Rollup merge of #70486 - Mark-Simulacrum:unicode-shrink, r=dtolnay

Shrink Unicode tables (even more)

This shrinks the Unicode tables further, building upon the wins in #68232 (the previous counts differ due to an interim Unicode version update, see #69929.

The new data structure is slower by around 3x, on the benchmark of looking up every Unicode scalar value in each data set sequentially in every data set included. Note that for ASCII, the exposed functions on `char` optimize with direct branches, so ASCII will retain the same performance regardless of internal optimizations (or the reverse). Also, note that the size reduction due to the skip list (from where the performance losses come) is around 40%, and, as a result, I believe the performance loss is acceptable, as the routines are still quite fast. Anywhere where this is hot, should probably be using a custom data structure anyway (e.g., a raw bitset) or something optimized for frequently seen values, etc.

This PR updates both the bitset data structure, and introduces a new data structure similar to a skip list. For more details, see the [main.rs] of the table generator, which describes both. The commits mostly work individually and document size wins.

As before, this is tested on all valid chars to have the same results as nightly (and the canonical Unicode data sets), happily, no bugs were found.

[main.rs]: https://github.com/rust-lang/rust/blob/fb4a715e18b/src/tools/unicode-table-generator/src/main.rs

Set             | Previous |  New  |  % of old  | Codepoints | Ranges |
----------------|---------:|------:|-----------:|-----------:|-------:|
Alphabetic      |     3055 |  1599 |        52% |     132875 |    695 |
Case Ignorable  |     2136 |   949 |        44% |       2413 |    410 |
Cased           |      934 |   359 |        38% |       4286 |    141 |
Cc              |       43 |     9 |        20% |         65 |      2 |
Grapheme Extend |     1774 |   813 |        46% |       1979 |    344 |
Lowercase       |      985 |   867 |        88% |       2344 |    652 |
N               |     1266 |   419 |        33% |       1781 |    133 |
Uppercase       |      934 |   777 |        83% |       1911 |    643 |
White_Space     |      140 |    37 |        26% |         25 |     10 |
----------------|----------|-------|------------|------------|--------|
Total           |    11267 |  5829 |        51% |     -      |   -    |

Author: Dylan-DPC

src/libcore/unicode/mod.rs

@@ -32,28 +32,3 @@ pub use unicode_data::lowercase::lookup as Lowercase;
 pub use unicode_data::n::lookup as N;
 pub use unicode_data::uppercase::lookup as Uppercase;
 pub use unicode_data::white_space::lookup as White_Space;
-
-#[inline(always)]
-fn range_search<const N: usize, const N1: usize, const N2: usize>(
-    needle: u32,
-    chunk_idx_map: &[u8; N],
-    (last_chunk_idx, last_chunk_mapping): (u16, u8),
-    bitset_chunk_idx: &[[u8; 16]; N1],
-    bitset: &[u64; N2],
-) -> bool {
-    let bucket_idx = (needle / 64) as usize;
-    let chunk_map_idx = bucket_idx / 16;
-    let chunk_piece = bucket_idx % 16;
-    let chunk_idx = if chunk_map_idx >= N {
-        if chunk_map_idx == last_chunk_idx as usize {
-            last_chunk_mapping
-        } else {
-            return false;
-        }
-    } else {
-        chunk_idx_map[chunk_map_idx]
-    };
-    let idx = bitset_chunk_idx[(chunk_idx as usize)][chunk_piece];
-    let word = bitset[(idx as usize)];
-    (word & (1 << (needle % 64) as u64)) != 0
-}

src/libcore/unicode/unicode_data.rs

@@ -1,9 +1,83 @@
+//! This implements the core logic of the compression scheme used to compactly
+//! encode Unicode properties.
+//!
+//! We have two primary goals with the encoding: we want to be compact, because
+//! these tables often end up in ~every Rust program (especially the
+//! grapheme_extend table, used for str debugging), including those for embedded
+//! targets (where space is important). We also want to be relatively fast,
+//! though this is more of a nice to have rather than a key design constraint.
+//! It is expected that libraries/applications which are performance-sensitive
+//! to Unicode property lookups are extremely rare, and those that care may find
+//! the tradeoff of the raw bitsets worth it. For most applications, a
+//! relatively fast but much smaller (and as such less cache-impacting, etc.)
+//! data set is likely preferable.
+//!
+//! We have two separate encoding schemes: a skiplist-like approach, and a
+//! compressed bitset. The datasets we consider mostly use the skiplist (it's
+//! smaller) but the lowercase and uppercase sets are sufficiently sparse for
+//! the bitset to be worthwhile -- for those sets the biset is a 2x size win.
+//! Since the bitset is also faster, this seems an obvious choice. (As a
+//! historical note, the bitset was also the prior implementation, so its
+//! relative complexity had already been paid).
+//!
+//! ## The bitset
+//!
+//! The primary idea is that we 'flatten' the Unicode ranges into an enormous
+//! bitset. To represent any arbitrary codepoint in a raw bitset, we would need
+//! over 17 kilobytes of data per character set -- way too much for our
+//! purposes.
+//!
+//! First, the raw bitset (one bit for every valid `char`, from 0 to 0x10FFFF,
+//! not skipping the small 'gap') is associated into words (u64) and
+//! deduplicated. On random data, this would be useless; on our data, this is
+//! incredibly beneficial -- our data sets have (far) less than 256 unique
+//! words.
+//!
+//! This gives us an array that maps `u8 -> word`; the current algorithm does
+//! not handle the case of more than 256 unique words, but we are relatively far
+//! from coming that close.
+//!
+//! With that scheme, we now have a single byte for every 64 codepoints.
+//!
+//! We further chunk these by some constant N (between 1 and 64 per group,
+//! dynamically chosen for smallest size), and again deduplicate and store in an
+//! array (u8 -> [u8; N]).
+//!
+//! The bytes of this array map into the words from the bitset above, but we
+//! apply another trick here: some of these words are similar enough that they
+//! can be represented by some function of another word. The particular
+//! functions chosen are rotation, inversion, and shifting (right).
+//!
+//! ## The skiplist
+//!
+//! The skip list arose out of the desire for an even smaller encoding than the
+//! bitset -- and was the answer to the question "what is the smallest
+//! representation we can imagine?". However, it is not necessarily the
+//! smallest, and if you have a better proposal, please do suggest it!
+//!
+//! This is a relatively straightforward encoding. First, we break up all the
+//! ranges in the input data into offsets from each other, essentially a gap
+//! encoding. In practice, most gaps are small -- less than u8::MAX -- so we
+//! store those directly. We make use of the larger gaps (which are nicely
+//! interspersed already) throughout the dataset to index this data set.
+//!
+//! In particular, each run of small gaps (terminating in a large gap) is
+//! indexed in a separate dataset. That data set stores an index into the
+//! primary offset list and a prefix sum of that offset list. These are packed
+//! into a single u32 (11 bits for the offset, 21 bits for the prefix sum).
+//!
+//! Lookup proceeds via a binary search in the index and then a straightforward
+//! linear scan (adding up the offsets) until we reach the needle, and then the
+//! index of that offset is utilized as the answer to whether we're in the set
+//! or not.
+
 use std::collections::{BTreeMap, HashMap};
 use std::ops::Range;
 use ucd_parse::Codepoints;
 
 mod case_mapping;
 mod raw_emitter;
+mod skiplist;
 mod unicode_download;
 
 use raw_emitter::{emit_codepoints, RawEmitter};
@@ -152,9 +226,17 @@ fn main() {
         std::process::exit(1);
     });
 
+    // Optional test path, which is a Rust source file testing that the unicode
+    // property lookups are correct.
+    let test_path = std::env::args().nth(2);
+
     let unicode_data = load_data();
     let ranges_by_property = &unicode_data.ranges;
 
+    if let Some(path) = test_path {
+        std::fs::write(&path, generate_tests(&write_location, &ranges_by_property)).unwrap();
+    }
+
     let mut total_bytes = 0;
     let mut modules = Vec::new();
     for (property, ranges) in ranges_by_property {
@@ -163,7 +245,16 @@ fn main() {
         emit_codepoints(&mut emitter, &ranges);
 
         modules.push((property.to_lowercase().to_string(), emitter.file));
-        println!("{:15}: {} bytes, {} codepoints", property, emitter.bytes_used, datapoints,);
+        println!(
+            "{:15}: {} bytes, {} codepoints in {} ranges ({} - {}) using {}",
+            property,
+            emitter.bytes_used,
+            datapoints,
+            ranges.len(),
+            ranges.first().unwrap().start,
+            ranges.last().unwrap().end,
+            emitter.desc,
+        );
         total_bytes += emitter.bytes_used;
     }
 
@@ -173,7 +264,10 @@ fn main() {
         "///! This file is generated by src/tools/unicode-table-generator; do not edit manually!\n",
     );
 
-    table_file.push_str("use super::range_search;\n\n");
+    // Include the range search function
+    table_file.push('\n');
+    table_file.push_str(include_str!("range_search.rs"));
+    table_file.push('\n');
 
     table_file.push_str(&version());
 
@@ -236,26 +330,110 @@ fn fmt_list<V: std::fmt::Debug>(values: impl IntoIterator<Item = V>) -> String {
     out
 }
 
+fn generate_tests(data_path: &str, ranges: &[(&str, Vec<Range<u32>>)]) -> String {
+    let mut s = String::new();
+    s.push_str("#![allow(incomplete_features, unused)]\n");
+    s.push_str("#![feature(const_generics)]\n\n");
+    s.push_str("\n#[allow(unused)]\nuse std::hint;\n");
+    s.push_str(&format!("#[path = \"{}\"]\n", data_path));
+    s.push_str("mod unicode_data;\n\n");
+
+    s.push_str("\nfn main() {\n");
+
+    for (property, ranges) in ranges {
+        s.push_str(&format!(r#"    println!("Testing {}");"#, property));
+        s.push('\n');
+        s.push_str(&format!("    {}_true();\n", property.to_lowercase()));
+        s.push_str(&format!("    {}_false();\n", property.to_lowercase()));
+        let mut is_true = Vec::new();
+        let mut is_false = Vec::new();
+        for ch_num in 0..(std::char::MAX as u32) {
+            if std::char::from_u32(ch_num).is_none() {
+                continue;
+            }
+            if ranges.iter().any(|r| r.contains(&ch_num)) {
+                is_true.push(ch_num);
+            } else {
+                is_false.push(ch_num);
+            }
+        }
+
+        s.push_str(&format!("    fn {}_true() {{\n", property.to_lowercase()));
+        generate_asserts(&mut s, property, &is_true, true);
+        s.push_str("    }\n\n");
+        s.push_str(&format!("    fn {}_false() {{\n", property.to_lowercase()));
+        generate_asserts(&mut s, property, &is_false, false);
+        s.push_str("    }\n\n");
+    }
+
+    s.push_str("}");
+    s
+}
+
+fn generate_asserts(s: &mut String, property: &str, points: &[u32], truthy: bool) {
+    for range in ranges_from_set(points) {
+        if range.end == range.start + 1 {
+            s.push_str(&format!(
+                "        assert!({}unicode_data::{}::lookup({:?}), \"{}\");\n",
+                if truthy { "" } else { "!" },
+                property.to_lowercase(),
+                std::char::from_u32(range.start).unwrap(),
+                range.start,
+            ));
+        } else {
+            s.push_str(&format!("        for chn in {:?}u32 {{\n", range));
+            s.push_str(&format!(
+                "            assert!({}unicode_data::{}::lookup(std::char::from_u32(chn).unwrap()), \"{{:?}}\", chn);\n",
+                if truthy { "" } else { "!" },
+                property.to_lowercase(),
+            ));
+            s.push_str("        }\n");
+        }
+    }
+}
+
+fn ranges_from_set(set: &[u32]) -> Vec<Range<u32>> {
+    let mut ranges = set.iter().map(|e| (*e)..(*e + 1)).collect::<Vec<Range<u32>>>();
+    merge_ranges(&mut ranges);
+    ranges
+}
+
 fn merge_ranges(ranges: &mut Vec<Range<u32>>) {
     loop {
         let mut new_ranges = Vec::new();
         let mut idx_iter = 0..(ranges.len() - 1);
+        let mut should_insert_last = true;
         while let Some(idx) = idx_iter.next() {
             let cur = ranges[idx].clone();
             let next = ranges[idx + 1].clone();
             if cur.end == next.start {
-                let _ = idx_iter.next(); // skip next as we're merging it in
+                if idx_iter.next().is_none() {
+                    // We're merging the last element
+                    should_insert_last = false;
+                }
                 new_ranges.push(cur.start..next.end);
             } else {
+                // We're *not* merging the last element
+                should_insert_last = true;
                 new_ranges.push(cur);
             }
         }
-        new_ranges.push(ranges.last().unwrap().clone());
+        if should_insert_last {
+            new_ranges.push(ranges.last().unwrap().clone());
+        }
         if new_ranges.len() == ranges.len() {
             *ranges = new_ranges;
             break;
         } else {
             *ranges = new_ranges;
         }
     }
+
+    let mut last_end = None;
+    for range in ranges {
+        if let Some(last) = last_end {
+            assert!(range.start > last, "{:?}", range);
+        }
+        last_end = Some(range.end);
+    }
 }

src/tools/unicode-table-generator/src/main.rs

@@ -0,0 +1,93 @@
+#[inline(always)]
+fn bitset_search<
+    const N: usize,
+    const CHUNK_SIZE: usize,
+    const N1: usize,
+    const CANONICAL: usize,
+    const CANONICALIZED: usize,
+>(
+    needle: u32,
+    chunk_idx_map: &[u8; N],
+    bitset_chunk_idx: &[[u8; CHUNK_SIZE]; N1],
+    bitset_canonical: &[u64; CANONICAL],
+    bitset_canonicalized: &[(u8, u8); CANONICALIZED],
+) -> bool {
+    let bucket_idx = (needle / 64) as usize;
+    let chunk_map_idx = bucket_idx / CHUNK_SIZE;
+    let chunk_piece = bucket_idx % CHUNK_SIZE;
+    let chunk_idx = if let Some(&v) = chunk_idx_map.get(chunk_map_idx) {
+        v
+    } else {
+        return false;
+    };
+    let idx = bitset_chunk_idx[chunk_idx as usize][chunk_piece] as usize;
+    let word = if let Some(word) = bitset_canonical.get(idx) {
+        *word
+    } else {
+        let (real_idx, mapping) = bitset_canonicalized[idx - bitset_canonical.len()];
+        let mut word = bitset_canonical[real_idx as usize];
+        let should_invert = mapping & (1 << 6) != 0;
+        if should_invert {
+            word = !word;
+        }
+        // Lower 6 bits
+        let quantity = mapping & ((1 << 6) - 1);
+        if mapping & (1 << 7) != 0 {
+            // shift
+            word >>= quantity as u64;
+        } else {
+            word = word.rotate_left(quantity as u32);
+        }
+        word
+    };
+    (word & (1 << (needle % 64) as u64)) != 0
+}
+
+fn decode_prefix_sum(short_offset_run_header: u32) -> u32 {
+    short_offset_run_header & ((1 << 21) - 1)
+}
+
+fn decode_length(short_offset_run_header: u32) -> usize {
+    (short_offset_run_header >> 21) as usize
+}
+
+#[inline(always)]
+fn skip_search<const SOR: usize, const OFFSETS: usize>(
+    needle: u32,
+    short_offset_runs: &[u32; SOR],
+    offsets: &[u8; OFFSETS],
+) -> bool {
+    // Note that this *cannot* be past the end of the array, as the last
+    // element is greater than std::char::MAX (the largest possible needle).
+    //
+    // So, we cannot have found it (i.e. Ok(idx) + 1 != length) and the correct
+    // location cannot be past it, so Err(idx) != length either.
+    //
+    // This means that we can avoid bounds checking for the accesses below, too.
+    let last_idx =
+        match short_offset_runs.binary_search_by_key(&(needle << 11), |header| header << 11) {
+            Ok(idx) => idx + 1,
+            Err(idx) => idx,
+        };
+
+    let mut offset_idx = decode_length(short_offset_runs[last_idx]);
+    let length = if let Some(next) = short_offset_runs.get(last_idx + 1) {
+        decode_length(*next) - offset_idx
+    } else {
+        offsets.len() - offset_idx
+    };
+    let prev =
+        last_idx.checked_sub(1).map(|prev| decode_prefix_sum(short_offset_runs[prev])).unwrap_or(0);
+
+    let total = needle - prev;
+    let mut prefix_sum = 0;
+    for _ in 0..(length - 1) {
+        let offset = offsets[offset_idx];
+        prefix_sum += offset as u32;
+        if prefix_sum > total {
+            break;
+        }
+        offset_idx += 1;
+    }
+    offset_idx % 2 == 1
+}