diff --git a/Cargo.lock b/Cargo.lock
index c071a2e11d269..2271800fe327f 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2683,6 +2683,7 @@ dependencies = [
 name = "rustc_data_structures"
 version = "0.0.0"
 dependencies = [
+ "arrayref 0.3.5 (registry+https://github.com/rust-lang/crates.io-index)",
  "cfg-if 0.1.6 (registry+https://github.com/rust-lang/crates.io-index)",
  "ena 0.13.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "graphviz 0.0.0",
diff --git a/src/librustc_data_structures/Cargo.toml b/src/librustc_data_structures/Cargo.toml
index 63e44d82a28c3..0bf45bf5f6b62 100644
--- a/src/librustc_data_structures/Cargo.toml
+++ b/src/librustc_data_structures/Cargo.toml
@@ -10,6 +10,7 @@ path = "lib.rs"
 crate-type = ["dylib"]
 
 [dependencies]
+arrayref = "0.3.5"
 ena = "0.13"
 log = "0.4"
 jobserver_crate = { version = "0.1", package = "jobserver" }
diff --git a/src/librustc_data_structures/ahash.rs b/src/librustc_data_structures/ahash.rs
new file mode 100644
index 0000000000000..ba55a64c561be
--- /dev/null
+++ b/src/librustc_data_structures/ahash.rs
@@ -0,0 +1,271 @@
+pub use rustc_hash::{FxHasher, FxHashMap, FxHashSet};
+use crate::convert::Convert;
+use std::hash::{Hasher};
+use std::slice;
+use std::mem;
+use std::sync::atomic::AtomicBool;
+use std::sync::atomic::Ordering::Relaxed;
+use arrayref::*;
+
+const BUFFER_SIZE: usize = 1024;
+
+pub struct BufferedHasher {
+    cursor: usize,
+    aes: AHasher,
+    buffer: [u8; BUFFER_SIZE],
+}
+
+impl ::std::fmt::Debug for BufferedHasher {
+    fn fmt(&self, f: &mut ::std::fmt::Formatter<'_>) -> ::std::fmt::Result {
+        write!(f, "{:?} [{}]", self.aes, self.cursor)
+    }
+}
+
+#[inline(never)]
+#[no_mangle]
+#[target_feature(enable = "aes")]
+unsafe fn ahash_test(input: &[u8]) -> u64 {
+    let mut a = AHasher::new_with_keys(67, 87);
+    a.write(input);
+    a.finish128().0
+}
+
+#[inline(never)]
+#[no_mangle]
+fn ahash_test_one(b: &mut BufferedHasher, input: u64) {
+    b.write_u64(input);
+}
+
+static AES_ENABLED: AtomicBool = AtomicBool::new(true);
+
+impl BufferedHasher {
+    #[inline(always)]
+    pub fn new() -> Self {
+        BufferedHasher {
+            aes: AHasher::new_with_keys(0, 0),
+            cursor: 0,
+            buffer: unsafe { std::mem::uninitialized() },
+        }
+    }
+
+    #[inline(always)]
+    unsafe fn flush(&mut self) {
+        self.aes.write(self.buffer.get_unchecked(0..self.cursor));
+    }
+
+    #[inline(never)]
+    #[cold]
+    unsafe fn flush_cold(&mut self) {
+        if likely!(AES_ENABLED.load(Relaxed)) {
+            self.flush()
+        } else {
+            panic!("no aes");
+        }
+    }
+
+    #[inline(always)]
+    fn short_write_gen<T>(&mut self, x: T) {
+        let bytes = unsafe {
+            slice::from_raw_parts(&x as *const T as *const u8, mem::size_of::<T>())
+        };
+        self.short_write(bytes);
+    }
+
+    #[inline(always)]
+    fn short_write(&mut self, data: &[u8]) {
+        let mut cursor = self.cursor;
+        let len = data.len();
+        if unlikely!(cursor + len > BUFFER_SIZE) {
+            unsafe { self.flush_cold() };
+            cursor = 0;
+        }
+        unsafe {
+            self.buffer.get_unchecked_mut(cursor..(cursor + len)).copy_from_slice(data);
+        }
+        self.cursor = cursor + len;
+    }
+
+    #[inline(never)]
+    #[target_feature(enable = "aes")]
+    unsafe fn finish128_aes(mut self) -> (u64, u64) {
+        self.flush();
+        self.aes.finish128()
+    }
+
+    #[inline]
+    pub fn finish128(self) -> (u64, u64) {
+        if likely!(AES_ENABLED.load(Relaxed)) {
+            unsafe {
+                self.finish128_aes()
+            }
+        } else {
+            panic!("no aes");
+        }
+    }
+}
+
+impl Hasher for BufferedHasher {
+    #[inline]
+    fn write_u8(&mut self, i: u8) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_u16(&mut self, i: u16) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_u32(&mut self, i: u32) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_u64(&mut self, i: u64) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_usize(&mut self, i: usize) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_i8(&mut self, i: i8) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_i16(&mut self, i: i16) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_i32(&mut self, i: i32) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_i64(&mut self, i: i64) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write_isize(&mut self, i: isize) {
+        self.short_write_gen(i);
+    }
+
+    #[inline]
+    fn write(&mut self, data: &[u8]) {
+        if likely!(data.len() < BUFFER_SIZE / 10) {
+            self.short_write(data);
+        } else {
+            unsafe {
+                self.aes.write(data);
+            }
+        }
+    }
+
+    fn finish(&self) -> u64 {
+        panic!("cannot provide valid 64 bit hashes")
+    }
+}
+
+///Just a simple bit pattern.
+const PAD : u128 = 0xF0E1_D2C3_B4A5_9687_7869_5A4B_3C2D_1E0F;
+
+#[derive(Debug, Clone)]
+pub struct AHasher {
+    buffer: [u64; 2],
+}
+
+impl AHasher {
+    #[inline]
+    pub(crate) fn new_with_keys(key0: u64, key1: u64) -> AHasher {
+        AHasher { buffer: [key0, key1] }
+    }
+
+    #[inline]
+    #[target_feature(enable = "aes")]
+    unsafe fn write(&mut self, input: &[u8]) {
+        let mut data = input;
+        let length = data.len() as u64;
+        //This will be scrambled by the first AES round in any branch.
+        self.buffer[1] ^= length;
+        //A 'binary search' on sizes reduces the number of comparisons.
+        if data.len() >= 8 {
+            if data.len() > 16 {
+                if data.len() > 128 {
+                    let mut par_block: u128 = self.buffer.convert();
+                    while data.len() > 128 {
+                        let (b1, rest) = data.split_at(16);
+                        let b1: u128 = (*as_array!(b1, 16)).convert();
+                        par_block = aeshash(par_block, b1);
+                        data = rest;
+                        let (b2, rest) = data.split_at(16);
+                        let b2: u128 = (*as_array!(b2, 16)).convert();
+                        self.buffer = aeshash(self.buffer.convert(), b2).convert();
+                        data = rest;
+                    }
+                    self.buffer = aeshash(self.buffer.convert(), par_block).convert();
+                }
+                while data.len() > 32 {
+                    //len 33-128
+                    let (block, rest) = data.split_at(16);
+                    let block: u128 = (*as_array!(block, 16)).convert();
+                    self.buffer = aeshash(self.buffer.convert(),block).convert();
+                    data = rest;
+                }
+                //len 17-32
+                let block = (*array_ref!(data, 0, 16)).convert();
+                self.buffer = aeshash(self.buffer.convert(),block).convert();
+                let block = (*array_ref!(data, data.len()-16, 16)).convert();
+                self.buffer = aeshash(self.buffer.convert(),block).convert();
+            } else {
+                //len 8-16
+                let block: [u64; 2] = [(*array_ref!(data, 0, 8)).convert(),
+                    (*array_ref!(data, data.len()-8, 8)).convert()];
+                self.buffer = aeshash(self.buffer.convert(),block.convert()).convert();
+            }
+        } else {
+            if data.len() >= 2 {
+                if data.len() >= 4 {
+                    //len 4-7
+                    let block: [u32; 2] = [(*array_ref!(data, 0, 4)).convert(),
+                        (*array_ref!(data, data.len()-4, 4)).convert()];
+                    let block: [u64;2] = [block[1] as u64, block[0] as u64];
+                    self.buffer = aeshash(self.buffer.convert(),block.convert()).convert()
+                } else {
+                    //len 2-3
+                    let block: [u16; 2] = [(*array_ref!(data, 0, 2)).convert(),
+                        (*array_ref!(data, data.len()-2, 2)).convert()];
+                    let block: u32 = block.convert();
+                    self.buffer = aeshash(self.buffer.convert(), block as u128).convert();
+                }
+            } else {
+                if data.len() > 0 {
+                    //len 1
+                    self.buffer = aeshash(self.buffer.convert(), data[0] as u128).convert();
+                }
+            }
+        }
+    }
+    #[inline]
+    #[target_feature(enable = "aes")]
+    unsafe fn finish128(self) -> (u64, u64) {
+        let result: [u64; 2] = aeshash(aeshash(self.buffer.convert(), PAD), PAD).convert();
+        (result[0], result[1])
+    }
+}
+
+#[inline]
+#[target_feature(enable = "aes")]
+unsafe fn aeshash(value: u128, xor: u128) -> u128 {
+    use std::mem::transmute;
+    #[cfg(target_arch = "x86")]
+    use core::arch::x86::*;
+    #[cfg(target_arch = "x86_64")]
+    use std::arch::x86_64::*;
+    let value = transmute(value);
+    transmute(_mm_aesdec_si128(value, transmute(xor)))
+}
diff --git a/src/librustc_data_structures/convert.rs b/src/librustc_data_structures/convert.rs
new file mode 100644
index 0000000000000..1b47b81629c45
--- /dev/null
+++ b/src/librustc_data_structures/convert.rs
@@ -0,0 +1,81 @@
+use std::mem::transmute;
+
+pub(crate) trait Convert<To> {
+    fn convert(self) -> To;
+    fn convert_ref(&self) -> &To;
+    fn convert_mut_ref(&mut self) -> &mut To;
+}
+macro_rules! convert {
+    ($from:ty, $to:ty) => {
+        impl Convert<$to> for $from {
+            #[inline(always)]
+            fn convert(self) -> $to {
+                unsafe { transmute(self) }
+            }
+            #[inline(always)]
+            fn convert_ref(&self) -> &$to {
+                unsafe { transmute(self) }
+            }
+            #[inline(always)]
+            fn convert_mut_ref(&mut self) -> &mut $to {
+                unsafe { transmute(self) }
+            }
+        }
+        impl Convert<$from> for $to {
+            #[inline(always)]
+            fn convert(self) -> $from {
+                unsafe { transmute(self) }
+            }
+            #[inline(always)]
+            fn convert_ref(&self) -> &$from {
+                unsafe { transmute(self) }
+            }
+            #[inline(always)]
+            fn convert_mut_ref(&mut self) -> &mut $from {
+                unsafe { transmute(self) }
+            }
+        }
+    };
+}
+convert!(u128, [u64; 2]);
+convert!(u128, [u32; 4]);
+convert!(u128, [u16; 8]);
+convert!(u128, [u8; 16]);
+convert!([u64; 2], [u32; 4]);
+convert!([u64; 2], [u16; 8]);
+convert!([u64; 2], [u8; 16]);
+convert!([u32; 4], [u16; 8]);
+convert!([u32; 4], [u8; 16]);
+convert!([u16; 8], [u8; 16]);
+convert!(u64, [u32; 2]);
+convert!(u64, [u16; 4]);
+convert!(u64, [u8; 8]);
+convert!([u32; 2], [u16; 4]);
+convert!([u32; 2], [u8; 8]);
+convert!(u32, [u16; 2]);
+convert!(u32, [u8; 4]);
+convert!([u16; 2], [u8; 4]);
+convert!(u16, [u8; 2]);
+
+convert!([f64; 2], [u8; 16]);
+convert!([f32; 4], [u8; 16]);
+convert!(f64, [u8; 8]);
+convert!([f32; 2], [u8; 8]);
+convert!(f32, [u8; 4]);
+
+
+
+macro_rules! as_array {
+    ($input:expr, $len:expr) => {{
+        {
+            #[inline]
+            fn as_array<T>(slice: &[T]) -> &[T; $len] {
+                assert_eq!(slice.len(), $len);
+                unsafe {
+                    &*(slice.as_ptr() as *const [_; $len])
+                }
+            }
+            as_array($input)
+        }
+    }}
+}
diff --git a/src/librustc_data_structures/lib.rs b/src/librustc_data_structures/lib.rs
index a1d7ab8856daa..029cfb6c483f0 100644
--- a/src/librustc_data_structures/lib.rs
+++ b/src/librustc_data_structures/lib.rs
@@ -69,6 +69,9 @@ macro_rules! unlikely {
       }
 }
 
+#[macro_use]
+pub mod convert;
+pub mod ahash;
 pub mod macros;
 pub mod svh;
 pub mod base_n;
diff --git a/src/librustc_data_structures/stable_hasher.rs b/src/librustc_data_structures/stable_hasher.rs
index 19343a9250df3..cdb5ed84cb50b 100644
--- a/src/librustc_data_structures/stable_hasher.rs
+++ b/src/librustc_data_structures/stable_hasher.rs
@@ -1,7 +1,7 @@
 use std::hash::{Hash, Hasher, BuildHasher};
 use std::marker::PhantomData;
 use std::mem;
-use crate::sip128::SipHasher128;
+use crate::ahash::BufferedHasher;
 use crate::indexed_vec;
 use crate::bit_set;
 
@@ -13,8 +13,7 @@ use crate::bit_set;
 /// hashing and the architecture dependent `isize` and `usize` types are
 /// extended to 64 bits if needed.
 pub struct StableHasher<W> {
-    state: SipHasher128,
-    bytes_hashed: u64,
+    state: BufferedHasher,
     width: PhantomData<W>,
 }
 
@@ -29,10 +28,10 @@ pub trait StableHasherResult: Sized {
 }
 
 impl<W: StableHasherResult> StableHasher<W> {
+    #[inline(always)]
     pub fn new() -> Self {
         StableHasher {
-            state: SipHasher128::new_with_keys(0, 0),
-            bytes_hashed: 0,
+            state: BufferedHasher::new(),
             width: PhantomData,
         }
     }
@@ -60,11 +59,6 @@ impl<W> StableHasher<W> {
     pub fn finalize(self) -> (u64, u64) {
         self.state.finish128()
     }
-
-    #[inline]
-    pub fn bytes_hashed(&self) -> u64 {
-        self.bytes_hashed
-    }
 }
 
 impl<W> Hasher for StableHasher<W> {
@@ -75,37 +69,31 @@ impl<W> Hasher for StableHasher<W> {
     #[inline]
     fn write(&mut self, bytes: &[u8]) {
         self.state.write(bytes);
-        self.bytes_hashed += bytes.len() as u64;
     }
 
     #[inline]
     fn write_u8(&mut self, i: u8) {
         self.state.write_u8(i);
-        self.bytes_hashed += 1;
     }
 
     #[inline]
     fn write_u16(&mut self, i: u16) {
         self.state.write_u16(i.to_le());
-        self.bytes_hashed += 2;
     }
 
     #[inline]
     fn write_u32(&mut self, i: u32) {
         self.state.write_u32(i.to_le());
-        self.bytes_hashed += 4;
     }
 
     #[inline]
     fn write_u64(&mut self, i: u64) {
         self.state.write_u64(i.to_le());
-        self.bytes_hashed += 8;
     }
 
     #[inline]
     fn write_u128(&mut self, i: u128) {
         self.state.write_u128(i.to_le());
-        self.bytes_hashed += 16;
     }
 
     #[inline]
@@ -114,37 +102,31 @@ impl<W> Hasher for StableHasher<W> {
         // platforms. This is important for symbol hashes when cross compiling,
         // for example.
         self.state.write_u64((i as u64).to_le());
-        self.bytes_hashed += 8;
     }
 
     #[inline]
     fn write_i8(&mut self, i: i8) {
         self.state.write_i8(i);
-        self.bytes_hashed += 1;
     }
 
     #[inline]
     fn write_i16(&mut self, i: i16) {
         self.state.write_i16(i.to_le());
-        self.bytes_hashed += 2;
     }
 
     #[inline]
     fn write_i32(&mut self, i: i32) {
         self.state.write_i32(i.to_le());
-        self.bytes_hashed += 4;
     }
 
     #[inline]
     fn write_i64(&mut self, i: i64) {
         self.state.write_i64(i.to_le());
-        self.bytes_hashed += 8;
     }
 
     #[inline]
     fn write_i128(&mut self, i: i128) {
         self.state.write_i128(i.to_le());
-        self.bytes_hashed += 16;
     }
 
     #[inline]
@@ -153,7 +135,6 @@ impl<W> Hasher for StableHasher<W> {
         // platforms. This is important for symbol hashes when cross compiling,
         // for example.
         self.state.write_i64((i as i64).to_le());
-        self.bytes_hashed += 8;
     }
 }
 
diff --git a/src/tools/tidy/src/deps.rs b/src/tools/tidy/src/deps.rs
index 30fe327cac4ff..1218ae09876fa 100644
--- a/src/tools/tidy/src/deps.rs
+++ b/src/tools/tidy/src/deps.rs
@@ -59,6 +59,7 @@ const WHITELIST: &[Crate<'_>] = &[
     Crate("adler32"),
     Crate("aho-corasick"),
     Crate("arrayvec"),
+    Crate("arrayref"),
     Crate("atty"),
     Crate("backtrace"),
     Crate("backtrace-sys"),