dpnp
dpnp copied to clipboard
IntelPython
Data Parallel Extension for NumPy
I'd like to see numpy used with [numexpr](https://numexpr.readthedocs.io/en/latest/mkl.html) in [comparison_generator.py](https://github.com/IntelPython/dpnp/blob/23090ce9285dd0bd00fd790ea25f1a3fc1843855/doc/comparison_generator.py).
I think the stride computation in dpnp arrays is not correct. ```python >>> import numpy as np >>> import dpnp >>> a = np.arange(0, 10, dtype=dpnp.int64) >>> a.strides (8,) >>>...
NumPy ufuncs have an attribute called [`nin`](https://numpy.org/doc/stable/reference/generated/numpy.ufunc.nin.html). `nin` returns the number of input arguments accepted by the ufunc. dpnp presently does not provide the attribute. The attribute is needed to...
I've caught strange behavior that does not raise error, but produce wired results. Looks like initialization writes out of buffer. ```python import dpnp size = 1000_000_000 a = dpnp.ones(size) b...
Results for _Game of life_ example (running on a laptop with 11th Gen processor and Iris Xe graphics): | example | numpy | dpnp CPU | dpnp GPU | size...
The PR pins DPC++ and OneMKL versions to pre-released ones from upcoming `2024.0` release. The PR updates cmake files to use `MKL::MKL_SYCL` instead of alias on old name `MKL::MKL_DPCPP`. The...
Results from dpbench of _gpairs_ workload (running on a laptop with 11th Gen processor and Iris Xe graphics): | workload | numpy | dpnp CPU | dpnp GPU | size...
``` x = dpnp.asarray([1.0, 2.0, 3.0]) y = 1.0/x ``` `y` is numpy array vs. dpnp array Complete reproducer is here https://gist.github.com/samaid/b4510dbcd7285ceed36b42dd96acf30c
``` python def asnumpy(input, order='C'): """ Returns the NumPy array with input data. Notes ----- This function works exactly the same as :obj:`numpy.asarray`. """ ``` ``` python def astype(x1, dtype,...
