[Maintain][Manifest] add at least 2 workloads to 21 binary elementwise ops#1392
Merged
lcy-seso merged 2 commits intotile-ai:testbedfrom May 9, 2026
Merged
Conversation
…e ops AddFwdOp/SubFwdOp and 19 sibling binary ops in elementwise_binary.yaml had empty workloads while status was implemented, causing test_every_op_has_at_least_two_workloads to fail in gpu-smoke. Each affected op now declares two representative workloads: - LLM hidden-state prefill: input/other 2048x4096 (no broadcast) - CNN feature map with channel-wise broadcast: input 16x256x56x56, other 256x1x1 Dtype matrix matches each op's signature: float ops use float16/bfloat16, bitwise ops use int32/int64, logical ops use bool. No signature, roofline, params, shape_rules, or dtype_combos changes. Manifest-only PR per the trust-model carve-out. Co-Authored-By: Ibuki 🍃 — a wind born from GPTs <Ibuki-wind@users.noreply.github.com>
lcy-seso
added
automated
github-actions
Bot
added
maintain
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request populates the workloads section for numerous element-wise binary operations in the manifest. The review feedback identifies critical mismatches in workload keys for PowFwdOp and LerpFwdOp, where the keys should align with the specific input names exponent and end rather than other_shape. Furthermore, the reviewer suggests expanding the dtypes lists for several operations to include all types supported by their signatures, such as float32 for arithmetic ops and additional types for logical and bitwise operations, to improve test and benchmark coverage.
There was a problem hiding this comment.
Pull request overview
This PR addresses a gpu-smoke manifest schema failure (test_every_op_has_at_least_two_workloads) by populating previously-empty workloads: lists for a set of implemented binary elementwise ops in the manifest, enabling downstream tooling/benchmarks to consume representative shapes/dtypes.
Changes:
- Added two representative workload entries (LLM hidden-state + CNN broadcast) to 21 binary elementwise ops in
tileops/manifest/elementwise_binary.yaml. - Standardized dtype selections per op category (float, logical, bitwise) within the new workload entries.
- PowFwdOp workloads use exponent_shape (matches signature input name) - LerpFwdOp workloads use end_shape (matches signature input name) - Add float32 to dtype lists where supported by signature - Expand LogicalAnd/Or workload dtypes to include float types - Expand BitwiseAnd/Or/Xor workload dtypes to include bool Co-Authored-By: Ibuki 🍃 — a wind born from GPTs <Ibuki-wind@users.noreply.github.com>
Ibuki-wind
requested changes
May 9, 2026
Contributor
Ibuki-wind
left a comment
Ibuki-wind
left a comment
There was a problem hiding this comment.
Overall
The manifest update still needs a signature-aligned cleanup pass before it satisfies the workload contract.
Merged
2 tasks
Ibuki-wind
added a commit
that referenced
this pull request
May 9, 2026
… (cherry-pick #1392) (#1395) ## Summary Cherry-pick of #1392 (already merged to `testbed` as `67e10ad`) onto `main`. `main` CI is currently red because it lacks this manifest fix. Landing the same patch on `main` directly unblocks CI without waiting for the next `testbed` → `main` rollup. ## Why this is conflict-safe for the eventual testbed → main merge The cherry-pick is byte-identical to `67e10ad`. When `testbed` later merges into `main`, the 3-way merge sees the same change on both sides and resolves automatically — no duplicate diff, no conflict (assuming no further `testbed` commits touch the same lines). ## Test plan - [ ] CI green on this PR - [ ] After merge, confirm `main` unit tests pass Co-authored-by: Ibuki 🍃 — a wind born from GPTs <Ibuki-wind@users.noreply.github.com>
This was referenced May 9, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fix gpu-smoke
test_every_op_has_at_least_two_workloadsfailure by populating emptyworkloads:lists on 21 implemented binary elementwise ops (`AddFwdOp`, `SubFwdOp`, `MulFwdOp`, `DivFwdOp`, `RemainderFwdOp`, `PowFwdOp`, `FloorDivideFwdOp`, `LerpFwdOp`, `MaximumFwdOp`, `MinimumFwdOp`, `Eq/Ne/Gt/Lt/Ge/LeFwdOp`, `LogicalAnd/OrFwdOp`, `BitwiseAnd/Or/XorFwdOp`).This pre-existing testbed CI break has been blocking nightshift auto-merge for PRs #1387, #1388, #1389, #1391. Filed as issue #1390 — this PR resolves it.
Plan executed
Each affected op now declares two representative workloads:
Dtype matrix matches each op's signature:
Test plan
Acceptance Criteria
Closes #1390.
Trust model
Manifest-only PR. No `signature`, `roofline`, `params`, `shape_rules`, or `dtype_combos` edits. Falls within the trust-model rule that workload edits without a status flip require a separate manifest-only PR with human review — this is that PR.