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

[JulianCloudNTH]

Stack from ghstack (oldest at bottom):

• -> [ExecuTorch][WebGPU] Convert remaining native tests to GTest #20706
• [ExecuTorch][WebGPU] Use requiredFeatures instance API on native + emscripten Dawn #20652
• [ExecuTorch][WebGPU] 2D-fold mul + permute dispatch (lift 65535 1D cap) #20651
• [ExecuTorch][WebGPU] 2D compute dispatch tests — prefill golden + fold unit test #20584
• [ExecuTorch][WebGPU] 2D compute dispatch — lift the 65535 per-dim cap (prefill path) #20583
• [ExecuTorch][WebGPU] Dynamic-shape integration test (allocate-at-max + per-op resize) #20582
• [ExecuTorch][WebGPU] Dynamic resize hook for slice_copy (dynamic start/end) #20581
• [ExecuTorch][WebGPU] Dynamic resize hook for SDPA (live seq-len S) #20580
• [ExecuTorch][WebGPU] Dynamic resize hook for view_copy #20579
• [ExecuTorch][WebGPU] Dynamic resize hooks for sigmoid and select_copy #20578
• [ExecuTorch][WebGPU] Dynamic resize hooks for add and mul #20577
• [ExecuTorch][WebGPU] Dynamic resize hook for linear_q4gsw #20576
• [ExecuTorch][WebGPU] Dynamic resize hooks for rms_norm, embedding, rope #20575
• [ExecuTorch][WebGPU] Dynamic tensor-shape resize engine core #20574
• [ExecuTorch][WebGPU] SymInt arithmetic ops (add/sub/mul/floordiv) for dynamic shapes #20573

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). 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.

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).
• 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; the existing probe/build/run logic is unchanged.

Differential Revision: D110536636

[pytorch-bot]

🔗 Helpful Links

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

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 102e7c6 with merge base 73c259e ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

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]

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

[claude]

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