Edward Z. Yang de8d81275a Do not decompose in functionalization/proxy tensor if autograd wouldn't have decomposed (#164939) ...

This fixes AOTAutograd rms_norm not being bitwise equivalent to
eager, because it avoids a decomposition.  You can force the
decomposition by having the decomposition in the dispatch table,
but if eager mode wouldn't have decomposed (because it went to the fused
one), we now default to preserving the fused call by default.

This largely reverts https://github.com/pytorch/pytorch/pull/103275/ for view ops. This means that in inference mode we could hit the wrong C++ kernel; if this occurs we should just SymInt'ify the C++ kernel.

Another neat side effect of this change is that Inductor's generated kernels for rms_norm now have rms_norm in their name.

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164939
Approved by: https://github.com/bdhirsh

2025-10-11 01:03:55 +00:00

..

_autoheuristic

[BE] fix typos in torchgen/ (#156083)

2025-06-17 19:25:50 +00:00

aoti

Back out "Revert D81959389" (#163905)

2025-09-30 07:05:13 +00:00

api

Reapply "Make functionalization ViewMeta serializable with pickle. (#143712)" (#163769)

2025-09-25 10:27:37 +00:00

decompositions

Revert "Use absolute path path.resolve() -> path.absolute() (#129409)"

2025-01-04 14:17:20 +00:00

dest

Revert "[Reland] Use std::string_view in torchgen (#158625)"

2025-09-13 07:52:50 +00:00

fuse

Revert "[Inductor] addmm + activation function fusion (#158137)"

2025-08-15 15:34:09 +00:00

operator_versions

[3/N] Use internal linkage in C++ files (#151297)

2025-05-05 17:48:39 +00:00

selective_build

[BE]: ruff PLC0207 - use maxsplit kwarg (#160107)

2025-08-08 03:14:59 +00:00

shape_functions

[BE] fix typos in torchgen/ (#156083)

2025-06-17 19:25:50 +00:00

static_runtime

Revert "[Reland] Use std::string_view in torchgen (#158625)"

2025-09-13 07:52:50 +00:00

__init__.py

Package config/template files with torchgen (#78942)

2022-06-07 13:33:55 +00:00

BUCK.oss

rename BUILD.buck to BUCK.oss (#78792)

2022-06-03 07:23:16 +00:00

BUILD.bazel

Rename tools/codegen to torchgen (#76275)

2022-04-25 01:38:06 +00:00

build.bzl

Revert "[CI] update flake8 and mypy lint dependencies (#158720)"

2025-07-28 22:02:10 +00:00

code_template.py

[BE] Ensure generated stub files by gen_pyi are properly formatted (#150730)

2025-05-17 12:30:40 +00:00

context.py

[5/N] Apply ruff UP035 rule (#164423)

2025-10-02 07:31:11 +00:00

gen_aoti_c_shim.py

Add new_empty (with dtype argument only) to torch::stable (#159508)

2025-08-20 00:50:42 +00:00

gen_backend_stubs.py

[BE][CI] bump ruff to 0.9.2: multiline assert statements (#144546)

2025-02-27 20:46:16 +00:00

gen_functionalization_type.py

Do not decompose in functionalization/proxy tensor if autograd wouldn't have decomposed (#164939)

2025-10-11 01:03:55 +00:00

gen_lazy_tensor.py

[5/N] Apply ruff UP035 rule (#164423)

2025-10-02 07:31:11 +00:00

gen_schema_utils.py

[BE] fix typos in torchgen/ (#156083)

2025-06-17 19:25:50 +00:00

gen_vmap_plumbing.py

Revert "Enable all SIM rules except disabled ones (#164645)"

2025-10-05 19:32:21 +00:00

gen.py

[5/N] Apply ruff UP035 rule (#164423)

2025-10-02 07:31:11 +00:00

local.py

[BE][CI] bump ruff to 0.9.2: multiline assert statements (#144546)

2025-02-27 20:46:16 +00:00

model.py

More ruff SIM fixes (#164695)

2025-10-09 03:24:50 +00:00

native_function_generation.py

[aotd] capture rrelu_with_noise noise mutation in compile (#141867)

2024-12-04 12:18:58 +00:00

utils.py

[TorchGen] Remove unused variables and function imports (#164538)

2025-10-03 20:49:36 +00:00

yaml_utils.py

[BE][CI] bump ruff to 0.9.2: multiline assert statements (#144546)

2025-02-27 20:46:16 +00:00