sp00n.82
Lt. Junior Grade
- Registriert
- Dez. 2020
- Beiträge
- 338
Mysticial, der Author von y-Cruncher, hatte das hier auf overlock.net gepostet bzgl. der neuen Versionen:
Mysticial schrieb:y-cruncher went through a major rewrite starting from v0.8.1 to v0.8.5. Basically that massive collection of old algorithms which had been accumulating for 15 years finally got cleaned out and replaced with a smaller number of modernized ones which run better (faster) on modern hardware. (and also modernized from a coding perspective)
So all those algorithms are gone. Completely removed from the codebase as they've become a developer maintenance burden.
Already Removed: HNT, N32, N64, VST, C17
To be Removed: SFT, FFT (will be removed in v0.8.6)
New Algorithms: N63 (rewrite of N64), VT3 (rewrite of VST and C17)
To be Added: SFTv4, FFTv4 (rewrites of SFT and FFT) (will be added in v0.8.5)
Since VST, N64, SFT, and FFT have been (or will be) directly replaced, the command line options for them will alias to their replacements. (VST -> VT3, N64 -> N63, etc...)
HNT and N32 have no replacements. They are mathematically obsolete and thus there is no reason to rewrite them.
VT3 is a weird one actually since it's the rewrite of either VST or C17 depending on the binary. VST and C17 are mathematically identical, but differ in how they are implemented. (one uses pure floating-point while the other uses either pure integer or a mix of int/FP)
When I rewrote the algorithm (eventually named VT3), I decided to combine them with the choice of implementation (VST or C17) being a compile-time flag based on the binary.
As much as I wanted to keep both simultaneously for the purpose of stress-testing, the new design didn't allow for it. And I wasn't doing to duplicate 50k lines of code just for stress testing.
- VT3 behaves like C17 on 18-CNL and 22-ZN4.
- Otherwise, it behaves like VST on all other binaries.
- VT3 on the upcoming Zen5 binary (24-ZN5) will also use the C17 implementation.