[CPU] Refactor mamtul node

[EgorDuplensky]

No description provided.

[Copilot]

Pull Request Overview

This PR refactors the MatMul node implementation in Intel CPU plugin to align with a new executor framework. The refactoring modernizes the code architecture by replacing the legacy implementation with a factory-based executor pattern that supports multiple backend implementations.

Key changes include:

• Refactored MatMul node to use new executor factory pattern
• Simplified header dependencies and removed legacy code
• Updated MatMul small executor to align with new interface
• Added comprehensive MatMul implementations with support selection

Reviewed Changes

Copilot reviewed 17 out of 18 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
src/plugins/intel_cpu/src/nodes/matmul.h	Refactored header to use executor factory pattern, removed legacy members
src/plugins/intel_cpu/src/nodes/matmul.cpp	Complete rewrite using executor factory, simplified initialization and execution
src/plugins/intel_cpu/src/nodes/executors/x64/matmul_small.hpp	Updated interface to inherit from new Executor base class
src/plugins/intel_cpu/src/nodes/executors/x64/matmul_small.cpp	Refactored implementation to work with new executor framework
src/plugins/intel_cpu/src/nodes/executors/variable_executor.hpp	Added precreate parameter for optional executor initialization
src/plugins/intel_cpu/src/nodes/executors/matmul_implementations.cpp	New file defining available MatMul executor implementations
src/plugins/intel_cpu/src/nodes/executors/matmul_config.hpp	Extended MatMul attributes structure with additional configuration
src/plugins/intel_cpu/src/nodes/executors/implementations.hpp	Added MatMul to executor implementations registry
Multiple DNNL files	Updated to support MatMul attributes and maintain compatibility with FC attributes

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[EgorDuplensky]

@maxnick Could you please take a look

[maxnick]

LGTM

[maxnick]

Nit. If we move dynamicQuantizationGroupSize behind bool declarations, we can reduce the size of MatMulAttrs struct by 8 bytes, due to data alignment:

struct MatMulAttrs {
    bool transposeA = false;
    bool transposeB = false;
    bool withBias = false;
    bool weightsNonTransposed = false;
    bool sparseWeights = false;

    bool constantWeights = false;
    bool fcSemantic = false;

    uint64_t dynamicQuantizationGroupSize = 0;

    // DQ scales for quantization
    std::vector<float> dqScales;

    // Post-operations for fused operations
    PostOps postOps;
};

[EgorDuplensky]

Applied