1552 |
Self::new(unsafe { $impl_fn(self.x, rhs.x) }) |
| 1553 |
} |
| 1554 |
} |
| 1555 |
}; |
| 1556 |
} |
| 1557 |
impl_bitop!(u32x4x2_avx2, BitXor, bitxor, _mm256_xor_si256); |
| 1558 |
impl_bitop!(u32x4x2_avx2, BitOr, bitor, _mm256_or_si256); |
| 1559 |
impl_bitop!(u32x4x2_avx2, BitAnd, bitand, _mm256_and_si256); |
| 1560 |
impl_bitop!(u32x4x2_avx2, AndNot, andnot, _mm256_andnot_si256); |
| 1561 |
impl_bitop!(u32x4x2_avx2, Add, add, _mm256_add_epi32); |
| 1562 |
|
| 1563 |
impl<NI> Not for u32x4x2_avx2<NI> { |
| 1564 |
type Output = Self; |
| 1565 |
#[inline(always)] |
| 1566 |
fn not(self) -> Self::Output { |
| 1567 |
unsafe { |
| 1568 |
let f = _mm256_set1_epi8(-0x7f); |
| 1569 |
Self::new(f) ^ self |
| 1570 |
} |
| 1571 |
} |
| 1572 |
} |
| 1573 |
|
| 1574 |
impl<NI> BSwap for u32x4x2_avx2<NI> { |
| 1575 |
shuf_lane_bytes!(bswap, 0x0c0d_0e0f_0809_0a0b, 0x0405_0607_0001_0203); |
| 1576 |
} |
| 1577 |
|
| 1578 |
impl<NI> From<x2<u128x1_sse2<YesS3, YesS4, NI>, G0>> for u32x4x2_avx2<NI> |
| 1579 |
where |
| 1580 |
NI: Copy, |
| 1581 |
{ |
| 1582 |
#[inline(always)] |
| 1583 |
fn from(x: x2<u128x1_sse2<YesS3, YesS4, NI>, G0>) -> Self { |
| 1584 |
Self::new(unsafe { _mm256_setr_m128i(x.0[0].x, x.0[1].x) }) |
| 1585 |
} |
| 1586 |
} |
| 1587 |
|
| 1588 |
impl<NI> LaneWords4 for u32x4x2_avx2<NI> { |
| 1589 |
#[inline(always)] |
| 1590 |
fn shuffle_lane_words1230(self) -> Self { |
| 1591 |
Self::new(unsafe { _mm256_shuffle_epi32(self.x, 0b1001_0011) }) |
| 1592 |
} |
| 1593 |
#[inline(always)] |
| 1594 |
fn shuffle_lane_words2301(self) -> Self { |
| 1595 |
Self::new(unsafe { _mm256_shuffle_epi32(self.x, 0b0100_1110) }) |
| 1596 |
} |
| 1597 |
#[inline(always)] |
| 1598 |
fn shuffle_lane_words3012(self) -> Self { |
| 1599 |
Self::new(unsafe { _mm256_shuffle_epi32(self.x, 0b0011_1001) }) |
| 1600 |
} |
| 1601 |
} |
| 1602 |
|
| 1603 |
/////////////////////////////////////////////////////////////////////////////////////////// |
| 1604 |
|
| 1605 |
pub type u32x4x4_avx2<NI> = x2<u32x4x2_avx2<NI>, G0>; |
| 1606 |
impl<NI: Copy> u32x4x4<Avx2Machine<NI>> for u32x4x4_avx2<NI> {} |
| 1607 |
|
| 1608 |
impl<NI: Copy> Store<vec512_storage> for u32x4x4_avx2<NI> { |
| 1609 |
#[inline(always)] |
| 1610 |
unsafe fn unpack(p: vec512_storage) -> Self { |
| 1611 |
Self::new([ |
| 1612 |
u32x4x2_avx2::unpack(p.avx[0]), |
| 1613 |
u32x4x2_avx2::unpack(p.avx[1]), |
| 1614 |
]) |
| 1615 |
} |
| 1616 |
} |
| 1617 |
impl<NI: Copy> MultiLane<[u32x4_sse2<YesS3, YesS4, NI>; 4]> for u32x4x4_avx2<NI> { |
| 1618 |
#[inline(always)] |
| 1619 |
fn to_lanes(self) -> [u32x4_sse2<YesS3, YesS4, NI>; 4] { |
| 1620 |
let [a, b] = self.0[0].to_lanes(); |
| 1621 |
let [c, d] = self.0[1].to_lanes(); |
| 1622 |
[a, b, c, d] |
| 1623 |
} |
| 1624 |
#[inline(always)] |
| 1625 |
fn from_lanes(x: [u32x4_sse2<YesS3, YesS4, NI>; 4]) -> Self { |
| 1626 |
let ab = u32x4x2_avx2::from_lanes([x[0], x[1]]); |
| 1627 |
let cd = u32x4x2_avx2::from_lanes([x[2], x[3]]); |
| 1628 |
Self::new([ab, cd]) |
| 1629 |
} |
| 1630 |
} |
| 1631 |
impl<NI: Copy> Vec4<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x4_avx2<NI> { |
| 1632 |
#[inline(always)] |
| 1633 |
fn extract(self, i: u32) -> u32x4_sse2<YesS3, YesS4, NI> { |
| 1634 |
match i { |
| 1635 |
0 => self.0[0].extract(0), |
| 1636 |
1 => self.0[0].extract(1), |
| 1637 |
2 => self.0[1].extract(0), |
| 1638 |
3 => self.0[1].extract(1), |
| 1639 |
_ => panic!(), |
| 1640 |
} |
| 1641 |
} |
| 1642 |
#[inline(always)] |
| 1643 |
fn insert(self, w: u32x4_sse2<YesS3, YesS4, NI>, i: u32) -> Self { |
| 1644 |
Self::new(match i { |
| 1645 |
0 | 1 => [self.0[0].insert(w, i), self.0[1]], |
| 1646 |
2 | 3 => [self.0[0], self.0[1].insert(w, i - 2)], |
| 1647 |
_ => panic!(), |
| 1648 |
}) |
| 1649 |
} |
| 1650 |
} |
| 1651 |
impl<NI: Copy> Vec4Ext<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x4_avx2<NI> { |
| 1652 |
#[inline(always)] |
| 1653 |
fn transpose4(a: Self, b: Self, c: Self, d: Self) -> (Self, Self, Self, Self) { |
| 1654 |
/* |
| 1655 |
* a00:a01 a10:a11 |
| 1656 |
* b00:b01 b10:b11 |
| 1657 |
* c00:c01 c10:c11 |
| 1658 |
* d00:d01 d10:d11 |
| 1659 |
* => |
| 1660 |
* a00:b00 c00:d00 |
| 1661 |
* a01:b01 c01:d01 |
| 1662 |
* a10:b10 c10:d10 |
| 1663 |
* a11:b11 c11:d11 |
| 1664 |
*/ |
| 1665 |
unsafe { |
| 1666 |
let ab00 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[0].x, b.0[0].x, 0x20)); |
| 1667 |
let ab01 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[0].x, b.0[0].x, 0x31)); |
| 1668 |
let ab10 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[1].x, b.0[1].x, 0x20)); |
| 1669 |
let ab11 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[1].x, b.0[1].x, 0x31)); |
| 1670 |
let cd00 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[0].x, d.0[0].x, 0x20)); |
| 1671 |
let cd01 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[0].x, d.0[0].x, 0x31)); |
| 1672 |
let cd10 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[1].x, d.0[1].x, 0x20)); |
| 1673 |
let cd11 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[1].x, d.0[1].x, 0x31)); |
| 1674 |
( |
| 1675 |
Self::new([ab00, cd00]), |
| 1676 |
Self::new([ab01, cd01]), |
| 1677 |
Self::new([ab10, cd10]), |
| 1678 |
Self::new([ab11, cd11]), |
| 1679 |
) |
| 1680 |
} |
| 1681 |
} |
| 1682 |
} |
| 1683 |
impl<NI: Copy> Vector<[u32; 16]> for u32x4x4_avx2<NI> { |
| 1684 |
#[inline(always)] |
| 1685 |
fn to_scalars(self) -> [u32; 16] { |
| 1686 |
transmute!(self) |
| 1687 |
} |
| 1688 |
} |
| 1689 |
impl<NI: Copy> From<u32x4x4_avx2<NI>> for vec512_storage { |
| 1690 |
#[inline(always)] |
| 1691 |
fn from(x: u32x4x4_avx2<NI>) -> Self { |
| 1692 |
Self { |
| 1693 |
avx: [ |
| 1694 |
vec256_storage { avx: x.0[0].x }, |
| 1695 |
vec256_storage { avx: x.0[1].x }, |
| 1696 |
], |
| 1697 |
} |
| 1698 |
} |
| 1699 |
} |
| 1700 |
impl<NI: Copy> From<x4<u128x1_sse2<YesS3, YesS4, NI>>> for u32x4x4_avx2<NI> { |
| 1701 |
#[inline(always)] |
| 1702 |
fn from(x: x4<u128x1_sse2<YesS3, YesS4, NI>>) -> Self { |
| 1703 |
Self::new(unsafe { |
| 1704 |
[ |
| 1705 |
u32x4x2_avx2::new(_mm256_setr_m128i(x.0[0].x, x.0[1].x)), |
| 1706 |
u32x4x2_avx2::new(_mm256_setr_m128i(x.0[2].x, x.0[3].x)), |
| 1707 |
] |
| 1708 |
}) |
| 1709 |
} |
| 1710 |
} |
| 1711 |
} |