|
| 1 | +// Copyright 2017 The Rust Project Developers. See the COPYRIGHT |
| 2 | +// file at the top-level directory of this distribution and at |
| 3 | +// http://rust-lang.org/COPYRIGHT. |
| 4 | +// |
| 5 | +// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
| 6 | +// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
| 7 | +// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your |
| 8 | +// option. This file may not be copied, modified, or distributed |
| 9 | +// except according to those terms. |
| 10 | + |
| 11 | +// Spans are encoded using 2-bit tag and 4 different encoding formats for each tag. |
| 12 | +// Three formats are used for keeping span data inline, |
| 13 | +// the fourth one contains index into out-of-line span interner. |
| 14 | +// The encoding formats for inline spans were obtained by optimizing over crates in rustc/libstd. |
| 15 | +// See https://internals.rust-lang.org/t/rfc-compiler-refactoring-spans/1357/28 |
| 16 | + |
| 17 | +use super::*; |
| 18 | + |
| 19 | +/// A compressed span. |
| 20 | +/// Contains either fields of `SpanData` inline if they are small, or index into span interner. |
| 21 | +/// The primary goal of `Span` is to be as small as possible and fit into other structures |
| 22 | +/// (that's why it uses `packed` as well). Decoding speed is the second priority. |
| 23 | +/// See `SpanData` for the info on span fields in decoded representation. |
| 24 | +#[derive(Clone, Copy, PartialEq, Eq, Hash)] |
| 25 | +#[repr(packed)] |
| 26 | +pub struct Span(u32); |
| 27 | + |
| 28 | +/// Dummy span, both position and length are zero, syntax context is zero as well. |
| 29 | +/// This span is kept inline and encoded with format 0. |
| 30 | +pub const DUMMY_SP: Span = Span(0); |
| 31 | + |
| 32 | +impl Span { |
| 33 | + #[inline] |
| 34 | + pub fn new(lo: BytePos, hi: BytePos, ctxt: SyntaxContext) -> Self { |
| 35 | + encode(&match lo <= hi { |
| 36 | + true => SpanData { lo, hi, ctxt }, |
| 37 | + false => SpanData { lo: hi, hi: lo, ctxt }, |
| 38 | + }) |
| 39 | + } |
| 40 | + |
| 41 | + #[inline] |
| 42 | + pub fn data(self) -> SpanData { |
| 43 | + decode(self) |
| 44 | + } |
| 45 | +} |
| 46 | + |
| 47 | +// Tags |
| 48 | +const TAG_INLINE0: u32 = 0b00; |
| 49 | +const TAG_INLINE1: u32 = 0b01; |
| 50 | +const TAG_INLINE2: u32 = 0b10; |
| 51 | +const TAG_INTERNED: u32 = 0b11; |
| 52 | +const TAG_MASK: u32 = 0b11; |
| 53 | + |
| 54 | +// Fields indexes |
| 55 | +const BASE_INDEX: usize = 0; |
| 56 | +const LEN_INDEX: usize = 1; |
| 57 | +const CTXT_INDEX: usize = 2; |
| 58 | + |
| 59 | +// Tag = 0b00, inline format 0. |
| 60 | +// ----------------------------------- |
| 61 | +// | base 31:8 | len 7:2 | tag 1:0 | |
| 62 | +// ----------------------------------- |
| 63 | +const INLINE0_SIZES: [u32; 3] = [24, 6, 0]; |
| 64 | +const INLINE0_OFFSETS: [u32; 3] = [8, 2, 2]; |
| 65 | + |
| 66 | +// Tag = 0b01, inline format 1. |
| 67 | +// ----------------------------------- |
| 68 | +// | base 31:10 | len 9:2 | tag 1:0 | |
| 69 | +// ----------------------------------- |
| 70 | +const INLINE1_SIZES: [u32; 3] = [22, 8, 0]; |
| 71 | +const INLINE1_OFFSETS: [u32; 3] = [10, 2, 2]; |
| 72 | + |
| 73 | +// Tag = 0b10, inline format 2. |
| 74 | +// ------------------------------------------------ |
| 75 | +// | base 31:14 | len 13:13 | ctxt 12:2 | tag 1:0 | |
| 76 | +// ------------------------------------------------ |
| 77 | +const INLINE2_SIZES: [u32; 3] = [18, 1, 11]; |
| 78 | +const INLINE2_OFFSETS: [u32; 3] = [14, 13, 2]; |
| 79 | + |
| 80 | +// Tag = 0b11, interned format. |
| 81 | +// ------------------------ |
| 82 | +// | index 31:3 | tag 1:0 | |
| 83 | +// ------------------------ |
| 84 | +const INTERNED_INDEX_SIZE: u32 = 30; |
| 85 | +const INTERNED_INDEX_OFFSET: u32 = 2; |
| 86 | + |
| 87 | +fn encode(sd: &SpanData) -> Span { |
| 88 | + let (base, len, ctxt) = (sd.lo.0, sd.hi.0 - sd.lo.0, sd.ctxt.0); |
| 89 | + |
| 90 | + // Can we fit the span data into this encoding? |
| 91 | + let fits = |sizes: [u32; 3]| { |
| 92 | + (base >> sizes[BASE_INDEX]) == 0 && (len >> sizes[LEN_INDEX]) == 0 && |
| 93 | + (ctxt >> sizes[CTXT_INDEX]) == 0 |
| 94 | + }; |
| 95 | + // Turn fields into a single `u32` value. |
| 96 | + let compose = |offsets: [u32; 3], tag| { |
| 97 | + (base << offsets[BASE_INDEX]) | (len << offsets[LEN_INDEX]) | |
| 98 | + (ctxt << offsets[CTXT_INDEX]) | tag |
| 99 | + }; |
| 100 | + |
| 101 | + let val = if fits(INLINE0_SIZES) { |
| 102 | + compose(INLINE0_OFFSETS, TAG_INLINE0) |
| 103 | + } else if fits(INLINE1_SIZES) { |
| 104 | + compose(INLINE1_OFFSETS, TAG_INLINE1) |
| 105 | + } else if fits(INLINE2_SIZES) { |
| 106 | + compose(INLINE2_OFFSETS, TAG_INLINE2) |
| 107 | + } else { |
| 108 | + let index = with_span_interner(|interner| interner.intern(sd)); |
| 109 | + if (index >> INTERNED_INDEX_SIZE) == 0 { |
| 110 | + (index << INTERNED_INDEX_OFFSET) | TAG_INTERNED |
| 111 | + } else { |
| 112 | + panic!("too many spans in a crate"); |
| 113 | + } |
| 114 | + }; |
| 115 | + Span(val) |
| 116 | +} |
| 117 | + |
| 118 | +fn decode(span: Span) -> SpanData { |
| 119 | + let val = span.0; |
| 120 | + |
| 121 | + // Extract a field at position `pos` having size `size`. |
| 122 | + let extract = |pos, size| { |
| 123 | + let mask = ((!0u32) as u64 >> (32 - size)) as u32; // Can't shift u32 by 32 |
| 124 | + (val >> pos) & mask |
| 125 | + }; |
| 126 | + |
| 127 | + let (base, len, ctxt) = match val & TAG_MASK { |
| 128 | + TAG_INLINE0 => ( |
| 129 | + extract(INLINE0_OFFSETS[BASE_INDEX], INLINE0_SIZES[BASE_INDEX]), |
| 130 | + extract(INLINE0_OFFSETS[LEN_INDEX], INLINE0_SIZES[LEN_INDEX]), |
| 131 | + extract(INLINE0_OFFSETS[CTXT_INDEX], INLINE0_SIZES[CTXT_INDEX]), |
| 132 | + ), |
| 133 | + TAG_INLINE1 => ( |
| 134 | + extract(INLINE1_OFFSETS[BASE_INDEX], INLINE1_SIZES[BASE_INDEX]), |
| 135 | + extract(INLINE1_OFFSETS[LEN_INDEX], INLINE1_SIZES[LEN_INDEX]), |
| 136 | + extract(INLINE1_OFFSETS[CTXT_INDEX], INLINE1_SIZES[CTXT_INDEX]), |
| 137 | + ), |
| 138 | + TAG_INLINE2 => ( |
| 139 | + extract(INLINE2_OFFSETS[BASE_INDEX], INLINE2_SIZES[BASE_INDEX]), |
| 140 | + extract(INLINE2_OFFSETS[LEN_INDEX], INLINE2_SIZES[LEN_INDEX]), |
| 141 | + extract(INLINE2_OFFSETS[CTXT_INDEX], INLINE2_SIZES[CTXT_INDEX]), |
| 142 | + ), |
| 143 | + TAG_INTERNED => { |
| 144 | + let index = extract(INTERNED_INDEX_OFFSET, INTERNED_INDEX_SIZE); |
| 145 | + return with_span_interner(|interner| *interner.get(index)); |
| 146 | + } |
| 147 | + _ => unreachable!() |
| 148 | + }; |
| 149 | + SpanData { lo: BytePos(base), hi: BytePos(base + len), ctxt: SyntaxContext(ctxt) } |
| 150 | +} |
| 151 | + |
| 152 | +#[derive(Default)] |
| 153 | +struct SpanInterner { |
| 154 | + spans: HashMap<SpanData, u32>, |
| 155 | + span_data: Vec<SpanData>, |
| 156 | +} |
| 157 | + |
| 158 | +impl SpanInterner { |
| 159 | + fn intern(&mut self, span_data: &SpanData) -> u32 { |
| 160 | + if let Some(index) = self.spans.get(span_data) { |
| 161 | + return *index; |
| 162 | + } |
| 163 | + |
| 164 | + let index = self.spans.len() as u32; |
| 165 | + self.span_data.push(*span_data); |
| 166 | + self.spans.insert(*span_data, index); |
| 167 | + index |
| 168 | + } |
| 169 | + |
| 170 | + fn get(&self, index: u32) -> &SpanData { |
| 171 | + &self.span_data[index as usize] |
| 172 | + } |
| 173 | +} |
| 174 | + |
| 175 | +// If an interner exists in TLS, return it. Otherwise, prepare a fresh one. |
| 176 | +fn with_span_interner<T, F: FnOnce(&mut SpanInterner) -> T>(f: F) -> T { |
| 177 | + thread_local!(static INTERNER: RefCell<SpanInterner> = { |
| 178 | + RefCell::new(SpanInterner::default()) |
| 179 | + }); |
| 180 | + INTERNER.with(|interner| f(&mut *interner.borrow_mut())) |
| 181 | +} |
0 commit comments