cpu loop R72: 2D gemm_bias batch2 family (294→298)

[chenxingqiang]

Summary

CPU optimization loop R72 — closes batch=2 inference naming for the full 2D gemm_bias compound family.

Inventory: 294 → 298 (+4 builders)

New builders

Builder	Contract
gemm_bias_batch2	2D [M,K] @ [K,N] + [1,N] bias
gemm_bias_relu_batch2	same + ReLU
gemm_bias_gelu_batch2	same + GELU
gemm_bias_silu_batch2	same + SiLU

Verification

make test-cpu-value-verify
# 298 passed, 1 skipped

Stacked on R71 (cursor/cpu-loop-r71-layernorm-gated-batch2-78f5).

R73 candidates

• gemm_gelu_batch2 / gemm_relu_batch2 / gemm_silu_batch2
• rms_norm_linear_batch2 + GELU/ReLU/SiLU

  

Summary by Sourcery

Add batch=2 coverage for the 2D gemm_bias operator family and update CPU certification inventory counts accordingly.

New Features:

• Introduce batch=2 builders for 2D gemm_bias with plain, ReLU, GELU, and SiLU variants.

Documentation:

• Extend AGENTS.md with a new Loop R72 entry documenting the 2D gemm_bias batch=2 contracts and verification status.

Tests:

• Register the new gemm_bias batch=2 builders in parity and inventory tests and increase planned value-verify count from 294 to 298.

[sourcery-ai]

Reviewer's Guide

Add batch=2 2D gemm_bias CPU builders (with ReLU/GELU/SiLU variants), register them in parity/inventory tests, and bump the planned value-verify inventory count from 294 to 298, documenting Loop R72 in AGENTS.md.

Flow diagram for new 2D gemm_bias batch2 CPU builders and activation variants

flowchart LR
    gemm_bias_batch2["gemm_bias_batch2\n[M,K] @ [K,N] + [1,N] bias"]

    gemm_bias_relu_batch2["gemm_bias_relu_batch2\nGEMM + bias + ReLU"]
    gemm_bias_gelu_batch2["gemm_bias_gelu_batch2\nGEMM + bias + GELU"]
    gemm_bias_silu_batch2["gemm_bias_silu_batch2\nGEMM + bias + SiLU"]

    gemm_bias_batch2 --> gemm_bias_relu_batch2
    gemm_bias_batch2 --> gemm_bias_gelu_batch2
    gemm_bias_batch2 --> gemm_bias_silu_batch2

Loading

File-Level Changes

Change	Details	Files
Introduce 2D gemm_bias batch=2 builders and reference implementations for plain and activated variants.	Add builder functions that construct kernel graphs for gemm_bias_batch2 and activation variants using float16 inputs with shapes [8,32] @ [32,16] + [1,16]. Define torch-based reference computations (matmul + bias, then optional ReLU/GELU/SiLU) cast back to float16 for each builder. Return graph, inputs, and reference tensors from each builder for use in value-verify tests.	tests/integration/cpu_op_builders.py
Wire new batch=2 gemm_bias builders into general ML parity and builder registries.	Register the new gemm_bias_batch2 and its ReLU/GELU/SiLU variants in the cpu_op_builders registry dict used by tests. Extend the list of builder keys used in general ML PyTorch parity tests to include the new batch=2 gemm_bias contracts.	tests/integration/cpu_op_builders.py tests/integration/test_cpu_general_ml_pytorch_parity.py
Update planned CPU value-verify certification counts from 294 to 298 to account for the new builders.	Change assertions that planned_value_verify_count() equals 294 to expect 298 in certification profile and inventory tests. Adjust mocked certification profiles to expect 298 passed value-verify tests instead of 294, keeping skipped and failed counts consistent. Update cert_profile_from_stages calls to pass planned_value_verify=298 so alignment checks remain valid.	tests/integration/test_cpu_certification_profile.py tests/integration/test_cpu_inventory_planned.py
Document Loop R72 and finalize Loop R71 metadata in AGENTS.md.	Update Loop R71 entry to reference PR cpu loop R71: layernorm act + gated MLP batch2 (290→294) #67 now that it is merged, keeping its verification count at 294 passed. Add a new Loop R72 entry describing closure of 2D gemm_bias batch=2 contracts, listing the new builders and stating 298 passed verifications.	AGENTS.md

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sourcery-ai]

Hey - I've left some high level feedback:

• The four new build_gemm_bias_*_batch2 helpers duplicate most of their setup logic; consider factoring out a shared builder factory that takes the activation op as a parameter to reduce repetition and potential for inconsistencies.
• The planned value-verify count 298 is now hardcoded in multiple tests; it may be more robust to source this from a single constant or helper to avoid future mismatches when the inventory changes again.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- The four new `build_gemm_bias_*_batch2` helpers duplicate most of their setup logic; consider factoring out a shared builder factory that takes the activation op as a parameter to reduce repetition and potential for inconsistencies.
- The planned value-verify count `298` is now hardcoded in multiple tests; it may be more robust to source this from a single constant or helper to avoid future mismatches when the inventory changes again.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}