Hi,
I was looking at v8-splay.js from the V8 benchmarks recently. It's an
awful benchmark. It contains this function:
function GeneratePayloadTree(depth, key) {
if (depth == 0) {
return {
array : [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ],
string : 'String for key ' + key + ' in leaf node'
};
} else {
return {
left: GeneratePayloadTree(depth - 1, key),
right: GeneratePayloadTree(depth - 1, key)
};
}
}
This function is called many times with 'depth' set to 5 and 'key' set
to a randomly-generated double. With these parameters it creates the
same string "String for key <whatever> in leaf node" 32 times, once
per leaf call. This involves a double-to-string conversion and then
two string concatenations.
The benchmark has this comment at the top:
// This benchmark is based on a JavaScript log processing module used
// by the V8 profiler to generate execution time profiles for runs of
// JavaScript applications, and it effectively measures how fast the
// JavaScript engine is at allocating nodes and reclaiming the memory
// used for old nodes. Because of the way splay trees work, the engine
// also has to deal with a lot of changes to the large tree object
// graph.
This isn't true. Because GeneratePayloadTree() is so stupid, the
benchmark is actually primarily measuring the speed of
double-to-string conversions. The following trivial change is all
that's required to make the benchmark less stupid:
Index: tests/v8-v4/v8-splay.js
===================================================================
--- tests/v8-v4/v8-splay.js (revision 58474)
+++ tests/v8-v4/v8-splay.js (working copy)
@@ -45,7 +45,7 @@
if (depth == 0) {
return {
array : [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ],
- string : 'String for key ' + key + ' in leaf node'
+ string : key
};
} else {
return {
@@ -69,7 +69,7 @@
do {
key = GenerateKey();
} while (splayTree.find(key) != null);
- splayTree.insert(key, GeneratePayloadTree(kSplayTreePayloadDepth, key));
+ splayTree.insert(key, GeneratePayloadTree(kSplayTreePayloadDepth, 'String for
key ' + key + ' in leaf node'));
return key;
}
I tested with Mozilla's TraceMonkey JS engine, it ran about 2.8x
faster with this change in place.
Any chance this change could make it into the next version of V8? I'm
happy to file a bug if someone can tell me where.
N
--
v8-users mailing list
[email protected]
http://groups.google.com/group/v8-users
