Re: [Gems-users] Ruby x86 patch

[Dan Gibson <degibson@xxxxxxxx>]

Hello!

SPARC's memory_transaction structure (passed to Ruby from Simics) contains 
a pre-decoded variable for determining if an access is privileged.  x86's 
structs have a great number of access types defined, in lieu of the single 
bit from SPARC. Each of these access types may be handled differently if 
you wish. The "vanilla" version that the x86 patch uses simply treats 
(nearly) all accesses as non-privileged.

To change operations from priv to non-priv or vice versa, modify the switch 
statement in function SIMICS_is_priv_x86 in ruby/simics/interface.C.

As for your other problems, if you can give some specific information (eg 
Ruby configuration parameters, if you have changed the defaults and/or 
simulation error outputs), I will try to reproduce your problems.

Regards,
Dan

At 11:49 AM 10/21/2005 +0200, you wrote:

> I have tested x86 ruby module, with several configurations (2 & 4 procs), 
> and I
> obtained courious results. Only 1st. processor misses, no supervisor mode
> accesses, no messages for chips others than 1st.
>
> Here is a stats file extract:
> ...
> Total_misses: 9864
> total_misses: 9864 [ 9864 0 ]
> user_misses: 9864 [ 9864 0 ]
> supervisor_misses: 0 [ 0 0 ]
>
> instruction_executed: 88731013 [ 41394406 47336607 ]
> cycles_per_instruction: 0.533477 [ 0.571768 0.499994 ]
> misses_per_thousand_instructions: 0.111167 [ 0.238293 0 ]
> .....
>   L1D_cache_access_mode_type_UserMode:   143677    100%
> ...
> MessageBuffer: [Chip 0 0, L1Cache, mandatoryQueue_in] stats - msgs:143677 
> full:0
> MessageBuffer: [Chip 1 0, L1Cache, mandatoryQueue_in] stats - msgs:0 full:0
> .....
>
> I repeated tests with sparc ruby module and I think results are OK. The 
> extract
> from stats file:
> ....
> Total_misses: 19120
> total_misses: 19120 [ 14741 4379 ]
> user_misses: 5209 [ 4125 1084 ]
> supervisor_misses: 13911 [ 10616 3295 ]
>
> instruction_executed: 32813149 [ 15066729 17746420 ]
> cycles_per_instruction: 0.676977 [ 0.737179 0.625866 ]
> misses_per_thousand_instructions: 0.582693 [ 0.978381 0.246754 ]
> .....
>
> Is this due to x86 beta version?
>
> Thanks
>
> -------------------------------------------------
> This mail sent through IMP: http://horde.org/imp/
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users