optimizing the bitset used in register allocation

also fixing an enum bug

Signed-off-by: Jakub Doka <jakub.doka2@gmail.com>

lang/src/backend/hbvm.rs

@@ -8,7 +8,7 @@ use {
         utils::{EntSlice, EntVec},
     },
     alloc::{boxed::Box, collections::BTreeMap, string::String, vec::Vec},
-    core::{assert_matches::debug_assert_matches, mem, ops::Range},
+    core::{assert_matches::debug_assert_matches, mem, ops::Range, usize},
     hbbytecode::{self as instrs, *},
     reg::Reg,
 };
@@ -606,9 +606,8 @@ impl TokenKind {
         Some(ops[size.ilog2() as usize])
     }
 
-    fn unop(&self, dst: ty::Id, src: ty::Id) -> Option<fn(u8, u8) -> EncodedInstr> {
-        let src_idx =
-            src.simple_size().unwrap_or_else(|| panic!("{:?}", src.expand())).ilog2() as usize;
+    fn unop(&self, dst: ty::Id, src: ty::Id, tys: &Types) -> Option<fn(u8, u8) -> EncodedInstr> {
+        let src_idx = tys.size_of(src).ilog2() as usize;
         Some(match self {
             Self::Sub => [
                 |a, b| sub8(a, reg::ZERO, b),
@@ -654,6 +653,13 @@ enum PLoc {
     WideReg(Reg, u16),
     Ref(Reg, u32),
 }
+impl PLoc {
+    fn reg(self) -> u8 {
+        match self {
+            PLoc::Reg(r, _) | PLoc::WideReg(r, _) | PLoc::Ref(r, _) => r,
+        }
+    }
+}
 
 struct ParamAlloc(Range<Reg>);
 

lang/src/backend/hbvm/regalloc.rs

@@ -1,4 +1,5 @@
 use {
+    super::ParamAlloc,
     crate::{
         backend::hbvm::{
             reg::{self, Reg},
@@ -32,7 +33,9 @@ impl HbvmBackend {
 
         let mut res = mem::take(&mut self.ralloc);
 
-        Regalloc::run(nodes, tys, &mut res);
+        let special_reg_count = 13u8;
+
+        Regalloc::run(nodes, tys, &mut res, special_reg_count as _);
 
         '_open_function: {
             self.emit(instrs::addi64(reg::STACK_PTR, reg::STACK_PTR, 0));
@@ -40,12 +43,12 @@ impl HbvmBackend {
         }
 
         if let Some(PLoc::Ref(..)) = tys.parama(sig.ret).0 {
-            res.node_to_reg[MEM as usize] = res.bundles.len() as u8 + 1;
-            res.bundles.push(Bundle::new(0));
+            res.node_to_reg[MEM as usize] = res.general_bundles.len() as u8 + 1;
+            res.general_bundles.push(Bundle::default());
         }
 
-        let reg_offset = if tail { reg::RET + 12 } else { reg::RET_ADDR + 1 };
-        let bundle_count = res.bundles.len() + (reg_offset as usize);
+        let reg_offset = if tail { special_reg_count } else { reg::RET_ADDR + 1 };
+        let bundle_count = res.general_bundles.len() + (reg_offset as usize);
 
         res.node_to_reg.iter_mut().filter(|r| **r != 0).for_each(|r| {
             if *r == u8::MAX {
@@ -325,6 +328,7 @@ impl HbvmBackend {
                                 node.ty,
                                 tys.inner_of(nodes[node.inputs[1]].ty)
                                     .unwrap_or(nodes[node.inputs[1]].ty),
+                                tys,
                             )
                             .unwrap_or_else(|| {
                                 panic!(
@@ -521,7 +525,7 @@ impl HbvmBackend {
             if tail {
                 bundle_count.saturating_sub(reg::RET_ADDR as _)
             } else {
-                self.ralloc.bundles.len()
+                self.ralloc.general_bundles.len()
             },
             tail,
         )
@@ -770,7 +774,13 @@ impl Nodes {
         nid
     }
 
-    fn uses_of(&self, nid: Nid, types: &Types, stack: &mut Vec<Nid>, buf: &mut Vec<(Nid, Nid)>) {
+    fn uses_of(
+        &self,
+        nid: Nid,
+        types: &Types,
+        stack: &mut Vec<Nid>,
+        buf: &mut Vec<(Nid, Nid, Reg)>,
+    ) {
         debug_assert!(stack.is_empty());
         debug_assert!(buf.is_empty());
 
@@ -786,13 +796,36 @@ impl Nodes {
                     continue;
                 }
                 if self.is_unlocked(o) {
-                    buf.push((self.use_block_of(exp, o), o));
+                    buf.push((self.use_block_of(exp, o), o, self.use_reg_of(exp, o)));
                 } else {
                     stack.push(o);
                 }
             }
         }
     }
+
+    fn init_loc_of(&self, def: Nid, types: &Types) -> Reg {
+        if self[def].kind == Kind::Arg {
+            let mut parama = ParamAlloc(0..11);
+            let (_, ploc) = self[VOID]
+                .outputs
+                .iter()
+                .skip(ARG_START)
+                .map(|&n| (n, parama.next(self[n].ty, types)))
+                .find(|&(n, _)| n == def)
+                .unwrap();
+
+            return ploc.unwrap().reg();
+        }
+
+        255
+    }
+
+    fn use_reg_of(&self, def: Nid, usage: Nid) -> Reg {
+        if matches!(self[usage].kind, Kind::Return { .. }) {}
+
+        255
+    }
 }
 
 struct Regalloc<'a> {
@@ -815,27 +848,35 @@ impl<'a> Regalloc<'a> {
         self.res.backrefs[nid as usize]
     }
 
-    fn run(ctx: &'a Nodes, tys: &'a Types, res: &'a mut Res) {
-        Self { nodes: ctx, tys, res }.run_low();
+    fn run(ctx: &'a Nodes, tys: &'a Types, res: &'a mut Res, special_count: usize) {
+        Self { nodes: ctx, tys, res }.run_low(special_count);
     }
 
-    fn run_low(&mut self) {
-        self.res.bundles.clear();
+    fn run_low(&mut self, special_count: usize) {
+        self.res.general_bundles.clear();
         self.res.node_to_reg.clear();
         #[cfg(debug_assertions)]
         self.res.marked.clear();
         self.res.node_to_reg.resize(self.nodes.vreg_count(), 0);
+        self.res.call_set.clear();
+
+        for (i, &instr) in self.res.instrs.iter().enumerate() {
+            if self.nodes[instr].kind.is_call() {
+                self.res.call_set.add_one(i);
+            }
+        }
 
         debug_assert!(self.res.dfs_buf.is_empty());
 
         let mut uses_buf = Vec::new();
-        let mut bundle = Bundle::new(self.res.instrs.len());
+        let mut range_buf = Vec::new();
+        let mut bundle = Bundle::default();
         self.res.visited.clear(self.nodes.len());
 
         for i in (0..self.res.blocks.len()).rev() {
             for [a, rest @ ..] in self.nodes.phi_inputs_of(self.res.blocks[i].entry) {
                 if self.res.visited.set(a) {
-                    self.append_bundle(a, &mut bundle, None, &mut uses_buf);
+                    self.append_bundle(a, &mut bundle, None, &mut uses_buf, &mut range_buf);
                 }
 
                 for r in rest {
@@ -848,6 +889,7 @@ impl<'a> Regalloc<'a> {
                         &mut bundle,
                         Some(self.res.node_to_reg[a as usize] as usize - 1),
                         &mut uses_buf,
+                        &mut range_buf,
                     );
                 }
             }
@@ -858,7 +900,7 @@ impl<'a> Regalloc<'a> {
             if self.nodes[inst].has_no_value() || self.res.visited.get(inst) || inst == 0 {
                 continue;
             }
-            self.append_bundle(inst, &mut bundle, None, &mut uses_buf);
+            self.append_bundle(inst, &mut bundle, None, &mut uses_buf, &mut range_buf);
         }
         self.res.instrs = instrs;
     }
@@ -868,12 +910,16 @@ impl<'a> Regalloc<'a> {
         inst: Nid,
         tmp: &mut Bundle,
         prefered: Option<usize>,
-        uses_buf: &mut Vec<(Nid, Nid)>,
+        uses_buf: &mut Vec<(Nid, Nid, Reg)>,
+        range_buf: &mut Vec<Range<usize>>,
     ) {
         let dom = self.nodes.idom_of(inst);
         self.res.dfs_seem.clear(self.nodes.len());
         self.nodes.uses_of(inst, self.tys, &mut self.res.dfs_buf, uses_buf);
-        for (cursor, uinst) in uses_buf.drain(..) {
+        let mut prefered_reg = reg::ZERO;
+        for (cursor, uinst, reg) in uses_buf.drain(..) {
+            prefered_reg = prefered_reg.min(reg);
+
             if !self.res.dfs_seem.set(uinst) {
                 continue;
             }
@@ -903,8 +949,22 @@ impl<'a> Regalloc<'a> {
                 range.end = new;
                 debug_assert!(range.start < range.end, "{:?} {inst} {uinst}", range);
 
-                tmp.add(range);
+                range_buf.push(range)
             });
+
+            range_buf.sort_unstable_by_key(|r| r.start);
+            range_buf.dedup_by(|a, b| {
+                if b.end == a.start {
+                    b.end = a.end;
+                    true
+                } else {
+                    false
+                }
+            });
+
+            for range in range_buf.drain(..) {
+                tmp.add(range);
+            }
         }
 
         if tmp.is_empty() {
@@ -913,23 +973,23 @@ impl<'a> Regalloc<'a> {
         }
 
         if let Some(prefered) = prefered
-            && !self.res.bundles[prefered].overlaps(tmp)
+            && !self.res.general_bundles[prefered].overlaps(tmp)
         {
-            self.res.bundles[prefered].merge(tmp);
+            self.res.general_bundles[prefered].merge(tmp);
             tmp.clear();
             self.res.node_to_reg[inst as usize] = prefered as Reg + 1;
             return;
         }
 
-        match self.res.bundles.iter_mut().enumerate().find(|(_, b)| !b.overlaps(tmp)) {
+        match self.res.general_bundles.iter_mut().enumerate().find(|(_, b)| !b.overlaps(tmp)) {
             Some((i, other)) => {
                 other.merge(tmp);
                 tmp.clear();
                 self.res.node_to_reg[inst as usize] = i as Reg + 1;
             }
             None => {
-                self.res.bundles.push(tmp.take());
-                self.res.node_to_reg[inst as usize] = self.res.bundles.len() as Reg;
+                self.res.general_bundles.push(tmp.take());
+                self.res.node_to_reg[inst as usize] = self.res.general_bundles.len() as Reg;
             }
         }
     }
@@ -980,7 +1040,8 @@ pub(super) struct Res {
     instrs: Vec<Nid>,
     backrefs: Vec<u16>,
 
-    bundles: Vec<Bundle>,
+    general_bundles: Vec<Bundle>,
+    call_set: Bundle,
     node_to_reg: Vec<Reg>,
 
     visited: BitSet,
@@ -991,37 +1052,83 @@ pub(super) struct Res {
 }
 
 struct Bundle {
-    taken: Vec<bool>,
+    start: usize,
+    end: usize,
+    usage: BitSet,
+}
+
+impl Default for Bundle {
+    fn default() -> Self {
+        Self { start: usize::MAX, end: 0, usage: Default::default() }
+    }
 }
 
 impl Bundle {
-    fn new(size: usize) -> Self {
-        Self { taken: vec![false; size] }
-    }
-
     fn add(&mut self, range: Range<usize>) {
-        self.taken[range].fill(true);
+        debug_assert!(!range.is_empty());
+        debug_assert!(range.start / BitSet::UNIT >= self.start || self.start == usize::MAX);
+        self.start = self.start.min(range.start / BitSet::UNIT);
+        self.end = self.end.max(range.end.div_ceil(BitSet::UNIT));
+        let proj_range =
+            range.start - self.start * BitSet::UNIT..range.end - self.start * BitSet::UNIT;
+        self.usage.set_range(proj_range)
     }
 
-    fn overlaps(&self, other: &Self) -> bool {
-        self.taken.iter().zip(other.taken.iter()).any(|(a, b)| a & b)
+    fn overlaps(&self, othr: &Self) -> bool {
+        let overlap = self.start.max(othr.start)..self.end.min(othr.end);
+
+        if overlap.start >= overlap.end {
+            return false;
+        }
+
+        let [mut sslot, mut oslot] = [0, 0];
+        let sunits =
+            &self.usage.units(&mut sslot)[overlap.start - self.start..overlap.end - self.start];
+        let ounits =
+            &othr.usage.units(&mut oslot)[overlap.start - othr.start..overlap.end - othr.start];
+
+        debug_assert_eq!(sunits.len(), ounits.len());
+
+        let res = sunits.iter().zip(ounits).any(|(a, b)| (a & b) != 0);
+        res
     }
 
-    fn merge(&mut self, other: &Self) {
-        debug_assert!(!self.overlaps(other));
-        self.taken.iter_mut().zip(other.taken.iter()).for_each(|(a, b)| *a |= *b);
+    fn merge(&mut self, othr: &Self) {
+        debug_assert!(!self.overlaps(othr));
+        debug_assert!(self.start <= othr.start || self.start == usize::MAX);
+
+        self.usage.reserve((othr.end - self.start) * BitSet::UNIT);
+        self.start = self.start.min(othr.start);
+        self.end = self.end.max(othr.end);
+
+        let sunits =
+            &mut self.usage.units_mut().unwrap()[othr.start - self.start..othr.end - self.start];
+        let mut oslot = 0;
+        let ounits = othr.usage.units(&mut oslot);
+        sunits.iter_mut().zip(ounits).for_each(|(a, b)| *a |= *b);
     }
 
     fn clear(&mut self) {
-        self.taken.fill(false);
+        self.start = usize::MAX;
+        self.end = 0;
+        self.usage.clear_as_is();
     }
 
     fn is_empty(&self) -> bool {
-        !self.taken.contains(&true)
+        self.end == 0
     }
 
     fn take(&mut self) -> Self {
-        mem::replace(self, Self::new(self.taken.len()))
+        let mut new = Self { start: 0, ..Self::default() };
+        new.merge(self);
+        self.clear();
+        new
+    }
+
+    fn add_one(&mut self, i: usize) {
+        self.start = self.start.min(i / BitSet::UNIT);
+        self.end = self.end.max(i.div_ceil(BitSet::UNIT));
+        self.usage.set(i as _);
     }
 }
 

lang/src/lib.rs

@@ -24,7 +24,8 @@
     iter_next_chunk,
     pointer_is_aligned_to,
     maybe_uninit_fill,
-    array_chunks
+    array_chunks,
+    array_windows
 )]
 #![warn(clippy::dbg_macro)]
 #![expect(internal_features)]

lang/src/ty.rs

@@ -184,11 +184,13 @@ impl Id {
     }
 
     pub fn is_unsigned(self) -> bool {
-        matches!(self.repr(), U8..=UINT) || self.is_never()
+        matches!(self.repr(), U8..=UINT)
+            || self.is_never()
+            || matches!(self.expand(), Kind::Enum(_))
     }
 
     pub fn is_integer(self) -> bool {
-        matches!(self.repr(), U8..=INT) || self.is_never()
+        self.is_signed() || self.is_unsigned()
     }
 
     pub fn is_never(self) -> bool {

lang/src/utils.rs

@@ -1,4 +1,3 @@
-#![expect(dead_code)]
 use {
     alloc::alloc,
     core::{
@@ -7,7 +6,7 @@ use {
         hint::unreachable_unchecked,
         marker::PhantomData,
         mem::MaybeUninit,
-        ops::{Deref, DerefMut, Not},
+        ops::{Deref, DerefMut, Not, Range},
         ptr::Unique,
     },
 };
@@ -32,9 +31,10 @@ pub fn is_screaming_case(str: &str) -> Result<(), &'static str> {
 }
 
 type Nid = u16;
+type BitSetUnit = usize;
 
 pub union BitSet {
-    inline: usize,
+    inline: BitSetUnit,
     alloced: Unique<AllocedBitSet>,
 }
 
@@ -78,9 +78,9 @@ impl Default for BitSet {
 }
 
 impl BitSet {
-    const FLAG: usize = 1 << (Self::UNIT - 1);
+    const FLAG: BitSetUnit = 1 << (Self::UNIT - 1);
     const INLINE_ELEMS: usize = Self::UNIT - 1;
-    const UNIT: usize = core::mem::size_of::<usize>() * 8;
+    pub const UNIT: usize = core::mem::size_of::<BitSetUnit>() * 8;
 
     pub fn with_capacity(len: usize) -> Self {
         let mut s = Self::default();
@@ -92,7 +92,7 @@ impl BitSet {
         unsafe { self.inline & Self::FLAG != 0 }
     }
 
-    fn data_and_len(&self) -> (&[usize], usize) {
+    fn data_and_len(&self) -> (&[BitSetUnit], usize) {
         unsafe {
             if self.is_inline() {
                 (core::slice::from_ref(&self.inline), Self::INLINE_ELEMS)
@@ -100,16 +100,16 @@ impl BitSet {
                 let small_vec = self.alloced.as_ref();
                 (
                     core::slice::from_raw_parts(
-                        &small_vec.data as *const _ as *const usize,
+                        &small_vec.data as *const _ as *const BitSetUnit,
                         small_vec.cap,
                     ),
-                    small_vec.cap * core::mem::size_of::<usize>() * 8,
+                    small_vec.cap * Self::UNIT,
                 )
             }
         }
     }
 
-    fn data_mut_and_len(&mut self) -> (&mut [usize], usize) {
+    fn data_mut_and_len(&mut self) -> (&mut [BitSetUnit], usize) {
         unsafe {
             if self.is_inline() {
                 (core::slice::from_mut(&mut self.inline), INLINE_ELEMS)
@@ -117,7 +117,7 @@ impl BitSet {
                 let small_vec = self.alloced.as_mut();
                 (
                     core::slice::from_raw_parts_mut(
-                        &mut small_vec.data as *mut _ as *mut usize,
+                        &mut small_vec.data as *mut _ as *mut BitSetUnit,
                         small_vec.cap,
                     ),
                     small_vec.cap * Self::UNIT,
@@ -163,7 +163,7 @@ impl BitSet {
         let (ptr, prev_len) = unsafe {
             if self.is_inline() {
                 let ptr = alloc::alloc(layout);
-                *ptr.add(off).cast::<usize>() = self.inline & !Self::FLAG;
+                *ptr.add(off).cast::<BitSetUnit>() = self.inline & !Self::FLAG;
                 (ptr, 1)
             } else {
                 let prev_len = self.alloced.as_ref().cap;
@@ -174,7 +174,7 @@ impl BitSet {
         unsafe {
             MaybeUninit::fill(
                 core::slice::from_raw_parts_mut(
-                    ptr.add(off).cast::<MaybeUninit<usize>>().add(prev_len),
+                    ptr.add(off).cast::<MaybeUninit<BitSetUnit>>().add(prev_len),
                     slot_count - prev_len,
                 ),
                 0,
@@ -187,7 +187,7 @@ impl BitSet {
     fn layout(slot_count: usize) -> (core::alloc::Layout, usize) {
         unsafe {
             core::alloc::Layout::new::<AllocedBitSet>()
-                .extend(Layout::array::<usize>(slot_count).unwrap_unchecked())
+                .extend(Layout::array::<BitSetUnit>(slot_count).unwrap_unchecked())
                 .unwrap_unchecked()
         }
     }
@@ -205,6 +205,10 @@ impl BitSet {
 
     pub fn clear(&mut self, len: usize) {
         self.reserve(len);
+        self.clear_as_is();
+    }
+
+    pub fn clear_as_is(&mut self) {
         if self.is_inline() {
             unsafe { self.inline &= Self::FLAG };
         } else {
@@ -212,7 +216,11 @@ impl BitSet {
         }
     }
 
-    pub fn units<'a>(&'a self, slot: &'a mut usize) -> &'a [usize] {
+    pub fn approx_unit_cap(&self) -> usize {
+        self.data_and_len().0.len()
+    }
+
+    pub fn units<'a>(&'a self, slot: &'a mut BitSetUnit) -> &'a [BitSetUnit] {
         if self.is_inline() {
             *slot = unsafe { self.inline } & !Self::FLAG;
             core::slice::from_ref(slot)
@@ -221,36 +229,47 @@ impl BitSet {
         }
     }
 
+    pub fn units_mut(&mut self) -> Option<&mut [BitSetUnit]> {
+        self.is_inline().not().then(|| self.data_mut_and_len().0)
+    }
+
     pub fn reserve(&mut self, len: usize) {
         if len > self.data_and_len().1 {
             self.grow(len.next_power_of_two().max(4 * Self::UNIT));
         }
     }
 
-    pub fn units_mut(&mut self) -> Result<&mut [usize], &mut InlineBitSetView> {
-        if self.is_inline() {
-            Err(unsafe {
-                core::mem::transmute::<&mut usize, &mut InlineBitSetView>(&mut self.inline)
-            })
-        } else {
-            Ok(self.data_mut_and_len().0)
+    pub fn set_range(&mut self, proj_range: Range<usize>) {
+        if proj_range.is_empty() {
+            return;
         }
-    }
-}
 
-pub struct InlineBitSetView(usize);
+        self.reserve(proj_range.end);
+        let (units, _) = self.data_mut_and_len();
 
-impl InlineBitSetView {
-    pub(crate) fn add_mask(&mut self, tmp: usize) {
-        debug_assert!(tmp & BitSet::FLAG == 0);
-        self.0 |= tmp;
+        if proj_range.start / Self::UNIT == (proj_range.end - 1) / Self::UNIT {
+            debug_assert!(proj_range.len() <= Self::UNIT);
+            let mask = ((1 << proj_range.len()) - 1) << (proj_range.start % Self::UNIT);
+            units[proj_range.start / Self::UNIT] |= mask;
+        } else {
+            let fill_range = proj_range.start.div_ceil(Self::UNIT)..proj_range.end / Self::UNIT;
+            units[fill_range].fill(BitSetUnit::MAX);
+
+            let prefix_len = Self::UNIT - proj_range.start % Self::UNIT;
+            let prefix_mask = ((1 << prefix_len) - 1) << (proj_range.start % Self::UNIT);
+            units[proj_range.start / Self::UNIT] |= prefix_mask;
+
+            let postfix_len = proj_range.end % Self::UNIT;
+            let postfix_mask = (1 << postfix_len) - 1;
+            units[proj_range.end / Self::UNIT] |= postfix_mask;
+        }
     }
 }
 
 pub struct BitSetIter<'a> {
     index: usize,
-    current: usize,
-    remining: &'a [usize],
+    current: BitSetUnit,
+    remining: &'a [BitSetUnit],
 }
 
 impl Iterator for BitSetIter<'_> {
@@ -270,7 +289,7 @@ impl Iterator for BitSetIter<'_> {
 
 struct AllocedBitSet {
     cap: usize,
-    data: [usize; 0],
+    data: [BitSetUnit; 0],
 }
 
 #[cfg(test)]