frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

[inductor] Allow backends to register their own custom config object (#158254)

Browse Source 
An out of tree backend can have its own configuration options that the user can enable to control inductor compilation. These config options need to be taken into account when calculating the key that is used to determine cache miss / hits. This PR allows out of tree backends to specify a custom config module that has the same type as `torch._inductor.config` that can be used to control codegen (in addition to the default config), and will be used when creating the cache key.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/158254
Approved by: https://github.com/eellison
This commit is contained in:
Mwiza Kunda
2025-07-23 15:56:06 +00:00
committed by PyTorch MergeBot
parent 7d296d5c19
commit d3d9bc1c31
5 changed files with 101 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
torch/_inductor/codecache.py
View File

@ -52,6 +52,7 @@ from torch._dynamo.exc import SkipFrame
from torch._dynamo.utils import CompileEventLogger, counters, dynamo_timed
from torch._inductor import config, exc, metrics
from torch._inductor.codegen.common import (
custom_backend_codegen_configs,
custom_backend_passes,
init_backend_registration,
)
@ -854,6 +855,13 @@ class FxGraphHashDetails:
map(self._get_custom_pass_detail, custom_backend_passes.values())
)
# Save custom inductor codegen configs
self.custom_backend_codegen_configs = {
device: custom_config.save_config_portable(ignore_private_configs=False)
for device, custom_config in custom_backend_codegen_configs.items()
if custom_config is not None
}
# This is mainly added to handle these two inductor configs, which are (unfortunately)
# sometimes cache safe:
# - _pre_fusion_custom_pass