rustv32i/rvdc/src/lib.rs
Noratrieb e9a689aa1a
Some checks failed
Rust / build (push) Has been cancelled
start with 64 bit
2025-03-26 21:10:53 +01:00

1811 lines
63 KiB
Rust

#![doc = include_str!("../README.md")]
#![no_std]
#![deny(missing_docs)]
use core::fmt::{self, Debug, Display};
use core::ops::RangeInclusive;
/// The register size of the ISA, RV32 or RV64.
#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub enum Xlen {
/// 32 bit
Rv32,
/// 64 bit
Rv64,
}
impl Xlen {
/// Whether this is [`Xlen::Rv32`].
pub fn is_32(self) -> bool {
matches!(self, Self::Rv32)
}
/// Whether this is [`Xlen::Rv64`].
pub fn is_64(self) -> bool {
matches!(self, Self::Rv64)
}
}
/// A decoded RISC-V integer register.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub struct Reg(pub u8);
impl Reg {
/// The zero register `zero` (`x0`)
pub const ZERO: Reg = Reg(0);
/// The return address register `ra` (`x1`)
pub const RA: Reg = Reg(1);
/// The stack pointer register `sp` (`x2`)
pub const SP: Reg = Reg(2);
/// The global pointer register `gp` (`x3`)
pub const GP: Reg = Reg(3);
/// The thread pointer register `tp` (`x4`)
pub const TP: Reg = Reg(4);
/// Saved register `s0` (`x8`)
pub const S0: Reg = Reg(8);
/// Saved register frame pointer `fp` (`s0`, `x8`)
pub const FP: Reg = Reg(8);
/// Saved register `s1` (`x9`)
pub const S1: Reg = Reg(9);
/// Saved register `s2` (`x18`)
pub const S2: Reg = Reg(18);
/// Saved register `s3` (`x19`)
pub const S3: Reg = Reg(19);
/// Saved register `s4` (`x20`)
pub const S4: Reg = Reg(20);
/// Saved register `s5` (`x21`)
pub const S5: Reg = Reg(21);
/// Saved register `s6` (`x22`)
pub const S6: Reg = Reg(22);
/// Saved register `s7` (`x23`)
pub const S7: Reg = Reg(23);
/// Saved register `s8` (`x24`)
pub const S8: Reg = Reg(24);
/// Saved register `s9` (`x25`)
pub const S9: Reg = Reg(25);
/// Saved register `s10` (`x26`)
pub const S10: Reg = Reg(26);
/// Saved register `s11` (`x27`)
pub const S11: Reg = Reg(27);
/// Argument/return value register `a0` (`x10`)
pub const A0: Reg = Reg(10);
/// Argument/return value register `a1` (`x11`)
pub const A1: Reg = Reg(11);
/// Argument register `a2` (`x12`)
pub const A2: Reg = Reg(12);
/// Argument register `a3` (`x13`)
pub const A3: Reg = Reg(13);
/// Argument register `a4` (`x14`)
pub const A4: Reg = Reg(14);
/// Argument register `a5` (`x15`)
pub const A5: Reg = Reg(15);
/// Argument register `a6` (`x16`)
pub const A6: Reg = Reg(16);
/// Argument register `a7` (`x17`)
pub const A7: Reg = Reg(17);
/// Temporary register `t0` (`x5`)
pub const T0: Reg = Reg(5);
/// Temporary register `t1` (`x6`)
pub const T1: Reg = Reg(6);
/// Temporary register `t2` (`x7`)
pub const T2: Reg = Reg(7);
/// Temporary register `t3` (`x28`)
pub const T3: Reg = Reg(28);
/// Temporary register `t4` (`x29`)
pub const T4: Reg = Reg(29);
/// Temporary register `t5` (`x30`)
pub const T5: Reg = Reg(30);
/// Temporary register `t6` (`x31`)
pub const T6: Reg = Reg(31);
}
impl Display for Reg {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
let n = self.0;
match n {
0 => write!(f, "zero"),
1 => write!(f, "ra"),
2 => write!(f, "sp"),
3 => write!(f, "gp"),
4 => write!(f, "tp"),
5..=7 => write!(f, "t{}", n - 5),
8 => write!(f, "s0"),
9 => write!(f, "s1"),
10..=17 => write!(f, "a{}", n - 10),
18..=27 => write!(f, "s{}", n - 18 + 2),
28..=31 => write!(f, "t{}", n - 28 + 3),
_ => unreachable!("invalid register"),
}
}
}
/// An immediate in an instruction.
/// This represents the real value that will be put in the register,
/// so sign extension has been performed if necessary, and for instructions
/// like `lui` the value will have been shifted.
///
/// This type is XLEN-agnostic, use the XLEN-specific accessors to get the correct value.
#[derive(Copy, Clone, PartialEq, Eq, Hash)]
pub struct Imm(u64);
impl Imm {
/// The immediate `0`.
/// Useful as a shortcut for `Imm::new_u32(0)` and for patterns.
pub const ZERO: Self = Self::new_u32(0);
/// Create a new immediate from the (if necessary) sign-extended value.
pub const fn new_i32(value: i32) -> Self {
Self(value as i64 as u64)
}
/// Create a new immediate from the (if necessary) zero-extended value.
pub const fn new_u32(value: u32) -> Self {
Self(value as u64)
}
/// Get the `u32` (RV32) value of the immediate.
pub const fn as_u32(self) -> u32 {
self.0 as u32
}
/// Get the `i32` (RV32) value of the immediate.
pub const fn as_i32(self) -> i32 {
self.0 as i32
}
/// Get the `u64` (RV64) value of the immediate.
pub const fn as_u64(self) -> u64 {
self.0 as u64
}
/// Get the `i64` (RV64) value of the immediate.
pub const fn as_i64(self) -> i64 {
self.0 as i64
}
}
impl From<i32> for Imm {
fn from(value: i32) -> Self {
Self::new_i32(value)
}
}
impl From<u32> for Imm {
fn from(value: u32) -> Self {
Self::new_u32(value)
}
}
impl From<Imm> for u32 {
fn from(value: Imm) -> Self {
value.as_u32()
}
}
impl From<Imm> for i32 {
fn from(value: Imm) -> Self {
value.as_i32()
}
}
/// A RISC-V instruction.
///
/// Every variant is a different instruction, with immediates as `u32`.
/// For instructions that sign-extend immediates, the immediates will have been
/// sign-extended already, so the value can be used as-is.
/// For instructions that have immediates in the upper bits (`lui`, `auipc`),
/// the shift will have been done already, so the value can also be used as-is.
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
#[rustfmt::skip]
#[expect(missing_docs)] // enum variant fields
#[non_exhaustive]
pub enum Inst {
/// Load Upper Immediate
Lui { uimm: Imm, dest: Reg },
/// Add Upper Immediate to PC
Auipc { uimm: Imm, dest: Reg },
/// Jump And Link
Jal { offset: Imm, dest: Reg },
/// Jump And Link Register (indirect)
Jalr { offset: Imm, base: Reg, dest: Reg },
/// Branch Equal
Beq { offset: Imm, src1: Reg, src2: Reg },
/// Branch Not Equal
Bne { offset: Imm, src1: Reg, src2: Reg },
/// Branch Less Than (signed)
Blt { offset: Imm, src1: Reg, src2: Reg },
/// Branch Greater or Equal (signed)
Bge { offset: Imm, src1: Reg, src2: Reg },
/// Branch Less Than Unsigned
Bltu { offset: Imm, src1: Reg, src2: Reg },
/// Branch Greater or Equal Unsigned
Bgeu { offset: Imm, src1: Reg, src2: Reg },
/// Load Byte (sign-ext)
Lb { offset: Imm, dest: Reg, base: Reg },
/// Load Unsigned Byte (zero-ext)
Lbu { offset: Imm, dest: Reg, base: Reg },
/// Load Half (sign-ext)
Lh { offset: Imm, dest: Reg, base: Reg },
/// Load Unsigned Half (zero-ext)
Lhu { offset: Imm, dest: Reg, base: Reg },
/// Load Word
Lw { offset: Imm, dest: Reg, base: Reg },
/// Store Byte
Sb { offset: Imm, src: Reg, base: Reg },
/// Store Half
Sh { offset: Imm, src: Reg, base: Reg },
/// Store Word
Sw { offset: Imm, src: Reg, base: Reg },
/// Add Immediate
Addi { imm: Imm, dest: Reg, src1: Reg },
/// Add Immediate 32-bit (**RV64 only**)
AddiW { imm: Imm, dest: Reg, src1: Reg },
/// Set Less Than Immediate (signed)
Slti { imm: Imm, dest: Reg, src1: Reg },
/// Set Less Than Immediate Unsigned
Sltiu { imm: Imm, dest: Reg, src1: Reg },
/// XOR Immediate
Xori { imm: Imm, dest: Reg, src1: Reg },
/// OR Immediate
Ori { imm: Imm, dest: Reg, src1: Reg },
/// AND Immediate
Andi { imm: Imm, dest: Reg, src1: Reg },
/// Shift Left Logical Immediate
Slli { imm: Imm, dest: Reg, src1: Reg },
/// Shift Left Logical Immediate 32-bit (**RV64 only**)
SlliW { imm: Imm, dest: Reg, src1: Reg },
/// Shift Right Logical Immediate (unsigned)
Srli { imm: Imm, dest: Reg, src1: Reg },
/// Shift Right Logical Immediate (unsigned) 32-bit (**RV64 only**)
SrliW { imm: Imm, dest: Reg, src1: Reg },
/// Shift Right Arithmetic Immediate (signed)
Srai { imm: Imm, dest: Reg, src1: Reg },
/// Shift Right Arithmetic Immediate (signed) 32-bit (**RV64 only**)
SraiW { imm: Imm, dest: Reg, src1: Reg },
/// Add
Add { dest: Reg, src1: Reg, src2: Reg },
/// Subtract
Sub { dest: Reg, src1: Reg, src2: Reg },
/// Shift Left Logical
Sll { dest: Reg, src1: Reg, src2: Reg },
/// Set Less Than (signed)
Slt { dest: Reg, src1: Reg, src2: Reg },
/// Set Less Than Unsigned
Sltu { dest: Reg, src1: Reg, src2: Reg },
/// XOR
Xor { dest: Reg, src1: Reg, src2: Reg },
/// Shift Right Logical (unsigned)
Srl { dest: Reg, src1: Reg, src2: Reg },
/// Shift Right Arithmetic (unsigned)
Sra { dest: Reg, src1: Reg, src2: Reg },
/// OR
Or { dest: Reg, src1: Reg, src2: Reg },
/// AND
And { dest: Reg, src1: Reg, src2: Reg },
/// Memory Fence
Fence { fence: Fence },
/// ECALL, call into environment
Ecall,
/// EBREAK, break into debugger
Ebreak,
// ------------- M extension -------------
/// Multiply
Mul { dest: Reg, src1: Reg, src2: Reg },
/// Mul Upper Half Signed-Signed
Mulh { dest: Reg, src1: Reg, src2: Reg },
/// Mul Upper Half Signed-Unsigned
Mulhsu { dest: Reg, src1: Reg, src2: Reg },
/// Mul Upper Half Unsigned-Unsigned
Mulhu { dest: Reg, src1: Reg, src2: Reg },
/// Divide (signed)
Div { dest: Reg, src1: Reg, src2: Reg },
/// Divide Unsigned
Divu { dest: Reg, src1: Reg, src2: Reg },
/// Remainder (signed)
Rem { dest: Reg, src1: Reg, src2: Reg },
/// Remainder Unsigned
Remu { dest: Reg, src1: Reg, src2: Reg },
// ------------- A extension -------------
/// Load-Reserved Word
LrW {
order: AmoOrdering,
dest: Reg,
addr: Reg,
},
/// Store-Conditional Word
ScW {
order: AmoOrdering,
dest: Reg,
addr: Reg,
src: Reg,
},
/// Atomic Memory Operation
AmoW {
order: AmoOrdering,
op: AmoOp,
dest: Reg,
addr: Reg,
src: Reg,
},
}
/// The details of a RISC-V `fence` instruction.
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
pub struct Fence {
/// The `fm` field of the instruction.
/// - `0b0000` is a normal fence
/// - `0b1000` with `rw,rw` implies a `fence.tso`
pub fm: u8,
/// The predecessor set.
pub pred: FenceSet,
/// The sucessor set.
pub succ: FenceSet,
/// The `rd` field of the instruction. Currently always zero.
pub dest: Reg,
/// The `rs1` field of the instruction. Currently always zero.
pub src: Reg,
}
/// The affected parts of a fence.
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
#[expect(missing_docs)]
pub struct FenceSet {
pub device_input: bool,
pub device_output: bool,
pub memory_read: bool,
pub memory_write: bool,
}
/// An atomic memory ordering for instructions from the A extension.
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
pub enum AmoOrdering {
/// No bits.
Relaxed,
/// `aq`
Acquire,
/// `rl`
Release,
/// `aq`, `rl`
SeqCst,
}
/// An atomic memory operations from the Zaamo extension.
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
pub enum AmoOp {
/// Swap
Swap,
/// ADD
Add,
/// XOR
Xor,
/// AND
And,
/// OR
Or,
/// Signed minimum
Min,
/// Signed maximum
Max,
/// Unsigned minimum
Minu,
/// Unsigned maximum
Maxu,
}
/// The error used for invalid instructions containing information about the instruction and error.
///
/// Note that this is also returned for the defined illegal instruction of all zero.
pub struct DecodeError {
/// The instruction bytes that failed to decode.
pub instruction: u32,
/// Which field of the instruction contained unexpected bits.
pub unexpected_field: &'static str,
}
impl Fence {
/// Whether this is a `fence.tso`.
/// `fm=0b1000` and `RW,RW`
pub fn is_tso(&self) -> bool {
matches!(
self,
Self {
fm: 0b1000,
pred: FenceSet {
device_input: _,
device_output: _,
memory_read: true,
memory_write: true
},
succ: FenceSet {
device_input: _,
device_output: _,
memory_read: true,
memory_write: true
},
..
}
)
}
/// Whether this fence indicates a `pause` assembler pseudoinstruction.
pub fn is_pause(&self) -> bool {
self.pred
== FenceSet {
device_input: false,
device_output: false,
memory_read: false,
memory_write: true,
}
&& self.succ
== FenceSet {
device_input: false,
device_output: false,
memory_read: false,
memory_write: false,
}
&& self.dest == Reg::ZERO
&& self.src == Reg::ZERO
}
}
impl AmoOrdering {
/// Create a new [`AmoOrdering`] from the two ordering bits.
pub fn from_aq_rl(aq: bool, rl: bool) -> Self {
match (aq, rl) {
(false, false) => Self::Relaxed,
(true, false) => Self::Acquire,
(false, true) => Self::Release,
(true, true) => Self::SeqCst,
}
}
}
impl Debug for Inst {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
Display::fmt(&self, f)
}
}
/// Prints the instruction in disassembled form.
///
/// Note that the precise output here is not considered stable.
impl Display for Inst {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match *self {
Inst::Lui { uimm, dest } => write!(f, "lui {dest}, {}", uimm.as_u32() >> 12),
Inst::Auipc { uimm, dest } => write!(f, "auipc {dest}, {}", uimm.as_u32() >> 12),
Inst::Jal { offset, dest } => {
if dest.0 == 0 {
write!(f, "j {}", offset.as_i32())
} else {
write!(f, "jal {dest}, {}", offset.as_i32())
}
}
Inst::Jalr { offset, base, dest } => {
if dest == Reg::ZERO && offset.as_u32() == 0 && base == Reg::RA {
write!(f, "ret")
} else {
write!(f, "jalr {dest}, {}({base})", offset.as_i32())
}
}
Inst::Beq { offset, src1, src2 } => {
write!(f, "beq {src1}, {src2}, {}", offset.as_i32())
}
Inst::Bne { offset, src1, src2 } => {
write!(f, "bne {src1}, {src2}, {}", offset.as_i32())
}
Inst::Blt { offset, src1, src2 } => {
write!(f, "blt {src1}, {src2}, {}", offset.as_i32())
}
Inst::Bge { offset, src1, src2 } => {
write!(f, "bge {src1}, {src2}, {}", offset.as_i32())
}
Inst::Bltu { offset, src1, src2 } => {
write!(f, "bltu {src1}, {src2}, {}", offset.as_i32())
}
Inst::Bgeu { offset, src1, src2 } => {
write!(f, "bgeu {src1}, {src2}, {}", offset.as_i32())
}
Inst::Lb { offset, dest, base } => write!(f, "lb {dest}, {}({base})", offset.as_i32()),
Inst::Lbu { offset, dest, base } => {
write!(f, "lbu {dest}, {}({base})", offset.as_i32())
}
Inst::Lh { offset, dest, base } => write!(f, "lh {dest}, {}({base})", offset.as_i32()),
Inst::Lhu { offset, dest, base } => {
write!(f, "lhu {dest}, {}({base})", offset.as_i32())
}
Inst::Lw { offset, dest, base } => write!(f, "lw {dest}, {}({base})", offset.as_i32()),
Inst::Sb { offset, src, base } => write!(f, "sb {src}, {}({base})", offset.as_i32()),
Inst::Sh { offset, src, base } => write!(f, "sh {src}, {}({base})", offset.as_i32()),
Inst::Sw { offset, src, base } => write!(f, "sw {src}, {}({base})", offset.as_i32()),
Inst::Addi { imm, dest, src1 } => {
if dest.0 == 0 && src1.0 == 0 && imm.as_u32() == 0 {
write!(f, "nop")
} else if src1.0 == 0 {
write!(f, "li {dest}, {}", imm.as_i32())
} else if imm.as_u32() == 0 {
write!(f, "mv {dest}, {src1}")
} else {
write!(f, "addi {dest}, {src1}, {}", imm.as_i32())
}
}
Inst::AddiW { imm, dest, src1 } => {
if imm.as_u32() == 0 {
write!(f, "sext.w {dest}, {src1}")
} else {
write!(f, "addi.w {dest}, {src1}, {}", imm.as_i32())
}
}
Inst::Slti {
imm,
dest,
src1: rs1,
} => write!(f, "slti {dest}, {rs1}, {}", imm.as_i32()),
Inst::Sltiu {
imm,
dest,
src1: rs1,
} => write!(f, "sltiu {dest}, {rs1}, {}", imm.as_i32()),
Inst::Andi {
imm,
dest,
src1: rs1,
} => write!(f, "andi {dest}, {rs1}, {}", imm.as_i32()),
Inst::Ori {
imm,
dest,
src1: rs1,
} => write!(f, "ori {dest}, {rs1}, {}", imm.as_i32()),
Inst::Xori {
imm,
dest,
src1: rs1,
} => write!(f, "xori {dest}, {rs1}, {}", imm.as_i32()),
Inst::Slli {
imm,
dest,
src1: rs1,
} => write!(f, "slli {dest}, {rs1}, {}", imm.as_i32()),
Inst::SlliW {
imm,
dest,
src1: rs1,
} => write!(f, "slliw {dest}, {rs1}, {}", imm.as_i32()),
Inst::Srli {
imm,
dest,
src1: rs1,
} => write!(f, "srli {dest}, {rs1}, {}", imm.as_i32()),
Inst::SrliW {
imm,
dest,
src1: rs1,
} => write!(f, "srliw {dest}, {rs1}, {}", imm.as_i32()),
Inst::Srai {
imm,
dest,
src1: rs1,
} => write!(f, "srai {dest}, {rs1}, {}", imm.as_i32()),
Inst::SraiW {
imm,
dest,
src1: rs1,
} => write!(f, "sraiw {dest}, {rs1}, {}", imm.as_i32()),
Inst::Add { dest, src1, src2 } => {
write!(f, "add {dest}, {src1}, {src2}")
}
Inst::Sub { dest, src1, src2 } => write!(f, "sub {dest}, {src1}, {src2}"),
Inst::Sll { dest, src1, src2 } => write!(f, "sll {dest}, {src1}, {src2}"),
Inst::Slt { dest, src1, src2 } => write!(f, "slt {dest}, {src1}, {src2}"),
Inst::Sltu { dest, src1, src2 } => write!(f, "sltu {dest}, {src1}, {src2}"),
Inst::Xor { dest, src1, src2 } => write!(f, "xor {dest}, {src1}, {src2}"),
Inst::Srl { dest, src1, src2 } => write!(f, "srl {dest}, {src1}, {src2}"),
Inst::Sra { dest, src1, src2 } => write!(f, "sra {dest}, {src1}, {src2}"),
Inst::Or { dest, src1, src2 } => write!(f, "or {dest}, {src1}, {src2}"),
Inst::And { dest, src1, src2 } => write!(f, "and {dest}, {src1}, {src2}"),
Inst::Fence { fence } => match fence.fm {
_ if fence.is_tso() => write!(f, "fence.tso"),
0b0000 if fence.is_pause() => {
write!(f, "pause")
}
_ => write!(f, "fence {},{}", fence.pred, fence.succ),
},
Inst::Ecall => write!(f, "ecall"),
Inst::Ebreak => write!(f, "ebreak"),
Inst::Mul { dest, src1, src2 } => write!(f, "mul {dest}, {src1}, {src2}"),
Inst::Mulh { dest, src1, src2 } => write!(f, "mulh {dest}, {src1}, {src2}"),
Inst::Mulhsu { dest, src1, src2 } => write!(f, "mulhsu {dest}, {src1}, {src2}"),
Inst::Mulhu { dest, src1, src2 } => write!(f, "mulhu {dest}, {src1}, {src2}"),
Inst::Div { dest, src1, src2 } => write!(f, "div {dest}, {src1}, {src2}"),
Inst::Divu { dest, src1, src2 } => write!(f, "divu {dest}, {src1}, {src2}"),
Inst::Rem { dest, src1, src2 } => write!(f, "rem {dest}, {src1}, {src2}"),
Inst::Remu { dest, src1, src2 } => write!(f, "remu {dest}, {src1}, {src2}"),
Inst::LrW { order, dest, addr } => write!(f, "lr.w{order} {dest}, ({addr})",),
Inst::ScW {
order,
dest,
addr,
src,
} => write!(f, "sc.w{order} {dest}, {src}, ({addr})"),
Inst::AmoW {
order,
op,
dest,
addr,
src,
} => write!(f, "amo{op}.w{order} {dest}, {src}, ({addr})",),
}
}
}
impl Display for FenceSet {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
let mut has = false;
if self.device_input {
has = true;
write!(f, "i")?;
}
if self.device_output {
has = true;
write!(f, "o")?;
}
if self.memory_read {
has = true;
write!(f, "r")?;
}
if self.memory_write {
has = true;
write!(f, "w")?;
}
if !has {
write!(f, "0")?;
}
Ok(())
}
}
impl Display for AmoOrdering {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
AmoOrdering::Relaxed => write!(f, ""),
AmoOrdering::Acquire => write!(f, ".aq"),
AmoOrdering::Release => write!(f, ".rl"),
AmoOrdering::SeqCst => write!(f, ".aqrl"),
}
}
}
impl Display for AmoOp {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
AmoOp::Swap => write!(f, "swap"),
AmoOp::Add => write!(f, "add"),
AmoOp::Xor => write!(f, "xor"),
AmoOp::And => write!(f, "and"),
AmoOp::Or => write!(f, "or"),
AmoOp::Min => write!(f, "min"),
AmoOp::Max => write!(f, "max"),
AmoOp::Minu => write!(f, "minu"),
AmoOp::Maxu => write!(f, "maxu"),
}
}
}
impl Debug for DecodeError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("DecodeError")
.field("instruction", &format_args!("{:0>32b}", self.instruction))
.field("unexpected_field", &self.unexpected_field)
.finish()
}
}
impl Display for DecodeError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
write!(
f,
"failed to decode instruction '{:0>32b}' because of field '{}'",
self.instruction, self.unexpected_field
)
}
}
impl core::error::Error for DecodeError {}
fn sign_extend(value: u32, size: u32) -> u32 {
let right = u32::BITS - size;
(((value << right) as i32) >> right) as u32
}
#[derive(Clone, Copy)]
struct InstCode(u32);
impl InstCode {
fn extract(self, range: RangeInclusive<u32>) -> u32 {
let end_span = 32 - (range.end() + 1);
(self.0 << (end_span)) >> (end_span + range.start())
}
fn immediate_u(self, mappings: &[(RangeInclusive<u32>, u32)]) -> Imm {
let mut imm = 0;
for (from, to) in mappings {
let value = self.extract(from.clone());
imm |= value << to;
}
Imm::new_u32(imm)
}
fn immediate_s(self, mappings: &[(RangeInclusive<u32>, u32)]) -> Imm {
let mut imm = 0;
let mut size = 0;
for (from, to) in mappings {
let value = self.extract(from.clone());
imm |= value << to;
let this_size = from.end() - from.start() + 1;
size = size.max(*to + this_size);
}
Imm::new_i32(sign_extend(imm, size) as i32)
}
fn opcode(self) -> u32 {
self.0 & 0b1111111
}
fn funct3(self) -> u32 {
self.extract(12..=14)
}
fn funct7(self) -> u32 {
self.extract(25..=31)
}
fn rs1(self) -> Reg {
Reg(self.extract(15..=19) as u8)
}
fn rs2(self) -> Reg {
Reg(self.extract(20..=24) as u8)
}
fn rs2_imm(self) -> u32 {
self.extract(20..=24)
}
fn rd(self) -> Reg {
Reg(self.extract(7..=11) as u8)
}
fn imm_i(self) -> Imm {
self.immediate_s(&[(20..=31, 0)])
}
fn imm_s(self) -> Imm {
self.immediate_s(&[(25..=31, 5), (7..=11, 0)])
}
fn imm_b(self) -> Imm {
self.immediate_s(&[(31..=31, 12), (7..=7, 11), (25..=30, 5), (8..=11, 1)])
}
fn imm_u(self) -> Imm {
self.immediate_u(&[(12..=31, 12)])
}
fn imm_j(self) -> Imm {
self.immediate_s(&[(31..=31, 20), (21..=30, 1), (20..=20, 11), (12..=19, 12)])
}
}
#[derive(Clone, Copy)]
struct InstCodeC(u16);
impl InstCodeC {
fn extract(self, range: RangeInclusive<u32>) -> u32 {
let end_span = u16::BITS - (range.end() + 1);
((self.0 << (end_span)) >> (end_span + range.start())) as u32
}
fn immediate_u(self, mappings: &[(RangeInclusive<u32>, u32)]) -> Imm {
let mut imm = 0;
for (from, to) in mappings {
let value = self.extract(from.clone());
imm |= value << to;
}
Imm::new_u32(imm)
}
fn immediate_s(self, mappings: &[(RangeInclusive<u32>, u32)]) -> Imm {
let mut imm = 0;
let mut size = 0;
for (from, to) in mappings {
assert!(from.start() <= from.end());
let value = self.extract(from.clone());
imm |= value << to;
let this_size = from.end() - from.start() + 1;
size = size.max(*to + this_size);
}
Imm::new_i32(sign_extend(imm, size) as i32)
}
fn quadrant(self) -> u16 {
self.0 & 0b11
}
fn funct3(self) -> u32 {
self.extract(13..=15)
}
fn funct2(self) -> u32 {
self.extract(10..=11)
}
/// rd/rs1 (7..=11)
fn rd(self) -> Reg {
Reg(self.extract(7..=11) as u8)
}
/// rs2 (2..=6)
fn rs2(self) -> Reg {
Reg(self.extract(2..=6) as u8)
}
/// rs1' (7..=9)
fn rs1_short(self) -> Reg {
let smol_reg = self.extract(7..=9);
// map to x8..=x15
Reg((smol_reg + 8) as u8)
}
/// rs2' (2..=4)
fn rs2_short(self) -> Reg {
let smol_reg = self.extract(2..=4);
// map to x8..=x15
Reg((smol_reg + 8) as u8)
}
}
impl From<InstCodeC> for InstCode {
fn from(value: InstCodeC) -> Self {
Self(value.0 as u32)
}
}
/// Whether the decoded instruction was a compressed instruction or not.
/// If it was compressed, only the first two bytes were used.
/// If it was not compressed, all four bytes are consumed.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum IsCompressed {
/// Normal 4-byte instruction
No,
/// Compressed 2-byte instruction
Yes,
}
fn decode_error(instruction: impl Into<InstCode>, unexpected_field: &'static str) -> DecodeError {
DecodeError {
instruction: instruction.into().0,
unexpected_field,
}
}
impl Inst {
/// Whether the first byte of an instruction indicates a compressed or uncompressed instruction.
///
/// # Examples
///
/// ```rust
/// // addi sp, sp, -0x20 (compressed)
/// let x = 0x1101_u32;
/// assert!(rvdc::Inst::first_byte_is_compressed(x.to_le_bytes()[0]));
/// let x = 0x1101_u16;
/// assert!(rvdc::Inst::first_byte_is_compressed(x.to_le_bytes()[0]));
/// ```
///
/// ```rust
/// // auipc t1, 0xa
/// let x = 0x0000a317_u32;
/// assert!(!rvdc::Inst::first_byte_is_compressed(x.to_le_bytes()[0]));
/// ```
pub fn first_byte_is_compressed(byte: u8) -> bool {
(byte & 0b11) != 0b11
}
/// Decode an instruction from four bytes.
///
/// The instruction may be compressed, in which case only two bytes are consumed.
/// Even in these cases, the full next four bytes must be passed.
///
/// If the caller wants to avoid reading more bytes than necessary, [`Self::first_byte_is_compressed`]
/// can be used to check, read the required bytes, and then call [`Self::decode_compressed`] or
/// [`Self::decode_normal`] directly.
pub fn decode(code: u32, xlen: Xlen) -> Result<(Inst, IsCompressed), DecodeError> {
let is_compressed = (code & 0b11) != 0b11;
if is_compressed {
Ok((
Self::decode_compressed(code as u16, xlen)?,
IsCompressed::Yes,
))
} else {
Ok((Self::decode_normal(code, xlen)?, IsCompressed::No))
}
}
/// Decode a known compressed instruction from its two bytes.
///
/// # Example
/// ```rust
/// // Compressed addi sp, sp, -0x20
/// let x = 0x1101_u16;
/// let expected = rvdc::Inst::Addi { imm: rvdc::Imm::new_i32(-0x20), dest: rvdc::Reg::SP, src1: rvdc::Reg::SP };
///
/// let inst = rvdc::Inst::decode_compressed(x).unwrap();
/// assert_eq!(inst, expected);
/// ```
pub fn decode_compressed(code: u16, xlen: Xlen) -> Result<Inst, DecodeError> {
let code = InstCodeC(code);
if code.0 == 0 {
return Err(decode_error(code, "null instruction"));
}
let inst = match code.quadrant() {
// C0
0b00 => match code.funct3() {
// C.ADDI4SPN -> addi \rd', sp, \imm
0b000 => {
let imm =
code.immediate_u(&[(5..=5, 3), (6..=6, 2), (7..=10, 6), (11..=12, 4)]);
if imm.as_u32() == 0 {
return Err(decode_error(code, "uimm=0 for C.ADDISPN is reserved"));
}
Inst::Addi {
imm,
dest: code.rs2_short(),
src1: Reg::SP,
}
}
// C.LW -> lw \dest \offset(\base)
0b010 => Inst::Lw {
offset: code.immediate_u(&[(10..=12, 3), (5..=5, 6), (6..=6, 2)]),
dest: code.rs2_short(),
base: code.rs1_short(),
},
// C.SW -> sw \src, \offset(\base)
0b110 => Inst::Sw {
offset: code.immediate_u(&[(10..=12, 3), (5..=5, 6), (6..=6, 2)]),
src: code.rs2_short(),
base: code.rs1_short(),
},
_ => return Err(decode_error(code, "C0 funct3")),
},
// C1
0b01 => match code.funct3() {
// C.ADDI -> addi \rd, \rd, \imm
0b000 => Inst::Addi {
imm: code.immediate_s(&[(2..=6, 0), (12..=12, 5)]),
dest: code.rd(),
src1: code.rd(),
},
// C.JAL -> jal ra, \offset
0b001 => Inst::Jal {
offset: code.immediate_s(&[
(2..=2, 5),
(3..=5, 1),
(6..=6, 7),
(7..=7, 6),
(8..=8, 10),
(9..=10, 8),
(11..=11, 4),
(12..=12, 11),
]),
dest: Reg::RA,
},
// C.LI -> addi \rd, zero, \imm
0b010 => Inst::Addi {
imm: code.immediate_s(&[(2..=6, 0), (12..=12, 5)]),
dest: code.rd(),
src1: Reg::ZERO,
},
// Arithmetic instructions
0b100 => {
let bit12 = code.extract(12..=12);
match code.funct2() {
// C.SRLI -> srli \rd', \rd', \imm
0b00 => {
if bit12 != 0 {
return Err(decode_error(code, "C.SRLI imm"));
}
Inst::Srli {
imm: code.immediate_u(&[(2..=6, 0), (12..=12, 5)]),
dest: code.rs1_short(),
src1: code.rs1_short(),
}
}
// C.SRAI -> srai \rd', \rd', \imm
0b01 => {
if bit12 != 0 {
return Err(decode_error(code, "C.SRLI imm"));
}
Inst::Srai {
imm: code.immediate_u(&[(2..=6, 0), (12..=12, 5)]),
dest: code.rs1_short(),
src1: code.rs1_short(),
}
}
// C.ANDI -> andi \rd', \rd', \imm
0b10 => Inst::Andi {
imm: code.immediate_u(&[(2..=6, 0), (12..=12, 5)]),
dest: code.rs1_short(),
src1: code.rs1_short(),
},
0b11 => {
if bit12 != 0 {
return Err(decode_error(code, "C1 Arith bit 12"));
}
let funct2 = code.extract(5..=6);
match funct2 {
// C.SUB -> sub \rd', \rd', \rs2'
0b00 => Inst::Sub {
dest: code.rs1_short(),
src1: code.rs1_short(),
src2: code.rs2_short(),
},
// C.XOR -> xor \rd', \rd', \rs2'
0b01 => Inst::Xor {
dest: code.rs1_short(),
src1: code.rs1_short(),
src2: code.rs2_short(),
},
// C.OR -> or \rd', \rd', \rs2'
0b10 => Inst::Or {
dest: code.rs1_short(),
src1: code.rs1_short(),
src2: code.rs2_short(),
},
// C.AND -> and \rd', \rd', \rs2'
0b11 => Inst::And {
dest: code.rs1_short(),
src1: code.rs1_short(),
src2: code.rs2_short(),
},
_ => unreachable!("only two bits"),
}
}
_ => unreachable!("only two bits"),
}
}
// C.J -> jal zero, \offset
0b101 => Inst::Jal {
offset: code.immediate_s(&[
(2..=2, 5),
(3..=5, 1),
(6..=6, 7),
(7..=7, 6),
(8..=8, 10),
(9..=10, 8),
(11..=11, 4),
(12..=12, 11),
]),
dest: Reg::ZERO,
},
0b011 => {
match code.rd().0 {
// C.ADDI16SP -> addi sp, sp, \imm
2 => Inst::Addi {
imm: code.immediate_s(&[
(2..=2, 5),
(3..=4, 7),
(5..=5, 6),
(6..=6, 4),
(12..=12, 9),
]),
dest: Reg::SP,
src1: Reg::SP,
},
// C.LUI -> lui \rd, \imm
_ => {
let uimm = code.immediate_s(&[(2..=6, 12), (12..=12, 17)]);
if uimm.as_u32() == 0 {
return Err(decode_error(code, "C.LUI imm"));
}
Inst::Lui {
uimm,
dest: code.rd(),
}
}
}
}
// C.BEQZ -> beq \rs1', zero, \offset
0b110 => Inst::Beq {
offset: code.immediate_s(&[
(2..=2, 5),
(3..=4, 1),
(5..=6, 6),
(10..=11, 3),
(12..=12, 8),
]),
src1: code.rs1_short(),
src2: Reg::ZERO,
},
// C.BEQZ -> bne \rs1', zero, \offset
0b111 => Inst::Bne {
offset: code.immediate_s(&[
(2..=2, 5),
(3..=4, 1),
(5..=6, 6),
(10..=11, 3),
(12..=12, 8),
]),
src1: code.rs1_short(),
src2: Reg::ZERO,
},
_ => return Err(decode_error(code, "C1 funct3")),
},
// C2
0b10 => match code.funct3() {
// C.SLLI -> slli \rd, \rd, \imm
0b000 => {
if code.extract(12..=12) != 0 {
return Err(decode_error(code, "C.SLLI shift amount must be zero"));
}
Inst::Slli {
imm: code.immediate_u(&[(2..=6, 0), (12..=12, 5)]),
dest: code.rd(),
src1: code.rd(),
}
}
// C.LWSP -> lw \reg \offset(sp)
0b010 => {
let dest = code.rd();
if dest.0 == 0 {
return Err(decode_error(code, "C.LWSP rd"));
}
Inst::Lw {
offset: code.immediate_u(&[(12..=12, 5), (4..=6, 2), (2..=3, 6)]),
dest,
base: Reg::SP,
}
}
0b100 => {
let bit = code.extract(12..=12);
let rs2 = code.rs2();
let rd_rs1 = code.rd();
match (bit, rd_rs1.0, rs2.0) {
// C.JR -> jalr zero, 0(\rs1)
(0, _, 0) => {
if rd_rs1.0 == 0 {
return Err(decode_error(code, "C.JR rs1"));
}
Inst::Jalr {
offset: Imm::ZERO,
base: rd_rs1,
dest: Reg::ZERO,
}
}
// C.MV -> add \rd, x0, \rs2
(0, _, _) => Inst::Add {
dest: code.rd(),
src1: Reg::ZERO,
src2: code.rs2(),
},
// C.EBREAK -> ebreak
(1, 0, 0) => Inst::Ebreak,
// C.JALR -> jalr ra, 0(\rs1)
(1, _, 0) if rd_rs1.0 != 0 => Inst::Jalr {
offset: Imm::ZERO,
base: rd_rs1,
dest: Reg::RA,
},
// C.ADD -> add \rd, \rd, \rs2
(1, _, _) => Inst::Add {
dest: rd_rs1,
src1: rd_rs1,
src2: rs2,
},
_ => return Err(decode_error(code, "C2 funct=100 inst")),
}
}
// C.SWSP -> sw \reg \offset(sp)
0b110 => Inst::Sw {
offset: code.immediate_u(&[(7..=8, 6), (9..=12, 2)]),
src: code.rs2(),
base: Reg::SP,
},
_ => return Err(decode_error(code, "C2 funct3")),
},
_ => return Err(decode_error(code, "instruction is not compressed")),
};
Ok(inst)
}
/// Decode a normal (not compressed) instruction.
pub fn decode_normal(code: u32, xlen: Xlen) -> Result<Inst, DecodeError> {
let code = InstCode(code);
let inst = match code.opcode() {
// LUI
0b0110111 => Inst::Lui {
uimm: code.imm_u(),
dest: code.rd(),
},
// AUIPC
0b0010111 => Inst::Auipc {
uimm: code.imm_u(),
dest: code.rd(),
},
// JAL
0b1101111 => Inst::Jal {
offset: code.imm_j(),
dest: code.rd(),
},
// JALR
0b1100111 => match code.funct3() {
0b000 => Inst::Jalr {
offset: code.imm_i(),
base: code.rs1(),
dest: code.rd(),
},
_ => return Err(decode_error(code, "JALR funct3")),
},
// BRANCH
0b1100011 => match code.funct3() {
0b000 => Inst::Beq {
offset: code.imm_b(),
src1: code.rs1(),
src2: code.rs2(),
},
0b001 => Inst::Bne {
offset: code.imm_b(),
src1: code.rs1(),
src2: code.rs2(),
},
0b100 => Inst::Blt {
offset: code.imm_b(),
src1: code.rs1(),
src2: code.rs2(),
},
0b101 => Inst::Bge {
offset: code.imm_b(),
src1: code.rs1(),
src2: code.rs2(),
},
0b110 => Inst::Bltu {
offset: code.imm_b(),
src1: code.rs1(),
src2: code.rs2(),
},
0b111 => Inst::Bgeu {
offset: code.imm_b(),
src1: code.rs1(),
src2: code.rs2(),
},
_ => return Err(decode_error(code, "BRANCH funct3")),
},
// LOAD
0b0000011 => match code.funct3() {
0b000 => Inst::Lb {
offset: code.imm_i(),
dest: code.rd(),
base: code.rs1(),
},
0b001 => Inst::Lh {
offset: code.imm_i(),
dest: code.rd(),
base: code.rs1(),
},
0b010 => Inst::Lw {
offset: code.imm_i(),
dest: code.rd(),
base: code.rs1(),
},
0b100 => Inst::Lbu {
offset: code.imm_i(),
dest: code.rd(),
base: code.rs1(),
},
0b101 => Inst::Lhu {
offset: code.imm_i(),
dest: code.rd(),
base: code.rs1(),
},
_ => return Err(decode_error(code, "LOAD funct3")),
},
// STORE
0b0100011 => match code.funct3() {
0b000 => Inst::Sb {
offset: code.imm_s(),
src: code.rs2(),
base: code.rs1(),
},
0b001 => Inst::Sh {
offset: code.imm_s(),
src: code.rs2(),
base: code.rs1(),
},
0b010 => Inst::Sw {
offset: code.imm_s(),
src: code.rs2(),
base: code.rs1(),
},
_ => return Err(decode_error(code, "STORE funct3")),
},
// OP-IMM
0b0010011 => match code.funct3() {
0b000 => Inst::Addi {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
0b010 => Inst::Slti {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
0b011 => Inst::Sltiu {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
0b100 => Inst::Xori {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
0b110 => Inst::Ori {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
0b111 => Inst::Andi {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
0b001 => {
if code.funct7() != 0 {
return Err(decode_error(code, "slli shift overflow"));
}
Inst::Slli {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
}
}
0b101 => match code.funct7() {
0b0000000 => Inst::Srli {
imm: Imm::new_u32(code.rs2_imm()),
dest: code.rd(),
src1: code.rs1(),
},
0b0100000 => Inst::Srai {
imm: Imm::new_u32(code.rs2_imm()),
dest: code.rd(),
src1: code.rs1(),
},
_ => return Err(decode_error(code, "OP-IMM funct7")),
},
_ => return Err(decode_error(code, "OP-IMM funct3")),
},
// OP-IMM-32
0b0011011 => {
if !xlen.is_64() {
return Err(decode_error(code, "OP-IMM-32 only on RV64"));
}
match code.funct3() {
0b000 => Inst::AddiW {
imm: code.imm_i(),
dest: code.rd(),
src1: code.rs1(),
},
// SLLIW
0b001 => {
if code.funct7() != 0 {
return Err(decode_error(code, "SLLIW funct7"));
}
Inst::SlliW {
imm: Imm::new_u32(code.rs2_imm()),
dest: code.rd(),
src1: code.rs1(),
}
}
0b101 => match code.funct7() {
0b0000000 => Inst::SrliW {
imm: Imm::new_u32(code.rs2_imm()),
dest: code.rd(),
src1: code.rs1(),
},
0b0100000 => Inst::SraiW {
imm: Imm::new_u32(code.rs2_imm()),
dest: code.rd(),
src1: code.rs1(),
},
_ => return Err(decode_error(code, "OP-IMM-32 funct7")),
},
_ => return Err(decode_error(code, "OP-IMM-32 funct3")),
}
}
// OP
0b0110011 => {
let (dest, src1, src2) = (code.rd(), code.rs1(), code.rs2());
match (code.funct3(), code.funct7()) {
(0b000, 0b0000000) => Inst::Add { dest, src1, src2 },
(0b000, 0b0100000) => Inst::Sub { dest, src1, src2 },
(0b001, 0b0000000) => Inst::Sll { dest, src1, src2 },
(0b010, 0b0000000) => Inst::Slt { dest, src1, src2 },
(0b011, 0b0000000) => Inst::Sltu { dest, src1, src2 },
(0b100, 0b0000000) => Inst::Xor { dest, src1, src2 },
(0b101, 0b0000000) => Inst::Srl { dest, src1, src2 },
(0b101, 0b0100000) => Inst::Sra { dest, src1, src2 },
(0b110, 0b0000000) => Inst::Or { dest, src1, src2 },
(0b111, 0b0000000) => Inst::And { dest, src1, src2 },
(0b000, 0b0000001) => Inst::Mul { dest, src1, src2 },
(0b001, 0b0000001) => Inst::Mulh { dest, src1, src2 },
(0b010, 0b0000001) => Inst::Mulhsu { dest, src1, src2 },
(0b011, 0b0000001) => Inst::Mulhu { dest, src1, src2 },
(0b100, 0b0000001) => Inst::Div { dest, src1, src2 },
(0b101, 0b0000001) => Inst::Divu { dest, src1, src2 },
(0b110, 0b0000001) => Inst::Rem { dest, src1, src2 },
(0b111, 0b0000001) => Inst::Remu { dest, src1, src2 },
_ => return Err(decode_error(code, "OP funct3/funct7")),
}
}
// MISC-MEM
0b0001111 => {
let fm = code.extract(28..=31);
let pred = FenceSet {
device_input: code.extract(27..=27) == 1,
device_output: code.extract(26..=26) == 1,
memory_read: code.extract(25..=25) == 1,
memory_write: code.extract(24..=24) == 1,
};
let succ = FenceSet {
device_input: code.extract(23..=23) == 1,
device_output: code.extract(22..=22) == 1,
memory_read: code.extract(21..=21) == 1,
memory_write: code.extract(20..=20) == 1,
};
match code.funct3() {
0b000 => Inst::Fence {
fence: Fence {
fm: fm as u8,
pred,
succ,
dest: code.rd(),
src: code.rs1(),
},
},
_ => return Err(decode_error(code, "MISC-MEM funct3")),
}
}
// SYSTEM
0b1110011 => {
if code.0 == 0b11000000000000000001000001110011 {
return Err(decode_error(code, "unimp instruction"));
}
if code.rd().0 != 0 {
return Err(decode_error(code, "SYSTEM rd"));
}
if code.funct3() != 0 {
return Err(decode_error(code, "SYSTEM funct3"));
}
if code.rs1().0 != 0 {
return Err(decode_error(code, "SYSTEM rs1"));
}
match code.imm_i().as_u32() {
0b000000000000 => Inst::Ecall,
0b000000000001 => Inst::Ebreak,
_ => return Err(decode_error(code, "SYSTEM imm")),
}
}
// AMO
0b00101111 => {
// width must be W
if code.funct3() != 0b010 {
return Err(decode_error(code, "AMO width funct3"));
}
let kind = code.extract(27..=31);
let aq = code.extract(26..=26) == 1;
let rl = code.extract(25..=25) == 1;
let order = AmoOrdering::from_aq_rl(aq, rl);
match kind {
// LR
0b00010 => {
if code.rs2().0 != 0 {
return Err(decode_error(code, "AMO.LR rs2"));
}
Inst::LrW {
order,
dest: code.rd(),
addr: code.rs1(),
}
}
// SC
0b00011 => Inst::ScW {
order,
dest: code.rd(),
addr: code.rs1(),
src: code.rs2(),
},
_ => {
let op = match kind {
0b00001 => AmoOp::Swap,
0b00000 => AmoOp::Add,
0b00100 => AmoOp::Xor,
0b01100 => AmoOp::And,
0b01000 => AmoOp::Or,
0b10000 => AmoOp::Min,
0b10100 => AmoOp::Max,
0b11000 => AmoOp::Minu,
0b11100 => AmoOp::Maxu,
_ => return Err(decode_error(code, "AMO op funct7")),
};
Inst::AmoW {
order,
op,
dest: code.rd(),
addr: code.rs1(),
src: code.rs2(),
}
}
}
}
_ => return Err(decode_error(code, "opcode")),
};
Ok(inst)
}
}
#[cfg(test)]
mod tests {
extern crate std;
use std::prelude::rust_2024::*;
use std::fmt::Write as _;
use std::io::Write as _;
use object::Object;
use object::ObjectSection;
use crate::Fence;
use crate::FenceSet;
use crate::Imm;
use crate::Inst;
use crate::Reg;
use crate::Xlen;
#[test]
#[cfg_attr(not(slow_tests), ignore = "cfg(slow_tests) not enabled")]
fn exhaustive_decode_no_panic() {
for i in 0..u32::MAX {
if (i % (2 << 25)) == 0 {
let percent = i as f32 / (u32::MAX as f32);
let done = (100.0 * percent) as usize;
std::print!("\r{}{}", "#".repeat(done), "-".repeat(100 - done));
std::io::stdout().flush().unwrap();
}
let _ = Inst::decode(i, Xlen::Rv32);
}
let _ = Inst::decode(u32::MAX, Xlen::Rv32);
for i in 0..u32::MAX {
if (i % (2 << 25)) == 0 {
let percent = i as f32 / (u32::MAX as f32);
let done = (100.0 * percent) as usize;
std::print!("\r{}{}", "#".repeat(done), "-".repeat(100 - done));
std::io::stdout().flush().unwrap();
}
let _ = Inst::decode(i, Xlen::Rv64);
}
let _ = Inst::decode(u32::MAX, Xlen::Rv64);
}
#[test]
fn size_of_instruction() {
assert!(size_of::<Inst>() <= 12);
}
const TEST_SECTION_NAME: &str = ".text.rvdctest";
/// Some instruction fields are reserved and not printed in the assembly,
/// but should be ignored instead of rejected by the decoder.
/// We filter out non-canonical forms of these instructions as they do not roundtrip.
fn is_inst_supposed_to_roundtrip(inst: &Inst) -> bool {
match inst {
// Canonical fence.tso
Inst::Fence {
fence:
Fence {
fm: 0b1000,
pred:
FenceSet {
device_input: false,
device_output: false,
memory_read: true,
memory_write: true,
},
succ:
FenceSet {
device_input: false,
device_output: false,
memory_read: true,
memory_write: true,
},
src: Reg::ZERO,
dest: Reg::ZERO,
},
} => true,
// Only canonical normal fence with x0 and x0
Inst::Fence {
fence:
Fence {
fm: 0b0000,
pred: _,
succ: _,
src: Reg::ZERO,
dest: Reg::ZERO,
},
} => true,
// All other fences are reserved
Inst::Fence { .. } => false,
_ => true,
}
}
fn is_compressed_inst_supposed_to_roundtrip(inst: &Inst) -> bool {
match inst {
// HINT
Inst::Addi {
dest: Reg::ZERO, ..
} => false,
// This does roundtrip, but only through C.MV, not C.ADDI
Inst::Addi { imm: Imm::ZERO, .. } => false,
// This does rountrip, but not through C.ADDI
Inst::Addi {
dest: Reg::SP,
src1: Reg::SP,
..
} => false,
// HINT
Inst::Slli {
dest: Reg::ZERO, ..
}
| Inst::Slli { imm: Imm::ZERO, .. }
| Inst::Srli {
dest: Reg::ZERO, ..
}
| Inst::Srli { imm: Imm::ZERO, .. }
| Inst::Srai {
dest: Reg::ZERO, ..
}
| Inst::Srai { imm: Imm::ZERO, .. } => false,
// HINT
Inst::Lui {
dest: Reg::ZERO, ..
} => false,
_ => true,
}
}
// turn this up for debugging, only run the test directly in these cases
const SKIP_CHUNKS: u32 = 0;
#[test]
fn ensure_no_chunks_are_skipped() {
assert_eq!(SKIP_CHUNKS, 0);
}
#[test]
#[cfg_attr(not(slow_tests), ignore = "cfg(slow_tests) not enabled")]
fn normal_clang_roundtrip() {
const CHUNKS: u32 = 128;
const CHUNK_SIZE: u32 = u32::MAX / CHUNKS;
for (chunk_idx, start) in ((SKIP_CHUNKS * CHUNK_SIZE)..u32::MAX)
.step_by(CHUNK_SIZE as usize)
.enumerate()
{
let chunk_idx = chunk_idx + SKIP_CHUNKS as usize;
let start_time = std::time::Instant::now();
let insts = (start..=start.saturating_add(CHUNK_SIZE))
.filter_map(|code| Some((code, Inst::decode_normal(code, Xlen::Rv32).ok()?)))
.filter(|(_, inst)| is_inst_supposed_to_roundtrip(inst))
.collect::<Vec<_>>();
let mut text = std::format!(".section {TEST_SECTION_NAME}\n.globl _start\n_start:\n");
for (_, inst) in &insts {
writeln!(text, " {inst}").unwrap();
}
let data = clang_assemble(&text, "-march=rv32ima_zihintpause");
for (i, result_code) in data.chunks(4).enumerate() {
let result_code = u32::from_le_bytes(result_code.try_into().unwrap());
assert_eq!(
insts[i].0, result_code,
"failed to rountrip!\n\
instruction `{:0>32b}` failed to rountrip\n\
resulted in `{:0>32b}` instead.\n\
disassembly of original instruction: `{}`",
insts[i].0, result_code, insts[i].1
);
}
let elapsed = start_time.elapsed();
let chunk_number = chunk_idx + 1;
let remaining = elapsed * (CHUNKS.saturating_sub(chunk_number as u32));
writeln!(
std::io::stdout(),
"Completed chunk {chunk_number}/{CHUNKS} (estimated {remaining:?} remaining)",
)
.unwrap();
}
}
#[test]
#[ignore = "this doesn't quite work yet because there is often a non-canonical encoding"]
fn compressed_clang_roundtrip() {
let insts = (0..=u16::MAX)
.filter_map(|code| Some((code, Inst::decode_compressed(code, Xlen::Rv32).ok()?)))
.filter(|(_, inst)| is_compressed_inst_supposed_to_roundtrip(inst))
.collect::<Vec<_>>();
let mut text = std::format!(".section {TEST_SECTION_NAME}\n.globl _start\n_start:\n");
for (_, inst) in &insts {
writeln!(text, " {inst}").unwrap();
}
let data = clang_assemble(&text, "-march=rv32imac");
for (i, result_code) in data.chunks(2).enumerate() {
assert!(
Inst::first_byte_is_compressed(result_code[0]),
"failed to roundtrip {i}th instruction!\n\
instruction `{:0>16b}` resulted in an uncompressed instruction from clang\n\
disassembly of original instruction: `{}`",
insts[i].0,
insts[i].1
);
let result_code = u16::from_le_bytes(result_code.try_into().unwrap());
assert_eq!(
insts[i].0, result_code,
"failed to rountrip {i}th instruction!\n\
instruction `{:0>16b}` failed to rountrip\n\
resulted in `{:0>16b}` instead.\n\
disassembly of original instruction: `{}`",
insts[i].0, result_code, insts[i].1
);
}
}
fn clang_assemble(text: &str, march_flag: &str) -> Vec<u8> {
let tmp = tempfile::tempdir().unwrap();
let path = tmp.path().join("16.s");
let bin_path = tmp.path().join("16.o");
std::fs::write(&path, text).unwrap();
let mut clang = std::process::Command::new("clang");
clang.args(["-target", "riscv32-unknown-none-elf", march_flag, "-c"]);
clang.arg(path);
clang.arg("-o");
clang.arg(&bin_path);
let out = clang.output().unwrap();
if !out.status.success() {
panic!(
"failed to run clang:\n{}",
String::from_utf8_lossy(&out.stderr)
);
}
let obj = std::fs::read(bin_path).unwrap();
let obj = object::File::parse(&*obj).unwrap();
let section = obj.section_by_name(TEST_SECTION_NAME).unwrap();
let data = section.data().unwrap();
data.to_owned()
}
}