[Snippets][CPU] Used Plugin Reorder for weights repacking #31024
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
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
3 times, most recently
from
86853be to
1bdd058
Compare
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
from
1bdd058 to
2da7160
Compare
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
2 times, most recently
from
569bebe to
2a7d9ed
Compare
aobolensk
reviewed
Jun 20, 2025
src/plugins/intel_cpu/src/transformations/snippets/x64/op/brgemm_utils.cpp
Show resolved
Hide resolved
...c/transformations/snippets/x64/pass/lowered/expressions/brgemm_copy_b_buffer_expressions.cpp
Show resolved
Hide resolved
v-Golubev
reviewed
Jun 23, 2025
src/plugins/intel_cpu/src/transformations/snippets/x64/pass/repack_matmul_weights.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_cpu/src/transformations/snippets/x64/pass/repack_matmul_weights.cpp
Show resolved
Hide resolved
...gins/intel_cpu/src/transformations/snippets/x64/pass/lowered/external_repacking_adjuster.cpp
Show resolved
Hide resolved
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
from
e2c1dc4 to
2d339c4
Compare
a-sidorova
commented
Jun 23, 2025
There was a problem hiding this comment.
@EgorDuplensky may I ask you to review dnnl_utils part which was updated after our offline discussion? Thank you in advance!
v-Golubev
reviewed
Jun 23, 2025
aobolensk
approved these changes
Jun 23, 2025
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
from
2d339c4 to
105ac58
Compare
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
from
105ac58 to
62afcda
Compare
a-sidorova
force-pushed
the
feature/snippets/use_reorder_for_weights
branch
from
62afcda to
837ab7c
Compare
EgorDuplensky
approved these changes
Jun 26, 2025
riverlijunjie
pushed a commit
to riverlijunjie/openvino-1
that referenced
this pull request
Jul 3, 2025
…olkit#31024) ### Details: - *Renamed `RepackedInput` to `InputRepacker` since this structure doesn't contain repacked data - it contains data and information for repacking* - *Fixed the pass `FuseTransposeBrgemm` - previously this imp calls matcher pass in callback which stores matched `Transpose` ops. It means that after `Transpose` elimination (fusing into `Brgemm`), this matcher pass prolongated life time of `Transpose` - the parent of `Transpose` still had this op in consumers* - *Implemented the data flow pass `RepackMatMulWeights` which calls `Reorder::reorderData` to align we plugin CPU dnnl impl. Locally launched tests with following settings: default (as it is), using blocked weights for all precision, using blocked shapes for second input of MatMuls if it is not constant (in MHA for example). All tests are passed.* ### Tickets: - *169331*
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.
Details:
RepackedInputtoInputRepackersince this structure doesn't contain repacked data - it contains data and information for repackingFuseTransposeBrgemm- previously this imp calls matcher pass in callback which stores matchedTransposeops. It means that afterTransposeelimination (fusing intoBrgemm), this matcher pass prolongated life time ofTranspose- the parent ofTransposestill had this op in consumersRepackMatMulWeightswhich callsReorder::reorderDatato align we plugin CPU dnnl impl. Locally launched tests with following settings: default (as it is), using blocked weights for all precision, using blocked shapes for second input of MatMuls if it is not constant (in MHA for example). All tests are passed.Tickets: