2 ms·
This will be an unpopular opinion in some quarters, but I'm ready for RISC-VI. Or -6. RISC-V is pretty close, so work on verifying -V should mostly carry over
by ncmncm 6y ago
This will be an unpopular opinion in some quarters, but I'm ready for RISC-VI. Or -6.
RISC-V is pretty close, so work on verifying -V should mostly carry over to -6. A MIPS engineer posted a list, recently, of errata that mostly would not be hard to fix, such as (IIRC) having separate opcodes for call and return so the microarchitecture can optimize them independently.
I would add making the ABI bool true value ~0 instead of 1; and a POPCNT instruction in the base set, not dependent on a big "bitmanip" extension package.
The extensions infrastructure is itself a problem, illustrated by the last point. Often, certain key parts of an extension are disproportionately useful, but you don't get the sink without a whole kitchen attached.
We probably won't get a new mainstream architecture for decades, so it is worth getting this one right before too much is committed. It would be disruptive to try to fix -V, but -6 work can go on in parallel.