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.

Sign up for GitHub

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

Jump to bottom

[Feature]: Offload compression #2031

Open
LunNova opened this issue Oct 9, 2024 · 1 comment
Open

[Feature]: Offload compression #2031

LunNova opened this issue Oct 9, 2024 · 1 comment
Labels
change request A code change requested by external or internal stakeholders triaged The issue has been reviewed by a team member and prioritized

Comments

@LunNova
Copy link

LunNova commented Oct 9, 2024 •
edited
Loading

Suggestion Description

TensileCreateLibrary should support offload compression.

rocBLAS uses TensileCreateLibrary and can hit 3GB of hsaco, co and dat/yaml files under lib/rocblas/library/.

Compressing each file individually with zstd reduces this to ~150MB. Compressing everything into a single archive with 7zip reduces this to 70MB.

clang-offload-bundler and clang --offload-arch=... already support compression with --compress and --offload-compress flags.

rocBLAS tries to compress its kernels with these options ROCm/rocBLAS@bb81a83 but it barely does anything because most space is used by tensile which unbundles libraries.

Operating System

N/A

GPU

N/A

ROCm Component

Tensile
LunNova referenced this issue in ROCm/rocBLAS Oct 9, 2024
@TorreZuk
compress code objects (#2707)
bb81a83 
* compress code objects by default
* adds CMake option to control and tests compiler support
@bstefanuk bstefanuk added change request A code change requested by external or internal stakeholders triaged The issue has been reviewed by a team member and prioritized labels Oct 15, 2024
@bstefanuk
Copy link
Contributor

bstefanuk commented Oct 15, 2024 •
edited
Loading

Good issue, thanks for raising this request @LunNova. This would be a nice improvement to help save disk space.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
change request A code change requested by external or internal stakeholders triaged The issue has been reviewed by a team member and prioritized
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@LunNova @bstefanuk