Switch to the new client-side phishing model that uses Murmurhash for word hashes.

Switch to the new client-side phishing model that uses Murmurhash for word hashes.

This gives about a 50% performance improvement on extraction of page term features.

BUG=96060
TEST=updated existing tests


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=100858

[noelutz, 2011-09-10 00:33:18]

On 2011/09/10 00:17:21, Brian Ryner wrote:

LGTM

Nicely done.
noe.

[mattm, 2011-09-10 01:44:25]

http://codereview.chromium.org/7866011/diff/1/chrome/renderer/safe_browsing/m...
File chrome/renderer/safe_browsing/murmurhash3_util.cc (right):

http://codereview.chromium.org/7866011/diff/1/chrome/renderer/safe_browsing/m...
chrome/renderer/safe_browsing/murmurhash3_util.cc:12:
MurmurHash3_x86_32(str.data(), str.size(), seed, &output);
Do you have any data on how this performs on non-x86 platforms?

[Brian Ryner, 2011-09-12 18:31:16]

On 2011/09/10 01:44:25, mattm wrote:
>
http://codereview.chromium.org/7866011/diff/1/chrome/renderer/safe_browsing/m...
> File chrome/renderer/safe_browsing/murmurhash3_util.cc (right):
> 
>
http://codereview.chromium.org/7866011/diff/1/chrome/renderer/safe_browsing/m...
> chrome/renderer/safe_browsing/murmurhash3_util.cc:12:
> MurmurHash3_x86_32(str.data(), str.size(), seed, &output);
> Do you have any data on how this performs on non-x86 platforms?

I was able to test that the hash works as expected on ARM, at least under QEMU. 
Aside from that, I don't have any actual non-x86 hardware to test on, but I'd
assume that this will always be faster than SHA-256.

There are a couple of compile fixes that were necessary for ARM, which I'll
upload shortly.

[Brian Ryner, 2011-09-12 21:51:49]

Ok, please take another look.

[mattm, 2011-09-13 00:45:39]

lgtm

[commit-bot: I haz the power, 2011-09-13 04:21:48]

Change committed as 100858