How should I define my own hash function for unordered_map in C++

Question

I have a database of image patches i.e. a 300*300 images with each patch of size 60*60. Thus giving me a grid of 5*5. I want to store some information against each of these patch. Since, my database can consist of millions of images, I wanted to use unordered_map so that finding a patch become easy.

My patch information consist of imgId, x_position of patch and y_position of patch (all integers). I don't have much experience with unordered_map. But going through some tutorials, I realised that I will have to use my own hash function. Can anyone suggest some efficient way of storing the above info in unordered_map with proper hash function.

Shall your hash be based on the identifying handle you gave, or on the underlying data in the database? — Deduplicator, Sep 17 '14 at 15:31
Use `std::hash` and see http://stackoverflow.com/questions/2590677/how-do-i-combine-hash-values-in-c0x — Mark Ransom, Sep 17 '14 at 15:34
@Deduplicator I need to find values on the basis of imgId,x and y position. That is the combination of them is primary key. — user3747190, Sep 17 '14 at 16:16

David Schwartz · Answer 1 · 2014-09-17T16:29:46.583

1

You can use any hash function you like to combine the three integers. Here's a classic Knuth 32-bit combiner:

int hash(int v1, int v2, int v3)
{
  int v = v1;
  v *= 2654435741;
  v += v2;
  v *= 2654435761;
  v ^= v3;
  v *= 2654435789;
  return v;
}

edited Sep 17 '14 at 16:29

answered Sep 17 '14 at 16:23

David Schwartz

179,497
17
214
278

I thought that is preferable that the 3 constants should be prime but different... Why use the same 3 times? – Basile Starynkevitch Sep 17 '14 at 16:28
@BasileStarynkevitch It doesn't make all that much difference. I changed them, for no particular reason. ;) – David Schwartz Sep 17 '14 at 16:29

How should I define my own hash function for unordered_map in C++

1 Answers1