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.

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

change the name of activation kernel #32374

Closed

Conversation

AnnaTrainingG
Copy link
Contributor

@AnnaTrainingG AnnaTrainingG commented Apr 20, 2021

PR types

Breaking changes

PR changes

OPs

Describe

change the name of activation kernel from REGISTER_ACTIVATION_GPU_KERNEL to REGISTER_ACTIVATION_HIGH_PERFORMANCE_KERNEL in activation_op.cu

序号 算子名称 reduce调用方式 适配说明 PR链接
0 reduce_sum KernelFunction 完成适配 #32885
1 reduce_mean KernelFunction 完成适配 #32885
2 reduce_max KernelFunction 完成适配 #32974
3 reduce_min KernelFunction 完成适配 #32974
4 reduce_any KernelFunction 完成适配 #32974
5 reduce_all KernelFunction 完成适配 #32974
6 reduce_prod KernelFunction 完成适配 #32974
7 broadcast 反向 KernelFunction 进行中  
8 argmax KernelFunction 未开始  
9 argmin KernelFunction 未开始  
10 avgpool KernelFunction 针对特定规模  
11 sumpool KernelFunction 针对特定规模  
工作序号 说明 预计完成时间 当前工作状态 文档及代码链接
1 reduce性能优化方案设计 4月8日 完成 http://agroup.baidu.com/paddle-perf/md/article/3839656
         
2 方案原型验证: 4月16日 完成 https://github.com/niuliling123/cudaTest/blob/main/add_same_shape/BlockingReduce.cu
  2.1 blocking load数据   完成  
  2.2 reduction:使用shareMem进行reduce 还是直接load 就求和 完成  
         
3 性能优化升级及通用性设计   完成 #32697
  3.1 向量化   完成  
  3.2 kernel实例化,根据size大小进行循环展开等   完成  
  3.3 load/compute 并行   完成  
         
4 进行通用性适配 5月11日 5月30日   #32884
  4.1 多维度reduce性能优化   进行中  
  4.2 对argmax/argmin/norm等op中的reduce操作进行替换   进行中  
axis case pytorch paddle_old paddle_new Perf Optimization perf diff.
axis = 1 [2  512 2048] 20.069 58.785 13.95 -76.27% -43.86%
axis = 1 [2  128 1024] 5.864 9.161 3.931 -57.09% -49.17%
axis = 1 [2  30522   1024] 296.54 4734 296.019 -93.75% -0.18%
axis = 1 [2  1024    16] 5.163 2.678 3.859 44.10% -33.79%
axis = 1 [2  256 12800] 33.255 173.95 37.317 -78.55% 10.89%
axis = 1 [2  256 10240] 27.616 120.5 30.512 -74.68% 9.49%
axis = 1 [2  1024    1280] 18.325 66.735 16.546 -75.21% -10.75%
axis = 1 [2  32768   1280] 390.39 8531.5 389.891 -95.43% -0.13%
axis = 1 [2  30522   10240] 1420.7 71629 2880.216 -95.98% 50.67%
axis = 1 [2  1024    1280] 17.84 57.585 16.525 -71.30% -7.96%
axis = 1 [2  32768   1280] 390.53 8500 389.821 -95.41% -0.18%
axis = 1 [2  2560    10240] 244.58 1157.2 247.311 -78.63% 1.10%
axis = 1 [2  10240   1280] 132.68 1228.3 129.352 -89.47% -2.57%
axis = 1 [2  32768   2560] 762.9 19096 767.538 -95.98% 0.60%
axis = 1 [2  30522   1024] 296.87 4708.1 296.251 -93.71% -0.21%

@paddle-bot-old
Copy link

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@AnnaTrainingG AnnaTrainingG changed the title Activation name change the name of activation kernel Apr 20, 2021
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.

1 participant