Fix pokeSTArray, some optimizations #32
Conversation
| while(++i < n) { | ||
| var n = arr.length >>> 0; | ||
| var as = new Array(n); | ||
| for (var i = 0 >>> 0; i < n; i++) |
There was a problem hiding this comment.
Just to ensure unsigned values. Probably useless in this case (not in the case of peekSTArray, where it saves a comparison). Do you want me to remove this one?
There was a problem hiding this comment.
I'm just trying to hint the JIT that both values are unsigned:
(-1) < ([].length)
true
(-1>>>0) < ([].length >>> 0)
false
�
I guess this is completely useless since JITs are now quite smart.
There was a problem hiding this comment.
We really need a good benchmarking library, I'm trying to wrap benchmark.js right now.
There was a problem hiding this comment.
asmjs style? I know x|0 is recognised as "it's an int" type hint, but not sure if there is one for uints.
There was a problem hiding this comment.
Looks like >>>0 is indeed the hint for unsigned.
Take a look at the first benchmark in http://mbebenita.github.io/LLJS/
Change the loop limit to something like 0x7fffffff in all 3 tests, it seems int is the fastest type.
There was a problem hiding this comment.
I've seen very disparate results across browsers/nodes. I'll remove the hints.
|
Looks good to me, thanks. Anyone else want to comment? |
|
I don't think it's very likely at all, but isn't Perhaps this is an absurd thing to worry about though. |
|
Should I leave that as normal numbers with 2 comparisons? The important thing here is really the preallocation and the wrong check... |
|
It's not something I feel strongly about, just thought it was worth discussing. I'd be happy to merge as it is really. @paf31? |
Fix pokeSTArray, some optimizations
|
👍 |
|
I might try and make an array that large to see how long it even takes to iterate over it... |
|
:) |
It seems pokeSTArray allows poking one element past the end.