Closed
[Requirement|需求建议]: Conv2d和Conv3d在Ascend960上支持weight内置Transpose #8765
luyakun_hwAdd on  8 days agoClosed at  4 days ago
luyakun_hw
luyakun_hwCollaborators
8 days ago Create

Backgroud(背景信息)

Conv2d和Conv3d在Ascend960上支持weight内置Transpose

Origin(信息来源)

NA

Benefit / Necessity (价值/作用)

Design(设计方案)

likedislike
luyakun_hwluyakun_hwCollaborators
8 days ago Added label: requirement
luyakun_hw
luyakun_hwCollaborators
8 days ago Comments:

/assign

likedislike
CANN-robotCANN-robotCollaborators
8 days ago Assign luyakun_hw as responsible person
yuning_chen
yuning_chenCollaborators
5 days ago Comments:

收到需求建议,欢迎您参加 ops-nn sig 例会评审并申报相关议题,最近的例会议题申报及会议时间可见:https://etherpad-cann.meeting.osinfra.cn/p/sig-ops-nn

本条要点: Backgroud(背景信息)

likedislike
CANN-robotCANN-robotCollaborators
4 days ago Close issue
CANN-robotCANN-robotCollaborators
4 days ago changed custom state from 待办的 to 已完成
CANN-robotCANN-robotCollaborators
4 days ago Added label: resolved
chenjiayi_
chenjiayi_Collaborators
2 days ago Comments:

补充说明(输出质量)

背景:Conv2d 与 Conv3d 需要在 Ascend960 上支持 weight 内置 Transpose,避免额外 transpose 算子。

影响:减少一次权重 transpose 的搬运和内存占用。

方案/范围:在 def/attr 中增加内置 transpose 开关,tiling/kernel 支持对应权重布局;补充 960 用例;当前未检索到关联 PR。

验收标准:960 上带内置 transpose 的 conv 精度通过;无开关路径不变;文档说明支持范围。

当前状态:本 issue 当前为 closed;暂无关联 PR。

likedislike