Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update 05Qnnpack.md #279

Merged
merged 1 commit into from
Aug 18, 2024
Merged

Update 05Qnnpack.md #279

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion 04Inference/06Kernel/05Qnnpack.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,7 +94,7 @@ Marat Dukhan 于 2019 年离开 Meta 来到谷歌之后，发表了一篇名为

这标志着 QNNPACK 与其他 GEMM 实现之间的重要区别：其他库会重新打包 A 和 B 矩阵，以更好地利用缓存层次结构，希望通过大量计算来摊销打包开销，而 QNNPACK 则针对 A 和 B 面板可以适应 L1 缓存的情况进行优化。因此，**它旨在消除所有非计算必需的内存转换**。

间接卷积算法的计算同样也是基于对输出的切分，计算 MR × NR 的小块。与传统 GEMM 方法不同的是，其将将整个 K 维全部在计算 Kernel 中处理完，消除了输出部分和的访存。这里所说的「将整个 K 维全部」并不是指 K 维不进行拆分，而是指拆分后不和其他维度交换，实际计算中 K 维会以 $2^n$ 为基础进行拆分。
间接卷积算法的计算同样也是基于对输出的切分，计算 MR × NR 的小块。如图19.5.4所示，与传统 GEMM 方法不同的是，其将将整个 K 维全部在计算 Kernel 中处理完，消除了输出部分和的访存。这里所说的「将整个 K 维全部」并不是指 K 维不进行拆分，而是指拆分后不和其他维度交换，实际计算中 K 维会以 $2^n$ 为基础进行拆分。

![间接卷积算法示意图](images/05QNNPACK04.png)

Expand Down