)]}'
{
  "commit": "48073298f8a637d40eeeb213eac44706bf3f0ea8",
  "tree": "06a2e0b09ab4589f134e2efdf5cb3e21adcbfd11",
  "parents": [
    "16c9bb2ea958e6ca41f46f48b4bdb738524e9459"
  ],
  "author": {
    "name": "Joseph Huber",
    "email": "huberjn@outlook.com",
    "time": "Tue Sep 29 07:55:59 2026 -0500"
  },
  "committer": {
    "name": "GitHub",
    "email": "noreply@github.com",
    "time": "Tue Sep 29 07:55:59 2026 -0500"
  },
  "message": "[Offload] Try loading dynamic libraries at global scope first (#227020)\n\nSummary:\nThe intention of the dynamic path is to provide basic linking in cases\nwhere the SDK is not avaialble at the user\u0027s build time. Previously this\ndid a distinct `dlopen` call on the library.\n\nHowever, in  cases where an existing HSA implementation exists at global\nscope this can cause multiple instances to be open at the same time.\nSpecifically this is problematic if mixing with another implementation\n(Like HIP or CUDA) or trying to intercept functions (like compiler-rt).\nThe compiler-rt interceptors currently carry interceptors into `dlsym`\nwhen we should instead just load these at global scope, this makes them\nbehave more similar to a direct link in these scenarios, which was the\nintention.\n\nThe fallback remains as the current behavior, should only trigger in\ncases where there is an existing HSA in the process.",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "6cb81f06dd9c457ce03f8d310b84ea3010e7c72f",
      "old_mode": 33188,
      "old_path": "offload/plugins-nextgen/amdgpu/dynamic_hsa/hsa.cpp",
      "new_id": "13bd2fd064002bf89fce24fe2dec847b5ce4ec14",
      "new_mode": 33188,
      "new_path": "offload/plugins-nextgen/amdgpu/dynamic_hsa/hsa.cpp"
    },
    {
      "type": "modify",
      "old_id": "4c7c22da2f046b09a19774651d4e728c7c1a5bc0",
      "old_mode": 33188,
      "old_path": "offload/plugins-nextgen/cuda/dynamic_cuda/cuda.cpp",
      "new_id": "f8c46099b6f494092043af68da1d72d1924c6cd2",
      "new_mode": 33188,
      "new_path": "offload/plugins-nextgen/cuda/dynamic_cuda/cuda.cpp"
    }
  ]
}