Build the custom ops AOT libs as C++20 - #21689
Open
shoumikhin wants to merge 1 commit into
Open
Conversation
ATen requires C++20, so every target that compiles a real PyTorch header needs to opt in. Most already do, but the two custom ops AOT libs were missed: custom_ops_aot_lib compiles torch/library.h through op_sdpa_aot.cpp, and the lib built by gen_custom_ops_aot_lib compiles the generated RegisterSchema.cpp, which includes the same header. Both still build at C++17 against the pinned PyTorch, so this is only visible when building against a newer PyTorch, where they fail on std::strong_ordering in c10/util/intrusive_ptr.h.
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21689
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 3 Unrelated FailuresAs of commit 44ac399 with merge base 48741ac ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
A few ExecuTorch build targets compile real PyTorch headers, instead of the small copy of
c10that ships inside this repo. ATen (the PyTorch tensor library) now needs C++20, so every one of those targets has to ask for C++20 explicitly.Most already do:
util,portable_lib,_training_lib, and the LLM runner all setCXX_STANDARD 20. Two were missed:custom_ops_aot_libinextension/llm/custom_ops/CMakeLists.txt. It compilesop_sdpa_aot.cpp, which includes<torch/library.h>.gen_custom_ops_aot_libintools/cmake/Codegen.cmake. It compiles the generatedRegisterSchema.cpp, which includes the same header.Both still build at C++17 against the currently pinned PyTorch, so nothing is broken today. The problem shows up when building against a newer PyTorch, where the compile fails inside
c10/util/intrusive_ptr.h, which usesstd::strong_ordering, a C++20 feature:The fix is two lines, in the same form the other targets already use:
set_target_properties(custom_ops_aot_lib PROPERTIES CXX_STANDARD 20)Test plan
Configured on Linux x86_64 with
EXECUTORCH_BUILD_KERNELS_LLM_AOT=ONandEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON, then compared the compile flags incompile_commands.jsonbefore and after the change:Exactly the two intended targets change, and no other target is affected.
Also compiled every source file shared by these targets under C++20 to confirm none of them relies on C++17-only behavior (7 files, all pass).