Simulating multi threaded benchmarks

[Yasas Senevirathne]

Hi,

I'm trying to simulate the bare metal multithreaded benchmarks that come with riscv-tests on a multicore Ariane OpenPiton setup using VCS. This includes mt-memcy, mt-vvadd, mt-matmul, and mm. I tried out both -march=rv64imafdc_zicsr and rv64imafdc_zicsr_zifence machine types when compiling the workloads as well. However, in both cases, the simulation fails with a bad trap.

50151500 : Simulation -> FAIL(HIT BAD TRAP)

Is there anything I'm doing wrong? I also tried adding thread count, thread stride, and finish mask as arguments based on the simulation manual.

sims -sys=manycore -vcs_run -x_tiles=2 -y_tiles=1 mt-memcpy.riscv -midas_args=-DTHREAD_COUNT=2 -midas_args=-DTHREAD_STRIDE=1 -precompiled -ariane -rtl_timeout 1000000 

sims -sys=manycore -vcs_run -x_tiles=2 -y_tiles=1 mt-memcpy.riscv -midas_args=-DTHREAD_COUNT=2 -midas_args=-DTHREAD_STRIDE=2 -finish_mask=11 -precompiled -ariane -rtl_timeout 1000000 

Thank you,

Yasas

[Jonathan Balkind]

Hi Yasas,

Hm interesting. I'm not sure if I have run the multithreaded tests before so I'm trying to think about what to expect. Do the other single threaded workloads work for you?

In terms of your commands, I'd mention that any -midas_ arguments to sims are for SPARC only so they won't do anything here. Whether they're there or not will have no effect. The finish mask and tile arguments look correct (for multithread those are important). It could be worth looking at the tests to see under what circumstances the test might branch to the `fail` label. That's what we dump from the binary using `nm` or similar in order to check the PC for pass/fail.

Also this kind of setup issue might be easier discussed on zulip if you're up for that.

One last note: This specific email somehow got triggered as spam by google groups. I don't think there's anything you can do about it but when I went to approve your other message it was there and I hadn't been informed with a spam digest for some reason. Just a heads up.

Thanks,

Jon

--
You received this message because you are subscribed to the Google Groups "OpenPiton Discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpiton+...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/openpiton/db4b366b-0797-4423-a714-56f2e69e3ba0n%40googlegroups.com.