Boyuan Feng a6b1ef1717 [GraphPartition] cache get_free_symbol_uses (#166338) ... Graph partition relies on `get_free_symbol_uses()` to collect symbol inputs. ee7434be82/torch/_inductor/scheduler.py (L4869-L4885) I empirically observed that `get_free_symbol_uses()` becomes slower for larger graphs. Specifically, I tried to aten fallback for torchtitan which results in 10k+ aten nodes. When processing the 600-th node, it takes seconds to `get_free_symbol_uses()` for 1 node. Why? Because `get_free_symbol_uses()` may recursively call another `get_free_symbol_uses()`, which could recursively run many times. ee7434be82/torch/_inductor/ir.py (L4541-L4543) This PR fixes the issue by caching the results of `get_free_symbol_uses()`. I validated on torchtitan that the issue is fixed. Pull Request resolved: https://github.com/pytorch/pytorch/pull/166338 Approved by: https://github.com/eellison		2025-10-31 02:50:10 +00:00
..
_awaits
_C	Introduce a new API torch.xpu.get_per_process_memory_fraction (#165511)	2025-10-30 19:30:09 +00:00
_C_flatbuffer
_custom_op	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
_decomp	Revert "address DDE in matmul decomp (#166541)"	2025-10-30 18:11:33 +00:00
_dispatch	Fix pyrefly ignores 1/n (#166239)	2025-10-26 00:44:10 +00:00
_dynamo	Fix torch.full with dynamic tensor fill_value in torch.compile (#166554)	2025-10-31 00:56:02 +00:00
_export	Use 'is' in callable comparisons (#166624)	2025-10-30 19:00:09 +00:00
_functorch	Use 'is' in callable comparisons (#166624)	2025-10-30 19:00:09 +00:00
_higher_order_ops	[1/N] Remove unused loop variables (#166258)	2025-10-30 12:22:25 +00:00
_inductor	[GraphPartition] cache get_free_symbol_uses (#166338)	2025-10-31 02:50:10 +00:00
_lazy	Fix pyrefly ignores 1/n (#166239)	2025-10-26 00:44:10 +00:00
_library	Make PT2 compile backprop through custom op without autograd key a hard error (#166367)	2025-10-30 18:43:07 +00:00
_logging	Clean up unused Pyrefly suppressions (#166178)	2025-10-25 05:32:21 +00:00
_numpy	Enable PLW0127 in ruff (#165851)	2025-10-21 03:30:57 +00:00
_prims	Fix pyrefly ignore syntax (#166438)	2025-10-29 00:02:21 +00:00
_prims_common	Fix pyrefly ignore syntax (#166438)	2025-10-29 00:02:21 +00:00
_refs	[3/N] fix typo in other folders (#166606)	2025-10-30 10:30:40 +00:00
_strobelight	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
_subclasses	Fix pyrefly errors on main (#166548)	2025-10-30 16:47:27 +00:00
_vendor
accelerator	Fix pyrefly ignores 1/n (#166239)	2025-10-26 00:44:10 +00:00
amp	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
ao	[Code Clean] Clean asserts in torch/ao/quantization/fx/* (#165420)	2025-10-30 20:53:36 +00:00
autograd	[1/N] Remove unused loop variables (#166258)	2025-10-30 12:22:25 +00:00
backends	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
compiler	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
contrib
cpu	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
csrc	Introduce a new API torch.xpu.get_per_process_memory_fraction (#165511)	2025-10-30 19:30:09 +00:00
cuda	[ROCm][CUDA] add unit test utility busy_wait_for_flag (#166218)	2025-10-29 22:40:23 +00:00
distributed	[CP][BE][2/2] Refactor the code structure (#166501)	2025-10-30 22:07:07 +00:00
distributions	Fix pyrelfy ignore syntax in distributions and ao (#166248)	2025-10-26 22:13:48 +00:00
export	Use 'is' in callable comparisons (#166624)	2025-10-30 19:00:09 +00:00
fft
func
futures
fx	Use 'is' in callable comparisons (#166624)	2025-10-30 19:00:09 +00:00
headeronly	Add a compile-time flag to trigger verbose logging for device-side asserts (#166171)	2025-10-30 19:43:46 +00:00
jit	[1/N] Remove unused loop variables (#166258)	2025-10-30 12:22:25 +00:00
legacy
lib	[2/N] Mark unused parameters in C++ code (#165121)	2025-10-15 03:04:39 +00:00
linalg
masked	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
monitor
mps
mtia	Fix pyrefly ignores 1/n (#166239)	2025-10-26 00:44:10 +00:00
multiprocessing	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
nativert	[triton][sigmoid] Fix kernel cache and serialization issue for triton sigmoid + CUDA kernel bug (#166568)	2025-10-30 14:54:54 +00:00
nested	[3/N] fix typo in other folders (#166606)	2025-10-30 10:30:40 +00:00
nn	Revert "[2/N] Add strict parameter to Python zip calls (#166257)"	2025-10-30 13:20:00 +00:00
numa	Fix pyrefly errors on main (#166548)	2025-10-30 16:47:27 +00:00
onnx	[1/N] Remove unused loop variables (#166258)	2025-10-30 12:22:25 +00:00
optim	Revert "[2/N] Add strict parameter to Python zip calls (#166257)"	2025-10-30 13:20:00 +00:00
package	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
profiler	Revert "[2/N] Add strict parameter to Python zip calls (#166257)"	2025-10-30 13:20:00 +00:00
quantization	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
signal	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
sparse	Revert "[2/N] Add strict parameter to Python zip calls (#166257)"	2025-10-30 13:20:00 +00:00
special
testing	[ROCm] moved gfx1100 back to experimental status for AOTriton (#166397)	2025-10-30 21:43:01 +00:00
utils	[DebugMode] store stringify args by default (#166347)	2025-10-30 22:12:23 +00:00
xpu	Introduce a new API torch.xpu.get_per_process_memory_fraction (#165511)	2025-10-30 19:30:09 +00:00
__config__.py
__future__.py
__init__.py	Fix pyrefly ignore syntax (#166438)	2025-10-29 00:02:21 +00:00
_appdirs.py
_classes.py
_compile.py
_custom_ops.py
_environment.py
_guards.py	Fix pyrefly ignores 1/n (#166239)	2025-10-26 00:44:10 +00:00
_jit_internal.py	Fix pyrefly ignore syntax (#166438)	2025-10-29 00:02:21 +00:00
_linalg_utils.py
_lobpcg.py	Fix pyrefly ignore syntax (#166438)	2025-10-29 00:02:21 +00:00
_lowrank.py
_meta_registrations.py	[Inductor][Triton][FP8] Support deepseek-style scaling in Inductor (#164404)	2025-10-28 03:38:54 +00:00
_namedtensor_internals.py
_ops.py	Use 'is' in callable comparisons (#166624)	2025-10-30 19:00:09 +00:00
_python_dispatcher.py
_size_docs.py
_sources.py
_storage_docs.py
_streambase.py
_tensor_docs.py
_tensor_str.py	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
_tensor.py	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
_thread_safe_fork.py
_torch_docs.py	Clarrifying input output angle unit in the docs for trigonometric fun… (#161248)	2025-10-18 11:53:48 +00:00
_utils_internal.py	Fix pyrefly error syntax (2/n) (#166448)	2025-10-29 00:36:40 +00:00
_utils.py	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
_VF.py
_vmap_internals.py
_weights_only_unpickler.py	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
CMakeLists.txt	[ROCm] Use a ROCm version string without hash. (#166336)	2025-10-28 03:53:55 +00:00
custom_class_detail.h
custom_class.h
extension.h
functional.py	[3/N] fix typo in other folders (#166606)	2025-10-30 10:30:40 +00:00
header_only_apis.txt	Move toUnderlying to headeronly (#165694)	2025-10-22 05:31:16 +00:00
hub.py	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
library.h
library.py	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
overrides.py	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
py.typed
quasirandom.py
random.py	Fix flake8 B028 warnings (#166224)	2025-10-26 06:18:55 +00:00
return_types.py
script.h
serialization.py	Fix syntax for pyrefly errors (#166496)	2025-10-29 20:00:25 +00:00
storage.py	Fix pyrefly ignores 1/n (#166239)	2025-10-26 00:44:10 +00:00
torch_version.py
types.py	Enable PLC0414 on ruff (#165828)	2025-10-22 04:56:52 +00:00
version.py.tpl