holey-bytes/hbvm/src/vmrun.rs
2024-09-14 11:26:54 +02:00

555 lines
25 KiB
Rust

//! Welcome to the land of The Great Dispatch Loop
//!
//! Have fun
use {
super::{bmc::BlockCopier, mem::Memory, value::ValueVariant, Vm, VmRunError, VmRunOk},
crate::{
mem::{addr::AddressOp, Address},
value::CheckedDivRem,
},
core::{cmp::Ordering, ops},
hbbytecode::{
OpsN, OpsO, OpsP, OpsRB, OpsRD, OpsRH, OpsRR, OpsRRA, OpsRRAH, OpsRRB, OpsRRD, OpsRRH,
OpsRRO, OpsRROH, OpsRRP, OpsRRPH, OpsRRR, OpsRRRR, OpsRW, RoundingMode,
},
};
macro_rules! handler {
($self:expr, |$ty:ident ($($ident:pat),* $(,)?)| $expr:expr) => {{
#[allow(unused_unsafe)]
let $ty($($ident),*) = unsafe { $self.decode::<$ty>() };
#[allow(clippy::no_effect)] let e = $expr;
$self.bump_pc::<$ty>();
e
}};
}
impl<Mem, const TIMER_QUOTIENT: usize> Vm<Mem, TIMER_QUOTIENT>
where
Mem: Memory,
{
/// Execute program
///
/// Program can return [`VmRunError`] if a trap handling failed
#[cfg_attr(feature = "nightly", repr(align(4096)))]
pub fn run(&mut self) -> Result<VmRunOk, VmRunError> {
use hbbytecode::opcode::*;
loop {
// Big match
//
// Contribution guide:
// - Zero register shall never be overwitten. It's value has to always be 0.
// - Prefer `Self::read_reg` and `Self::write_reg` functions
// - Try to use `handler!` macro for decoding and then bumping program counter
// - Prioritise speed over code size
// - Memory is cheap, CPUs not that much
// - Do not heap allocate at any cost
// - Yes, user-provided trap handler may allocate,
// but that is not our »fault«.
// - Unsafe is kinda must, but be sure you have validated everything
// - Your contributions have to pass sanitizers, fuzzer and Miri
// - Strictly follow the spec
// - The spec does not specify how you perform actions, in what order,
// just that the observable effects have to be performed in order and
// correctly.
// - Yes, we assume you run 64 bit CPU. Else ?conradluget a better CPU
// sorry 8 bit fans, HBVM won't run on your Speccy :(
unsafe {
match self.memory.prog_read::<u8>(self.pc as _) {
UN => {
self.bump_pc::<OpsN>();
return Err(VmRunError::Unreachable);
}
TX => {
self.bump_pc::<OpsN>();
return Ok(VmRunOk::End);
}
NOP => handler!(self, |OpsN()| ()),
ADD8 => self.binary_op(u8::wrapping_add),
ADD16 => self.binary_op(u16::wrapping_add),
ADD32 => self.binary_op(u32::wrapping_add),
ADD64 => self.binary_op(u64::wrapping_add),
SUB8 => self.binary_op(u8::wrapping_sub),
SUB16 => self.binary_op(u16::wrapping_sub),
SUB32 => self.binary_op(u32::wrapping_sub),
SUB64 => self.binary_op(u64::wrapping_sub),
MUL8 => self.binary_op(u8::wrapping_mul),
MUL16 => self.binary_op(u16::wrapping_mul),
MUL32 => self.binary_op(u32::wrapping_mul),
MUL64 => self.binary_op(u64::wrapping_mul),
AND => self.binary_op::<u64>(ops::BitAnd::bitand),
OR => self.binary_op::<u64>(ops::BitOr::bitor),
XOR => self.binary_op::<u64>(ops::BitXor::bitxor),
SLU8 => self.binary_op_shift::<u8>(u8::wrapping_shl),
SLU16 => self.binary_op_shift::<u16>(u16::wrapping_shl),
SLU32 => self.binary_op_shift::<u32>(u32::wrapping_shl),
SLU64 => self.binary_op_shift::<u64>(u64::wrapping_shl),
SRU8 => self.binary_op_shift::<u8>(u8::wrapping_shr),
SRU16 => self.binary_op_shift::<u16>(u16::wrapping_shr),
SRU32 => self.binary_op_shift::<u32>(u32::wrapping_shr),
SRU64 => self.binary_op_shift::<u64>(u64::wrapping_shr),
SRS8 => self.binary_op_shift::<i8>(i8::wrapping_shr),
SRS16 => self.binary_op_shift::<i16>(i16::wrapping_shr),
SRS32 => self.binary_op_shift::<i32>(i32::wrapping_shr),
SRS64 => self.binary_op_shift::<i64>(i64::wrapping_shr),
CMPU => handler!(self, |OpsRRR(tg, a0, a1)| self.cmp(
tg,
a0,
self.read_reg(a1).cast::<u64>()
)),
CMPS => handler!(self, |OpsRRR(tg, a0, a1)| self.cmp(
tg,
a0,
self.read_reg(a1).cast::<i64>()
)),
DIRU8 => self.dir::<u8>(),
DIRU16 => self.dir::<u16>(),
DIRU32 => self.dir::<u32>(),
DIRU64 => self.dir::<u64>(),
DIRS8 => self.dir::<i8>(),
DIRS16 => self.dir::<i16>(),
DIRS32 => self.dir::<i32>(),
DIRS64 => self.dir::<i64>(),
NEG => handler!(self, |OpsRR(tg, a0)| {
// Bit negation
self.write_reg(tg, self.read_reg(a0).cast::<u64>().wrapping_neg())
}),
NOT => handler!(self, |OpsRR(tg, a0)| {
// Logical negation
self.write_reg(tg, u64::from(self.read_reg(a0).cast::<u64>() == 0));
}),
SXT8 => handler!(self, |OpsRR(tg, a0)| {
self.write_reg(tg, self.read_reg(a0).cast::<i8>() as i64)
}),
SXT16 => handler!(self, |OpsRR(tg, a0)| {
self.write_reg(tg, self.read_reg(a0).cast::<i16>() as i64)
}),
SXT32 => handler!(self, |OpsRR(tg, a0)| {
self.write_reg(tg, self.read_reg(a0).cast::<i32>() as i64)
}),
ADDI8 => self.binary_op_imm(u8::wrapping_add),
ADDI16 => self.binary_op_imm(u16::wrapping_add),
ADDI32 => self.binary_op_imm(u32::wrapping_add),
ADDI64 => self.binary_op_imm(u64::wrapping_add),
MULI8 => self.binary_op_imm(u8::wrapping_mul),
MULI16 => self.binary_op_imm(u16::wrapping_mul),
MULI32 => self.binary_op_imm(u32::wrapping_mul),
MULI64 => self.binary_op_imm(u64::wrapping_mul),
ANDI => self.binary_op_imm::<u64>(ops::BitAnd::bitand),
ORI => self.binary_op_imm::<u64>(ops::BitOr::bitor),
XORI => self.binary_op_imm::<u64>(ops::BitXor::bitxor),
SLUI8 => self.binary_op_ims::<u8>(u8::wrapping_shl),
SLUI16 => self.binary_op_ims::<u16>(u16::wrapping_shl),
SLUI32 => self.binary_op_ims::<u32>(u32::wrapping_shl),
SLUI64 => self.binary_op_ims::<u64>(u64::wrapping_shl),
SRUI8 => self.binary_op_ims::<u8>(u8::wrapping_shr),
SRUI16 => self.binary_op_ims::<u16>(u16::wrapping_shr),
SRUI32 => self.binary_op_ims::<u32>(u32::wrapping_shr),
SRUI64 => self.binary_op_ims::<u64>(u64::wrapping_shr),
SRSI8 => self.binary_op_ims::<i8>(i8::wrapping_shr),
SRSI16 => self.binary_op_ims::<i16>(i16::wrapping_shr),
SRSI32 => self.binary_op_ims::<i32>(i32::wrapping_shr),
SRSI64 => self.binary_op_ims::<i64>(i64::wrapping_shr),
CMPUI => handler!(self, |OpsRRD(tg, a0, imm)| { self.cmp(tg, a0, imm) }),
CMPSI => handler!(self, |OpsRRD(tg, a0, imm)| { self.cmp(tg, a0, imm as i64) }),
CP => handler!(self, |OpsRR(tg, a0)| self.write_reg(tg, self.read_reg(a0))),
SWA => handler!(self, |OpsRR(r0, r1)| {
// Swap registers
match (r0, r1) {
(0, 0) => (),
(dst, 0) | (0, dst) => self.write_reg(dst, 0_u64),
(r0, r1) => {
core::ptr::swap(
self.registers.get_unchecked_mut(usize::from(r0)),
self.registers.get_unchecked_mut(usize::from(r1)),
);
}
}
}),
LI8 => handler!(self, |OpsRB(tg, imm)| self.write_reg(tg, imm)),
LI16 => handler!(self, |OpsRH(tg, imm)| self.write_reg(tg, imm)),
LI32 => handler!(self, |OpsRW(tg, imm)| self.write_reg(tg, imm)),
LI64 => handler!(self, |OpsRD(tg, imm)| self.write_reg(tg, imm)),
LRA => handler!(self, |OpsRRO(tg, reg, off)| self.write_reg(
tg,
self.pcrel(off).wrapping_add(self.read_reg(reg).cast::<i64>()).get(),
)),
// Load. If loading more than register size, continue on adjecent registers
LD => handler!(self, |OpsRRAH(dst, base, off, count)| self
.load(dst, base, off, count)?),
// Store. Same rules apply as to LD
ST => handler!(self, |OpsRRAH(dst, base, off, count)| self
.store(dst, base, off, count)?),
LDR => handler!(self, |OpsRROH(dst, base, off, count)| self.load(
dst,
base,
self.pcrel(off).get(),
count
)?),
STR => handler!(self, |OpsRROH(dst, base, off, count)| self.store(
dst,
base,
self.pcrel(off).get(),
count
)?),
BMC => {
// Block memory copy
match if let Some(copier) = &mut self.copier {
// There is some copier, poll.
copier.poll(&mut self.memory)
} else {
// There is none, make one!
let OpsRRH(src, dst, count) = self.decode();
self.copier = Some(BlockCopier::new(
Address::new(self.read_reg(src).cast()),
Address::new(self.read_reg(dst).cast()),
count as _,
));
self.copier
.as_mut()
.unwrap_unchecked() // SAFETY: We just assigned there
.poll(&mut self.memory)
} {
// We are done, shift program counter
core::task::Poll::Ready(Ok(())) => {
self.copier = None;
self.bump_pc::<OpsRRH>();
}
// Error, shift program counter (for consistency)
// and yield error
core::task::Poll::Ready(Err(e)) => {
return Err(e.into());
}
// Not done yet, proceed to next cycle
core::task::Poll::Pending => (),
}
}
BRC => handler!(self, |OpsRRB(src, dst, count)| {
// Block register copy
if src.checked_add(count).is_none() || dst.checked_add(count).is_none() {
return Err(VmRunError::RegOutOfBounds);
}
core::ptr::copy(
self.registers.get_unchecked(usize::from(src)),
self.registers.get_unchecked_mut(usize::from(dst)),
usize::from(count),
);
}),
JMP => {
let OpsO(off) = self.decode();
self.pc = self.pc.wrapping_add(off);
}
JAL => {
// Jump and link. Save PC after this instruction to
// specified register and jump to reg + relative offset.
let OpsRRO(save, reg, offset) = self.decode();
self.write_reg(save, self.pc.next::<OpsRRO>());
self.pc = self.pcrel(offset).wrapping_add(self.read_reg(reg).cast::<i64>());
}
JALA => {
// Jump and link. Save PC after this instruction to
// specified register and jump to reg
let OpsRRA(save, reg, offset) = self.decode();
self.write_reg(save, self.pc.next::<OpsRRA>());
self.pc =
Address::new(self.read_reg(reg).cast::<u64>().wrapping_add(offset));
}
// Conditional jumps, jump only to immediates
JEQ => self.cond_jmp::<u64>(Ordering::Equal),
JNE => {
let OpsRRP(a0, a1, ja) = self.decode();
if self.read_reg(a0).cast::<u64>() != self.read_reg(a1).cast::<u64>() {
self.pc = self.pcrel(ja);
} else {
self.bump_pc::<OpsRRP>();
}
}
JLTS => self.cond_jmp::<i64>(Ordering::Less),
JGTS => self.cond_jmp::<i64>(Ordering::Greater),
JLTU => self.cond_jmp::<u64>(Ordering::Less),
JGTU => self.cond_jmp::<u64>(Ordering::Greater),
ECA => {
// So we don't get timer interrupt after ECALL
if TIMER_QUOTIENT != 0 {
self.timer = self.timer.wrapping_add(1);
}
self.bump_pc::<OpsN>();
return Ok(VmRunOk::Ecall);
}
EBP => {
self.bump_pc::<OpsN>();
return Ok(VmRunOk::Breakpoint);
}
FADD32 => self.binary_op::<f32>(ops::Add::add),
FADD64 => self.binary_op::<f64>(ops::Add::add),
FSUB32 => self.binary_op::<f32>(ops::Sub::sub),
FSUB64 => self.binary_op::<f64>(ops::Sub::sub),
FMUL32 => self.binary_op::<f32>(ops::Mul::mul),
FMUL64 => self.binary_op::<f64>(ops::Mul::mul),
FDIV32 => self.binary_op::<f32>(ops::Div::div),
FDIV64 => self.binary_op::<f64>(ops::Div::div),
FMA32 => self.fma::<f32>(),
FMA64 => self.fma::<f64>(),
FINV32 => handler!(self, |OpsRR(tg, reg)| self
.write_reg(tg, 1. / self.read_reg(reg).cast::<f32>())),
FINV64 => handler!(self, |OpsRR(tg, reg)| self
.write_reg(tg, 1. / self.read_reg(reg).cast::<f64>())),
FCMPLT32 => self.fcmp::<f32>(Ordering::Less),
FCMPLT64 => self.fcmp::<f64>(Ordering::Less),
FCMPGT32 => self.fcmp::<f32>(Ordering::Greater),
FCMPGT64 => self.fcmp::<f64>(Ordering::Greater),
ITF32 => handler!(self, |OpsRR(tg, reg)| self
.write_reg(tg, self.read_reg(reg).cast::<i64>() as f32)),
ITF64 => handler!(self, |OpsRR(tg, reg)| self
.write_reg(tg, self.read_reg(reg).cast::<i64>() as f64)),
FTI32 => handler!(self, |OpsRRB(tg, reg, mode)| self.write_reg(
tg,
crate::float::f32toint(
self.read_reg(reg).cast::<f32>(),
RoundingMode::try_from(mode)
.map_err(|()| VmRunError::InvalidOperand)?,
),
)),
FTI64 => handler!(self, |OpsRRB(tg, reg, mode)| self.write_reg(
tg,
crate::float::f64toint(
self.read_reg(reg).cast::<f64>(),
RoundingMode::try_from(mode)
.map_err(|()| VmRunError::InvalidOperand)?,
),
)),
FC32T64 => handler!(self, |OpsRR(tg, reg)| self
.write_reg(tg, self.read_reg(reg).cast::<f32>() as f64)),
FC64T32 => handler!(self, |OpsRRB(tg, reg, mode)| self.write_reg(
tg,
crate::float::conv64to32(
self.read_reg(reg).cast(),
RoundingMode::try_from(mode)
.map_err(|()| VmRunError::InvalidOperand)?,
),
)),
LRA16 => handler!(self, |OpsRRP(tg, reg, imm)| self.write_reg(
tg,
(self.pc + self.read_reg(reg).cast::<u64>() + imm + 3_u16).get(),
)),
LDR16 => handler!(self, |OpsRRPH(dst, base, off, count)| self.load(
dst,
base,
self.pcrel(off).get(),
count
)?),
STR16 => handler!(self, |OpsRRPH(dst, base, off, count)| self.store(
dst,
base,
self.pcrel(off).get(),
count
)?),
JMP16 => {
let OpsP(off) = self.decode();
self.pc = self.pcrel(off);
}
op => return Err(VmRunError::InvalidOpcode(op)),
}
}
if TIMER_QUOTIENT != 0 {
self.timer = self.timer.wrapping_add(1);
if self.timer % TIMER_QUOTIENT == 0 {
return Ok(VmRunOk::Timer);
}
}
}
}
/// Bump instruction pointer
#[inline(always)]
fn bump_pc<T: Copy>(&mut self) {
self.pc = self.pc.wrapping_add(core::mem::size_of::<T>()).wrapping_add(1);
}
/// Decode instruction operands
#[inline(always)]
unsafe fn decode<T: Copy>(&mut self) -> T {
unsafe { self.memory.prog_read::<T>(self.pc + 1_u64) }
}
/// Load
#[inline(always)]
unsafe fn load(
&mut self,
dst: u8,
base: u8,
offset: u64,
count: u16,
) -> Result<(), VmRunError> {
let n: u8 = match dst {
0 => 1,
_ => 0,
};
unsafe {
self.memory.load(
self.ldst_addr_uber(dst, base, offset, count, n)?,
self.registers.as_mut_ptr().add(usize::from(dst) + usize::from(n)).cast(),
usize::from(count).saturating_sub(n.into()),
)
}?;
Ok(())
}
/// Store
#[inline(always)]
unsafe fn store(
&mut self,
dst: u8,
base: u8,
offset: u64,
count: u16,
) -> Result<(), VmRunError> {
unsafe {
self.memory.store(
self.ldst_addr_uber(dst, base, offset, count, 0)?,
self.registers.as_ptr().add(usize::from(dst)).cast(),
count.into(),
)
}?;
Ok(())
}
/// Three-way comparsion
#[inline(always)]
unsafe fn cmp<T: ValueVariant + Ord>(&mut self, to: u8, reg: u8, val: T) {
self.write_reg(to, self.read_reg(reg).cast::<T>().cmp(&val) as i64);
}
/// Perform binary operating over two registers
#[inline(always)]
unsafe fn binary_op<T: ValueVariant>(&mut self, op: impl Fn(T, T) -> T) {
let OpsRRR(tg, a0, a1) = unsafe { self.decode() };
self.write_reg(tg, op(self.read_reg(a0).cast::<T>(), self.read_reg(a1).cast::<T>()));
self.bump_pc::<OpsRRR>();
}
/// Perform binary operation over register and immediate
#[inline(always)]
unsafe fn binary_op_imm<T: ValueVariant>(&mut self, op: impl Fn(T, T) -> T) {
#[derive(Clone, Copy)]
#[repr(packed)]
struct OpsRRImm<I>(OpsRR, I);
let OpsRRImm::<T>(OpsRR(tg, reg), imm) = unsafe { self.decode() };
self.write_reg(tg, op(self.read_reg(reg).cast::<T>(), imm));
self.bump_pc::<OpsRRImm<T>>();
}
/// Perform binary operation over register and shift immediate
#[inline(always)]
unsafe fn binary_op_shift<T: ValueVariant>(&mut self, op: impl Fn(T, u32) -> T) {
let OpsRRR(tg, a0, a1) = unsafe { self.decode() };
self.write_reg(tg, op(self.read_reg(a0).cast::<T>(), self.read_reg(a1).cast::<u32>()));
self.bump_pc::<OpsRRR>();
}
/// Perform binary operation over register and shift immediate
#[inline(always)]
unsafe fn binary_op_ims<T: ValueVariant>(&mut self, op: impl Fn(T, u32) -> T) {
let OpsRRB(tg, reg, imm) = unsafe { self.decode() };
self.write_reg(tg, op(self.read_reg(reg).cast::<T>(), imm.into()));
self.bump_pc::<OpsRRB>();
}
/// Fused division-remainder
#[inline(always)]
unsafe fn dir<T: ValueVariant + CheckedDivRem>(&mut self) {
handler!(self, |OpsRRRR(td, tr, a0, a1)| {
let a0 = self.read_reg(a0).cast::<T>();
let a1 = self.read_reg(a1).cast::<T>();
if let Some(div) = a0.checked_div(a1) {
self.write_reg(td, div);
} else {
self.write_reg(td, -1_i64);
}
if let Some(rem) = a0.checked_rem(a1) {
self.write_reg(tr, rem);
} else {
self.write_reg(tr, a0);
}
});
}
/// Fused multiply-add
#[inline(always)]
unsafe fn fma<T>(&mut self)
where
T: ValueVariant + core::ops::Mul<Output = T> + core::ops::Add<Output = T>,
{
handler!(self, |OpsRRRR(tg, a0, a1, a2)| {
let a0 = self.read_reg(a0).cast::<T>();
let a1 = self.read_reg(a1).cast::<T>();
let a2 = self.read_reg(a2).cast::<T>();
self.write_reg(tg, a0 * a1 + a2)
});
}
/// Float comparsion
#[inline(always)]
unsafe fn fcmp<T: PartialOrd + ValueVariant>(&mut self, nan: Ordering) {
handler!(self, |OpsRRR(tg, a0, a1)| {
let a0 = self.read_reg(a0).cast::<T>();
let a1 = self.read_reg(a1).cast::<T>();
self.write_reg(tg, (a0.partial_cmp(&a1).unwrap_or(nan) as i8 + 1) as u8)
});
}
/// Calculate pc-relative address
#[inline(always)]
fn pcrel(&self, offset: impl AddressOp) -> Address {
self.pc.wrapping_add(offset)
}
/// Jump at `PC + #3` if ordering on `#0 <=> #1` is equal to expected
#[inline(always)]
unsafe fn cond_jmp<T: ValueVariant + Ord>(&mut self, expected: Ordering) {
let OpsRRP(a0, a1, ja) = unsafe { self.decode() };
if self.read_reg(a0).cast::<T>().cmp(&self.read_reg(a1).cast::<T>()) == expected {
self.pc = self.pcrel(ja);
} else {
self.bump_pc::<OpsRRP>();
}
}
/// Load / Store Address check-computation überfunction
#[inline(always)]
unsafe fn ldst_addr_uber(
&self,
dst: u8,
base: u8,
offset: u64,
size: u16,
adder: u8,
) -> Result<Address, VmRunError> {
let reg = dst.checked_add(adder).ok_or(VmRunError::RegOutOfBounds)?;
if usize::from(reg) * 8 + usize::from(size) > 2048 {
Err(VmRunError::RegOutOfBounds)
} else {
self.read_reg(base)
.cast::<u64>()
.checked_add(offset)
.and_then(|x| x.checked_add(adder.into()))
.ok_or(VmRunError::AddrOutOfBounds)
.map(Address::new)
}
}
}