PTX mode environment set up running tensor core application

[Shouzhe Zhang]

Hello all, 

I was wondering is there anyone successfully running the tensor core application in PTX mode. If so, could you please share the information about the environment set up? For instance, the version of cuda/cudnn/cutlass and gcc/g++?

Thanks in advance, any help would be appreciated!

Shouzhe Zhang

[Rajesh Shashi Kumar]

Hi,

The documentation to run PTX simulations is present here:
https://github.com/accel-sim/accel-sim-framework/blob/dev/README.md

If you have trouble setting up dependencies on your local machine, you can use the docker image as suggested in the README. I have recently documented the steps that I used to run PTX simulations on the docker image in the following gist.

https://gist.github.com/rajesh-s/770e290d127b2484fb3818d69c97bb1d

Thanks,
Rajesh

--
You received this message because you are subscribed to the Google Groups "accel-sim" group.
To unsubscribe from this group and stop receiving emails from it, send an email to accel-sim+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/accel-sim/0c5f8755-0cb1-451e-9048-3ff060c26334n%40googlegroups.com.

[Shouzhe Zhang]

Hello,

Thanks for your reply! Wish you are doing good.

I have gone through the documentation and README, also tried the docker file you provided. There is no problem when running non-tensor-core application like Rodinia and polybench. However, when I wanted to run the tensor-core application of cutlass(cutlass_perf_test), the following error occured.

volta884_gemm_cta_rasterization_tn.sm_70.ptx:738 Syntax error:

        mma.sync.aligned.m8n8k4.row.col.f32.f16.f16.f32 {%f1653,%f1654,%f1655,%f1656,%f1657,%f1658,%f1659,%f1660}, {%r3302,%r3301}, {%r3294,%r3293}, {%f6788,%f6661,%f6787,%f6662,%f6664,%f6663,%f6666,%f6665};

(There was also a little arrow point to "sync".)

Could you please help me out with this issue, thanks a lot!

Regards,

Shouzhe Zhang

[Shouzhe Zhang]

Hello,

There was an additional error message when running the cutlass benchmark in PTX mode:

cutlass_perf_test: cuda_api_object.h:82: void CUctx_st::add_ptxinfo(const char*, const gpgpu_ptx_sim_info&): Assertion `s != NULL' failed.

Aborted (core dumped)

Could you please have a look, thank you!

Regards,

Shouzhe Zhang

[Shouzhe Zhang]

Hello,

I have also tried the deepbench-tencore, all 3 benchmarks will end with error 

Segmentation fault (core dumped)

Regard,

Shouzhe Zhang