openmlsys-zh/zh_chapters/chapter_accelerator/summary.md at ea4d0d178e9217712d85ef60277af4c358e7ab5d

mirror of https://github.com/openmlsys/openmlsys-zh.git synced 2026-03-31 01:11:48 +08:00

Files

anyin233 00db02dbfd fix: fix equation rendering by changing the toolchain to mathjax (#493 )

* docs: update README and build guide

* fix: escape * and _ inside math to prevent markdown emphasis corruption

* fix: configure MathJax to use TeX (Computer Modern) font

* feat: enhance markdown processing with label and figure collection

* fix: remove duplicate bibliography directives from chapter summaries

References are already handled at the chapter level, so the
:bibliography: directives in summary pages are redundant and cause
rendering issues.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

2026-03-12 06:21:56 +00:00

753 B

Raw Blame History

总结

面向深度学习计算任务，加速器通常都是由多种片上缓存以及多种运算单元组成来提升性能。
未来性能增长需要依赖架构上的改变，即需要利用可编程的硬件加速器来实现性能突破。
出于计算效率和易用性等原因，加速器一般会具有多个等级的编程方式，包括：算子库层级，编程原语层级和指令层级。
越底层的编程方式越能够灵活地控制加速器，但同时对程序员的能力要求也越高。

扩展阅读

CUDA编程指导 CUDA
昇腾社区 Ascend
MLIR应用进展 MLIR

753 B Raw Blame History

总结

扩展阅读

753 B

Raw Blame History