Terminology (and associated types)

rainwoodman / vast

vala and scientific numerical computation

11 stars 1 forks source link

Terminology (and associated types) #12

Closed arteymix closed 7 years ago

arteymix commented 8 years ago

All this should probably end-up in the wiki.

Dimension

Number of independent axes.

dimension is size_t as it defines the size of strides and shape

Axis

ssize_t from the outside world for negative indexing
size_t internally for indexing strides and shape
plural form is axes

Index

ssize_t[] to allow negative indexing
length is implicitly given by the dimension
computes a size_t offset using strides and shape

Offset

size_t to index the raw memory

Strides

ssize_t[] to allow negative indexing
length is implicitly given by the dimension

Shape

size_t[] as it define the length of each axes
length is implicitly given by the dimension

Scalar

The scalar are the atoms of a dense array.

scalar_size is a size_t
scalar_type is a GType

Raw memory

GBytes that wraps a uint8[] array

arteymix commented 8 years ago

Just to mention, all the size_t and ssize_t are refeering to gsize and gssize as described in https://developer.gnome.org/glib/stable/glib-Basic-Types.html They are alias with introspection.

rainwoodman commented 8 years ago

1.In general offset could be negative. I know it is impossible if it is GBytes. If it turns out necessary we can always flip.

in numpy, the argument newshape to reshape(newshape) is signed. If a newshape is [-1] it will be automatically inferred from the total size and the non -1 elements. Very handy feature. but it means sometimes shape can be ssize_t too.

Would it hurt if we make all offsets / shapes are ssize_t?

arteymix commented 8 years ago

I like to keep everything unsigned internally and signed externally to compute reverse offset/index.

We never really expose offsets, only via indexes which are already signed.

For reshape wouldn't it be nicer to fill smallest size first? Instead of having -1 for inferred dimension, we could have something like size_t.MAX for unconstrained axes.

arr.reshape ({2, 2, size_t.MAX}); // fill 2 by 2 and as much as possible in the last axis

rainwoodman commented 8 years ago

That's a great idea!

arteymix commented 7 years ago

I moved everything into the wiki: https://github.com/rainwoodman/vast/wiki/Types-and-Terminology

I will copy the comment about reshaping in #7.