JuliaGPU / JuliaGPU/AMDGPU.jl

BUG: `using AMDGPU` hangs indefinitely during precompilation on Windows on main branch

Open
#1,083 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug
Dominant language
Julia
Stars
344
Forks
79
Avg merge
2d 23h
Merged PRs (30d)
27

Description

Questionnaire

  1. Does ROCm works for you outside of Julia, e.g. C/C++/Python?: Yes

  2. Post output of rocminfo.

N/A (AMD ROCm on Windows does not provide the `rocminfo` utility).
  1. Post output of AMDGPU.versioninfo() if possible.
# This is printed in v2.8.0
┌───────────┬──────────────────┬───────────┬───────────────────────────────────────────
│ Available │ Name             │ Version   │ Path                                     ⋯
├───────────┼──────────────────┼───────────┼───────────────────────────────────────────
│     +     │ LLD              │ -         │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     +     │ Device Libraries │ -         │ C:\\Users\\MASKED\\.julia\\artif ⋯
│     +     │ HIP              │ 7.2.60201 │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     +     │ rocBLAS          │ 5.2.0     │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     +     │ rocSOLVER        │ 3.32.0    │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     +     │ rocSPARSE        │ 4.2.0     │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     +     │ rocRAND          │ 4.2.0     │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     +     │ rocFFT           │ 1.0.36    │ C:\\Program Files\\AMD\\ROCm\\7.2\\bin\\ ⋯
│     -     │ hipTENSOR        │ -         │ -                                        ⋯
│     -     │ MIOpen           │ -         │ -                                        ⋯
└───────────┴──────────────────┴───────────┴───────────────────────────────────────────
                                                                       1 column omitted

AMDGPU devices
┌────┬────────────────────────┬──────────┬───────────┬────────────┬───────────────┐
│ Id │                   Name │ GCN arch │ Wavefront │     Memory │ Shared Memory │
├────┼────────────────────────┼──────────┼───────────┼────────────┼───────────────┤
│  1 │ AMD Radeon RX 7900 XTX │  gfx1100 │        32 │ 23.984 GiB │    64.000 KiB │
└────┴────────────────────────┴──────────┴───────────┴────────────┴───────────────┘```

Reproducing the bug

  1. Describe what's not working.

On Windows (Windows 11, Julia 1.13.0, ROCm 7.2), using AMDGPU (or Pkg.precompile()) on the main branch hangs indefinitely during precompilation and never finishes.

  • On release v2.8.0 and on main branch just before the problematic commit (commit b3e831ef): Precompilation successfully completes.
  • On main branch (commit 63c6c404): Precompilation hangs indefinitely with a frozen background Julia worker process:
(@v1.13) pkg> up
    Updating registry at `C:\Users\MASKED\.julia\registries\General.toml`
 Installed StaticArrays ─ v1.9.22
   Project No packages added to or removed from `C:\Users\MASKED\.julia\environments\v1.13\Project.toml`
  Updating `C:\Users\MASKED\.julia\environments\v1.13\Manifest.toml`
⌃ [61eb1bfa] ↑ GPUCompiler v2.6.0 ⇒ v2.7.0
  [90137ffa] ↑ StaticArrays v1.9.21 ⇒ v1.9.22
  [cc5c0156] ↑ AMDGPU_LLVM_Backend_jll v22.1.8+2 ⇒ v23.1.1+1
⌃ [f52de702] + LLVMDowngrader_jll v0.10.0+1
  [873c0968] - ROCmDeviceLibs_jll v7.0.2+0
        Info Packages marked with ⌃ have new versions available and may be upgradable.
Precompiling packages  ──────────────╸─────────  11/12
  ◐ AMDGPU

I could not find the root cause.

  1. Provide MWE to reproduce it (if possible).
]precompile AMDGPU

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Reproduce the hang on Windows with Julia 1.13.0 and ROCm 7.2 using ]precompile AMDGPU, then compare behavior between commits b3e831ef and 63c6c404. Trace the AMDGPU precompilation path and the package changes between those commits; done means AMDGPU precompilation completes instead of leaving a frozen worker process.

Written by the indexing model from the issue text.

Assessment

Tech stack
julia
Domain
hpc
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
50/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.