From 5afc63a2aef113d49944b6ee5f7a743198a6aff1 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Steinbrink?= <bsteinbr@gmail.com>
Date: Sat, 1 Feb 2014 14:56:51 +0100
Subject: [PATCH] Optimize u64_to_{le,be}_bytes

LLVM fails to properly optimize the shifts used to convert the source
value to the right endianess. The resulting assembly copies the value
to the stack one byte at a time even when there's no conversion required
(e.g. u64_to_le_bytes on a little endian machine).

Instead of doing the conversion ourselves using shifts, we can use the
existing intrinsics to perform the endianess conversion and then
transmute the value to get a fixed vector of its bytes.

Before:

test be_i8  ... bench:     21442 ns/iter (+/- 70)
test be_i16 ... bench:     21447 ns/iter (+/- 45)
test be_i32 ... bench:     23832 ns/iter (+/- 63)
test be_i64 ... bench:     26887 ns/iter (+/- 267)

test le_i8  ... bench:     21442 ns/iter (+/- 56)
test le_i16 ... bench:     21448 ns/iter (+/- 36)
test le_i32 ... bench:     23825 ns/iter (+/- 153)
test le_i64 ... bench:     26271 ns/iter (+/- 138)

After:

test be_i8  ... bench:     21438 ns/iter (+/- 10)
test be_i16 ... bench:     21441 ns/iter (+/- 15)
test be_i32 ... bench:     19057 ns/iter (+/- 6)
test be_i64 ... bench:     21439 ns/iter (+/- 34)

test le_i8  ... bench:     21438 ns/iter (+/- 19)
test le_i16 ... bench:     21439 ns/iter (+/- 8)
test le_i32 ... bench:     21439 ns/iter (+/- 19)
test le_i64 ... bench:     21438 ns/iter (+/- 22)
---
 src/libstd/io/extensions.rs | 42 +++++++++++++------------------------
 1 file changed, 14 insertions(+), 28 deletions(-)

diff --git a/src/libstd/io/extensions.rs b/src/libstd/io/extensions.rs
index f6e87212d17be..548dc3efe92f0 100644
--- a/src/libstd/io/extensions.rs
+++ b/src/libstd/io/extensions.rs
@@ -51,23 +51,16 @@ impl<'r, R: Reader> Iterator<u8> for Bytes<'r, R> {
 }
 
 pub fn u64_to_le_bytes<T>(n: u64, size: uint, f: |v: &[u8]| -> T) -> T {
+    use unstable::intrinsics::{to_le16, to_le32, to_le64};
+    use cast::transmute;
+
+    // LLVM fails to properly optimize this when using shifts instead of the to_le* intrinsics
     assert!(size <= 8u);
     match size {
       1u => f(&[n as u8]),
-      2u => f(&[n as u8,
-              (n >> 8) as u8]),
-      4u => f(&[n as u8,
-              (n >> 8) as u8,
-              (n >> 16) as u8,
-              (n >> 24) as u8]),
-      8u => f(&[n as u8,
-              (n >> 8) as u8,
-              (n >> 16) as u8,
-              (n >> 24) as u8,
-              (n >> 32) as u8,
-              (n >> 40) as u8,
-              (n >> 48) as u8,
-              (n >> 56) as u8]),
+      2u => f(unsafe { transmute::<i16, [u8, ..2]>(to_le16(n as i16)) }),
+      4u => f(unsafe { transmute::<i32, [u8, ..4]>(to_le32(n as i32)) }),
+      8u => f(unsafe { transmute::<i64, [u8, ..8]>(to_le64(n as i64)) }),
       _ => {
 
         let mut bytes: ~[u8] = ~[];
@@ -84,23 +77,16 @@ pub fn u64_to_le_bytes<T>(n: u64, size: uint, f: |v: &[u8]| -> T) -> T {
 }
 
 pub fn u64_to_be_bytes<T>(n: u64, size: uint, f: |v: &[u8]| -> T) -> T {
+    use unstable::intrinsics::{to_be16, to_be32, to_be64};
+    use cast::transmute;
+
+    // LLVM fails to properly optimize this when using shifts instead of the to_be* intrinsics
     assert!(size <= 8u);
     match size {
       1u => f(&[n as u8]),
-      2u => f(&[(n >> 8) as u8,
-              n as u8]),
-      4u => f(&[(n >> 24) as u8,
-              (n >> 16) as u8,
-              (n >> 8) as u8,
-              n as u8]),
-      8u => f(&[(n >> 56) as u8,
-              (n >> 48) as u8,
-              (n >> 40) as u8,
-              (n >> 32) as u8,
-              (n >> 24) as u8,
-              (n >> 16) as u8,
-              (n >> 8) as u8,
-              n as u8]),
+      2u => f(unsafe { transmute::<i16, [u8, ..2]>(to_be16(n as i16)) }),
+      4u => f(unsafe { transmute::<i32, [u8, ..4]>(to_be32(n as i32)) }),
+      8u => f(unsafe { transmute::<i64, [u8, ..8]>(to_be64(n as i64)) }),
       _ => {
         let mut bytes: ~[u8] = ~[];
         let mut i = size;