Move rust memchr impl to libcore

Author: Manishearth

src/libcore/slice/memchr.rs

@@ -0,0 +1,224 @@
+// Copyright 2015 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+//
+// Original implementation taken from rust-memchr
+// Copyright 2015 Andrew Gallant, bluss and Nicolas Koch
+
+use cmp;
+use mem;
+
+const LO_U64: u64 = 0x0101010101010101;
+const HI_U64: u64 = 0x8080808080808080;
+
+// use truncation
+const LO_USIZE: usize = LO_U64 as usize;
+const HI_USIZE: usize = HI_U64 as usize;
+
+/// Return `true` if `x` contains any zero byte.
+///
+/// From *Matters Computational*, J. Arndt
+///
+/// "The idea is to subtract one from each of the bytes and then look for
+/// bytes where the borrow propagated all the way to the most significant
+/// bit."
+#[inline]
+fn contains_zero_byte(x: usize) -> bool {
+    x.wrapping_sub(LO_USIZE) & !x & HI_USIZE != 0
+}
+
+#[cfg(target_pointer_width = "32")]
+#[inline]
+fn repeat_byte(b: u8) -> usize {
+    let mut rep = (b as usize) << 8 | b as usize;
+    rep = rep << 16 | rep;
+    rep
+}
+
+#[cfg(target_pointer_width = "64")]
+#[inline]
+fn repeat_byte(b: u8) -> usize {
+    let mut rep = (b as usize) << 8 | b as usize;
+    rep = rep << 16 | rep;
+    rep = rep << 32 | rep;
+    rep
+}
+
+/// Return the first index matching the byte `a` in `text`.
+pub fn memchr(x: u8, text: &[u8]) -> Option<usize> {
+    // Scan for a single byte value by reading two `usize` words at a time.
+    //
+    // Split `text` in three parts
+    // - unaligned initial part, before the first word aligned address in text
+    // - body, scan by 2 words at a time
+    // - the last remaining part, < 2 word size
+    let len = text.len();
+    let ptr = text.as_ptr();
+    let usize_bytes = mem::size_of::<usize>();
+
+    // search up to an aligned boundary
+    let mut offset = ptr.align_offset(usize_bytes);
+    if offset > 0 {
+        offset = cmp::min(offset, len);
+        if let Some(index) = text[..offset].iter().position(|elt| *elt == x) {
+            return Some(index);
+        }
+    }
+
+    // search the body of the text
+    let repeated_x = repeat_byte(x);
+
+    if len >= 2 * usize_bytes {
+        while offset <= len - 2 * usize_bytes {
+            unsafe {
+                let u = *(ptr.offset(offset as isize) as *const usize);
+                let v = *(ptr.offset((offset + usize_bytes) as isize) as *const usize);
+
+                // break if there is a matching byte
+                let zu = contains_zero_byte(u ^ repeated_x);
+                let zv = contains_zero_byte(v ^ repeated_x);
+                if zu || zv {
+                    break;
+                }
+            }
+            offset += usize_bytes * 2;
+        }
+    }
+
+    // find the byte after the point the body loop stopped
+    text[offset..].iter().position(|elt| *elt == x).map(|i| offset + i)
+}
+
+/// Return the last index matching the byte `a` in `text`.
+pub fn memrchr(x: u8, text: &[u8]) -> Option<usize> {
+    // Scan for a single byte value by reading two `usize` words at a time.
+    //
+    // Split `text` in three parts
+    // - unaligned tail, after the last word aligned address in text
+    // - body, scan by 2 words at a time
+    // - the first remaining bytes, < 2 word size
+    let len = text.len();
+    let ptr = text.as_ptr();
+    let usize_bytes = mem::size_of::<usize>();
+
+    // search to an aligned boundary
+    let end_align = (ptr as usize + len) & (usize_bytes - 1);
+    let mut offset;
+    if end_align > 0 {
+        offset = if end_align >= len { 0 } else { len - end_align };
+        if let Some(index) = text[offset..].iter().rposition(|elt| *elt == x) {
+            return Some(offset + index);
+        }
+    } else {
+        offset = len;
+    }
+
+    // search the body of the text
+    let repeated_x = repeat_byte(x);
+
+    while offset >= 2 * usize_bytes {
+        unsafe {
+            let u = *(ptr.offset(offset as isize - 2 * usize_bytes as isize) as *const usize);
+            let v = *(ptr.offset(offset as isize - usize_bytes as isize) as *const usize);
+
+            // break if there is a matching byte
+            let zu = contains_zero_byte(u ^ repeated_x);
+            let zv = contains_zero_byte(v ^ repeated_x);
+            if zu || zv {
+                break;
+            }
+        }
+        offset -= 2 * usize_bytes;
+    }
+
+    // find the byte before the point the body loop stopped
+    text[..offset].iter().rposition(|elt| *elt == x)
+}
+
+// test fallback implementations on all platforms
+#[test]
+fn matches_one() {
+    assert_eq!(Some(0), memchr(b'a', b"a"));
+}
+
+#[test]
+fn matches_begin() {
+    assert_eq!(Some(0), memchr(b'a', b"aaaa"));
+}
+
+#[test]
+fn matches_end() {
+    assert_eq!(Some(4), memchr(b'z', b"aaaaz"));
+}
+
+#[test]
+fn matches_nul() {
+    assert_eq!(Some(4), memchr(b'\x00', b"aaaa\x00"));
+}
+
+#[test]
+fn matches_past_nul() {
+    assert_eq!(Some(5), memchr(b'z', b"aaaa\x00z"));
+}
+
+#[test]
+fn no_match_empty() {
+    assert_eq!(None, memchr(b'a', b""));
+}
+
+#[test]
+fn no_match() {
+    assert_eq!(None, memchr(b'a', b"xyz"));
+}
+
+#[test]
+fn matches_one_reversed() {
+    assert_eq!(Some(0), memrchr(b'a', b"a"));
+}
+
+#[test]
+fn matches_begin_reversed() {
+    assert_eq!(Some(3), memrchr(b'a', b"aaaa"));
+}
+
+#[test]
+fn matches_end_reversed() {
+    assert_eq!(Some(0), memrchr(b'z', b"zaaaa"));
+}
+
+#[test]
+fn matches_nul_reversed() {
+    assert_eq!(Some(4), memrchr(b'\x00', b"aaaa\x00"));
+}
+
+#[test]
+fn matches_past_nul_reversed() {
+    assert_eq!(Some(0), memrchr(b'z', b"z\x00aaaa"));
+}
+
+#[test]
+fn no_match_empty_reversed() {
+    assert_eq!(None, memrchr(b'a', b""));
+}
+
+#[test]
+fn no_match_reversed() {
+    assert_eq!(None, memrchr(b'a', b"xyz"));
+}
+
+#[test]
+fn each_alignment_reversed() {
+    let mut data = [1u8; 64];
+    let needle = 2;
+    let pos = 40;
+    data[pos] = needle;
+    for start in 0..16 {
+        assert_eq!(Some(pos - start), memrchr(needle, &data[start..]));
+    }
+}

src/libcore/slice/mod.rs

@@ -50,6 +50,11 @@ use mem;
 use marker::{Copy, Send, Sync, Sized, self};
 use iter_private::TrustedRandomAccess;
 
+#[unstable(feature = "slice_internals", issue = "0",
+           reason = "exposed from core to be reused in std; use the memchr crate")]
+/// Pure rust memchr implementation, taken from rust-memchr
+pub mod memchr;
+
 mod rotate;
 mod sort;
 

src/libstd/lib.rs

@@ -302,6 +302,7 @@
 #![feature(sip_hash_13)]
 #![feature(slice_bytes)]
 #![feature(slice_concat_ext)]
+#![feature(slice_internals)]
 #![feature(slice_patterns)]
 #![feature(staged_api)]
 #![feature(stmt_expr_attributes)]

src/libstd/sys/redox/memchr.rs

@@ -11,4 +11,4 @@
 // Original implementation taken from rust-memchr
 // Copyright 2015 Andrew Gallant, bluss and Nicolas Koch
 
-pub use sys_common::memchr::fallback::{memchr, memrchr};
+pub use core::slice::memchr::{memchr, memrchr};

src/libstd/sys/unix/memchr.rs

@@ -50,7 +50,7 @@ pub fn memrchr(needle: u8, haystack: &[u8]) -> Option<usize> {
 
     #[cfg(not(target_os = "linux"))]
     fn memrchr_specific(needle: u8, haystack: &[u8]) -> Option<usize> {
-        ::sys_common::memchr::fallback::memrchr(needle, haystack)
+        ::core::slice::memchr::memrchr(needle, haystack)
     }
 
     memrchr_specific(needle, haystack)

src/libstd/sys/wasm/memchr.rs

@@ -8,4 +8,4 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-pub use sys_common::memchr::fallback::{memchr, memrchr};
+pub use core::slice::memchr::{memchr, memrchr};

src/libstd/sys/windows/memchr.rs

@@ -12,4 +12,4 @@
 // Copyright 2015 Andrew Gallant, bluss and Nicolas Koch
 
 // Fallback memchr is fastest on windows
-pub use sys_common::memchr::fallback::{memchr, memrchr};
+pub use core::slice::memchr::{memchr, memrchr};