[Inductor][Optimus] Fix group fusion stride layout #2442
Conversation
This file contains 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
mengluy0125
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
mengluy0125
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
|
This pull request was exported from Phabricator. Differential Revision: D61888433 |
|
This pull request was exported from Phabricator. Differential Revision: D61888433 |
mengluy0125
added a commit
to mengluy0125/benchmark
that referenced
this pull request
Aug 28, 2024
Summary: Pull Request resolved: pytorch#2442 context: https://fb.workplace.com/groups/1075192433118967/permalink/1401282167176657/ moving the changes to the group gemm op has compilation errors, see details in D55606636 Differential Revision: D61888433
mengluy0125
force-pushed
the
export-D61888433
branch
from
d218f8b to
0bc66e5
Compare
mengluy0125
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
mengluy0125
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
mengluy0125
added a commit
to mengluy0125/pytorch
that referenced
this pull request
Aug 28, 2024
Summary: X-link: pytorch/benchmark#2442 context: https://fb.workplace.com/groups/1075192433118967/permalink/1401282167176657/ moving the changes to the group gemm op has compilation errors, see details in D55606636 Test Plan: # local reproduce ``` CUDA_LAUNCH_BLOCKING=1 buck2 run mode/opt //scripts/jackiexu0313/pt2:local_model_with_pt2 -- --test_mode batch-split-group --model_type "afoc" --flow_id 544109991 ``` Counter({'pattern_matcher_nodes': 1215, 'pattern_matcher_count': 1090, 'normalization_pass': 430, 'remove_split_with_size_one_pass': 416, 'batch_aten_mul': 13, 'scmerge_split_sections_removed': 11, 'scmerge_cat_removed': 5, 'scmerge_cat_added': 4, 'batch_linear_post_grad': 4, 'scmerge_split_removed': 3, 'batch_aten_sub': 2, 'batch_layernorm': 1, 'group_linear': 1}) ``` CUDA_VISIBLE_DEVICES=3 OC_CAUSE=1 buck2 run mode/opt //scripts/jackiexu0313/pt2:local_model_with_pt2 -- --test_mode group-batch-split --model_type "cmf_shrink" --flow_id 587303213 ``` P1551948670 Counter({'pattern_matcher_nodes': 2244, 'pattern_matcher_count': 1738, 'normalization_pass': 404, 'extern_calls': 370, 'benchmarking.TritonBenchmarker.benchmark_gpu': 293, 'remove_split_with_size_one_pass': 269, 'merge_splits_pass': 74, 'normalization_aten_pass': 56, 'batch_aten_mul': 11, 'fxgraph_cache_miss': 10, 'group_linear': 9, 'scmerge_split_sections_removed': 5, 'scmerge_split_removed': 4, 'scmerge_cat_removed': 4, 'unbind_stack_pass': 4, 'batch_sigmoid': 2, 'batch_linear': 2, 'move_reshape_out_of_split_stack_pass': 2, 'batch_aten_sub': 2, 'batch_aten_add': 2, 'batch_layernorm': 1, 'scmerge_split_added': 1, 'scmerge_cat_added': 1, 'split_stack_to_cats_pass': 1, 'split_cat_to_slices_pass': 1, 'benchmarking.TritonBenchmarker.triton_do_bench': 1, 'batch_relu': 1}) # e2e ### AFOC baseline: f545589474 proposal: f545589302 {F1474302182} ### cmf shrink ads_dper3:0e442d2994ad1421377489d53ef99593 training_platform:be4b7015f1582fb1760bd72cf83ff38d baseline f635512197 baseline + group_fusion f635975547 The group fusion can be enabled but has qps regression by using group fusion. Differential Revision: D61888433
Summary: X-link: pytorch/pytorch#134696 Pull Request resolved: pytorch#2442 context: https://fb.workplace.com/groups/1075192433118967/permalink/1401282167176657/ moving the changes to the group gemm op has compilation errors, see details in D55606636 Differential Revision: D61888433
|
This pull request was exported from Phabricator. Differential Revision: D61888433 |
mengluy0125
force-pushed
the
export-D61888433
branch
from
0bc66e5 to
878b2b5
Compare
mengluy0125
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
mengluy0125
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
mengluy0125
added a commit
to mengluy0125/pytorch
that referenced
this pull request
Aug 28, 2024
Summary: Pull Request resolved: pytorch#134696 X-link: pytorch/benchmark#2442 context: https://fb.workplace.com/groups/1075192433118967/permalink/1401282167176657/ moving the changes to the group gemm op has compilation errors, see details in D55606636 Test Plan: # local reproduce ``` CUDA_LAUNCH_BLOCKING=1 buck2 run mode/opt //scripts/jackiexu0313/pt2:local_model_with_pt2 -- --test_mode batch-split-group --model_type "afoc" --flow_id 544109991 ``` Counter({'pattern_matcher_nodes': 1215, 'pattern_matcher_count': 1090, 'normalization_pass': 430, 'remove_split_with_size_one_pass': 416, 'batch_aten_mul': 13, 'scmerge_split_sections_removed': 11, 'scmerge_cat_removed': 5, 'scmerge_cat_added': 4, 'batch_linear_post_grad': 4, 'scmerge_split_removed': 3, 'batch_aten_sub': 2, 'batch_layernorm': 1, 'group_linear': 1}) ``` CUDA_VISIBLE_DEVICES=3 OC_CAUSE=1 buck2 run mode/opt //scripts/jackiexu0313/pt2:local_model_with_pt2 -- --test_mode group-batch-split --model_type "cmf_shrink" --flow_id 587303213 ``` P1551948670 Counter({'pattern_matcher_nodes': 2244, 'pattern_matcher_count': 1738, 'normalization_pass': 404, 'extern_calls': 370, 'benchmarking.TritonBenchmarker.benchmark_gpu': 293, 'remove_split_with_size_one_pass': 269, 'merge_splits_pass': 74, 'normalization_aten_pass': 56, 'batch_aten_mul': 11, 'fxgraph_cache_miss': 10, 'group_linear': 9, 'scmerge_split_sections_removed': 5, 'scmerge_split_removed': 4, 'scmerge_cat_removed': 4, 'unbind_stack_pass': 4, 'batch_sigmoid': 2, 'batch_linear': 2, 'move_reshape_out_of_split_stack_pass': 2, 'batch_aten_sub': 2, 'batch_aten_add': 2, 'batch_layernorm': 1, 'scmerge_split_added': 1, 'scmerge_cat_added': 1, 'split_stack_to_cats_pass': 1, 'split_cat_to_slices_pass': 1, 'benchmarking.TritonBenchmarker.triton_do_bench': 1, 'batch_relu': 1}) # e2e ### AFOC baseline: f545589474 proposal: f545589302 {F1474302182} ### cmf shrink ads_dper3:0e442d2994ad1421377489d53ef99593 training_platform:be4b7015f1582fb1760bd72cf83ff38d baseline f635512197 baseline + group_fusion f635975547 {F1832419326} {F1832419319} {F1832419401} The group fusion can be enabled but has qps regression by using group fusion, will do a dive deep study. Differential Revision: D61888433
mengluy0125
changed the title
Yuzhen11
approved these changes
Aug 28, 2024
|
@xuzhao9 Is this the correct place to change this file? I remember you mentioned this |
|
@FindHao Exporting to pytorch/benchmark is optional. We will keep this PR and when it merges in fbcode or pytorch/pytorch, the changes will automatically be applied to pytorch/benchmark. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary:
context:
https://fb.workplace.com/groups/1075192433118967/permalink/1401282167176657/
moving the changes to the group gemm op has compilation errors, see details in D55606636
Differential Revision: D61888433