From aeb3bf33a9bbf6dd662e9e570accf56c57eafbd7 Mon Sep 17 00:00:00 2001
From: anematode <timothy.herchen@gmail.com>
Date: Wed, 31 Dec 2025 17:44:43 -0800
Subject: [PATCH] port get_changed_pieces to ARM NEON

passed STC:
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 71968 W: 18833 L: 18489 D: 34646
Ptnml(0-2): 192, 7310, 20643, 7640, 199
https://tests.stockfishchess.org/tests/view/69509e5c572093c1986d7a0a

closes https://github.com/official-stockfish/Stockfish/pull/6512

No functional change
---
 src/nnue/nnue_accumulator.cpp | 12 ++++++++++++
 src/search.cpp                |  7 ++++---
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/src/nnue/nnue_accumulator.cpp b/src/nnue/nnue_accumulator.cpp
index 763fb7a5e..338d291e8 100644
--- a/src/nnue/nnue_accumulator.cpp
+++ b/src/nnue/nnue_accumulator.cpp
@@ -645,6 +645,18 @@ Bitboard get_changed_pieces(const std::array<Piece, SQUARE_NB>& oldPieces,
         sameBB |= static_cast<Bitboard>(equalMask) << i;
     }
     return ~sameBB;
+#elif defined(USE_NEON)
+    uint8x16x4_t old_v = vld4q_u8(reinterpret_cast<const uint8_t*>(oldPieces.data()));
+    uint8x16x4_t new_v = vld4q_u8(reinterpret_cast<const uint8_t*>(newPieces.data()));
+    auto         cmp   = [=](const int i) { return vceqq_u8(old_v.val[i], new_v.val[i]); };
+
+    uint8x16_t cmp0_1 = vsriq_n_u8(cmp(1), cmp(0), 1);
+    uint8x16_t cmp2_3 = vsriq_n_u8(cmp(3), cmp(2), 1);
+    uint8x16_t merged = vsriq_n_u8(cmp2_3, cmp0_1, 2);
+    merged            = vsriq_n_u8(merged, merged, 4);
+    uint8x8_t sameBB  = vshrn_n_u16(vreinterpretq_u16_u8(merged), 4);
+
+    return ~vget_lane_u64(vreinterpret_u64_u8(sameBB), 0);
 #else
     Bitboard changed = 0;
 
diff --git a/src/search.cpp b/src/search.cpp
index 05f9b47fa..a880a741f 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -65,7 +65,7 @@ using namespace Search;
 namespace {
 
 constexpr int SEARCHEDLIST_CAPACITY = 32;
-constexpr int mainHistoryDefault = 68;
+constexpr int mainHistoryDefault    = 68;
 using SearchedList                  = ValueList<Move, SEARCHEDLIST_CAPACITY>;
 
 // (*Scalers):
@@ -313,9 +313,10 @@ void Search::Worker::iterative_deepening() {
 
     lowPlyHistory.fill(97);
 
-    for (Color c: {WHITE, BLACK})
+    for (Color c : {WHITE, BLACK})
         for (int i = 0; i < UINT_16_HISTORY_SIZE; i++)
-            mainHistory[c][i] = (mainHistory[c][i] - mainHistoryDefault) * 3 / 4 + mainHistoryDefault;
+            mainHistory[c][i] =
+              (mainHistory[c][i] - mainHistoryDefault) * 3 / 4 + mainHistoryDefault;
 
     // Iterative deepening loop until requested to stop or the target depth is reached
     while (++rootDepth < MAX_PLY && !threads.stop