Skip to content

Conversation

@RedCarrottt
Copy link
Contributor

In order to resolve Issue #112, I modified GPGPU-Sim so that it extracts PTX files from multiple libraries.

Existing GPGPU-Sim cannot run both convolution layer and FC layer since each layer requires libcudnn or libcublas. Therefore, in Issue #112, it results in PTX not implementation error when simulating libcublas call.

After applying this patch, mnistCUDNN in https://github.com/gpgpu-sim/gpgpu-sim_simulations works well with Titan V config(Volta architecture).

You can the path of multiple libraries as following:
(each library path is separated by colons(:).)

export PYTORCH_BIN=/usr/local/cuda/lib64/libcudnn.so:/usr/local/cuda/lib64/libcublas.so

nassosliak pushed a commit to nassosliak/gpgpu-sim_distribution that referenced this pull request Oct 1, 2025
* Update the AccelSim test script to target the repo specified by the user.

* Change the warning for a missing ACCELSIM_REPO environment variable to an error and update workflow file to point to a temperory fix for circular dependency issue

* fix the repo name

* Change back to use dev branch
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant