Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Bug]: rocblas taking over an hour to install, roclas-bench unavailable after installation #1516

Open
obhalerao97 opened this issue Nov 7, 2024 · 4 comments
Assignees

Comments

@obhalerao97
Copy link

obhalerao97 commented Nov 7, 2024

Describe the bug

Issue 1: Installing rocblas is taking over an hour on MI300. When running ./install -d, it gets stuck here for over an hour.
image

Issue 2: Once installation is complete and when I run rocBLAS/scripts/performance/shakespeare.sh, error thrown: ./rocblas-bench: No such file or directory
I have tried to locate rocblas-bench in build/release/clients/staging/, but after installation this is what the folder looks like:
image

To Reproduce

  1. Pull container: https://github.com/powderluv/vllm-docs
  2. exec into container
  3. git clone https://github.com/ROCm/rocBLAS.git
  4. cd rocBLAS
  5. ./install.sh -d

Tried running it without the container too, but installation fails.

Issue: rocblas-bench not found anywhere, and it takes over an hour to install.

Expected behavior

  1. I expected the installation to take less than an hour
  2. I expected to find the rocblas-bench binary in order to run shakespeare.sh

Environment

Hardware used: AMD MI300 8 GPUS

image
Software version
rocm-core 6.2.0.60200-66~20.04
rocblas 4.2.0.60200-66~20.04

environment.txt

@TorreZuk
Copy link
Contributor

TorreZuk commented Nov 7, 2024

Hopefully we can solve your problem, but just need to clarify your goals.
Do you really need to build from source, and can't install the pre-built packages?
sudo apt install rocblas
sudo apt install rocblas-clients
See some more details for what you are doing:
https://rocm.docs.amd.com/projects/rocBLAS/en/latest/install/Linux_Install_Guide.html#linux-install

If you need to build and install from source you need the -c for the clients including rocblas-bench.
Use -a gfx942 to cut your rebuild time down significantly if you only want to run on the node for which you provided details.
./install.sh -d -c -a gfx942

@TorreZuk TorreZuk self-assigned this Nov 7, 2024
@obhalerao97
Copy link
Author

@TorreZuk , thanks for your prompt response! My main goal is to be able to do: rocblas-bench -f gemm -r s -m 4000 -n 4000 -k 4000 --lda 4000 --ldb 4000 --ldc 4000 --transposeA N \ --transposeB T

with varying precisions and gemm sizes. But it seems like I'm unable to install rocblas-bench in any way.

@TorreZuk
Copy link
Contributor

TorreZuk commented Nov 8, 2024

Try the command I mention and let us know how it goes. It will build rocblas-bench:
./install.sh -d -c -a gfx942

@TorreZuk
Copy link
Contributor

Provided information to install pre-built packages and also way to build with significantly lower compute load so I will close issue later this week unless reporter provides feedback.

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

No branches or pull requests

2 participants