aztecher / hcpu-rv32i

hcpu-rv32i is a RISC-V RV32I compliant software CPU implementation using Haskell + Clash

hcpu-rv32i

hcpu-rv32i is a RISC-V RV32I compliant software CPU implementation using Haskell + Clash

Supported/Unsupported Instruction Sets

hcpu-rv32 now support some basic instruction sets of RV32I.
However, some instruction sets, s.t. FENCE, FENCEI, ... are not implemented yet.

Supported instruction sets

• R-type Format
  • ADD
  • SUB
  • AND
  • OR
  • XOR
  • SLL
  • SRA
  • SRL
  • SLT
  • SLTU
• I-type Format
  • ADDI
  • SLTI
  • SLTIU
  • XORI
  • ORI
  • ANDI
  • SLLI
  • SRAI
  • SRLI
  • JALR
  • LB
  • LH
  • LW
  • LBU
  • LHU
• S-type Format
  • SB
  • SH
  • SW
• B-type Format
  • BEQ
  • BNE
  • BLT
  • BGE
  • BLTU
  • BGEU
• U-type Format
  • LUI
  • AUIPC
• J-type Format
  • JAL

Unsupported instruction sets (2021.10.03)

• I-type Format
  • FENCE
  • FENCEI
  • EBREAK
  • ECALL
  • CSRRC
  • CSRRS
  • CSRRW
  • CSRRCI
  • CSRRSI
  • CSRRWI

TODO

• Add Test
• Refactor Codes
• Consider relation between Vec n Instruction and address space
• Create fibonacci function and check it's execution
• Compile to Verilog-HDL
• Check working on compiled Verilog-HDL
• Check working on FPGA

Usage

Now, you have to hardcode instruction that you want to execute.
Please check some example in src/RV32I/Programs/Example.hs and test codes.

For example, If you want to run addImm program in Example.hs, you can execute it by GHCi like bellow.

λ > Prelude.take 10 $ sample (cpu initCPUState (programmedRAM addImm) :: Signal System Registers)
...

or, you can use runCPU' helper function.

λ > runCPU' addImm 0 10

sample function generates infinite list of result of CPU execution, and addImm code will loop by jumping same addresss, so you take some length from this.

You can see the 'fibonacci program' in src/RV32I/Programs/Fib.hs as a more complex example.

Clash

In this repository, we use Clash version 1.4.5 and can check its version as follows

stack run -- clash --version
Clash, version 1.4.5 (using clash-lib, version: 1.4.5)

Known Issue

Take too much time to compile this program. In my local computer, total compilation time is about 9h!

cabal run clash -- RV32I.Top --verilog
Up to date
Loaded package environment from /Users/mikiyaf/Documents/haskell/Clash/hcpu-rv32i/.ghc.environment.x86_64-darwin-8.10.7
GHC: Parsing and optimising modules took: 8.366s
GHC: Loading external modules from interface files took: 0.008s
GHC: Parsing annotations took: 0.010s
Clash: Parsing and compiling primitives took 0.176s
GHC+Clash: Loading modules cumulatively took 13.695s
Clash: Compiling RV32I.Top.topEntity
Clash: Normalization took 8h57m46.924s
Clash: Netlist generation took 2m0.349s
Clash: Total compilation took 9h0m3.538s

and generaed code is total 70000+ LoC verilog file.