DLX RISC uProcessor

Authors:

• Alfredo Paolino
• Vincenzo Petrolo
• Diamante Simone Crescenzo

Outline

• DLX RISC uProcessor
  • Outline
  • Datapath
  • Instruction Memory
  • Data memory
  • Control Unit
  • Simulation – I-type tests
  • Simulation – R-type tests
  • Simulation – Iterative Division
  • Synthesis
  • Physical layout
  • Technical report
  • Possible improvements

Datapath

• ALU : It is described, using a mixed behavioral-structural description in order to use the Pentium-4 adder in case of additions or subtractions.
• Register File : It is composed of 32 registers and has 2 read ports and 1 write port.
• Adders : Ripple carry adders used for computing next Program Counter.
• Multiplexers : Are used to choose between input sources in the Register File, Program Counter and ALU.
• Pipeline registers : Are used to store the results from previous stage.

Instruction Memory

• Size : 2 kB
• Word : 32 bits
• It contains the firmware that is loaded into the microprocessor.
• Asynchronous memory.

Data memory

• Size : 2 kB
• Word size: 32 bits
• It stores data coming from registers.
• Asynchronous memory.

Control Unit

• LUT size : 62 lines
• Control Word: 9 bits
• Hardwired implementation
• Modular control word generation through std_logic_vector concatenation to improve readability during debugging phase.

Simulation – I-type tests

• In order to verify that the instruction set for I- type instructions works, we wrote a simple assembly program that performs:
  • Additions;
  • Subtractions;
  • Mask operations through logical instructions (i.e. AND/OR/XOR etc..).
  • Eventually the program halts.

Simulation – R-type tests

• In order to verify that the instruction set for R- type instructions works, we wrote a simple assembly program that performs:
  • Additions;
  • Subtractions;
  • Shift operations.
  • Eventually the program halts.

Simulation – Iterative Division

• In conclusion, to create a more complete program we went through the iterative division simulation.
• The algorithm is executed after a call to a procedure (JAL) and performs 81/27.
• When the procedure ends it stores the result into memory and load again into another register. Then it returns to the caller.
• Eventually the program halts.

Synthesis

• Starting from a clock frequency of 50 MHz we reached up to 1 GHz, without major synthesis optimizations.
• Eventually 2 GHz clock frequency goal was achieved using aggressive optimization flow with the usage of compile_ultra, set_dont_touch to avoid removal of skewing registers and set_max_delay to constrain the quasi-critical paths.

Physical layout

• We followed the flow for the place & route phase.
• After the post-routing optimization phase, we run a static power analysis and got an estimated power of 500 mW running at 1 GHz.
• The total power estimated from PrimeTime (1 GHz) was: 426 mW

Technical report

• The final step was to produce a report describing more in detail what is shown in this presentation.

Possible improvements

For a future version, some of the possible improvements:

• Data Hazard Unit
• Branch Prediction Unit
• Floating Point Unit
• Instruction/Data cache
• Extended Instruction set
• Reducing to 1 the Branch delay slot
• Verification using UVM