perf(compute): tile checked lane kernels consistently

[connortsui20]

Summary

Adding Narrow changed LLVM's unrolling of the existing checked 64-bit multiplication loop on Linux ARM, even though the loop's source was unchanged. Chunking map_checked_into gives LLVM a fixed trip count for the checked loop, as the other unmasked lane kernels already do.

Known limitation: chunking also regresses other benchmarks, particularly unsigned multiplication on AVX2 and AVX512. Making the chunk length const generic and removing forced inlining made the regressions worse, so that experiment was reverted.

Changes

Uses the same chunk size as the other unmasked lane kernels, with a runtime count and an #[inline(always)] helper. Failure evidence combines across full chunks and the remainder.

[codspeed]

Merging this PR will regress 9 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

⚡ 17 improved benchmarks
❌ 9 regressed benchmarks
✅ 2079 untouched benchmarks
⏩ 503 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	mul_u16_nonnull_avx2	11.2 µs	20.8 µs	-45.9%
❌	WallTime	mul_u16_nonnull_avx512	11.3 µs	19.5 µs	-42.22%
❌	WallTime	mul_u8_nonnull_avx512	22.5 µs	29.7 µs	-24.41%
❌	WallTime	mul_u32_nonnull_avx2	12.8 µs	16.9 µs	-24.07%
❌	WallTime	mul_u8_nonnull_avx2	20.5 µs	26 µs	-20.94%
❌	WallTime	add_u32_nonnull_avx2	13 µs	15.4 µs	-15.86%
❌	WallTime	mul_u32_nonnull_avx512	12 µs	13.8 µs	-13.33%
❌	WallTime	add_u32_nonnull_neon	13.5 µs	15.5 µs	-12.93%
❌	WallTime	add_i32_nonnull_avx512	12 µs	13.5 µs	-10.9%
⚡	WallTime	mul_u64_nonnull_avx512	52.7 µs	29 µs	+81.43%
⚡	WallTime	mul_i32_nonnull_avx2	32.7 µs	18.1 µs	+80.4%
⚡	WallTime	mul_i32_nullable_avx2	33.8 µs	19.3 µs	+74.94%
⚡	WallTime	mul_u64_nonnull_neon	39.4 µs	24.7 µs	+59.27%
⚡	WallTime	mul_i16_nonnull_avx2	29.2 µs	21.5 µs	+35.84%
⚡	WallTime	multiply_shapes_neon[(32768, PerRowPerRow)]	38.9 µs	30.2 µs	+29.02%
⚡	WallTime	mul_i64_nonnull_neon	38.6 µs	30.2 µs	+27.71%
⚡	WallTime	mul_i8_nonnull_avx2	29.9 µs	24.7 µs	+20.89%
⚡	WallTime	add_constant_shapes_avx2[(32768, PerRowConstant)]	12.3 µs	10.9 µs	+12.67%
⚡	WallTime	add_constant_shapes_avx2[(32768, ConstantPerRow)]	12.3 µs	10.9 µs	+12.39%
⚡	WallTime	mul_i8_nonnull_neon	25.6 µs	22.8 µs	+12.19%
...	...	...	...	...	...

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/checked-lane-chunks (47db41a) with develop (80aaa32)²}

Footnotes

1. 503 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 develop (23a59e3) during the generation of this report, so 80aaa32 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩