Changing One "If" To "While" Caused An Unexpected Shift In A Kernel Benchmark This Week

[phoronix]

Phoronix: Changing One "If" To "While" Caused An Unexpected Shift In A Kernel Benchmark This Week

Several months back you may recall that Linux 5.9 kernel regression we noted that in turn was bisected to code introduced by Linus Torvalds around page lock fairness. That was ultimately worked out in time with allowing a control over the page lock (un)fairness to address the regressed workloads while being fair enough to satisfy his original change. But now this week for Linux 5.11, Linus Torvalds has again altered the behavior. It then ended up causing a PostgreSQL database server performance regression but fortunately any impact should be very minimal and hopefully not appearing in any real-world situation...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-If-While-PL-Fairness-5.11

[slalomsk8er]

If I change a "if" to anything else, stuff normally breaks and "if" I tested I will know before my colleges do

[slalomsk8er]

Low quality post. Why not read the article and post something relevant?

Did read the article and nice case of pot calling the kettle black

[Aryma]

what a joke

[Sethox]

Low quality post. Why not read the article and post something relevant?

The joke serves 2 points, 1 is the actual joke the other is the valid criticism you are supposed to do (because of human error).

[carewolf]

If I change a "if" to anything else, stuff normally breaks and "if" I tested I will know before my colleges do

Being able to replace if with while is pretty common.

[turboNOMAD]

In the news: changing a conditional code, which runs 0 or 1 times, to a loop which may run from 0 to N times until the condition is no longer true, MAY INCREASE THE TIME THIS CODE TAKES. Who would have thought?

In other news: the Sun rose in the east today.

[gens]

In the news: changing a conditional code, which runs 0 or 1 times, to a loop which may run from 0 to N times until the condition is no longer true, MAY INCREASE THE TIME THIS CODE TAKES. Who would have thought?

In other news: the Sun rose in the east today.

It probably doesn't loop more then once, most of the time.

The difference is one jmp instruction. That may change alignment of code, the size of the code the cpu "transpiles" for itself, and another thing i forgot now.
So it's probably as linus said.

[turboNOMAD]

It probably doesn't loop more then once, most of the time.

I think so too. My guess it that PostgreSQL benchmark mentioned in the article is one of the few cases when it does, due to nature of its workload. I don't think it has to do anything with code alignment. Never seen noticeable performance changes due to one extra instruction, although I am more familiar with such micro-optimizations for controllers such as STM32, it's been some years since I did this wizardry on x86.