1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
use crate::rmeta::*;

use rustc_index::vec::Idx;
use rustc_serialize::opaque::Encoder;
use rustc_serialize::Encoder as _;
use std::convert::TryInto;
use std::marker::PhantomData;
use std::num::NonZeroUsize;
use tracing::debug;

/// Helper trait, for encoding to, and decoding from, a fixed number of bytes.
/// Used mainly for Lazy positions and lengths.
/// Unchecked invariant: `Self::default()` should encode as `[0; BYTE_LEN]`,
/// but this has no impact on safety.
pub(super) trait FixedSizeEncoding: Default {
    const BYTE_LEN: usize;

    // FIXME(eddyb) convert to and from `[u8; Self::BYTE_LEN]` instead,
    // once that starts being allowed by the compiler (i.e. lazy normalization).
    fn from_bytes(b: &[u8]) -> Self;
    fn write_to_bytes(self, b: &mut [u8]);

    // FIXME(eddyb) make these generic functions, or at least defaults here.
    // (same problem as above, needs `[u8; Self::BYTE_LEN]`)
    // For now, a macro (`fixed_size_encoding_byte_len_and_defaults`) is used.

    /// Read a `Self` value (encoded as `Self::BYTE_LEN` bytes),
    /// from `&b[i * Self::BYTE_LEN..]`, returning `None` if `i`
    /// is not in bounds, or `Some(Self::from_bytes(...))` otherwise.
    fn maybe_read_from_bytes_at(b: &[u8], i: usize) -> Option<Self>;
    /// Write a `Self` value (encoded as `Self::BYTE_LEN` bytes),
    /// at `&mut b[i * Self::BYTE_LEN..]`, using `Self::write_to_bytes`.
    fn write_to_bytes_at(self, b: &mut [u8], i: usize);
}

// HACK(eddyb) this shouldn't be needed (see comments on the methods above).
macro_rules! fixed_size_encoding_byte_len_and_defaults {
    ($byte_len:expr) => {
        const BYTE_LEN: usize = $byte_len;
        fn maybe_read_from_bytes_at(b: &[u8], i: usize) -> Option<Self> {
            const BYTE_LEN: usize = $byte_len;
            // HACK(eddyb) ideally this would be done with fully safe code,
            // but slicing `[u8]` with `i * N..` is optimized worse, due to the
            // possibility of `i * N` overflowing, than indexing `[[u8; N]]`.
            let b = unsafe {
                std::slice::from_raw_parts(b.as_ptr() as *const [u8; BYTE_LEN], b.len() / BYTE_LEN)
            };
            b.get(i).map(|b| FixedSizeEncoding::from_bytes(b))
        }
        fn write_to_bytes_at(self, b: &mut [u8], i: usize) {
            const BYTE_LEN: usize = $byte_len;
            // HACK(eddyb) ideally this would be done with fully safe code,
            // see similar comment in `read_from_bytes_at` for why it can't yet.
            let b = unsafe {
                std::slice::from_raw_parts_mut(
                    b.as_mut_ptr() as *mut [u8; BYTE_LEN],
                    b.len() / BYTE_LEN,
                )
            };
            self.write_to_bytes(&mut b[i]);
        }
    };
}

impl FixedSizeEncoding for u32 {
    fixed_size_encoding_byte_len_and_defaults!(4);

    fn from_bytes(b: &[u8]) -> Self {
        let mut bytes = [0; Self::BYTE_LEN];
        bytes.copy_from_slice(&b[..Self::BYTE_LEN]);
        Self::from_le_bytes(bytes)
    }

    fn write_to_bytes(self, b: &mut [u8]) {
        b[..Self::BYTE_LEN].copy_from_slice(&self.to_le_bytes());
    }
}

// NOTE(eddyb) there could be an impl for `usize`, which would enable a more
// generic `Lazy<T>` impl, but in the general case we might not need / want to
// fit every `usize` in `u32`.
impl<T> FixedSizeEncoding for Option<Lazy<T>> {
    fixed_size_encoding_byte_len_and_defaults!(u32::BYTE_LEN);

    fn from_bytes(b: &[u8]) -> Self {
        Some(Lazy::from_position(NonZeroUsize::new(u32::from_bytes(b) as usize)?))
    }

    fn write_to_bytes(self, b: &mut [u8]) {
        let position = self.map_or(0, |lazy| lazy.position.get());
        let position: u32 = position.try_into().unwrap();

        position.write_to_bytes(b)
    }
}

impl<T> FixedSizeEncoding for Option<Lazy<[T]>> {
    fixed_size_encoding_byte_len_and_defaults!(u32::BYTE_LEN * 2);

    fn from_bytes(b: &[u8]) -> Self {
        Some(Lazy::from_position_and_meta(
            <Option<Lazy<T>>>::from_bytes(b)?.position,
            u32::from_bytes(&b[u32::BYTE_LEN..]) as usize,
        ))
    }

    fn write_to_bytes(self, b: &mut [u8]) {
        self.map(|lazy| Lazy::<T>::from_position(lazy.position)).write_to_bytes(b);

        let len = self.map_or(0, |lazy| lazy.meta);
        let len: u32 = len.try_into().unwrap();

        len.write_to_bytes(&mut b[u32::BYTE_LEN..]);
    }
}

/// Random-access table (i.e. offering constant-time `get`/`set`), similar to
/// `Vec<Option<T>>`, but without requiring encoding or decoding all the values
/// eagerly and in-order.
/// A total of `(max_idx + 1) * <Option<T> as FixedSizeEncoding>::BYTE_LEN` bytes
/// are used for a table, where `max_idx` is the largest index passed to
/// `TableBuilder::set`.
pub(super) struct Table<I: Idx, T>
where
    Option<T>: FixedSizeEncoding,
{
    _marker: PhantomData<(fn(&I), T)>,
    // NOTE(eddyb) this makes `Table` not implement `Sized`, but no
    // value of `Table` is ever created (it's always behind `Lazy`).
    _bytes: [u8],
}

/// Helper for constructing a table's serialization (also see `Table`).
pub(super) struct TableBuilder<I: Idx, T>
where
    Option<T>: FixedSizeEncoding,
{
    // FIXME(eddyb) use `IndexVec<I, [u8; <Option<T>>::BYTE_LEN]>` instead of
    // `Vec<u8>`, once that starts working (i.e. lazy normalization).
    // Then again, that has the downside of not allowing `TableBuilder::encode` to
    // obtain a `&[u8]` entirely in safe code, for writing the bytes out.
    bytes: Vec<u8>,
    _marker: PhantomData<(fn(&I), T)>,
}

impl<I: Idx, T> Default for TableBuilder<I, T>
where
    Option<T>: FixedSizeEncoding,
{
    fn default() -> Self {
        TableBuilder { bytes: vec![], _marker: PhantomData }
    }
}

impl<I: Idx, T> TableBuilder<I, T>
where
    Option<T>: FixedSizeEncoding,
{
    pub(crate) fn set(&mut self, i: I, value: T) {
        // FIXME(eddyb) investigate more compact encodings for sparse tables.
        // On the PR @michaelwoerister mentioned:
        // > Space requirements could perhaps be optimized by using the HAMT `popcnt`
        // > trick (i.e. divide things into buckets of 32 or 64 items and then
        // > store bit-masks of which item in each bucket is actually serialized).
        let i = i.index();
        let needed = (i + 1) * <Option<T>>::BYTE_LEN;
        if self.bytes.len() < needed {
            self.bytes.resize(needed, 0);
        }

        Some(value).write_to_bytes_at(&mut self.bytes, i);
    }

    pub(crate) fn encode(&self, buf: &mut Encoder) -> Lazy<Table<I, T>> {
        let pos = buf.position();
        buf.emit_raw_bytes(&self.bytes).unwrap();
        Lazy::from_position_and_meta(NonZeroUsize::new(pos as usize).unwrap(), self.bytes.len())
    }
}

impl<I: Idx, T> LazyMeta for Table<I, T>
where
    Option<T>: FixedSizeEncoding,
{
    type Meta = usize;

    fn min_size(len: usize) -> usize {
        len
    }
}

impl<I: Idx, T> Lazy<Table<I, T>>
where
    Option<T>: FixedSizeEncoding,
{
    /// Given the metadata, extract out the value at a particular index (if any).
    #[inline(never)]
    pub(super) fn get<'a, 'tcx, M: Metadata<'a, 'tcx>>(&self, metadata: M, i: I) -> Option<T> {
        debug!("Table::lookup: index={:?} len={:?}", i, self.meta);

        let start = self.position.get();
        let bytes = &metadata.blob()[start..start + self.meta];
        <Option<T>>::maybe_read_from_bytes_at(bytes, i.index())?
    }

    /// Size of the table in entries, including possible gaps.
    pub(super) fn size(&self) -> usize {
        self.meta / <Option<T>>::BYTE_LEN
    }
}