DRAM modeling

[hex0102]

Hi Francisco,

I looked through the accelerator code and try to understand how the interaction between the STONNE engine and DRAM is modeled. Can you briefly explain how DRAM bandwidth/latency is modeled? From the paper, you mentioned Gbuffer has double buffering and DRAM is modeled with DRAMsim. Can you also point me to the corresponding codes?

Thanks. Xin

[francisco-munoz]

Hi @hex0102 , sorry for the late reply. The part that controls the memory accesses is in SDMemory.cpp (dense). However, right now the DRAM part mentioned in the paper is not published in the repo yet. I will publish it soon, but I've been very busy lately and want to test everything before I do the final push. I am sorry for the inconvenience. I will try to push it as soon as possible.

[isai-roman]

Dear all,

We are currently testing STONNE and we have interest in the memory hierarchy simulation. Is the DRAM functionality already published? Please let us know your comments.

Best regards,

Isai Roman.

[Adrian-2105]

Hi,

This STONNE version does not perform an accurate memory hierarchy simulation. For that reason, SST-STONNE has been released, which integrates STONNE into the SST simulator, allowing to implement a complete and accurate simulation of the memory hierarchy (including cache, DRAM, etc).

Currently, we still do not have any official/complete documentation on its repository since there is some pending work to be done. However, @francisco-munoz made a tutorial recently about how to use it and it should be enough to start using the tool. I attach the useful links you could need:

• ASPLOS Tutorial 2023: https://stonne-simulator.github.io/ASPLOSTUT.html. Take a look to the (Hands-on) SST-STONNE part
• ASPLOS Tutorial 2023 Materials: https://github.com/stonne-simulator/tutorials/tree/main/ASPLOS23
• SST-STONNE repository: https://github.com/stonne-simulator/sst-elements-with-stonne (use branch sparse_dataflows_project since it is the most updated)

To use SST-STONNE, I recommend you to use directly our Docker image, which already contains all the environment ready to use the new version of the simulator: https://github.com/stonne-simulator/dockerfile. However, if you want to do a manual installation of the environment then try to replicate the installation commands listed on the Dockerfile or contact me if you find any problem during the installation.

I hope it can help you :-)

Best regards, Adrián

[isai-roman]

Hi Adrian,

Great! Let me take a look on it and I will let you know how it’s working. Thank you!

Regards,

Isai