This *almost* fixes the SPARC_FS regression. There are the following
stat differences:

  system.cpu.num_int_register_writes 2108336490 2100562807   -7773683   
-0.37%
  system.cpu.num_int_register_reads 4304894311 4305540407     646096   
+0.02%

I'm pretty sure these are from the way I forced the fp operation to be
done between setting the rounding modes, specifically how the FSR which
specifies the rounding mode is read once in execute itself and then
possibly again in the function with the actual FP operation. It's a
little strange that writes go down though. I'm not sure why that
happens. I'm willing to live with the minor stat noise to get things to
work properly again.

Gabe

On 11/27/11 05:20, Gabe Black wrote:
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://reviews.m5sim.org/r/914/
> -----------------------------------------------------------
>
> Review request for Default, Ali Saidi, Gabe Black, Steve Reinhardt, and 
> Nathan Binkert.
>
>
> Summary
> -------
>
> SPARC: Isolate FP operations enough to prevent code/rounding mode reordering.
>
>
> Diffs
> -----
>
>   src/arch/sparc/isa/formats/basic.isa c226cd1e5e5e 
>
> Diff: http://reviews.m5sim.org/r/914/diff
>
>
> Testing
> -------
>
>
> Thanks,
>
> Gabe
>
> _______________________________________________
> gem5-dev mailing list
> [email protected]
> http://m5sim.org/mailman/listinfo/gem5-dev

_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev

Reply via email to