Skip to content

NanoBind Ref Counting issue for GroupedGemm #202

Description

@asglover

The following warning appears when running the test suite after the grouped gemms got re-added. I will investigate and fix.

nanobind: leaked 8 instances!
 - leaked instance 0x7edfe487ebc8 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F64"
 - leaked instance 0x7edfe460b8c8 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F32"
 - leaked instance 0x7edfe47fb9c8 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F64"
 - leaked instance 0x7edff459c2c8 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F32"
 - leaked instance 0x7edfef4bfdc8 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F32"
 - leaked instance 0x7edfe47fa948 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F64"
 - leaked instance 0x7edfe4a22848 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F64"
 - leaked instance 0x7edfe5689dc8 of type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F32"
nanobind: leaked 3 types!
 - leaked type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F64"
 - leaked type "openequivariance._torch.extlib.oeq_stable_cuda.GroupMM_F32"
 - leaked type "openequivariance._torch.extlib.oeq_stable_cuda.DeviceProp"
nanobind: leaked 11 functions!
 - leaked function ""
 - leaked function ""
 - leaked function ""
 - leaked function "__init__"
 - leaked function ""
 - leaked function "__init__"
 - leaked function ""
 - leaked function "group_gemm"
 - leaked function ""
 - leaked function "group_gemm"
 - ... skipped remainder
nanobind: this is likely caused by a reference counting issue in the binding code.
See https://nanobind.readthedocs.io/en/latest/refleaks.html

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions