Automation to bump rocm-libraries submodule daily

.github/workflows/bump_rocm_libraries.yml

@@ -0,0 +1,44 @@
+name: Bump rocm-libraries submodule
+
+on:
+  workflow_dispatch:
+  schedule:
+    # 2:00 AM PST → 10:00 UTC, Monday–Friday
+    - cron: "0 10 * * 1-5"
+permissions:
+  contents: write
+  pull-requests: write
+jobs:
+  bump-submodules:
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+      - uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
+        with:
+          python-version: '3.12'
+      - name: Configure Git Identity
+        run: |
+          git config --global user.name "therockbot"
+          git config --global user.email "[email protected]"
+
+      - name: Set branch name
+        id: set-branch
+        run: |
+          DATE=$(date +%Y%m%d)
+          BRANCH="github-action/bump-rocm-libraries-submodule-$DATE"
+          echo "branch-name=$BRANCH" >> $GITHUB_OUTPUT
+
+      - name: Generate GitHub App token
+        uses: actions/create-github-app-token@7e473efe3cb98aa54f8d4bac15400b15fad77d94 # v2.2.0
+        id: generate-token
+        with:
+          app-id: ${{ secrets.PULL_REQUEST_APP_ID }}
+          private-key: ${{ secrets.PULL_REQUEST_APP_KEY }}
+
+      - name: Run bump_submodules.py
+        env:
+          GH_TOKEN: ${{ steps.generate-token.outputs.token }}
+        run: |
+          python3 ./build_tools/bump_submodules.py --push-branch \
+          --branch-name ${{ steps.set-branch.outputs.branch-name }} \
+          --components rocm-libraries"

build_tools/bump_submodules.py

@@ -2,9 +2,9 @@
 """Helper script to bump TheRock's submodules, doing the following:
  * (Optional) Creates a new branch
  * Updates submodules from remote using `fetch_sources.py`
- * Creares a commit and tries to apply local patches
+ * Creates a commit and tries to apply local patches
  * (Optional) Pushed the new branch to origin
-
+ * (Optional) Create pull request from the new branch to origin (requires gh cli installed)
 The submodules to bump can be specified via `--components`.
 
 Examples:
@@ -27,6 +27,7 @@
 import shlex
 import subprocess
 import sys
+import shutil
 
 THIS_SCRIPT_DIR = Path(__file__).resolve().parent
 THEROCK_DIR = THIS_SCRIPT_DIR.parent
@@ -36,7 +37,6 @@ def log(*args, **kwargs):
     print(*args, **kwargs)
     sys.stdout.flush()
 
-
 def exec(args: list[str | Path], cwd: Path):
     args = [str(arg) for arg in args]
     log(f"++ Exec [{cwd}]$ {shlex.join(args)}")
@@ -63,6 +63,69 @@ def pin_ck():
         cwd=THEROCK_DIR / "ml-libs" / "composable_kernel",
     )
 
+def get_component_submodules(component: str) -> list[str]:
+    """
+    Returns the list of submodule paths belonging to the component.
+    Reads .gitmodules to ensure accuracy.
+    """
+
+    gitmodules_path = THEROCK_DIR / ".gitmodules"
+    content = gitmodules_path.read_text()
+
+    paths = []
+
+    for line in content.splitlines():
+        line = line.strip()
+        if line.startswith("path = "):
+            sub_path = line.split("=", 1)[1].strip()
+            if sub_path == component or sub_path.startswith(component + "/"):
+                paths.append(sub_path)
+    return paths
+
+
+def get_submodule_hash(path: str) -> str:
+    """
+    Returns the commit hash for a submodule path.
+    """
+    output = subprocess.check_output(
+        ["git", "submodule", "status", path],
+        cwd=str(THEROCK_DIR),
+        text=True
+    ).strip()
+
+    commit = output.split()[0].lstrip("+ -")
+    return commit
+
+
+def get_hashes_for_component(component: str) -> dict:
+    """
+    Returns {submodule_path: hash} for the component.
+    """
+    submodules = get_component_submodules(component)
+
+    hashes = {}
+    for sm in submodules:
+        try:
+            h = get_submodule_hash(sm)
+            hashes[sm] = h
+        except Exception as e:
+            log(f"Failed to read hash for {sm}: {e}")
+    return hashes
+
+
+def detect_hash_change(old: dict, new: dict):
+    """
+    Return (old_hash, new_hash, path) for the first changed submodule.
+    """
+    for path in new:
+        old_h = old.get(path)
+        new_h = new[path]
+        if old_h != new_h:
+            log(f"CHANGE DETECTED in {path}: {old_h} -> {new_h}")
+            return old_h, new_h, path
+
+    log("No changes detected for component")
+    return None, None, None
 
 def parse_components(components: list[str]) -> list[list]:
     arguments = []
@@ -123,13 +186,8 @@ def parse_components(components: list[str]) -> list[list]:
 
 
 def run(args: argparse.Namespace, fetch_args: list[str], system_projects: list[str]):
-    date = datetime.today().strftime("%Y%m%d")
-
-    if args.create_branch or args.push_branch:
-        exec(
-            ["git", "checkout", "-b", args.branch_name],
-            cwd=THEROCK_DIR,
-        )
+    component = args.components[0]
+    old_hashes = get_hashes_for_component(component)
 
     if system_projects:
         projects_args = ["--system-projects"] + system_projects
@@ -151,10 +209,28 @@ def run(args: argparse.Namespace, fetch_args: list[str], system_projects: list[s
     if args.pin_ck:
         pin_ck()
 
-    exec(
-        ["git", "commit", "-a", "-m", "Bump submodules " + date],
-        cwd=THEROCK_DIR,
+    # Detect new hashes BEFORE commit/branch creation
+    new_hashes = get_hashes_for_component(component)
+    old_h, new_h, changed_path = detect_hash_change(old_hashes, new_hashes)
+    if not old_h or not new_h:
+        log("No submodule changes detected – aborting without commit, branch, or PR.")
+        return
+
+    if args.create_branch or args.push_branch or args.create_pr:
+        exec(["git", "checkout", "-b", args.branch_name], cwd=THEROCK_DIR)
+
+    # Prepare commit/PR title/body
+    pr_title = f"Bump {component} from {old_h[:7]} to {new_h[:7]}"
+    pr_body = (
+        f"Bump happened on {datetime.today().strftime('%m%d%Y')}\n\n"
+        "### Submodule changes:\n"
+        + "".join(
+            f"- `{sm}`: {old_hashes.get(sm)} → {new_hashes.get(sm)}\n"
+            for sm in new_hashes
+        )
     )
+    commit_msg = pr_title + "\n\n" + pr_body
+    exec(["git", "commit", "-a", "-m", commit_msg], cwd=THEROCK_DIR)
 
     try:
         exec(
@@ -169,8 +245,21 @@ def run(args: argparse.Namespace, fetch_args: list[str], system_projects: list[s
         exec(
             ["git", "push", "-u", "origin", args.branch_name],
             cwd=THEROCK_DIR,
-        )
-
+        )  
+        # Create PR
+        if args.create_pr:
+            if not shutil.which("gh"):
+                raise SystemExit("ERROR: --create-pr requires the `gh` CLI.\n")      
+            exec(
+                [
+                    "gh", "pr", "create",
+                    "--title", pr_title,
+                    "--body", pr_body,
+                    "--head", args.branch_name,
+                    "--base", "main",
+                ],
+                cwd=THEROCK_DIR,
+            )
 
 def main(argv):
     parser = argparse.ArgumentParser(prog="bump_submodules")
@@ -214,10 +303,16 @@ def main(argv):
                   profiler
              """,
     )
+    parser.add_argument(
+        "--create-pr",
+        default=False,
+        action=argparse.BooleanOptionalAction,
+        help="Create a GitHub PR (requires `gh` CLI)",
+    )
     args = parser.parse_args(argv)
     fetch_args, system_projects = parse_components(args.components)
     run(args, fetch_args, system_projects)
 
 
 if __name__ == "__main__":
-    main(sys.argv[1:])
+    main(sys.argv[1:])