Hi Corona, (small world
)
Quote:
Originally Posted by
Corona688
You could try using Linux futexes, which handle the nonblocking case completely in userspace, but I'm not sure what that'd do for memory barriers
I hadn't heard of futexes until you mentioned them, but I did some reading and it seems they still use atomic instructions to update shared variables. In that case I could just use one of GCC's built-in atomic operations like "__sync_fetch_ and_ add" or "__sync_bool_compare_and_swap" as described here...
Atomic Builtins - Using the GNU Compiler Collection (GCC)
The thing with these is they use the asm op-code "lock", which issues a hardware lock on the data-bus effectively locking every other process out of memory. Because I'm writing an application that should be scalable for a system with many cores, I'm discouraged by this.
Quote:
Besides, what if you need to move it to MIPS or something?
Could be a possibility, I believe they have made advances into highly parrallel architectures recently, but the project is at a research stage right now so if I can get it to work well on x86 that's good enough for now. I like the sound of this idea though...
Quote:
Could you perhaps reorder it to put prefix, instance, and state in order in memory? You could assemble the data in an MMX or SSE register, then overwrite several structure members in one assembly op.
This could be a good solution, but I'm not sure how to do it. Do you have any examples of similar code as a guide?
Quote:
One thought does occur to me. How large are these structures?
prefix and instance are both uint32_t while state is an enum (guess that means its a uint32_t also?).