Bug 140123 – REGRESSION (r177876): 35% regression in Parser/html5-full-render

• Status: RESOLVED
• Resolution: FIXED
• Component: Layout and Rendering

[Antti Koivisto]

https://perf.webkit.org/#mode=charts&chartList=%5B%5B%22mac-mavericks%22%2C%22Parser%2Fhtml5-full-render%3ATime%22%5D%5D

[Antti Koivisto]

Created attachment 244053 [details]
patch

[Darin Adler]

Comment on attachment 244053 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=244053&action=review

> Source/WebCore/platform/graphics/SimpleFontData.cpp:443
>  RefPtr<SimpleFontData> SimpleFontData::systemFallbackFontDataForCharacter(UChar32 c, const FontDescription& description, bool isForPlatformFont) const

Would be nice to name it "character" instead of "c".

> Source/WebCore/platform/graphics/SimpleFontData.cpp:455
> +        if (c <= 0xFFFF) {

I think we should use U_IS_BMP here.

> Source/WebCore/platform/graphics/SimpleFontData.cpp:480
> +        Vector<std::pair<UChar32, unsigned>> toRemove;

Inline capacity here, maybe?

> Source/WebCore/platform/graphics/SimpleFontData.cpp:484
> +        for (auto& entry : characterMap) {
> +            if (entry.value == this)
> +                toRemove.append(entry.key);
> +        }

Should we have a map in the reverse direction to avoid this iteration?

[Antti Koivisto]

https://trac.webkit.org/r177979

[Antti Koivisto]

> Should we have a map in the reverse direction to avoid this iteration?

I suspect it is fine as-is. The GlyphPageTree code did a comparable crawl and I don't think we ever had problems with it.

[Antti Koivisto]

...and https://trac.webkit.org/r177984

[Antti Koivisto]

The bot says performance is back where it was.