-
Notifications
You must be signed in to change notification settings - Fork 291
endianness of murmurhash #23
Comments
MurmurHashNeutral2 is twice as slow as the little-endian only version. We should only use it if we detect a big-endian architecture. https://sites.google.com/site/murmurhash/ |
But in practice codegenning? (mobile) On Jul 11, 2011, at 1:47 PM, alastairreply@reply.github.com wrote:
|
true - less than 1/10th of a second over a 5 minute song. |
Need to dig into the whole endian thing a bit more - 16bit LE PCM won't read correctly on a BE machine anyway, so there's a chance that more parts of the codegen need tweaking. |
for my reference, stuff on detecting endianness: |
I think we're running into this problem - we codgenned on x86 to populate our database, then got our android app to codegen ten second samples from the mic - and nothing matches. I'm thinking this is because ARM is little endian and x86 is big ? How could this work in the IOS sample? or is that intended for the desktop and not iPhones ? |
from ashwin:
MurmurHash2 used gives different results on big endian and little endian machines. Looks like if this is fixed, it might break codegen backward compatibility? Could someone explain if this was by design or a bug? How will the matching work if the codes are different?
Edit Fingerprint.css, replace MurmurHash2 with:
Looks like it is the same on little endian machines:
Machine used was 64bit Intel.
The text was updated successfully, but these errors were encountered: