This project consists of a minimal CPU called (SDVU) that is able to process binaries generated by the compiler (SDVC). Its behavior is reproduced by a simulator (SDVS) written in Python. All the binaries are generated from compiling a simplification of the DiVinE description language into an single-static-assignment form (Simple DiVinE or SDVE). The BEEM benchmark used to characterize DVE models is translated to SDVE to provide samples to test the whole ecosystem (SDVE BEEM).
The different memories of the project are filled and generated from a given file. They can be regenerated using the gen_memories script. This Python script uses Jinja2 as its templating engine. The templates can be found in templates/ and the generation can be run with :
$ python templates/gen_memories.py The default values for memories are adding.6.cfg and adding.6.out for the CONFIG_MEMORY and PROGRAM_MEMORY respectively.
The sim.sh script utility analyses and run the corresponding testbench for each one of the entity. Running the script for the ALU can be done with:
$ ./sim.sh alu Running the script for sdvu or sdvu_top requires the analysis of the different entities used.
The SDVU project is a CPU composed of several elements:
Config Memory: Memory holding the values of the different global variables (see SDVC for more info on the ISA and global variables).
Program Memory: Memory holding the compiled program, i.e. 32-bits instructions.
SDVU: Processing unit consisting of several sub-parts.
- Control Unit: Finite State Machine that controls the state of the program and enables the entities corresponding to the current state.
- Program Counter: Address of the next instruction in the program memory.
- Decoder: Transforms a raw 32-bits instruction to the different elements (register selectors, immediate values, op code, etc.).
- Arithmetic Logic Unit: Processor of binary operations (without DIV/MOD).
- Registers: 16 registers to hold the result of the ALU and handles LOAD and STORE operations (see SDVC for more info on the ISA and register usage).
The project is synthesized with Vivado 2019.2 (tested with Vivado 2018.3)