wide/
lib.rs

Help
1#![no_std]
2#![allow(non_camel_case_types)]
3#![warn(clippy::doc_markdown)]
4#![warn(clippy::missing_inline_in_public_items)]
5#![allow(clippy::eq_op)]
6#![allow(clippy::excessive_precision)]
7#![allow(clippy::let_and_return)]
8#![allow(clippy::unusual_byte_groupings)]
9#![allow(clippy::misrefactored_assign_op)]
10#![cfg_attr(test, allow(clippy::approx_constant))]
11
12//! A crate to help you go wide.
13//!
14//! This crate provides SIMD-compatible data types.
15//!
16//! When possible, explicit SIMD is used with all the math operations here. As a
17//! fallback, the fact that all the lengths of a fixed length array are doing
18//! the same thing will often make LLVM notice that it should use SIMD
19//! instructions to complete the task. In the worst case, the code just becomes
20//! totally scalar (though the math is still correct, at least).
21//!
22//! ## Crate Features
23//!
24//! * `std`: This causes the feature to link to `std`.
25//!   * Currently this just improves the performance of `sqrt` when an explicit
26//!     SIMD `sqrt` isn't available.
27
28// Note(Lokathor): Due to standard library magic, the std-only methods for f32
29// and f64 will automatically be available simply by declaring this.
30#[cfg(feature = "std")]
31extern crate std;
32
33// TODO
34// Add/Sub/Mul/Div with constant
35// Shuffle left/right/by index
36
37use core::{
38  fmt::{
39    Binary, Debug, Display, LowerExp, LowerHex, Octal, UpperExp, UpperHex,
40  },
41  ops::*,
42};
43
44#[allow(unused_imports)]
45use safe_arch::*;
46
47use bytemuck::*;
48
49#[cfg(feature = "serde")]
50use serde::{ser::SerializeTuple, Deserialize, Serialize};
51
52#[macro_use]
53mod macros;
54
55macro_rules! pick {
56  ($(if #[cfg($($test:meta),*)] {
57      $($if_tokens:tt)*
58    })else+ else {
59      $($else_tokens:tt)*
60    }) => {
61    pick!{
62      @__forests [ ] ;
63      $( [ {$($test),*} {$($if_tokens)*} ], )*
64      [ { } {$($else_tokens)*} ],
65    }
66  };
67  (if #[cfg($($if_meta:meta),*)] {
68      $($if_tokens:tt)*
69    } $(else if #[cfg($($else_meta:meta),*)] {
70      $($else_tokens:tt)*
71    })*) => {
72    pick!{
73      @__forests [ ] ;
74      [ {$($if_meta),*} {$($if_tokens)*} ],
75      $( [ {$($else_meta),*} {$($else_tokens)*} ], )*
76    }
77  };
78  (@__forests [$($not:meta,)*];) => {
79    /* halt expansion */
80  };
81  (@__forests [$($not:meta,)*]; [{$($m:meta),*} {$($tokens:tt)*}], $($rest:tt)*) => {
82    #[cfg(all( $($m,)* not(any($($not),*)) ))]
83    pick!{ @__identity $($tokens)* }
84    pick!{ @__forests [ $($not,)* $($m,)* ] ; $($rest)* }
85  };
86  (@__identity $($tokens:tt)*) => {
87    $($tokens)*
88  };
89}
90
91// TODO: make these generic over `mul_add`? Worth it?
92
93macro_rules! polynomial_2 {
94  ($x:expr, $c0:expr, $c1:expr, $c2:expr $(,)?) => {{
95    let x = $x;
96    let x2 = x * x;
97    x2.mul_add($c2, x.mul_add($c1, $c0))
98  }};
99}
100
101macro_rules! polynomial_3 {
102  ($x:expr, $c0:expr, $c1:expr, $c2:expr, $c3:expr $(,)?) => {{
103    let x = $x;
104    let x2 = x * x;
105    $c3.mul_add(x, $c2).mul_add(x2, $c1.mul_add(x, $c0))
106  }};
107}
108
109macro_rules! polynomial_4 {
110  ($x:expr, $c0:expr, $c1:expr, $c2:expr ,$c3:expr, $c4:expr $(,)?) => {{
111    let x = $x;
112    let x2 = x * x;
113    let x4 = x2 * x2;
114    $c3.mul_add(x, $c2).mul_add(x2, $c1.mul_add(x, $c0)) + $c4 * x4
115  }};
116}
117
118macro_rules! polynomial_5 {
119  ($x:expr, $c0:expr, $c1:expr, $c2:expr, $c3:expr, $c4:expr, $c5:expr $(,)?) => {{
120    let x = $x;
121    let x2 = x * x;
122    let x4 = x2 * x2;
123    $c3
124      .mul_add(x, $c2)
125      .mul_add(x2, $c5.mul_add(x, $c4).mul_add(x4, $c1.mul_add(x, $c0)))
126  }};
127}
128
129macro_rules! polynomial_5n {
130  ($x:expr, $c0:expr, $c1:expr, $c2:expr, $c3:expr, $c4:expr $(,)?) => {{
131    let x = $x;
132    let x2 = x * x;
133    let x4 = x2 * x2;
134    x2.mul_add(x.mul_add($c3, $c2), (x4.mul_add($c4 + x, x.mul_add($c1, $c0))))
135  }};
136}
137
138macro_rules! polynomial_6 {
139  ($x:expr, $c0:expr, $c1:expr, $c2:expr, $c3:expr, $c4:expr, $c5:expr ,$c6:expr $(,)?) => {{
140    let x = $x;
141    let x2 = x * x;
142    let x4 = x2 * x2;
143    x4.mul_add(
144      x2.mul_add($c6, x.mul_add($c5, $c4)),
145      x2.mul_add(x.mul_add($c3, $c2), x.mul_add($c1, $c0)),
146    )
147  }};
148}
149
150macro_rules! polynomial_6n {
151  ($x:expr, $c0:expr, $c1:expr, $c2:expr, $c3:expr, $c4:expr, $c5:expr $(,)?) => {{
152    let x = $x;
153    let x2 = x * x;
154    let x4 = x2 * x2;
155    x4.mul_add(
156      x.mul_add($c5, x2 + $c4),
157      x2.mul_add(x.mul_add($c3, $c2), x.mul_add($c1, $c0)),
158    )
159  }};
160}
161
162macro_rules! polynomial_8 {
163  ($x:expr, $c0:expr, $c1:expr, $c2:expr, $c3:expr, $c4:expr, $c5:expr,  $c6:expr, $c7:expr, $c8:expr $(,)?) => {{
164    let x = $x;
165    let x2 = x * x;
166    let x4 = x2 * x2;
167    let x8 = x4 * x4;
168    x4.mul_add(
169      x2.mul_add($c7.mul_add(x, $c6), x.mul_add($c5, $c4)),
170      x8.mul_add($c8, x2.mul_add(x.mul_add($c3, $c2), x.mul_add($c1, $c0))),
171    )
172  }};
173}
174
175macro_rules! polynomial_13 {
176  // calculates polynomial c13*x^13 + c12*x^12 + ... + c1*x + c0
177  ($x:expr,  $c2:expr, $c3:expr, $c4:expr, $c5:expr,$c6:expr, $c7:expr, $c8:expr,$c9:expr, $c10:expr, $c11:expr, $c12:expr, $c13:expr  $(,)?) => {{
178    let x = $x;
179    let x2 = x * x;
180    let x4 = x2 * x2;
181    let x8 = x4 * x4;
182    x8.mul_add(
183      x4.mul_add(
184        x.mul_add($c13, $c12),
185        x2.mul_add(x.mul_add($c11, $c10), x.mul_add($c9, $c8)),
186      ),
187      x4.mul_add(
188        x2.mul_add(x.mul_add($c7, $c6), x.mul_add($c5, $c4)),
189        x2.mul_add(x.mul_add($c3, $c2), x),
190      ),
191    )
192  }};
193}
194
195macro_rules! polynomial_13m {
196  // return  ((c8+c9*x) + (c10+c11*x)*x2 + (c12+c13*x)*x4)*x8 + (((c6+c7*x)*x2 +
197  // (c4+c5*x))*x4 + ((c2+c3*x)*x2 + x));
198  ($x:expr,  $c2:expr, $c3:expr, $c4:expr, $c5:expr,$c6:expr, $c7:expr, $c8:expr,$c9:expr, $c10:expr, $c11:expr, $c12:expr, $c13:expr  $(,)?) => {{
199    let x = $x;
200    let x2 = x * x;
201    let x4 = x2 * x2;
202    let x8 = x4 * x4;
203
204    x8.mul_add(
205      x4.mul_add(
206        x.mul_add($c13, $c12),
207        x2.mul_add(x.mul_add($c11, $c10), x.mul_add($c9, $c8)),
208      ),
209      x4.mul_add(
210        x2.mul_add(x.mul_add($c7, $c6), x.mul_add($c5, $c4)),
211        x2.mul_add(x.mul_add($c3, $c2), x),
212      ),
213    )
214  }};
215}
216
217mod f32x8_;
218pub use f32x8_::*;
219
220mod f32x4_;
221pub use f32x4_::*;
222
223mod f64x4_;
224pub use f64x4_::*;
225
226mod f64x2_;
227pub use f64x2_::*;
228
229mod i8x16_;
230pub use i8x16_::*;
231
232mod i16x16_;
233pub use i16x16_::*;
234
235mod i8x32_;
236pub use i8x32_::*;
237
238mod i16x8_;
239pub use i16x8_::*;
240
241mod i32x4_;
242pub use i32x4_::*;
243
244mod i32x8_;
245pub use i32x8_::*;
246
247mod i64x2_;
248pub use i64x2_::*;
249
250mod i64x4_;
251pub use i64x4_::*;
252
253mod u8x16_;
254pub use u8x16_::*;
255
256mod u8x32_;
257pub use u8x32_::*;
258
259mod u16x8_;
260pub use u16x8_::*;
261
262mod u16x16_;
263pub use u16x16_::*;
264
265mod u32x4_;
266pub use u32x4_::*;
267
268mod u32x8_;
269pub use u32x8_::*;
270
271mod u64x2_;
272pub use u64x2_::*;
273
274mod u64x4_;
275pub use u64x4_::*;
276
277#[allow(dead_code)]
278fn generic_bit_blend<T>(mask: T, y: T, n: T) -> T
279where
280  T: Copy + BitXor<Output = T> + BitAnd<Output = T>,
281{
282  n ^ ((n ^ y) & mask)
283}
284
285/// given `type.op(type)` and type is `Copy`, impls `type.op(&type)`
286macro_rules! bulk_impl_op_ref_self_for {
287  ($(($op:ident, $method:ident) => [$($t:ty),+]),+ $(,)?) => {
288    $( // do each trait/list matching given
289      $( // do the current trait for each type in its list.
290        impl $op<&Self> for $t {
291          type Output = Self;
292          #[inline]
293          #[must_use]
294          fn $method(self, rhs: &Self) -> Self::Output {
295            self.$method(*rhs)
296          }
297        }
298      )+
299    )+
300  };
301}
302
303bulk_impl_op_ref_self_for! {
304  (Add, add) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
305  (Sub, sub) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
306  (Mul, mul) => [f32x8, f32x4, f64x4, f64x2, i16x8, i16x16, i32x8, i32x4, u16x8, u16x16],
307  (Div, div) => [f32x8, f32x4, f64x4, f64x2],
308  (BitAnd, bitand) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16,u32x8, u32x4, u64x4, u64x2],
309  (BitOr, bitor) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
310  (BitXor, bitxor) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
311}
312
313/// given `type.op(rhs)` and type is Copy, impls `type.op_assign(rhs)`
314macro_rules! bulk_impl_op_assign_for {
315  ($(($op:ident<$rhs:ty>, $method:ident, $method_assign:ident) => [$($t:ty),+]),+ $(,)?) => {
316    $( // do each trait/list matching given
317      $( // do the current trait for each type in its list.
318        impl $op<$rhs> for $t {
319          #[inline]
320          fn $method_assign(&mut self, rhs: $rhs) {
321            *self = self.$method(rhs);
322          }
323        }
324      )+
325    )+
326  };
327}
328
329// Note: remember to update bulk_impl_op_ref_self_for first or this will give
330// weird errors!
331bulk_impl_op_assign_for! {
332  (AddAssign<Self>, add, add_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
333  (AddAssign<&Self>, add, add_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
334  (SubAssign<Self>, sub, sub_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
335  (SubAssign<&Self>, sub, sub_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x4, u64x2],
336  (MulAssign<Self>, mul, mul_assign) => [f32x8, f32x4, f64x4, f64x2, i16x8, i16x16, i32x8, i32x4, u16x8, u16x16],
337  (MulAssign<&Self>, mul, mul_assign) => [f32x8, f32x4, f64x4, f64x2, i16x8, i16x16, i32x8, i32x4, u16x8, u16x16],
338  (DivAssign<Self>, div, div_assign) => [f32x8, f32x4, f64x4, f64x2],
339  (DivAssign<&Self>, div, div_assign) => [f32x8, f32x4, f64x4, f64x2],
340  (BitAndAssign<Self>, bitand, bitand_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, u16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u32x8, u32x4, u64x4, u64x2],
341  (BitAndAssign<&Self>, bitand, bitand_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, u16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u32x8, u32x4, u64x4, u64x2],
342  (BitOrAssign<Self>, bitor, bitor_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, u16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u32x8, u32x4, u64x4, u64x2],
343  (BitOrAssign<&Self>, bitor, bitor_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, u16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u32x8, u32x4, u64x4, u64x2],
344  (BitXorAssign<Self>, bitxor, bitxor_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, u16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u32x8, u32x4, u64x4, u64x2],
345  (BitXorAssign<&Self>, bitxor, bitxor_assign) => [f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, u16x16, i32x8, i32x4, i64x2, u8x32, u8x16, u16x8, u32x8, u32x4, u64x4, u64x2],
346}
347
348macro_rules! impl_simple_neg {
349  ($($t:ty),+ $(,)?) => {
350    $(
351      impl Neg for $t {
352        type Output = Self;
353        #[inline]
354        #[must_use]
355        fn neg(self) -> Self::Output {
356          Self::default() - self
357        }
358      }
359      impl Neg for &'_ $t {
360        type Output = $t;
361        #[inline]
362        #[must_use]
363        fn neg(self) -> Self::Output {
364          <$t>::default() - *self
365        }
366      }
367    )+
368  };
369}
370
371impl_simple_neg! {
372  f32x8, f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x2, u64x4
373}
374
375// only works for 128 bit values
376macro_rules! impl_simple_not {
377  ($($t:ty),+ $(,)?) => {
378    $(
379      impl Not for $t {
380        type Output = Self;
381        #[inline]
382        #[must_use]
383        fn not(self) -> Self::Output {
384          self ^ cast::<u128, $t>(u128::MAX)
385        }
386      }
387      impl Not for &'_ $t {
388        type Output = $t;
389        #[inline]
390        #[must_use]
391        fn not(self) -> Self::Output {
392          *self ^ cast::<u128, $t>(u128::MAX)
393        }
394      }
395    )+
396  };
397}
398
399impl_simple_not! {
400  f32x4, i8x16, i16x8, i32x4, i64x2, u8x16, u16x8, u32x4, u64x2,
401}
402
403macro_rules! impl_simple_sum {
404  ($($t:ty),+ $(,)?) => {
405    $(
406      impl<RHS> core::iter::Sum<RHS> for $t where $t: AddAssign<RHS> {
407        #[inline]
408        fn sum<I: Iterator<Item = RHS>>(iter: I) -> Self {
409          let mut total = Self::zeroed();
410          for val in iter {
411            total += val;
412          }
413          total
414        }
415      }
416    )+
417  };
418}
419
420impl_simple_sum! {
421  f32x4, f64x4, f64x2, i8x32, i8x16, i16x8, i16x16, i32x8, i32x4, i64x4, i64x2, u8x32, u8x16, u16x8, u16x16, u32x8, u32x4, u64x2, u64x4
422}
423
424macro_rules! impl_floating_product {
425  ($($t:ty),+ $(,)?) => {
426    $(
427      impl<RHS> core::iter::Product<RHS> for $t where $t: MulAssign<RHS> {
428        #[inline]
429        fn product<I: Iterator<Item = RHS>>(iter: I) -> Self {
430          let mut total = Self::from(1.0);
431          for val in iter {
432            total *= val;
433          }
434          total
435        }
436      }
437    )+
438  };
439}
440
441impl_floating_product! {
442  f32x8, f32x4, f64x4, f64x2
443}
444
445macro_rules! impl_integer_product {
446  ($($t:ty),+ $(,)?) => {
447    $(
448      impl<RHS> core::iter::Product<RHS> for $t where $t: MulAssign<RHS> {
449        #[inline]
450        fn product<I: Iterator<Item = RHS>>(iter: I) -> Self {
451          let mut total = Self::from(1);
452          for val in iter {
453            total *= val;
454          }
455          total
456        }
457      }
458    )+
459  };
460}
461
462impl_integer_product! {
463  i16x8, i32x4, i32x8,
464}
465
466/// impls `From<a> for b` by just calling `cast`
467macro_rules! impl_from_a_for_b_with_cast {
468  ($(($arr:ty, $simd:ty)),+  $(,)?) => {
469    $(impl From<$arr> for $simd {
470      #[inline]
471      #[must_use]
472      fn from(arr: $arr) -> Self {
473        cast(arr)
474      }
475    }
476    impl From<$simd> for $arr {
477      #[inline]
478      #[must_use]
479      fn from(simd: $simd) -> Self {
480        cast(simd)
481      }
482    })+
483  };
484}
485
486impl_from_a_for_b_with_cast! {
487  ([f32;8], f32x8),
488  ([f32;4], f32x4), ([f64;4], f64x4), ([f64;2], f64x2),
489  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2), ([i64;4], i64x4),
490  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2), ([u64;4], u64x4),
491}
492
493macro_rules! impl_from_single_value {
494  ($(([$elem:ty;$len:expr], $simd:ty)),+  $(,)?) => {
495    $(impl From<$elem> for $simd {
496      /// Splats the single value given across all lanes.
497      #[inline]
498      #[must_use]
499      fn from(elem: $elem) -> Self {
500        cast([elem; $len])
501      }
502    }
503    impl $simd {
504      #[inline]
505      #[must_use]
506      pub fn splat(elem: $elem) -> $simd {
507        cast([elem; $len])
508      }
509    })+
510  };
511}
512
513impl_from_single_value! {
514  ([f32;8], f32x8),
515  ([f32;4], f32x4), ([f64;4], f64x4), ([f64;2], f64x2),
516  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2), ([i64;4], i64x4),
517  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2), ([u64;4], u64x4),
518}
519
520/// formatter => [(arr, simd)+],+
521macro_rules! impl_formatter_for {
522  ($($trait:ident => [$(($arr:ty, $simd:ty)),+]),+ $(,)?) => {
523    $( // do per trait
524      $( // do per simd type
525        impl $trait for $simd {
526          #[allow(clippy::missing_inline_in_public_items)]
527          fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result {
528            let a: $arr = cast(*self);
529            write!(f, "(")?;
530            for (x, a_ref) in a.iter().enumerate() {
531              if x > 0 {
532                write!(f, ", ")?;
533              }
534              $trait::fmt(a_ref, f)?;
535            }
536            write!(f, ")")
537          }
538        }
539      )+
540    )+
541  }
542}
543
544impl_formatter_for! {
545  Binary => [([u32;8], f32x8), ([u32;4], f32x4), ([u64;4], f64x4), ([u64;2], f64x2),
546  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
547  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
548  Debug => [([f32;8], f32x8), ([f32;4], f32x4), ([f64;4], f64x4), ([f64;2], f64x2),
549  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
550  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
551  Display => [([f32;8], f32x8), ([f32;4], f32x4), ([f64;4], f64x4), ([f64;2], f64x2),
552  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
553  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
554  LowerExp => [([f32;8], f32x8), ([f32;4], f32x4), ([u64;4], f64x4), ([u64;2], f64x2),
555  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
556  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
557  LowerHex => [([u32;8], f32x8), ([u32;4], f32x4), ([u64;4], f64x4), ([u64;2], f64x2),
558  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
559  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
560  Octal => [([u32;8], f32x8), ([u32;4], f32x4), ([u64;4], f64x4), ([u64;2], f64x2),
561  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
562  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
563  UpperExp => [([u32;8], f32x8), ([u32;4], f32x4), ([u64;4], f64x4), ([u64;2], f64x2),
564  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
565  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
566  UpperHex => [([u32;8], f32x8), ([u32;4], f32x4), ([u64;4], f64x4), ([u64;2], f64x2),
567  ([i8;32], i8x32), ([i8;16], i8x16), ([i16;8], i16x8), ([i16;16], i16x16), ([i32;8], i32x8), ([i32;4], i32x4), ([i64;2], i64x2),([i64;4], i64x4),
568  ([u8;32], u8x32), ([u8;16], u8x16), ([u16;8], u16x8), ([u16;16], u16x16), ([u32;8], u32x8), ([u32;4], u32x4), ([u64;2], u64x2),([u64;4], u64x4)],
569}
570
571// With const generics this could be simplified I hope
572macro_rules! from_array {
573  ($ty:ty,$dst:ty,$dst_wide:ident,32) => {
574    impl From<&[$ty]> for $dst_wide {
575      #[inline]
576      fn from(src: &[$ty]) -> $dst_wide {
577        match src.len() {
578          32 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst, src[26] as $dst, src[27] as $dst, src[28] as $dst, src[29] as $dst, src[30] as $dst, src[31] as $dst,]),
579          31 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst, src[26] as $dst, src[27] as $dst, src[28] as $dst, src[29] as $dst, src[30] as $dst,0 as $dst,]),
580          30 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst, src[26] as $dst, src[27] as $dst, src[28] as $dst, src[29] as $dst,0 as $dst,0 as $dst,]),
581          29 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst, src[26] as $dst, src[27] as $dst, src[28] as $dst,0 as $dst,0 as $dst,0 as $dst,]),
582          28 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst, src[26] as $dst, src[27] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
583          27 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst, src[26] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
584          26 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst, src[25] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
585          25 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst, src[24] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
586          24 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst, src[23] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
587          23 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst, src[22] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
588          22 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst, src[21] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
589          21 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst, src[20] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
590          20 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst, src[19] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
591          19 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst, src[18] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
592          18 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst, src[17] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
593          17 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst, src[16] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
594          16 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
595          15 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
596          14 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
597          13 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
598          12 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
599          11 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
600          10 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
601          9 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
602          8 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
603          7 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
604          6 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
605          5 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
606          4 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
607          3 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
608          2 => $dst_wide::from([src[0] as $dst, src[1] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
609          1 => $dst_wide::from([src[0] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
610          _ => panic!(
611            "Converting from an array larger than what can be stored in $dst_wide"
612          ),
613        }
614      }
615    }
616  };
617  ($ty:ty,$dst:ty,$dst_wide:ident,16) => {
618    impl From<&[$ty]> for $dst_wide {
619      #[inline]
620      fn from(src: &[$ty]) -> $dst_wide {
621        match src.len() {
622          16 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst, src[15] as $dst,]),
623          15 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst, src[14] as $dst,0 as $dst,]),
624          14 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst, src[13] as $dst,0 as $dst,0 as $dst,]),
625          13 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst, src[12] as $dst,0 as $dst,0 as $dst,0 as $dst,]),
626          12 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst, src[11] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
627          11 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst, src[10] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
628          10 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst, src[9] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
629          9 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst, src[8] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
630          8 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
631          7 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
632          6 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
633          5 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
634          4 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
635          3 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
636          2 => $dst_wide::from([src[0] as $dst, src[1] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
637          1 => $dst_wide::from([src[0] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
638          _ => panic!(
639            "Converting from an array larger than what can be stored in $dst_wide"
640          ),
641        }
642      }
643    }
644  };
645  ($ty:ty,$dst:ty,$dst_wide:ident,8) => {
646    impl From<&[$ty]> for $dst_wide {
647      #[inline]
648      fn from(src: &[$ty]) -> $dst_wide {
649        match src.len() {
650          8 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst, src[7] as $dst,]),
651          7 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst, src[6] as $dst,0 as $dst,]),
652          6 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst, src[5] as $dst,0 as $dst,0 as $dst,]),
653          5 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst, src[4] as $dst,0 as $dst,0 as $dst,0 as $dst,]),
654          4 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
655          3 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
656          2 => $dst_wide::from([src[0] as $dst, src[1] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
657          1 => $dst_wide::from([src[0] as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
658          0 => $dst_wide::from([0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,0 as $dst,]),
659          _ => panic!(
660            "Converting from an array larger than what can be stored in $dst_wide"
661          ),
662        }
663      }
664    }
665  };
666  ($ty:ty,$dst:ty,$dst_wide:ident,4) => {
667    impl From<&[$ty]> for $dst_wide {
668      #[inline]
669      fn from(src: &[$ty]) -> $dst_wide {
670        match src.len() {
671          4 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst, src[3] as $dst,]),
672          3 => $dst_wide::from([src[0] as $dst, src[1] as $dst, src[2] as $dst,0 as $dst,]),
673          2 => $dst_wide::from([src[0] as $dst, src[1] as $dst,0 as $dst,0 as $dst,]),
674          1 => $dst_wide::from([src[0] as $dst,0 as $dst,0 as $dst,0 as $dst,]),
675          _ => panic!(
676            "Converting from an array larger than what can be stored in $dst_wide"
677          ),
678        }
679      }
680    }
681  };
682}
683
684from_array!(i8, i8, i8x32, 32);
685from_array!(i8, i8, i8x16, 16);
686from_array!(i8, i32, i32x8, 8);
687from_array!(u8, u8, u8x16, 16);
688from_array!(u8, u8, u8x32, 32);
689from_array!(i16, i16, i16x16, 16);
690from_array!(u16, u16, u16x16, 16);
691from_array!(i32, i32, i32x8, 8);
692from_array!(f32, f32, f32x8, 8);
693from_array!(f32, f32, f32x4, 4);
694from_array!(f64, f64, f64x4, 4);
695from_array!(u64, u64, u64x4, 4);
696from_array!(i64, i64, i64x4, 4);
697
698#[allow(unused)]
699fn software_sqrt(x: f64) -> f64 {
700  use core::num::Wrapping;
701  type wu32 = Wrapping<u32>;
702  const fn w(u: u32) -> wu32 {
703    Wrapping(u)
704  }
705  let mut z: f64;
706  let sign: wu32 = w(0x80000000);
707  let mut ix0: i32;
708  let mut s0: i32;
709  let mut q: i32;
710  let mut m: i32;
711  let mut t: i32;
712  let mut i: i32;
713  let mut r: wu32;
714  let mut t1: wu32;
715  let mut s1: wu32;
716  let mut ix1: wu32;
717  let mut q1: wu32;
718  // extract data
719
720  pick! {
721    if #[cfg(target_endian = "little")]
722    {
723      let [low, high]: [u32; 2] = cast(x);
724      ix0 = high as i32;
725      ix1 = w(low);
726    }
727    else
728    {
729      let [high, low]: [u32; 2] = cast(x);
730      ix0 = high as i32;
731      ix1 = w(low);
732    }
733  }
734
735  // inf and nan
736  {
737    if x.is_nan() {
738      return f64::NAN;
739    }
740    if ix0 & 0x7ff00000 == 0x7ff00000 {
741      return x * x + x;
742    }
743  }
744  // handle zero
745  {
746    if ix0 <= 0 {
747      if ((ix0 & (!sign).0 as i32) | (ix1.0 as i32)) == 0 {
748        return x;
749      } else if ix0 < 0 {
750        return (x - x) / (x - x);
751      }
752    }
753  }
754  // normalize
755  {
756    m = ix0 >> 20;
757    if m == 0 {
758      // subnormal
759      while ix0 == 0 {
760        m -= 21;
761        ix0 |= (ix1 >> 11).0 as i32;
762        ix1 <<= 21;
763      }
764      i = 0;
765      while ix0 & 0x00100000 == 0 {
766        ix0 <<= 1;
767        i += 1;
768      }
769      m -= i - 1;
770      ix0 |= (ix1.0 >> (31 - i)) as i32;
771      ix1 <<= i as usize;
772    }
773    // un-bias exponent
774    m -= 1023;
775    ix0 = (ix0 & 0x000fffff) | 0x00100000;
776    if (m & 1) != 0 {
777      // odd m, double the input to make it even
778      ix0 += ix0 + ((ix1 & sign) >> 31).0 as i32;
779      ix1 += ix1;
780    }
781    m >>= 1;
782  }
783  // generate sqrt bit by bit
784  {
785    ix0 += ix0 + ((ix1 & sign) >> 31).0 as i32;
786    ix1 += ix1;
787    // q and q1 store the sqrt(x);
788    q = 0;
789    q1 = w(0);
790    s0 = 0;
791    s1 = w(0);
792    // our bit that moves from right to left
793    r = w(0x00200000);
794    while r != w(0) {
795      t = s0 + (r.0 as i32);
796      if t <= ix0 {
797        s0 = t + (r.0 as i32);
798        ix0 -= t;
799        q += (r.0 as i32);
800      }
801      ix0 += ix0 + ((ix1 & sign) >> 31).0 as i32;
802      ix1 += ix1;
803      r >>= 1;
804    }
805    r = sign;
806    while r != w(0) {
807      t1 = s1 + r;
808      t = s0;
809      if (t < ix0) || ((t == ix0) && (t1 <= ix1)) {
810        s1 = t1 + r;
811        if t1 & sign == sign && (s1 & sign) == w(0) {
812          s0 += 1;
813        }
814        ix0 -= t;
815        if ix1 < t1 {
816          ix0 -= 1;
817        }
818        ix1 -= t1;
819        q1 += r;
820      }
821      ix0 += ix0 + ((ix1 & sign) >> 31).0 as i32;
822      ix1 += ix1;
823      r >>= 1;
824    }
825  }
826  // use floating add to find out rounding direction
827  {
828    if ix0 | (ix1.0 as i32) != 0 {
829      z = 1.0 - 1.0e-300;
830      if z >= 1.0 {
831        z = 1.0 + 1.0e-300;
832        if q1 == w(0xffffffff) {
833          q1 = w(0);
834          q += 1;
835        } else if z > 1.0 {
836          if q1 == w(0xfffffffe) {
837            q += 1;
838          }
839          q1 += w(2);
840        } else {
841          q1 += q1 & w(1);
842        }
843      }
844    }
845  }
846  // finish up
847  ix0 = (q >> 1) + 0x3fe00000;
848  ix1 = q1 >> 1;
849  if q & 1 == 1 {
850    ix1 |= sign;
851  }
852  ix0 += m << 20;
853
854  pick! {
855    if #[cfg(target_endian = "little")]
856    {
857      cast::<[u32; 2], f64>([ix1.0, ix0 as u32])
858    }
859    else
860    {
861      cast::<[u32; 2], f64>([ix0 as u32, ix1.0])
862    }
863  }
864}
865
866#[test]
867fn test_software_sqrt() {
868  assert!(software_sqrt(f64::NAN).is_nan());
869  assert_eq!(software_sqrt(f64::INFINITY), f64::INFINITY);
870  assert_eq!(software_sqrt(0.0), 0.0);
871  assert_eq!(software_sqrt(-0.0), -0.0);
872  assert!(software_sqrt(-1.0).is_nan());
873  assert!(software_sqrt(f64::NEG_INFINITY).is_nan());
874  assert_eq!(software_sqrt(4.0), 2.0);
875  assert_eq!(software_sqrt(9.0), 3.0);
876  assert_eq!(software_sqrt(16.0), 4.0);
877  assert_eq!(software_sqrt(25.0), 5.0);
878  assert_eq!(software_sqrt(5000.0 * 5000.0), 5000.0);
879}
880
881pub trait CmpEq<Rhs = Self> {
882  type Output;
883  fn cmp_eq(self, rhs: Rhs) -> Self::Output;
884}
885
886pub trait CmpGt<Rhs = Self> {
887  type Output;
888  fn cmp_gt(self, rhs: Rhs) -> Self::Output;
889}
890
891pub trait CmpGe<Rhs = Self> {
892  type Output;
893  fn cmp_ge(self, rhs: Rhs) -> Self::Output;
894}
895
896pub trait CmpNe<Rhs = Self> {
897  type Output;
898  fn cmp_ne(self, rhs: Rhs) -> Self::Output;
899}
900
901pub trait CmpLt<Rhs = Self> {
902  type Output;
903  fn cmp_lt(self, rhs: Rhs) -> Self::Output;
904}
905
906pub trait CmpLe<Rhs = Self> {
907  type Output;
908  fn cmp_le(self, rhs: Rhs) -> Self::Output;
909}
910
911macro_rules! bulk_impl_const_rhs_op {
912  (($op:ident,$method:ident) => [$(($lhs:ty,$rhs:ty),)+]) => {
913    $(
914    impl $op<$rhs> for $lhs {
915      type Output = Self;
916      #[inline]
917      #[must_use]
918      fn $method(self, rhs: $rhs) -> Self::Output {
919        self.$method(<$lhs>::splat(rhs))
920      }
921    }
922    )+
923  };
924}
925
926bulk_impl_const_rhs_op!((CmpEq, cmp_eq) => [(f64x4, f64), (f64x2, f64), (f32x4,f32), (f32x8,f32),]);
927bulk_impl_const_rhs_op!((CmpLt, cmp_lt) => [(f64x4, f64), (f64x2, f64), (f32x4,f32), (f32x8,f32),]);
928bulk_impl_const_rhs_op!((CmpGt, cmp_gt) => [(f64x4, f64), (f64x2, f64), (f32x4,f32), (f32x8,f32),]);
929bulk_impl_const_rhs_op!((CmpNe, cmp_ne) => [(f64x4, f64), (f64x2, f64), (f32x4,f32), (f32x8,f32),]);
930bulk_impl_const_rhs_op!((CmpLe, cmp_le) => [(f64x4, f64), (f64x2, f64), (f32x4,f32), (f32x8,f32),]);
931bulk_impl_const_rhs_op!((CmpGe, cmp_ge) => [(f64x4, f64), (f64x2, f64), (f32x4,f32), (f32x8,f32),]);
932
933macro_rules! impl_serde {
934  ($i:ident, [$t:ty; $len:expr]) => {
935    #[cfg(feature = "serde")]
936    impl Serialize for $i {
937      #[inline]
938      fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
939      where
940        S: serde::Serializer,
941      {
942        let array = self.as_array_ref();
943        let mut seq = serializer.serialize_tuple($len)?;
944        for e in array {
945          seq.serialize_element(e)?;
946        }
947        seq.end()
948      }
949    }
950
951    #[cfg(feature = "serde")]
952    impl<'de> Deserialize<'de> for $i {
953      #[inline]
954      fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
955      where
956        D: serde::Deserializer<'de>,
957      {
958        Ok(<[$t; $len]>::deserialize(deserializer)?.into())
959      }
960    }
961  };
962}
963
964impl_serde!(f32x8, [f32; 8]);
965impl_serde!(f32x4, [f32; 4]);
966impl_serde!(f64x4, [f64; 4]);
967impl_serde!(f64x2, [f64; 2]);
968impl_serde!(i8x16, [i8; 16]);
969impl_serde!(i16x16, [i16; 16]);
970impl_serde!(i8x32, [i8; 32]);
971impl_serde!(i16x8, [i16; 8]);
972impl_serde!(i32x4, [i32; 4]);
973impl_serde!(i32x8, [i32; 8]);
974impl_serde!(i64x2, [i64; 2]);
975impl_serde!(i64x4, [i64; 4]);
976impl_serde!(u8x16, [u8; 16]);
977impl_serde!(u8x32, [u8; 32]);
978impl_serde!(u16x8, [u16; 8]);
979impl_serde!(u16x16, [u16; 16]);
980impl_serde!(u32x4, [u32; 4]);
981impl_serde!(u32x8, [u32; 8]);
982impl_serde!(u64x2, [u64; 2]);
983impl_serde!(u64x4, [u64; 4]);
wide/lib.rs

wide/
lib.rs
