4 ms·
Whilst true for multiply, divide is considerably more complex. You have to step up to the M3 to get a divide instruction on the ARM side. This is why the new Zm
by bpye 4y ago
Whilst true for multiply, divide is considerably more complex. You have to step up to the M3 to get a divide instruction on the ARM side. This is why the new Zmmul is valuable over the existing M extension.
- monocasa 4y agoAgreed, but the point was that state machine, iterative, blocking pipeline stages are pretty common even in RISC cores, including particularly ascetic RISC-V and ARM cores. But I can get how at ~12k gates you don't even want the slowest iterative divide unit.