Subject: Re: [boost] unordered_map 32bit VS 64bit
From: Daniel James (dnljms_at_[hidden])
Date: 2012-01-26 02:29:05
On 26 January 2012 07:11, Andrey Semashev <andrey.semashev_at_[hidden]> wrote:
> Is it possible to always use power-of-2 number of buckets and bitwise
> operations instead of modulus division?
Since the container can use arbitrary hash functions, and the
standard's quality requirements for hash functions aren't very high,
it's problematic. It might be possible to use a mix function on the
hash value and then use a power of 2, although that adds to the
platform specific issues, which I've been trying to avoid - I've had
enough problems dealing with varying compilers. I'll have to look into
how much faster it is (if at all). I originally developed this on 32
bit machines and it didn't seem worth it there.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk