Hi Matt,
I'm truly grateful for your help.


| |
林胜
|
|
[email protected]
|
---- Replied Message ----
| From | Poremba, Matthew<[email protected]> |
| Date | 8/26/2022 05:34 |
| To | Matt Sinclair<[email protected]> ,
The gem5 Users mailing list<[email protected]> ,
Dutu, Alexandru<[email protected]> ,
Beckmann, Brad<[email protected]> |
| Subject | [gem5-users] Re: GCN3_X86 How to differentiate memory request from 
CPU or GPU? |

[AMD Official Use Only - General]





Hi Lin,

 

 

If you only want stats, another option is to look at Ruby’s collated stats for 
cache transitions that would trigger memory requests. These have names like 
“system.ruby.TCC_Controller.I.RdBlk,” “system.ruby.TCC_Controller.IV.RdBlk,” 
“system.ruby.TCC_Controller.I.WrVicBlk,” etc. from the GPU. There are similar 
names in MOESI_AMD_Base-L3cache for CPU.

 

If you want to determine this during simulation time, that requires a longer 
response.

 

 

-Matt

 

From: Matt Sinclair <[email protected]>
Sent: Thursday, August 25, 2022 12:09 PM
To: The gem5 Users mailing list <[email protected]>; Poremba, Matthew 
<[email protected]>; Dutu, Alexandru <[email protected]>; Beckmann, 
Brad <[email protected]>
Subject: Re: [gem5-users] GCN3_X86 How to differentiate memory request from CPU 
or GPU?

 

[CAUTION: External Email]

Hi Lin,

 

What sort of stats are you looking for?  The gem5 stats already give you 
information like GPU/CPU L1 and L2 accesses.  So if you are just looking for 
how many accesses the CPU vs. GPU issue to the caches, looking at these stats 
should be sufficient.

 

Thanks,

Matt

 

On Thu, Aug 25, 2022 at 5:20 AM [email protected] <[email protected]> wrote:

Hello,

For some reason, I want to know the memory request source. Is it CPU or GPU. 
Does anyone know how to do this? thanks.

 

|

|

林胜

|
|

[email protected]

|

_______________________________________________
gem5-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
_______________________________________________
gem5-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to