Comparison composition from eq/lt

[AdamGS]

No description provided.

[codspeed-hq]

Merging this PR will degrade performance by 11.35%

⚡ 4 improved benchmarks
❌ 8 regressed benchmarks
✅ 135 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	pack_16_to_3_stack	2.4 µs	1.9 µs	+26.65%
⚡	unchecked_unpack_16_from_3_stack	2.1 µs	1.6 µs	+28.51%
⚡	unpack_16_from_3_stack	2 µs	1.5 µs	+33.41%
❌	bitpacking_cmp_seq[2]	6.8 µs	8.5 µs	-20.19%
❌	bitpacking_cmp_seq[3]	6.9 µs	8.6 µs	-20.3%
❌	bitpacking_cmp_seq[5]	7.1 µs	8.8 µs	-19.48%
❌	bitpacking_cmp_seq[7]	7.4 µs	9.1 µs	-19.47%
❌	bitpacking_cmp_unpack[2]	3.2 µs	5 µs	-35.24%
❌	bitpacking_cmp_unpack[3]	3.4 µs	5.2 µs	-34.15%
❌	bitpacking_cmp_unpack[5]	3.6 µs	5.4 µs	-32.62%
❌	bitpacking_cmp_unpack[7]	3.8 µs	5.6 µs	-32.24%
⚡	for_pack_16_to_3_stack	1.8 µs	1.3 µs	+35.16%

Tip

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

---

_{Comparing adamg/cmp-structure-refactor (d28a5cd) with develop (2489a1d)}

[joseph-isaacs]

Can you include local benchmark if they differ to codspeed. I woud expect AVX-512 to be interesting here

[joseph-isaacs]

This could be done without two kernels.

$x > c \iff x >= c+1$ (apart from at the boundary)