This repo contains a BloomFilter chisel module, a scala-implemented reference model, several hash functions and test code.

This implementation is a 2-staged bloom filter, which means lookup result will be given in next cycle (limited by SyncReadMem).

sbt test

A new hash function class should inherite HashFunc, implement hash_scala to return the hash value for scala model, and hash_chisel to return the hardware implementation. Hash functions should consider input_width as width of data to be hashed, and use hash_width as width of hash value. HashFunc_Modulo is a good example.

BloomFilter and BloomFilterModel requires BloomFilterParams as parameter. The BloomFilterParameter requires following input when instantiates:

val hash_funcs: Seq[HashFunc] // a sequence of HashFunc object
val data_width: Int           // input data width
val array_size: Int           // BloomFilter array size

1. Test clear memory

2. Test insert data and readback

3. Test pipelined lookup/insert

4. Test Chisel implementation against Scala model

Following hash functions are implemented:

1. Modulo (hash = input % hash_width)

2. Murmur3 Wikipedia

3. FNV Wikipedia

1. Currently the BloomFilter chisel module will instantiates n_hash_funcs + 1 number of memory write ports (extra 1 port is for clearing chisel SyncReadMem). However even in worst case only n_hash_funcs write ports will be used at same time (write and clear won't happen at same time). The extra port can be merged.

2. Futher more, clearing speed can be increased by using existing hash function write ports.