v1.0.0 Checklist

[fluidnumericsJoe]

• Have a basic memory management interface for kaolin arrays for CPU (shared memory) and GPU (single GPU) #4
• Have basic operator overloading functionality for = for work between allocatable <-> kaolin_array, kaolin_array <-> kaolin_array #5
• Provide common initializers #6
• Save kaolin_array to HDF5 #7
• Easy to use logging features for use in Kaolin as well as other projects, e.g. kaolin_logger class; log to stdout,stderr or parseable file, e.g. json #8
• Exception handling utilities to use within Kaolin as as in other projects #9

[garrettbyrd]

• save to HDF5

[garrettbyrd]

Operation Overloading

Usual operators

= (assignment)

1. Automatically assign GPU array to CPU array and vice versa. E.g., karr_cpu = karr_gpu.
2. Automatically configure operation output based on device of array being assigned. E.g., karr_cpu2 = dot(karr_gpu, karr_cpu1) automatically places output on cpu. How to decide which device to perform operation on?

+ (addition)

1. Overload + to perform addition if arrays are the same shape (element-wise addition)
2. Related to (2) above, allow for addition of arrays on difference devices.

- (subtraction)

1. Element-wise subtraction
2. Similar to (2) above

* (multiplication)

1. Element-wise multiplication
2. Similar to (2) above
3. Allow scalar multiplication (consider arrays of shape (1) as scalar)

/ (division)

1. Element-wise division
2. Similar to (2) above
3. Similar to (3) above (scalar division)

Other operators?

@ (dot product)

• Inherited from PyTorch

^ (matrix exponentiation?)

• ???