From 0452a8ed4bf529699047218fcf624ead56b6e168 Mon Sep 17 00:00:00 2001
From: Nico Weber <thakis@chromium.org>
Date: Sat, 25 Mar 2023 21:18:38 +0100
Subject: [PATCH] LibCrypto: Start sometimes hardware-accelerating crc32

Takes

    % time Build/lagom/gunzip -c \
        /Users/thakis/Downloads/trace_bug.json.gz > /dev/null

from 4s to 3.9s on my MBP.
---
 .../Libraries/LibCrypto/Checksum/CRC32.cpp     | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/Userland/Libraries/LibCrypto/Checksum/CRC32.cpp b/Userland/Libraries/LibCrypto/Checksum/CRC32.cpp
index 0a0ed2e861..805bff2546 100644
--- a/Userland/Libraries/LibCrypto/Checksum/CRC32.cpp
+++ b/Userland/Libraries/LibCrypto/Checksum/CRC32.cpp
@@ -11,6 +11,22 @@
 
 namespace Crypto::Checksum {
 
+#if defined(__aarch64__) && defined(__ARM_FEATURE_CRC32)
+
+void CRC32::update(ReadonlyBytes data)
+{
+    // FIXME: Does this require runtime checking on rpi?
+    //        (Maybe the instruction is present on the rpi4 but not on the rpi3?)
+
+    // FIXME: Use __builtin_arm_crc32d() for aligned middle part.
+    for (size_t i = 0; i < data.size(); i++)
+        m_state = __builtin_arm_crc32b(m_state, data.at(i));
+};
+
+    // FIXME: On Intel, use _mm_crc32_u8 / _mm_crc32_u64 if available (SSE 4.2).
+
+#else
+
 static constexpr auto generate_table()
 {
     Array<u32, 256> data {};
@@ -39,6 +55,8 @@ void CRC32::update(ReadonlyBytes data)
     }
 };
 
+#endif
+
 u32 CRC32::digest()
 {
     return ~m_state;