Thanks a lot to make me clear!

Most of benchmarks except x264 could run to the end without crash (although 
having invalid opcode somewhere). I will keep this mail-list update when I test 
all benchmarks.

Thanks,
HZ




________________________________
From: avadh patel <[email protected]>
To: Hz Xes <[email protected]>
Cc: marss86 <[email protected]>
Sent: Sat, July 31, 2010 8:58:35 AM
Subject: Re: [marss86-devel] Invalid opcode ?




On Fri, Jul 30, 2010 at 4:14 PM, Hz Xes <[email protected]> wrote:

Hi
> 
>More questions for Marss86, very appreciated the support!
> 
>(1)
>When I run parsec benchmarks using downloaded parsec_image, the running log 
>shows in one more place of Invalid opcode as following:
>Invalid opcode at 0x4ea118: split_invalid_basic_blocks 0, first_insn_in_bb? 0

Ignore this as long as the program does not try to execute from this IP 
location. Due to some branch-predictor issues, sometimes it try to fetch 
instructions from locations that does not have valid instructions. And MARSS 
still hasn't implemented all MMX instructions so that might also be the issue.
Some of benchmark can run to the end, but x264 can not.

x264 uses some specific MMX or SSE instruction that is not yet implemented, so 
it is crashing. If you can find the instruction that cause this, please let us 
know so we can try to implement it.

For the simulations that could finish the runs, are the produced statistics 
valid?  And for those simulations that can not finished, did I make mistakes 
for 
running the simulations or Marss86 has not fully support those applications?

Yes. All the benchmarks that finish correctly, we have validated their output 
with output from real machine to make sure that simulation is correct. Most of 
the statistics has to be valid based on the model implemented in the simulator. 
For the simulations that did not finish, did they crash? or they just took too 
long to finish?
 
(2) 
>Could you please let me know what’s the typical “simconfig” configuration for 
>cache/memory systems to get detail latency and bandwidth statistics? I don’t 
>quite understand the difference of “-use-memory-model”, 
>“-use-new-memory-system”, and “-use-memory-hierarchy”. Could they be used 
>together?

Interesting question. Actually I didnt knew that there are three misleading 
options.. 
-use-memory-model  : NOT USED
-use-new-memory-system : USED and its set by default . This options uses new 
memory hierarchy designed from scratch and it supports coherent caches for 
multicore system.
-use-memory-hierarchy : NOT USED

As far as I know, you dont have to specify anything here. By default it will 
use 
new memory system.

- Avadh
--HZ
>
>_______________________________________________
>http://www.marss86.org
>Marss86-Devel mailing list
>[email protected]
>https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>
>



      
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel

Reply via email to