roberto-trani / mphf_benchmark Goto Github PK

Cmph is too old. Another CHD implement optimized for throughput: https://github.com/PeterRK/fastCHD

Hello,

I'm Piotr Beling, PhD from University of Łódź.
For my own research, I began to develop a fork of mphf_benchmark available at:
https://github.com/beling/mphf_benchmark

In my fork I:

• I've added xor-shifts 32 and 64 generators for keys,
• I've updated PTHash to the newest version and added its support for its partitioned variant,
• some methods (especially PTHash) are run also with multiple threads,
• only one variant of the method can be run with new variant flag,
• I fixed the bug that cause coping all the keys.

Please feel free to copy some or all my changes to your branch. I especially recommended the following one that fixes the bug with keys duplication:
beling@953a44c

Best regards
Piotr Beling

When I try to run bbhash with strings keys, then the program ends with MPHF contains out of range values exception.
bbhash works very well with integer keys.

Have you noticed this problem Giulio? Is it the fault of mphf_benchmark or bbhash?
I don't see anything suspicious in the mphf_benchmark code, while the bbhash code is a bit unreadable to me.

Example run:
cat uk-2005-nat.urls | ./mphf_benchmark bbhash

And output:
4 threads available for multi-threaded calculations Reading keys from stdin Read 39459925 keys, with average length 71.37 Algorithm BBhash(gamma=1, num_threads=1) Average Construction time: 57s 778ms (57.778s) Average Space usage: 3.06 bits/key terminate called after throwing an instance of 'std::runtime_error' what(): MPHF contains out of range values