Skip to content

Add LruFusionCache when direct bindings is used #2806

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

Open
rdspring1 wants to merge 4 commits into
base: main
Choose a base branch
from
Open

Add LruFusionCache when direct bindings is used #2806

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
954fbf1
Add LruFusionCache when direct bindings is used
rdspring1 Oct 29, 2025
fa29d3b
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Dec 15, 2025
989273a
Merge branch 'main' into direct_lru
kshitij12345 Dec 19, 2025
f3aedb6
Merge branch 'main' into direct_lru
rdspring1 Jan 23, 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
13 changes: 12 additions & 1 deletion thunder/executors/nvfuserex_impl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@
# NOTE This impl file is here because nvFuser may not be available, so it's imported conditionally
# by nvfuserex.py when nvFuser is available.

DIRECT_BINDINGS_SUPPORTED_VERSION = LooseVersion("0.2.34")
DIRECT_BINDINGS_SUPPORTED_VERSION = LooseVersion("0.2.35")
Copy link
Collaborator

@kshitij12345 kshitij12345 Dec 18, 2025

Choose a reason for hiding this comment

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

Is this the minimum version which ships with LruFusionCache?

Copy link
Collaborator Author

@rdspring1 rdspring1 Dec 19, 2025

Choose a reason for hiding this comment

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

Yes

DTENSOR_SUPPORTED_VERSION = LooseVersion("0.2.28")
if nvfuser_version() >= DIRECT_BINDINGS_SUPPORTED_VERSION:
import nvfuser_direct as nvfuser
Expand Down Expand Up @@ -298,6 +298,17 @@ def multidevice_schedule(fd: FusionDefinition, in_dtensors: list[Proxy]) -> None
in_tv.set_allocation_domain(in_tv.get_loop_domain(), new_contiguity=True)


# This function wraps nvfuser_direct's LruFusionCache with a version check.
def FusionCacheDecorator(func: callable):
# For legacy bindings, the decorator does nothin.
if nvfuser_version() < DIRECT_BINDINGS_SUPPORTED_VERSION:
return func
from nvfuser_direct import LruFusionCache

return LruFusionCache(max_fusions=16384)(func)
Copy link
Collaborator

@kshitij12345 kshitij12345 Dec 18, 2025

Choose a reason for hiding this comment

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

Default value for max_fusions is 16384, we can remove this here.

Copy link
Collaborator

@crcrpar crcrpar Dec 19, 2025

Choose a reason for hiding this comment

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

qq: will there be an option of setting cache size? Or it would not be that useful

Copy link
Collaborator Author

@rdspring1 rdspring1 Dec 19, 2025

Choose a reason for hiding this comment

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

Usually we pick a reasonable number to avoid out-of-memory issues. We never thought to change it at runtime.



@FusionCacheDecorator
def create_fd(
bsyms: list[BoundSymbol],
input_descriptors: Sequence[type | tuple[tuple[int, ...], tuple[bool, ...], tuple[int, ...]]],
Expand Down
Loading