Title: [231187] trunk
- Revision
- 231187
- Author
- mmaxfi...@apple.com
- Date
- 2018-04-30 17:23:24 -0700 (Mon, 30 Apr 2018)
Log Message
Improve the performance of FontCascadeDescription's effectiveFamilies
https://bugs.webkit.org/show_bug.cgi?id=184720
PerformanceTests:
Reviewed by Simon Fraser.
This performance test calls SystemFontDatabase::systemFontCascadeList() around 2,000,000 times (before
this patch is applied), which is roughly equivalent to the page we found the performance problem on.
The calling pattern is roughly equivalent in this test.
* Layout/system-ui.html: Added.
Source/WebCore:
<rdar://problem/38970927>
Reviewed by Simon Fraser.
The page that had the performance problem renders many different Chinese characters in system-ui
with only a small number of individual fonts. It turns out we were calling into the system-ui
machinery for each character in order to opportunistically start loading data URLs (see also:
https://bugs.webkit.org/show_bug.cgi?id=175845). These data URLS will never represent the system
font, so we don't need to invoke the system-ui machinery at all.
This patch makes a 92x performance improvement on the associated performance test. This test is
designed to test Chinese text rendered with system-ui.
Performance test: Layout/system-ui.html
* platform/graphics/FontCascadeFonts.cpp:
(WebCore::opportunisticallyStartFontDataURLLoading):
Modified Paths
Added Paths
Diff
Modified: trunk/PerformanceTests/ChangeLog (231186 => 231187)
--- trunk/PerformanceTests/ChangeLog 2018-05-01 00:06:14 UTC (rev 231186)
+++ trunk/PerformanceTests/ChangeLog 2018-05-01 00:23:24 UTC (rev 231187)
@@ -1,3 +1,16 @@
+2018-04-30 Myles C. Maxfield <mmaxfi...@apple.com>
+
+ Improve the performance of FontCascadeDescription's effectiveFamilies
+ https://bugs.webkit.org/show_bug.cgi?id=184720
+
+ Reviewed by Simon Fraser.
+
+ This performance test calls SystemFontDatabase::systemFontCascadeList() around 2,000,000 times (before
+ this patch is applied), which is roughly equivalent to the page we found the performance problem on.
+ The calling pattern is roughly equivalent in this test.
+
+ * Layout/system-ui.html: Added.
+
2018-03-08 Antti Koivisto <an...@apple.com>
Update StyleBench version number in page title to 0.3
Added: trunk/PerformanceTests/Layout/system-ui.html (0 => 231187)
--- trunk/PerformanceTests/Layout/system-ui.html (rev 0)
+++ trunk/PerformanceTests/Layout/system-ui.html 2018-05-01 00:23:24 UTC (rev 231187)
@@ -0,0 +1,36 @@
+<!DOCTYPE html>
+<html>
+<head>
+<meta charset="utf-8">
+<script src=""
+</head>
+<body>
+<div id="target" style="width: 300px; display: none; font: 42px 'system-ui', '-apple-system';" lang="zh-CN"></div>
+<script>
+var target = document.getElementById("target");
+var style = target.style;
+
+var s = "";
+var length = 10000;
+var startCode = 0x4E00;
+for (var i = 0; i < length; ++i) {
+ s = s + String.fromCharCode(i + startCode);
+}
+
+
+function test() {
+ if (window.internals)
+ window.internals.invalidateFontCache();
+
+ style.display = "block";
+ target.offsetLeft;
+ target.textContent = s;
+ target.offsetLeft;
+ target.textContent = "";
+ style.display = "none";
+}
+
+PerfTestRunner.measureRunsPerSecond({ run: test });
+</script>
+</body>
+</html>
Modified: trunk/Source/WebCore/ChangeLog (231186 => 231187)
--- trunk/Source/WebCore/ChangeLog 2018-05-01 00:06:14 UTC (rev 231186)
+++ trunk/Source/WebCore/ChangeLog 2018-05-01 00:23:24 UTC (rev 231187)
@@ -1,3 +1,25 @@
+2018-04-30 Myles C. Maxfield <mmaxfi...@apple.com>
+
+ Improve the performance of FontCascadeDescription's effectiveFamilies
+ https://bugs.webkit.org/show_bug.cgi?id=184720
+ <rdar://problem/38970927>
+
+ Reviewed by Simon Fraser.
+
+ The page that had the performance problem renders many different Chinese characters in system-ui
+ with only a small number of individual fonts. It turns out we were calling into the system-ui
+ machinery for each character in order to opportunistically start loading data URLs (see also:
+ https://bugs.webkit.org/show_bug.cgi?id=175845). These data URLS will never represent the system
+ font, so we don't need to invoke the system-ui machinery at all.
+
+ This patch makes a 92x performance improvement on the associated performance test. This test is
+ designed to test Chinese text rendered with system-ui.
+
+ Performance test: Layout/system-ui.html
+
+ * platform/graphics/FontCascadeFonts.cpp:
+ (WebCore::opportunisticallyStartFontDataURLLoading):
+
2018-04-30 Jer Noble <jer.no...@apple.com>
<img src="" does not display on ios despite Accept: video/* advertisement
Modified: trunk/Source/WebCore/platform/graphics/FontCascadeFonts.cpp (231186 => 231187)
--- trunk/Source/WebCore/platform/graphics/FontCascadeFonts.cpp 2018-05-01 00:06:14 UTC (rev 231186)
+++ trunk/Source/WebCore/platform/graphics/FontCascadeFonts.cpp 2018-05-01 00:23:24 UTC (rev 231187)
@@ -389,14 +389,8 @@
// asynchronous, and this code doesn't actually fix the race - it just makes it more likely for the two fonts to tie in the race.
if (!fontSelector)
return;
- for (unsigned i = 0; i < description.effectiveFamilyCount(); ++i) {
- auto visitor = WTF::makeVisitor([&](const AtomicString& family) {
- fontSelector->opportunisticallyStartFontDataURLLoading(description, family);
- }, [&](const FontFamilyPlatformSpecification&) {
- });
- const auto& currentFamily = description.effectiveFamilyAt(i);
- WTF::visit(visitor, currentFamily);
- }
+ for (unsigned i = 0; i < description.familyCount(); ++i)
+ fontSelector->opportunisticallyStartFontDataURLLoading(description, description.familyAt(i));
}
GlyphData FontCascadeFonts::glyphDataForVariant(UChar32 character, const FontCascadeDescription& description, FontVariant variant, unsigned fallbackIndex)
_______________________________________________
webkit-changes mailing list
webkit-changes@lists.webkit.org
https://lists.webkit.org/mailman/listinfo/webkit-changes
