riscv-opcodes

This repo enumerates standard RISC-V instruction opcodes and control and status registers. It also contains a script to convert them into several formats (C, Scala, LaTeX).

Artifacts (encoding.h, latex-tables, etc) from this repo are used in other tools and projects like Spike, PK, RISC-V Manual, etc.

Project Structure

├── constants.py    # contains variables, constants and data-structures used in parse.py
├── encoding.h      # the template encoding.h file
├── LICENSE         # license file
├── Makefile        # makefile to generate artifacts
├── parse.py        # python file to perform checks on the instructions and generate artifacts
├── README.md       # this file
├── rv*             # instruction opcode files
└── unratified      # contains unratified instruction opcode files

File Naming Policy

This project follows a very specific file structure to define the instruction encodings. All files containing instruction encodings start with the prefix rv. These files can either be present in the root directory (if the instructions have been ratified) or the unratified directory. The exact file-naming policy and location is as mentioned below:

1. rv_x - contains instructions common within the 32-bit and 64-bit modes of extension X.
2. rv32_x - contains instructions present in rv32x only (absent in rv64x e.g.. brev8)
3. rv64_x - contains instructions present in rv64x only (absent in rv32x, e.g. addw)
4. rv_x_y - contains instructions when both extension X and Y are available/enabled. It is recommended to follow canonical ordering for such file names as specified by the spec.
5. unratified - this directory will also contain files similar to the above policies, but will correspond to instructions which have not yet been ratified.

When an instruction is present in multiple extensions and the spec is vague in defining the extension which owns the instruction, the instruction encoding must be placed in the first canonically ordered extension and should be imported(via the $import keyword) in the remaining extensions.

Encoding Syntax

The encoding syntax uses $ to indicate keywords. As of now 2 keywords have been identified : $import and $pseudo_op (described below). The syntax also uses :: as a means to define the relationship between extension and instruction. .. is used to defined bit ranges. We use # to define comments in the files. All comments must be in a separate line. In-line comments are not supported.

Instruction syntaxes used in this project are broadly categorized into three:

• regular instructions :- these are instructions which hold a unique opcode in the encoding space. A very generic syntax guideline for these instructions is as follows:

  <instruction name> <arguments>
  

  where <argument> is either <bit encoding> or <variable argument>.

  Examples:

  lui     rd imm20 6..2=0x0D 1..0=3
  beq     bimm12hi rs1 rs2 bimm12lo 14..12=0 6..2=0x18 1..0=3
  

  The bit encodings are usually of 2 types:

  • single bit assignment : here the value of a single bit is assigned using syntax <bit-position>=<value>. For e.g. 6=1 means bit 6 should be 1. Here the value must be 1 or 0.
  • range assignment: here a range of bits is assigned a value using syntax: <msb>..<lsb>=<val>. For e.g. 31..24=0xab. The value here can be either unsigned integer, hex (0x) or binary (0b).

• pseudo_instructions (a.k.a pseudo_ops) - These are instructions which are aliases of regular instructions. Their encodings force certain restrictions over the regular instruction. The syntax for such instructions uses the $pseudo_op keyword as follows:

  $pseudo_op <extension>::<base-instruction> <instruction name> <instruction args> <bit-encodings>
  

  Here the <extension> specifies the extension which contains the base instruction. <base-instruction> indicates the name of the instruction this pseudo-instruction is an alias of. The remaining fields are the same as the regular instruction syntax, where all the args and the fields of the pseudo instruction are specified.

  Example:

  $pseudo_op rv_zicsr::csrrs frflags rd 19..15=0 31..20=0x001 14..12=2 6..2=0x1C 1..0=3
  

  If a ratified instruction is a pseudo_op of a regular unratified instruction, it is recommended to maintain this pseudo_op relationship i.e. define the new instruction as a pseudo_op of the unratified regular instruction, as this avoids existence of overlapping opcodes for users who are experimenting with unratified extensions as well.

• imported_instructions - these are instructions which are borrowed from an extension into a new/different extension/sub-extension. Only regular instructions can be imported. Pseudo-op or already imported instructions cannot be imported. Example:

  $import rv32_zkne::aes32esmi
  

RESTRICTIONS

Following are the restrictions one should keep in mind while defining $pseudo_ops and $imported_ops

• Pseudo-op or already imported instructions cannot be imported again in another file. One should always import base-instructions only.
• While defining a $pseudo_op, the base-instruction itself cannot be a $pseudo_op

Flow for parse.py

The parse.py python file is used to perform checks on the current set of instruction encodings and also generates multiple artifacts : latex tables, encoding.h header file, etc. This section will provide a brief overview of the flow within the python file.

To start with, parse.py creates a list of all rv* files currently checked into the repo (including those inside the unratified directory as well). It then starts parsing each file line by line. In the first pass, we only capture regular instructions and ignore the imported or pseudo instructions. For each regular instruction, the following checks are performed :

• for range-assignment syntax, the msb position must be higher than the lsb position
• for range-assignment syntax, the value of the range must representable in the space identified by msb and lsb
• values for the same bit positions should not be defined multiple times.
• All bit positions must be accounted for (either as args or constant value fields)

Once the above checks are passed for a regular instruction, we then create a dictionary for this instruction which contains the following fields:

• encoding : contains a 32-bit string defining the encoding of the instruction. Here - is used to represent instruction argument fields
• extension : string indicating which extension/filename this instruction was picked from
• mask : a 32-bit hex value indicating the bits of the encodings that must be checked for legality of that instruction
• match : a 32-bit hex value indicating the values the encoding must take for the bits which are set as 1 in the mask above
• variable_fields : This is list of args required by the instruction

The above dictionary elements are added to a main instr_dict dictionary under the instruction node. This process continues until all regular instructions have been processed. In the second pass, we now process the $pseudo_op instructions. Here, we first check if the base-instruction of this pseudo instruction exists in the relevant extension/filename or not. If it is present, the the remaining part of the syntax undergoes the same checks as above. Once the checks pass and if the base-instruction is not already added to the main instr_dict then the pseudo-instruction is added to the list. In the third, and final, pass we process the imported instructions.

The case where the base-instruction for a pseudo-instruction may not be present in the main instr_dict after the first pass is if the only a subset of extensions are being processed such that the base-instruction is not included.

Artifact Generation and Usage

The following artifacts can be generated using parse.py:

• instr_dict.json : This is always generated by parse.py and contains the entire main dictionary instr\_dict in JSON format. Note, in this file the dots in an instruction are replaced with underscores. In previous versions of this project the generated file was instr_dict.yaml. Note that JSON is a subset of YAML so the file can still be read by any YAML parser.
• encoding.out.h : this is the header file that is used by tools like spike, pk, etc
• instr-table.tex : the latex table of instructions used in the riscv-unpriv spec
• priv-instr-table.tex : the latex table of instruction used in the riscv-priv spec
• inst.chisel : chisel code to decode instructions
• inst.sverilog : system verilog code to decode instructions
• inst.rs : rust code containing mask and match variables for all instructions
• inst.spinalhdl : spinalhdl code to decode instructions
• inst.go : go code to decode instructions

To generate all the above artifacts for all instructions currently checked in, simply run make from the root-directory. This should print the following log on the command-line:

Running with args : ['./parse.py', '-c', '-go', '-chisel', '-sverilog', '-rust', '-latex', '-spinalhdl', 'rv*', 'unratified/rv*']
Extensions selected : ['rv*', 'unratified/rv*']
INFO:: encoding.out.h generated successfully
INFO:: inst.chisel generated successfully
INFO:: inst.spinalhdl generated successfully
INFO:: inst.sverilog generated successfully
INFO:: inst.rs generated successfully
INFO:: inst.go generated successfully
INFO:: instr-table.tex generated successfully
INFO:: priv-instr-table.tex generated successfully

By default all extensions are enabled. To select only a subset of extensions you can change the EXTENSIONS variable of the makefile to contains only the file names of interest. For example if you want only the I and M extensions you can do the following:

make EXTENSIONS='rv*_i rv*_m'

Which will print the following log:

Running with args : ['./parse.py', '-c', '-chisel', '-sverilog', '-rust', '-latex', 'rv32_i', 'rv64_i', 'rv_i', 'rv64_m', 'rv_m']
Extensions selected : ['rv32_i', 'rv64_i', 'rv_i', 'rv64_m', 'rv_m']
INFO:: encoding.out.h generated successfully
INFO:: inst.chisel generated successfully
INFO:: inst.sverilog generated successfully
INFO:: inst.rs generated successfully
INFO:: instr-table.tex generated successfully
INFO:: priv-instr-table.tex generated successfully

If you only want a specific artifact you can use one or more of the following targets : c, rust, chisel, sverilog, latex

You can use the clean target to remove all artifacts.

Adding a new extension

To add a new extension of instructions, create an appropriate rv* file based on the policy defined in File Structure. Run make from the root directory to ensure that all checks pass and all artifacts are created correctly. A successful run should print the following log on the terminal:

Running with args : ['./parse.py', '-c', '-chisel', '-sverilog', '-rust', '-latex', 'rv*', 'unratified/rv*']
Extensions selected : ['rv*', 'unratified/rv*']
INFO:: encoding.out.h generated successfully
INFO:: inst.chisel generated successfully
INFO:: inst.sverilog generated successfully
INFO:: inst.rs generated successfully
INFO:: instr-table.tex generated successfully
INFO:: priv-instr-table.tex generated successfully

Create a PR for review.

Enabling Debug logs in parse.py

To enable debug logs in parse.py change level=logging.INFO to level=logging.DEBUG and run the python command. You will now see debug statements on the terminal like below:

DEBUG:: Collecting standard instructions first
DEBUG:: Parsing File: ./rv_i
DEBUG::      Processing line: lui     rd imm20 6..2=0x0D 1..0=3
DEBUG::      Processing line: auipc   rd imm20 6..2=0x05 1..0=3
DEBUG::      Processing line: jal     rd jimm20                          6..2=0x1b 1..0=3
DEBUG::      Processing line: jalr    rd rs1 imm12              14..12=0 6..2=0x19 1..0=3
DEBUG::      Processing line: beq     bimm12hi rs1 rs2 bimm12lo 14..12=0 6..2=0x18 1..0=3
DEBUG::      Processing line: bne     bimm12hi rs1 rs2 bimm12lo 14..12=1 6..2=0x18 1..0=3

How do I find where an instruction is defined?

You can use grep "^\s*<instr-name>" rv* unratified/rv* OR run make and open instr_dict.json and search for the instruction you are looking for. Within that instruction the extension field will indicate which file the instruction was picked from.