[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen by JCNTH · Pull Request #20728 · pytorch/executorch

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

JCNTH · 2026-07-05T01:19:51Z

Stack from ghstack (oldest at bottom):

Additive coverage for the template engine added in the previous diff.

Key additions (test/test_wgsl_codegen.py, WgslTemplateEngineTest):

preprocess$if/$else branch selection + ${...} substitution + guarded-body indentation.
the DTYPE x VEC variant matrix and SUFFIX rules ("" suppresses; defaults to str(VALUE)).
parse_template_spec — minimal-spec expansion, duplicate-key rejection (via the UniqueKeyLoader), top-level-key validation.
the 3 WGSL type-helpers (buffer_scalar_type / buffer_gvec_type / accum_scalar_type).
a byte-identity round-trip reproducing both committed rms_norm headers from the template + rms_norm.yaml.
a source-of-truth sync-lock asserting the shared $-block transpiler helpers and the UniqueKeyLoader stay character-identical to gen_vulkan_spv.py (guards against silent drift).
@exported-using-ghexport

Differential Revision: D110660006

[ghstack-poisoned]

pytorch-bot · 2026-07-05T01:19:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20728

📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

As of commit 090cdc3 with merge base f4b01a8 ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner-mypy (gh)
pull / android / build-android (gh)
pull / test-binary-size-linux-gcc / linux-job (gh)
RuntimeError: Command docker exec -t b4545e369e383e61e93d3c116efb4f9ec77f0eebe9ad42713cb3dc2127ac0cea /exec failed with exit code 1
pull / unittest / linux / linux-job (gh)
RuntimeError: Command docker exec -t 38626472d2dcd78dc934ecbb59fbdfaede195c81badad824a9381242c472882d /exec failed with exit code 1
pull / unittest-editable / linux / linux-job (gh)
RuntimeError: Command docker exec -t a8f7d874cc2a7d8df38e271971df86d02f903d6ed794f0272bfa564a758821e2 /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest-nxp-neutron / linux-job (gh) (trunk failure)
##[error]The operation was canceled.

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

linux-foundation-easycla · 2026-07-05T01:19:57Z

The committers listed above are authorized under a signed CLA.

✅ login: JCNTH / name: Julian Ng-Thow-Hing (2a0675a, 92e5a83)

github-actions · 2026-07-05T01:20:39Z

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.

JCNTH · 2026-07-05T01:23:02Z

@claude throoughly review, check for overengineering, correctness, and areaas for modularization

claude · 2026-07-05T01:23:52Z

Claude finished @JCNTH's task in 2m 40s —— View job

Review: WGSL shader-variant codegen unit tests

Read the test file
Read the codegen source (gen_wgsl_headers.py) under test
Cross-check the sync-lock against gen_vulkan_spv.py
Assess correctness, overengineering, modularization

Solid, well-targeted test suite. The names are descriptive, the docstrings capture the why (the DD-009 enable f16 note and the raw-string-terminator case are especially good), and every assertion I traced lines up with the actual API in gen_wgsl_headers.py. I verified the three shared helpers are byte-identical between the two files, and all files the tests reference (rms_norm.wgsl/.json/.h) exist. Verdict: approve with minor suggestions — nothing blocking.

Note: I reviewed by inspection + cross-file diffing; the sandbox blocked me from actually running pytest, so I did not get a green run locally.

Correctness

_function_source (lines 36–61) is the one piece with real fragility. Its body-end heuristic — "stop at the next non-blank column-0 line" — is correct for the current layout (escape is followed by a # comment, preprocess by class SPVGenerator:, both at column 0), but it will silently truncate if any helper ever gains a line that starts at column 0 inside the body (e.g. a multi-line string, or a ) closing a call at column 0). For a test whose whole purpose is guarding against drift, the parser guarding it shouldn't itself be position-sensitive. See the modularization note for a sturdier approach.
The two-phase signature skip (lines 50–53) correctly handles preprocess's multi-line signature whose ) -> str: sits at column 0 — good catch by the author. Worth noting the loop assumes the first line ending in : is the signature terminator; a default value like x: Dict = {} ending a wrapped signature line would fool it. Not a problem for these three functions, but it's the same class of latent fragility.

Overengineering

Reasonable overall — this isn't gold-plated. The single candidate is the sync-lock's hand-rolled text parser: ~26 lines of column heuristics to enforce char-identity of 3 functions. Since gen_wgsl_headers.pyalready imports these functions, the webgpu side can just use inspect.getsource(getattr(g, fn)) and only the (un-importable) Vulkan side needs source extraction — and that extraction is more robustly done with ast than regex/column-counting. That removes the fragility and shrinks the helper.
Mild redundancy: test_rms_norm_template_roundtrip_byte_identical (416) overlaps test_committed_headers_match_generator (134) — the committed rms_norm headers are already validated through headers_for_shader, which runs the same template expansion. The round-trip test earns its keep by exercising preprocess + render_headerdirectly (bypassing spec parsing) and pinning the concrete variant names, so I'd keep it — just calling out the overlap.

Modularization

Suggested: extract via ast instead of column heuristics.ast.parse(text) + walk for FunctionDef with the target name + ast.get_source_segment(text, node) gives the exact source span with no indentation assumptions, and works identically for both files. This replaces _function_source entirely and is pure stdlib (consistent with the generator's stdlib-only constraint). Fix this →
The two test classes (WgslCodegenTest / WgslTemplateEngineTest) and the module-level loader are cleanly separated. _write_spec is an appropriate local helper. No changes needed here.

Coverage gaps (optional, not blocking)

generate_variant_combinations/parse_template_spec are inherited from the Vulkan engine and have branches the tests don't reach. If webgpu will exercise them, worth a line each:

The int/uint rejection (gen_wgsl_headers.py:240-243) — a one-line assertRaises(ValueError) test.
The RANGE expansion (lines 226-236), including the _1/_2 suffix naming — currently only VALUE is tested.
The invalid variant-key assert in parse_template_spec (line 292) — the top-level-key mismatch is covered, but not an unknown key inside a shader_variants entry.
The combination/combinationN branch (lines 205-221) is untested. Likely out of scope for webgpu today; a log-style comment noting it's intentionally uncovered would prevent a false sense of completeness.

Nit

test_render_header_shape:111 asserts assertNotIn("kUpdateCacheWorkgroupSize ", h) (trailing space) to reject an untyped symbol — clever, but a touch cryptic; a short comment (# no untyped WorkgroupSize symbol, only X/Y/Z) would help the next reader.

None of the above blocks the PR — the suggestions are hardening the drift-guard and closing a few inherited-branch gaps. Nice work on the byte-identity and DD-009 regression cases.
· branch gh/JCNTH/2/head

[ghstack-poisoned]

SS-JIA

Review automatically exported from Phabricator review in Meta.

[ghstack-poisoned]

Pull Request resolved: #20728 Additive coverage for the template engine added in the previous diff. Key additions (`test/test_wgsl_codegen.py`, `WgslTemplateEngineTest`): - `preprocess` `$if`/`$else` branch selection + `${...}` substitution + guarded-body indentation. - the DTYPE x VEC variant matrix and `SUFFIX` rules (`""` suppresses; defaults to `str(VALUE)`). - `parse_template_spec` — minimal-spec expansion, duplicate-key rejection (via the `UniqueKeyLoader`), top-level-key validation. - the 3 WGSL type-helpers (`buffer_scalar_type` / `buffer_gvec_type` / `accum_scalar_type`). - a byte-identity round-trip reproducing both committed `rms_norm` headers from the template + `rms_norm.yaml`. - a source-of-truth sync-lock asserting the shared `$`-block transpiler helpers and the `UniqueKeyLoader` stay character-identical to `gen_vulkan_spv.py` (guards against silent drift). ghstack-source-id: 401515160 @exported-using-ghexport Differential Revision: [D110660006](https://our.internmc.facebook.com/intern/diff/D110660006/)

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Update
2a0675a
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 01:19 — with GitHub Actions Inactive

meta-claBot 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 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 01:47 — with GitHub Actions Inactive

Update
92e5a83
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 02:27 — with GitHub Actions Inactive

meta-codesyncBot added the meta-exported label Jul 5, 2026

JCNTH temporarily deployed to cadence July 5, 2026 02:55 — with GitHub Actions Inactive

Update
3246801
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 04:31 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 5, 2026 04:57 — with GitHub Actions Failure

Update
d3a4296
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 5, 2026 06:18 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 5, 2026 06:46 — with GitHub Actions Inactive

This was referenced Jul 6, 2026
[ExecuTorch][WebGPU] Add f16-multiply variant of the steel q4gsw prefill GEMM #20752
Merged
[ExecuTorch][WebGPU] Test coverage for the f16-multiply steel q4gsw GEMM #20753
Merged

Update
5a8dc5b
[ghstack-poisoned]

This was referenced Jul 7, 2026
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20770
Closed
[ExecuTorch][WebGPU] Add opt-in native f16 KV cache (−280 MiB, default-OFF) #20772
Merged
[ExecuTorch][WebGPU] Test coverage for the f16 KV cache #20773
Merged

JCNTH temporarily deployed to cadence July 7, 2026 18:30 — with GitHub Actions Inactive

SS-JIA approved these changes Jul 8, 2026
View reviewed changes

Update
df81e69
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 03:45 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 04:14 — with GitHub Actions Inactive

Update
d5eb0a8
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 17:53 — with GitHub Actions Inactive

JCNTH had a problem deploying to cadence July 9, 2026 18:22 — with GitHub Actions Error

Update
42983af
[ghstack-poisoned]

JCNTH had a problem deploying to cadence July 9, 2026 18:36 — with GitHub Actions Error

Update
090cdc3
[ghstack-poisoned]

JCNTH temporarily deployed to cadence July 9, 2026 18:58 — with GitHub Actions Inactive

JCNTH temporarily deployed to cadence July 9, 2026 19:27 — with GitHub Actions Inactive

meta-codesyncBot merged commit 706bce5 into gh/JCNTH/2/baseJul 9, 2026
175 of 183 checks passed

meta-codesyncBot deleted the gh/JCNTH/2/head branch July 9, 2026 21:39

meta-codesyncBottemporarily deployed to cherry-pick-bot July 9, 2026 21:39 Inactive

pytorchbot mentioned this pull request Jul 9, 2026
[ExecuTorch][WebGPU] Unit tests for the WGSL shader-variant codegen #20813
Merged

Uh oh!

Conversation

JCNTH commented Jul 5, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-botBot commented Jul 5, 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/20728

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

Uh oh!

linux-foundation-easyclaBot commented Jul 5, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actionsBot commented Jul 5, 2026

This PR needs a release notes: label

Uh oh!

JCNTH commented Jul 5, 2026

Uh oh!

claudeBot commented Jul 5, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review: WGSL shader-variant codegen unit tests

Correctness

Overengineering

Modularization

Coverage gaps (optional, not blocking)

Nit

Uh oh!

SS-JIA left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

JCNTH commented Jul 5, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-botBot commented Jul 5, 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/20728

❌ 5 New Failures, 2 Pending, 1 Unrelated Failure

Uh oh!

linux-foundation-easyclaBot commented Jul 5, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actionsBot commented Jul 5, 2026

This PR needs a release notes: label

Uh oh!

JCNTH commented Jul 5, 2026

Uh oh!

claudeBot commented Jul 5, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review: WGSL shader-variant codegen unit tests

Correctness

Overengineering

Modularization

Coverage gaps (optional, not blocking)

Nit

Uh oh!

SS-JIA left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

JCNTH commented Jul 5, 2026•
edited
Loading

pytorch-botBot commented Jul 5, 2026•
edited
Loading

linux-foundation-easyclaBot commented Jul 5, 2026•
edited
Loading

This PR needs a `release notes:` label

claudeBot commented Jul 5, 2026•
edited
Loading

JCNTH commented Jul 5, 2026•
edited
Loading

pytorch-botBot commented Jul 5, 2026•
edited
Loading

linux-foundation-easyclaBot commented Jul 5, 2026•
edited
Loading

This PR needs a `release notes:` label

claudeBot commented Jul 5, 2026•
edited
Loading

JCNTH commented Jul 5, 2026•
edited
Loading

pytorch-botBot commented Jul 5, 2026•
edited
Loading

linux-foundation-easyclaBot commented Jul 5, 2026•
edited
Loading

This PR needs a `release notes:` label

claudeBot commented Jul 5, 2026•
edited
Loading

JCNTH commented Jul 5, 2026•
edited
Loading

pytorch-botBot commented Jul 5, 2026•
edited
Loading

linux-foundation-easyclaBot commented Jul 5, 2026•
edited
Loading