NAME

nbperf — compute a perfect hash function

SYNOPSIS

nbperf

[-ps] [-a algorithm] [-c utilisation] [-h hash] [-i iterations] [-m map-file] [-n name] [-o output] [input]

DESCRIPTION

nbperf reads a number of keys one per line from standard input or input. It computes a minimal perfect hash function and writes it to stdout or output. The default algorithm is “chm”.

The -m argument instructs nbperf to write the resulting key mapping to map-file. Each line gives the result of the hash function for the corresponding input key.

The parameter utilisation determines the space efficiency.

Supported arguments for -a:

chm

This results in an order preserving minimal perfect hash function. The utilisation must be at least 2, the default. The number of iterations needed grows if the utilisation is very near to 2.

chm3

Similar to chm. The resulting hash function needs three instead of two table lookups when compared to chm. The utilisation must be at least 1.24, the default. This makes the output for chm3 noticable smaller than the output for chm.

bpz

This results in a non-order preserving minimal perfect hash function. Output size is approximately 2.79 bit per key for the default value of utilisation, 1.24. This is also the smallest supported value.

Supported arguments for -h:

mi_vector_hash

Platform-independent version of Jenkins parallel hash. See mi_vector_hash(3).

The number of iterations can be limited with -i. nbperf outputs a function matching uint32_t hash(const void * restrict, size_t) to stdout. The function expects the key length as second argument, for strings not including the terminating NUL. It is the responsibility of the caller to pass in only valid keys or compare the resulting index to the key. The function name can be changed using -n name. If the -s flag is specified, it will be static.

After each failing iteration, a dot is written to stderr.

nbperf checks for duplicate keys on the first iteration that passed basic hash distribution tests. In that case, an error message is printed and the program terminates.

If the -p flag is specified, the hash function is seeded in a stable way. This may take longer than the normal random seed, but ensures that the output is the same for repeated invocations as long as the input is constant.

EXIT STATUS

The nbperf utility exits 0 on success, and >0 if an error occurs.

SEE ALSO

mi_vector_hash(3)

AUTHORS

Jörg Sonnenberger