Execute tensor product functions with RowFn

[connortsui20]

Summary

• Tracking Issue: Epic: Row-oriented scalar functions #9128
• Depends on: Execute tensor L2 norm with RowFn #9768

Moves inner product and cosine similarity to RowFn without changing their arithmetic contracts.

Changes

Inner product keeps the existing left-to-right multiply-and-sum order. Cosine computes that dot product followed by both L2 norms and the existing zero-denominator guard, with no encoded reduction, reassociation, or constant-norm shortcut. Tests cover nulls, empty and zero-width rows, IEEE overflow and underflow, and bitwise agreement between constant encodings and materialized rows.

[codspeed-hq]

Merging this PR will regress 5 benchmarks

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 25 improved benchmarks
❌ 5 regressed benchmarks
✅ 2157 untouched benchmarks
🆕 10 new benchmarks
⏩ 218 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	mul_u64_nonnull_neon	15.4 µs	20.5 µs	-24.89%
❌	Simulation	decompress[u64, (4000, 1024)]	70.5 µs	86 µs	-18.03%
❌	WallTime	mul_i64_nonnull_neon	17.3 µs	20.1 µs	-13.81%
❌	WallTime	multiply_shapes_neon[(16384, PerRowPerRow)]	17.4 µs	20.1 µs	-13.43%
❌	WallTime	words_gather_scalar_avx2[65536]	8.3 µs	9.4 µs	-11.9%
⚡	Simulation	column_x_constant[2]	517.8 µs	131 µs	×4
⚡	Simulation	column_x_constant[32]	473.9 µs	131 µs	×3.6
⚡	WallTime	filtered_owned_i64_avx2[NineNullsInTen]	36.8 µs	13.3 µs	×2.8
⚡	WallTime	filtered_owned_i64_avx512[NineNullsInTen]	36.9 µs	13.4 µs	×2.7
⚡	WallTime	filtered_owned_i64_avx2[OneNullInEight]	56.1 µs	22.2 µs	×2.5
⚡	Simulation	column_x_constant[256]	742.4 µs	306 µs	×2.4
⚡	WallTime	filtered_owned_i64_avx512[OneNullInEight]	62.3 µs	26.3 µs	×2.4
⚡	WallTime	filtered_sink_i64_avx2[NineNullsInTen]	36.8 µs	16.4 µs	×2.3
⚡	WallTime	filtered_sink_i64_avx512[NineNullsInTen]	36.9 µs	16.5 µs	×2.2
⚡	Simulation	column_x_extension_constant[256]	573.8 µs	274.4 µs	×2.1
⚡	Simulation	column_x_column[2]	201 µs	101.8 µs	+97.52%
⚡	WallTime	filtered_sink_i64_avx512[OneNullInEight]	62.2 µs	31.5 µs	+97.47%
⚡	Simulation	column_x_extension_constant[2]	274 µs	147.9 µs	+85.31%
⚡	Simulation	nullable[2]	591.4 µs	323.9 µs	+82.57%
⚡	Simulation	non_nullable[2]	593.6 µs	329 µs	+80.44%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ct/row-fn-tensor-products-v2 (4bd332d) with ct/row-fn-tensor-l2-v2 (929ae6d)²}

Footnotes

1. 218 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on ct/row-fn-tensor-l2-v2 (c11ae04) during the generation of this report, so b771900 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩