Optimize LowLevelHash by breaking dependency between final loads and previous len/ptr updates.

No functional change.

PiperOrigin-RevId: 619979819
Change-Id: I71a4ae14272bcdf1da2e201ed286f95213ecb5e2

absl/hash/internal/low_level_hash.cc

@@ -37,6 +37,7 @@ uint64_t LowLevelHashLenGt16(const void* data, size_t len, uint64_t seed,
   PrefetchToLocalCache(data);
   const uint8_t* ptr = static_cast<const uint8_t*>(data);
   uint64_t starting_length = static_cast<uint64_t>(len);
+  const uint8_t* last_16_ptr = ptr + starting_length - 16;
   uint64_t current_state = seed ^ salt[0];
 
   if (len > 64) {
@@ -97,15 +98,12 @@ uint64_t LowLevelHashLenGt16(const void* data, size_t len, uint64_t seed,
     uint64_t b = absl::base_internal::UnalignedLoad64(ptr + 8);
 
     current_state = Mix(a ^ salt[1], b ^ current_state);
-
-    ptr += 16;
-    len -= 16;
   }
 
   // We now have a data `ptr` with at least 1 and at most 16 bytes. But we can
   // safely read from `ptr + len - 16`.
-  uint64_t a = absl::base_internal::UnalignedLoad64(ptr + len - 16);
-  uint64_t b = absl::base_internal::UnalignedLoad64(ptr + len - 8);
+  uint64_t a = absl::base_internal::UnalignedLoad64(last_16_ptr);
+  uint64_t b = absl::base_internal::UnalignedLoad64(last_16_ptr + 8);
 
   return Mix(a ^ salt[1] ^ starting_length, b ^ current_state);
 }