CuPy 7.0.0b2: CUDA NumPy兼容多维数组库

需积分: 33 186 浏览量更新于2024-07-16 收藏 1.06MB PDF 举报

CuPy文档概述了CuPy项目，这是一个针对CUDA的NumPy兼容多维数组库，由Preferred Networks, Inc. 和 Preferred Infrastructure, Inc. 开发。CuPy的核心是`cupy.ndarray`类，它提供了对NumPy接口的子集支持。这个子集包括： 1. 基本索引：支持整数、切片、新轴和省略运算符（Ellipsis）进行数组访问。 2. 高级索引：除了部分使用布尔掩码的复杂模式外，大部分高级索引操作也是支持的。 3. 数据类型：CuPy支持多种基本数据类型，如bool_、int8、int16、int32、int64、uint8等，以及浮点数(float16、float32、float64)和复数(complex64、complex128)。 4. 数组创建函数：提供诸如`empty`、`ones_like`、`diag`等用于生成空数组、与已有数组形状相似的全1数组和对角线数组等常用功能。 5. 数组操作：包含如`reshape`、`rollaxis`、`concatenate`等操作，用于调整数组形状和合并数组。 6. 广播操作：支持各种运算符实现元素级别的广播计算。 7. 元素wise操作：除了一些复数运算函数外，大多数通用函数都得到支持。 8. 线性代数：包括矩阵乘法、分解（如Cholesky分解、SVD等），利用cuBLAS加速。 9. 沿轴的聚合函数：如求和、最大值、argmax等。 10. 性能特性：用户可以自定义CUDA内核执行元素操作和聚集操作，通过优化内联多个操作来提高效率；支持可配置的内存分配器和内存池；提供cuDNN工具。 CuPy文档还包括章节如教程、参考手册、API兼容性政策、贡献指南、安装指南、升级指南以及许可证信息，为开发者提供了全面的学习和使用资源。此外，文档还列出了Python模块索引和索引，方便用户查找特定功能和内容。总体上，CuPy作为一个高效的深度学习加速工具，为CUDA环境下的科学计算和机器学习任务提供了强大的支持。

CuPy Documentation, Release 7.0.0b2

(continued from previous page)

>>> x_np = np.arange(10)

>>> y_np = np.arange(10)[::-1]

>>> squared_diff(x_np, y_np)

array([81, 49, 25, 9, 1, 1, 9, 25, 49, 81])

At the ﬁrst function call, the fused function analyzes the original function based on the abstracted information of

arguments (e.g. their dtypes and ndims) and creates and caches an actual CUDA kernel. From the second function call

with the same input types, the fused function calls the previously cached kernel, so it is highly recommended to reuse

the same decorated functions instead of decorating local functions that are deﬁned multiple times.

cupy.fuse() also supports simple reduction kernel.

>>> @cp.fuse()

... def sum_of_products(x, y):

... return cupy.sum(x

y, axis = -1)

You can specify the kernel name by using the kernel_name keyword argument as follows:

>>> @cp.fuse(kernel_name='squared_diff')

... def squared_diff(x, y):

... return (x - y)

(x - y)

Note: Currently, cupy.fuse() can fuse only simple elementwise and reduction operations. Most other routines

(e.g. cupy.matmul(), cupy.reshape()) are not supported.

12 Chapter 2. Tutorial

CuPy Documentation, Release 7.0.0b2

• kth (int or sequence of ints) – Element index to partition by. If supplied with

a sequence of k-th it will partition all elements indexed by k-th of them into their sorted

position at once.

• axis (int or None) – Axis along which to sort. Default is -1, which means sort along

the last axis. If None is supplied, the array is ﬂattened before sorting.

Returns Array of the same type and shape as a.

Return type cupy.ndarray

See also:

cupy.argpartition() for full documentation, numpy.ndarray.argpartition()

argsort(self, axis=-1) → ndarray

Returns the indices that would sort an array with stable sorting

Parameters axis (int or None) – Axis along which to sort. Default is -1, which means

sort along the last axis. If None is supplied, the array is ﬂattened before sorting.

Returns Array of indices that sort the array.

Return type cupy.ndarray

See also:

cupy.argsort() for full documentation, numpy.ndarray.argsort()

astype(self, dtype, order=u’K’, casting=None, subok=None, copy=True) → ndarray

Casts the array to given data type.

Parameters

• dtype – Type speciﬁer.

• order ({'C', 'F', 'A', 'K'}) – Row-major (C-style) or column-major (Fortran-

style) order. When order is ‘A’, it uses ‘F’ if a is column-major and uses ‘C’ otherwise.

And when order is ‘K’, it keeps strides as closely as possible.

• copy (bool) – If it is False and no cast happens, then this method returns the array itself.

Otherwise, a copy is returned.

Returns If copy is False and no cast is required, then the array itself is returned. Otherwise, it

returns a (possibly casted) copy of the array.

Note: This method currently does not support casting, and subok arguments.

See also:

numpy.ndarray.astype()

choose(self, choices, out=None, mode=u’raise’)

clip(self, a_min=None, a_max=None, out=None) → ndarray

Returns an array with values limited to [a_min, a_max].

See also:

cupy.clip() for full documentation, numpy.ndarray.clip()

conj(self ) → ndarray

16 Chapter 3. Reference Manual

剩余320页未读，继续阅读

ENOCH_Q

粉丝: 49
资源: 20

CuPy 7.0.0b2: CUDA NumPy兼容多维数组库

SmallPebble:使用NumPyCuPy用Python从头开始编写的最小深度学习库

Python库 | cupy_cuda80-6.3.0-cp37-cp37m-manylinux1_x86_64.whl

cupy‑9.6.0+cuda114‑cp38‑cp38‑win_amd64.whl

cupy-release-tools：CuPy释放工具

linux 安装cupy

anaconda安装cupy

ubuntu安装cupy

linux安装cupy

python安装cupy

cupy ModuleNotFoundError: No module named 'cupy'

最新资源