From 041e1704fcd9cc932a4fa587c43d32ed9dcb9712 Mon Sep 17 00:00:00 2001
From: Lzu Tao <taolzu@gmail.com>
Date: Sun, 29 Dec 2019 15:47:13 +0000
Subject: [PATCH 1/2] use of wmemchr for faster searching in [u16]

---
 src/libstd/sys/windows/mod.rs | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/src/libstd/sys/windows/mod.rs b/src/libstd/sys/windows/mod.rs
index 74dd41fd50147..cb75e8122fdd5 100644
--- a/src/libstd/sys/windows/mod.rs
+++ b/src/libstd/sys/windows/mod.rs
@@ -81,10 +81,20 @@ pub fn decode_error_kind(errno: i32) -> ErrorKind {
     }
 }
 
+pub fn wmemchr(needle: u16, haystack: &[u16]) -> Option<usize> {
+    extern "C" {
+        fn wmemchr(s: *const u16, c: u16, n: usize) -> *mut u16;
+    }
+    let len = haystack.len();
+    let ptr = haystack.as_ptr();
+    let p = unsafe { wmemchr(ptr, needle, len) };
+    if p.is_null() { None } else { Some((p as usize - ptr as usize) / 2) }
+}
+
 pub fn to_u16s<S: AsRef<OsStr>>(s: S) -> crate::io::Result<Vec<u16>> {
     fn inner(s: &OsStr) -> crate::io::Result<Vec<u16>> {
         let mut maybe_result: Vec<u16> = s.encode_wide().collect();
-        if maybe_result.iter().any(|&u| u == 0) {
+        if wmemchr(0, &maybe_result).is_some() {
             return Err(crate::io::Error::new(
                 ErrorKind::InvalidInput,
                 "strings passed to WinAPI cannot contain NULs",
@@ -214,7 +224,7 @@ fn wide_char_to_multi_byte(
 }
 
 pub fn truncate_utf16_at_nul(v: &[u16]) -> &[u16] {
-    match v.iter().position(|c| *c == 0) {
+    match wmemchr(0, v) {
         // don't include the 0
         Some(i) => &v[..i],
         None => v,

From 89bc23643bc4ba979f28d6df8c091813c89c36a9 Mon Sep 17 00:00:00 2001
From: lzutao <taolzu@gmail.com>
Date: Thu, 2 Apr 2020 02:43:23 +0000
Subject: [PATCH 2/2] Use unrolled loop

---
 src/libstd/sys/windows/mod.rs | 52 +++++++++++++++++++++++++++++------
 1 file changed, 43 insertions(+), 9 deletions(-)

diff --git a/src/libstd/sys/windows/mod.rs b/src/libstd/sys/windows/mod.rs
index cb75e8122fdd5..d745e87a07258 100644
--- a/src/libstd/sys/windows/mod.rs
+++ b/src/libstd/sys/windows/mod.rs
@@ -81,20 +81,54 @@ pub fn decode_error_kind(errno: i32) -> ErrorKind {
     }
 }
 
-pub fn wmemchr(needle: u16, haystack: &[u16]) -> Option<usize> {
-    extern "C" {
-        fn wmemchr(s: *const u16, c: u16, n: usize) -> *mut u16;
-    }
-    let len = haystack.len();
+pub fn unrolled_find_u16s(needle: u16, haystack: &[u16]) -> Option<usize> {
     let ptr = haystack.as_ptr();
-    let p = unsafe { wmemchr(ptr, needle, len) };
-    if p.is_null() { None } else { Some((p as usize - ptr as usize) / 2) }
+    let mut len = haystack.len();
+    let mut start = &haystack[..];
+
+    // For performance reasons unfold the loop eight times.
+    while len >= 8 {
+        if start[0] == needle {
+            return Some((start.as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[1] == needle {
+            return Some((start[1..].as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[2] == needle {
+            return Some((start[2..].as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[3] == needle {
+            return Some((start[3..].as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[4] == needle {
+            return Some((start[4..].as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[5] == needle {
+            return Some((start[5..].as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[6] == needle {
+            return Some((start[6..].as_ptr() as usize - ptr as usize) / 2);
+        }
+        if start[7] == needle {
+            return Some((start[7..].as_ptr() as usize - ptr as usize) / 2);
+        }
+
+        start = &start[8..];
+        len -= 8;
+    }
+
+    for (i, c) in start.iter().enumerate() {
+        if *c == needle {
+            return Some((start.as_ptr() as usize - ptr as usize) / 2 + i);
+        }
+    }
+    None
 }
 
 pub fn to_u16s<S: AsRef<OsStr>>(s: S) -> crate::io::Result<Vec<u16>> {
     fn inner(s: &OsStr) -> crate::io::Result<Vec<u16>> {
         let mut maybe_result: Vec<u16> = s.encode_wide().collect();
-        if wmemchr(0, &maybe_result).is_some() {
+        if unrolled_find_u16s(0, &maybe_result).is_some() {
             return Err(crate::io::Error::new(
                 ErrorKind::InvalidInput,
                 "strings passed to WinAPI cannot contain NULs",
@@ -224,7 +258,7 @@ fn wide_char_to_multi_byte(
 }
 
 pub fn truncate_utf16_at_nul(v: &[u16]) -> &[u16] {
-    match wmemchr(0, v) {
+    match unrolled_find_u16s(0, v) {
         // don't include the 0
         Some(i) => &v[..i],
         None => v,