Hello

sorry I was unresponsive for a while, other non-m5 related activities 
intruded.

Even though I was idle, our cluster here was not.
I ran all of spec2k full reference inputs on m5 for Alpha, x86, 
and x86_64 and put the preliminary results here:
  http://www.csl.cornell.edu/~vince/projects/m5

The x86 (32-bit) results are more or less what I've reported on the list 
earlier.

The x86 (64-bit) results with SSE compiled binaries I had high hopes 
for, but unfortunately there are sitll many issues.

 Many of the benchmarks complained about unimplemented fp/SSE 
 instructions. Most notably:
    fwait, fnstcw_Mw, fldcw_Mw, 
    movdqa_Vo_Wo, movdqa_Wo_Vo, 
    movdqu_Vo_Wo, movdqu_Wo_Vo, 
    prefetch_nta

 The main missing syscalls (besides the ones in my previous patchs) are 
 rt_sigaction and stat.


For Alpha of course the outlook is much better (I ran the benchmarks as 
a baseline).  vpr.place and facerec both give wrong results, and 
there are lots of warnings about non-standard trapping modes not being 
supported, but in general all the rest of the benchmarks ran completely 
(and correctly at a first glance).

I plan to start addressing some of the system-call problems with small, 
carefully tested patches.  Those patches will supercede the 
previous ones I posted.  Progress might be slow for a while though.
I'll also be glad to handle the ioctl changes.

I also don't object to the previously posted systemcall patches by other 
people being merged.  I should have checked harder before doing mine, 
would have saved a lot of duplicated effort.

Vince

_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev

Reply via email to