Question about the use of long instead of int32_t in some intrinsic definitions about rvkrypto-fips HOT 2 CLOSED

This is a convention adopted from the B intrinsic; I can't speak for its elegance. However I can tell you which instructions have it: The long is used specifically when the same instruction format is applicable to both RV32 and RV64, as is the case with sha256 instructions. The convention assumes that "long" is an int32_t on RV32 and int64_t on RV64. If we put the type as int32_t then it would require a cast to the native 64-bit types, which in turn might generate an implicit sign extension instruction (or clearing of the upper half in case of the receiving variable being an unsigned type). At least with code shared between platforms, this would not be ideal, right? It's there to essentially avoid performance problems from unnecessary casts.

from rvkrypto-fips.

Oh I got it! That's a nice design, thanks!

from rvkrypto-fips.