This Java program simulates some functionalities of complex processor architectures. The project was part of the course Advanced Computer Architecture at the University of Bristol in the fall of 2016. In order to understand how modern processors achieve low CPI (cycles per instruction), I designed a whole new computer architecture from scratch. It speculatively processes MIPS-like instructions in parallel pipelines.
This final version implements 2-way superscalar, out-of-order execution of instructions with a dynamic branch prediction scheme. A GUI allows to step through the execution of a program cycle-by-cycle to inspect the contents of the registers, pipeline stages, etc.
The 5-stage pipelines are configured as follows:
These instructions will get you a copy of the project up and running on your local machine.
To run javac from the command line you'll need to have the Java SDK installed.
Secondly, to compile the java source code, we'll use ant. macOS users can
brew install ant
verify that it works using ant -v. The output should look like this:
Apache Ant(TM) version 1.10.5 compiled on July 10 2018
Trying the default build file: build.xml
Buildfile: build.xml does not exist!
Build failed
Here's how to compile the code.
Move to the directory containing the build.xml file
cd superscalar_processor_OoO
and compile
ant
Verify that the bin directory now contains 3 folders with compiled Java code in .class format, in addition to the programs folder containing the benchmark programs.
Four benchmark programs are included:
- A recursive algorithm that calculates the greatest common divisor (GCD) of two numbers.
- Bubblesort
- Vector addition
- Dotproduct of two 10-dimensional vectors
Note that these MIPS-like programs were compiled by hand. Depending on the processor's configuration, NOP instructions have to be inserted to ensure correct execution.
To start the GUI, simply
cd bin
and, to load the GCD program for example,
java benchmarks.GCD true
The first argument selects the benchmark program. Valid options are GCD, BubbleSort, VectorAddition or DotProduct.
The second argument enables the branch predictor. It can be either true or false.
This will open four windows (initially empty):
- Registers
- Data Memory
- Instruction Memory
- Controls
Click the step button in the Controls window to advance the program by one cycle.
Click execute all to run the program until it's finished.
Each cycle, the simulator prints detailed status info & results to stdout.
Close any window to terminate.
- Julien Schuermans - julienschuermans
- Many thanks to David May & Simon McIntosh-Smith from the University of Bristol for their guidance and feedback.
- During this project I relied heavily on Computer Architecture - A Quantitative Approach, Hennessy & Patterson