passwordmaker-rs/src/passwordmaker/base_conversion, branch v0.2.1 A Rust reimplementation of the PasswordMaker Pro password generation logic. https://git.grois.info/passwordmaker-rs/atom/src/passwordmaker/base_conversion?h=v0.2.1 2023-01-20T23:26:24Z Minor: Restore compatibility with Rust 1.52 2023-01-20T23:26:24Z Andreas Grois andi@grois.info 2023-01-20T23:26:24Z urn:sha1:283313e8619d629757a62487d8656fb3781dd554 Fix doc error. 2022-11-04T21:02:20Z Andreas Grois andi@grois.info 2022-11-04T21:02:20Z urn:sha1:fb5bd1a86ff5b62e1272c8b66e4dd3d860b74e3f More Clippy lints. 2022-11-04T19:51:12Z Andreas Grois andi@grois.info 2022-11-04T19:51:12Z urn:sha1:36b7ec5ea805196749c7f10f1d8e03ae03564f2b Now Clippy is happy. Some clippy lints. 2022-11-03T23:12:14Z Andreas Grois andi@grois.info 2022-11-03T23:12:14Z urn:sha1:344264e03d7635b9bd2688390100d3b9f623c58a Some were fixed, some were ignored because they seem to make a (negative) performance impact. Test that compare iterative conversion with div. 2022-11-03T21:05:24Z Andreas Grois andi@grois.info 2022-11-03T21:05:24Z urn:sha1:db2e6ce5b648fa513009863e81a13f0e64281a78 These tests compare the code for iterative base conversion with the alternative formula that spits out digits in reverse order. Base Conv: Combine shifting and padding. 2022-10-27T06:56:51Z Andreas Grois andi@grois.info 2022-10-27T06:56:51Z urn:sha1:6ee6a65b30bdbd4956b18518009f25a0b7db0979 Having one function that does both seems to be measurably faster, though I would have expected the compiler to inline and generate comparable assembly. Well, seems it didn't. Minor: Remove an unnecessary clone. 2022-10-26T21:47:18Z Andreas Grois andi@grois.info 2022-10-26T21:47:18Z urn:sha1:37d19e9e9171c819d92afe395e30359aa059ee15 Minor, remove a pointless type conversion. 2022-10-26T19:39:52Z Andreas Grois andi@grois.info 2022-10-26T19:39:52Z urn:sha1:09d9c9e7d8d30c7f718837a44cabc90142f85b3e Seems not to have any performance impact, but still, cleaner this way. Use MulAssign in BaseConversion. 2022-10-26T15:59:27Z Andreas Grois andi@grois.info 2022-10-26T15:59:27Z urn:sha1:839e1096dde8e1e934a82d1cf0c4d3a43dc69f38 Benchmarks show that MulAssign is quite a bit faster than Mul, especially since in this case it's mathematically proven that the multiplication cannot overflow. Also, removed skipping of leading zeros in long division. With the switch to multiplication after the first iteration, the chance that there actually are any leading zeros is on the order of 2^-26. I think at least. In any case, it's small. Base Conv: Mul dividend instead of div divisor. 2022-10-25T19:27:47Z Andreas Grois andi@grois.info 2022-10-25T19:27:47Z urn:sha1:c43c86380e00b018d37dc9b8dbed275c7b6d264d As it turns out, the speed gained by lowering the number of digits in divisor using repeated division is much less than the speed gained by using multiplication of the dividend instead of division of divisor.