I think the summary of this thread on outstanding PRs is: > RNG-82: Add XorShift1024StarPhi generator > RNG-70: Add new XoShiRo generators
Merge but: - Ensure those generators with more than one constructor build the same sequence if the seed data is the same. This is to satisfy package requirement that "The "provider" must specify one way for setting the seed. For a given seed, the generated sequence must always be the same.” The new generators have utility constructors where for example an int[] seed of length 2 can be specified as 'int seed0, int seed1'. This is just a different way to specify and set the same exact seed. So this is OK. However using 'long seed0, long seed1' would not be OK as this is not the same seed data. - Test the correlation between XorShift1024Star and XorShift1024StarPhi and possibly mark XOR_SHIFT_1024_S deprecated The test for this is outstanding. A composite fails DieHarder. It passes BigCrush but there is a possible issue with passing the int sequence to the TestU01 BigCrush test suite that is under investigation. > > RNG-81: NumberFactory to sample all rationals between 0 and 1. Merge. > > RNG-78: Added a ThreadLocalRandomSource. Merge. Then update all the benchmarks to include the new providers. Note: A separate thread is discussing the order for benchmark reports. This is whether the order of the generators in the quality benchmarks should match the order of the RandomSource enum. Currently it does not (this requires some simple file rearrangement) but all new generators should be added to the end of the benchmark list so the existing report output files are not affected. Save the commit tag for a speed benchmark reference. > RNG-76: Added primitive long constructor to SplitMix64 Merge as the first commit for improving the construction speed benchmarking. Then update the ProviderBuilder with ideas to improve RNG construction… (Note: with no breaking changes ;) ) Alex --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
