Add sym_int prim op for symbolic integer casting#20679
Merged
Merged
Conversation
torch.export with dynamic shapes (Dim.AUTO) can emit sym_int nodes when symbolic float expressions need integer conversion. Without a registered prim op and C++ kernel, these models fail at export or crash at runtime. Adds Python op registration, mapping entry, and C++ kernel handling Int (passthrough), Double (truncate toward zero), and Bool (0/1).
psiddh
requested review from
JacobSzwejbka,
larryliu0820 and
manuelcandales
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20679
Note: Links to docs will display an error until the docs builds have been completed. ❌ 4 New Failures, 1 Cancelled Job, 2 Unrelated FailuresAs of commit 4ebf64a with merge base ee990d7 ( NEW FAILURES - The following jobs have failed:
CANCELLED JOB - The following job was cancelled. Please retry:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds ExecuTorch support for torch.sym_int emitted by torch.export when symbolic float expressions must be cast back to integers for shape computations, preventing export failures and runtime crashes for dynamic-shape models.
Changes:
- Register a new ExecuTorch prim op
executorch_prim::sym_int.Scalarand implement its C++ kernel (int passthrough, double→int64 truncation toward zero, bool→0/1). - Add Python-side op schema/registration and map
torch.sym_intto the ExecuTorch prim op during lowering. - Add C++ and Python emit tests validating the new op is present in exported graphs and executes correctly at runtime.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| kernels/prim_ops/test/prim_ops_test.cpp | Extends prim-op registration tests and adds value-level correctness tests for sym_int. |
| kernels/prim_ops/register_prim_ops.cpp | Registers the executorch_prim::sym_int.Scalar kernel with type handling for int/double/bool. |
| exir/passes/executorch_prim_ops_registry.py | Adds schema + implementation binding for sym_int and maps torch.sym_int to the ExecuTorch prim op. |
| exir/emit/test/test_emit.py | Adds an end-to-end emit/export-to-runtime test that ensures torch.sym_int appears and executes correctly under dynamic shapes. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
GregoryComer
approved these changes
Jul 1, 2026
The original model used sym_int(sym_float(n)) which is an identity round-trip that PyTorch's symbolic evaluator folds away at trace time. Use a bool→float→int chain instead, which can't be simplified. Also drop the fragile graph node assertion — the C++ unit test validates the kernel directly.
The bool→float→int chain (sym_not→sym_float→sym_int) produces TruncToInt(ToFloat(s0 <= 5)) which triggers a PyTorch value_ranges bug: simple_sympify can't handle ToFloat(False). Use the simple sym_float→sym_int round-trip instead. The symbolic evaluator folds it to identity, but the test still validates the full export→edge→executorch pipeline with dynamic shapes. The C++ unit test validates the sym_int kernel directly.
Comment on lines
+421
to
+423
| } else if (a.isDouble()) { | ||
| out = EValue(static_cast<int64_t>(a.toDouble())); | ||
| } else if (a.isBool()) { |
Comment on lines
+2476
to
+2483
| exported_program = torch.export.export( | ||
| model, (test_inputs[0],), dynamic_shapes=dynamic_shapes | ||
| ) | ||
|
|
||
| edge_program = to_edge( | ||
| exported_program, | ||
| compile_config=exir.EdgeCompileConfig(_check_ir_validity=False), | ||
| ) |
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.
torch.export with dynamic shapes (Dim.AUTO) can emit sym_int nodes when symbolic float expressions need integer conversion. Without a registered prim op and C++ kernel, these models fail at export or crash at runtime.
Adds Python op registration, mapping entry, and C++ kernel handling Int (passthrough), Double (truncate toward zero), and Bool (0/1).