WIP adding next_pc

crates/cpu/src/config.rs

@@ -34,6 +34,8 @@ pub struct CpuConfig {
     pub units: Vec<UnitConfig>,
     pub out_reg_num_width: usize,
     pub fetch_width: NonZeroUsize,
+    pub max_branches_per_fetch: NonZeroUsize,
+    pub log2_fetch_width_in_bytes: u8,
     /// default value for [`UnitConfig::max_in_flight`]
     pub default_unit_max_in_flight: NonZeroUsize,
     pub rob_size: NonZeroUsize,
@@ -47,6 +49,13 @@ impl CpuConfig {
         };
         v
     };
+    pub const DEFAULT_MAX_BRANCHES_PER_FETCH: NonZeroUsize = {
+        let Some(v) = NonZeroUsize::new(1) else {
+            unreachable!();
+        };
+        v
+    };
+    pub const DEFAULT_LOG2_FETCH_WIDTH_IN_BYTES: u8 = 3;
     pub const DEFAULT_UNIT_MAX_IN_FLIGHT: NonZeroUsize = {
         let Some(v) = NonZeroUsize::new(8) else {
             unreachable!();
@@ -58,6 +67,8 @@ impl CpuConfig {
             units,
             out_reg_num_width: Self::DEFAULT_OUT_REG_NUM_WIDTH,
             fetch_width: Self::DEFAULT_FETCH_WIDTH,
+            max_branches_per_fetch: Self::DEFAULT_MAX_BRANCHES_PER_FETCH,
+            log2_fetch_width_in_bytes: Self::DEFAULT_LOG2_FETCH_WIDTH_IN_BYTES,
             default_unit_max_in_flight: Self::DEFAULT_UNIT_MAX_IN_FLIGHT,
             rob_size,
         }
@@ -117,4 +128,21 @@ impl CpuConfig {
         UnitToRegAlloc[mop_ty][extra_out_ty][self.unit_num_width()][self.out_reg_num_width]
             [self.non_const_unit_nums().len()]
     }
+    pub fn fetch_width_in_bytes(&self) -> usize {
+        1usize
+            .checked_shl(self.log2_fetch_width_in_bytes.into())
+            .expect("log2_fetch_width_in_bytes is too big")
+    }
 }
+
+#[hdl(get(|c| c.fetch_width.get()))]
+pub type CpuConfigFetchWidth<C: PhantomConstGet<CpuConfig>> = DynSize;
+
+#[hdl(get(|c| c.max_branches_per_fetch.get()))]
+pub type CpuConfigMaxBranchesPerFetch<C: PhantomConstGet<CpuConfig>> = DynSize;
+
+#[hdl(get(|c| c.log2_fetch_width_in_bytes.into()))]
+pub type CpuConfigLog2FetchWidthInBytes<C: PhantomConstGet<CpuConfig>> = DynSize;
+
+#[hdl(get(|c| c.fetch_width_in_bytes()))]
+pub type CpuConfigFetchWidthInBytes<C: PhantomConstGet<CpuConfig>> = DynSize;

crates/cpu/src/lib.rs

@@ -2,6 +2,7 @@
 // See Notices.txt for copyright information
 pub mod config;
 pub mod instruction;
+pub mod next_pc;
 pub mod reg_alloc;
 pub mod register;
 pub mod unit;

crates/cpu/src/next_pc.rs

@@ -0,0 +1,728 @@
+// SPDX-License-Identifier: LGPL-3.0-or-later
+// See Notices.txt for copyright information
+
+//! [Next-Instruction Logic](https://git.libre-chip.org/libre-chip/grant-tracking/issues/10)
+//!
+//! The basic idea here is that there's a `next_pc` stage that sends predicted fetch PCs to the `fetch` stage,
+//! the `fetch` stage's outputs eventually end up in the `decode` stage,
+//! after the `decode` stage there's a `post_decode` stage (that may run in the same clock cycle as `decode`)
+//! that checks that the fetched instructions' kinds match the predicted instruction kinds and that feeds
+//! information back to the `fetch` stage to cancel fetches that need to be predicted differently.
+
+use crate::{config::CpuConfig, util::array_vec::ArrayVec};
+use fayalite::{
+    int::{UIntInRange, UIntInRangeInclusive, UIntInRangeType},
+    prelude::*,
+    sim::{ForkJoinScope, value::SimOnlyValueTrait},
+    util::ready_valid::ReadyValid,
+};
+
+#[hdl]
+pub enum PredictedCond {
+    Taken,
+    Fallthrough,
+}
+
+#[hdl]
+pub struct PredictedFallthrough {}
+
+#[hdl]
+pub enum BranchPredictionKind<CondKind> {
+    Branch(HdlOption<CondKind>),
+    IndirectBranch(HdlOption<CondKind>),
+    Call(HdlOption<CondKind>),
+    IndirectCall(HdlOption<CondKind>),
+    Ret(HdlOption<CondKind>),
+}
+
+#[hdl(get(|c| c.max_branches_per_fetch.get() - 1))]
+pub type NextPcPredictionMaxBranchesBeforeLast<C: PhantomConstGet<CpuConfig>> = DynSize;
+
+#[hdl(no_static)]
+pub struct NextPcPrediction<C: PhantomConstGet<CpuConfig>> {
+    pub fetch_pc: UInt<64>,
+    pub async_interrupt: Bool,
+    pub branches_before_last: ArrayVec<
+        BranchPredictionKind<PredictedFallthrough>,
+        NextPcPredictionMaxBranchesBeforeLast<C>,
+    >,
+    pub last_branch: HdlOption<BranchPredictionKind<PredictedCond>>,
+    pub last_branch_target_pc: UInt<64>,
+}
+
+pub const FETCH_BLOCK_ID_WIDTH: usize = FetchBlockIdInt::BITS as usize;
+type FetchBlockIdInt = u8;
+
+#[hdl]
+pub struct NextPcToFetchInterfaceInner {
+    pub next_fetch_pc: UInt<64>,
+    pub fetch_block_id: UInt<{ FETCH_BLOCK_ID_WIDTH }>,
+    pub in_progress_fetches_to_cancel: UInt<8>,
+}
+
+#[hdl(no_static)]
+pub struct NextPcToFetchInterface<C: PhantomConstGet<CpuConfig>> {
+    pub inner: ReadyValid<NextPcToFetchInterfaceInner>,
+    pub config: C,
+}
+
+#[hdl]
+/// WIP version of decoded instruction just good enough to represent stuff needed for [`next_pc()`] since the actual instruction definition isn't finalized yet. This will be replaced at a later point.
+pub enum WipDecodedInsnKind {
+    NonBranch,
+    Branch(UInt<64>),
+    BranchCond(UInt<64>),
+    IndirectBranch,
+    IndirectBranchCond,
+    Call(UInt<64>),
+    CallCond(UInt<64>),
+    IndirectCall,
+    IndirectCallCond,
+    Ret,
+    RetCond,
+    /// not actually an instruction read from memory, covers stuff like external interrupts, page faults, memory errors, and so on.
+    Interrupt(UInt<64>),
+}
+
+#[hdl]
+/// WIP version of decoded instruction just good enough to represent stuff needed for [`next_pc()`] since the actual instruction definition isn't finalized yet. This will be replaced at a later point.
+pub struct WipDecodedInsn {
+    pub fetch_block_id: UInt<8>,
+    pub id: UInt<12>,
+    pub pc: UInt<64>,
+    pub kind: WipDecodedInsnKind,
+}
+
+#[hdl(no_static)]
+/// handles updating speculative branch predictor state (e.g. branch histories) when instructions retire,
+/// as well as updating state when a branch instruction is mis-speculated.
+pub struct NextPcToRetireInterface<C: PhantomConstGet<CpuConfig>> {
+    // TODO: add needed fields
+    pub config: C,
+}
+
+#[hdl(no_static)]
+pub struct DecodeToPostDecodeInterface<C: PhantomConstGet<CpuConfig>> {
+    // TODO: add needed fields
+    pub config: C,
+}
+
+#[hdl(no_static)]
+pub struct PostDecodeOutputInterface<C: PhantomConstGet<CpuConfig>> {
+    // TODO: add needed fields
+    pub config: C,
+}
+
+#[hdl]
+enum BranchPredictionState {
+    StronglyNotTaken,
+    WeaklyNotTaken,
+    WeaklyTaken,
+    StronglyTaken,
+}
+
+impl BranchPredictionState {
+    #[must_use]
+    #[hdl]
+    fn is_taken(this: &SimValue<Self>) -> bool {
+        #[hdl(sim)]
+        match this {
+            Self::StronglyNotTaken => false,
+            Self::WeaklyNotTaken => false,
+            Self::WeaklyTaken => true,
+            Self::StronglyTaken => true,
+        }
+    }
+    #[must_use]
+    #[hdl]
+    fn towards_taken(this: &SimValue<Self>) -> SimValue<Self> {
+        (#[hdl(sim)]
+        match this {
+            Self::StronglyNotTaken => BranchPredictionState.WeaklyNotTaken(),
+            Self::WeaklyNotTaken => BranchPredictionState.WeaklyTaken(),
+            Self::WeaklyTaken => BranchPredictionState.StronglyTaken(),
+            Self::StronglyTaken => BranchPredictionState.StronglyTaken(),
+        })
+        .to_sim_value()
+    }
+    #[must_use]
+    #[hdl]
+    fn towards_not_taken(this: &SimValue<Self>) -> SimValue<Self> {
+        (#[hdl(sim)]
+        match this {
+            Self::StronglyNotTaken => BranchPredictionState.StronglyNotTaken(),
+            Self::WeaklyNotTaken => BranchPredictionState.StronglyNotTaken(),
+            Self::WeaklyTaken => BranchPredictionState.WeaklyNotTaken(),
+            Self::StronglyTaken => BranchPredictionState.WeaklyTaken(),
+        })
+        .to_sim_value()
+    }
+}
+
+impl SimValueDefault for BranchPredictionState {
+    fn sim_value_default(self) -> SimValue<Self> {
+        self.WeaklyNotTaken().to_sim_value()
+    }
+}
+
+#[derive(Copy, Clone, Debug)]
+#[must_use]
+enum ResetStatus {
+    Done,
+    Working,
+}
+
+impl ResetStatus {
+    fn and(self, other: Self) -> Self {
+        match (self, other) {
+            (ResetStatus::Done, ResetStatus::Done) => ResetStatus::Done,
+            (ResetStatus::Done | ResetStatus::Working, ResetStatus::Working)
+            | (ResetStatus::Working, ResetStatus::Done) => ResetStatus::Working,
+        }
+    }
+}
+
+trait SimValueDefault: Type {
+    fn sim_value_default(self) -> SimValue<Self>;
+}
+
+impl<T: SimOnlyValueTrait> SimValueDefault for SimOnly<T> {
+    fn sim_value_default(self) -> SimValue<Self> {
+        SimOnlyValue::<T>::default().to_sim_value_with_type(self)
+    }
+}
+
+impl<T: Type> SimValueDefault for HdlOption<T> {
+    fn sim_value_default(self) -> SimValue<Self> {
+        self.HdlNone().to_sim_value_with_type(self)
+    }
+}
+
+impl SimValueDefault for Bool {
+    fn sim_value_default(self) -> SimValue<Self> {
+        false.to_sim_value()
+    }
+}
+
+impl<Width: Size> SimValueDefault for UIntType<Width> {
+    fn sim_value_default(self) -> SimValue<Self> {
+        self.zero().to_sim_value()
+    }
+}
+
+trait ResetSteps: Type {
+    fn reset_step(this: &mut SimValue<Self>, step: usize) -> ResetStatus;
+}
+
+impl<T: SimValueDefault, Len: Size> ResetSteps for ArrayType<T, Len> {
+    fn reset_step(this: &mut SimValue<Self>, step: usize) -> ResetStatus {
+        let element = SimValue::ty(this).element();
+        let len = SimValue::ty(this).len();
+        if step < len {
+            this[step] = element.sim_value_default();
+        }
+        if step.saturating_add(1) >= len {
+            ResetStatus::Done
+        } else {
+            ResetStatus::Working
+        }
+    }
+}
+
+#[hdl]
+struct CallStack {
+    return_addresses: Array<UInt<64>, { CallStack::SIZE }>,
+    len: UIntInRangeInclusive<0, { CallStack::SIZE }>,
+}
+
+impl CallStack {
+    const SIZE: usize = 16;
+}
+
+impl SimValueDefault for CallStack {
+    #[hdl]
+    fn sim_value_default(self) -> SimValue<Self> {
+        #[hdl(sim)]
+        CallStack {
+            // something other than zero so you can see the values getting reset
+            return_addresses: [!0u64; Self::SIZE],
+            len: 0usize.to_sim_value_with_type(self.len),
+        }
+    }
+}
+
+impl ResetSteps for CallStack {
+    #[hdl]
+    fn reset_step(this: &mut SimValue<Self>, _step: usize) -> ResetStatus {
+        #[hdl(sim)]
+        let CallStack {
+            return_addresses,
+            len,
+        } = this;
+        // return_addresses is implemented as a shift register, so it can be all reset at once
+        return_addresses.fill(0u64.to_sim_value());
+        **len = 0;
+        ResetStatus::Done
+    }
+}
+
+#[hdl]
+struct BranchTargetBuffer {
+    branch_pc_to_target_map: Array<HdlOption<(UInt<64>, UInt<64>)>, { BranchTargetBuffer::SIZE }>,
+}
+
+impl BranchTargetBuffer {
+    const SIZE: usize = 16;
+}
+
+impl SimValueDefault for BranchTargetBuffer {
+    #[hdl]
+    fn sim_value_default(self) -> SimValue<Self> {
+        #[hdl(sim)]
+        BranchTargetBuffer {
+            // something other than zero so you can see the values getting reset
+            branch_pc_to_target_map: [HdlSome((0u64, 0u64)); Self::SIZE],
+        }
+    }
+}
+
+impl ResetSteps for BranchTargetBuffer {
+    #[hdl]
+    fn reset_step(this: &mut SimValue<Self>, step: usize) -> ResetStatus {
+        #[hdl(sim)]
+        let BranchTargetBuffer {
+            branch_pc_to_target_map,
+        } = this;
+        ResetSteps::reset_step(branch_pc_to_target_map, step)
+    }
+}
+
+#[hdl]
+struct BranchHistory {
+    history: Array<Bool, { BranchHistory::SIZE }>,
+    /// exclusive
+    tail: UIntInRange<0, { BranchHistory::SIZE }>,
+    /// inclusive, always at or after tail, always at or before speculative_head
+    non_speculative_head: UIntInRange<0, { BranchHistory::SIZE }>,
+    /// inclusive, always at or after both tail and non_speculative_head
+    speculative_head: UIntInRange<0, { BranchHistory::SIZE }>,
+}
+
+impl ResetSteps for BranchHistory {
+    #[hdl]
+    fn reset_step(this: &mut SimValue<Self>, step: usize) -> ResetStatus {
+        #[hdl(sim)]
+        let Self {
+            history,
+            tail,
+            non_speculative_head,
+            speculative_head,
+        } = this;
+        **tail = 0;
+        **non_speculative_head = 0;
+        **speculative_head = 0;
+        ResetSteps::reset_step(history, step)
+    }
+}
+
+impl SimValueDefault for BranchHistory {
+    #[hdl]
+    fn sim_value_default(self) -> SimValue<Self> {
+        #[hdl(sim)]
+        BranchHistory {
+            // something other than zero so you can see the values getting reset
+            history: [true; Self::SIZE],
+            tail: 0usize.to_sim_value_with_type(self.tail),
+            non_speculative_head: 0usize.to_sim_value_with_type(self.non_speculative_head),
+            speculative_head: 0usize.to_sim_value_with_type(self.speculative_head),
+        }
+    }
+}
+
+enum BranchHistoryTryPushSpeculativeError {
+    NoSpace,
+}
+
+enum BranchHistoryTryPushNonSpeculativeError {
+    NoSpace,
+    Misprediction { speculated: bool },
+}
+
+impl BranchHistory {
+    const LOG2_SIZE: usize = 8;
+    const SIZE: usize = 1 << Self::LOG2_SIZE;
+    fn next_pos(pos: usize) -> usize {
+        (pos + 1) % Self::SIZE
+    }
+    fn prev_pos(pos: usize) -> usize {
+        (pos + Self::SIZE - 1) % Self::SIZE
+    }
+    fn history_from_head<const N: usize>(this: &SimValue<Self>, head: usize) -> [bool; N] {
+        let mut retval = [false; N];
+        let mut pos = head;
+        for entry in &mut retval {
+            if pos == *this.tail {
+                break;
+            }
+            *entry = *this.history[pos];
+            pos = Self::prev_pos(pos);
+        }
+        retval
+    }
+    fn delete_speculative_history(this: &mut SimValue<Self>) {
+        let non_speculative_head = *this.non_speculative_head;
+        *this.speculative_head = non_speculative_head;
+    }
+    fn recent_history_including_speculative<const N: usize>(this: &SimValue<Self>) -> [bool; N] {
+        let head = *this.speculative_head;
+        Self::history_from_head(this, head)
+    }
+    fn speculative_full(this: &SimValue<Self>) -> bool {
+        let speculative_head = *this.speculative_head;
+        Self::next_pos(speculative_head) == *this.tail
+    }
+    fn try_push_speculative(
+        this: &mut SimValue<Self>,
+        value: bool,
+    ) -> Result<(), BranchHistoryTryPushSpeculativeError> {
+        if Self::speculative_full(this) {
+            Err(BranchHistoryTryPushSpeculativeError::NoSpace)
+        } else {
+            let speculative_head = Self::next_pos(*this.speculative_head);
+            *this.speculative_head = speculative_head;
+            *this.history[speculative_head] = value;
+            Ok(())
+        }
+    }
+    fn try_push_non_speculative(
+        this: &mut SimValue<Self>,
+        value: bool,
+    ) -> Result<(), BranchHistoryTryPushNonSpeculativeError> {
+        let speculative_head = *this.speculative_head;
+        let non_speculative_head = *this.non_speculative_head;
+        if speculative_head == non_speculative_head {
+            Err(BranchHistoryTryPushNonSpeculativeError::NoSpace)
+        } else {
+            let pos = Self::next_pos(non_speculative_head);
+            let speculated = *this.history[pos];
+            if speculated != value {
+                Err(BranchHistoryTryPushNonSpeculativeError::Misprediction { speculated })
+            } else {
+                *this.non_speculative_head = pos;
+                Ok(())
+            }
+        }
+    }
+}
+
+#[hdl]
+struct Queue<T, Capacity: Size> {
+    data: ArrayType<T, Capacity>,
+    /// inclusive
+    head: UIntInRangeType<ConstUsize<0>, Capacity>,
+    /// exclusive
+    tail: UIntInRangeType<ConstUsize<0>, Capacity>,
+}
+
+impl<T: Type, Capacity: Size> Queue<T, Capacity> {
+    fn capacity(self) -> usize {
+        self.data.len()
+    }
+    fn next_pos(self, pos: usize) -> usize {
+        assert_ne!(self.capacity(), 0);
+        (pos + 1) % self.capacity()
+    }
+    fn prev_pos(self, pos: usize) -> usize {
+        assert_ne!(self.capacity(), 0);
+        (pos + self.capacity() - 1) % self.capacity()
+    }
+    fn is_empty(this: &SimValue<Self>) -> bool {
+        this.head == this.tail
+    }
+    fn is_full(this: &SimValue<Self>) -> bool {
+        let head = *this.head;
+        let tail = *this.tail;
+        SimValue::ty(this).next_pos(head) == tail
+    }
+    fn try_push(this: &mut SimValue<Self>, value: impl ToSimValueWithType<T>) -> Result<(), ()> {
+        if Self::is_full(this) {
+            Err(())
+        } else {
+            let head = *this.head;
+            let head = SimValue::ty(this).next_pos(head);
+            *this.head = head;
+            let data = &mut this.data[head];
+            *data = value.to_sim_value_with_type(SimValue::ty(data));
+            Ok(())
+        }
+    }
+}
+
+impl<T: SimValueDefault, Capacity: Size> SimValueDefault for Queue<T, Capacity> {
+    #[hdl]
+    fn sim_value_default(self) -> SimValue<Self> {
+        let Self { data, head, tail } = self;
+        #[hdl(sim)]
+        Queue::<T, Capacity> {
+            data: repeat(
+                data.element().sim_value_default(),
+                Capacity::from_usize(data.len()),
+            ),
+            head: 0usize.to_sim_value_with_type(head),
+            tail: 0usize.to_sim_value_with_type(tail),
+        }
+    }
+}
+
+impl<T: SimValueDefault, Capacity: Size> ResetSteps for Queue<T, Capacity> {
+    #[hdl]
+    fn reset_step(this: &mut SimValue<Self>, step: usize) -> ResetStatus {
+        #[hdl(sim)]
+        let Queue::<T, Capacity> { data, head, tail } = this;
+        **head = 0;
+        **tail = 0;
+        ResetSteps::reset_step(data, step)
+    }
+}
+
+#[hdl]
+struct FetchQueueEntry {
+    fetch_block_id: UInt<{ FETCH_BLOCK_ID_WIDTH }>,
+}
+
+impl SimValueDefault for FetchQueueEntry {
+    #[hdl]
+    fn sim_value_default(self) -> SimValue<Self> {
+        #[hdl(sim)]
+        FetchQueueEntry {
+            fetch_block_id: 0 as FetchBlockIdInt,
+        }
+    }
+}
+
+const BRANCH_PREDICTOR_LOG2_SIZE: usize = 8;
+const BRANCH_PREDICTOR_SIZE: usize = 1 << BRANCH_PREDICTOR_LOG2_SIZE;
+
+#[hdl]
+pub struct NextPcState<C: PhantomConstGet<CpuConfig>> {
+    speculative_call_stack: CallStack,
+    non_speculative_call_stack: CallStack,
+    branch_target_buffer: BranchTargetBuffer,
+    branch_history: BranchHistory,
+    branch_predictor: Array<BranchPredictionState, { BRANCH_PREDICTOR_SIZE }>,
+    fetching_queue: Queue<FetchQueueEntry, ConstUsize<{ 1 << FETCH_BLOCK_ID_WIDTH }>>,
+    pc: UInt<64>,
+    fetch_block_id: UInt<{ FETCH_BLOCK_ID_WIDTH }>,
+    config: C,
+}
+
+impl<C: Type + PhantomConstGet<CpuConfig>> NextPcState<C> {
+    fn next_fetch_pc(this: &SimValue<Self>) -> u64 {
+        let pc = u64::try_from(this.pc.to_bigint()).expect("in range");
+        pc & (!0u64 << SimValue::ty(&this.config).get().log2_fetch_width_in_bytes)
+    }
+    fn branch_predictor_index(this: &SimValue<Self>, pc: u64) -> usize {
+        let mut history = 0u64;
+        let history_bits: [bool; BRANCH_PREDICTOR_LOG2_SIZE] =
+            BranchHistory::recent_history_including_speculative(&this.branch_history);
+        for history_bit in history_bits {
+            history <<= 1;
+            if history_bit {
+                history |= 1;
+            }
+        }
+        let mut t = history;
+        t ^= t.rotate_left(5) & !pc.rotate_right(3);
+        t ^= pc;
+        t ^= !t.rotate_left(2) & t.rotate_left(4);
+        let mut retval = 0;
+        for i in (0..BRANCH_PREDICTOR_LOG2_SIZE).step_by(BRANCH_PREDICTOR_LOG2_SIZE) {
+            retval ^= t >> i;
+        }
+        retval as usize % BRANCH_PREDICTOR_SIZE
+    }
+}
+
+impl SimValueDefault for NextPcState<PhantomConst<CpuConfig>> {
+    #[hdl]
+    fn sim_value_default(self) -> SimValue<Self> {
+        let Self {
+            speculative_call_stack,
+            non_speculative_call_stack,
+            branch_target_buffer,
+            branch_history,
+            branch_predictor: _,
+            fetching_queue,
+            pc: _,
+            fetch_block_id: _,
+            config,
+        } = self;
+        #[hdl(sim)]
+        Self {
+            speculative_call_stack: speculative_call_stack.sim_value_default(),
+            non_speculative_call_stack: non_speculative_call_stack.sim_value_default(),
+            branch_target_buffer: branch_target_buffer.sim_value_default(),
+            branch_history: branch_history.sim_value_default(),
+            // use something other than the default so you can see the reset progress
+            branch_predictor: std::array::from_fn(|_| {
+                BranchPredictionState::towards_not_taken(&BranchPredictionState.sim_value_default())
+            }),
+            fetching_queue: fetching_queue.sim_value_default(),
+            // use something other than the default so you can see the reset progress
+            pc: !0u64,
+            // use something other than the default so you can see the reset progress
+            fetch_block_id: !0u8,
+            config,
+        }
+    }
+}
+
+impl<C: Type + PhantomConstGet<CpuConfig>> ResetSteps for NextPcState<C> {
+    #[hdl]
+    fn reset_step(this: &mut SimValue<Self>, step: usize) -> ResetStatus {
+        #[hdl(sim)]
+        let NextPcState::<C> {
+            speculative_call_stack,
+            non_speculative_call_stack,
+            branch_target_buffer,
+            branch_history,
+            branch_predictor,
+            fetching_queue,
+            pc,
+            fetch_block_id,
+            config: _,
+        } = this;
+        **pc = 0u64.into(); // match Microwatt's reset PC
+        **fetch_block_id = 0u8.into();
+        let speculative_call_stack = ResetSteps::reset_step(speculative_call_stack, step);
+        let non_speculative_call_stack = ResetSteps::reset_step(non_speculative_call_stack, step);
+        let branch_target_buffer = ResetSteps::reset_step(branch_target_buffer, step);
+        let branch_history = ResetSteps::reset_step(branch_history, step);
+        let branch_predictor = ResetSteps::reset_step(branch_predictor, step);
+        let fetching_queue = ResetSteps::reset_step(fetching_queue, step);
+        speculative_call_stack
+            .and(non_speculative_call_stack)
+            .and(branch_target_buffer)
+            .and(branch_history)
+            .and(branch_predictor)
+            .and(fetching_queue)
+    }
+}
+
+#[hdl_module(extern)]
+pub fn next_pc(config: PhantomConst<CpuConfig>) {
+    #[hdl]
+    let cd: ClockDomain = m.input();
+    #[hdl]
+    let to_fetch: NextPcToFetchInterface<PhantomConst<CpuConfig>> =
+        m.output(NextPcToFetchInterface[config]);
+    #[hdl]
+    let state_for_debug: NextPcState<PhantomConst<CpuConfig>> = m.output(NextPcState[config]);
+    m.register_clock_for_past(cd.clk);
+    #[hdl]
+    async fn run(
+        scope: ForkJoinScope<'_>,
+        mut sim: ExternModuleSimulationState,
+        cd: Expr<ClockDomain>,
+        to_fetch: Expr<NextPcToFetchInterface<PhantomConst<CpuConfig>>>,
+        state_expr: Expr<NextPcState<PhantomConst<CpuConfig>>>,
+    ) {
+        let config = Expr::ty(state_expr.config);
+        let mut state = sim.read(state_expr).await;
+        for step in 0usize.. {
+            sim.write(state_expr, state).await;
+            sim.wait_for_clock_edge(cd.clk).await;
+            state = sim.read_past(state_expr, cd.clk).await;
+            let reset_status = ResetSteps::reset_step(&mut state, step);
+            match reset_status {
+                ResetStatus::Done => break,
+                ResetStatus::Working => {}
+            }
+        }
+        scope.spawn_detached(|_, mut sim: ExternModuleSimulationState| async move {
+            loop {
+                let state = sim.read(state_expr).await;
+                if Queue::is_full(&state.fetching_queue) {
+                    sim.write(to_fetch.inner.data, HdlNone()).await;
+                } else {
+                    sim.write(
+                        to_fetch.inner.data,
+                        HdlSome(
+                            #[hdl(sim)]
+                            NextPcToFetchInterfaceInner {
+                                next_fetch_pc: NextPcState::next_fetch_pc(&state),
+                                fetch_block_id: state.fetch_block_id,
+                                in_progress_fetches_to_cancel: 0u8, // TODO: implement
+                            },
+                        ),
+                    )
+                    .await;
+                }
+                sim.wait_for_changes([state_expr], None).await;
+            }
+        });
+        loop {
+            sim.write(state_expr, state).await;
+            sim.wait_for_clock_edge(cd.clk).await;
+            state = sim.read_past(state_expr, cd.clk).await;
+            let next_fetch_pc = NextPcState::next_fetch_pc(&state);
+            if Queue::is_full(&state.fetching_queue) {
+                continue;
+            }
+            if sim.read_past_bool(to_fetch.inner.ready, cd.clk).await {
+                let fetch_block_id =
+                    FetchBlockIdInt::try_from(state.fetch_block_id.to_bigint()).expect("in range");
+                Queue::try_push(
+                    &mut state.fetching_queue,
+                    #[hdl(sim)]
+                    FetchQueueEntry { fetch_block_id },
+                )
+                .expect("checked is_full above");
+                // TODO: handle instructions not aligned with fetch blocks
+                let mut new_pc =
+                    next_fetch_pc.wrapping_add(config.get().fetch_width_in_bytes() as u64);
+                for entry in &state.branch_target_buffer.branch_pc_to_target_map {
+                    #[hdl(sim)]
+                    match entry {
+                        HdlNone => continue,
+                        HdlSome(entry) => {
+                            #[hdl(sim)]
+                            let (entry_pc, entry_target) = entry;
+                            if *entry_pc == state.pc {
+                                new_pc = entry_target
+                                    .to_bigint()
+                                    .try_into()
+                                    .expect("known to be in range");
+                                break;
+                            }
+                        }
+                    }
+                }
+                *state.pc = new_pc.into();
+                *state.fetch_block_id = fetch_block_id.wrapping_add(1).into();
+            }
+        }
+        // TODO: finish
+    }
+    m.extern_module_simulation_fn(
+        (cd, to_fetch, state_for_debug),
+        |(cd, to_fetch, state_for_debug), mut sim| async move {
+            sim.write(
+                state_for_debug,
+                Expr::ty(state_for_debug).sim_value_default(),
+            )
+            .await;
+            sim.resettable(
+                cd,
+                |mut sim: ExternModuleSimulationState| async move {
+                    sim.write(to_fetch.inner.data, HdlNone()).await;
+                },
+                |mut sim: ExternModuleSimulationState, ()| async move {
+                    sim.fork_join_scope(|scope, sim| run(scope, sim, cd, to_fetch, state_for_debug))
+                        .await
+                },
+            )
+            .await;
+        },
+    );
+}

crates/cpu/tests/next_pc.rs

@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: LGPL-3.0-or-later
+// See Notices.txt for copyright information
+
+use cpu::{
+    config::{CpuConfig, UnitConfig},
+    next_pc::next_pc,
+    unit::UnitKind,
+};
+use fayalite::{prelude::*, sim::vcd::VcdWriterDecls, util::RcWriter};
+use std::num::NonZeroUsize;
+
+#[hdl]
+#[test]
+fn test_next_pc() {
+    let _n = SourceLocation::normalize_files_for_tests();
+    let mut config = CpuConfig::new(
+        vec![
+            UnitConfig::new(UnitKind::AluBranch),
+            UnitConfig::new(UnitKind::AluBranch),
+        ],
+        NonZeroUsize::new(20).unwrap(),
+    );
+    config.fetch_width = NonZeroUsize::new(2).unwrap();
+    let m = next_pc(PhantomConst::new_sized(config));
+    let mut sim = Simulation::new(m);
+    let mut writer = RcWriter::default();
+    sim.add_trace_writer(VcdWriterDecls::new(writer.clone()));
+    let to_fetch = sim.io().to_fetch;
+    sim.write_clock(sim.io().cd.clk, false);
+    sim.write_reset(sim.io().cd.rst, true);
+    sim.write_bool(to_fetch.inner.ready, true);
+    for _cycle in 0..300 {
+        sim.advance_time(SimDuration::from_nanos(500));
+        sim.write_clock(sim.io().cd.clk, true);
+        sim.advance_time(SimDuration::from_nanos(500));
+        sim.write_clock(sim.io().cd.clk, false);
+        sim.write_reset(sim.io().cd.rst, false);
+    }
+    // FIXME: vcd is just whatever next_pc does now, which isn't known to be correct
+    let vcd = String::from_utf8(writer.take()).unwrap();
+    println!("####### VCD:\n{vcd}\n#######");
+    if vcd != include_str!("expected/next_pc.vcd") {
+        panic!();
+    }
+}