676700 - Slightly optimize js::IndexToId

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jeff Walden [:Waldo]]

Attachment: Patch

Sparse/dense patchwork needed js::IndexToId, which I then implemented, before discovering it had already been done.  But the current implementation isn't as efficient as this one (it creates an intermediate string rather than atomize directly from a jschar buffer), so let's improve it slightly.

Comment 1

[Luke Wagner [:luke]]

IIRC, js_AtomizeString(js_IntToString()) is only a few branches (and obviously a function call) worse than the proposed IndexToId.  Can you measure the perf improvement for any use of IndexToId?

Btw, there is a flagrantly redundant branch in js_AtomizeString perhaps you could sweep up...

Comment 2

[Luke Wagner [:luke]]

If nothing else, it seems like we should have an inline reusable function to factor out the core backwards-filling index-to-chars algorithm.

Comment 3

[Jeff Walden [:Waldo]]

Attachment: v2

It's not the js_AtomizeString call that contributes extra time.  It's that js_NumberToString must allocate a new JSString, and then that JSString immediately becomes garbage after the atom is created.  So a little time, a little more GC pressure, neither particularly great (if not too horrible generally -- although, one of the JSON parser optimizations I did was to eliminate an extra string creation like this, and it showed up in perf numbers).

I was thinking that loop was a bit duplicated, but it looked a little annoying to abstract it out.  This patch does that.

Comment 4

[Luke Wagner [:luke]]

Ohhh, I knew Atomize converted non-atom strings to JSAtoms, but I forgot that it doesn't do this to string argument passed to js_AtomizeString.  It seems like it should.  A quick scan of js_AtomizeString shows at least a couple of places that exhibit the immediately-garbage string pattern you just mentioned.

Comment 5

[Luke Wagner [:luke]]

However, if the input is already atomized, then the new string is still garbage, so your approach is better.

Comment 6

[Luke Wagner [:luke]]

Comment on attachment 550912 [details] [diff] [review]
v2

Review of attachment 550912 [details] [diff] [review]:
-----------------------------------------------------------------

Wow, I didn't realize it was x3!  r+ with nits.

::: js/src/jsatominlines.h
@@ +142,5 @@
> + * than end.
> + */
> +template <typename T>
> +inline mozilla::RangedPtr<T>
> +WriteIndex(uint32 index, mozilla::RangedPtr<T> end)

As much as I like short names, this seems too short for how weird it is.  How about BackfillIndexInCharBuffer?

Second idea: how about a symbolic UINT32_CHAR_BUFFER_LENGTH constant that is statically asserted correct here and can be reused/asserted in the 3 other places?

Comment 7

[Jeff Walden [:Waldo]]

http://hg.mozilla.org/integration/mozilla-inbound/rev/699b68990d54

Comment 8

[Marco Bonardo [:mak]]

http://hg.mozilla.org/mozilla-central/rev/699b68990d54