iframe-proxy

JulianCloudNTH · 2026-07-02T23:49:30Z

Stack from ghstack (oldest at bottom):

Convert the remaining hand-rolled int main() + printf/bool ok native tests to GTest so the whole backends/webgpu/test/ suite is uniform, filterable via --gtest_filter, and self-reporting (extends the GTest conversion already applied to test_dynamic_shape). The five converted files are a harness-only change — every test case, tensor shape, tolerance, artifact filename, and skip condition is preserved 1:1, only the pass/fail reporting mechanism changes — and this diff additionally wires the already-GTest webgpu_dynamic_shape_test into the CI runner so the dynamic-shape suite actually executes.

Key changes:

test/test_webgpu_native.cpp, test/native/test_dispatch_order.cpp, test/native/test_index.cpp, test/native/test_scratch_buffer.cpp, test/native/test_update_cache.cpp — main+printf/bool ok accumulator → TEST() cases using EXPECT_*/ASSERT_*; each keeps a custom main() that brings up the WebGPU device once then RUN_ALL_TESTS() (device-absent still SKIPs by returning 0). test_index/test_webgpu_native use inclusive EXPECT_LE(err, tol) to match the original err > tol fail gate exactly.
CMakeLists.txt — move every native-test target into the if(TARGET GTest::gtest) block, linking GTest::gtest.
scripts/test_webgpu_native_ci.sh — add -DEXECUTORCH_BUILD_TESTS=ON to the native-test configure so the now-gtest-gated targets are defined, and wire webgpu_dynamic_shape_test into the runner: export its .ptes + goldens via export_dynamic_shape_cases, add it to the built/run target list behind the same --target help probe, and run it guarded (mirroring the index test).
test/test_build_webgpu.sh — add -DEXECUTORCH_BUILD_TESTS=ON so the local build script (which builds the now-gtest-gated targets unconditionally) still finds them.
@exported-using-ghexport

Differential Revision: D110536636

[ghstack-poisoned]

pytorch-bot · 2026-07-02T23:49:34Z

github-actions · 2026-07-02T23:53:55Z

This PR needs a `release notes:` label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

JulianCloudNTH · 2026-07-03T00:40:21Z

@claude Review thoroughly. Check for stale code / bugs.

claude · 2026-07-03T03:20:21Z

Claude finished @JulianCloudNTH's task in 8m 7s —— View job

Review: WebGPU native tests → GTest

Reviewed all 7 files against origin/gh/JulianCloudNTH/85/base. The conversion is careful and faithful — cases, shapes, tolerances, artifact names, skip conditions, and the device-absent→return 0 SKIP behavior are preserved, and every deleted sweep wrapper maps cleanly onto a TEST(). ASSERT_* inside the void helpers is the correct pattern here (it returns from the helper, matching the old return false). One real bug and a couple of minor notes below.

🔴 Sibling build script will break: `test_build_webgpu.sh` still lacks `-DEXECUTORCH_BUILD_TESTS=ON`

This PR moves webgpu_native_test, webgpu_dispatch_order_test, webgpu_scratch_buffer_test, webgpu_update_cache_test, webgpu_index_test inside the if(TARGET GTest::gtest) block (CMakeLists.txt:190-209). GTest::gtest only exists when googletest is added as a subdirectory, which happens only under BUILD_TESTING/EXECUTORCH_BUILD_TESTS:

third-party/CMakeLists.txt:18   if(BUILD_TESTING) add_subdirectory(googletest) endif()
CMakeLists.txt:253              if(EXECUTORCH_BUILD_TESTS) include(CTest) else() set(BUILD_TESTING OFF)

You correctly added -DEXECUTORCH_BUILD_TESTS=ON to scripts/test_webgpu_native_ci.sh. But the sibling backends/webgpu/test/test_build_webgpu.sh — the documented local end-to-end script — configures without it (test_build_webgpu.sh:84-95) and then builds these targets unconditionally:

cmake --build "${NATIVE_BUILD_DIR}" --target webgpu_native_test -j${NPROC}          # :97
cmake --build "${NATIVE_BUILD_DIR}" --target webgpu_dispatch_order_test -j${NPROC}  # :98
cmake --build "${NATIVE_BUILD_DIR}" --target webgpu_scratch_buffer_test -j${NPROC}  # :99

Unlike the CI script, this one does not probe --target help first, so if no system GTest is picked up by the find_package(GTest QUIET) fallback, these targets won't be defined and the build fails with "unknown target". Before this PR the targets were defined regardless of EXECUTORCH_BUILD_TESTS, so the script worked. Add -DEXECUTORCH_BUILD_TESTS=ON to the configure at test_build_webgpu.sh:84. Fix this →

🟡 Minor: pass/fail boundary changed from inclusive to exclusive in a few tests

Several gates moved from if (err > tol) FAIL (pass iff <= tol) to EXPECT_LT(err, tol) (pass iff < tol), e.g. test_index.cpp (EXPECT_LT(max_abs_err, 1e-3f) / EXPECT_LT(max_rel_err, 1e-3f)), test_update_cache.cpp / test_webgpu_native.cpp update_cache (EXPECT_LT(max_abs_err, 1e-3f)). The AND/OR logic is preserved; only the exact-boundary case (err == 1e-3f) flips from pass to fail. This is essentially never hit in practice and is stricter, not looser — but it's a deviation from the "tolerance preserved 1:1" claim. dispatch_order avoided this by keeping the exact expression via EXPECT_FALSE(a && b); the others could do the same with EXPECT_LE if strict 1:1 is desired. Non-blocking.

🟢 Things I checked that are correct

test_query_pool_roundtrip(*get_default_webgpu_context()) — safe: RUN_ALL_TESTS() only runs after set_default_webgpu_context(&ctx), and the device-absent path returns before it.
SDPA expect_reject path (ASSERT_NE(err, Error::Ok) + return) preserves the original PASS/FAIL semantics.
SdpaAllFamiliesRanWhenDirSet recomputes .pte presence independently but is equivalent to the old sdpa_ran && replay_ran && dyn_ran && incache_ran guard.
required/heavy gating in the q4gsw & SDPA sweeps correctly maps FAIL → ADD_FAILURE() (non-fatal, so the loop continues, matching the old ok = ... && ok) and SKIP → printf/GTEST_SKIP.
quant_within_tol moved to a full definition (forward-decl removed) with identical body; no other refs dangling.
CI script --target help probe still guards the CI path, so the webgpu_index_test move into the gated block is covered there.

Nice, mechanical, well-commented conversion overall — the one thing to fix before landing is the test_build_webgpu.sh configure flag.
• gh/JulianCloudNTH/85/head

[ghstack-poisoned]

Pull Request resolved: #20706 Convert the remaining hand-rolled `int main()` + printf/`bool ok` native tests to GTest so the whole `backends/webgpu/test/` suite is uniform, filterable via `--gtest_filter`, and self-reporting (extends the GTest conversion already applied to `test_dynamic_shape`). The five converted files are a harness-only change — every test case, tensor shape, tolerance, artifact filename, and skip condition is preserved 1:1, only the pass/fail reporting mechanism changes — and this diff additionally wires the already-GTest `webgpu_dynamic_shape_test` into the CI runner so the dynamic-shape suite actually executes. Key changes: - `test/test_webgpu_native.cpp`, `test/native/test_dispatch_order.cpp`, `test/native/test_index.cpp`, `test/native/test_scratch_buffer.cpp`, `test/native/test_update_cache.cpp` — `main`+`printf`/`bool ok` accumulator → `TEST()` cases using `EXPECT_*`/`ASSERT_*`; each keeps a custom `main()` that brings up the WebGPU device once then `RUN_ALL_TESTS()` (device-absent still SKIPs by returning 0). `test_index`/`test_webgpu_native` use inclusive `EXPECT_LE(err, tol)` to match the original `err > tol` fail gate exactly. - `CMakeLists.txt` — move every native-test target into the `if(TARGET GTest::gtest)` block, linking `GTest::gtest`. - `scripts/test_webgpu_native_ci.sh` — add `-DEXECUTORCH_BUILD_TESTS=ON` to the native-test configure so the now-gtest-gated targets are defined, and wire `webgpu_dynamic_shape_test` into the runner: export its `.pte`s + goldens via `export_dynamic_shape_cases`, add it to the built/run target list behind the same `--target help` probe, and run it guarded (mirroring the `index` test). - `test/test_build_webgpu.sh` — add `-DEXECUTORCH_BUILD_TESTS=ON` so the local build script (which builds the now-gtest-gated targets unconditionally) still finds them. ghstack-source-id: 399790083 @exported-using-ghexport Differential Revision: [D110536636](https://our.internmc.facebook.com/intern/diff/D110536636/)

[ghstack-poisoned]

Pull Request resolved: #20706 Convert the remaining hand-rolled `int main()` + printf/`bool ok` native tests to GTest so the whole `backends/webgpu/test/` suite is uniform, filterable via `--gtest_filter`, and self-reporting (extends the GTest conversion already applied to `test_dynamic_shape`). The five converted files are a harness-only change — every test case, tensor shape, tolerance, artifact filename, and skip condition is preserved 1:1, only the pass/fail reporting mechanism changes — and this diff additionally wires the already-GTest `webgpu_dynamic_shape_test` into the CI runner so the dynamic-shape suite actually executes. Key changes: - `test/test_webgpu_native.cpp`, `test/native/test_dispatch_order.cpp`, `test/native/test_index.cpp`, `test/native/test_scratch_buffer.cpp`, `test/native/test_update_cache.cpp` — `main`+`printf`/`bool ok` accumulator → `TEST()` cases using `EXPECT_*`/`ASSERT_*`; each keeps a custom `main()` that brings up the WebGPU device once then `RUN_ALL_TESTS()` (device-absent still SKIPs by returning 0). `test_index`/`test_webgpu_native` use inclusive `EXPECT_LE(err, tol)` to match the original `err > tol` fail gate exactly. - `CMakeLists.txt` — move every native-test target into the `if(TARGET GTest::gtest)` block, linking `GTest::gtest`. - `scripts/test_webgpu_native_ci.sh` — add `-DEXECUTORCH_BUILD_TESTS=ON` to the native-test configure so the now-gtest-gated targets are defined, and wire `webgpu_dynamic_shape_test` into the runner: export its `.pte`s + goldens via `export_dynamic_shape_cases`, add it to the built/run target list behind the same `--target help` probe, and run it guarded (mirroring the `index` test). - `test/test_build_webgpu.sh` — add `-DEXECUTORCH_BUILD_TESTS=ON` so the local build script (which builds the now-gtest-gated targets unconditionally) still finds them. ghstack-source-id: 399809713 @exported-using-ghexport Differential Revision: [D110536636](https://our.internmc.facebook.com/intern/diff/D110536636/)

[ghstack-poisoned]

Pull Request resolved: #20706 Convert the remaining hand-rolled `int main()` + printf/`bool ok` native tests to GTest so the whole `backends/webgpu/test/` suite is uniform, filterable via `--gtest_filter`, and self-reporting (extends the GTest conversion already applied to `test_dynamic_shape`). The five converted files are a harness-only change — every test case, tensor shape, tolerance, artifact filename, and skip condition is preserved 1:1, only the pass/fail reporting mechanism changes — and this diff additionally wires the already-GTest `webgpu_dynamic_shape_test` into the CI runner so the dynamic-shape suite actually executes. Key changes: - `test/test_webgpu_native.cpp`, `test/native/test_dispatch_order.cpp`, `test/native/test_index.cpp`, `test/native/test_scratch_buffer.cpp`, `test/native/test_update_cache.cpp` — `main`+`printf`/`bool ok` accumulator → `TEST()` cases using `EXPECT_*`/`ASSERT_*`; each keeps a custom `main()` that brings up the WebGPU device once then `RUN_ALL_TESTS()` (device-absent still SKIPs by returning 0). `test_index`/`test_webgpu_native` use inclusive `EXPECT_LE(err, tol)` to match the original `err > tol` fail gate exactly. - `CMakeLists.txt` — move every native-test target into the `if(TARGET GTest::gtest)` block, linking `GTest::gtest`. - `scripts/test_webgpu_native_ci.sh` — add `-DEXECUTORCH_BUILD_TESTS=ON` to the native-test configure so the now-gtest-gated targets are defined, and wire `webgpu_dynamic_shape_test` into the runner: export its `.pte`s + goldens via `export_dynamic_shape_cases`, add it to the built/run target list behind the same `--target help` probe, and run it guarded (mirroring the `index` test). - `test/test_build_webgpu.sh` — add `-DEXECUTORCH_BUILD_TESTS=ON` so the local build script (which builds the now-gtest-gated targets unconditionally) still finds them. ghstack-source-id: 399812941 @exported-using-ghexport Differential Revision: [D110536636](https://our.internmc.facebook.com/intern/diff/D110536636/)

Update

102e7c6

[ghstack-poisoned]

JulianCloudNTH requested review from kirklandsign and larryliu0820 as code owners July 2, 2026 23:49

JulianCloudNTH temporarily deployed to cadence July 2, 2026 23:49 — with GitHub Actions Inactive

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jul 2, 2026

Update

ec749ff

[ghstack-poisoned]

JulianCloudNTH had a problem deploying to cadence July 3, 2026 20:28 — with GitHub Actions Error

JulianCloudNTH temporarily deployed to cadence July 3, 2026 20:28 — with GitHub Actions Inactive

meta-codesync Bot added the meta-exported label Jul 3, 2026

Update

566efe3

[ghstack-poisoned]

JulianCloudNTH temporarily deployed to cadence July 3, 2026 20:52 — with GitHub Actions Inactive

JulianCloudNTH had a problem deploying to cadence July 3, 2026 21:20 — with GitHub Actions Error

JulianCloudNTH requested a review from psiddh July 3, 2026 21:27

Update

004e1d9

[ghstack-poisoned]

JulianCloudNTH temporarily deployed to cadence July 3, 2026 21:37 — with GitHub Actions Inactive

JulianCloudNTH deployed to cadence July 3, 2026 22:07 — with GitHub Actions Active

Sunbelt Computer Software

PL/B Language Development and Support

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[ExecuTorch][WebGPU] Convert remaining native tests to GTest#20706

[ExecuTorch][WebGPU] Convert remaining native tests to GTest#20706
JulianCloudNTH wants to merge 4 commits into
gh/JulianCloudNTH/85/basefrom
gh/JulianCloudNTH/85/head

JulianCloudNTH commented Jul 2, 2026 •

edited

Loading

Uh oh!

pytorch-bot Bot commented Jul 2, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 2, 2026

Uh oh!

JulianCloudNTH commented Jul 3, 2026

Uh oh!

claude Bot commented Jul 3, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Sunbelt Computer Software

PL/B Language Development and Support

Uh oh!

Conversation

JulianCloudNTH commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20706

❌ 1 New Failure

Uh oh!

github-actions Bot commented Jul 2, 2026

This PR needs a release notes: label

Uh oh!

JulianCloudNTH commented Jul 3, 2026

Uh oh!

claude Bot commented Jul 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review: WebGPU native tests → GTest

🔴 Sibling build script will break: test_build_webgpu.sh still lacks -DEXECUTORCH_BUILD_TESTS=ON

🟡 Minor: pass/fail boundary changed from inclusive to exclusive in a few tests

🟢 Things I checked that are correct

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

JulianCloudNTH commented Jul 2, 2026 •

edited

Loading

pytorch-bot Bot commented Jul 2, 2026 •

edited

Loading

This PR needs a `release notes:` label

claude Bot commented Jul 3, 2026 •

edited

Loading

🔴 Sibling build script will break: `test_build_webgpu.sh` still lacks `-DEXECUTORCH_BUILD_TESTS=ON`