perf(fp8): Remove redundant clone calls from FP8 Linear Kernel

[R3hankhan123]

Description

Remove Redundant clone calls from FP8 Linear Kernel

Related Issues

Fixes #840

Test Plan

curl request

curl http://localhost:8000/v1/chat/completions   -H "Content-Type: application/json"   -d '{
    "model": "ibm-granite/granite-4.1-8b-fp8",
    "messages": [
      {
        "role": "user",
        "content": "Hello! Explain what a transformer model is in one paragraph."
      }
    ],
    "max_tokens": 20,
    "temperature": 0.7
  }' | jq

output

"role": "assistant",
        "content": "A transformer model is a type of deep learning architecture designed to process and generate sequential data, such as",

Checklist

• I have read the contributing guidelines
• My code follows the project's code style (run bash format.sh)
• I have added tests for my changes (if applicable)
• I have updated the documentation (if applicable)
• My commits include a Signed-off-by: line (DCO compliance)

[github-actions]

👋 Hi! Thank you for contributing.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, run ./format.sh.
Now you are good to go 🚀.

We also recommend installing prek and configuring it to check your code before every local commit.

[tdoublep]

/caude-review

[tdoublep]

Posted by Claude Code on behalf of @tdoublep.

I ran this on the device. The removed clones really are redundant: output is bitwise identical to main across 9 shape/M combinations, still contiguous at offset 0, and nothing reuses the buffer. Looks good.

Two notes:

1. It also fixes a bug. On main, 3-D input that needs M padding comes back with wrong numbers (off by ~2-3 vs. the same rows passed as 2-D). Trimming before the reshape is what fixes it. Worth saying so in the description.
2. Nothing tests this. test_scaled_mm_apply uses num_tokens 1/4/128, which all skip M padding, and it only checks dtype and shape — so the branch added here never runs in CI. Please add 5 or 130 and compare values.

[joerunde]

/claude-review

[github-actions]

These two cases (5, 130) do add M-padding, but the test still only asserts dtype and shape and only ever passes 2-D x. The bug this PR fixes is 3-D input that needs M-padding: the pre-clone() reshape reads the padded storage and returns wrong numbers. On device I confirmed it — feeding the same 130 rows as (13, 10, 128) gives max|out3d - out2d| = 2.999 on main and 0.0 on this branch. Neither of those facts is checked here: no 3-D shape is exercised (so the reshape branch never runs in CI), and no value comparison is done (so even a 2-D numeric regression would pass). The docstring's claim that these cases are "verifying the trim-before-reshape path" is therefore not accurate. Please add a 3-D case and compare apply_weights output against a reference (e.g. the 2-D result reshaped, or a dequantized fp16 matmul), as tdoublep asked in the review.

[joerunde]

@R3hankhan123 looks like this might need one more pass then we should be good to go

[github-actions]

Reviewed on Spyre hardware. The clone removals in _join and the final return are safe, and moving the clone() to before the reshape is a real correctness fix, not just a perf change: feeding 130 rows as a 3-D (13, 10, 128) tensor gives max|out3d - out2d| = 2.999 on main versus 0.0 on this branch — the padded storage was corrupting the reshape on main. The fp8 tests run in the smoke shards on single-card (spyre_pf_x1) runners, which is correct for these 128×128 shapes, and all 16 cases pass.

One issue worth blocking on: the fix is untested. The new num_tokens 5/130 add M-padding but the tests still pass only 2-D x (so the reshape branch never runs in CI) and still assert only dtype/shape (so a numeric regression would pass). This is the value-comparison tdoublep asked for in the review; it hasn't been done yet. I left an inline comment with a concrete suggestion and one docstring-accuracy note. The two prior inline threads (the _join docstring and the block comment) are addressed — I resolved the first and left a one-line grammar nit on the second. Also worth doing per tdoublep's note: mention the bug fix in the PR description. Net: 1 significant gap (missing test coverage) plus 2 minor nits.