Draft
Conversation
jit.py: "|".join(running_hash) iterated over characters of the concatenated string instead of per-kernel hash values, causing cache key collisions between different ExternalFunction configurations. utils.py: compile_external_kernel silently returned when source_file did not exist, producing a confusing downstream linker error. Now raises FileNotFoundError with the missing path in the message. Regression tests added in test/python/test_jit_utils_bugs.py; they run without hardware or XRT. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…lar import - Delete test/python/test_jit_utils_bugs.py (tested local reimplementations of hash_module and compile_external_kernel rather than the real functions) - Add test/python/npu-xrt/test_jit_utils.py with 12 tests against the real hash_module and compile_external_kernel using real MLIR modules built via iron and real Peano compilation; all 12 pass on hardware - Fix circular import in python/utils/jit.py: top-level `from aie.iron.kernel import ExternalFunction` triggered aie.iron.__init__ which imports jit, causing ImportError when jit.py was the first module loaded; move the import inside decorator() and _filter_tensor_args(), matching the existing deferred-import pattern in that file Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…und (#2879) Closures differing only in captured value produced identical in-memory cache keys because _create_function_cache_key hashed only co_code, co_consts, co_names, and defaults — none of which change when a free variable changes. Add _closure_key() to extract cell_contents from a callable's closure and include it in the hash tuple, with fallbacks for unhashable values (repr()) and deleted cells (ValueError). Apply to both the args and kwargs branches of _create_function_cache_key. Remove the has_closures / effective_use_cache workaround from jit.py and the now-unused _has_closure helper. The workaround was disabling the in-memory cache for all closures and forcing a full disk recompile on every call, even when a valid cached artifact already existed. Add two tests to test_jit_utils.py: - test_closure_cache_key_distinguishes_captured_values: unit test verifying different captured values produce different cache keys - test_jit_closure_parametrize: parametrized hardware test confirming correct output for add_value in {1, 2, 3} Co-Authored-By: Claude Opus 4.6 <[email protected]>
Replace identity-based hash() fallback in _closure_key with a
_cell_val_to_key helper that uses a priority chain:
1. value-hash — O(1) for types with a proper value-based __hash__
(int, str, float, tuple, frozenset, custom immutables)
2. pickle — SHA-256 of pickle.dumps(val) captures full __dict__
state for mutable objects regardless of whether
__eq__ / __hash__ / __repr__ are defined
3. pickle_dict — fallback for locally-defined classes (not picklable
by qualified name); pickle __dict__ directly instead
4. repr — last resort for non-picklable identity-hashable objects
Also include variable names in closure key tuples for precision.
Adds four new unit tests covering: mutable object with no __repr__,
in-place list mutation, key stability without mutation, and the
no-closure fast path.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
The jit() decorator previously had an is_placed parameter to select between two execution modes. This removes the flag and auto-detects the mode from the decorated function's return value: - If the function returns None the module was built implicitly via @device/@core/@runtime_sequence context managers (placed style). - If the function returns a module it is used directly (unplaced style, Program.resolve_program). All call sites that passed is_placed=False explicitly are updated to use the simpler @iron.jit() / @iron.jit form. Comments referencing the is_placed parameter are removed from example files. A new test (test_jit_placed_style.py) is added to cover the placed- style JIT code path, which had no automated test coverage. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Comment on lines
+187
to
+188
| f.write( | ||
| """extern "C" { |
Contributor
There was a problem hiding this comment.
[black] reported by reviewdog 🐶
Suggested change
| f.write( | |
| """extern "C" { | |
| f.write("""extern "C" { |
Comment on lines
+192
to
+193
| }""" | ||
| ) |
Contributor
There was a problem hiding this comment.
[black] reported by reviewdog 🐶
Suggested change
| }""" | |
| ) | |
| }""") |
25 tasks
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
1 participant
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.
Remove is_placed flag from JIT decorator
Summary
returns None, the module was built implicitly via @device/@core/@runtime_sequence context managers (placed style); if it returns a module object, that
module is used directly (unplaced style via Program.resolve_program).
automated test coverage.
Test plan
test_algorithms
🤖 Generated with https://claude.com/claude-code