18487918. PERFORMING MATRIX MULTIPLICATION IN A STREAMING PROCESSOR simplified abstract (QUALCOMM Incorporated)

PERFORMING MATRIX MULTIPLICATION IN A STREAMING PROCESSOR

Organization Name

QUALCOMM Incorporated

Inventor(s)

Yun Du of San Diego CA (US)

Gang Zhong of San Diego CA (US)

Fei Wei of San Diego CA (US)

Yibin Zhang of San Diego CA (US)

Jing Han of San Jose CA (US)

Hongjiang Shang of San Diego CA (US)

Elina Kamenetskaya of Belmont MA (US)

Minjie Huang of San Diego CA (US)

Alexei Vladimirovich Bourd of San Diego CA (US)

Chun Yu of Rancho Santa Fe CA (US)

Andrew Evan Gruber of Arlington MA (US)

Eric Demers of San Diego CA (US)

PERFORMING MATRIX MULTIPLICATION IN A STREAMING PROCESSOR - A simplified explanation of the abstract

This abstract first appeared for US patent application 18487918 titled 'PERFORMING MATRIX MULTIPLICATION IN A STREAMING PROCESSOR

Simplified Explanation

The present disclosure relates to methods and apparatus for compute processing, specifically improving the performance of matrix multiplication in a streaming processor.

• The disclosed techniques involve executing load instructions to load input data and weight data from memory to a second memory using a load control unit.
• The techniques also involve performing a matrix multiplication operation using the loaded input data and weight data to generate an output matrix using an ALU component.
• The output matrix is then stored at a general-purpose register accessible to the ALU component.

Potential applications of this technology:

• This technology can be applied in various fields that require matrix multiplication operations, such as machine learning, image processing, and scientific simulations.
• It can be used in high-performance computing systems to accelerate matrix-based computations and improve overall system performance.

Problems solved by this technology:

• Matrix multiplication is a computationally intensive operation, and improving its performance can significantly enhance the efficiency of various applications that rely on it.
• By optimizing the process of loading input and weight data and performing the matrix multiplication operation, this technology addresses the challenge of efficiently processing large matrices in real-time.

Benefits of this technology:

• Improved performance: The disclosed techniques enable faster and more efficient matrix multiplication, leading to overall improved performance in applications that rely on this operation.
• Enhanced scalability: The methods and apparatus described in the patent application can be implemented in various computing systems, allowing for scalable and adaptable solutions.
• Reduced memory access: By utilizing a second memory and general-purpose registers, this technology minimizes the need for frequent memory access, which can significantly improve processing speed.

Original Abstract Submitted

The present disclosure relates to methods and apparatus for compute processing. For example, disclosed techniques facilitate improving performance of matrix multiplication in streaming processor. Aspects of the present disclosure can execute, with a load control unit, a first load instruction to load a set of input data of an input matrix from a first memory to a second memory. Aspects of the present disclosure can also execute, with the load control unit, a second load instruction to load a set of weight data of a weight matrix from the first memory to the second memory. Additionally, aspects of the present disclosure can perform, with an ALU component, a matrix multiplication operation using the set of input data and the set of weight data to generate an output matrix. Further, aspects of the present disclosure can store the output matrix at a general purpose register accessible to the ALU component.