From 61e9e82bdc10110b74bdeb973cc542c7b73a4ae2 Mon Sep 17 00:00:00 2001
From: LemonBoy <thatlemon@gmail.com>
Date: Sun, 13 Sep 2020 21:12:21 +0200
Subject: [PATCH] std: Make the CRC32 calculation slightly faster

Speed up a little the slicing-by-8 code path by replacing the
(load+shift+xor)*4 sequence with a single u32 load plus a xor.

Before:

```
iterative:  1018 MiB/s [000000006c3b110d]
small keys:  1075 MiB/s [0035bf3dcac00000]
```

After:

```
iterative:  1114 MiB/s [000000006c3b110d]
small keys:  1324 MiB/s [0035bf3dcac00000]
```
---
 lib/std/hash/crc.zig | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/lib/std/hash/crc.zig b/lib/std/hash/crc.zig
index 37695df8b3..6290369fca 100644
--- a/lib/std/hash/crc.zig
+++ b/lib/std/hash/crc.zig
@@ -71,10 +71,7 @@ pub fn Crc32WithPoly(comptime poly: Polynomial) type {
                 const p = input[i .. i + 8];
 
                 // Unrolling this way gives ~50Mb/s increase
-                self.crc ^= (@as(u32, p[0]) << 0);
-                self.crc ^= (@as(u32, p[1]) << 8);
-                self.crc ^= (@as(u32, p[2]) << 16);
-                self.crc ^= (@as(u32, p[3]) << 24);
+                self.crc ^= std.mem.readIntLittle(u32, p[0..4]);
 
                 self.crc =
                     lookup_tables[0][p[7]] ^