3 ms·
It’s ARM64, so why would they be needed? I imagine Mach-O doesn’t differ significantly when the arch changes.
by api 6y ago
It’s ARM64, so why would they be needed? I imagine Mach-O doesn’t differ significantly when the arch changes.
- Someone 6y agoHaving the same instruction set doesn’t imply that the same sequence of instructions as before will be optimal. If so, Apple telling them which formerly optimal sequences now have better replacements would help. Also, do we know their new CPU is pure ARM64, and doesn’t have any Apple-specific extensions? I may not have looked hard enough, but can’t find any statement about that on developer.apple.com. Edit (after reading https://debugger.medium.com/why-is-apples-m1-chip-so-fast-3262b158cba2 https://debugger.medium.com/why-is-apples-m1-chip-so-fast-32...): Apple probably also knows when it makes sense to offload some code from golang’s standard library to one of pieces of specialized silicon on the M1.
- geoah 6y agoI think that @bradfitz’s issue was that apple claimed in their presentation that they would be contributing (publishing patches) to golang in order to get it working on apple silicon. Which never happened. Possible patches could either be for codesigning that was required on silicon, or simply optimisations for things that go might have architecture specific implementations?
- wmf 6y agoClearly something is needed because Go 1.15 doesn't support it and 1.16 won't be available for months. Maybe the announcement was after the 1.15 feature freeze though.