onnxruntime
Add hipified SkipLayerNorm code for ROCmEP
#12107
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
22
Changes
View On
GitHub
Add hipified SkipLayerNorm code for ROCmEP
#12107
zhangyaobit
merged 22 commits into
microsoft:master
from
ROCm:hubertlu/hipify_skiplayernorm
First attempt for half2 vectorized memory access in SkipLayerNorm
0557af6d
Add some functions for debugging
27d0c313
Clean up the code
90e6c2f2
Clean up the code
71bd2ec2
Generalize the vectorized kernels with aligned_vector and remove cuda…
2ff0aafd
Add a unit test for a larger input size
3a262afe
Fix some Lint C++ warnings
be39627a
Use ILP = 4 for the vectorized kernels
9d0d5525
Rewrite the vectorized kernel and templatize ComputeSkipLayerNorm
116cc742
Use conditional operator for input_v
3f63b805
Refactor LaunchSkipLayerNormKernel and replace the original SkipLayer…
3d50998c
Clean some comments and rename the layernorm function
7d242cad
Use ComputeSkipLayerNorm to replace LaunchSkipLayerNormKernel
1977b638
Resolve a Lint C++ warning
36c7c500
Merge remote-tracking branch 'upstream/master' into hubertlu/skiplaye…
f424cb3d
Fix SkipLayerNormBatch1_Float16_vec output data
7900b316
Add hipified code of bert SkipLayerNorm for ROCmEP
9f34a91c
Resolve some Lint C++ warnings
08782876
Merge remote-tracking branch 'upstream/master' into hubertlu/hipify_s…
4a4ba2db
Resolve some Lint C++ warnings
8b278438
Resolve some Lint C++ warnings
41e86634
Resolve Python formatting issue
7f123f55
zhangyaobit
requested a review
from
zhangyaobit
3 years ago
zhangyaobit
approved these changes on 2022-07-07
zhangyaobit
merged
dbcf54aa
into master
3 years ago
Login to write a write a comment.
Login via GitHub
Reviewers
zhangyaobit
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub