Skip to content

Conversation

@Jiawei-Shao
Copy link
Contributor

Description

This patch removes all the logic that handles bias in MatMul::ComputeInternal().

Motivation and Context

As MatMul operator doesn't support bias as its input, we don't need to handle bias in MatMul::ComputeInternal().

As `MatMul` operator doesn't support `bias` as its input, we don't
need to handle `bias` in `MatMul::ComputeInternal`.
return intel::ApplyMatMulIntel(context, Activation(), inputs, output_tensor);
}

return ComputeMatMul(&context, Activation(), inputs, output_tensor, false);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe it should be return ComputeMatMul(&context, Activation(), inputs, output_tensor, true); instead of return ComputeMatMul(&context, Activation(), inputs, output_tensor, false);. The previous one will prevent you go to the splitK path.

Please consider directly return ComputeMatMul(&context, Activation(), inputs, output_tensor, true) in line 118 and move all necessary optimizations in ComputeMatMul to reduce duplicated code.

@daijh
Copy link
Contributor

daijh commented Jan 30, 2026

Is this bias reserved for fusing MatMul + Add sequence.

@Jiawei-Shao Jiawei-Shao marked this pull request as draft February 2, 2026 08:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants