Skip to content

[ET-VK] Don't specify memory layouts when testing #7324

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
merged 1 commit into from
Dec 13, 2024
+43 −86
Merged

[ET-VK] Don't specify memory layouts when testing #7324

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 8 additions & 9 deletions backends/vulkan/_passes/tag_memory_meta_pass.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,8 +23,6 @@

from executorch.exir.pass_base import ExportPass, PassResult

from torch._subclasses.fake_tensor import FakeTensor

from torch.fx.passes.tools_common import NodeList
from torch.fx.passes.utils.fuser_utils import topo_sort

Expand Down Expand Up @@ -138,9 +136,7 @@ def propose_node_storage(
return storage

for arg in node.args:
if isinstance(arg, torch.fx.Node) and isinstance(
arg.meta["val"], FakeTensor
):
if isinstance(arg, torch.fx.Node) and utils.is_tensor_node(arg):
storage = utils.get_node_storage_type(arg)
if storage is not None and storage in valid_storage_types:
return storage
Expand Down Expand Up @@ -178,9 +174,7 @@ def propose_node_layout(
return layout

for arg in node.args:
if isinstance(arg, torch.fx.Node) and isinstance(
arg.meta["val"], FakeTensor
):
if isinstance(arg, torch.fx.Node) and utils.is_tensor_node(arg):
layout = utils.get_node_memory_layout(arg)
if layout is not None and layout in valid_layouts:
return layout
Expand All @@ -202,14 +196,19 @@ def should_annotate(self, node) -> bool:
if not isinstance(node, torch.fx.Node):
return False

if not isinstance(node.meta["val"], FakeTensor):
if not utils.is_tensor_node(node):
return False

# Storage type and memory layout for tensorref will be determined at runtime
# so there's no use in setting those attributes ahead of time.
if node.meta.get("vkdg_tensorref", False):
return False

# Skip annotating output node. The output tensors should be annotated by the
# time the output node is observed.
if node.op == "output":
return False

return True

def should_delay_annotation(self, node: torch.fx.Node) -> bool:
Expand Down
Loading
Loading