asb wrote:

I failed to report back in the other review thread about the discussion at the 
previous RISC-V LLVM sync-up call. I think it's a fair summary to say we 
concluded was that gating the intrinsics on a flag was a good idea (though not 
a strict requirement if it was going to slow things down too much), and that we 
probably shouldn't agonise over the precise form this flag takes. What you 
propose here seems OK to me, but please add something to RISCVUsage indicating 
that in some cases an extension may be non-experimental but the intrinsics 
still experimental and `-menable-experimental-extensions` must be passed to 
Clang to use them. I also think a Clang release note might make sense, such as 
"It is now possible for intrinsics to be experimental while the extension 
itself is non-experimental. This is currently the case for the vector crypto 
extensions (Zvk*, Zvbb, Zvbc) where `-menable-experimental-extension` must 
still be passed to Clang in order to enable use of the C intrinsics."
cfe-commits mailing list

Reply via email to