The reason for this implementation is to increase the performance of the QNICE CPU, and to use techniques from formal verification to prove its correctness.
This version of the QNICE CPU (from the QNICE-FPGA project) is not a drop-in replacement, for the following three reasons:
- This design uses the Wishbone memory bus.
- This design uses separate instruction and data interfaces.
- This design expects (at least) one clock cycle delay when reading from instruction and/or data memory.
However, it should be a simple operation to modify the QNICE-FPGA project to support this implementation.
The overall idea of this implementation is to convert each instruction into a sequence of micro-operations, such as:
- Read from memory to source operand buffer
- Read from memory to destination operand buffer
- Write to memory
- Write to register
The reason is that e.g. the instruction ADD @R0, @R1 performs two memory
reads (from @R0 and @R1) and one memory write (to @R1). Since only one
memory operation is possible in each clock cycle, such an instruction will
need to be serialized and will take a total of three clock cycles.
Please go to the doc directory for more in-depth descrition of the architecture and the design.
The current makefile supports the following targets:
make sim: Run simulationmake system.bit: Run synthesis using Vivadomake synth: Run synthesis using yosysmake formal: Run formal verificationmake clean: Remove all generated files
The simulation and synthesis options assemble and run the test program in
test/prog.asm.