A NumPy-compatible matrix library accelerated by CUDA

CuPy is an open-source matrix library accelerated with NVIDIA CUDA.
CuPy provides GPU accelerated computing with Python.
CuPy uses CUDA-related libraries including cuBLAS, cuDNN, cuRand, cuSolver, cuSPARSE, cuFFT and NCCL to make full use of the GPU architecture.

The figure shows CuPy speedup over NumPy.
Most of them perform well on a GPU using CuPy out of the box. CuPy speeds up some operations more than 100X.
You can read original benchmark article in Single-GPU CuPy Speedups (RAPIDS AI).

CuPy's interface is highly compatible with NumPy; in most cases it can be used as a drop-in replacement.
All you need to do is just replace `numpy`

with `cupy`

in your Python code.
Basics of CuPy (Tutorial) is usefull to learn first step of CuPy.

CuPy supports various methods, indexing, data types, broadcasting and more.
Comparison Table (Reference Manual) shows a list of NumPy / SciPy APIs and its corresponding CuPy implementations.

```
>>> import cupy as cp
>>> x = cp.arange(6).reshape(2, 3).astype('f')
>>> x
array([[ 0., 1., 2.],
[ 3., 4., 5.]], dtype=float32)
>>> x.sum(axis=1)
array([ 3., 12.], dtype=float32)
```

The easiest way to install CuPy is to use pip.
CuPy provides Wheels (precompiled binary packages) for the recommended environments. These packages include cuDNN and NCCL.
Please read Install CuPy (Installation Guide).

CuPy can be installed from source code.
The install script in the source code automatically detects installed versions of CUDA, cuDNN and NCCL in your environment.

```
# For CUDA 8.0
pip install cupy-cuda80
# For CUDA 9.0
pip install cupy-cuda90
# For CUDA 9.1
pip install cupy-cuda91
# For CUDA 9.2
pip install cupy-cuda92
# For CUDA 10.0
pip install cupy-cuda100
# For CUDA 10.1
pip install cupy-cuda101
# Install CuPy from source
pip install cupy
```

You can easily make a custom CUDA kernel if you want to make your code run faster, requiring only a small code snippet of C++.
CuPy automatically wraps and compiles it to make a CUDA binary.
Compiled binaries are cached and reused in subsequent runs.
Please read User-Defined Kernels (Tutorial).

And, you can also use raw CUDA kernel via Raw modules (Tutorial).

```
>>> x = cp.arange(6, dtype='f').reshape(2, 3)
>>> y = cp.arange(3, dtype='f')
>>> kernel = cp.ElementwiseKernel(
... 'float32 x, float32 y', 'float32 z',
... '''if (x - 2 > y) {
... z = x * y;
... } else {
... z = x + y;
... }''',
... 'my_kernel')
>>> kernel(x, y)
array([[ 0., 2., 4.],
[ 0., 4., 10.]], dtype=float32)
```