FullO3CPU<Impl>::tick() is not called every cycle. It's only called
when the full cpu is busy. You want to create your own function to do
this and use an EventWrapper to create an event that calls that
function. Every time your function is called you will schedule() at
the bottom of the function to tell M5 when you want that function
called again. Search the code for EventWrapper and you'll find plenty
examples of this being done.
Ali
On Sep 21, 2007, at 7:32 AM, Jun Pang wrote:
Thank you very much for your soon answer. I think you mean I could
add my algorithm in the function Event::reschedule ( Tick t),
so I put a printf in it, but it does not print anything on the
screen. But when I add the printf in the
FullO3CPU<Impl>::tick() function, it prints a number every cycle. I
think the function Event::reschedule( Tick t)
is not called every cycle but the FullO3CPU<Impl>::tick() is. Maybe
I could just add my algorithm here and put
a counter here to deal with frequency difference between CPU and
SDRAM. If my idea is not feasible or it will
have some bad effects, please tell me.
Thank you million times.
On 9/21/07, Ali Saidi <[EMAIL PROTECTED]> wrote:
On Sep 20, 2007, at 9:09 AM, Jun Pang wrote:
Ali,
Thank you very much for your detailed reply.
The implementation of my algorithm must execute every cycle. In
simplescalar, there is a sim-main function with a loop in it to
emulate the pipeline
every cycle. Is there the same function in M5? I think I should
put my algorithm in that function to execute every cycle just like
a stage of pipeline.
In that case you should create a function and every time it's
called at the end reschedule and event re-call it in x
picoseconds. Unlike Simplescalar M5 normally uses one picosecond
as a unit of simulation, so you could convert that to the memory
clock frequency or whatever you want. Take a look at the tick()
function in one of the simple CPUs to see how it's done.
In any recent version of M5 PhysicalMemory is used for both full
system and syscall emulation and the MainMemory object no longer
exists.
Ali
In the mail list 2006, Steve said
"MainMemory is used for running applications(like you are doing,
via the Process object), while PhysicalMemory is
used in full-system mode (when you're booting an operating
system)...... Unless you start using full-system
mode, you will notuse PhysicalMemory or the MemoryController
object (which serves as a switch to direct
memory accesses either to a PhysicalMemory module or a memory-
mapped device, depending on the physical
address)."
Does that mean the SE mode cannot be configured with a physical
memory? And if we want to emulate a SDRAM, I must
use the FS mode?
Thanks a lot
On 9/20/07, Ali Saidi <[EMAIL PROTECTED]> wrote:
On Sep 19, 2007, at 12:09 PM, laymanyang wrote:
Hi all,
I am new to M5 and I have some questions. If they look stupid,
please forgive me.
I want to add an out-of-order scheduling algorithm to physical
memory access. I need to deal with the packet to the physical
memory every cycle. Where should I add the algorithm to?
Where ever you want. Derriving from the PhysicalMemory class
probably makes sense.
Does the PhysicalMemory::MemoryPort::recvTiming(Packet *pkt)
Function execute every cycle or just when the packet comes?
Just went a packet is received.
Can I just add the scheduling here?
You could, but it would be better to add it somewhere else in the
physical memory object. You don't need to be called every cycle,
just the cycles where you have work to do. So you should schedule
an event that calls you function in n cycles. This is done
frequently by the caches and the devices that support DMA. Take a
look at dev/nsgige* and mem/caches/* for examples on how to
schedule events in the future.
And which kind of simulation model should I choose, Full System
or Systemcall Emulation?
Which ever one you need to do the experiments you want to do.
I notice that the latency from
memory->calculateLatency(pkt )is sent to EventQueue by
SimpleTimingPort::SendEvent(this, pkt, time).What is the function
of the EventQueue?
It is a list of all the event the simulator needs to process. The
simulator processes the next event in the event queue until there
are no more events or the simulator exits. You can schedule events
so that you get a call back when a certain amount of time has
elapsed (for example to send a response in 100ns).
Since only the latency is transferred to the eventqueue but the
pkt is not. How do they work collaboratively? What is the
relationship between the CPU and the EventQueue?
See above.
Ali
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
