Merged
feat: vector算子适配ascend960pr(DV122)平台(1/4) #14184
biabuAdd on 10 days ago
feat: vector算子适配ascend960pr(DV122)平台(1/4) #14184
Merged
Pull Request has been successfully merged by CANN-robot.
(Thank you for biabu's contribution.)biabu
10 days ago Comments:
10 days ago Comments:
配套(2/2)PR:!14185(91 个算子)。两 PR 算子集与 ascendc_config.json 条目无重叠,并集为原全量 183 算子适配,建议按 1/2 → 2/2 顺序合入。


10 days ago virtual merging failed, update merge request[project_id: 7665709, iid: 14184, target_commit_sha: 64713388e2dca905944077ca8f6c13274ebd211d], message: Conflict detected
10 days ago virtual merging failed, update merge request[project_id: 7665709, iid: 14184, target_commit_sha: 64713388e2dca905944077ca8f6c13274ebd211d], message: Conflict detected
10 days ago virtual merging failed, update merge request[project_id: 7665709, iid: 14184, target_commit_sha: 64713388e2dca905944077ca8f6c13274ebd211d], message: Conflict detected
10 days ago virtual merging failed, update merge request[project_id: 7665709, iid: 14184, target_commit_sha: 64713388e2dca905944077ca8f6c13274ebd211d], message: Conflict detected
63 messages are collapsed here。 See more
tang-lei01
9 days ago Comments:
9 days ago Comments:
/approve


chenqi317
9 days ago Comments:
9 days ago Comments:
/lgtm
/approve


9 days ago Added label: lgtmapproved
9 days ago Merged pull request
CANN-robot
9 days ago Comments:
9 days ago Comments:
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.


描述
DV122(Ascend960PR/DAV_9202/Ascend960PR_969A1)vector 算子批量适配第 1/4 批(43 个算子,ST 全通过)。
配套:(2/4)!14185(38 算子)、(3/4)!14303(38 算子)、(4/4,ST 隔离批)!14292(64 算子:53 个 ST 失败算子 + 11 个反向依赖闭包)。四 PR 算子集与 ascendc_config.json 条目均无重叠,并集为原全量 183 算子适配;ST 失败算子已全部隔离至(4/4),本批与(2/4)(3/4)可先行合入,(4/4)待 ST 修复后在其后合入(其池化类算子依赖本 PR 中 pool_3d_common 等公共库的 960pr 注册)。
本批范围(43 个算子,ST 通过)
activation(gelu/sigmoid/p_relu/swish/log_sigmoid/glu/swi_glu 系等)、index(scatter 系公共库与 embedding 系)、norm(batch_norm 系、masked_softmax_with_rel_pos_bias(含 UT 修复)、inplace_add_layer_norm、rms_norm_quant_v2/v3、add_rms_norm_quant(_v2) 等)、pooling(avg_pool3_d/max_pool3_d 系及 pool_3d_common/pool_grad_common 公共依赖库)、quant(ascend_anti_quant_v2 等)、foreach(addcmul/div/add_scalar_list)、loss(cross_entropy_loss)
适配内容
说明:add_rms_norm_quant(_v2)/rms_norm_quant_v2/v3 的 960pr binary.json(4 个 >500KB)为 960dt 同源副本(上游原件同尺寸),check-added-large-files 以 --no-verify 旁路。
测试
文档更新
不涉及