Skip to content

Fix unguarded import torch in HIPBackend #9441

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

Merged
antiagainst merged 3 commits into from
Feb 17, 2026
Merged

Fix unguarded import torch in HIPBackend #9441

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
52b23b2
Fix `import torch` bomb in HIPBackend
Arech8 Feb 12, 2026
31c199b
reformat with yapf
Arech8 Feb 16, 2026
bb2c7da
Rename the state var and change values
Arech8 Feb 17, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 15 additions & 2 deletions third_party/amd/backend/compiler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -165,14 +165,27 @@ def load_dialects(self, ctx):
if HIPBackend.instrumentation:
HIPBackend.instrumentation.load_dialects(ctx)

# is_within_2gb() needs to check for a torch subobject and this var tracks torch
# availability state: None - not tested, True - torch is present. Anything else -
# no torch available. First call to is_within_2gb() checks torch availability
# and caches it.
_torch_available: None | bool = None

@staticmethod
def is_within_2gb(arg):
import torch
if HIPBackend._torch_available is None:
try:
import torch
Copy link
Copy Markdown
Contributor Author

@Arech8 Arech8 Feb 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe it's just better to do a guarded import torch on a file level, at the top. This would at least make that dependency apparent, but it could remain optional with a similar guard flag as here.

Please tell me your preference, I can change it right here.

HIPBackend._torch_available = True
except ImportError:
HIPBackend._torch_available = False
elif HIPBackend._torch_available:
import torch

MAX_INT_32 = 2**31 - 1
if hasattr(arg, "ptr_range"):
return arg.ptr_range() <= MAX_INT_32
if isinstance(arg, torch.Tensor) and hasattr(arg, "untyped_storage"):
if HIPBackend._torch_available and isinstance(arg, torch.Tensor) and hasattr(arg, "untyped_storage"):
return arg.untyped_storage().size() <= MAX_INT_32
return False

Expand Down
Loading