perf: avoid floating-point formatting in microtime

[henderkes]

Convert to long instead of double and save the division. Extra gain because zend_strpprintf() takes the cheaper integer formatting path instead of having to call zend_dtoa().

This isn't a bug fix so I guess it should target master? @devnexen

[TimWolla]

This isn't a bug fix so I guess it should target master?

Yes.

[TimWolla]

I'm wondering if microtime() without true is something that actually happens in the real world, but looks correct and simple enough to me.

[henderkes]

Yes, it saved half a second in phpbench. Though I guess arguable whether that's real-world usage. Stumbled upon this randomly when tracing back codegen differences between gcc tailcall (patch open) and clang tailcall.

Edit: sorry, @staabm, I meant phpbench. Not phpstan.

[staabm]

do these lines have the same problem?

ext/standard/uniqid.c:72-79 also uses %.8F for a generated seed string.
sapi/phpdbg/phpdbg_out.c:101-105 uses %.8F in debug logging output.

[henderkes]

Yes, but I'll create a new PR since this one was already approved.

[staabm]

to re-iterate this and hopefully I got this right:

microtime(true) is 10x faster than microtime(), right?
https://gist.github.com/staabm/ae867df5f19f88c56cd4de4418763fd3
(before this PR)

hrtime(true) and hrtime() are equal fast though.