[Docs] Add operator library. (#47)

Author: zhaowt96

docs.js

@@ -171,8 +171,8 @@
       titleEn: "RuyiAI Operator Library",
       desc: "面向 RISC-V 的原生高性能算子库",
       descEn: "Native high-performance operator library for RISC-V",
-      markdown: "欢迎查阅 RuyiAI 算子库文档。\n\n（待补充 TBD）",
-      markdownEn: "Welcome to RuyiAI Operator Library documentation.\n\n(TBD)"
+      markdownUrl: "docs/operator-lib/operator-lib.md",
+      bodyOnly: true
     },
     runtime: {
       title: "RuyiAI 运行时环境",

docs/operator-lib/operator-lib.en.md

@@ -0,0 +1,9 @@
+# RuyiAI Operator Library
+
+The RuyiAI Operator Library targets efficient operator execution needs in AI computing and builds a unified operator support system that is tightly coordinated with the compilation pipeline. After graph-representation fusion and compilation processing, generated operator representations are further connected to the operator library, which provides implementation, adaptation, and invocation capabilities. This closes the key loop from graph compilation to operator execution. Through deep collaboration with the compiler, the RuyiAI Operator Library not only carries low-level operator implementations, but also plays a critical role in connecting model representation, compilation outputs, and target hardware execution.
+
+![RuyiAI Operator Library](/docs/operator-lib/operator-lib.png)
+
+The capabilities of the RuyiAI Operator Library mainly come from three sources. First, **vendor-customized operators** for specific hardware platforms, used to fully exploit the performance advantages of specialized architectures and extension instructions. Second, **porting and reuse of existing high-performance operator libraries**, which brings mature operator capabilities into a unified system and improves baseline coverage and performance. Third, **AI-generated operators** based on automation techniques, which improve new-operator construction efficiency and platform adaptation capability. Through multi-source integration, the RuyiAI Operator Library is gradually forming an open, extensible, and continuously evolving operator implementation system.
+
+At operator execution time, RuyiAI further builds an operator-selection method oriented to target hardware characteristics. The system matches and selects among candidate operator implementations by considering key hardware features such as vector-register resources, array scale, memory hierarchy, and accumulation resources, so that each platform can use a more suitable execution path. Because different hardware platforms vary significantly in compute organization, data access patterns, and resource configuration, the same operator often requires different implementations to reach better performance. With this hardware-aware operator-selection mechanism, RuyiAI efficiently connects operator-library capabilities with target-platform capabilities, improving overall execution efficiency and deployment adaptability.

docs/operator-lib/operator-lib.md

@@ -0,0 +1,9 @@
+# RuyiAI 算子库
+
+RuyiAI 算子库面向人工智能计算中的高效算子执行需求，构建了与编译流程紧密协同的统一算子支撑体系。系统在图表示融合和编译处理之后，将生成的算子表示进一步对接到算子库中，由算子库提供对应的实现、适配与调用能力，从而打通从图编译到算子执行的关键链路。通过与编译器的深度协同，RuyiAI 算子库不仅承载底层算子实现，也承担连接模型表达、编译结果与目标硬件执行的重要作用。
+
+![RuyiAI 算子库](/docs/operator-lib/operator-lib.png)
+
+RuyiAI 算子库的能力来源主要包括三个方面：一是面向特定硬件平台的**厂商定制算子**，用于充分发挥专用架构和扩展指令的性能优势；二是已有**高性能算子库的移植与复用**，将成熟算子能力引入统一体系，提升基础算子覆盖与性能基础；三是基于自动化技术形成的**AI 生成算子**，用于提高新算子的构建效率与平台适配能力。通过多来源融合，RuyiAI 算子库逐步形成开放、可扩展、可持续演进的算子实现体系。
+
+在算子执行阶段，RuyiAI 进一步构建了面向目标硬件特征的算子选择方法。系统会结合向量寄存器资源、阵列规模、存储层次以及累加资源等关键硬件特征，对不同候选算子实现进行匹配与选择，从而为不同平台选择更合适的执行路径。由于不同硬件在计算组织、数据访问和资源配置等方面存在显著差异，同一算子往往需要采用不同实现方式才能获得更优性能。RuyiAI 通过硬件感知的算子选择机制，实现算子库能力与目标平台能力之间的高效对接，提升整体执行效率与部署适配能力。