[Copilot]

Description

• Compute block IDs per quantized-axis row:

  block_id = row * blocks_per_row + column / block_size;

• Add CUDA regression coverage for a 4-bit, non-block-aligned quantized axis.

Motivation and Context

CUDA previously derived the scale and zero-point block from the flat data index. Non-block-aligned quantized axes therefore crossed row boundaries and silently produced incorrect results.

• Fixes GatherBlockQuantized CUDA kernel uses the flat index for the block id, giving wrong results when the quantized axis is not a multiple of block_size #32663

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

🟡 Changes recommended

Packed zero-point indexing remains incorrect for rows with a non-pack-aligned block count, and the CUDA test can silently run on CPU.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Fixes CUDA block indexing for non-aligned quantized-axis rows and adds regression coverage.

Changes:

• Computes scale block IDs per quantized-axis row.
• Passes the unpacked quantized-axis dimension to the CUDA kernel.
• Adds a 4-bit CUDA regression test.

File summaries

File	Description
gather_block_quantized.cc	Supplies the logical quantized-axis dimension.
gather_block_quantized.cu	Corrects row-aware block indexing.
gather_block_quantized.cuh	Updates CUDA kernel parameters.
gather_block_quantized_op_test.cc	Adds non-aligned-axis coverage.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 2
• Review effort level: Balanced

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Tianlei Wu (tianleiwu)]

Re-reviewed the per-row scale indexing, the independently padded packed-zero-point indexing, and the CUDA-specific test setup. The earlier packing and CPU-fallback concerns are addressed; no additional actionable findings.

An independent host arithmetic check matches all 72 values in the odd-block regression and verifies padded byte indices for 2/4/8-bit packing. git diff --check passes. These are not CUDA runtime tests, and this head currently has no reported CI checks, so please obtain current-head CUDA validation before merging. No approval submitted.

[Akshay Sonawane (apsonawane)]

AI-assisted source review at d43a594486cdec442fc4f9a739ca5a6fd299b018.

Traced packed input expansion, gather offsets, dense scale indexing, and separately row-padded packed zero points against the schema and CPU implementation. Also checked CUDA-only test selection and disabled CPU fallback. The prior packed-zero-point and fallback concerns are resolved at this head.

No new actionable findings in this pass. Independent host-side indexing probes covered supported 2/4/8-bit examples and reproduced the new odd-block expected values. Reported head checks are 100 successful and one neutral. No CUDA compilation, GPU execution, or local ORT suite was performed.