[MLton] approaches to parallelism in ML
kmacy at fsmware.com
Tue Oct 10 16:02:12 PDT 2006
> The link that Matthew sent
> showed an average 35% hit for using the C codegen. My rule of thumb
> is no worse than a 2X slowdown, and usually less than 50%. This on
> x86 of course. It's quite possible that the hit on other, less
> forgiving, platforms is worse.
> As a rough guess, yes. Although my above point about instability does
> apply. LLVM is a less stable target than SPARC assembly.
Right. I've been burned by external dependencies many times before. The
difference is small enough that, at least for integer code, it probably
doesn't matter on a heavily threaded architecture. What would otherwise
be a pipeline stall just lets other strands run on the core. 64-bit
support and concurrency are certainly a much bigger deal.
More information about the MLton