[QNN-EP] Einsum equation ReduceSum Multiply on broadcast X

[qti-yuduo]

[QNN-EP] Einsum equation ReduceSum Multiply on broadcast X

[HectorSVC]

/azp run Linux QNN CI Pipeline, Win_TRT_Minimal_CUDA_Test_CI, Windows ARM64 QNN CI Pipeline, Windows x64 QNN CI Pipeline, Windows GPU Doc Gen CI Pipeline

[azure-pipelines]

Azure Pipelines successfully started running 5 pipeline(s).

[Copilot]

Pull Request Overview

This PR adds support for a new Einsum equation pattern "ReduceSum Multiply on broadcast X" (e.g., "bhwc,wkc->bhwk") to the QNN provider. The implementation handles cases where the first input needs broadcasting and both inputs are multiplied before being reduced along the contraction axis.

• Adds pattern recognition for the new ReduceSum Multiply broadcast equation type
• Implements the equation by reshaping, multiplying with broadcasting, and reducing along the contraction axis
• Includes comprehensive test coverage across CPU, HTP, and GPU backends with appropriate disabling for known issues

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
onnxruntime/test/providers/qnn/einsum_op_test.cc	Adds test cases for the new equation pattern across different backends, with some disabled due to QAIRT limitations
onnxruntime/core/providers/qnn/builder/opbuilder/einsum_op_builder.cc	Implements pattern detection and graph construction for ReduceSum Multiply broadcast operations

Comments suppressed due to low confidence (1)

onnxruntime/test/providers/qnn/einsum_op_test.cc:192

• The test name 'EinsumRank4MatMulTransposeAll2' is misleading as this test was moved and replaced with a different equation pattern. The name should reflect that this is testing the ReduceSum Multiply broadcast functionality, not MatMul transpose.

TEST_F(QnnCPUBackendTests, EinsumRank4MatMulTransposeAll2) {