Hi everyone,
I'm trying to run same workload by multi cores in SE mode in order to test some cache coherence issues. (Using risc-v core)
For example, I add a process to cpu[0] by using "system.cpu[0].workload = process". It works, cpu[0] will run the workload correctly in the simulation. Then I try to add this process to cpu[1] by using "system.cpu[1].workload = process". In simulation, only cpu[0] will run the workload. By checking generated config file, cpu[1] do not have its own workload. It will sleep whole time in the simulation. This is reasonable. So I try to create 2 processes(Using same binary) and bind them to 2 cores. In this time, both core will run its own workload correctly. This is because they loaded executables separately and accessed from different memory addresses which will not cause any cache coherence issues like read shared in their L1 cache. After checking code, I found that address transfer will happened in MMUTranslationGen(). For different processes, each memory allocation is independent.
Here is my question:
Thanks + regards,
Ziyang
Hi Ziyang,
Based on my understanding, in SE mode, you can get both cores to execute
the same binary but they will run using different addresses because they
are two processes and have their own address space. If you want to test
cache coherence, you might want to run a multi-threaded program that has
shared data accessed by different threads mapped to different cores.
Best,
Jiayi
On Thu, Jun 16, 2022 at 4:23 PM Peng, Ziyang ziyang.peng@intel.com wrote:
Hi everyone,
I’m trying to run same workload by multi cores in SE mode in order to test
some cache coherence issues. (Using risc-v core)
For example, I add a process to cpu[0] by using “system.cpu[0].workload
= process”. It works, cpu[0] will run the workload correctly in the
simulation. Then I try to add this process to cpu[1] by using “system.cpu[1].workload
= process”. In simulation, only cpu[0] will run the workload. By
checking generated config file, cpu[1] do not have its own workload. It
will sleep whole time in the simulation. This is reasonable. So I try to
create 2 processes(Using same binary) and bind them to 2 cores. In this
time, both core will run its own workload correctly. This is because they
loaded executables separately and accessed from different memory addresses
which will not cause any cache coherence issues like read shared in their
L1 cache. After checking code, I found that address transfer will happened
in MMUTranslationGen(). For different processes, each memory allocation is
independent.
Here is my question:
1. Is there any way to let both core execute same binary in same
memory address in SE mode?
2. Can I let a core run same binary serval times in one simulation?
3. Is there any tools to test cache coherence protocol in gem5 ?
Thanks + regards,
Ziyang
gem5-users mailing list -- gem5-users@gem5.org
To unsubscribe send an email to gem5-users-leave@gem5.org
Thanks Jiayi. Form my understanding, Risc-v core do not support multi-threaded programs in SE mode? But I can take a try with X86 core at first. Thanks for your help!
Thanks + regards,
Ziyang
From: Jiayi Huang jyhuang91@gmail.com
Sent: Saturday, June 18, 2022 12:22 AM
To: The gem5 Users mailing list gem5-users@gem5.org
Subject: [gem5-users] Re: [GEM5] Executing same workload by different cores in SE mode
Hi Ziyang,
Based on my understanding, in SE mode, you can get both cores to execute the same binary but they will run using different addresses because they are two processes and have their own address space. If you want to test cache coherence, you might want to run a multi-threaded program that has shared data accessed by different threads mapped to different cores.
Best,
Jiayi
On Thu, Jun 16, 2022 at 4:23 PM Peng, Ziyang <ziyang.peng@intel.commailto:ziyang.peng@intel.com> wrote:
Hi everyone,
I’m trying to run same workload by multi cores in SE mode in order to test some cache coherence issues. (Using risc-v core)
For example, I add a process to cpu[0] by using “system.cpu[0].workload = process”. It works, cpu[0] will run the workload correctly in the simulation. Then I try to add this process to cpu[1] by using “system.cpu[1].workload = process”. In simulation, only cpu[0] will run the workload. By checking generated config file, cpu[1] do not have its own workload. It will sleep whole time in the simulation. This is reasonable. So I try to create 2 processes(Using same binary) and bind them to 2 cores. In this time, both core will run its own workload correctly. This is because they loaded executables separately and accessed from different memory addresses which will not cause any cache coherence issues like read shared in their L1 cache. After checking code, I found that address transfer will happened in MMUTranslationGen(). For different processes, each memory allocation is independent.
Here is my question:
Thanks + regards,
Ziyang
gem5-users mailing list -- gem5-users@gem5.orgmailto:gem5-users@gem5.org
To unsubscribe send an email to gem5-users-leave@gem5.orgmailto:gem5-users-leave@gem5.org