Dev1/winskuo/gh /aihub remove code by winskuo-quic · Pull Request #2 · CodeLinaro/executorch

winskuo-quic · 2026-04-27T02:46:06Z

Summary

[PLEASE REMOVE] See CONTRIBUTING.md's Pull Requests for ExecuTorch PR guidelines.

[PLEASE REMOVE] If this PR closes an issue, please add a Fixes #<issue-id> line.

[PLEASE REMOVE] If this PR introduces a fix or feature that should be the upcoming release notes, please add a "Release notes: " label. For a list of available release notes labels, check out CONTRIBUTING.md's Pull Requests.

Test plan

[PLEASE REMOVE] How did you test this PR? Please write down any manual commands you used and note down tests that you have written if applicable.

Summary: Attempt 3 to check numel and nbytes overflow. This time we defer checking dynamic sized inputs until their size is realized. Reviewed By: lucylq Differential Revision: D98148157

@larryliu0820

…inear (pytorch#19117) ## Summary The bilinear grid_sampler_2d portable kernel computes interpolation weights via subtractions like `(ix_se - ix)` where both operands are close integer-valued coordinates in pixel space. In fp16 (10 bits of mantissa) that's classic catastrophic cancellation — the result has only a handful of significant bits. The downstream weighted-sum accumulation then loses further precision. Measured on a unit test exercising interior grid points with fp16 inputs, the kernel drifts by ~0.1 absolute from an fp32 reference. That's visible as incorrect depth / flow output near non-integer sample points, which is most of them. ## Fix An `AccType<CTYPE>` trait mapping `Half` and `BFloat16` to `float`, leaving every other dtype unchanged. Used for intermediate coordinate, weight computation, and `out_val` accumulation. Loads cast `CTYPE -> ACC`; the final store casts `ACC -> CTYPE` once. Only internal math is promoted; memory layout / public API / tensor dtypes are unchanged. ```cpp template <typename CTYPE> using AccType = std::conditional_t< std::is_same_v<CTYPE, executorch::aten::Half> || std::is_same_v<CTYPE, executorch::aten::BFloat16>, float, CTYPE>; ``` ## Effects - **fp32 / Int / any non-half dtype**: `AccType<T>` is `T`, so the generated code is byte-identical. No behavior change. - **Half / BFloat16**: `max_abs` vs an fp32 reference drops from **~0.1 to 0** on the shapes I tested (N=1..2, C=7..64, H/W up to 96, both `align_corners` values). - **Perf**: a handful of fp16↔fp32 conversions per output element. Not measurable at op level; well within the portable kernel's scalar cost envelope. ## Scope Only touches the bilinear interpolation path. The nearest-mode path doesn't do weighted-sum accumulation and doesn't have the cancellation issue — left alone in this change. ## Test plan - [x] Builds clean for Android arm64 and host (Apple Clang 21). - [x] Verified numerically via a standalone harness that runs the kernel with matched fp32 / fp16 inputs and compares against an fp32-then-downcast reference. All shapes pass within a single fp16 ULP (or are bit-exact). fp32 tests remain bit-identical to the pre-change kernel. - [x] Existing `kernels/test/op_grid_sampler_2d_test.cpp` unit tests continue to pass (both fp32 shapes that were previously tested, and the fp16 path I'm specifically fixing). Happy to add an fp16-specific test case to `op_grid_sampler_2d_test.cpp` if useful for CI coverage here — just let me know the preferred approach. cc @larryliu0820 @manuelcandales

Differential Revision: D101728720 Pull Request resolved: pytorch#19052

@metascroy

### Summary Fixespytorch#18924 Extends `aten.bitwise_not` support in the MLX delegate to handle integer tensors, not just boolean tensors. Previously the handler only dispatched to `LogicalNotNode` for `bool` and raised `NotImplementedError` for all other dtypes. This adds a dedicated `BitwiseInvertNode` backed by `mlx::core::bitwise_invert`, and updates the handler to dispatch based on dtype: - `bool` → `LogicalNotNode` (unchanged) - `int32`, `int64` → `BitwiseInvertNode` #### Changes: - `serialization/schema.fbs`: add `BitwiseInvertNode` table and append to `OpNode` union - `runtime/MLXInterpreter.h`: add `exec_bitwise_invert()` and dispatch case - `ops.py`: update `_bitwise_not_handler` to dispatch to `BitwiseInvertNode` for integers - `test/test_ops.py`: add `bitwise_not_int` test for `int32` and `int64` ### Test plan All tests were ran on a machine with an Apple M1 Pro CPU, macOS 26.4.1. - `python3 -m py_compile backends/mlx/ops.py backends/mlx/test/test_ops.py` - `python3 backends/mlx/serialization/generate.py` - `python3 -m executorch.backends.mlx.test.run_all_tests bitwise_not_int` ### Test output ``` ============================================================ TEST SUMMARY ============================================================ Passed: 6 Failed: 0 ============================================================ ``` cc @metascroy --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Scott Roy <161522778+metascroy@users.noreply.github.com>

Differential Revision: D102070375 Pull Request resolved: pytorch#19074

pytorch#19075 failed to cp to main

Add Closeable interface so Module can be used with try-with-resources. close() delegates to destroy(). Also make destroy() idempotent by checking mHybridData.isValid() before calling resetNative(), satisfying the Closeable contract. This commit was authored with the help of Claude.

TrainingModule: implement Closeable, replace Log.e + silent empty returns with IllegalStateException throws. Add checkNotDestroyed() guard on all public methods. SGD: throw IllegalStateException instead of bare RuntimeException when optimizer is destroyed. AsrModule: throw ExecutorchRuntimeException instead of bare RuntimeException on transcription failure. ExecuTorchRuntime.validateFilePath: throw IllegalArgumentException instead of bare RuntimeException, with descriptive message. JNI constructors: wrap ExecuTorchJni and ExecuTorchLlmJni constructor bodies in try-catch so C++ exceptions become ExecutorchRuntimeException instead of generic RuntimeException. This commit was authored with the help of Claude. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Differential Revision: D102425537 Pull Request resolved: pytorch#19128

…_nhwc operator + tests Differential Revision: D96507563 Pull Request resolved: pytorch#18479

…ch#19028 (pytorch#19133) ## Summary Reverts the following Android PRs: - pytorch#19099 — Android: consistent error types across all modules - pytorch#19124 — Android: Module implements Closeable - pytorch#19092 — Android: improve error diagnostics for LlmModule and exceptions - pytorch#19028 — Ignored Module tests: provide required input tensor Authored with Claude.

Differential Revision: D102488314 Pull Request resolved: pytorch#19134

Differential Revision: D102385104 Pull Request resolved: pytorch#19122

Differential Revision: D102493794 Pull Request resolved: pytorch#19136

…rch#19137) Differential Revision: D102493838 Pull Request resolved: pytorch#19137

…ualcomm/README.md

Add observe_pass, a decorator that wraps any PassBase subclass or callable pass to automatically collect input/output graphs via Observatory. Names are derived from the class/function name and deduplicated by collect() itself (#2, pytorch#3, ...) so repeated calls never silently overwrite records. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

… region Adds the third top-level Region in the typical AOT+device flow: Session "<script-name>" ├── quantization/ (pipeline_graph_collector, AOT) ├── edge/ (pipeline_graph_collector, AOT) └── device/ (this lens; on-device runtime work) ├── adb.execute #1 ├── adb.execute #2 └── ... Region opens lazily on the first patched SimpleADB operation (via `_ensure_device_region`, called from `note_simple_adb`), so a CLI run with no on-device inference doesn't get an empty group cluttering the tree view. The region is closed in `on_session_end` via a `contextlib.ExitStack`. Each inference record (`adb.execute #N`) lives **directly** under `device` — no per-call sub-region — so the region holds multiple records as designed in RFC §4.5. Disabled lens (`config={"adb": {"enabled": False}}`) skips region opening entirely. tests/test_adb_lens_regions.py (new): - 6 tests covering: lazy device-region opening, first SimpleADB call triggers it, multiple records share the one region, disabled-lens does not open it, on_session_end closes the ExitStack, idempotent _ensure_device_region. Verification: source ~/executorch/.venv-11/bin/activate source ~/executorch/qaisw-*/bin/envsetup.sh export QNN_SDK_ROOT=~/executorch/qaisw-v2.41.0.251111144900_191416 PYTHONPATH=~ python -m pytest \ ~/executorch/backends/qualcomm/debugger/observatory/tests/ -v -> 19 passed (6 new + 13 existing AdbLens tests). Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

In compare mode each archive column already shows the label in its sticky header (commit 5ae2a72), so the `<label>/` that compare_archives prepends to every record and session name renders as redundant noise inside the column and squeezes the natural-width layout from c9e7bda. Strip is display-only — `rec.name` / `session.name` in `state.data` stay prefixed so graph_ref resolution and the global record map keep working. A `displayPrefix` arg threads through `renderRecordItem`, `_renderSessionDashboardLink`, `renderIndexFlat`, and `renderIndexTree`; `_renderArchiveColumn` passes `archive.label`. Single-archive paths pass no prefix and are unchanged. Collision suffixes (`A/foo #2`) strip cleanly to `foo #2` since only the head matches. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Add observe_pass, a decorator that wraps any PassBase subclass or callable pass to automatically collect input/output graphs via Observatory. Names are derived from the class/function name and deduplicated by collect() itself (#2, pytorch#3, ...) so repeated calls never silently overwrite records. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

… region Adds the third top-level Region in the typical AOT+device flow: Session "<script-name>" ├── quantization/ (pipeline_graph_collector, AOT) ├── edge/ (pipeline_graph_collector, AOT) └── device/ (this lens; on-device runtime work) ├── adb.execute #1 ├── adb.execute #2 └── ... Region opens lazily on the first patched SimpleADB operation (via `_ensure_device_region`, called from `note_simple_adb`), so a CLI run with no on-device inference doesn't get an empty group cluttering the tree view. The region is closed in `on_session_end` via a `contextlib.ExitStack`. Each inference record (`adb.execute #N`) lives **directly** under `device` — no per-call sub-region — so the region holds multiple records as designed in RFC §4.5. Disabled lens (`config={"adb": {"enabled": False}}`) skips region opening entirely. tests/test_adb_lens_regions.py (new): - 6 tests covering: lazy device-region opening, first SimpleADB call triggers it, multiple records share the one region, disabled-lens does not open it, on_session_end closes the ExitStack, idempotent _ensure_device_region. Verification: source ~/executorch/.venv-11/bin/activate source ~/executorch/qaisw-*/bin/envsetup.sh export QNN_SDK_ROOT=~/executorch/qaisw-v2.41.0.251111144900_191416 PYTHONPATH=~ python -m pytest \ ~/executorch/backends/qualcomm/debugger/observatory/tests/ -v -> 19 passed (6 new + 13 existing AdbLens tests). Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

In compare mode each archive column already shows the label in its sticky header (commit 5ae2a72), so the `<label>/` that compare_archives prepends to every record and session name renders as redundant noise inside the column and squeezes the natural-width layout from c9e7bda. Strip is display-only — `rec.name` / `session.name` in `state.data` stay prefixed so graph_ref resolution and the global record map keep working. A `displayPrefix` arg threads through `renderRecordItem`, `_renderSessionDashboardLink`, `renderIndexFlat`, and `renderIndexTree`; `_renderArchiveColumn` passes `archive.label`. Single-archive paths pass no prefix and are unchanged. Collision suffixes (`A/foo #2`) strip cleanly to `foo #2` since only the head matches. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

### Summary - fill the test gap for newly added op - utils test ### Test plan pytest backends/qualcomm/tests/rework/utils/test.py

…2095) ### Summary ETDump records an intermediate tensor by handing the tensor's data pointer to a data sink, and every sink reads those bytes with a plain host read: ```cpp memcpy(cur_data_begin, ptr, length); ``` When the tensor lives on an accelerator that pointer is not host memory, so the read segfaults. A program placed on CUDA crashes as soon as tracing is turned on, which is exactly when someone is trying to debug it. Returning an error instead would not help. All four callers wrap the result in `ET_CHECK_MSG`, so an error aborts the process rather than skipping the tensor. This change brings the data back to host memory first. When the tensor is not on CPU, ETDump looks up the allocator registered for that device type, stages the bytes into a temporary host buffer with `copy_device_to_host`, writes that buffer to the sink and frees it. A tensor on CPU keeps the old path and copies nothing extra. If no allocator is registered for the device, ETDump now reports `NotFound` and logs the device type instead of reading the pointer anyway. ### Test plan Two new test files, each with a CMake target and a Buck target. `devtools/etdump/tests/etdump_device_test.cpp` registers the existing `MockCudaAllocator`, which backs its device memory with host memory, and has two tests. One logs a tensor tagged as CUDA and checks both that ETDump went through the allocator and that the bytes reached the debug buffer. The other logs a tensor on CPU and checks that the allocator was not used at all, so the CPU path is unchanged. `devtools/etdump/tests/etdump_device_no_allocator_test.cpp` covers the case where nothing is registered for the device. The registry is a process wide static with no way to remove an entry, so that case needs a binary that never registers anything, which is why it is a second file. `devtools/etdump/tests/CMakeLists.txt` was not referenced by any parent `CMakeLists.txt`, so nothing in that directory was built by CMake. This adds `add_subdirectory(tests)` to `devtools/etdump/CMakeLists.txt` under `BUILD_TESTING`, so both new tests are picked up by `ctest`, which is how the C++ tests run. The pre-existing `sdk_etdump_tests` target stays out of the CMake build. It compiles `etdump_test.cpp`, which includes `etdump_filter.h`, which needs re2, and the devtools build does not pull re2 in. It is now guarded on re2 being available rather than being silently unreachable. With this change both new binaries pass under `ctest`: ``` 1/2 Test #1: etdump_device_test ................ Passed 2/2 Test #2: etdump_device_no_allocator_test ... Passed ``` With `etdump_flatcc.cpp` reverted to the old code and everything rebuilt, both fail: ``` Expected equality of these values: g_mock_cuda.d2h_count_ Which is: 0 1 Death test: etdump_gen.log_evalue(EValue(tensor)) Result: failed to die. ``` Also reproduced the real crash on one NVIDIA H100, with a small program that allocates through `cudaMalloc`, tags a tensor as CUDA and logs it: ``` before: Segmentation fault (core dumped) after: debug buffer holds 1.5 2.5 3.5 4.5 ``` Checked that `etdump_flatcc.cpp` still compiles with `-DUSE_ATEN_LIB`. `clang-format` reports no changes needed on the four touched C++ files. ### Landing order pytorch#22058 adds device-planned arenas to the Python bindings that ETDump can be handed pointers into, and without this change `BufferDataSink::write` would `memcpy` device memory from the host. This should land before or together with it. ### Not covered The ATen mode branch of the device type conversion only compiles. There is no ATen mode CMake build to run it in, so nothing here executes it. It is reachable in principle: `runtime/executor/tensor_parser_aten.cpp` reads the serialized device type and index, including CUDA, and builds the ATen tensor on that device. An earlier version of this description said ATen tensors carry no device metadata, which is wrong. `LogTensorOnCpuDoesNotStageThroughTheAllocator` passes with the production change reverted, since CPU tensors already went straight to the data sink. It documents the CPU path rather than locking the fix; the other two tests are the ones that require the new branch. --------- Co-authored-by: r <r@e>

winskuo-quic · 2026-04-27T02:46:06Z

Summary

[PLEASE REMOVE] See CONTRIBUTING.md's Pull Requests for ExecuTorch PR guidelines.

[PLEASE REMOVE] If this PR closes an issue, please add a Fixes #<issue-id> line.

[PLEASE REMOVE] If this PR introduces a fix or feature that should be the upcoming release notes, please add a "Release notes: " label. For a list of available release notes labels, check out CONTRIBUTING.md's Pull Requests.

Test plan

[PLEASE REMOVE] How did you test this PR? Please write down any manual commands you used and note down tests that you have written if applicable.

Summary: Attempt 3 to check numel and nbytes overflow. This time we defer checking dynamic sized inputs until their size is realized. Reviewed By: lucylq Differential Revision: D98148157

@larryliu0820

…inear (pytorch#19117) ## Summary The bilinear grid_sampler_2d portable kernel computes interpolation weights via subtractions like `(ix_se - ix)` where both operands are close integer-valued coordinates in pixel space. In fp16 (10 bits of mantissa) that's classic catastrophic cancellation — the result has only a handful of significant bits. The downstream weighted-sum accumulation then loses further precision. Measured on a unit test exercising interior grid points with fp16 inputs, the kernel drifts by ~0.1 absolute from an fp32 reference. That's visible as incorrect depth / flow output near non-integer sample points, which is most of them. ## Fix An `AccType<CTYPE>` trait mapping `Half` and `BFloat16` to `float`, leaving every other dtype unchanged. Used for intermediate coordinate, weight computation, and `out_val` accumulation. Loads cast `CTYPE -> ACC`; the final store casts `ACC -> CTYPE` once. Only internal math is promoted; memory layout / public API / tensor dtypes are unchanged. ```cpp template <typename CTYPE> using AccType = std::conditional_t< std::is_same_v<CTYPE, executorch::aten::Half> || std::is_same_v<CTYPE, executorch::aten::BFloat16>, float, CTYPE>; ``` ## Effects - **fp32 / Int / any non-half dtype**: `AccType<T>` is `T`, so the generated code is byte-identical. No behavior change. - **Half / BFloat16**: `max_abs` vs an fp32 reference drops from **~0.1 to 0** on the shapes I tested (N=1..2, C=7..64, H/W up to 96, both `align_corners` values). - **Perf**: a handful of fp16↔fp32 conversions per output element. Not measurable at op level; well within the portable kernel's scalar cost envelope. ## Scope Only touches the bilinear interpolation path. The nearest-mode path doesn't do weighted-sum accumulation and doesn't have the cancellation issue — left alone in this change. ## Test plan - [x] Builds clean for Android arm64 and host (Apple Clang 21). - [x] Verified numerically via a standalone harness that runs the kernel with matched fp32 / fp16 inputs and compares against an fp32-then-downcast reference. All shapes pass within a single fp16 ULP (or are bit-exact). fp32 tests remain bit-identical to the pre-change kernel. - [x] Existing `kernels/test/op_grid_sampler_2d_test.cpp` unit tests continue to pass (both fp32 shapes that were previously tested, and the fp16 path I'm specifically fixing). Happy to add an fp16-specific test case to `op_grid_sampler_2d_test.cpp` if useful for CI coverage here — just let me know the preferred approach. cc @larryliu0820 @manuelcandales

Differential Revision: D101728720 Pull Request resolved: pytorch#19052

@metascroy