|
@@ -54,24 +54,29 @@ float SDL_rand_float(void)
|
|
|
return (SDL_rand() >> (32-24)) * 0x1p-24f;
|
|
|
}
|
|
|
|
|
|
-/* A fast psuedo-random number generator.
|
|
|
- * Not suitable for cryptography or gambling
|
|
|
- */
|
|
|
Uint32 SDL_rand_r(Uint64 *state)
|
|
|
{
|
|
|
if (!state) {
|
|
|
return 0;
|
|
|
}
|
|
|
|
|
|
- // Multiplier from Table 6 of
|
|
|
- // Steele GL, Vigna S. Computationally easy, spectrally good multipliers
|
|
|
- // for congruential pseudorandom number generators.
|
|
|
- // Softw Pract Exper. 2022;52(2):443-458. doi: 10.1002/spe.3030
|
|
|
+ // The C and A parameters of this LCG have been chosen based on hundreds
|
|
|
+ // of core-hours of testing with PractRand and TestU01's Crush.
|
|
|
+ // Using a 32-bit A improves performance on 32-bit architectures.
|
|
|
+ // C can be any odd number, but < 256 generates smaller code on ARM32
|
|
|
+ // These values perform as well as a full 64-bit implementation against
|
|
|
+ // Crush and PractRand. Plus, their worst-case performance is better
|
|
|
+ // than common 64-bit constants when tested against PractRand using seeds
|
|
|
+ // with only a single bit set.
|
|
|
|
|
|
- // 32-bit 'a' improves performance on 32-bit architectures
|
|
|
- // 'c' can be any odd number, but < 256 generates smaller code on some arch
|
|
|
- *state = *state * 0xf9b25d65ul + 0xFD;
|
|
|
+ // We tested all 32-bit and 33-bit A with all C < 256 from a v2 of:
|
|
|
+ // Steele GL, Vigna S. Computationally easy, spectrally good multipliers
|
|
|
+ // for congruential pseudorandom number generators.
|
|
|
+ // Softw Pract Exper. 2022;52(2):443-458. doi: 10.1002/spe.3030
|
|
|
+ // https://arxiv.org/abs/2001.05304v2
|
|
|
|
|
|
- // Only return top 32 bits because they have a longer period
|
|
|
- return (Uint32)(*state >> 32);
|
|
|
+ *state = *state * 0xff1cd035ul + 0x05;
|
|
|
+
|
|
|
+ // Only return top 32 bits because they have a longer period
|
|
|
+ return (Uint32)(*state >> 32);
|
|
|
}
|