FPGA-GPU

Check out the YouTube video!

6.205 Final Project by Laker and Hanfei

Steps to program the GPU

1. Write an assembly-style program (e.g., matrix-mult.txt)
2. Convert into machine code using python isa.py <program_name.txt>
3. Update the program file in BRAM initialization in hdl/controller.sv
4. Update the instruction count parameter when initializating controller.sv
5. Run testbench sim/top_level_tb.sv or real deal hdl/top_level.sv

Steps to add an ISA command

1. Update the comment in the enum for ISA in controller.sv
2. Possibly add a case statement in controller.sv, or update memory_valid_for_memory_out if it's a memory command
3. Possibly add a case statement in memory.sv if it's a memory command
4. Update isa.py to conform to the new arguments you use (return as below, and add to str_to_command dictionary)

return (op_code << 28) + (reg_a << 24) + (immediate << 8) + (reg_b << 4) + (reg_c << 0)

5. Run isa.py data/matrix-mult.txt, then iverilog -g2012 -o sim.out sim/top_level_tb.sv hdl/..., then vvp sim.out, then open dump.vcd in GTKWave and verify that the output of the matrix multiplication remains correct (\BRAM[5][95:0] in GTKWave takes on value 0x00000000001600320013002B)