Neither bitmanip nor crypto specks require intrinsics.
This is probably not an argument against intrinsics: most of bitmanip instructions are emitted by compiler and crypto finds its usage in such libraries as OpenSSL (where generics are already written).
The main argument against intrinsics -- generic code could probably be optimized better than one with intrinsics.
I.e.
is presumably better than multiple intrinsic calls.
Intrinsics are about software but not about instruction set architecture -- so they should be probably developing in riscv-software task group (as it was discussed about crypto intrinsics at the latest meeting).
Neither bitmanip nor crypto specks require intrinsics. This is probably not an argument against intrinsics: most of bitmanip instructions are emitted by compiler and crypto finds its usage in such libraries as OpenSSL (where generics are already written).
The main argument against intrinsics -- generic code could probably be optimized better than one with intrinsics. I.e.
is presumably better than multiple intrinsic calls.
Intrinsics are about software but not about instruction set architecture -- so they should be probably developing in riscv-software task group (as it was discussed about crypto intrinsics at the latest meeting).