1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586 |
- //! Helper functions for pseudorandom number generation using LCG, see https://pubs.opengroup.org/onlinepubs/9699919799.2018edition/functions/drand48.html
- use platform::types::*;
- /* The default element buffer for the linear congruential generator's
- * sequence. Implemented using a c_ushort array for consistency between
- * the drand48()/lrand48()/mrand48() and erand48()/nrand48()/jrand48()
- * functions, and with STASHED_XI (see below). */
- pub static mut DEFAULT_XI: [c_ushort; 3] = [0; 3];
- // Used by seed48() (returns a pointer to this array).
- pub static mut STASHED_XI: [c_ushort; 3] = [0; 3];
- /* Multiplier and addend, which may be set through lcong48(). Default
- * values as specified in POSIX. */
- const A_DEFAULT_VALUE: u64 = 0x5deece66d;
- const C_DEFAULT_VALUE: u16 = 0xb;
- pub static mut A: u64 = A_DEFAULT_VALUE;
- pub static mut C: u16 = C_DEFAULT_VALUE;
- /// Used by `srand48()` and `seed48()`.
- pub unsafe fn reset_a_and_c() {
- A = A_DEFAULT_VALUE;
- C = C_DEFAULT_VALUE;
- }
- /// Build a 48-bit integer from a size-3 array of unsigned short.
- ///
- /// Takes a pointer argument due to the inappropriate C function
- /// signatures generated from Rust's sized arrays, see
- /// https://github.com/eqrion/cbindgen/issues/171
- pub unsafe fn ushort_arr3_to_uint48(arr_ptr: *const c_ushort) -> u64 {
- let arr = [*arr_ptr.offset(0), *arr_ptr.offset(1), *arr_ptr.offset(2)];
-
- /* Cast via u16 to ensure we get only the lower 16 bits of each
- * element, as specified by POSIX. */
- u64::from(arr[0] as u16) | (u64::from(arr[1] as u16) << 16) | (u64::from(arr[2] as u16) << 32)
- }
- /// Set a size-3 array of unsigned short from a 48-bit integer.
- pub unsafe fn set_ushort_arr3_from_uint48(arr_ptr: *mut c_ushort, value: u64) {
- *arr_ptr.offset(0) = c_ushort::from(value as u16);
- *arr_ptr.offset(1) = c_ushort::from((value >> 16) as u16);
- *arr_ptr.offset(2) = c_ushort::from((value >> 32) as u16);
- }
- /// Advances the buffer from the input argument to the next element in
- /// the linear congruential generator's sequence.
- ///
- /// Modifies the passed argument in-place and returns the new value as a
- /// u64. The input argument must be a size-3 array.
- pub unsafe fn generator_step(xi_arr_ptr: *mut c_ushort) -> u64 {
- let old_xi: u64 = ushort_arr3_to_uint48(xi_arr_ptr);
-
- /* The recurrence relation of the linear congruential generator,
- * X_(n+1) = (a * X_n + c) % m,
- * with m = 2**48. The multiplication and addition can overflow a
- * u64, but we just let it wrap since we take mod 2**48 anyway. */
- let new_xi: u64 = A.wrapping_mul(old_xi).wrapping_add(u64::from(C)) & 0xffff_ffff_ffff;
-
- set_ushort_arr3_from_uint48(xi_arr_ptr, new_xi);
- new_xi
- }
- /// Get a C `double` from a 48-bit integer (for `drand48()` and
- /// `erand48()`).
- pub fn x_to_float64(x: u64) -> c_double {
- /* We set the exponent to 0, and the 48-bit integer is copied into the high
- * 48 of the 52 significand bits. The value then lies in the range
- * [1.0, 2.0), from which we simply subtract 1.0. */
- f64::from_bits(0x3ff0_0000_0000_0000_u64 | (x << 4)) - 1.0f64
- }
- /// Get the high 31 bits of a 48-bit integer (for `lrand48()` and
- /// `nrand48()`).
- pub fn x_to_uint31(x: u64) -> c_long {
- (x >> 17) as c_long
- }
- /// Get the high 32 bits, signed, of a 48-bit integer (for `mrand48()`
- /// and `jrand48()`).
- pub fn x_to_int32(x: u64) -> c_long {
- // Cast via i32 to ensure we get the sign correct
- c_long::from((x >> 16) as i32)
- }
|