1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579
use core::{mem::MaybeUninit, num::NonZeroUsize, ptr};
use awint_internals::*;
use const_fn::const_fn;
use crate::Bits;
/// Rotates a slice range so that `mid` is at element 0.
///
/// Rust's slice rotation function was highly optimized by me, but I had to copy
/// it from https://github.com/rust-lang/rust/blob/master/library/core/src/slice/mod.rs
/// and specialize it to work as `const`.
///
/// # Safety
///
/// The range `[mid-left, mid+right)` must be valid for reading and writing
#[const_fn(cfg(feature = "const_support"))]
const unsafe fn usize_rotate(mut left: usize, mut mid: *mut usize, mut right: usize) {
unsafe {
type BufType = [usize; 32];
loop {
if (right == 0) || (left == 0) {
return
}
if left + right < 24 {
// Algorithm 1
let x = mid.sub(left);
let mut tmp: usize = x.read();
let mut i = right;
let mut gcd = right;
loop {
let tmp_tmp = x.add(i).read();
x.add(i).write(tmp);
tmp = tmp_tmp;
if i >= left {
i -= left;
if i == 0 {
x.write(tmp);
break
}
if i < gcd {
gcd = i;
}
} else {
i += right;
}
}
const_for!(start in {1..gcd} {
tmp = x.add(start).read();
i = start + right;
loop {
let tmp_tmp = x.add(i).read();
x.add(i).write(tmp);
tmp = tmp_tmp;
if i >= left {
i -= left;
if i == start {
x.add(start).write(tmp);
break;
}
} else {
i += right;
}
}
});
return
// I have tested this with Miri to make sure it doesn't complain
} else if left <= 32 || right <= 32 {
// Algorithm 2
let mut rawarray = MaybeUninit::<(BufType, [usize; 0])>::uninit();
let buf = rawarray.as_mut_ptr() as *mut usize;
let dim = mid.sub(left).add(right);
if left <= right {
ptr::copy_nonoverlapping(mid.sub(left), buf, left);
ptr::copy(mid, mid.sub(left), right);
ptr::copy_nonoverlapping(buf, dim, left);
} else {
ptr::copy_nonoverlapping(mid, buf, right);
ptr::copy(mid.sub(left), dim, left);
ptr::copy_nonoverlapping(buf, mid.sub(left), right);
}
return
} else if left >= right {
// Algorithm 3
loop {
ptr::swap_nonoverlapping(mid.sub(right), mid, right);
mid = mid.sub(right);
left -= right;
if left < right {
break
}
}
} else {
// Algorithm 3, `left < right`
loop {
ptr::swap_nonoverlapping(mid.sub(left), mid, left);
mid = mid.add(left);
right -= left;
if right < left {
break
}
}
}
}
}
}
// `usize_rotate` has some unusually large thresholds for some branches that
// don't get tested well by Miri in fuzz.rs, so test them here
#[test]
fn usize_rotate_test() {
let mut buf = [123usize; 123];
for k in 0..123 {
unsafe { usize_rotate(k, buf.as_mut_ptr().add(k), 123 - k) }
}
}
/// # Bit permutation
impl Bits {
/// Shift-left-assigns at the digit level
#[inline]
#[const_fn(cfg(feature = "const_support"))]
pub(crate) const fn digit_shl_assign(&mut self, s: NonZeroUsize) {
// Should get optimized away when this function is inlined
assert!(s.get() < self.bw());
let s = digits(s);
if s == 0 {
return
}
unsafe {
// below performs this:
//const_for!(i in {s..self.len()}.rev() {
// *self.get_unchecked_mut(i) = self.get_unchecked(i - s);
//});
// If this shift overflows, it does not result in UB because it falls back to
// the more relaxed `ptr:copy`
if s.wrapping_shl(1) >= self.len() {
// We cannot call multiple `as_ptr` or `as_mut_ptr` at the same time, because
// they come from the same allocation. One would invalidate the other, and we
// would run into stacked borrow issues.
let ptr = self.as_mut_ptr();
ptr::copy_nonoverlapping(ptr, ptr.add(s), self.len() - s);
} else {
let ptr = self.as_mut_ptr();
ptr::copy(ptr, ptr.add(s), self.len() - s);
}
self.digit_set(false, 0..s, false);
}
}
/// Shift-left-assigns according to extra bits
#[inline]
#[const_fn(cfg(feature = "const_support"))]
pub(crate) const fn subdigit_shl_assign(&mut self, s: NonZeroUsize, clear_unused_bits: bool) {
let s = extra(s);
if s != 0 {
// TODO benchmark this strategy vs dual unroll
const_for!(i in {1..self.len()}.rev() {
unsafe {
*self.get_unchecked_mut(i) =
(self.get_unchecked(i - 1) >> (BITS - s)) | (self.get_unchecked(i) << s);
}
});
*self.first_mut() <<= s;
}
if clear_unused_bits {
self.clear_unused_bits();
}
}
/// Shift-right-assigns at the digit level
#[inline]
#[const_fn(cfg(feature = "const_support"))]
pub(crate) const fn digit_shr_assign(
&mut self,
s: NonZeroUsize,
extension: bool,
clear_unused_bits: bool,
) {
assert!(s.get() < self.bw());
let s = digits(s);
if s == 0 {
if clear_unused_bits {
self.clear_unused_bits();
}
return
}
unsafe {
// below performs this:
//const_for!(i in {s..self.len()} {
// *self.get_unchecked_mut(i - s) = self.get_unchecked(i);
//});
// If this shift overflows, it does not result in UB because it falls back to
// the more relaxed `ptr:copy`
if s.wrapping_shl(1) >= self.len() {
// We cannot call multiple `as_ptr` or `as_mut_ptr` at the same time, because
// they come from the same allocation. One would invalidate the other, and we
// would run into stacked borrow issues.
let ptr = self.as_mut_ptr();
ptr::copy_nonoverlapping(ptr.add(s), ptr, self.len() - s);
} else {
let ptr = self.as_mut_ptr();
ptr::copy(ptr.add(s), ptr, self.len() - s);
}
if extension && (self.unused() != 0) {
// Safety: There are fewer digit shifts than digits
*self.get_unchecked_mut(self.len() - 1 - s) |= MAX << (BITS - self.unused());
}
self.digit_set(extension, (self.len() - s)..self.len(), clear_unused_bits);
}
}
/// Shift-right-assigns according to extra bits
#[inline]
#[const_fn(cfg(feature = "const_support"))]
pub(crate) const fn subdigit_shr_assign(
&mut self,
s: NonZeroUsize,
extension: bool,
clear_unused_bits: bool,
) {
let s = extra(s);
if s == 0 {
if clear_unused_bits {
self.clear_unused_bits();
}
return
}
unsafe {
// TODO benchmark this strategy vs dual unroll
const_for!(i in {0..(self.len() - 1)} {
*self.get_unchecked_mut(i) =
(self.get_unchecked(i) >> s) | (self.get_unchecked(i + 1) << (BITS - s));
});
*self.last_mut() >>= s;
if extension {
if (s + self.unused()) > BITS {
*self.last_mut() = MAX;
// handle bits that get shifted into the next to last digit
// Safety: it is not possible to reach this unless there are enough bits for the
// shift which has to be less than the bitwidth
*self.get_unchecked_mut(self.len() - 2) |=
MAX << ((2 * BITS) - s - self.unused());
} else {
*self.last_mut() |= MAX << (BITS - s - self.unused());
}
}
if clear_unused_bits {
self.clear_unused_bits();
}
}
}
/// Left-shifts-assigns by `s` bits. If `s >= self.bw()`, then
/// `None` is returned and the `Bits` are left unchanged.
///
/// Left shifts can act as a very fast multiplication by a power of two for
/// both the signed and unsigned interpretation of `Bits`.
#[const_fn(cfg(feature = "const_support"))]
#[must_use]
pub const fn shl_assign(&mut self, s: usize) -> Option<()> {
match NonZeroUsize::new(s) {
None => Some(()),
Some(s) if s.get() < self.bw() => {
self.digit_shl_assign(s);
self.subdigit_shl_assign(s, true);
Some(())
}
_ => None,
}
}
/// Logically-right-shift-assigns by `s` bits. If `s >= self.bw()`, then
/// `None` is returned and the `Bits` are left unchanged.
///
/// Logical right shifts do not copy the sign bit, and thus can act as a
/// very fast floored division by a power of two for the unsigned
/// interpretation of `Bits`.
#[const_fn(cfg(feature = "const_support"))]
#[must_use]
pub const fn lshr_assign(&mut self, s: usize) -> Option<()> {
match NonZeroUsize::new(s) {
None => Some(()),
Some(s) if s.get() < self.bw() => {
self.digit_shr_assign(s, false, false);
self.subdigit_shr_assign(s, false, true);
Some(())
}
_ => None,
}
}
/// Arithmetically-right-shift-assigns by `s` bits. If `s >= self.bw()`,
/// then `None` is returned and the `Bits` are left unchanged.
///
/// Arithmetic right shifts copy the sign bit, and thus can act as a very
/// fast _floored_ division by a power of two for the signed interpretation
/// of `Bits`.
#[const_fn(cfg(feature = "const_support"))]
#[must_use]
pub const fn ashr_assign(&mut self, s: usize) -> Option<()> {
match NonZeroUsize::new(s) {
None => Some(()),
Some(s) if s.get() < self.bw() => {
let extension = self.msb();
self.digit_shr_assign(s, extension, false);
self.subdigit_shr_assign(s, extension, true);
Some(())
}
_ => None,
}
}
/// Left-rotate-assigns by `s` bits. If `s >= self.bw()`, then
/// `None` is returned and the `Bits` are left unchanged.
///
/// This function is equivalent to the following:
/// ```
/// use awint::prelude::*;
/// let mut input = inlawi!(0x4321u16);
/// let mut output = inlawi!(0u16);
/// // rotate left by 4 bits or one hexadecimal digit
/// let shift = 4;
///
/// // temporary clone of the input
/// let mut tmp = ExtAwi::from(input);
/// cc!(input; output).unwrap();
/// if shift != 0 {
/// if shift >= input.bw() {
/// // the actual function would return `None`
/// panic!();
/// }
/// output.shl_assign(shift).unwrap();
/// tmp.lshr_assign(input.bw() - shift).unwrap();
/// output.or_assign(&tmp);
/// };
///
/// assert_eq!(output, inlawi!(0x3214u16));
/// let mut using_rotate = ExtAwi::from(input);
/// using_rotate.rotl_assign(shift).unwrap();
/// assert_eq!(using_rotate, extawi!(0x3214u16));
///
/// // Note that slices are typed in a little-endian order opposite of
/// // how integers are typed, but they still visually rotate in the
/// // same way. This means `Rust`s built in slice rotation is in the
/// // opposite direction to integers and `Bits`
/// let mut array = [4, 3, 2, 1];
/// array.rotate_left(1);
/// assert_eq!(array, [3, 2, 1, 4]);
/// assert_eq!(0x4321u16.rotate_left(4), 0x3214);
/// let mut x = inlawi!(0x4321u16);
/// x.rotl_assign(4).unwrap();
/// // `Bits` has the preferred endianness
/// assert_eq!(x, inlawi!(0x3214u16));
/// ```
///
/// Unlike the example above which needs cloning, this function avoids any
/// allocation and has many optimized branches for different input sizes and
/// shifts.
#[const_fn(cfg(feature = "const_support"))]
#[must_use]
pub const fn rotl_assign(&mut self, s: usize) -> Option<()> {
match NonZeroUsize::new(s) {
None => Some(()),
Some(s) if s.get() < self.bw() => {
let x = self;
// fast path and simplifies other code paths
if x.len() == 1 {
*x.last_mut() = ((x.last() >> (x.bw() - s.get())) | (x.last() << s.get()))
& (MAX >> (BITS - x.bw()));
return Some(())
}
// TODO implement faster `subdigit_rotate_right` branch for certain cases
let digits = digits(s);
// Note: this is not a bitwidth but a shift
let s0 = extra(s);
let extra = x.extra();
let mid_digit = x.len() - digits;
let p = x.as_mut_ptr();
// Safety: this satisfies the requirements of `usize_rotate`
unsafe {
usize_rotate(mid_digit, p.add(mid_digit), digits);
}
if extra != 0 && digits != 0 {
// fix unused bits left in the middle from the rotation
let wrap = x.last() >> extra;
unsafe {
subdigits_mut!(x, 0..digits, y, {
y.subdigit_shl_assign(NonZeroUsize::new_unchecked(BITS - extra), false)
});
}
*x.first_mut() |= wrap;
x.clear_unused_bits();
}
if s0 != 0 {
// apply subdigit rotation
let wrap = if extra == 0 {
x.last() >> (BITS - s0)
} else if s0 <= extra {
x.last() >> (extra - s0)
} else {
// bits from the second to last digit get rotated all the way through the
// extra bits. We have already handled `x.len() == 1`.
unsafe {
(x.last() << (s0 - extra))
| (x.get_unchecked(x.len() - 2) >> (BITS - s0 + extra))
}
};
x.subdigit_shl_assign(s, true);
*x.first_mut() |= wrap;
}
Some(())
}
_ => None,
}
}
/// Right-rotate-assigns by `s` bits. If `s >= self.bw()`, then
/// `None` is returned and the `Bits` are left unchanged.
///
/// See `Bits::rotl_assign` for more details.
#[const_fn(cfg(feature = "const_support"))]
#[must_use]
pub const fn rotr_assign(&mut self, s: usize) -> Option<()> {
let bw = self.bw();
if s == 0 {
return Some(())
} else if s >= bw {
return None
}
self.rotl_assign(bw - s)
}
/// Reverse-bit-order-assigns `self`. The least significant bit becomes the
/// most significant bit, the second least significant bit becomes the
/// second most significant bit, etc.
#[const_fn(cfg(feature = "const_support"))]
pub const fn rev_assign(&mut self) {
let len = self.len();
if len == 1 {
*self.last_mut() = self.last().reverse_bits() >> self.unused();
return
}
let halfway = len >> 1;
let odd = (len & 1) != 0;
if self.extra() == 0 {
unsafe {
const_for!(i in {0..halfway} {
// swap opposite reversed digits until reaching the halfway point
let tmp = self.get_unchecked(i).reverse_bits();
*self.get_unchecked_mut(i) = self.get_unchecked_mut(len - 1 - i).reverse_bits();
*self.get_unchecked_mut(len - 1 - i) = tmp;
});
if odd {
// reverse the digit in the middle inplace
let tmp = self.get_unchecked_mut(halfway);
*tmp = tmp.reverse_bits();
}
}
} else {
if len == 2 {
let tmp0 = self.first().reverse_bits();
let tmp1 = self.last().reverse_bits();
*self.first_mut() = tmp1 >> self.unused() | tmp0 << self.extra();
*self.last_mut() = tmp0 >> self.unused();
return
}
unsafe {
let unused = self.unused();
let extra = self.extra();
// There are four temporaries, two starting in the least significant digits and
// two starting in the most significant digits. `tmp0` starts initialized with
// zero, so that the new unused bits are set to zero. If there are 7 digits,
// then the temporary assignments look like:
//
// tmp0=0 | .... | .... | .... | .... | .... | .... | tmp3 |
// tmp0=0 | tmp1 | .... | .... | .... | .... | tmp2 | tmp3 |
// | done | tmp0 | .... | .... | .... | tmp2 | done |
// | done | tmp0 | tmp1 | .... | tmp3 | tmp2 | done |
// | done | done | tmp1 | .... | tmp3 | done | done |
// | done | done | tmp1 |tmp0&2| tmp3 | done | done |
// | done | done | done |tmp0&2| done | done | done |
// | done | done | done |bridge| done | done | done |
let mut i0 = 0;
let mut tmp0 = 0;
let mut tmp1;
let mut i2 = len - 1;
let mut tmp2;
let mut tmp3 = self.get_unchecked(i2).reverse_bits();
loop {
if i0 == halfway {
// bridge between the converging indexes
if odd {
*self.get_unchecked_mut(i2) = (tmp3 >> unused) | (tmp0 << extra);
}
break
}
tmp1 = self.get_unchecked(i0).reverse_bits();
tmp2 = self.get_unchecked(i2 - 1).reverse_bits();
*self.get_unchecked_mut(i0) = (tmp3 >> unused) | (tmp2 << extra);
*self.get_unchecked_mut(i2) = (tmp1 >> unused) | (tmp0 << extra);
i0 += 1;
i2 -= 1;
if i0 == halfway {
if odd {
*self.get_unchecked_mut(i0) = (tmp2 >> unused) | (tmp1 << extra);
}
break
}
tmp0 = self.get_unchecked(i0).reverse_bits();
tmp3 = self.get_unchecked(i2 - 1).reverse_bits();
*self.get_unchecked_mut(i0) = (tmp2 >> unused) | (tmp3 << extra);
*self.get_unchecked_mut(i2) = (tmp0 >> unused) | (tmp1 << extra);
i0 += 1;
i2 -= 1;
}
}
}
}
/// Funnel shift with power-of-two bitwidths. Returns `None` if
/// `2*self.bw() != rhs.bw() || 2^s.bw() != self.bw()`. A `self.bw()` sized
/// field is assigned to `self` from `rhs` starting from the bit position
/// `s`. The shift cannot overflow because of the restriction on the
/// bitwidth of `s`.
///
/// ```
/// use awint::prelude::*;
/// let mut lhs = inlawi!(0xffff_ffffu32);
/// let mut rhs = inlawi!(0xfedc_ba98_7654_3210u64);
/// // `lhs.bw()` must be a power of two, `s.bw()` here is
/// // `log_2(32) == 5`. The value of `s` is set to what bit
/// // of `rhs` should be the starting bit for `lhs`.
/// let mut s = inlawi!(12u5);
/// lhs.funnel(&rhs, &s).unwrap();
/// assert_eq!(lhs, inlawi!(0xa9876543_u32))
/// ```
#[const_fn(cfg(feature = "const_support"))]
#[must_use]
pub const fn funnel(&mut self, rhs: &Self, s: &Self) -> Option<()> {
// because we later call `s.to_usize()` and assume it fits within `s.bw()`
s.assert_cleared_unused_bits();
// We avoid overflow by checking in this order and with `BITS - 1` instead of
// `BITS`
if (s.bw() >= (BITS - 1))
|| ((1usize << s.bw()) != self.bw())
|| ((self.bw() << 1) != rhs.bw())
{
return None
}
let s = s.to_usize();
let digits = digits_u(s);
let bits = extra_u(s);
if bits == 0 {
// Safety: there are two nonoverlapping `Bits`, and no out-of-bounds can occur
// because of strict checks
unsafe {
ptr::copy_nonoverlapping(rhs.as_ptr().add(digits), self.as_mut_ptr(), self.len());
}
} else if self.bw() < BITS {
*self.first_mut() = rhs.first() >> bits;
} else {
// Safety: When `self.bw() >= BITS`, `digits + i + 1` can be at most
// `self.len() + digits` which cannot reach `rhs.len()`.
unsafe {
const_for!(i in {0..self.len()} {
*self.get_unchecked_mut(i) = (rhs.get_unchecked(digits + i) >> bits)
| (rhs.get_unchecked(digits + i + 1) << (BITS - bits));
});
}
}
self.clear_unused_bits();
Some(())
}
}