diff --git a/Cargo.lock b/Cargo.lock
index 01d8910042820..1f191cf79473f 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -730,7 +730,7 @@ dependencies = [
  "indexmap 2.0.0",
  "itertools",
  "limit",
- "line-index 0.1.0-pre.1 (registry+https://github.com/rust-lang/crates.io-index)",
+ "line-index 0.1.0-pre.1",
  "memchr",
  "nohash-hasher",
  "once_cell",
@@ -947,6 +947,8 @@ version = "0.0.0"
 [[package]]
 name = "line-index"
 version = "0.1.0-pre.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2cad96769710c1745e11d4f940a8ff36000ade4bbada4285b001cb8aa2f745ce"
 dependencies = [
  "nohash-hasher",
  "text-size",
@@ -954,9 +956,7 @@ dependencies = [
 
 [[package]]
 name = "line-index"
-version = "0.1.0-pre.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2cad96769710c1745e11d4f940a8ff36000ade4bbada4285b001cb8aa2f745ce"
+version = "0.1.0"
 dependencies = [
  "nohash-hasher",
  "text-size",
diff --git a/lib/line-index/Cargo.toml b/lib/line-index/Cargo.toml
index b267daa2050f9..6c0d06f471780 100644
--- a/lib/line-index/Cargo.toml
+++ b/lib/line-index/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "line-index"
-version = "0.1.0-pre.1"
+version = "0.1.0"
 description = "Maps flat `TextSize` offsets to/from `(line, column)` representation."
 license = "MIT OR Apache-2.0"
 repository = "https://github.com/rust-lang/rust-analyzer/tree/master/lib/line-index"
diff --git a/lib/line-index/src/lib.rs b/lib/line-index/src/lib.rs
index ad67d3f246eca..58db93a58909b 100644
--- a/lib/line-index/src/lib.rs
+++ b/lib/line-index/src/lib.rs
@@ -94,44 +94,7 @@ pub struct LineIndex {
 impl LineIndex {
     /// Returns a `LineIndex` for the `text`.
     pub fn new(text: &str) -> LineIndex {
-        let mut newlines = Vec::<TextSize>::with_capacity(16);
-        let mut line_wide_chars = IntMap::<u32, Box<[WideChar]>>::default();
-
-        let mut wide_chars = Vec::<WideChar>::new();
-        let mut cur_row = TextSize::from(0);
-        let mut cur_col = TextSize::from(0);
-        let mut line = 0u32;
-
-        for c in text.chars() {
-            let c_len = TextSize::of(c);
-            cur_row += c_len;
-            if c == '\n' {
-                newlines.push(cur_row);
-
-                // Save any wide characters seen in the previous line
-                if !wide_chars.is_empty() {
-                    let cs = std::mem::take(&mut wide_chars).into_boxed_slice();
-                    line_wide_chars.insert(line, cs);
-                }
-
-                // Prepare for processing the next line
-                cur_col = TextSize::from(0);
-                line += 1;
-                continue;
-            }
-
-            if !c.is_ascii() {
-                wide_chars.push(WideChar { start: cur_col, end: cur_col + c_len });
-            }
-
-            cur_col += c_len;
-        }
-
-        // Save any wide characters seen in the last line
-        if !wide_chars.is_empty() {
-            line_wide_chars.insert(line, wide_chars.into_boxed_slice());
-        }
-
+        let (newlines, line_wide_chars) = analyze_source_file(text);
         LineIndex {
             newlines: newlines.into_boxed_slice(),
             line_wide_chars,
@@ -235,3 +198,182 @@ impl LineIndex {
         self.len
     }
 }
+
+/// This is adapted from the rustc_span crate, https://github.com/rust-lang/rust/blob/master/compiler/rustc_span/src/analyze_source_file.rs
+fn analyze_source_file(src: &str) -> (Vec<TextSize>, IntMap<u32, Box<[WideChar]>>) {
+    assert!(src.len() < !0u32 as usize);
+    let mut lines = vec![];
+    let mut line_wide_chars = IntMap::<u32, Vec<WideChar>>::default();
+
+    // Calls the right implementation, depending on hardware support available.
+    analyze_source_file_dispatch(src, &mut lines, &mut line_wide_chars);
+
+    (lines, line_wide_chars.into_iter().map(|(k, v)| (k, v.into_boxed_slice())).collect())
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn analyze_source_file_dispatch(
+    src: &str,
+    lines: &mut Vec<TextSize>,
+    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
+) {
+    if is_x86_feature_detected!("sse2") {
+        // SAFETY: SSE2 support was checked
+        unsafe {
+            analyze_source_file_sse2(src, lines, multi_byte_chars);
+        }
+    } else {
+        analyze_source_file_generic(src, src.len(), TextSize::from(0), lines, multi_byte_chars);
+    }
+}
+
+/// Checks 16 byte chunks of text at a time. If the chunk contains
+/// something other than printable ASCII characters and newlines, the
+/// function falls back to the generic implementation. Otherwise it uses
+/// SSE2 intrinsics to quickly find all newlines.
+#[target_feature(enable = "sse2")]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+unsafe fn analyze_source_file_sse2(
+    src: &str,
+    lines: &mut Vec<TextSize>,
+    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
+) {
+    #[cfg(target_arch = "x86")]
+    use std::arch::x86::*;
+    #[cfg(target_arch = "x86_64")]
+    use std::arch::x86_64::*;
+
+    const CHUNK_SIZE: usize = 16;
+
+    let src_bytes = src.as_bytes();
+
+    let chunk_count = src.len() / CHUNK_SIZE;
+
+    // This variable keeps track of where we should start decoding a
+    // chunk. If a multi-byte character spans across chunk boundaries,
+    // we need to skip that part in the next chunk because we already
+    // handled it.
+    let mut intra_chunk_offset = 0;
+
+    for chunk_index in 0..chunk_count {
+        let ptr = src_bytes.as_ptr() as *const __m128i;
+        // We don't know if the pointer is aligned to 16 bytes, so we
+        // use `loadu`, which supports unaligned loading.
+        let chunk = _mm_loadu_si128(ptr.add(chunk_index));
+
+        // For character in the chunk, see if its byte value is < 0, which
+        // indicates that it's part of a UTF-8 char.
+        let multibyte_test = _mm_cmplt_epi8(chunk, _mm_set1_epi8(0));
+        // Create a bit mask from the comparison results.
+        let multibyte_mask = _mm_movemask_epi8(multibyte_test);
+
+        // If the bit mask is all zero, we only have ASCII chars here:
+        if multibyte_mask == 0 {
+            assert!(intra_chunk_offset == 0);
+
+            // Check for newlines in the chunk
+            let newlines_test = _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8));
+            let newlines_mask = _mm_movemask_epi8(newlines_test);
+
+            if newlines_mask != 0 {
+                // All control characters are newlines, record them
+                let mut newlines_mask = 0xFFFF0000 | newlines_mask as u32;
+                let output_offset = TextSize::from((chunk_index * CHUNK_SIZE + 1) as u32);
+
+                loop {
+                    let index = newlines_mask.trailing_zeros();
+
+                    if index >= CHUNK_SIZE as u32 {
+                        // We have arrived at the end of the chunk.
+                        break;
+                    }
+
+                    lines.push(TextSize::from(index) + output_offset);
+
+                    // Clear the bit, so we can find the next one.
+                    newlines_mask &= (!1) << index;
+                }
+            }
+            continue;
+        }
+
+        // The slow path.
+        // There are control chars in here, fallback to generic decoding.
+        let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
+        intra_chunk_offset = analyze_source_file_generic(
+            &src[scan_start..],
+            CHUNK_SIZE - intra_chunk_offset,
+            TextSize::from(scan_start as u32),
+            lines,
+            multi_byte_chars,
+        );
+    }
+
+    // There might still be a tail left to analyze
+    let tail_start = chunk_count * CHUNK_SIZE + intra_chunk_offset;
+    if tail_start < src.len() {
+        analyze_source_file_generic(
+            &src[tail_start..],
+            src.len() - tail_start,
+            TextSize::from(tail_start as u32),
+            lines,
+            multi_byte_chars,
+        );
+    }
+}
+
+#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
+// The target (or compiler version) does not support SSE2 ...
+fn analyze_source_file_dispatch(
+    src: &str,
+    lines: &mut Vec<TextSize>,
+    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
+) {
+    analyze_source_file_generic(src, src.len(), TextSize::from(0), lines, multi_byte_chars);
+}
+
+// `scan_len` determines the number of bytes in `src` to scan. Note that the
+// function can read past `scan_len` if a multi-byte character start within the
+// range but extends past it. The overflow is returned by the function.
+fn analyze_source_file_generic(
+    src: &str,
+    scan_len: usize,
+    output_offset: TextSize,
+    lines: &mut Vec<TextSize>,
+    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
+) -> usize {
+    assert!(src.len() >= scan_len);
+    let mut i = 0;
+    let src_bytes = src.as_bytes();
+
+    while i < scan_len {
+        let byte = unsafe {
+            // We verified that i < scan_len <= src.len()
+            *src_bytes.get_unchecked(i)
+        };
+
+        // How much to advance in order to get to the next UTF-8 char in the
+        // string.
+        let mut char_len = 1;
+
+        if byte == b'\n' {
+            lines.push(TextSize::from(i as u32 + 1) + output_offset);
+        } else if byte >= 127 {
+            // The slow path: Just decode to `char`.
+            let c = src[i..].chars().next().unwrap();
+            char_len = c.len_utf8();
+
+            let pos = TextSize::from(i as u32) + output_offset;
+
+            if char_len > 1 {
+                assert!((2..=4).contains(&char_len));
+                let mbc = WideChar { start: pos, end: pos + TextSize::from(char_len as u32) };
+                multi_byte_chars.entry(lines.len() as u32).or_default().push(mbc);
+            }
+        }
+
+        i += char_len;
+    }
+
+    i - scan_len
+}
diff --git a/lib/line-index/src/tests.rs b/lib/line-index/src/tests.rs
index 31c01c20ee36f..8f3762d191064 100644
--- a/lib/line-index/src/tests.rs
+++ b/lib/line-index/src/tests.rs
@@ -1,11 +1,120 @@
-use super::LineIndex;
-
-#[test]
-fn test_empty_index() {
-    let col_index = LineIndex::new(
-        "
-const C: char = 'x';
-",
-    );
-    assert_eq!(col_index.line_wide_chars.len(), 0);
+use crate::{LineIndex, TextSize, WideChar};
+
+macro_rules! test {
+    (
+        case: $test_name:ident,
+     text: $text:expr,
+     lines: $lines:expr,
+     multi_byte_chars: $multi_byte_chars:expr,
+    ) => {
+        #[test]
+        fn $test_name() {
+            let line_index = LineIndex::new($text);
+
+            let expected_lines: Vec<TextSize> =
+                $lines.into_iter().map(<TextSize as From<u32>>::from).collect();
+
+            assert_eq!(&*line_index.newlines, &*expected_lines);
+
+            let expected_mbcs: Vec<_> = $multi_byte_chars
+                .into_iter()
+                .map(|(line, (pos, end)): (u32, (u32, u32))| {
+                    (line, WideChar { start: TextSize::from(pos), end: TextSize::from(end) })
+                })
+                .collect();
+
+            assert_eq!(
+                line_index
+                    .line_wide_chars
+                    .iter()
+                    .flat_map(|(line, val)| std::iter::repeat(*line).zip(val.iter().copied()))
+                    .collect::<Vec<_>>(),
+                expected_mbcs
+            );
+        }
+    };
 }
+
+test!(
+    case: empty_text,
+    text: "",
+    lines: vec![],
+    multi_byte_chars: vec![],
+);
+
+test!(
+    case: newlines_short,
+    text: "a\nc",
+    lines: vec![2],
+    multi_byte_chars: vec![],
+);
+
+test!(
+    case: newlines_long,
+    text: "012345678\nabcdef012345678\na",
+    lines: vec![10, 26],
+    multi_byte_chars: vec![],
+);
+
+test!(
+    case: newline_and_multi_byte_char_in_same_chunk,
+    text: "01234β789\nbcdef0123456789abcdef",
+    lines: vec![11],
+    multi_byte_chars: vec![(0, (5, 7))],
+);
+
+test!(
+    case: newline_and_control_char_in_same_chunk,
+    text: "01234\u{07}6789\nbcdef0123456789abcdef",
+    lines: vec![11],
+    multi_byte_chars: vec![],
+);
+
+test!(
+    case: multi_byte_char_short,
+    text: "aβc",
+    lines: vec![],
+    multi_byte_chars: vec![(0, (1, 3))],
+);
+
+test!(
+    case: multi_byte_char_long,
+    text: "0123456789abcΔf012345β",
+    lines: vec![],
+    multi_byte_chars: vec![(0, (13, 15)), (0, (22, 24))],
+);
+
+test!(
+    case: multi_byte_char_across_chunk_boundary,
+    text: "0123456789abcdeΔ123456789abcdef01234",
+    lines: vec![],
+    multi_byte_chars: vec![(0, (15, 17))],
+);
+
+test!(
+    case: multi_byte_char_across_chunk_boundary_tail,
+    text: "0123456789abcdeΔ....",
+    lines: vec![],
+    multi_byte_chars: vec![(0, (15, 17))],
+);
+
+test!(
+    case: multi_byte_with_new_lines,
+    text: "01\t345\n789abcΔf01234567\u{07}9\nbcΔf",
+    lines: vec![7, 27],
+    multi_byte_chars: vec![(1, (13, 15)), (2, (29, 31))],
+);
+
+test!(
+    case: trailing_newline,
+    text: "0123456789\n",
+    lines: vec![11],
+    multi_byte_chars: vec![],
+);
+
+test!(
+    case: trailing_newline_chunk_boundary,
+    text: "0123456789abcde\n",
+    lines: vec![16],
+    multi_byte_chars: vec![],
+);
diff --git a/lib/line-index/tests/it.rs b/lib/line-index/tests/it.rs
deleted file mode 100644
index ce1c0bc6f143d..0000000000000
--- a/lib/line-index/tests/it.rs
+++ /dev/null
@@ -1,62 +0,0 @@
-use line_index::{LineCol, LineIndex, TextRange};
-
-#[test]
-fn test_line_index() {
-    let text = "hello\nworld";
-    let table = [
-        (00, 0, 0),
-        (01, 0, 1),
-        (05, 0, 5),
-        (06, 1, 0),
-        (07, 1, 1),
-        (08, 1, 2),
-        (10, 1, 4),
-        (11, 1, 5),
-    ];
-
-    let index = LineIndex::new(text);
-    for (offset, line, col) in table {
-        assert_eq!(index.line_col(offset.into()), LineCol { line, col });
-    }
-
-    let text = "\nhello\nworld";
-    let table = [(0, 0, 0), (1, 1, 0), (2, 1, 1), (6, 1, 5), (7, 2, 0)];
-    let index = LineIndex::new(text);
-    for (offset, line, col) in table {
-        assert_eq!(index.line_col(offset.into()), LineCol { line, col });
-    }
-}
-
-#[test]
-fn test_char_len() {
-    assert_eq!('メ'.len_utf8(), 3);
-    assert_eq!('メ'.len_utf16(), 1);
-}
-
-#[test]
-fn test_splitlines() {
-    fn r(lo: u32, hi: u32) -> TextRange {
-        TextRange::new(lo.into(), hi.into())
-    }
-
-    let text = "a\nbb\nccc\n";
-    let line_index = LineIndex::new(text);
-
-    let actual = line_index.lines(r(0, 9)).collect::<Vec<_>>();
-    let expected = vec![r(0, 2), r(2, 5), r(5, 9)];
-    assert_eq!(actual, expected);
-
-    let text = "";
-    let line_index = LineIndex::new(text);
-
-    let actual = line_index.lines(r(0, 0)).collect::<Vec<_>>();
-    let expected = vec![];
-    assert_eq!(actual, expected);
-
-    let text = "\n";
-    let line_index = LineIndex::new(text);
-
-    let actual = line_index.lines(r(0, 1)).collect::<Vec<_>>();
-    let expected = vec![r(0, 1)];
-    assert_eq!(actual, expected)
-}