1 //! Simple variable-time big integer implementation
4 use core::marker::PhantomData;
6 const WORD_COUNT_4096: usize = 4096 / 64;
7 const WORD_COUNT_256: usize = 256 / 64;
8 const WORD_COUNT_384: usize = 384 / 64;
10 // RFC 5702 indicates RSA keys can be up to 4096 bits
11 #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
12 pub(super) struct U4096([u64; WORD_COUNT_4096]);
14 #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
15 pub(super) struct U256([u64; WORD_COUNT_256]);
17 #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
18 pub(super) struct U384([u64; WORD_COUNT_384]);
20 pub(super) trait Int: Clone + Ord + Sized {
23 fn from_be_bytes(b: &[u8]) -> Result<Self, ()>;
24 fn limbs(&self) -> &[u64];
27 const ZERO: U256 = U256([0; 4]);
28 const BYTES: usize = 32;
29 fn from_be_bytes(b: &[u8]) -> Result<Self, ()> { Self::from_be_bytes(b) }
30 fn limbs(&self) -> &[u64] { &self.0 }
33 const ZERO: U384 = U384([0; 6]);
34 const BYTES: usize = 48;
35 fn from_be_bytes(b: &[u8]) -> Result<Self, ()> { Self::from_be_bytes(b) }
36 fn limbs(&self) -> &[u64] { &self.0 }
39 /// Defines a *PRIME* Modulus
40 pub(super) trait PrimeModulus<I: Int> {
42 const R_SQUARED_MOD_PRIME: I;
43 const NEGATIVE_PRIME_INV_MOD_R: I;
46 #[derive(Clone, Debug, PartialEq, Eq)] // Ord doesn't make sense cause we have an R factor
47 pub(super) struct U256Mod<M: PrimeModulus<U256>>(U256, PhantomData<M>);
49 #[derive(Clone, Debug, PartialEq, Eq)] // Ord doesn't make sense cause we have an R factor
50 pub(super) struct U384Mod<M: PrimeModulus<U384>>(U384, PhantomData<M>);
52 macro_rules! debug_unwrap { ($v: expr) => { {
54 debug_assert!(v.is_ok());
57 Err(e) => return Err(e),
61 // Various const versions of existing slice utilities
62 /// Const version of `&a[start..end]`
63 const fn const_subslice<'a, T>(a: &'a [T], start: usize, end: usize) -> &'a [T] {
64 assert!(start <= a.len());
65 assert!(end <= a.len());
66 assert!(end >= start);
67 let mut startptr = a.as_ptr();
68 startptr = unsafe { startptr.add(start) };
69 let len = end - start;
70 // The docs for from_raw_parts do not mention any requirements that the pointer be valid if the
71 // length is zero, aside from requiring proper alignment (which is met here). Thus,
72 // one-past-the-end should be an acceptable pointer for a 0-length slice.
73 unsafe { alloc::slice::from_raw_parts(startptr, len) }
76 /// Const version of `dest[dest_start..dest_end].copy_from_slice(source)`
78 /// Once `const_mut_refs` is stable we can convert this to a function
79 macro_rules! copy_from_slice {
80 ($dest: ident, $dest_start: expr, $dest_end: expr, $source: ident) => { {
81 let dest_start = $dest_start;
82 let dest_end = $dest_end;
83 assert!(dest_start <= $dest.len());
84 assert!(dest_end <= $dest.len());
85 assert!(dest_end >= dest_start);
86 assert!(dest_end - dest_start == $source.len());
88 while i < $source.len() {
89 $dest[i + dest_start] = $source[i];
95 /// Const version of a > b
96 const fn slice_greater_than(a: &[u64], b: &[u64]) -> bool {
97 debug_assert!(a.len() == b.len());
98 let len = if a.len() <= b.len() { a.len() } else { b.len() };
101 if a[i] > b[i] { return true; }
102 else if a[i] < b[i] { return false; }
108 /// Const version of a == b
109 const fn slice_equal(a: &[u64], b: &[u64]) -> bool {
110 debug_assert!(a.len() == b.len());
111 let len = if a.len() <= b.len() { a.len() } else { b.len() };
114 if a[i] != b[i] { return false; }
120 /// Adds a single u64 valuein-place, returning an overflow flag, in which case one out-of-bounds
121 /// high bit is implicitly included in the result.
123 /// Once `const_mut_refs` is stable we can convert this to a function
124 macro_rules! add_u64 { ($a: ident, $b: expr) => { {
129 let (v, carry) = $a[i].overflowing_add(add);
132 if add == 0 { break; }
140 /// Negates the given u64 slice.
142 /// Once `const_mut_refs` is stable we can convert this to a function
143 macro_rules! negate { ($v: ident) => { {
146 $v[i] ^= 0xffff_ffff_ffff_ffff;
152 /// Doubles in-place, returning an overflow flag, in which case one out-of-bounds high bit is
153 /// implicitly included in the result.
155 /// Once `const_mut_refs` is stable we can convert this to a function
156 macro_rules! double { ($a: ident) => { {
157 { let _: &[u64] = &$a; } // Force type resolution
159 let mut carry = false;
162 let next_carry = ($a[i] & (1 << 63)) != 0;
163 let (v, _next_carry_2) = ($a[i] << 1).overflowing_add(carry as u64);
165 debug_assert!(!_next_carry_2, "Adding one to 0x7ffff..*2 is only 0xffff..");
167 // Note that we can ignore _next_carry_2 here as we never need it - it cannot be set if
168 // next_carry is not set and at max 0xffff..*2 + 1 is only 0x1ffff.. (i.e. we can not need
179 macro_rules! define_add { ($name: ident, $len: expr) => {
180 /// Adds two $len-64-bit integers together, returning a new $len-64-bit integer and an overflow
181 /// bit, with the same semantics as the std [`u64::overflowing_add`] method.
182 const fn $name(a: &[u64], b: &[u64]) -> ([u64; $len], bool) {
183 debug_assert!(a.len() == $len);
184 debug_assert!(b.len() == $len);
185 let mut r = [0; $len];
186 let mut carry = false;
187 let mut i = $len - 1;
189 let (v, mut new_carry) = a[i].overflowing_add(b[i]);
190 let (v2, new_new_carry) = v.overflowing_add(carry as u64);
191 new_carry |= new_new_carry;
202 define_add!(add_2, 2);
203 define_add!(add_3, 3);
204 define_add!(add_4, 4);
205 define_add!(add_6, 6);
206 define_add!(add_8, 8);
207 define_add!(add_12, 12);
208 define_add!(add_16, 16);
209 define_add!(add_32, 32);
210 define_add!(add_64, 64);
211 define_add!(add_128, 128);
213 macro_rules! define_sub { ($name: ident, $name_abs: ident, $len: expr) => {
214 /// Subtracts the `b` $len-64-bit integer from the `a` $len-64-bit integer, returning a new
215 /// $len-64-bit integer and an overflow bit, with the same semantics as the std
216 /// [`u64::overflowing_sub`] method.
217 const fn $name(a: &[u64], b: &[u64]) -> ([u64; $len], bool) {
218 debug_assert!(a.len() == $len);
219 debug_assert!(b.len() == $len);
220 let mut r = [0; $len];
221 let mut carry = false;
222 let mut i = $len - 1;
224 let (v, mut new_carry) = a[i].overflowing_sub(b[i]);
225 let (v2, new_new_carry) = v.overflowing_sub(carry as u64);
226 new_carry |= new_new_carry;
236 /// Subtracts the `b` $len-64-bit integer from the `a` $len-64-bit integer, returning a new
237 /// $len-64-bit integer representing the absolute value of the result, as well as a sign bit.
239 const fn $name_abs(a: &[u64], b: &[u64]) -> ([u64; $len], bool) {
240 let (mut res, neg) = $name(a, b);
248 define_sub!(sub_2, sub_abs_2, 2);
249 define_sub!(sub_3, sub_abs_3, 3);
250 define_sub!(sub_4, sub_abs_4, 4);
251 define_sub!(sub_6, sub_abs_6, 6);
252 define_sub!(sub_8, sub_abs_8, 8);
253 define_sub!(sub_12, sub_abs_12, 12);
254 define_sub!(sub_16, sub_abs_16, 16);
255 define_sub!(sub_32, sub_abs_32, 32);
256 define_sub!(sub_64, sub_abs_64, 64);
257 define_sub!(sub_128, sub_abs_128, 128);
259 /// Multiplies two 128-bit integers together, returning a new 256-bit integer.
261 /// This is the base case for our multiplication, taking advantage of Rust's native 128-bit int
262 /// types to do multiplication (potentially) natively.
263 const fn mul_2(a: &[u64], b: &[u64]) -> [u64; 4] {
264 debug_assert!(a.len() == 2);
265 debug_assert!(b.len() == 2);
267 // Gradeschool multiplication is way faster here.
268 let (a0, a1) = (a[0] as u128, a[1] as u128);
269 let (b0, b1) = (b[0] as u128, b[1] as u128);
273 let (z1, i_carry) = z1i.overflowing_add(z1j);
276 let z2a = ((z2 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
277 let z1a = ((z1 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
278 let z0a = ((z0 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
279 let z2b = (z2 & 0xffff_ffff_ffff_ffff) as u64;
280 let z1b = (z1 & 0xffff_ffff_ffff_ffff) as u64;
281 let z0b = (z0 & 0xffff_ffff_ffff_ffff) as u64;
284 let (k, j_carry) = z0a.overflowing_add(z1b);
285 let (mut j, mut second_i_carry) = z1a.overflowing_add(z2b);
288 (j, new_i_carry) = j.overflowing_add(j_carry as u64);
289 debug_assert!(!second_i_carry || !new_i_carry);
290 second_i_carry |= new_i_carry;
293 let mut spurious_overflow;
294 (i, spurious_overflow) = i.overflowing_add(i_carry as u64);
295 debug_assert!(!spurious_overflow);
296 (i, spurious_overflow) = i.overflowing_add(second_i_carry as u64);
297 debug_assert!(!spurious_overflow);
302 const fn mul_3(a: &[u64], b: &[u64]) -> [u64; 6] {
303 debug_assert!(a.len() == 3);
304 debug_assert!(b.len() == 3);
306 let (a0, a1, a2) = (a[0] as u128, a[1] as u128, a[2] as u128);
307 let (b0, b1, b2) = (b[0] as u128, b[1] as u128, b[2] as u128);
319 let r5 = ((m4 >> 0) & 0xffff_ffff_ffff_ffff) as u64;
321 let r4a = ((m4 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
322 let r4b = ((m3a >> 0) & 0xffff_ffff_ffff_ffff) as u64;
323 let r4c = ((m3b >> 0) & 0xffff_ffff_ffff_ffff) as u64;
325 let r3a = ((m3a >> 64) & 0xffff_ffff_ffff_ffff) as u64;
326 let r3b = ((m3b >> 64) & 0xffff_ffff_ffff_ffff) as u64;
327 let r3c = ((m2a >> 0 ) & 0xffff_ffff_ffff_ffff) as u64;
328 let r3d = ((m2b >> 0 ) & 0xffff_ffff_ffff_ffff) as u64;
329 let r3e = ((m2c >> 0 ) & 0xffff_ffff_ffff_ffff) as u64;
331 let r2a = ((m2a >> 64) & 0xffff_ffff_ffff_ffff) as u64;
332 let r2b = ((m2b >> 64) & 0xffff_ffff_ffff_ffff) as u64;
333 let r2c = ((m2c >> 64) & 0xffff_ffff_ffff_ffff) as u64;
334 let r2d = ((m1a >> 0 ) & 0xffff_ffff_ffff_ffff) as u64;
335 let r2e = ((m1b >> 0 ) & 0xffff_ffff_ffff_ffff) as u64;
337 let r1a = ((m1a >> 64) & 0xffff_ffff_ffff_ffff) as u64;
338 let r1b = ((m1b >> 64) & 0xffff_ffff_ffff_ffff) as u64;
339 let r1c = ((m0 >> 0 ) & 0xffff_ffff_ffff_ffff) as u64;
341 let r0a = ((m0 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
343 let (r4, r3_ca) = r4a.overflowing_add(r4b);
344 let (r4, r3_cb) = r4.overflowing_add(r4c);
345 let r3_c = r3_ca as u64 + r3_cb as u64;
347 let (r3, r2_ca) = r3a.overflowing_add(r3b);
348 let (r3, r2_cb) = r3.overflowing_add(r3c);
349 let (r3, r2_cc) = r3.overflowing_add(r3d);
350 let (r3, r2_cd) = r3.overflowing_add(r3e);
351 let (r3, r2_ce) = r3.overflowing_add(r3_c);
352 let r2_c = r2_ca as u64 + r2_cb as u64 + r2_cc as u64 + r2_cd as u64 + r2_ce as u64;
354 let (r2, r1_ca) = r2a.overflowing_add(r2b);
355 let (r2, r1_cb) = r2.overflowing_add(r2c);
356 let (r2, r1_cc) = r2.overflowing_add(r2d);
357 let (r2, r1_cd) = r2.overflowing_add(r2e);
358 let (r2, r1_ce) = r2.overflowing_add(r2_c);
359 let r1_c = r1_ca as u64 + r1_cb as u64 + r1_cc as u64 + r1_cd as u64 + r1_ce as u64;
361 let (r1, r0_ca) = r1a.overflowing_add(r1b);
362 let (r1, r0_cb) = r1.overflowing_add(r1c);
363 let (r1, r0_cc) = r1.overflowing_add(r1_c);
364 let r0_c = r0_ca as u64 + r0_cb as u64 + r0_cc as u64;
366 let (r0, must_not_overflow) = r0a.overflowing_add(r0_c);
367 debug_assert!(!must_not_overflow);
369 [r0, r1, r2, r3, r4, r5]
372 macro_rules! define_mul { ($name: ident, $len: expr, $submul: ident, $add: ident, $subadd: ident, $sub: ident, $subsub: ident) => {
373 /// Multiplies two $len-64-bit integers together, returning a new $len*2-64-bit integer.
374 const fn $name(a: &[u64], b: &[u64]) -> [u64; $len * 2] {
375 // We could probably get a bit faster doing gradeschool multiplication for some smaller
376 // sizes, but its easier to just have one variable-length multiplication, so we do
377 // Karatsuba always here.
378 debug_assert!(a.len() == $len);
379 debug_assert!(b.len() == $len);
381 let a0 = const_subslice(a, 0, $len / 2);
382 let a1 = const_subslice(a, $len / 2, $len);
383 let b0 = const_subslice(b, 0, $len / 2);
384 let b1 = const_subslice(b, $len / 2, $len);
386 let z2 = $submul(a0, b0);
387 let z0 = $submul(a1, b1);
389 let (z1a_max, z1a_min, z1a_sign) =
390 if slice_greater_than(&a1, &a0) { (a1, a0, true) } else { (a0, a1, false) };
391 let (z1b_max, z1b_min, z1b_sign) =
392 if slice_greater_than(&b1, &b0) { (b1, b0, true) } else { (b0, b1, false) };
394 let z1a = $subsub(z1a_max, z1a_min);
395 debug_assert!(!z1a.1);
396 let z1b = $subsub(z1b_max, z1b_min);
397 debug_assert!(!z1b.1);
398 let z1m_sign = z1a_sign == z1b_sign;
400 let z1m = $submul(&z1a.0, &z1b.0);
401 let z1n = $add(&z0, &z2);
402 let mut z1_carry = z1n.1;
403 let z1 = if z1m_sign {
404 let r = $sub(&z1n.0, &z1m);
405 if r.1 { z1_carry ^= true; }
408 let r = $add(&z1n.0, &z1m);
409 if r.1 { z1_carry = true; }
413 let l = const_subslice(&z0, $len / 2, $len);
414 let (k, j_carry) = $subadd(const_subslice(&z0, 0, $len / 2), const_subslice(&z1, $len / 2, $len));
415 let (mut j, mut i_carry) = $subadd(const_subslice(&z1, 0, $len / 2), const_subslice(&z2, $len / 2, $len));
417 let new_i_carry = add_u64!(j, 1);
418 debug_assert!(!i_carry || !new_i_carry);
419 i_carry |= new_i_carry;
421 let mut i = [0; $len / 2];
422 let i_source = const_subslice(&z2, 0, $len / 2);
423 copy_from_slice!(i, 0, $len / 2, i_source);
425 let spurious_carry = add_u64!(i, 1);
426 debug_assert!(!spurious_carry);
429 let spurious_carry = add_u64!(i, 1);
430 debug_assert!(!spurious_carry);
433 let mut res = [0; $len * 2];
434 copy_from_slice!(res, $len * 2 * 0 / 4, $len * 2 * 1 / 4, i);
435 copy_from_slice!(res, $len * 2 * 1 / 4, $len * 2 * 2 / 4, j);
436 copy_from_slice!(res, $len * 2 * 2 / 4, $len * 2 * 3 / 4, k);
437 copy_from_slice!(res, $len * 2 * 3 / 4, $len * 2 * 4 / 4, l);
442 define_mul!(mul_4, 4, mul_2, add_4, add_2, sub_4, sub_2);
443 define_mul!(mul_6, 6, mul_3, add_6, add_3, sub_6, sub_3);
444 define_mul!(mul_8, 8, mul_4, add_8, add_4, sub_8, sub_4);
445 define_mul!(mul_16, 16, mul_8, add_16, add_8, sub_16, sub_8);
446 define_mul!(mul_32, 32, mul_16, add_32, add_16, sub_32, sub_16);
447 define_mul!(mul_64, 64, mul_32, add_64, add_32, sub_64, sub_32);
450 /// Squares a 128-bit integer, returning a new 256-bit integer.
452 /// This is the base case for our squaring, taking advantage of Rust's native 128-bit int
453 /// types to do multiplication (potentially) natively.
454 const fn sqr_2(a: &[u64]) -> [u64; 4] {
455 debug_assert!(a.len() == 2);
457 let (a0, a1) = (a[0] as u128, a[1] as u128);
459 let mut z1 = a0 * a1;
460 let i_carry = z1 & (1u128 << 127) != 0;
464 let z2a = ((z2 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
465 let z1a = ((z1 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
466 let z0a = ((z0 >> 64) & 0xffff_ffff_ffff_ffff) as u64;
467 let z2b = (z2 & 0xffff_ffff_ffff_ffff) as u64;
468 let z1b = (z1 & 0xffff_ffff_ffff_ffff) as u64;
469 let z0b = (z0 & 0xffff_ffff_ffff_ffff) as u64;
472 let (k, j_carry) = z0a.overflowing_add(z1b);
473 let (mut j, mut second_i_carry) = z1a.overflowing_add(z2b);
476 (j, new_i_carry) = j.overflowing_add(j_carry as u64);
477 debug_assert!(!second_i_carry || !new_i_carry);
478 second_i_carry |= new_i_carry;
481 let mut spurious_overflow;
482 (i, spurious_overflow) = i.overflowing_add(i_carry as u64);
483 debug_assert!(!spurious_overflow);
484 (i, spurious_overflow) = i.overflowing_add(second_i_carry as u64);
485 debug_assert!(!spurious_overflow);
490 macro_rules! define_sqr { ($name: ident, $len: expr, $submul: ident, $subsqr: ident, $subadd: ident) => {
491 /// Squares a $len-64-bit integers, returning a new $len*2-64-bit integer.
492 const fn $name(a: &[u64]) -> [u64; $len * 2] {
493 debug_assert!(a.len() == $len);
495 let hi = const_subslice(a, 0, $len / 2);
496 let lo = const_subslice(a, $len / 2, $len);
498 let v0 = $subsqr(lo);
499 let mut v1 = $submul(hi, lo);
500 let i_carry = double!(v1);
501 let v2 = $subsqr(hi);
503 let l = const_subslice(&v0, $len / 2, $len);
504 let (k, j_carry) = $subadd(const_subslice(&v0, 0, $len / 2), const_subslice(&v1, $len / 2, $len));
505 let (mut j, mut i_carry_2) = $subadd(const_subslice(&v1, 0, $len / 2), const_subslice(&v2, $len / 2, $len));
507 let mut i = [0; $len / 2];
508 let i_source = const_subslice(&v2, 0, $len / 2);
509 copy_from_slice!(i, 0, $len / 2, i_source);
512 let new_i_carry = add_u64!(j, 1);
513 debug_assert!(!i_carry_2 || !new_i_carry);
514 i_carry_2 |= new_i_carry;
517 let spurious_carry = add_u64!(i, 1);
518 debug_assert!(!spurious_carry);
521 let spurious_carry = add_u64!(i, 1);
522 debug_assert!(!spurious_carry);
525 let mut res = [0; $len * 2];
526 copy_from_slice!(res, $len * 2 * 0 / 4, $len * 2 * 1 / 4, i);
527 copy_from_slice!(res, $len * 2 * 1 / 4, $len * 2 * 2 / 4, j);
528 copy_from_slice!(res, $len * 2 * 2 / 4, $len * 2 * 3 / 4, k);
529 copy_from_slice!(res, $len * 2 * 3 / 4, $len * 2 * 4 / 4, l);
534 // TODO: Write an optimized sqr_3 (though secp384r1 is barely used)
535 const fn sqr_3(a: &[u64]) -> [u64; 6] { mul_3(a, a) }
537 define_sqr!(sqr_4, 4, mul_2, sqr_2, add_2);
538 define_sqr!(sqr_6, 6, mul_3, sqr_3, add_3);
539 define_sqr!(sqr_8, 8, mul_4, sqr_4, add_4);
540 define_sqr!(sqr_16, 16, mul_8, sqr_8, add_8);
541 define_sqr!(sqr_32, 32, mul_16, sqr_16, add_16);
542 define_sqr!(sqr_64, 64, mul_32, sqr_32, add_32);
544 macro_rules! dummy_pre_push { ($name: ident, $len: expr) => {} }
545 macro_rules! vec_pre_push { ($name: ident, $len: expr) => { $name.push([0; $len]); } }
547 macro_rules! define_div_rem { ($name: ident, $len: expr, $sub: ident, $heap_init: expr, $pre_push: ident $(, $const_opt: tt)?) => {
548 /// Divides two $len-64-bit integers, `a` by `b`, returning the quotient and remainder
550 /// Fails iff `b` is zero.
551 $($const_opt)? fn $name(a: &[u64; $len], b: &[u64; $len]) -> Result<([u64; $len], [u64; $len]), ()> {
552 if slice_equal(b, &[0; $len]) { return Err(()); }
555 let mut pow2s = $heap_init;
556 let mut pow2s_count = 0;
557 while slice_greater_than(a, &b_pow) {
558 $pre_push!(pow2s, $len);
559 pow2s[pow2s_count] = b_pow;
561 let double_overflow = double!(b_pow);
562 if double_overflow { break; }
564 let mut quot = [0; $len];
566 let mut pow2 = pow2s_count as isize - 1;
568 let b_pow = pow2s[pow2 as usize];
569 let overflow = double!(quot);
570 debug_assert!(!overflow);
571 if slice_greater_than(&rem, &b_pow) {
572 let (r, carry) = $sub(&rem, &b_pow);
573 debug_assert!(!carry);
579 if slice_equal(&rem, b) {
580 let overflow = add_u64!(quot, 1);
581 debug_assert!(!overflow);
582 Ok((quot, [0; $len]))
590 define_div_rem!(div_rem_2, 2, sub_2, [[0; 2]; 2 * 64], dummy_pre_push, const);
591 define_div_rem!(div_rem_4, 4, sub_4, [[0; 4]; 4 * 64], dummy_pre_push, const); // Uses 8 KiB of stack
592 define_div_rem!(div_rem_6, 6, sub_6, [[0; 6]; 6 * 64], dummy_pre_push, const); // Uses 18 KiB of stack!
593 #[cfg(debug_assertions)]
594 define_div_rem!(div_rem_8, 8, sub_8, [[0; 8]; 8 * 64], dummy_pre_push, const); // Uses 32 KiB of stack!
595 #[cfg(debug_assertions)]
596 define_div_rem!(div_rem_12, 12, sub_12, [[0; 12]; 12 * 64], dummy_pre_push, const); // Uses 72 KiB of stack!
597 define_div_rem!(div_rem_64, 64, sub_64, Vec::new(), vec_pre_push); // Uses up to 2 MiB of heap
598 #[cfg(debug_assertions)]
599 define_div_rem!(div_rem_128, 128, sub_128, Vec::new(), vec_pre_push); // Uses up to 8 MiB of heap
601 macro_rules! define_mod_inv { ($name: ident, $len: expr, $div: ident, $add: ident, $sub_abs: ident, $mul: ident) => {
602 /// Calculates the modular inverse of a $len-64-bit number with respect to the given modulus,
604 const fn $name(a: &[u64; $len], m: &[u64; $len]) -> Result<[u64; $len], ()> {
605 if slice_equal(a, &[0; $len]) || slice_equal(m, &[0; $len]) { return Err(()); }
607 let (mut s, mut old_s) = ([0; $len], [0; $len]);
612 let (mut old_s_neg, mut s_neg) = (false, false);
614 while !slice_equal(&r, &[0; $len]) {
615 let (quot, new_r) = debug_unwrap!($div(&old_r, &r));
617 let new_sa = $mul(", &s);
618 debug_assert!(slice_equal(const_subslice(&new_sa, 0, $len), &[0; $len]), "S overflowed");
619 let (new_s, new_s_neg) = match (old_s_neg, s_neg) {
621 let (new_s, overflow) = $add(&old_s, const_subslice(&new_sa, $len, new_sa.len()));
622 debug_assert!(!overflow);
626 let (new_s, overflow) = $add(&old_s, const_subslice(&new_sa, $len, new_sa.len()));
627 debug_assert!(!overflow);
631 let (new_s, overflow) = $add(&old_s, const_subslice(&new_sa, $len, new_sa.len()));
632 debug_assert!(!overflow);
635 (false, false) => $sub_abs(&old_s, const_subslice(&new_sa, $len, new_sa.len())),
647 // At this point old_r contains our GCD and old_s our first Bézout's identity coefficient.
648 if !slice_equal(const_subslice(&old_r, 0, $len - 1), &[0; $len - 1]) || old_r[$len - 1] != 1 {
651 debug_assert!(slice_greater_than(m, &old_s));
653 let (modinv, underflow) = $sub_abs(m, &old_s);
654 debug_assert!(!underflow);
655 debug_assert!(slice_greater_than(m, &modinv));
664 define_mod_inv!(mod_inv_2, 2, div_rem_2, add_2, sub_abs_2, mul_2);
665 define_mod_inv!(mod_inv_4, 4, div_rem_4, add_4, sub_abs_4, mul_4);
666 define_mod_inv!(mod_inv_6, 6, div_rem_6, add_6, sub_abs_6, mul_6);
668 define_mod_inv!(mod_inv_8, 8, div_rem_8, add_8, sub_abs_8, mul_8);
671 /// Constructs a new [`U4096`] from a variable number of big-endian bytes.
672 pub(super) fn from_be_bytes(bytes: &[u8]) -> Result<U4096, ()> {
673 if bytes.len() > 4096/8 { return Err(()); }
674 let u64s = (bytes.len() + 7) / 8;
675 let mut res = [0; WORD_COUNT_4096];
678 let pos = (u64s - i) * 8;
679 let start = bytes.len().saturating_sub(pos);
680 let end = bytes.len() + 8 - pos;
681 b[8 + start - end..].copy_from_slice(&bytes[start..end]);
682 res[i + WORD_COUNT_4096 - u64s] = u64::from_be_bytes(b);
687 /// Naively multiplies `self` * `b` mod `m`, returning a new [`U4096`].
689 /// Fails iff m is 0 or self or b are greater than m.
690 #[cfg(debug_assertions)]
691 fn mulmod_naive(&self, b: &U4096, m: &U4096) -> Result<U4096, ()> {
692 if m.0 == [0; WORD_COUNT_4096] { return Err(()); }
693 if self > m || b > m { return Err(()); }
695 let mul = mul_64(&self.0, &b.0);
697 let mut m_zeros = [0; 128];
698 m_zeros[WORD_COUNT_4096..].copy_from_slice(&m.0);
699 let (_, rem) = div_rem_128(&mul, &m_zeros)?;
700 let mut res = [0; WORD_COUNT_4096];
701 debug_assert_eq!(&rem[..WORD_COUNT_4096], &[0; WORD_COUNT_4096]);
702 res.copy_from_slice(&rem[WORD_COUNT_4096..]);
706 /// Calculates `self` ^ `exp` mod `m`, returning a new [`U4096`].
708 /// Fails iff m is 0, even, or self or b are greater than m.
709 pub(super) fn expmod_odd_mod(&self, mut exp: u32, m: &U4096) -> Result<U4096, ()> {
710 #![allow(non_camel_case_types)]
712 if m.0 == [0; WORD_COUNT_4096] { return Err(()); }
713 if m.0[WORD_COUNT_4096 - 1] & 1 == 0 { return Err(()); }
714 if self > m { return Err(()); }
716 let mut t = [0; WORD_COUNT_4096];
717 if &m.0[..WORD_COUNT_4096 - 1] == &[0; WORD_COUNT_4096 - 1] && m.0[WORD_COUNT_4096 - 1] == 1 {
720 t[WORD_COUNT_4096 - 1] = 1;
721 if exp == 0 { return Ok(U4096(t)); }
723 // Because m is not even, using 2^4096 as the Montgomery R value is always safe - it is
724 // guaranteed to be co-prime with any non-even integer.
726 type mul_ty = fn(&[u64], &[u64]) -> [u64; WORD_COUNT_4096 * 2];
727 type sqr_ty = fn(&[u64]) -> [u64; WORD_COUNT_4096 * 2];
728 type add_double_ty = fn(&[u64], &[u64]) -> ([u64; WORD_COUNT_4096 * 2], bool);
729 type sub_ty = fn(&[u64], &[u64]) -> ([u64; WORD_COUNT_4096], bool);
730 let (word_count, log_bits, mul, sqr, add_double, sub) =
731 if m.0[..WORD_COUNT_4096 / 2] == [0; WORD_COUNT_4096 / 2] {
732 if m.0[..WORD_COUNT_4096 * 3 / 4] == [0; WORD_COUNT_4096 * 3 / 4] {
733 fn mul_16_subarr(a: &[u64], b: &[u64]) -> [u64; WORD_COUNT_4096 * 2] {
734 debug_assert_eq!(a.len(), WORD_COUNT_4096);
735 debug_assert_eq!(b.len(), WORD_COUNT_4096);
736 debug_assert_eq!(&a[..WORD_COUNT_4096 * 3 / 4], &[0; WORD_COUNT_4096 * 3 / 4]);
737 debug_assert_eq!(&b[..WORD_COUNT_4096 * 3 / 4], &[0; WORD_COUNT_4096 * 3 / 4]);
738 let mut res = [0; WORD_COUNT_4096 * 2];
739 res[WORD_COUNT_4096 + WORD_COUNT_4096 / 2..].copy_from_slice(
740 &mul_16(&a[WORD_COUNT_4096 * 3 / 4..], &b[WORD_COUNT_4096 * 3 / 4..]));
743 fn sqr_16_subarr(a: &[u64]) -> [u64; WORD_COUNT_4096 * 2] {
744 debug_assert_eq!(a.len(), WORD_COUNT_4096);
745 debug_assert_eq!(&a[..WORD_COUNT_4096 * 3 / 4], &[0; WORD_COUNT_4096 * 3 / 4]);
746 let mut res = [0; WORD_COUNT_4096 * 2];
747 res[WORD_COUNT_4096 + WORD_COUNT_4096 / 2..].copy_from_slice(
748 &sqr_16(&a[WORD_COUNT_4096 * 3 / 4..]));
751 fn add_32_subarr(a: &[u64], b: &[u64]) -> ([u64; WORD_COUNT_4096 * 2], bool) {
752 debug_assert_eq!(a.len(), WORD_COUNT_4096 * 2);
753 debug_assert_eq!(b.len(), WORD_COUNT_4096 * 2);
754 debug_assert_eq!(&a[..WORD_COUNT_4096 * 3 / 2], &[0; WORD_COUNT_4096 * 3 / 2]);
755 debug_assert_eq!(&b[..WORD_COUNT_4096 * 3 / 2], &[0; WORD_COUNT_4096 * 3 / 2]);
756 let (add, overflow) = add_32(&a[WORD_COUNT_4096 * 3 / 2..], &b[WORD_COUNT_4096 * 3 / 2..]);
757 let mut res = [0; WORD_COUNT_4096 * 2];
758 res[WORD_COUNT_4096 * 3 / 2..].copy_from_slice(&add);
761 fn sub_16_subarr(a: &[u64], b: &[u64]) -> ([u64; WORD_COUNT_4096], bool) {
762 debug_assert_eq!(a.len(), WORD_COUNT_4096);
763 debug_assert_eq!(b.len(), WORD_COUNT_4096);
764 debug_assert_eq!(&a[..WORD_COUNT_4096 * 3 / 4], &[0; WORD_COUNT_4096 * 3 / 4]);
765 debug_assert_eq!(&b[..WORD_COUNT_4096 * 3 / 4], &[0; WORD_COUNT_4096 * 3 / 4]);
766 let (sub, underflow) = sub_16(&a[WORD_COUNT_4096 * 3 / 4..], &b[WORD_COUNT_4096 * 3 / 4..]);
767 let mut res = [0; WORD_COUNT_4096];
768 res[WORD_COUNT_4096 * 3 / 4..].copy_from_slice(&sub);
771 (16, 10, mul_16_subarr as mul_ty, sqr_16_subarr as sqr_ty, add_32_subarr as add_double_ty, sub_16_subarr as sub_ty)
773 fn mul_32_subarr(a: &[u64], b: &[u64]) -> [u64; WORD_COUNT_4096 * 2] {
774 debug_assert_eq!(a.len(), WORD_COUNT_4096);
775 debug_assert_eq!(b.len(), WORD_COUNT_4096);
776 debug_assert_eq!(&a[..WORD_COUNT_4096 / 2], &[0; WORD_COUNT_4096 / 2]);
777 debug_assert_eq!(&b[..WORD_COUNT_4096 / 2], &[0; WORD_COUNT_4096 / 2]);
778 let mut res = [0; WORD_COUNT_4096 * 2];
779 res[WORD_COUNT_4096..].copy_from_slice(
780 &mul_32(&a[WORD_COUNT_4096 / 2..], &b[WORD_COUNT_4096 / 2..]));
783 fn sqr_32_subarr(a: &[u64]) -> [u64; WORD_COUNT_4096 * 2] {
784 debug_assert_eq!(a.len(), WORD_COUNT_4096);
785 debug_assert_eq!(&a[..WORD_COUNT_4096 / 2], &[0; WORD_COUNT_4096 / 2]);
786 let mut res = [0; WORD_COUNT_4096 * 2];
787 res[WORD_COUNT_4096..].copy_from_slice(
788 &sqr_32(&a[WORD_COUNT_4096 / 2..]));
791 fn add_64_subarr(a: &[u64], b: &[u64]) -> ([u64; WORD_COUNT_4096 * 2], bool) {
792 debug_assert_eq!(a.len(), WORD_COUNT_4096 * 2);
793 debug_assert_eq!(b.len(), WORD_COUNT_4096 * 2);
794 debug_assert_eq!(&a[..WORD_COUNT_4096], &[0; WORD_COUNT_4096]);
795 debug_assert_eq!(&b[..WORD_COUNT_4096], &[0; WORD_COUNT_4096]);
796 let (add, overflow) = add_64(&a[WORD_COUNT_4096..], &b[WORD_COUNT_4096..]);
797 let mut res = [0; WORD_COUNT_4096 * 2];
798 res[WORD_COUNT_4096..].copy_from_slice(&add);
801 fn sub_32_subarr(a: &[u64], b: &[u64]) -> ([u64; WORD_COUNT_4096], bool) {
802 debug_assert_eq!(a.len(), WORD_COUNT_4096);
803 debug_assert_eq!(b.len(), WORD_COUNT_4096);
804 debug_assert_eq!(&a[..WORD_COUNT_4096 / 2], &[0; WORD_COUNT_4096 / 2]);
805 debug_assert_eq!(&b[..WORD_COUNT_4096 / 2], &[0; WORD_COUNT_4096 / 2]);
806 let (sub, underflow) = sub_32(&a[WORD_COUNT_4096 / 2..], &b[WORD_COUNT_4096 / 2..]);
807 let mut res = [0; WORD_COUNT_4096];
808 res[WORD_COUNT_4096 / 2..].copy_from_slice(&sub);
811 (32, 11, mul_32_subarr as mul_ty, sqr_32_subarr as sqr_ty, add_64_subarr as add_double_ty, sub_32_subarr as sub_ty)
814 (64, 12, mul_64 as mul_ty, sqr_64 as sqr_ty, add_128 as add_double_ty, sub_64 as sub_ty)
817 let mut r = [0; WORD_COUNT_4096 * 2];
818 r[WORD_COUNT_4096 * 2 - word_count - 1] = 1;
820 let mut m_inv_pos = [0; WORD_COUNT_4096];
821 m_inv_pos[WORD_COUNT_4096 - 1] = 1;
822 let mut two = [0; WORD_COUNT_4096];
823 two[WORD_COUNT_4096 - 1] = 2;
824 for _ in 0..log_bits {
825 let mut m_m_inv = mul(&m_inv_pos, &m.0);
826 m_m_inv[..WORD_COUNT_4096 * 2 - word_count].fill(0);
827 let m_inv = mul(&sub(&two, &m_m_inv[WORD_COUNT_4096..]).0, &m_inv_pos);
828 m_inv_pos[WORD_COUNT_4096 - word_count..].copy_from_slice(&m_inv[WORD_COUNT_4096 * 2 - word_count..]);
830 m_inv_pos[..WORD_COUNT_4096 - word_count].fill(0);
832 // We want the negative modular inverse of m mod R, so subtract m_inv from R.
833 let mut m_inv = m_inv_pos;
835 m_inv[..WORD_COUNT_4096 - word_count].fill(0);
836 debug_assert_eq!(&mul(&m_inv, &m.0)[WORD_COUNT_4096 * 2 - word_count..],
838 &[0xffff_ffff_ffff_ffff; WORD_COUNT_4096][WORD_COUNT_4096 - word_count..]);
840 debug_assert_eq!(&m_inv[..WORD_COUNT_4096 - word_count], &[0; WORD_COUNT_4096][..WORD_COUNT_4096 - word_count]);
842 let mont_reduction = |mu: [u64; WORD_COUNT_4096 * 2]| -> [u64; WORD_COUNT_4096] {
843 debug_assert_eq!(&mu[..WORD_COUNT_4096 * 2 - word_count * 2],
844 &[0; WORD_COUNT_4096 * 2][..WORD_COUNT_4096 * 2 - word_count * 2]);
845 let mut mu_mod_r = [0; WORD_COUNT_4096];
846 mu_mod_r[WORD_COUNT_4096 - word_count..].copy_from_slice(&mu[WORD_COUNT_4096 * 2 - word_count..]);
847 let mut v = mul(&mu_mod_r, &m_inv);
848 v[..WORD_COUNT_4096 * 2 - word_count].fill(0); // mod R
849 let t0 = mul(&v[WORD_COUNT_4096..], &m.0);
850 let (t1, t1_extra_bit) = add_double(&t0, &mu);
851 let mut t1_on_r = [0; WORD_COUNT_4096];
852 debug_assert_eq!(&t1[WORD_COUNT_4096 * 2 - word_count..], &[0; WORD_COUNT_4096][WORD_COUNT_4096 - word_count..],
853 "t1 should be divisible by r");
854 t1_on_r[WORD_COUNT_4096 - word_count..].copy_from_slice(&t1[WORD_COUNT_4096 * 2 - word_count * 2..WORD_COUNT_4096 * 2 - word_count]);
855 if t1_extra_bit || t1_on_r >= m.0 {
857 (t1_on_r, underflow) = sub(&t1_on_r, &m.0);
858 debug_assert_eq!(t1_extra_bit, underflow);
863 // Calculate R^2 mod m as ((2^DOUBLES * R) mod m)^(log_bits - LOG2_DOUBLES) mod R
864 let mut r_minus_one = [0xffff_ffff_ffff_ffffu64; WORD_COUNT_4096];
865 r_minus_one[..WORD_COUNT_4096 - word_count].fill(0);
866 // While we do a full div here, in general R should be less than 2x m (assuming the RSA
867 // modulus used its full bit range and is 1024, 2048, or 4096 bits), so it should be cheap.
868 // In cases with a nonstandard RSA modulus we may end up being pretty slow here, but we'll
870 // If we ever find a problem with this we should reduce R to be tigher on m, as we're
871 // wasting extra bits of calculation if R is too far from m.
872 let (_, mut r_mod_m) = debug_unwrap!(div_rem_64(&r_minus_one, &m.0));
873 let r_mod_m_overflow = add_u64!(r_mod_m, 1);
874 if r_mod_m_overflow || r_mod_m >= m.0 {
875 (r_mod_m, _) = sub_64(&r_mod_m, &m.0);
878 let mut r2_mod_m: [u64; 64] = r_mod_m;
879 const DOUBLES: usize = 32;
880 const LOG2_DOUBLES: usize = 5;
882 for _ in 0..DOUBLES {
883 let overflow = double!(r2_mod_m);
884 if overflow || r2_mod_m > m.0 {
885 (r2_mod_m, _) = sub_64(&r2_mod_m, &m.0);
888 for _ in 0..log_bits - LOG2_DOUBLES {
889 r2_mod_m = mont_reduction(sqr(&r2_mod_m));
891 // Clear excess high bits
892 for (m_limb, r2_limb) in m.0.iter().zip(r2_mod_m.iter_mut()) {
893 let clear_bits = m_limb.leading_zeros();
894 if clear_bits == 0 { break; }
895 *r2_limb &= !(0xffff_ffff_ffff_ffffu64 << (64 - clear_bits));
896 if *m_limb != 0 { break; }
898 debug_assert!(r2_mod_m < m.0);
900 // Calculate t * R and a * R as mont multiplications by R^2 mod m
901 let mut tr = mont_reduction(mul(&r2_mod_m, &t));
902 let mut ar = mont_reduction(mul(&r2_mod_m, &self.0));
904 #[cfg(debug_assertions)] {
905 debug_assert_eq!(r2_mod_m, U4096(r_mod_m).mulmod_naive(&U4096(r_mod_m), &m).unwrap().0);
906 debug_assert_eq!(&tr, &U4096(t).mulmod_naive(&U4096(r_mod_m), &m).unwrap().0);
907 debug_assert_eq!(&ar, &self.mulmod_naive(&U4096(r_mod_m), &m).unwrap().0);
912 tr = mont_reduction(mul(&tr, &ar));
915 ar = mont_reduction(sqr(&ar));
918 ar = mont_reduction(mul(&ar, &tr));
919 let mut resr = [0; WORD_COUNT_4096 * 2];
920 resr[WORD_COUNT_4096..].copy_from_slice(&ar);
921 Ok(U4096(mont_reduction(resr)))
925 const fn u64_from_bytes_a_panicking(b: &[u8]) -> u64 {
927 [a, b, c, d, e, f, g, h, ..] => {
928 ((*a as u64) << 8*7) |
929 ((*b as u64) << 8*6) |
930 ((*c as u64) << 8*5) |
931 ((*d as u64) << 8*4) |
932 ((*e as u64) << 8*3) |
933 ((*f as u64) << 8*2) |
934 ((*g as u64) << 8*1) |
941 const fn u64_from_bytes_b_panicking(b: &[u8]) -> u64 {
943 [_, _, _, _, _, _, _, _,
944 a, b, c, d, e, f, g, h, ..] => {
945 ((*a as u64) << 8*7) |
946 ((*b as u64) << 8*6) |
947 ((*c as u64) << 8*5) |
948 ((*d as u64) << 8*4) |
949 ((*e as u64) << 8*3) |
950 ((*f as u64) << 8*2) |
951 ((*g as u64) << 8*1) |
958 const fn u64_from_bytes_c_panicking(b: &[u8]) -> u64 {
960 [_, _, _, _, _, _, _, _,
961 _, _, _, _, _, _, _, _,
962 a, b, c, d, e, f, g, h, ..] => {
963 ((*a as u64) << 8*7) |
964 ((*b as u64) << 8*6) |
965 ((*c as u64) << 8*5) |
966 ((*d as u64) << 8*4) |
967 ((*e as u64) << 8*3) |
968 ((*f as u64) << 8*2) |
969 ((*g as u64) << 8*1) |
976 const fn u64_from_bytes_d_panicking(b: &[u8]) -> u64 {
978 [_, _, _, _, _, _, _, _,
979 _, _, _, _, _, _, _, _,
980 _, _, _, _, _, _, _, _,
981 a, b, c, d, e, f, g, h, ..] => {
982 ((*a as u64) << 8*7) |
983 ((*b as u64) << 8*6) |
984 ((*c as u64) << 8*5) |
985 ((*d as u64) << 8*4) |
986 ((*e as u64) << 8*3) |
987 ((*f as u64) << 8*2) |
988 ((*g as u64) << 8*1) |
995 const fn u64_from_bytes_e_panicking(b: &[u8]) -> u64 {
997 [_, _, _, _, _, _, _, _,
998 _, _, _, _, _, _, _, _,
999 _, _, _, _, _, _, _, _,
1000 _, _, _, _, _, _, _, _,
1001 a, b, c, d, e, f, g, h, ..] => {
1002 ((*a as u64) << 8*7) |
1003 ((*b as u64) << 8*6) |
1004 ((*c as u64) << 8*5) |
1005 ((*d as u64) << 8*4) |
1006 ((*e as u64) << 8*3) |
1007 ((*f as u64) << 8*2) |
1008 ((*g as u64) << 8*1) |
1009 ((*h as u64) << 8*0)
1015 const fn u64_from_bytes_f_panicking(b: &[u8]) -> u64 {
1017 [_, _, _, _, _, _, _, _,
1018 _, _, _, _, _, _, _, _,
1019 _, _, _, _, _, _, _, _,
1020 _, _, _, _, _, _, _, _,
1021 _, _, _, _, _, _, _, _,
1022 a, b, c, d, e, f, g, h, ..] => {
1023 ((*a as u64) << 8*7) |
1024 ((*b as u64) << 8*6) |
1025 ((*c as u64) << 8*5) |
1026 ((*d as u64) << 8*4) |
1027 ((*e as u64) << 8*3) |
1028 ((*f as u64) << 8*2) |
1029 ((*g as u64) << 8*1) |
1030 ((*h as u64) << 8*0)
1037 /// Constructs a new [`U256`] from a variable number of big-endian bytes.
1038 pub(super) fn from_be_bytes(bytes: &[u8]) -> Result<U256, ()> {
1039 if bytes.len() > 256/8 { return Err(()); }
1040 let u64s = (bytes.len() + 7) / 8;
1041 let mut res = [0; WORD_COUNT_256];
1044 let pos = (u64s - i) * 8;
1045 let start = bytes.len().saturating_sub(pos);
1046 let end = bytes.len() + 8 - pos;
1047 b[8 + start - end..].copy_from_slice(&bytes[start..end]);
1048 res[i + WORD_COUNT_256 - u64s] = u64::from_be_bytes(b);
1053 /// Constructs a new [`U256`] from a fixed number of big-endian bytes.
1054 pub(super) const fn from_32_be_bytes_panicking(bytes: &[u8; 32]) -> U256 {
1056 u64_from_bytes_a_panicking(bytes),
1057 u64_from_bytes_b_panicking(bytes),
1058 u64_from_bytes_c_panicking(bytes),
1059 u64_from_bytes_d_panicking(bytes),
1064 pub(super) const fn zero() -> U256 { U256([0, 0, 0, 0]) }
1065 pub(super) const fn one() -> U256 { U256([0, 0, 0, 1]) }
1066 pub(super) const fn three() -> U256 { U256([0, 0, 0, 3]) }
1069 impl<M: PrimeModulus<U256>> U256Mod<M> {
1070 const fn mont_reduction(mu: [u64; 8]) -> Self {
1071 #[cfg(debug_assertions)] {
1072 // Check NEGATIVE_PRIME_INV_MOD_R is correct. Since this is all const, the compiler
1073 // should be able to do it at compile time alone.
1074 let minus_one_mod_r = mul_4(&M::PRIME.0, &M::NEGATIVE_PRIME_INV_MOD_R.0);
1075 assert!(slice_equal(const_subslice(&minus_one_mod_r, 4, 8), &[0xffff_ffff_ffff_ffff; 4]));
1078 #[cfg(debug_assertions)] {
1079 // Check R_SQUARED_MOD_PRIME is correct. Since this is all const, the compiler
1080 // should be able to do it at compile time alone.
1081 let r_minus_one = [0xffff_ffff_ffff_ffff; 4];
1082 let (mut r_mod_prime, _) = sub_4(&r_minus_one, &M::PRIME.0);
1083 add_u64!(r_mod_prime, 1);
1084 let r_squared = sqr_4(&r_mod_prime);
1085 let mut prime_extended = [0; 8];
1086 let prime = M::PRIME.0;
1087 copy_from_slice!(prime_extended, 4, 8, prime);
1088 let (_, r_squared_mod_prime) = if let Ok(v) = div_rem_8(&r_squared, &prime_extended) { v } else { panic!() };
1089 assert!(slice_greater_than(&prime_extended, &r_squared_mod_prime));
1090 assert!(slice_equal(const_subslice(&r_squared_mod_prime, 4, 8), &M::R_SQUARED_MOD_PRIME.0));
1093 let mu_mod_r = const_subslice(&mu, 4, 8);
1094 let mut v = mul_4(&mu_mod_r, &M::NEGATIVE_PRIME_INV_MOD_R.0);
1095 const ZEROS: &[u64; 4] = &[0; 4];
1096 copy_from_slice!(v, 0, 4, ZEROS); // mod R
1097 let t0 = mul_4(const_subslice(&v, 4, 8), &M::PRIME.0);
1098 let (t1, t1_extra_bit) = add_8(&t0, &mu);
1099 let t1_on_r = const_subslice(&t1, 0, 4);
1100 let mut res = [0; 4];
1101 if t1_extra_bit || slice_greater_than(&t1_on_r, &M::PRIME.0) {
1103 (res, underflow) = sub_4(&t1_on_r, &M::PRIME.0);
1104 debug_assert!(t1_extra_bit == underflow);
1106 copy_from_slice!(res, 0, 4, t1_on_r);
1108 Self(U256(res), PhantomData)
1111 pub(super) const fn from_u256_panicking(v: U256) -> Self {
1112 assert!(v.0[0] <= M::PRIME.0[0]);
1113 if v.0[0] == M::PRIME.0[0] {
1114 assert!(v.0[1] <= M::PRIME.0[1]);
1115 if v.0[1] == M::PRIME.0[1] {
1116 assert!(v.0[2] <= M::PRIME.0[2]);
1117 if v.0[2] == M::PRIME.0[2] {
1118 assert!(v.0[3] < M::PRIME.0[3]);
1122 assert!(M::PRIME.0[0] != 0 || M::PRIME.0[1] != 0 || M::PRIME.0[2] != 0 || M::PRIME.0[3] != 0);
1123 Self::mont_reduction(mul_4(&M::R_SQUARED_MOD_PRIME.0, &v.0))
1126 pub(super) fn from_u256(mut v: U256) -> Self {
1127 debug_assert!(M::PRIME.0 != [0; 4]);
1128 debug_assert!(M::PRIME.0[0] > (1 << 63), "PRIME should have the top bit set");
1129 while v >= M::PRIME {
1130 let (new_v, spurious_underflow) = sub_4(&v.0, &M::PRIME.0);
1131 debug_assert!(!spurious_underflow);
1134 Self::mont_reduction(mul_4(&M::R_SQUARED_MOD_PRIME.0, &v.0))
1137 pub(super) fn from_modinv_of(v: U256) -> Result<Self, ()> {
1138 Ok(Self::from_u256(U256(mod_inv_4(&v.0, &M::PRIME.0)?)))
1141 /// Multiplies `self` * `b` mod `m`.
1143 /// Panics if `self`'s modulus is not equal to `b`'s
1144 pub(super) fn mul(&self, b: &Self) -> Self {
1145 Self::mont_reduction(mul_4(&self.0.0, &b.0.0))
1148 /// Doubles `self` mod `m`.
1149 pub(super) fn double(&self) -> Self {
1150 let mut res = self.0.0;
1151 let overflow = double!(res);
1152 if overflow || !slice_greater_than(&M::PRIME.0, &res) {
1154 (res, underflow) = sub_4(&res, &M::PRIME.0);
1155 debug_assert_eq!(overflow, underflow);
1157 Self(U256(res), PhantomData)
1160 /// Multiplies `self` by 3 mod `m`.
1161 pub(super) fn times_three(&self) -> Self {
1162 // TODO: Optimize this a lot
1163 self.mul(&U256Mod::from_u256(U256::three()))
1166 /// Multiplies `self` by 4 mod `m`.
1167 pub(super) fn times_four(&self) -> Self {
1168 // TODO: Optimize this somewhat?
1169 self.double().double()
1172 /// Multiplies `self` by 8 mod `m`.
1173 pub(super) fn times_eight(&self) -> Self {
1174 // TODO: Optimize this somewhat?
1175 self.double().double().double()
1178 /// Multiplies `self` by 8 mod `m`.
1179 pub(super) fn square(&self) -> Self {
1180 Self::mont_reduction(sqr_4(&self.0.0))
1183 /// Subtracts `b` from `self` % `m`.
1184 pub(super) fn sub(&self, b: &Self) -> Self {
1185 let (mut val, underflow) = sub_4(&self.0.0, &b.0.0);
1188 (val, overflow) = add_4(&val, &M::PRIME.0);
1189 debug_assert_eq!(overflow, underflow);
1191 Self(U256(val), PhantomData)
1194 /// Adds `b` to `self` % `m`.
1195 pub(super) fn add(&self, b: &Self) -> Self {
1196 let (mut val, overflow) = add_4(&self.0.0, &b.0.0);
1197 if overflow || !slice_greater_than(&M::PRIME.0, &val) {
1199 (val, underflow) = sub_4(&val, &M::PRIME.0);
1200 debug_assert_eq!(overflow, underflow);
1202 Self(U256(val), PhantomData)
1205 /// Returns the underlying [`U256`].
1206 pub(super) fn into_u256(self) -> U256 {
1207 let mut expanded_self = [0; 8];
1208 expanded_self[4..].copy_from_slice(&self.0.0);
1209 Self::mont_reduction(expanded_self).0
1214 /// Constructs a new [`U384`] from a variable number of big-endian bytes.
1215 pub(super) fn from_be_bytes(bytes: &[u8]) -> Result<U384, ()> {
1216 if bytes.len() > 384/8 { return Err(()); }
1217 let u64s = (bytes.len() + 7) / 8;
1218 let mut res = [0; WORD_COUNT_384];
1221 let pos = (u64s - i) * 8;
1222 let start = bytes.len().saturating_sub(pos);
1223 let end = bytes.len() + 8 - pos;
1224 b[8 + start - end..].copy_from_slice(&bytes[start..end]);
1225 res[i + WORD_COUNT_384 - u64s] = u64::from_be_bytes(b);
1230 /// Constructs a new [`U384`] from a fixed number of big-endian bytes.
1231 pub(super) const fn from_48_be_bytes_panicking(bytes: &[u8; 48]) -> U384 {
1233 u64_from_bytes_a_panicking(bytes),
1234 u64_from_bytes_b_panicking(bytes),
1235 u64_from_bytes_c_panicking(bytes),
1236 u64_from_bytes_d_panicking(bytes),
1237 u64_from_bytes_e_panicking(bytes),
1238 u64_from_bytes_f_panicking(bytes),
1243 pub(super) const fn zero() -> U384 { U384([0, 0, 0, 0, 0, 0]) }
1244 pub(super) const fn one() -> U384 { U384([0, 0, 0, 0, 0, 1]) }
1245 pub(super) const fn three() -> U384 { U384([0, 0, 0, 0, 0, 3]) }
1248 impl<M: PrimeModulus<U384>> U384Mod<M> {
1249 const fn mont_reduction(mu: [u64; 12]) -> Self {
1250 #[cfg(debug_assertions)] {
1251 // Check NEGATIVE_PRIME_INV_MOD_R is correct. Since this is all const, the compiler
1252 // should be able to do it at compile time alone.
1253 let minus_one_mod_r = mul_6(&M::PRIME.0, &M::NEGATIVE_PRIME_INV_MOD_R.0);
1254 assert!(slice_equal(const_subslice(&minus_one_mod_r, 6, 12), &[0xffff_ffff_ffff_ffff; 6]));
1257 #[cfg(debug_assertions)] {
1258 // Check R_SQUARED_MOD_PRIME is correct. Since this is all const, the compiler
1259 // should be able to do it at compile time alone.
1260 let r_minus_one = [0xffff_ffff_ffff_ffff; 6];
1261 let (mut r_mod_prime, _) = sub_6(&r_minus_one, &M::PRIME.0);
1262 add_u64!(r_mod_prime, 1);
1263 let r_squared = sqr_6(&r_mod_prime);
1264 let mut prime_extended = [0; 12];
1265 let prime = M::PRIME.0;
1266 copy_from_slice!(prime_extended, 6, 12, prime);
1267 let (_, r_squared_mod_prime) = if let Ok(v) = div_rem_12(&r_squared, &prime_extended) { v } else { panic!() };
1268 assert!(slice_greater_than(&prime_extended, &r_squared_mod_prime));
1269 assert!(slice_equal(const_subslice(&r_squared_mod_prime, 6, 12), &M::R_SQUARED_MOD_PRIME.0));
1272 let mu_mod_r = const_subslice(&mu, 6, 12);
1273 let mut v = mul_6(&mu_mod_r, &M::NEGATIVE_PRIME_INV_MOD_R.0);
1274 const ZEROS: &[u64; 6] = &[0; 6];
1275 copy_from_slice!(v, 0, 6, ZEROS); // mod R
1276 let t0 = mul_6(const_subslice(&v, 6, 12), &M::PRIME.0);
1277 let (t1, t1_extra_bit) = add_12(&t0, &mu);
1278 let t1_on_r = const_subslice(&t1, 0, 6);
1279 let mut res = [0; 6];
1280 if t1_extra_bit || slice_greater_than(&t1_on_r, &M::PRIME.0) {
1282 (res, underflow) = sub_6(&t1_on_r, &M::PRIME.0);
1283 debug_assert!(t1_extra_bit == underflow);
1285 copy_from_slice!(res, 0, 6, t1_on_r);
1287 Self(U384(res), PhantomData)
1290 pub(super) const fn from_u384_panicking(v: U384) -> Self {
1291 assert!(v.0[0] <= M::PRIME.0[0]);
1292 if v.0[0] == M::PRIME.0[0] {
1293 assert!(v.0[1] <= M::PRIME.0[1]);
1294 if v.0[1] == M::PRIME.0[1] {
1295 assert!(v.0[2] <= M::PRIME.0[2]);
1296 if v.0[2] == M::PRIME.0[2] {
1297 assert!(v.0[3] <= M::PRIME.0[3]);
1298 if v.0[3] == M::PRIME.0[3] {
1299 assert!(v.0[4] <= M::PRIME.0[4]);
1300 if v.0[4] == M::PRIME.0[4] {
1301 assert!(v.0[5] < M::PRIME.0[5]);
1307 assert!(M::PRIME.0[0] != 0 || M::PRIME.0[1] != 0 || M::PRIME.0[2] != 0
1308 || M::PRIME.0[3] != 0|| M::PRIME.0[4] != 0|| M::PRIME.0[5] != 0);
1309 Self::mont_reduction(mul_6(&M::R_SQUARED_MOD_PRIME.0, &v.0))
1312 pub(super) fn from_u384(mut v: U384) -> Self {
1313 debug_assert!(M::PRIME.0 != [0; 6]);
1314 debug_assert!(M::PRIME.0[0] > (1 << 63), "PRIME should have the top bit set");
1315 while v >= M::PRIME {
1316 let (new_v, spurious_underflow) = sub_6(&v.0, &M::PRIME.0);
1317 debug_assert!(!spurious_underflow);
1320 Self::mont_reduction(mul_6(&M::R_SQUARED_MOD_PRIME.0, &v.0))
1323 pub(super) fn from_modinv_of(v: U384) -> Result<Self, ()> {
1324 Ok(Self::from_u384(U384(mod_inv_6(&v.0, &M::PRIME.0)?)))
1327 /// Multiplies `self` * `b` mod `m`.
1329 /// Panics if `self`'s modulus is not equal to `b`'s
1330 pub(super) fn mul(&self, b: &Self) -> Self {
1331 Self::mont_reduction(mul_6(&self.0.0, &b.0.0))
1334 /// Doubles `self` mod `m`.
1335 pub(super) fn double(&self) -> Self {
1336 let mut res = self.0.0;
1337 let overflow = double!(res);
1338 if overflow || !slice_greater_than(&M::PRIME.0, &res) {
1340 (res, underflow) = sub_6(&res, &M::PRIME.0);
1341 debug_assert_eq!(overflow, underflow);
1343 Self(U384(res), PhantomData)
1346 /// Multiplies `self` by 3 mod `m`.
1347 pub(super) fn times_three(&self) -> Self {
1348 // TODO: Optimize this a lot
1349 self.mul(&U384Mod::from_u384(U384::three()))
1352 /// Multiplies `self` by 4 mod `m`.
1353 pub(super) fn times_four(&self) -> Self {
1354 // TODO: Optimize this somewhat?
1355 self.double().double()
1358 /// Multiplies `self` by 8 mod `m`.
1359 pub(super) fn times_eight(&self) -> Self {
1360 // TODO: Optimize this somewhat?
1361 self.double().double().double()
1364 /// Multiplies `self` by 8 mod `m`.
1365 pub(super) fn square(&self) -> Self {
1366 Self::mont_reduction(sqr_6(&self.0.0))
1369 /// Subtracts `b` from `self` % `m`.
1370 pub(super) fn sub(&self, b: &Self) -> Self {
1371 let (mut val, underflow) = sub_6(&self.0.0, &b.0.0);
1374 (val, overflow) = add_6(&val, &M::PRIME.0);
1375 debug_assert_eq!(overflow, underflow);
1377 Self(U384(val), PhantomData)
1380 /// Adds `b` to `self` % `m`.
1381 pub(super) fn add(&self, b: &Self) -> Self {
1382 let (mut val, overflow) = add_6(&self.0.0, &b.0.0);
1383 if overflow || !slice_greater_than(&M::PRIME.0, &val) {
1385 (val, underflow) = sub_6(&val, &M::PRIME.0);
1386 debug_assert_eq!(overflow, underflow);
1388 Self(U384(val), PhantomData)
1391 /// Returns the underlying [`U384`].
1392 pub(super) fn into_u384(self) -> U384 {
1393 let mut expanded_self = [0; 12];
1394 expanded_self[6..].copy_from_slice(&self.0.0);
1395 Self::mont_reduction(expanded_self).0
1404 impl PrimeModulus<U256> for P256 {
1405 const PRIME: U256 = U256::from_32_be_bytes_panicking(&hex_lit::hex!(
1406 "ffffffff00000001000000000000000000000000ffffffffffffffffffffffff"));
1407 const R_SQUARED_MOD_PRIME: U256 = U256::from_32_be_bytes_panicking(&hex_lit::hex!(
1408 "00000004fffffffdfffffffffffffffefffffffbffffffff0000000000000003"));
1409 const NEGATIVE_PRIME_INV_MOD_R: U256 = U256::from_32_be_bytes_panicking(&hex_lit::hex!(
1410 "ffffffff00000002000000000000000000000001000000000000000000000001"));
1414 impl PrimeModulus<U384> for P384 {
1415 const PRIME: U384 = U384::from_48_be_bytes_panicking(&hex_lit::hex!(
1416 "fffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffeffffffff0000000000000000ffffffff"));
1417 const R_SQUARED_MOD_PRIME: U384 = U384::from_48_be_bytes_panicking(&hex_lit::hex!(
1418 "000000000000000000000000000000010000000200000000fffffffe000000000000000200000000fffffffe00000001"));
1419 const NEGATIVE_PRIME_INV_MOD_R: U384 = U384::from_48_be_bytes_panicking(&hex_lit::hex!(
1420 "00000014000000140000000c00000002fffffffcfffffffafffffffbfffffffe00000000000000010000000100000001"));
1427 /// Read some bytes and use them to test bigint math by comparing results against the `ibig` crate.
1428 pub fn fuzz_math(input: &[u8]) {
1429 if input.len() < 32 || input.len() % 16 != 0 { return; }
1430 let split = core::cmp::min(input.len() / 2, 512);
1431 let (a, b) = input.split_at(core::cmp::min(input.len() / 2, 512));
1432 let b = &b[..split];
1434 let ai = ibig::UBig::from_be_bytes(&a);
1435 let bi = ibig::UBig::from_be_bytes(&b);
1437 let mut a_u64s = Vec::with_capacity(split / 8);
1438 for chunk in a.chunks(8) {
1439 a_u64s.push(u64::from_be_bytes(chunk.try_into().unwrap()));
1441 let mut b_u64s = Vec::with_capacity(split / 8);
1442 for chunk in b.chunks(8) {
1443 b_u64s.push(u64::from_be_bytes(chunk.try_into().unwrap()));
1446 macro_rules! test { ($mul: ident, $sqr: ident, $add: ident, $sub: ident, $div_rem: ident, $mod_inv: ident) => {
1447 let res = $mul(&a_u64s, &b_u64s);
1448 let mut res_bytes = Vec::with_capacity(input.len() / 2);
1450 res_bytes.extend_from_slice(&i.to_be_bytes());
1452 assert_eq!(ibig::UBig::from_be_bytes(&res_bytes), ai.clone() * bi.clone());
1454 debug_assert_eq!($mul(&a_u64s, &a_u64s), $sqr(&a_u64s));
1455 debug_assert_eq!($mul(&b_u64s, &b_u64s), $sqr(&b_u64s));
1457 let (res, carry) = $add(&a_u64s, &b_u64s);
1458 let mut res_bytes = Vec::with_capacity(input.len() / 2 + 1);
1459 if carry { res_bytes.push(1); } else { res_bytes.push(0); }
1461 res_bytes.extend_from_slice(&i.to_be_bytes());
1463 assert_eq!(ibig::UBig::from_be_bytes(&res_bytes), ai.clone() + bi.clone());
1465 let mut add_u64s = a_u64s.clone();
1466 let carry = add_u64!(add_u64s, 1);
1467 let mut res_bytes = Vec::with_capacity(input.len() / 2 + 1);
1468 if carry { res_bytes.push(1); } else { res_bytes.push(0); }
1469 for i in &add_u64s {
1470 res_bytes.extend_from_slice(&i.to_be_bytes());
1472 assert_eq!(ibig::UBig::from_be_bytes(&res_bytes), ai.clone() + 1);
1474 let mut double_u64s = b_u64s.clone();
1475 let carry = double!(double_u64s);
1476 let mut res_bytes = Vec::with_capacity(input.len() / 2 + 1);
1477 if carry { res_bytes.push(1); } else { res_bytes.push(0); }
1478 for i in &double_u64s {
1479 res_bytes.extend_from_slice(&i.to_be_bytes());
1481 assert_eq!(ibig::UBig::from_be_bytes(&res_bytes), bi.clone() * 2);
1483 let (quot, rem) = if let Ok(res) =
1484 $div_rem(&a_u64s[..].try_into().unwrap(), &b_u64s[..].try_into().unwrap()) {
1487 let mut quot_bytes = Vec::with_capacity(input.len() / 2);
1489 quot_bytes.extend_from_slice(&i.to_be_bytes());
1491 let mut rem_bytes = Vec::with_capacity(input.len() / 2);
1493 rem_bytes.extend_from_slice(&i.to_be_bytes());
1495 let (quoti, remi) = ibig::ops::DivRem::div_rem(ai.clone(), &bi);
1496 assert_eq!(ibig::UBig::from_be_bytes("_bytes), quoti);
1497 assert_eq!(ibig::UBig::from_be_bytes(&rem_bytes), remi);
1499 if ai != ibig::UBig::from(0u32) { // ibig provides a spurious modular inverse for 0
1500 let ring = ibig::modular::ModuloRing::new(&bi);
1501 let ar = ring.from(ai.clone());
1502 let invi = ar.inverse().map(|i| i.residue());
1504 if let Ok(modinv) = $mod_inv(&a_u64s[..].try_into().unwrap(), &b_u64s[..].try_into().unwrap()) {
1505 let mut modinv_bytes = Vec::with_capacity(input.len() / 2);
1507 modinv_bytes.extend_from_slice(&i.to_be_bytes());
1509 assert_eq!(invi.unwrap(), ibig::UBig::from_be_bytes(&modinv_bytes));
1511 assert!(invi.is_none());
1516 macro_rules! test_mod { ($amodp: expr, $bmodp: expr, $PRIME: expr, $len: expr, $into: ident, $div_rem_double: ident, $div_rem: ident, $mul: ident, $add: ident, $sub: ident) => {
1517 // Test the U256/U384Mod wrapper, which operates in Montgomery representation
1518 let mut p_extended = [0; $len * 2];
1519 p_extended[$len..].copy_from_slice(&$PRIME);
1521 let amodp_squared = $div_rem_double(&$mul(&a_u64s, &a_u64s), &p_extended).unwrap().1;
1522 assert_eq!(&amodp_squared[..$len], &[0; $len]);
1523 assert_eq!(&$amodp.square().$into().0, &amodp_squared[$len..]);
1525 let abmodp = $div_rem_double(&$mul(&a_u64s, &b_u64s), &p_extended).unwrap().1;
1526 assert_eq!(&abmodp[..$len], &[0; $len]);
1527 assert_eq!(&$amodp.mul(&$bmodp).$into().0, &abmodp[$len..]);
1529 let (aplusb, aplusb_overflow) = $add(&a_u64s, &b_u64s);
1530 let mut aplusb_extended = [0; $len * 2];
1531 aplusb_extended[$len..].copy_from_slice(&aplusb);
1532 if aplusb_overflow { aplusb_extended[$len - 1] = 1; }
1533 let aplusbmodp = $div_rem_double(&aplusb_extended, &p_extended).unwrap().1;
1534 assert_eq!(&aplusbmodp[..$len], &[0; $len]);
1535 assert_eq!(&$amodp.add(&$bmodp).$into().0, &aplusbmodp[$len..]);
1537 let (mut aminusb, aminusb_underflow) = $sub(&a_u64s, &b_u64s);
1538 if aminusb_underflow {
1540 (aminusb, overflow) = $add(&aminusb, &$PRIME);
1542 (aminusb, overflow) = $add(&aminusb, &$PRIME);
1546 let aminusbmodp = $div_rem(&aminusb, &$PRIME).unwrap().1;
1547 assert_eq!(&$amodp.sub(&$bmodp).$into().0, &aminusbmodp);
1550 if a_u64s.len() == 2 {
1551 test!(mul_2, sqr_2, add_2, sub_2, div_rem_2, mod_inv_2);
1552 } else if a_u64s.len() == 4 {
1553 test!(mul_4, sqr_4, add_4, sub_4, div_rem_4, mod_inv_4);
1554 let amodp = U256Mod::<fuzz_moduli::P256>::from_u256(U256(a_u64s[..].try_into().unwrap()));
1555 let bmodp = U256Mod::<fuzz_moduli::P256>::from_u256(U256(b_u64s[..].try_into().unwrap()));
1556 test_mod!(amodp, bmodp, fuzz_moduli::P256::PRIME.0, 4, into_u256, div_rem_8, div_rem_4, mul_4, add_4, sub_4);
1557 } else if a_u64s.len() == 6 {
1558 test!(mul_6, sqr_6, add_6, sub_6, div_rem_6, mod_inv_6);
1559 let amodp = U384Mod::<fuzz_moduli::P384>::from_u384(U384(a_u64s[..].try_into().unwrap()));
1560 let bmodp = U384Mod::<fuzz_moduli::P384>::from_u384(U384(b_u64s[..].try_into().unwrap()));
1561 test_mod!(amodp, bmodp, fuzz_moduli::P384::PRIME.0, 6, into_u384, div_rem_12, div_rem_6, mul_6, add_6, sub_6);
1562 } else if a_u64s.len() == 8 {
1563 test!(mul_8, sqr_8, add_8, sub_8, div_rem_8, mod_inv_8);
1564 } else if input.len() == 512*2 + 4 {
1565 let mut e_bytes = [0; 4];
1566 e_bytes.copy_from_slice(&input[512 * 2..512 * 2 + 4]);
1567 let e = u32::from_le_bytes(e_bytes);
1568 let a = U4096::from_be_bytes(&a).unwrap();
1569 let b = U4096::from_be_bytes(&b).unwrap();
1571 let res = if let Ok(r) = a.expmod_odd_mod(e, &b) { r } else { return };
1572 let mut res_bytes = Vec::with_capacity(512);
1574 res_bytes.extend_from_slice(&i.to_be_bytes());
1577 let ring = ibig::modular::ModuloRing::new(&bi);
1578 let ar = ring.from(ai.clone());
1579 assert_eq!(ar.pow(&e.into()).residue(), ibig::UBig::from_be_bytes(&res_bytes));
1587 fn u64s_to_u128(v: [u64; 2]) -> u128 {
1590 r |= (v[0] as u128) << 64;
1594 fn u64s_to_i128(v: [u64; 2]) -> i128 {
1597 r |= (v[0] as i128) << 64;
1603 let mut zero = [0u64; 2];
1605 assert_eq!(zero, [0; 2]);
1607 let mut one = [0u64, 1u64];
1609 assert_eq!(u64s_to_i128(one), -1);
1611 let mut minus_one: [u64; 2] = [u64::MAX, u64::MAX];
1613 assert_eq!(minus_one, [0, 1]);
1618 let mut zero = [0u64; 2];
1619 assert!(!double!(zero));
1620 assert_eq!(zero, [0; 2]);
1622 let mut one = [0u64, 1u64];
1623 assert!(!double!(one));
1624 assert_eq!(one, [0, 2]);
1626 let mut u64_max = [0, u64::MAX];
1627 assert!(!double!(u64_max));
1628 assert_eq!(u64_max, [1, u64::MAX - 1]);
1630 let mut u64_carry_overflow = [0x7fff_ffff_ffff_ffffu64, 0x8000_0000_0000_0000];
1631 assert!(!double!(u64_carry_overflow));
1632 assert_eq!(u64_carry_overflow, [u64::MAX, 0]);
1634 let mut max = [u64::MAX; 4];
1635 assert!(double!(max));
1636 assert_eq!(max, [u64::MAX, u64::MAX, u64::MAX, u64::MAX - 1]);
1640 fn mul_min_simple_tests() {
1643 let res = mul_2(&a, &b);
1644 assert_eq!(res, [0, 3, 10, 8]);
1646 let a = [0x1bad_cafe_dead_beef, 2424];
1647 let b = [0x2bad_beef_dead_cafe, 4242];
1648 let res = mul_2(&a, &b);
1649 assert_eq!(res, [340296855556511776, 15015369169016130186, 4248480538569992542, 10282608]);
1651 let a = [0xf6d9_f8eb_8b60_7a6d, 0x4b93_833e_2194_fc2e];
1652 let b = [0xfdab_0000_6952_8ab4, 0xd302_0000_8282_0000];
1653 let res = mul_2(&a, &b);
1654 assert_eq!(res, [17625486516939878681, 18390748118453258282, 2695286104209847530, 1510594524414214144]);
1656 let a = [0x8b8b_8b8b_8b8b_8b8b, 0x8b8b_8b8b_8b8b_8b8b];
1657 let b = [0x8b8b_8b8b_8b8b_8b8b, 0x8b8b_8b8b_8b8b_8b8b];
1658 let res = mul_2(&a, &b);
1659 assert_eq!(res, [5481115605507762349, 8230042173354675923, 16737530186064798, 15714555036048702841]);
1661 let a = [0x0000_0000_0000_0020, 0x002d_362c_005b_7753];
1662 let b = [0x0900_0000_0030_0003, 0xb708_00fe_0000_00cd];
1663 let res = mul_2(&a, &b);
1664 assert_eq!(res, [1, 2306290405521702946, 17647397529888728169, 10271802099389861239]);
1666 let a = [0x0000_0000_7fff_ffff, 0xffff_ffff_0000_0000];
1667 let b = [0x0000_0800_0000_0000, 0x0000_1000_0000_00e1];
1668 let res = mul_2(&a, &b);
1669 assert_eq!(res, [1024, 0, 483183816703, 18446743107341910016]);
1671 let a = [0xf6d9_f8eb_ebeb_eb6d, 0x4b93_83a0_bb35_0680];
1672 let b = [0xfd02_b9b9_b9b9_b9b9, 0xb9b9_b9b9_b9b9_b9b9];
1673 let res = mul_2(&a, &b);
1674 assert_eq!(res, [17579814114991930107, 15033987447865175985, 488855932380801351, 5453318140933190272]);
1676 let a = [u64::MAX; 2];
1677 let b = [u64::MAX; 2];
1678 let res = mul_2(&a, &b);
1679 assert_eq!(res, [18446744073709551615, 18446744073709551614, 0, 1]);
1684 fn test(a: [u64; 2], b: [u64; 2]) {
1685 let a_int = u64s_to_u128(a);
1686 let b_int = u64s_to_u128(b);
1688 let res = add_2(&a, &b);
1689 assert_eq!((u64s_to_u128(res.0), res.1), a_int.overflowing_add(b_int));
1691 let res = sub_2(&a, &b);
1692 assert_eq!((u64s_to_u128(res.0), res.1), a_int.overflowing_sub(b_int));
1695 test([0; 2], [0; 2]);
1696 test([0x1bad_cafe_dead_beef, 2424], [0x2bad_cafe_dead_cafe, 4242]);
1697 test([u64::MAX; 2], [u64::MAX; 2]);
1698 test([u64::MAX, 0x8000_0000_0000_0000], [0, 0x7fff_ffff_ffff_ffff]);
1699 test([0, 0x7fff_ffff_ffff_ffff], [u64::MAX, 0x8000_0000_0000_0000]);
1700 test([u64::MAX, 0], [0, u64::MAX]);
1701 test([0, u64::MAX], [u64::MAX, 0]);
1702 test([u64::MAX; 2], [0; 2]);
1703 test([0; 2], [u64::MAX; 2]);
1707 fn mul_4_simple_tests() {
1710 assert_eq!(mul_4(&a, &b),
1711 [0, 2, 4, 6, 8, 6, 4, 2]);
1713 let a = [0x1bad_cafe_dead_beef, 2424, 0x1bad_cafe_dead_beef, 2424];
1714 let b = [0x2bad_beef_dead_cafe, 4242, 0x2bad_beef_dead_cafe, 4242];
1715 assert_eq!(mul_4(&a, &b),
1716 [340296855556511776, 15015369169016130186, 4929074249683016095, 11583994264332991364,
1717 8837257932696496860, 15015369169036695402, 4248480538569992542, 10282608]);
1719 let a = [u64::MAX; 4];
1720 let b = [u64::MAX; 4];
1721 assert_eq!(mul_4(&a, &b),
1722 [18446744073709551615, 18446744073709551615, 18446744073709551615,
1723 18446744073709551614, 0, 0, 0, 1]);
1727 fn double_simple_tests() {
1728 let mut a = [0xfff5_b32d_01ff_0000, 0x00e7_e7e7_e7e7_e7e7];
1729 assert!(double!(a));
1730 assert_eq!(a, [18440945635998695424, 130551405668716494]);
1732 let mut a = [u64::MAX, u64::MAX];
1733 assert!(double!(a));
1734 assert_eq!(a, [18446744073709551615, 18446744073709551614]);