已合并
conv:conv3dv2/extendconv2d/quantconv2d/quantconv3d 新增ascend350平台支持 #10795
conv:conv3dv2/extendconv2d/quantconv2d/quantconv3d 新增ascend350平台支持 #10795
已合并
XianyongWang创建于 24 天前
XianyongWang
XianyongWang成员
24 天前

描述

4 个 conv 系算子新增 ascend350 平台支持。修改模式(350 与 ascend950 完全同配,两者同为 DAV_3510,走同一代码分支):

  • QuantConv2D / QuantConv3D:def 直加 AddConfig("ascend350", aicore_config)(同一 config对象注册两次);op_kernel/CMakeLists.txt 的 COMPUTE_UNITS 加 ascend350;op_host/config/ascend350/ 拷贝 950 版 binary.json(各 9 bin)
  • ExtendConv2D:def 为双参 SetAscendConfig helper 形态(无 dst_version 参数),helper 调用后直接 AddConfig("ascend350", aicoreConfig);CMake 为ascend950 ascend350 mc62;conv2d_postcube_to_extendconv2d_fusion_pass.h 的SUPPORT_SOC_LIST 增加 {"Ascend350", NpuArch::DAV_3510}(明文 soc 门禁,npuarch 同构覆盖不到,不加则 350 图模式融合不生效)
  • QuantConv3D 同上追加 conv3d_dequant_to_quantconv3d_fusion_pass.h 明文 soc
  • Conv3DV2:def 为独立 config 对象形态(SetAscendConfig(aicoreConfig95, "ascend950")),同对象直加 AddConfig("ascend350", aicoreConfig95);kernel CMake 两段结构仅改ascend950 段;binary.json 11 bin。conv3d_to_conv3dv2 融合 pass 无需修改:其门禁为平台 intrinsic 能力探测(Intrinsic_data_move_out2l1_dn2nz 等),DAV_3510 同构自动覆盖
  • UT:4 个 tests/ut/op_kernel/CMakeLists.txt 的 AddOpTestCase 增加 ascend350_355e
  • 2 个 def 文件版权头规范化(extend_conv2d_def / quant_conv3d_def,OAT 合规要求)

关联的Issue

关联Issue #6010

测试

编译验证通过:4 算子 self config 日志齐全;二进制产物 38 bin 全量产出; run 包 173M;soc_version=Ascend350_355e 注入 76 处(38 bin × 2)

  • ttk kernel 精度验证:60/60 全过、binary 匹配全命中(无在线编译回退):

    算子 用例数 binary 匹配 精度 PASS
    QuantConv2D 21 21/21 21/21
    Conv3DV2 17 17/17 17/17
    ExtendConv2D 11 11/11 11/11
    QuantConv3D 11 11/11 11/11

类型标签

AI/Agent生成声明

likedislike
Pull Request已成功合入, 合并人@CANN-robot
(感谢 XianyongWang 的贡献)
XianyongWangXianyongWang成员
24 天前 创建了 pull request,commit 7fa2c95a
XianyongWangXianyongWang成员
24 天前 关联了issue:[Requirement|需求建议]: conv部分算子还未适配ascend350平台
CANN-robotCANN-robot成员
24 天前 添加了label:cann-cla/yes
CANN-robot
CANN-robot成员
24 天前 评论:

CLA Signature Pass

XianyongWang, thanks for your pull request. All authors of the commits have signed the CLA. 👍

likedislike
CANN-robot
CANN-robot成员
24 天前 评论:

Thanks for your pull-request.
The full list of commands accepted by me can be found at here.
You can get sig-info at here.
You can self-configure the PR merge rules for this repository. For more details, please refer to Here.
For more, you also can visit HICANN.


PR Approval Progress

✅ Congratulations! All modules have met the lgtm and approve requirements.

Module Approval Details

module lgtm status approve status
*/*/op_host/*_def.cpp ✅ 王永光, 陈华锋, 王慈孟 (3/2) ✅ 王永光 (1/1)
conv ✅ 王慈孟, 陈华锋 (2/2) ✅ 陈华锋, 王慈孟 (2/1)

💡 Tip:

  • Committer can comment /approve or /lgtm
  • Commenting /approve implies both code review (lgtm) and intent to merge (approve)

CLA Signature Pass

XianyongWang, thanks for your pull request. All authors of the commits have signed the CLA. 👍

likedislike
此处折叠了81条消息 查看更多
CANN-robotCANN-robot成员
20 天前 添加了label:approved
CANN-robotCANN-robot成员
20 天前 关闭了关联的issue
CANN-robotCANN-robot成员
20 天前 合入了pull request
CANN-robot
CANN-robot成员
20 天前 评论:

Pull Request 已合并或已关闭。

If you want to solve this problem, you can click here to do it in the FAQs.

likedislike
conv2d_ci
19 天前 评论:
pipeline history
TimeNameResultChangeidTimeStamp
2026-09-24 21:32:25commit2026-09-23T20:57:02+08:0004ea3d4a
SC_Result :x: WARNINGIbb1360d
RDV_Result
IndexSubTaskResult
1rdv_tbe_1911ep_01 :white_check_mark: SUCCESS (100%%)
2rdv_tbe_1952_01 :white_check_mark: SUCCESS (100%%)
3rdv_tbe_1971_01 :x: FAILURE
4rdv_tbe_1971_02 :x: FAILURE (99.04%%)
5rdv_tbe_david_02 :white_check_mark: SUCCESS (100%%)
:x: FAILUREIbb1360d
本次未执行的rdv任务为
序号任务名未执行理由
1rdv_tbe_1971_01None
2rdv_tbe_1981_01None
31980None
41910None
5Smoke_1911mdc切换至br_feature_pegasus_v100r003c10b770_main
6rdv_tbe_1911rc_01暂时下线
7Smoke_nano暂无用例
8rdv_tbe_david_03None
9rdv_tbe_davidv122_05None
10rdv_tbe_david_02None
11smoke_1980暂无1980环境,仅看护红线用例
12Smoke_1910miniNone
13transpose该组的RDV疑似不需要运行:[Davinci_RDV_conv_back_V121, conv_backward_1982_rdv, conv_backward_1971_rdv, conv_transpose_1952_rdv]
14amct该组的RDV疑似不需要运行:[GPU_AMCT, GPU_AMCT_02, GPU_310_AMCT, CPU_AMCT, Smoke_rdv_AMCT_1980]
likedislike