[CPU][ARM]Non i32 conv bias support by alvoron · Pull Request #33072 · openvinotoolkit/openvino

alvoron · 2025-11-28T21:28:34Z

Details:

The PR enables Convolution non-i32 bias support. Before this PR only i32 Convolution bias was supported (due to ACL limitations).

ConvertConvolutionBias has been introduced. This transformation detects specific quantized Convolution patterns followed by Multiply and Add and inserts a Convert to i32 between the constant bias and the Add node.
AddTransformation is called for non-convolution bias only on ARM. Convolution bias is handled by ConvertConvolutionBias transformation on ARM.
The order of applying scales and shifts has been changed on ARM: bias, scale, fq on ARM vs scale, bias, fq on x86. It's needed to get specific postops order.
ConvertConvolutionBias transformation tests have been added to test the transformation.

Tickets:

CVS-180491

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_bias.cpp

src/plugins/intel_cpu/src/graph_optimizer.cpp

src/plugins/intel_cpu/tests/functional/custom/subgraph_tests/src/arm/conv_fq.cpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_bias.cpp

src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp

src/plugins/intel_cpu/tests/functional/custom/subgraph_tests/src/arm/conv_fq.cpp

src/plugins/intel_cpu/tests/unit/transformations/arm/convert_conv_bias.cpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_bias.cpp

src/plugins/intel_cpu/tests/unit/transformations/arm/convert_conv_bias.cpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_bias.cpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_dq_scales.hpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/conv_mul_add_fq_block.cpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/conv_mul_add_fq_block.hpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_dq_scales.cpp

src/plugins/intel_cpu/src/transformations/cpu_opset/arm/pass/convert_conv_dq_scales.hpp

Currently, copyright header consistency is checked manually during code review (e.g. [1](#32393 (comment)), [2](#33072 (comment)), [3](#32983 (comment)), [4](#31191 (comment))). The idea of this PR is to automate this process and save reviewers' time. ### Details: - *Introduced GHA workflow that validates copyright headers in C++ and Python files on PRs. If issues are found, it generates a patch file and fails the check with clear instructions.* - *Added some changes to the files which have copyright inconsistency in order to verify the added scripts* ### Tickets: - *N\A*

Currently, copyright header consistency is checked manually during code review (e.g. [1](openvinotoolkit#32393 (comment)), [2](openvinotoolkit#33072 (comment)), [3](openvinotoolkit#32983 (comment)), [4](openvinotoolkit#31191 (comment))). The idea of this PR is to automate this process and save reviewers' time. ### Details: - *Introduced GHA workflow that validates copyright headers in C++ and Python files on PRs. If issues are found, it generates a patch file and fails the check with clear instructions.* - *Added some changes to the files which have copyright inconsistency in order to verify the added scripts* ### Tickets: - *N\A*

### Details: The PR enables Convolution non-i32 bias support. Before this PR only i32 Convolution bias was supported (due to ACL limitations). - `ConvertConvolutionBias` has been introduced. This transformation detects specific quantized Convolution patterns followed by Multiply and Add and inserts a Convert to i32 between the constant bias and the Add node. - `AddTransformation` is called for non-convolution bias only on ARM. Convolution bias is handled by `ConvertConvolutionBias transformation` on ARM. - The order of applying scales and shifts has been changed on ARM: `bias, scale, fq` on ARM vs `scale, bias, fq` on x86. It's needed to get specific postops order. - `ConvertConvolutionBias` transformation tests have been added to test the transformation. ### Tickets: - CVS-180491 --------- Co-authored-by: Vladislav Golubev <vladislav.golubev@intel.com>

Currently, copyright header consistency is checked manually during code review (e.g. [1](openvinotoolkit#32393 (comment)), [2](openvinotoolkit#33072 (comment)), [3](openvinotoolkit#32983 (comment)), [4](openvinotoolkit#31191 (comment))). The idea of this PR is to automate this process and save reviewers' time. ### Details: - *Introduced GHA workflow that validates copyright headers in C++ and Python files on PRs. If issues are found, it generates a patch file and fails the check with clear instructions.* - *Added some changes to the files which have copyright inconsistency in order to verify the added scripts* ### Tickets: - *N\A*

### Details: The PR enables Convolution non-i32 bias support. Before this PR only i32 Convolution bias was supported (due to ACL limitations). - `ConvertConvolutionBias` has been introduced. This transformation detects specific quantized Convolution patterns followed by Multiply and Add and inserts a Convert to i32 between the constant bias and the Add node. - `AddTransformation` is called for non-convolution bias only on ARM. Convolution bias is handled by `ConvertConvolutionBias transformation` on ARM. - The order of applying scales and shifts has been changed on ARM: `bias, scale, fq` on ARM vs `scale, bias, fq` on x86. It's needed to get specific postops order. - `ConvertConvolutionBias` transformation tests have been added to test the transformation. ### Tickets: - CVS-180491 --------- Co-authored-by: Vladislav Golubev <vladislav.golubev@intel.com>

Currently, copyright header consistency is checked manually during code review (e.g. [1](openvinotoolkit#32393 (comment)), [2](openvinotoolkit#33072 (comment)), [3](openvinotoolkit#32983 (comment)), [4](openvinotoolkit#31191 (comment))). The idea of this PR is to automate this process and save reviewers' time. ### Details: - *Introduced GHA workflow that validates copyright headers in C++ and Python files on PRs. If issues are found, it generates a patch file and fails the check with clear instructions.* - *Added some changes to the files which have copyright inconsistency in order to verify the added scripts* ### Tickets: - *N\A*

### Details: The PR enables Convolution non-i32 bias support. Before this PR only i32 Convolution bias was supported (due to ACL limitations). - `ConvertConvolutionBias` has been introduced. This transformation detects specific quantized Convolution patterns followed by Multiply and Add and inserts a Convert to i32 between the constant bias and the Add node. - `AddTransformation` is called for non-convolution bias only on ARM. Convolution bias is handled by `ConvertConvolutionBias transformation` on ARM. - The order of applying scales and shifts has been changed on ARM: `bias, scale, fq` on ARM vs `scale, bias, fq` on x86. It's needed to get specific postops order. - `ConvertConvolutionBias` transformation tests have been added to test the transformation. ### Tickets: - CVS-180491 --------- Co-authored-by: Vladislav Golubev <vladislav.golubev@intel.com>

alvoron added 8 commits November 26, 2025 14:38

init

4e58c7a

address comments

06a154d

fixed comments

feabefb

fix clang-tidy and arm32 tests

3c43a84

fix clang-tidy

3ed4e83

fix clang

ba07c2d

added comment

08ddf4e

init

b2b7d94

github-actions bot added the category: CPU OpenVINO CPU plugin label Nov 28, 2025

subgraph test handles non-i32 bias case

1a5ca1d

github-actions bot added the category: LP transformations OpenVINO Low Precision transformations label Dec 8, 2025

register ConvertConvolutionBias separately

f8c7eb1

github-actions bot removed the category: LP transformations OpenVINO Low Precision transformations label Dec 8, 2025

alvoron marked this pull request as ready for review December 8, 2025 19:44

alvoron requested review from a team as code owners December 8, 2025 19:44

Register ConvertConvolutionBias on ARM only + cleanup

4e02b17

alvoron assigned v-Golubev Dec 8, 2025

v-Golubev reviewed Dec 9, 2025

View reviewed changes

alvoron added 2 commits December 9, 2025 18:34

addressed comments

ab71e7b

fix comparison of different signedness

680df8a

v-Golubev reviewed Dec 10, 2025

View reviewed changes

alvoron added 3 commits December 10, 2025 17:12

addressed comments

ab75009

trigger AddTransformation for non-convolution bias

1a7826a

fix tests

73ca02a

v-Golubev reviewed Dec 12, 2025

View reviewed changes

alvoron added 3 commits December 12, 2025 11:23

addressed comments

b20fe80

Merge branch 'master' into non_i32_conv_bias_support

3c75699

reverted legacy logic in ConvAndFQ

6690abb

alvoron added the platform: arm OpenVINO on ARM / ARM64 label Dec 19, 2025

fix risc-v

9c873b8

v-Golubev reviewed Feb 11, 2026

View reviewed changes

v-Golubev mentioned this pull request Feb 11, 2026

[CPU] Remove unused transformations utils #34075

Closed

alvoron added 4 commits February 11, 2026 20:14

addressed comments

2e93b12

fix clang-tidy

1561de2

fix clang-tidy

d717a20

temporary disabled ConvolutionTransformation

59b6639

v-Golubev approved these changes Feb 12, 2026

View reviewed changes

v-Golubev removed the do not merge label Feb 12, 2026

v-Golubev added this to the 2026.1 milestone Feb 12, 2026

alvoron added 3 commits February 12, 2026 14:48

fix clang-tidy

279e46b

disable smoke_ConvAndFQ_CPU tests

18b960d

Merge branch 'master' into non_i32_conv_bias_support

7f6d7eb

alvoron enabled auto-merge February 12, 2026 14:38

Add SKIP_IF_CURRENT_TEST_IS_DISABLED to int8 test

ae12a3f

alvoron added the category: LP transformations OpenVINO Low Precision transformations label Feb 12, 2026

fix Copyright year

a96d292

github-actions bot removed the category: LP transformations OpenVINO Low Precision transformations label Feb 12, 2026

Merge branch 'master' into non_i32_conv_bias_support

ed5f9c9

alvoron added this pull request to the merge queue Feb 13, 2026

Merged via the queue into openvinotoolkit:master with commit b88f5e7 Feb 13, 2026
239 of 241 checks passed

alvoron deleted the non_i32_conv_bias_support branch February 13, 2026 12:58

Conversation

alvoron commented Nov 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Details:

Tickets:

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

alvoron commented Nov 28, 2025 •

edited

Loading