https://flagos-ai.github.io/FlagGems/zh-cn/overview/Recent content in 简介 on FlagGemsHugozh-cnhttps://flagos-ai.github.io/FlagGems/zh-cn/overview/features/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/zh-cn/overview/features/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <!-- # Features Overview ## Rich Collection of Operators FlagGems features a large collection of PyTorch compatible operators. Refer to [operator supported](/FlagGems/references/operators/) and [experimental operators](/FlagGems/references/experiental/) for list of formally supported operators and experimental operators. --> <h1 id="功能特性概览">功能特性概览<a class="anchor" href="#%e5%8a%9f%e8%83%bd%e7%89%b9%e6%80%a7%e6%a6%82%e8%a7%88">#</a></h1> <h2 id="丰富的算子集合">丰富的算子集合<a class="anchor" href="#%e4%b8%b0%e5%af%8c%e7%9a%84%e7%ae%97%e5%ad%90%e9%9b%86%e5%90%88">#</a></h2> <p><em>FlagGems</em> 的一大特性是提供丰富的 PyTorch 兼容算子集合。 参阅<a href="https://flagos-ai.github.io/FlagGems/zh-cnreferences/operators/">支持的算子</a>和 <a href="https://flagos-ai.github.io/FlagGems/zh-cn/references/experimental/">实验性算子</a> 页面了解正式支持的算子集合以及当前处于实验阶段的其他算子。</p>https://flagos-ai.github.io/FlagGems/zh-cn/overview/platforms/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/zh-cn/overview/platforms/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <!-- # Platforms Supported --> <h1 id="支持的硬件平台">支持的硬件平台<a class="anchor" href="#%e6%94%af%e6%8c%81%e7%9a%84%e7%a1%ac%e4%bb%b6%e5%b9%b3%e5%8f%b0">#</a></h1> <!-- *FlagGems* supports 10+ backends. The currently supported platforms and their capabilities are listed below: --> <p><em>FlagGems</em> 支持超过 10 种不同后端硬件平台。目前支持的平台及这些平台的能力列举如下：</p>https://flagos-ai.github.io/FlagGems/zh-cn/overview/pointwise-dynamic/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/zh-cn/overview/pointwise-dynamic/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <!-- # Pointwise Dynamic Operators ## 1. Pointwise Operations --> <h1 id="关于逐点动态算子">关于逐点动态算子<a class="anchor" href="#%e5%85%b3%e4%ba%8e%e9%80%90%e7%82%b9%e5%8a%a8%e6%80%81%e7%ae%97%e5%ad%90">#</a></h1> <h2 id="pointwise-operations">1. 逐点操作<a class="anchor" href="#pointwise-operations">#</a></h2> <!-- Pointwise operators are trivial to parallelize. Most parallel programming guides begin with pointwise addition between 2 contiguous vectors. For [`vector_add` in Triton](https://triton-lang.org/main/getting-started/tutorials/01-vector-add.html#sphx-glr-getting-started-tutorials-01-vector-add-py), it is simple to implement a task partitioning schema that each CTA reads a contiguous range from each input vector and writes to a contiguous range of the output vector. --> <p>逐点算子（Pointwise operators）比较容易并行化执行。 大多数并行计算编程指南的开篇都会使用计算两个连续向量的单点加和操作作为示例。 对于 <a href="https://triton-lang.org/main/getting-started/tutorials/01-vector-add.html#sphx-glr-getting-started-tutorials-01-vector-add-py">Triton 语言中的 <code>vector_add</code></a> 而言，很容易实现一种任务切分模式，让每个<strong>块集群（Cooperative Thread Array，CTA）</strong> 从每个输入向量中读入一个连续的范围，向输出向量中的一个连续范围写出结果。</p>