Support Board
Date/Time: Thu, 01 Jan 2026 09:11:54 +0000
Post From: Upcoming change to using double instead of float or int
| [2025-12-31 12:44:07] |
| curious16 - Posts: 15 |
|
I read your announcement (ThreadID=103365) that you want to (or rather have to) move from integer volumes to supporting fractional volumes, and you plan to do this by moving to using double precision instead of float (single precision) for all the sc.Subgraph[] arrays. I fully understand that such fundamental changes sometimes are necessary. However, I must admit I am worried about forcing all data to be handled as 8 byte double precision instead of 4 byte float. For the vast majority of all data items the extra precision is not needed and causes overhead - extra storage, slower loading times, slower processing etc. SC is such a well-written, efficient application, really an exception in a world of poorly engineered, unnecessarily fat software. It would be a pity to see it being bulked up to accommodate some fairly edge-case scenarios. One idea which may perhaps be worth considering is to allow for different data types, e.g., to allow some sc.Subgraph arrays to be <int> or <float> while others are <double>. Of course there is advantage in having a uniform sc.BaseDataIn[][] array of arrays (e.g. addressing multiple values of the same bar using strides), but it seems that most data processing is done on the individual arrays. I often find that having some arrays set to integer would be helpful - maybe this major change is an opportunity to go beyond fixed <float> subgraphs. You surely will consider keeping SC as efficient as possible. Please keep up the excellent work and, if possible, provide options that allow for avoiding using <double> everywhere. Thanks for your great work and all the best for the New Year. |