You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I suppose so. MIPS has a flag that can be enabled in user-space that causes all load/store instructions to use a reverse byte endian, so a pattern I would imagine for a MIPS64 implementation of qReverse would be
Reverse64:
dsbh $2,$4
dshd $2,$2
which is how __builtin_bswap64 is implemented. So, in theory building qReverse on a gcc compiler targetting MIPS64 would already see speedups. So that work is implicitly already done.
RISC-V doesn't exactly lend itself to a fast in-register byte reverse though I don't think. I've never worked with either and can't think of an instruction that would lend itself for a qReverse acceleration.
is there any support for risc-v and mips in future ?
The text was updated successfully, but these errors were encountered: