SO业务编排-quicksort-PROSAGA-码农传奇

Benchmarks using online quick-bench.com:

Using lambda to compare - 1.8 or 2 slower

with do-not-optimize (is_sorted)

new with span passed by value
needed to install gcc/g++ 10 to get span<> header
Short arrays - clang is better

longer arrays - clang is also worse now

longer arrays with const& span<>

Bypassing span, using to_array - without pivot, O3 long arrays (ignore for now)

without pivot compare, span by value O3 long arrays

without pivot compare, span by const ref O3 long arrays

O2 - no const calling _impl directly (clang is faster)

more consts O2 - all const calling _impl directly (gcc - same, clang - worse )

O3 with const std::span<> long arrays (without pivot compare)

O3 with const std::span<> long arrays (with pivot compare) shows worse than without pivot compare

separate - almost the same

gcc - sort running before geeks

gcc sort - remove const& from func param a bit slower. so const does help

Godbolt

5/22/2021 after working with Jason Turner on Discord.

He found out that using iterator instead int for “lower” is much faster
the const & for span is really messing up things and is slower
clang - with all elements (fastest) - https://quick-bench.com/q/U-wfgBGsFVn7uUmBjNVZre7IK3E
gcc - 10.1 with all elements (fast) - https://quick-bench.com/q/Hqg-jpftC_acXRNMkhVyw0TUv10
gcc - 10.2 with all elements fast 1.2 https://quick-bench.com/q/MqdF-v8BG39a1LRiC8aCcK-RFVk
gcc - excluding pivot (still fast) - https://quick-bench.com/q/9Fw0tnbM6Etf-2NP-hb1FiP-GXc
gcc - 10.3 excluding pivot 10.3 https://quick-bench.com/q/SKO6R1gX5sMQgkLd86ruBgPiXgw
gcc - 10.3 all elements https://quick-bench.com/q/VQKNRWkR6M1jL4ZAT0B2OYSDvIY a bit slower