news 2026/6/13 10:48:55

CANN/PTO-ISA指令族契约规范

作者头像

张小明

前端开发工程师

1.2k 24
文章封面图
CANN/PTO-ISA指令族契约规范

6. Instruction families and contracts

【免费下载链接】pto-isaParallel Tile Operation (PTO) is a virtual instruction set architecture designed by Ascend CANN, focusing on tile-level operations. This repository offers high-performance, cross-platform tile operations across Ascend platforms.项目地址: https://gitcode.com/cann/pto-isa

6.1 Scope

This chapter defines family-level normative contracts. Per-op normative details remain indocs/isa/*.md.

6.2 Family taxonomy

PTO instruction families:

  1. synchronization and resource binding
  2. elementwise tile-tile operations
  3. tile-scalar and tile-immediate operations
  4. axis reduce and expand operations
  5. memory operations (GM <-> Tileand indexed variants)
  6. matrix multiply and GEMV operations
  7. data movement and layout transforms
  8. irregular/complex operations

The source-synchronized inventory is maintained bydocs/isa/manifest.yaml.

6.3 Common family contract

Every instruction family MUST define:

  • operand/result classes and position rules
  • semantic domain (valid-region handling)
  • required constraints (dtype/layout/location/shape)
  • synchronization/ordering implications
  • diagnostics behavior for illegal use
  • implementation-defined boundaries

6.4 Valid-region-first rule

Unless a specific instruction states otherwise:

  • semantics are defined only on the operation's valid domain
  • out-of-domain results are unspecified
  • family contracts MUST state domain-composition rules for multi-input operations

6.5 Family-level summaries

6.5.1 Synchronization and resource binding

IncludesTSYNC,TASSIGN, mode/config instructions. These operations define ordering or state-configuration effects and MUST preserve architecture ordering semantics.

6.5.2 Elementwise and scalar variants

Includes arithmetic, bitwise, compare, select, unary math, and scalar-fused forms. Operations MUST define per-element behavior and mode-specific constraints.

6.5.3 Reduce/expand families

Includes row/column reductions and broadcast-like expansions. Operations MUST define axis semantics and domain compatibility.

6.5.4 Memory families

Includes load/store/prefetch and indexed gather/scatter forms. Operations MUST define mapping between tile domains and memory domains.

6.5.5 Matrix families

IncludesTMATMUL*andTGEMV*families. Contracts MUST define accumulation domain, operand-role legality, and precision-mode interactions.

6.5.6 Movement/layout families

Includes extract/insert/reshape/transpose/fillpad-like transforms. Contracts MUST define index mapping and domain preservation rules.

6.5.7 Complex/irregular families

Includes sort/quant/partial/gather variants and other special operations. Contracts MUST explicitly identify implementation-defined portions.

6.6 Documentation contract for per-op pages

Each per-instruction page SHOULD follow Appendix B template sections:

  • Syntax
  • Operands
  • Semantics
  • Constraints
  • Diagnostics
  • Implementation-defined behavior
  • Compatibility notes

6.7 Coverage and synchronization policy

Family and instruction indexes MUST stay synchronized with:

  • docs/isa/manifest.yaml
  • include/pto/common/pto_instr.hpp
  • generated index/matrix tooling indocs/tools/

【免费下载链接】pto-isaParallel Tile Operation (PTO) is a virtual instruction set architecture designed by Ascend CANN, focusing on tile-level operations. This repository offers high-performance, cross-platform tile operations across Ascend platforms.项目地址: https://gitcode.com/cann/pto-isa

创作声明:本文部分内容由AI辅助生成(AIGC),仅供参考

版权声明: 本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若内容造成侵权/违法违规/事实不符,请联系邮箱:809451989@qq.com进行投诉反馈,一经查实,立即删除!
网站建设 2026/6/6 14:45:28

无人机/农机自动驾驶避坑指南:GNSS-RTK/INS紧组合为何比松组合更抗干扰?

无人机与农机自动驾驶避障技术&#xff1a;GNSS-RTK/INS紧组合的实战解析在农田、果园或建筑工地等复杂环境中&#xff0c;自动驾驶设备常面临卫星信号遮挡、多路径干扰等挑战。传统松组合方案在信号中断时定位误差可能迅速累积至米级&#xff0c;而紧组合技术通过深度耦合惯性…

作者头像 李华
网站建设 2026/6/8 16:11:51

用快马ai快速生成你的第一个vscode插件原型,十分钟验证创意

快速体验 打开 InsCode(快马)平台 https://www.inscode.net输入框内输入如下内容&#xff1a; 请生成一个vscode插件项目的初始代码&#xff0c;该插件功能是&#xff1a;在编辑器侧边栏显示一个自定义视图&#xff0c;用于快速查看和插入常用的代码片段&#xff0c;视图应包…

作者头像 李华
网站建设 2026/6/6 14:45:13

为什么92%的SaaS初创公司试用CSDN AI后30天内停用?——5个被官方文档刻意弱化的集成断点与1套即插即用评估清单

更多请点击&#xff1a; https://codechina.net 第一章&#xff1a;SaaS 企业适合用 CSDN AI 数字营销吗&#xff1f; CSDN AI 数字营销平台聚焦于开发者社区流量分发与技术内容智能触达&#xff0c;其核心能力包括技术博客自动摘要生成、关键词语义聚类、开发者画像建模及精准…

作者头像 李华
网站建设 2026/6/6 14:44:09

LikeC4物联网架构:边缘计算与云服务的可视化集成

LikeC4物联网架构&#xff1a;边缘计算与云服务的可视化集成 【免费下载链接】likec4 Visualize, collaborate, and evolve the software architecture with always actual and live diagrams from your code 项目地址: https://gitcode.com/GitHub_Trending/li/likec4 …

作者头像 李华