Skip to content

feat(conv): support grouped and depthwise convolution - #3

Open
yichao-mt wants to merge 1 commit into
masterfrom
yichao/feat_conv_group
Open

yichao-mt wants to merge 1 commit into
masterfrom
yichao/feat_conv_group

Conversation

@yichao-mt

Copy link
Copy Markdown
Collaborator

Summary

Conv 目前只支持 group=1,分组卷积与 depthwise 卷积都无法在 MUSA EP 上执行。
本 PR 让 muDNN 快路与自写 kernel 都支持 group > 1,并补上 CPU-vs-MUSA 的对照测试。
含分组 / depthwise 卷积的常见模型(例如 EfficientNet 系列)因此可以在 MUSA EP 上跑通。

Validation

  • Built successfully, or build not required for this change
  • Ran focused tests
  • Ran broader suite when touching shared runtime, allocator, stream, fusion, or data-transfer paths
  • Updated docs or generated docs when behavior changed — 未改动算子注册,musa/docs/ 无需更新

MUSA 5.2:

  ./build.sh                                  # OK
  python -m pytest test/ops/test_conv.py -q   # 6 passed
  bash test/run_all.sh -q                     # 1 failed, 1273 passed

@real-zhangzhe real-zhangzhe left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

在 4.3.8 / 5.1.0 测试了,没问题

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants