Re: [Jikesrvm-core] Opt compiler register assignment for PPC/Linux

[Chris H. <hof...@cs...>]

Dave,

Let me start out by looking at it. Adding some sort of distinguished 
system call operator seemed like a likely approach, and if there's 
already an example I can follow in the IA32 world that's probably all I 
need.

BTW, you may have noticed I've got the OSX regression going to the 
regression-list. I'm not sure why I got so many failures last night; 
when I rebuilt some of the configurations by hand on the same machine 
with the same source code they worked fine. We'll see what happens tonight.

Chris

David P Grove wrote:
> 
> Hi Chirs,
> 
>         Param passing for Java-Java and Java-C is quite different on 
> IA32.  The trick we use in the opt compiler for IA32 is the operator is 
> IA32_SYSCALL instead of CALL.  That's the right way to tackle things on 
> PowerPC too.  It randomly happens to be true that AIX/PowerPC C 
> conventions and our PowerPC Java-Java conventions are the same (and this 
> got baked into the PPC backend), but I don't think we want to 
> automatically assume that.
> 
>         I can dig up some time to put similar plumbing into the PPC 
> backend (calls vs. syscalls) but it may take a few days to get it in 
> place.  If you want to take a whack (learn some of the opt compiler), 
> probably the thing to do is look at how we do it on IA32 and try to do 
> something similar on PPC.  Just let me know if you want to do it, or 
> want me to do it.
> 
>         As a very short term, you could put a kludge into 
> OPT_GenerateMagic to bail out (throw a MagicNotImplementedException) 
> when we hit a problem syscall on Linux/PPC.  That would allow the 
> Linux/PPC night-sanity runs to got through while we get the right fix in 
> place.
> 
> --dave

-- 
Chris Hoffmann -- Dept. of Computer Science/UMass at Amherst
http://www-ali.cs.umass.edu/~hoffmann