json-stringify perf suggestions

[Ledion Bitincka]

First off, kudos to the team for the recent json stringify optimizations! 

While looking at the new (and old) code I noticed that the way the escaped size of a string is being estimated uses an optimization which could introduce large memory overhead. I don't think that optimizing away a multiplication is worth the memory trade off especially when dealing with serialization of large strings.

I would recommend that the function uses the proper max estimation of 6x the string length

size_t MaxEscapedStringLength(size_t length) { return length  * 6; }

--

Ledion Bitincka

Co-founder | CTO cribl.io |  @ledbit

Curious about our products?

[Claudia]

The code path is hit even for empty strings and short strings, so it may still be worth the extra ~30% overhead. Short strings dominate objects, and it's quite common to see large arrays of objects only using small property names.

Better would be to ensure capacity of 4*sizeof(char) if you really wanted to eliminate menory overhead.

Keep in mind that a multiplication on high-end chips usually takes about 3 cycles to complete, vs 1 cycle for an addition.

[Ledion Bitincka]

Yeah, it makes sense for small strings and when the serialized object doesn't exceed the 256 byte stack buffer. However, when objects don't fit and new buffers need to be allocated because of the overestimation the few cycles saved by not using multiplication will be dwarfed by the memory management operations. 

--
--
v8-dev mailing list
v8-...@googlegroups.com
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to v8-dev+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/v8-dev/2daed2c6-c146-472d-a7f5-879c755fa2cdn%40googlegroups.com.