[aarch64-port-dev ] Proposed patch for C2 volatile generation (resent)

[Andrew Dinn]

On 15/04/15 15:53, Andrew Haley wrote:
> On 04/15/2015 03:51 PM, Andrew Dinn wrote:
>> So, it seems the latest patch is looking good as a fix to make volatile
>> reads/writes use ldar<x>/stlr<x>. Andrew Haley, what else do we need to
>> do in order to proceed to commit?
> 
> OK, if you think it's sound, I'll kick it for a while.
> 
> I'd like it to be complete and correct, if at all possible.

I'm feeling very good about correct. Complete is a more complex matter.

There are clearly a lot of other cases where we generate dmb
instructions that are redundant. That includes various inlined unsafe
sequences as well as e.g. inlined allocations. So, there is still a lot
of room for the predicates that guard the 3 membar generation rules and
the ldar/stlr rules to be extended to detect other signature input node
sequences.

I don't yet know if we can detect unique node signatures for all cases
nor how easy it will be to make the predicates detect them. How
important each such addition is in terms of performance is also another
issue. I'll look into this to get some idea of what reward we can expect
from any expended effort (just say no to the phrase 'low hanging fruit' :-).

regards,


Andrew Dinn
-----------
Senior Principal Software Engineer
Red Hat UK Ltd
Registered in UK and Wales under Company Registration No. 3798903
Directors: Michael Cunningham (USA), Matt Parson (USA), Charlie Peters
(USA), Michael O'Neill (Ireland)