HHVM Is Now Running Even Faster, Beating PHP7 By Wider Margins

[phoronix]

Phoronix: HHVM Is Now Running Even Faster, Beating PHP7 By Wider Margins

The Facebook team working on the HHVM project for being a faster PHP interpreter and powering their Hack language have just come out of a two-week, open-source performance lockdown. Over the past two weeks they focused on making strides to make HHVM's compelling performance even better...

http://www.phoronix.com/scan.php?page=news_item&px=HHVM-Lockdown-Perf-PHP7

[Danny3]

Since PHP is open source, I never understoold why Facebook didn't improved PHP core
Why build a new interpreter when you can make your improvements to default php interpreter and submit the changes?

[mark45]

Since PHP is open source, I never understoold why Facebook didn't improved PHP core
Why build a new interpreter when you can make your improvements to default php interpreter and submit the changes?

From scratch is sometimes better and you're not bound by others/old paradigms.
Another reason, some dudes think open source contribution is made of ponnies and rainbows, in reality there's a lot of mistrust and dickheads in your way to improve many projects and they hold the keys.

[Morpheus]

As I see it, beyond the differences in architecture, HHVM and PHP 7 have two different visions (it's said in the article) : HHVM focus on a set of frameworks to support, while the "stock" PHP has to deal with the whole language, regardless of how frameworks are using it. That's why you can't use every PHP application on HHVM.

[TheOne]

I just don't understand how is it that an engine with JIT like HHVM is just slightly faster than PHP7, I thought it should be more faster than that...

[gggeek]

TheOne this was explained in multiple blog posts by the Zend guys: they also spent 2 years on developing a jit vm as an alternative to hhvm, only to find out that most of the time used by "real life" php applications (such as wikipedia - no one is using php to calculate prime numbers) is not spent executing calculations, bur rather in accessory tasks such as memory allocations and copying of data around. Changing the fundamental data structure used by php (the 'zend_value' used to represent any variable) was by itself a huge step forward, as it allows less copying of data and cache-trashing. Also think that hhvm can only do deep-optimization of code paths when the types of variables can be inferred. If you code php java-style, then you will get excellent results. Less so when you code php php4-style, with no type hinting and few objects.

[TheOne]

gggeek that makes sense, especially array zvals, they are slow and big. Just try to allocate a php array of 100000 elements and see how much ram it consumes.