Hi all!

I'm new to ChampSim, but trying to use it to simulate a workload that I gathered. I realized that my workload uses exclusively huge (2MB) pages to reduce TLB misses, so I tried to enable huge pages in the ChampSim config to get a more accurate simulation. However, I ran into a few confusing options:

First, I found page_size at the top of the config file, which I tried setting to 2097152. However, I then got the following issue when running champsim:

[VMEM] WARNING: physical memory size is smaller than virtual memory size.
terminate called after throwing an instance of 'std::invalid_argument'
  what():  Upper bound is not representable in the underlying type
Aborted (core dumped)

I then looked further down and also found the virtual_memory section of the config, which also has the pte_page_size parameter. However, I'm not sure how this is supposed to differ from the other page_size argument. If I try to set this to 2MiB as well, I get an immediate assertion failure:

champsim: /home/mgiordan/alliance/champsim/inc/extent.h:49: constexpr champsim::dynamic_extent::dynamic_extent(champsim::data::bits, champsim::data::bits): Assertion `upper >= lower' failed.
Aborted (core dumped)

It seems like this is an issue while creating next_pte_page inside of vmem.cc, but I'm not sure exactly why this is calculated the way it is: next_pte_page(champsim::dynamic_extent{champsim::data::bits{LOG2_PAGE_SIZE}, champsim::data::bits{champsim::lg2(champsim::data::bytes{pte_page_size}.count())}}, 0)

I'm pretty confused, and would really appreciate some clarification on what these two parameters mean and how they differ!
It also seems like I may need to increase the amount of physical memory, and I am not sure which knob I should tune to make that happen (does higher # banks imply more memory total?).

Finally, I also see the option num_levels, which is set to 5. I believe that I should set this to 4 for huge pages, because there is one fewer level of page table depth. Unfortunately, this doesn't seem to help with the issues above. I also would like to know if I should lower minor_fault_penalty by 20% because of the lowered num_levels, or if the lowering of num_levels will internally decrease the page walker latency.

Any advice is much appreciated! Thank you for maintaining a great project.