SpCL | paced Contrastive Learning with Hybrid Memory | Machine Learning library

This is a huge process, doing many things at once, some of them deeply nested in each other.

The function is very, very long, which means it probably needs to be abstracted into various methods and classes (better OOP).

You may also be running into the N+1 problem with some of your Laravel queries, although it's hard to say at a glance.

I would recommend using Laravel's Queues, and adding each employee's payroll calculation to the queue: https://laravel.com/docs/5.8/queues

You can then use a worker process to perform each one individually.

All the database updates and transactions means PHP is constantly going to-and-fro from your database. Try to perform as many operations in pure code as possible, and then once completed, write to the DB. (Where possible, this is not a blanket rule).

Initially I would say, consider creating a SalaryCalculator class with methods such as getAttendances() and calculateSalaryFromEmployeeAttendances().

Separately, create a CSV exporter class. Process all your payroll calculations first, store the results in the DB, and then convert to CSV on demand later.

You'll refactor this later once you are able to look at all the different parts, someone else may be able to suggest a better way to break it down, but in some ways there are no right answers... just start with some OOP and abstraction of methods, and it will get better.

Use microtime(true) to get and calculate the time differences between when you start and finish operations, to start to track how long each function runs for... and go looking for the big optimisations first. What is the slowest part? Why?

You could probably ask a dozen Stack Overflow questions to optimise each of those methods, and that's Ok!