Currently the JavaScriptCore Lexer allocates two buffers for storing the source to be parsed. When the lexer is cleared the buffers are swapped out with new buffers. This change allocates a big inline buffer during object creation so that the swap is only necessary if the size exceeds 512.
Created attachment 31500 [details] big inline buffers The patch is from Yong Li. I've tried testing it on Sunspider, but I'm getting inconsistent results with ToT QtWebKit. I don't even know if Sunspider would be relevant for this change, but perhaps someone else can look at it...
Comment on attachment 31500 [details] big inline buffers Looks good, but we need some solid perf results.
I believe we only have performance info on mobile. Should be easy to test it on a desktop build though.
Is any perf info available yet? It is hard to review this without a concrete understanding of the performance benefit.
Removing from the review queue for now as I've been unsuccessful in trying to measure a performance impact of this patch. The plan is to try and find such a performance impact on embedded hardware where the patch originated. Once we find such an impact, we'll revisit.
Not CE specific.
Sill an issue?
(In reply to comment #7) > Sill an issue? shouldn't be a big issue now. I was using a very slow memory allocator in that days.