3 ms·
Looks like it was inspired by http://keyj.emphy.de/minimp3/ http://keyj.emphy.de/minimp3/ and https://keyj.emphy.de/kjmp2/ https://keyj.emphy.de/kjmp2/ , always
by userbinator 8y ago
Looks like it was inspired by http://keyj.emphy.de/minimp3/ http://keyj.emphy.de/minimp3/ and https://keyj.emphy.de/kjmp2/ https://keyj.emphy.de/kjmp2/ , always good to see more people trying to implement multimedia codecs. The theory behind it all is not easy to understand, but in practice it turns into a bunch of arithmetic and table lookups.
That said, this one being floating-point and requiring intrinsics makes it less portable.
- lieff 8y agoYou absolutely correct, it's indeed inspired by Keyj`s minimp3. Float-point used because it's hard do with 4-byte fixed-point type and achieve ISO conformance. We must to use different dynamic ranges for different parts of decoder (i.e. emulate own floating point) or more than 4-byte fixed-point type. Only float-point support is needed, SSE/NEON intrinsics is not required and can be fully disabled by MINIMP3_NO_SIMD.