154910 - arena allocate PerFrameData and PerSpanData

Status: RESOLVED DUPLICATE of bug 250105

(Core :: Layout: Block and Inline, defect, P2)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I decided to try arena-allocating PerFrameData and PerSpanData (in nsLineLayout)
using the pres shell's dynamic stack arena.  I haven't yet found a testcase
where this gives a measurable performance improvement, but I suspect it helps a
little bit.

This patch also removes the free list -- I put in some printfs, and on the first
10 or so pages of jrgm's tests, we used the free list for about 10 of 100000 or
so allocations of these structures.  It seems that the usual case is just
creating them and then destroying them (as I suspected).

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: patch

Here's the patch I've been testing.

It might also be worth changing things like
nsBlockFrame::DoReflowInlineFramesMalloc to use the same arena.

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: slightly cleaner patch

Comment 3

[Chris Waterson]

Comment on attachment 89634 [details] [diff] [review]
slightly cleaner patch

r=/sr=waterson.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 89634 [details] [diff] [review]
slightly cleaner patch

>+    // The normal operator new is disallowed.
>+#if (__GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 95))
>+    void* operator new(size_t sz) throw () { return nsnull; }
>+#else
>+    void* operator new(size_t sz) { return nsnull; }
>+#endif

use CPP_THROW_NEW (here and for the real operator new()).

Same for PerSpanData

Could we remove the NS_LINELAYOUT_NUM_FRAMES and NS_LINELAYOUT_NUM_SPANS
defines too?

With that, r=bzbarsky

Comment 5

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Fix checked into trunk, 2002-08-06 06:51 PDT.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Reopening.  This fix was incorrect.  See bug 162516.

Comment 7

[Justin Wood (:Callek)]

Bug last commented in 2002, last touched in 2009, is this still valid