James Avery

Bohrium 100 times slower than numpy, time "unaccounted for"

3

When running with Bohrium or bh107, a crazy amount of time is being spent in "unaccounted for". The following code runs in 0.27 seconds in pure Numpy, but takes 25...

bh107: Can't multiply from the left

1

Binary operations with scalars are implemented only when the scalar is on the right. Negation isn't working either. Example: ```python import numpy as np import bh107 as bh xs =...

Both bh107 and bohrium explode when applying libmath functions (sin, cos, tanh, exp, etc.) to float64 data in OpenCL mode. ```python In [1]: import bohrium as bh In [2]: A...

OK, so the dtype is not defined in bh107: ```python In [1]: import bh107 as bh In [2]: a = bh.array([1,2,3,4],dtype=bh.float64) --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in ---->...

Indexing with arrays only seems to work with 1D arrays

3

Small example: ``` a = np.arange(10); ix = a[::-1]; print(a[ix]) # works fine b = a[:,None]*a[None,:]; print(b[:,ix]) # Explodes in array_create.py, line 147 print(b[ix,:]) # Explodes in array_create.py, line 147...

Indexing with array gives wrong result for negative entries

1

Example: ```python a = np.arange(1,5); ix = a[:,None]-a[None,:]; print(a[ix]) ``` Numpy result: ```python [[1 4 3 2] [2 1 4 3] [3 2 1 4] [4 3 2 1]] ```...

The standard NumPy eigensolver bohrium.linalg.eigh isn't implemented, and it tries to use numpy. **This is not the problem.** However, instead of handing it gracefully off to NumPy, it crashes: ```python...

Bohrium does not support ufunc `modf` it will be handled by the original NumPy.

1

modf (part of math.h, splits floats into integer and fractional part) is missing. Is that because it has two outputs per element?

No progress bar, even with test-example.cpp

2

I was very happy to find an easy-to use tqdm analogue for C++, then quickly saddened that it seems very difficult to actually get a progress bar. Here is what...