Matrix Accelerator Generator designed in CHISEL HDL for efficient General Matrix Multiply (GeMM) operations. The architecture employed is the innovative SIGMA Architecture, which enhances performance and scalability in matrix multiplication tasks.
CHISEL HDL: Developed using the CHISEL hardware description language, ensuring flexibility and control in hardware design.
GeMM Optimization: Tailored for General Matrix Multiply (GeMM) operations, a fundamental computational kernel in many numerical and machine learning applications.
SIGMA Architecture: Leveraging the SIGMA Architecture, known for its efficiency and scalability, to accelerate matrix multiplication tasks.
Performance: Designed with a focus on performance improvements, allowing for faster and more efficient matrix operations.
Clone the Repository:
git clone https://github.com/merledu/magma-si.git
Access the Project Directory:
cd magma-si
Compile the project using SBT (Scala Build Tool) with the following command:
sbt
Execute the tests with VCD output by issuing the following command:
testOnly magmasi.components.flexdpecom2test -- -DwriteVcd=1
Test case that has been executed in this manner:
1. // for input data base 1
c.io.i_data_bus(0).poke(1.U)
2. // for input data bus 2
c.io.i_data_bus2(0).poke(0.U)
3. // for i vn
c.io.i_vn(3).poke("b11101".U)
4. // for muxes
c.io.i_mux_bus(576).poke(1.B)
5. // stationary and valid
c.io.i_stationary.poke(1.B)
c.io.i_data_valid.poke(1.B)
Desire Output:
When you provide the correct input values for i_vn and muxes pins, you will receive the expected matrix output.
In this document, we illustrate the matrix multiplication process and conduct a comparative analysis between the results obtained from traditional matrix multiplication and the flexDpe multiplication which is use in magma-si .matrix multiplication