ais-check: Add better HIP detection

[derobins]

Adds better HIP detection and output

AIHIPFILE-156

[Copilot]

Pull request overview

Improves ais-check’s HIP runtime detection by scanning common ROCm/HIP installation locations and reporting which discovered HIP runtimes support AIS-related symbols.

Changes:

• Add filesystem-based discovery for libamdhip64.so across common ROCm/HIP paths.
• Probe each discovered HIP runtime for AIS symbol availability and store per-library support state.
• Print a per-HIP-library AIS support summary in non-quiet mode.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Enhances tools/ais-check to improve HIP runtime detection for AMD Infinity Storage (AIS) and to provide more detailed output about which HIP runtime libraries were found and whether each supports AIS.

Changes:

• Add filesystem/environment-based discovery of libamdhip64.so across multiple common ROCm install locations.
• Probe AIS symbol availability across all discovered HIP runtime libraries and aggregate results.
• Print a per-library AIS support summary in the non-quiet output.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[kurtmcmillan]

I think that ais-check should return success if the current environment supports the fastpath. I think this change breaks that. If the current environment is using ROCm 7.0, but ROCm 7.2 is also installed, would this script return success? If so, I don't think it should.

[riley-dixon]

Might be worth adding the word "redundant" (e.g. redundant symlinks) since /opt/rocm will likely be a duplicate entry (tbh I can't think of typical scenarios where it's not). To fix entirely, we could track ROCm installs by the realpath to get rid of symlinks.

[jordan-turbofish]

Played around with getting the absolute path when using find_library:

def dlinfo_path(library_name):
    class LinkMap(ctypes.Structure):
        _fields_ = [
            ("l_addr", ctypes.c_void_p),
            ("l_name", ctypes.c_char_p),
        ]

    dl_path = ctypes.util.find_library("dl")
    library_path = ctypes.util.find_library(library_name)
    if dl_path is None or library_path is None:
        return None

    try:
        libdl = ctypes.CDLL(dl_path)
        libwanted = ctypes.CDLL(library_path)
    except OSError:
        return None

    dlinfo = libdl.dlinfo
    dlinfo.argtypes = ctypes.c_void_p, ctypes.c_int, ctypes.c_void_p
    dlinfo.restype = ctypes.c_int

    lmptr = ctypes.POINTER(LinkMap)()
    err = dlinfo(libwanted._handle, 2, ctypes.byref(lmptr))
    if err != 0:
        return None

    abspath = lmptr.contents.l_name.decode('ascii')
    return abspath

print(f"amdhip64 path:{dlinfo_path(library_name='amdhip64')}")

Testing:

> python3 test.py
amdhip64 path:None
> LD_LIBRARY_PATH=/home/jpatters/hip-7.2.0-dbg/lib:/opt/rocm/lib python3 test.py
amdhip64 path:/home/jpatters/hip-7.2.0-dbg/lib/libamdhip64.so.7