LibCrypto: Don't compute 2*N remainders in Adler32

Otherwise Zlib decompression spends half of the time computing the
checksum.

Userland/Libraries/LibCrypto/Checksum/Adler32.cpp

@@ -12,10 +12,32 @@ namespace Crypto::Checksum {
 
 void Adler32::update(ReadonlyBytes data)
 {
- for (size_t i = 0; i < data.size(); i++) {
- m_state_a = (m_state_a + data.at(i)) % 65521;
- m_state_b = (m_state_b + m_state_a) % 65521;
+ // See https://github.com/SerenityOS/serenity/pull/24408#discussion_r1609051678
+ constexpr size_t iterations_without_overflow = 380368439;
+
+ u64 state_a = m_state_a;
+ u64 state_b = m_state_b;
+ while (data.size()) {
+ // You can verify that no overflow will happen here during at least
+ // `iterations_without_overflow` iterations using the following Python script:
+ //
+ // state_a = 65520
+ // state_b = 65520
+ // for i in range(380368439):
+ // state_a += 255
+ // state_b += state_a
+ // print(state_b < 2 ** 64)
+ auto chunk = data.slice(0, min(data.size(), iterations_without_overflow));
+ for (u8 byte : chunk) {
+ state_a += byte;
+ state_b += state_a;
+ }
+ state_a %= 65521;
+ state_b %= 65521;
+ data = data.slice(chunk.size());
  }
+ m_state_a = state_a;
+ m_state_b = state_b;
 }
 
 u32 Adler32::digest()