rapidsai / legate-boost

GBM implementation on Legate
https://rapidsai.github.io/legate-boost/
Apache License 2.0
8 stars 8 forks source link

Perform quantisation in 32 atomics #154

Closed RAMitchell closed 1 month ago

RAMitchell commented 2 months ago

Experiments show we can reduce precision down to 32 bits and still get accurate results. This allows us to increase the throughput of atomics in the histogram kernel.

The implementation is somewhat hacky, involving reinterpret casts, although the code changes are not large.

image