Some questions regarding the simulator code and the paper，thanks a lot

[zheweijushi]

(1) Understanding the simulator code:

In the VitCoD.py file, the PE_height parameter is defined, but it doesn't seem to be used anywhere.

Additionally, in line 173, the code _tile_q in range(math.ceil(Q.shape[1] / (args.PE_width/head))) uses PE_width, but it seems like PE_height should be used instead. Could you clarify this?

Additionally, could you please confirm if PE_width represents the number of MAC lanes and PE_height represents the number of MACs within each lane?

(2) Parallelism pattern:
In the paper, it seems the parallelism method for dense q*k Computing is described as follows:
for (head) //parallel in MAC_lines
for(L)
for(K) // in dense SpMM can be instead of global_tokens
for(dk) // parallel in 8 MAC

However, in the code, the parallelism appears to be:
for (L) // parallel in MAC_lines
for (K)
for (head)
for (dk) // parallel in 8 MAC

Could you please clarify what the actual parallelism method is?

(3) Number of MAC lines for encoding/decoding:
The paper does not seem to mention the specific number of MAC lines dedicated to encoding/decoding. Could you provide a typical value for this?

Thank you very much.