una-dinosauria / Rayuela.jl

Code for my PhD thesis. Library of quantization-based methods for fast similarity search in high dimensions. Presented at ECCV 18.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Support different codebook sizes in CUDA kernels

dryman opened this issue · comments

Does this package support other dimensions than 128?
We'd like to test it on GloVe, which is 100.

Yep, just add your data reading functions to

function read_dataset(

Does the following code in cuda matter?

// FIXME hard-coding 256 entries in each codebook, and 128 dimensions
const int H = 256; // size of each codebook
const int D = 128; // dimensionality of each vector

Oh right, sorry. The GPU part is hardcoded.

I checked again, since I definitely ran this code on the GPU for my thesis.

The function veccost that you linked above is not called from the Julia part. veccost2 is

// FIXME hard-coding 256 entries in each codebook
const int H = 256; // size of each codebook

which only hardcodes the number of entries in the codebook (not the dataset dimensionality). I'm changing the name of this issue to reflect that.

Also opened #41 to keep track of code deletion.