[Profile] Split record event into Global and Local for more accurate profile #62722
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
你的PR提交成功，感谢你对开源项目的贡献! |
plit record event into Global and Local for more accurate profile
HydrogenSulfate
force-pushed
the
fix_event_range
branch
2 times, most recently
from
3d481a4 to
c05787e
Compare
HydrogenSulfate
force-pushed
the
fix_event_range
branch
from
c05787e to
e6e2fb6
Compare
luotao1
approved these changes
Mar 15, 2024
zyfncg
approved these changes
Mar 15, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
Function optimization
PR changes
Others
Description
Pcard-75624
Fix the ambiguity of node execution event, the first segment is for calling GradNodeFunction, should be recognized as
XXXGradNodecomputation cost, the second segment is for potential gradient accumulation in backward queue, should not be count intoXXXGradNode, or will misleading users who profile paddle program with nsight or other visualization software.So this PR use
Local_XXXGradNodeto represent execution time ofXXXGradNodefunction,Global_XXXGradNodeto represent execution time ofLocal_XXXGradNodeplus potential gradient accumulation. Thus,Global_XXXGradNodeshould always be larger thanLocal_XXXGradNodeandLocal_XXXGradNodeis more significant for profiling.To achieve this target, this PR modifies
eager_gen.pyand several manuallyXXXnode.cc, and move event creation next to node execution for ignoring node(s) skipped in backward node queue(i.e.).
before:
after:
As is dipicted below,
Global_MultiplyGradNodeinclude 2 parts:Local_MultiplyGradNodeexecution(grad_output_tensors = (*node)( node_input_buffer->Buffers(), create_graph, is_general_grad);) and gradient accumulation(node_input_buffers_dict[next_node]->add(edge_rank.first, edge_rank.second, grad_output_tensor, create_graph);)