Hi.
Yes, you are right. The flexpoint generation runs for 20*25M cycles to generate 
20 flexpoints as you specified in the run_job command and  .run_job.rc.tcl file.
Flexpoint generation for each phase (you only specify one phase in 
.run_job.rc.tcl, namely phase 0) starts with a cold state.
However, after the first flexpoint in each phase is generated, upon the 
generation of each flexpoint (every 25M cycles), the simulator loads the most 
recent flexpoint generated (flexpoint n) and continues with functional 
simulation to generate the next flexpoint (flexpoint n+1). As a result, the 
caches are not empty.

Hope this answers your question.

--
Cansu

On Jul 24, 2013, at 6:17 PM, zhangwuxiang 
<[email protected]<mailto:[email protected]>> wrote:

Hi all,
When using CMP.L2Shared.Trace to generate flexpoints, the cache and branch 
prediction are warmed. But I feel confused about warming time of each flexpoint.
For example, I use this command to generate 28 flexpoints:
 run_job -postprocess 
"/home/zhangwuxiang/modify-flexus-4.1/scripts/postprocess_ckptgen.sh flexpoint 
20 cassandra" -cfg trace -run flexpoint -local -ckpt-gen CMP.L2Shared.Trace 
cassandra/1cpu
And according to .run_job.rc.tcl file, it executes this command:
{ cassandra/1cpu             baseline  0-0:0   "server_cpu0.cycle-break 
25000000;   instruction-fetch-mode instruction-fetch-trace" }
Does this mean that  it run for 25m cycles with CMP.L2Shared.Trace, and take a 
flexpoint. And then  it continues to run with CMP.L2Shared.Trace for next 25m 
cycles, and then take next flexpoint and so on? In other words, it run with 
CMP.L2Shared.Trace for 20*25m cycles continuously, and save a state of cache 
and branch prediction for each flexpoint.
But I am not sure whether this comprehension is right or not.
Another comprehension of the flexpoint : it run for 25m cycles with 
CMP.L2Shared.Trace, and take a flexpoint, it contains a flexstate with cache 
and branch prediction information and a simics checkpoint, and then start from 
this simics checkpoint , run next 25m cycles with CMP.L2Shared.Trace and then 
take flexpoint and so on. If this is right, at the start of every 25m cycles, 
the cache will be empty, isn't it?
I don't know how to confirm which comprehension is right. Can anyone tell me 
which is right and why it right?
Look forward your reply.

2013-07-24
________________________________
Best Regards!
MAIL: [email protected]<mailto:[email protected]>


Reply via email to