From 2deb08a52946379d4cebb1082e5d740d1d027122 Mon Sep 17 00:00:00 2001
From: SFisGOD <jonathandumale@gmail.com>
Date: Tue, 18 Aug 2020 18:54:28 +0800
Subject: [PATCH 01/52] Reintroduce last captures extension

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 34840 W: 3834 L: 3682 D: 27324
Ptnml(0-2): 153, 2767, 11455, 2865, 180
https://tests.stockfishchess.org/tests/view/5f3bb380b38d442594aabefc

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 15832 W: 890 L: 776 D: 14166
Ptnml(0-2): 17, 669, 6429, 785, 16
https://tests.stockfishchess.org/tests/view/5f3c46a0a95672ddd56c632a

closes https://github.com/official-stockfish/Stockfish/pull/3028

see also https://github.com/official-stockfish/Stockfish/pull/3020

Bench: 4348811
---
 src/search.cpp | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/search.cpp b/src/search.cpp
index 1d5bc5f7..7c839dfc 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1122,6 +1122,11 @@ moves_loop: // When in check, search starts from here
                && (pos.is_discovery_check_on_king(~us, move) || pos.see_ge(move)))
           extension = 1;
 
+      // Last captures extension
+      else if (   PieceValue[EG][pos.captured_piece()] > PawnValueEg
+               && pos.non_pawn_material() <= 2 * RookValueMg)
+          extension = 1;
+
       // Castling extension
       if (   type_of(move) == CASTLING
           && popcount(pos.pieces(us) & ~pos.pieces(PAWN) & (to_sq(move) & KingSide ? KingSide : QueenSide)) <= 2)

From a1ad8604a11459a94189f857e368d0fbb72da25d Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Wed, 19 Aug 2020 19:21:41 +0200
Subject: [PATCH 02/52] Send error message as an UCI info string

some GUIs do not show the error message when the engine terminates in the no-net case, as it is send to cerr.
Instead send it as an info string, which the GUI will more likely display.

closes https://github.com/official-stockfish/Stockfish/pull/3031

No functional change.
---
 src/evaluate.cpp | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index 1bd89353..c84d894f 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -53,10 +53,11 @@ namespace Eval {
         UCI::OptionsMap defaults;
         UCI::init(defaults);
 
-        std::cerr << "NNUE evaluation used, but the network file " << eval_file << " was not loaded successfully. "
-                  << "These network evaluation parameters must be available, and compatible with this version of the code. "
-                  << "The UCI option EvalFile might need to specify the full path, including the directory/folder name, to the file. "
-                  << "The default net can be downloaded from: https://tests.stockfishchess.org/api/nn/"+std::string(defaults["EvalFile"]) << std::endl;
+        sync_cout << "info string ERROR: NNUE evaluation used, but the network file " << eval_file << " was not loaded successfully." << sync_endl;
+        sync_cout << "info string ERROR: The UCI option EvalFile might need to specify the full path, including the directory/folder name, to the file." << sync_endl;
+        sync_cout << "info string ERROR: The default net can be downloaded from: https://tests.stockfishchess.org/api/nn/"+std::string(defaults["EvalFile"]) << sync_endl;
+        sync_cout << "info string ERROR: If the UCI option Use NNUE is set to true, network evaluation parameters compatible with the program must be available." << sync_endl;
+        sync_cout << "info string ERROR: The engine will be terminated now." << sync_endl;
         std::exit(EXIT_FAILURE);
     }
 

From daac86691de55fe388b9b727794c7d27f2b90d5c Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Thu, 20 Aug 2020 14:24:49 +0200
Subject: [PATCH 03/52] Set Use NNUE by default to true

Since the initial stages of the merge, progress has been made so that
this seems the best option now:

* NNUE is clearly stronger on most relevant hardware and time controls
* All of our CI and testing infrastructure has been adjusted
* The default net is easy to get (further ideas #3030)

fixes https://github.com/official-stockfish/Stockfish/issues/2861

closes https://github.com/official-stockfish/Stockfish/pull/3033

No functional change.
---
 src/ucioption.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/ucioption.cpp b/src/ucioption.cpp
index 2b66a475..ec83c7c8 100644
--- a/src/ucioption.cpp
+++ b/src/ucioption.cpp
@@ -78,7 +78,7 @@ void init(OptionsMap& o) {
   o["SyzygyProbeDepth"]      << Option(1, 1, 100);
   o["Syzygy50MoveRule"]      << Option(true);
   o["SyzygyProbeLimit"]      << Option(7, 0, 7);
-  o["Use NNUE"]              << Option(false, on_use_NNUE);
+  o["Use NNUE"]              << Option(true, on_use_NNUE);
   // The default must follow the format nn-[SHA256 first 12 digits].nnue
   // for the build process (profile-build and fishtest) to work.
   o["EvalFile"]              << Option("nn-82215d0fd0df.nnue", on_eval_file);

From 8b45b1c4907b4b2186441e02edd3b0c37f8b3269 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Fri, 21 Aug 2020 07:42:19 +0200
Subject: [PATCH 04/52] Deal with very old linux kernels

MADV_HUGEPAGE might not be available, for kernels before 2.6.38 (released 2011). Just skip the madvise.

closes https://github.com/official-stockfish/Stockfish/pull/3039

No functional change
---
 src/misc.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/misc.cpp b/src/misc.cpp
index 459ea100..56a3dcad 100644
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -367,7 +367,9 @@ void* aligned_ttmem_alloc(size_t allocSize, void*& mem) {
   size_t size = ((allocSize + alignment - 1) / alignment) * alignment; // multiple of alignment
   if (posix_memalign(&mem, alignment, size))
      mem = nullptr;
+#if defined(MADV_HUGEPAGE)
   madvise(mem, allocSize, MADV_HUGEPAGE);
+#endif
   return mem;
 }
 

From 15abcaedc1e32d4de913a2f7dea12578912371b7 Mon Sep 17 00:00:00 2001
From: gsobala <gsobala@gmail.com>
Date: Fri, 21 Aug 2020 11:28:53 +0100
Subject: [PATCH 05/52] Update Makefile for macOS

Changes to deal with compilation (particularly profile-build) on macOS.
(1) The default toolchain has gcc masquerading as clang,
    the previous Makefile was not picking up the required changes
    to the different profiling tools.
(2) The previous Makefile test for gccisclang occurred before
    a potential overwrite of CXX by COMPCXX
(3) llvm-profdata no longer runs as a command on macOS and
    instead is invoked by ``xcrun llvm-profdata``
(4) Needs to support use of true gcc using e.g.
    ``make build ... COMPCXX=g++-10``
(5) enable profile-build in travis for macOS

closes https://github.com/official-stockfish/Stockfish/pull/3043

No functional change
---
 .travis.yml  |  3 ++-
 AUTHORS      |  1 +
 src/Makefile | 16 ++++++++++++----
 3 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/.travis.yml b/.travis.yml
index 12596f1e..a029c4fc 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -71,7 +71,8 @@ script:
   - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-32-sse2 build && ../tests/signature.sh $benchref; fi
   - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-32 build && ../tests/signature.sh $benchref; fi
   - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=general-32 build && ../tests/signature.sh $benchref; fi
-  - if [[ "$TRAVIS_OS_NAME" == "linux" && "$COMP" == "gcc" ]]; then make clean && make -j2 ARCH=x86-64-modern profile-build && ../tests/signature.sh $benchref; fi
+  # workaround: exclude a custom version of llvm+clang, which doesn't find llvm-profdata on ubuntu
+  - if [[ "$TRAVIS_OS_NAME" != "linux" || "$COMP" == "gcc" ]]; then make clean && make -j2 ARCH=x86-64-modern profile-build && ../tests/signature.sh $benchref; fi
 
   # compile only for some more advanced architectures (might not run in travis)
   - make clean && make -j2 ARCH=x86-64-avx2 build
diff --git a/AUTHORS b/AUTHORS
index d8f4d30e..c96f870a 100644
--- a/AUTHORS
+++ b/AUTHORS
@@ -59,6 +59,7 @@ Fauzi Akram Dabat (FauziAkram)
 Felix Wittmann
 gamander
 Gary Heckman (gheckman)
+George Sobala (gsobala)
 gguliash
 Gian-Carlo Pascutto (gcp)
 Gontran Lemaire (gonlem)
diff --git a/src/Makefile b/src/Makefile
index 79c7333a..b969ba04 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -302,9 +302,6 @@ ifeq ($(COMP),gcc)
 	ifneq ($(KERNEL),Darwin)
 	   LDFLAGS += -Wl,--no-as-needed
 	endif
-
-	gccversion = $(shell $(CXX) --version)
-	gccisclang = $(findstring clang,$(gccversion))
 endif
 
 ifeq ($(COMP),mingw)
@@ -376,6 +373,7 @@ endif
 ifeq ($(KERNEL),Darwin)
 	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.14
 	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.14
+	XCRUN = xcrun
 endif
 
 # To cross-compile for Android, NDK version r21 or later is recommended.
@@ -407,6 +405,16 @@ ifdef COMPCXX
 	CXX=$(COMPCXX)
 endif
 
+### Sometimes gcc is really clang
+ifeq ($(COMP),gcc)
+	gccversion = $(shell $(CXX) --version)
+	gccisclang = $(findstring clang,$(gccversion))
+	ifneq ($(gccisclang),)
+		profile_make = clang-profile-make
+		profile_use = clang-profile-use
+	endif
+endif
+
 ### On mingw use Windows threads, otherwise POSIX
 ifneq ($(comp),mingw)
 	# On Android Bionic's C library comes with its own pthread implementation bundled in
@@ -798,7 +806,7 @@ clang-profile-make:
 	all
 
 clang-profile-use:
-	llvm-profdata merge -output=stockfish.profdata *.profraw
+	$(XCRUN) llvm-profdata merge -output=stockfish.profdata *.profraw
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
 	EXTRACXXFLAGS='-fprofile-instr-use=stockfish.profdata' \
 	EXTRALDFLAGS='-fprofile-use ' \

From e64b957274b94e89ad1a6e3ec4571c9082246a0a Mon Sep 17 00:00:00 2001
From: Unai Corzo <corzounai@gmail.com>
Date: Fri, 21 Aug 2020 09:24:25 +0200
Subject: [PATCH 06/52] Simplify away internal iterative deepening

Remove the iterative deepening step.
Instead, employ a depth reduction if the position is not in TT and on the PV.

STC https://tests.stockfishchess.org/tests/view/5f3ce6eaa95672ddd56c637e
LLR: 2.97 (-2.94,2.94) {-0.50,1.50}
Total: 41096 W: 4421 L: 4257 D: 32418
Ptnml(0-2): 207, 3259, 13460, 3407, 215

LTC (old) https://tests.stockfishchess.org/tests/view/5f3d7d4fa95672ddd56c640b
LLR: 2.92 (-2.94,2.94) {-1.50,0.50}
Total: 26032 W: 1320 L: 1309 D: 23403
Ptnml(0-2): 22, 1152, 10654, 1169, 19

LTC (new) https://tests.stockfishchess.org/tests/view/5f3e31e0a95672ddd56c6464
LLR: 2.95 (-2.94,2.94) {-0.75,0.25}
Total: 34160 W: 1844 L: 1766 D: 30550
Ptnml(0-2): 33, 1533, 13876, 1599, 39

bench: 3849173
---
 src/search.cpp | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index 7c839dfc..ba13680c 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -939,15 +939,11 @@ namespace {
             }
     }
 
-    // Step 11. Internal iterative deepening (~1 Elo)
-    if (depth >= 7 && !ttMove)
-    {
-        search<NT>(pos, ss, alpha, beta, depth - 7, cutNode);
-
-        tte = TT.probe(posKey, ttHit);
-        ttValue = ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
-        ttMove = ttHit ? tte->move() : MOVE_NONE;
-    }
+    // Step 11. If the position is not in TT, decrease depth by 2
+    if (   PvNode
+        && depth >= 6
+        && !ttMove)
+        depth -= 2;
 
 moves_loop: // When in check, search starts from here
 

From cbcb05ca092160137c166f84e7e9da3d6bb4e2d3 Mon Sep 17 00:00:00 2001
From: MJZ1977 <37274752+MJZ1977@users.noreply.github.com>
Date: Fri, 21 Aug 2020 10:57:34 +0200
Subject: [PATCH 07/52] Display classic and NNUE evaluation in trace mode

show both the classical and NNUE evaluation,
as well as the Final evaluation.

closes https://github.com/official-stockfish/Stockfish/pull/3042

No functional change.
---
 src/evaluate.cpp | 65 ++++++++++++++++++++++++++----------------------
 1 file changed, 35 insertions(+), 30 deletions(-)

diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index c84d894f..c66938d6 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -972,42 +972,47 @@ std::string Eval::trace(const Position& pos) {
 
   Value v;
 
-  if (Eval::useNNUE)
-  {
-      v = NNUE::evaluate(pos);
-  }
-  else
-  {
-      std::memset(scores, 0, sizeof(scores));
+  std::memset(scores, 0, sizeof(scores));
 
-      pos.this_thread()->contempt = SCORE_ZERO; // Reset any dynamic contempt
+  pos.this_thread()->contempt = SCORE_ZERO; // Reset any dynamic contempt
 
-      v = Evaluation<TRACE>(pos).value();
+  v = Evaluation<TRACE>(pos).value();
 
-      ss << std::showpoint << std::noshowpos << std::fixed << std::setprecision(2)
-         << "     Term    |    White    |    Black    |    Total   \n"
-         << "             |   MG    EG  |   MG    EG  |   MG    EG \n"
-         << " ------------+-------------+-------------+------------\n"
-         << "    Material | " << Term(MATERIAL)
-         << "   Imbalance | " << Term(IMBALANCE)
-         << "       Pawns | " << Term(PAWN)
-         << "     Knights | " << Term(KNIGHT)
-         << "     Bishops | " << Term(BISHOP)
-         << "       Rooks | " << Term(ROOK)
-         << "      Queens | " << Term(QUEEN)
-         << "    Mobility | " << Term(MOBILITY)
-         << " King safety | " << Term(KING)
-         << "     Threats | " << Term(THREAT)
-         << "      Passed | " << Term(PASSED)
-         << "       Space | " << Term(SPACE)
-         << "    Winnable | " << Term(WINNABLE)
-         << " ------------+-------------+-------------+------------\n"
-         << "       Total | " << Term(TOTAL);
-  }
+  ss << std::showpoint << std::noshowpos << std::fixed << std::setprecision(2)
+     << "     Term    |    White    |    Black    |    Total   \n"
+     << "             |   MG    EG  |   MG    EG  |   MG    EG \n"
+     << " ------------+-------------+-------------+------------\n"
+     << "    Material | " << Term(MATERIAL)
+     << "   Imbalance | " << Term(IMBALANCE)
+     << "       Pawns | " << Term(PAWN)
+     << "     Knights | " << Term(KNIGHT)
+     << "     Bishops | " << Term(BISHOP)
+     << "       Rooks | " << Term(ROOK)
+     << "      Queens | " << Term(QUEEN)
+     << "    Mobility | " << Term(MOBILITY)
+     << " King safety | " << Term(KING)
+     << "     Threats | " << Term(THREAT)
+     << "      Passed | " << Term(PASSED)
+     << "       Space | " << Term(SPACE)
+     << "    Winnable | " << Term(WINNABLE)
+     << " ------------+-------------+-------------+------------\n"
+     << "       Total | " << Term(TOTAL);
 
   v = pos.side_to_move() == WHITE ? v : -v;
 
-  ss << "\nFinal evaluation: " << to_cp(v) << " (white side)\n";
+  ss << "\nClassical evaluation: " << to_cp(v) << " (white side)\n";
+
+  if (Eval::useNNUE)
+  {
+      v = NNUE::evaluate(pos);
+      v = pos.side_to_move() == WHITE ? v : -v;
+      ss << "\nNNUE evaluation:      " << to_cp(v) << " (white side)\n";
+  }
+
+  v = evaluate(pos);
+  v = pos.side_to_move() == WHITE ? v : -v;
+  ss << "\nFinal evaluation:     " << to_cp(v) << " (white side)\n";
+
 
   return ss.str();
 }

From 34f67c57223d73ad40d583ccc033c75eb0df2453 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Fri, 21 Aug 2020 22:10:55 +0200
Subject: [PATCH 08/52] Explicitly rely on pthreads if possible

allows us to set the needed stacksize on thread creation.

Useful for environments with too small a default stack size (e.g. Alpine Linux with musl).

Passed STC, no regression:

LLR: 2.96 (-2.94,2.94) {-1.25,0.25}
Total: 17816 W: 1344 L: 1275 D: 15197
Ptnml(0-2): 30, 1057, 6682, 1092, 47
https://tests.stockfishchess.org/tests/view/5f402b5587a5c3c63d8f534d

closes https://github.com/official-stockfish/Stockfish/pull/3047

fixes https://github.com/official-stockfish/Stockfish/issues/3041

No functional change.
---
 src/Makefile           | 1 +
 src/thread_win32_osx.h | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/Makefile b/src/Makefile
index b969ba04..74ef87b9 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -417,6 +417,7 @@ endif
 
 ### On mingw use Windows threads, otherwise POSIX
 ifneq ($(comp),mingw)
+	CXXFLAGS += -DUSE_PTHREADS
 	# On Android Bionic's C library comes with its own pthread implementation bundled in
 	ifneq ($(OS),Android)
 		# Haiku has pthreads in its libroot, so only link it in on other platforms
diff --git a/src/thread_win32_osx.h b/src/thread_win32_osx.h
index c4b55a48..75ef5d9a 100644
--- a/src/thread_win32_osx.h
+++ b/src/thread_win32_osx.h
@@ -27,7 +27,7 @@
 /// The implementation calls pthread_create() with the stack size parameter
 /// equal to the linux 8MB default, on platforms that support it.
 
-#if defined(__APPLE__) || defined(__MINGW32__) || defined(__MINGW64__)
+#if defined(__APPLE__) || defined(__MINGW32__) || defined(__MINGW64__) || defined(USE_PTHREADS)
 
 #include <pthread.h>
 

From 3542033342f15625f808013b69aa8c2d274a2f91 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?St=C3=A9phane=20Nicolet?= <Stephane.Nicolet@u-paris2.fr>
Date: Sat, 22 Aug 2020 11:37:53 +0200
Subject: [PATCH 09/52] Instructions to build on older Macintosh

In recent Macs, it is possible to use the Clang compiler provided by Apple
to compile Stockfish out of the box, and this is the method used by default
in our Makefile (the Makefile sets the macosx-version-min=10.14 flag to select
the right libc++ library for the Clang compiler with recent c++17 support).

But it is quite possible to compile and run Stockfish on older Macs! Below
we describe a method to install a recent GNU compiler on these Macs, to get
the c++17 support. We have tested the following procedure to install gcc10 on
machines running Mac OS 10.7, Mac OS 10.9 and Mac OS 10.13:

1) install XCode for your machine.

2) install Apple command-line developer tools for XCode, by typing the following
   command in a Terminal:

```
      sudo xcode-select --install
```

3) go to the Stockfish "src" directory, then try a default build and run Stockfish:

```
      make clean
      make build
      make net
      ./stockfish
```

4) if step 3 worked, congrats! You have a compiler recent enough on your Mac
to compile Stockfish. If not, continue with step 5 to install GNU gcc10 :-)

5) install the MacPorts package manager (https://www.macports.org/install.php),
for instance using the fast method in the "macOS Package (.pkg) Installer"
section of the page.

6) use the "port" command to install the gcc10 package of MacPorts by typing the
following command:

```
    sudo port install gcc10
```

With this step, MacPorts will install the gcc10 compiler under the name "g++-mp-10"
in the /opt/local/bin directory:

```
   which g++-mp-10

   /opt/local/bin/g++-mp-10       <--- answer
```

7) You can now go back to the "src" directory of Stockfish, and try to build
Stockfish by pointing at the right compiler:

```
   make clean
   make build COMP=gcc COMPCXX=/opt/local/bin/g++-mp-10
   make net
   ./stockfish
```

8) Enjoy Stockfish on Macintosh!

See this pull request for further discussion:
https://github.com/official-stockfish/Stockfish/pull/3049

No functional change
---
 src/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/Makefile b/src/Makefile
index 74ef87b9..b0274504 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -370,7 +370,7 @@ else
 endif
 endif
 
-ifeq ($(KERNEL),Darwin)
+ifeq ($(KERNEL),Darwin) 
 	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.14
 	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.14
 	XCRUN = xcrun

From 5f1843c9cb55afcd3fb1da9e9dc4b0092f25d9f0 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Sat, 11 Jul 2020 16:59:33 +0200
Subject: [PATCH 10/52] Small trivial cleanups

closes https://github.com/official-stockfish/Stockfish/pull/2801

No functional change
---
 README.md        | 53 +++++++++++++++++++++++++-----------------------
 src/Makefile     |  2 +-
 src/bitboard.cpp | 12 ++++++++++-
 src/bitboard.h   | 10 ---------
 src/evaluate.cpp | 17 ++++++++--------
 src/material.cpp |  2 +-
 src/misc.cpp     | 17 ++++++++--------
 src/misc.h       |  8 --------
 src/movegen.cpp  |  2 +-
 src/movepick.cpp | 14 ++++++-------
 src/movepick.h   |  8 ++++----
 src/pawns.cpp    |  2 +-
 src/position.cpp |  4 ++--
 src/position.h   |  5 +++++
 src/search.cpp   | 15 +++++++-------
 src/timeman.cpp  | 18 ++++++++--------
 src/uci.cpp      |  2 +-
 17 files changed, 96 insertions(+), 95 deletions(-)

diff --git a/README.md b/README.md
index 7b6ddf4c..2cc88bf4 100644
--- a/README.md
+++ b/README.md
@@ -4,17 +4,17 @@
 [![Build Status](https://ci.appveyor.com/api/projects/status/github/official-stockfish/Stockfish?branch=master&svg=true)](https://ci.appveyor.com/project/mcostalba/stockfish/branch/master)
 
 [Stockfish](https://stockfishchess.org) is a free, powerful UCI chess engine
-derived from Glaurung 2.1. It features two evaluation functions, the classical
-evaluation based on handcrafted terms, and the NNUE evaluation based on
-efficiently updateable neural networks. The classical evaluation runs efficiently
-on most 64bit CPU architectures, while the NNUE evaluation benefits strongly from the
-vector intrinsics available on modern CPUs (avx2 or similar).
+derived from Glaurung 2.1. Stockfish is not a complete chess program and requires a
+UCI-compatible graphical user interface (GUI) (e.g. XBoard with PolyGlot, Scid,
+Cute Chess, eboard, Arena, Sigma Chess, Shredder, Chess Partner or Fritz) in order
+to be used comfortably. Read the documentation for your GUI of choice for information
+about how to use Stockfish with it.
 
-Stockfish is not a complete chess program and requires a
-UCI-compatible GUI (e.g. XBoard with PolyGlot, Scid, Cute Chess, eboard, Arena,
-Sigma Chess, Shredder, Chess Partner or Fritz) in order to be used comfortably.
-Read the documentation for your GUI of choice for information about how to use
-Stockfish with it.
+The Stockfish engine features two evaluation functions for chess, the classical
+evaluation based on handcrafted terms, and the NNUE evaluation based on efficiently
+updateable neural networks. The classical evaluation runs efficiently on most 64bit
+CPU architectures, while the NNUE evaluation benefits strongly from the vector
+intrinsics available on modern CPUs (avx2 or similar).
 
 
 ## Files
@@ -28,10 +28,13 @@ This distribution of Stockfish consists of the following files:
   * src, a subdirectory containing the full source code, including a Makefile
     that can be used to compile Stockfish on Unix-like systems.
 
-To use the NNUE evaluation an additional data file with neural network parameters
-needs to be downloaded. The filename for the default set can be found as the default
-value of the `EvalFile` UCI option, with the format
-`nn-[SHA256 first 12 digits].nnue` (e.g. nn-c157e0a5755b.nnue). This file can be downloaded from
+  * a file with the .nnue extension, storing the neural network for the NNUE 
+    evaluation.
+
+Note: to use the NNUE evaluation, the additional data file with neural network parameters
+needs to be downloaded. The filename for the default net can be found as the default
+value of the `EvalFile` UCI option, with the format `nn-[SHA256 first 12 digits].nnue`
+(for instance, `nn-c157e0a5755b.nnue`). This file can be downloaded from
 ```
 https://tests.stockfishchess.org/api/nn/[filename]
 ```
@@ -64,14 +67,6 @@ Currently, Stockfish has the following UCI options:
     The name of the file of the NNUE evaluation parameters. Depending on the GUI the
     filename should include the full path to the folder/directory that contains the file.
 
-  * #### Contempt
-    A positive value for contempt favors middle game positions and avoids draws,
-    effective for the classical evaluation only.
-
-  * #### Analysis Contempt
-    By default, contempt is set to prefer the side to move. Set this option to "White"
-    or "Black" to analyse with contempt for that side, or "Off" to disable contempt.
-
   * #### UCI_AnalyseMode
     An option handled by your GUI.
 
@@ -120,6 +115,14 @@ Currently, Stockfish has the following UCI options:
     Limit Syzygy tablebase probing to positions with at most this many pieces left
     (including kings and pawns).
 
+  * #### Contempt
+    A positive value for contempt favors middle game positions and avoids draws,
+    effective for the classical evaluation only.
+
+  * #### Analysis Contempt
+    By default, contempt is set to prefer the side to move. Set this option to "White"
+    or "Black" to analyse with contempt for that side, or "Off" to disable contempt.
+
   * #### Move Overhead
     Assume a time delay of x ms due to network and GUI overheads. This is useful to
     avoid losses on time in those cases.
@@ -138,7 +141,7 @@ Currently, Stockfish has the following UCI options:
   * #### Debug Log File
     Write all communication to and from the engine into a text file.
 
-## Classical and NNUE evaluation
+## A note on classical and NNUE evaluation
 
 Both approaches assign a value to a position that is used in alpha-beta (PVS) search
 to find the best move. The classical evaluation computes this value as a function
@@ -226,6 +229,7 @@ targets with corresponding descriptions.
     cd src
     make help
     make build ARCH=x86-64-modern
+    make net
 ```
 
 When not using the Makefile to compile (for instance with Microsoft MSVC) you
@@ -237,8 +241,7 @@ compiler you used to create your executable. These informations can
 be found by typing the following commands in a console:
 
 ```
-    ./stockfish
-    compiler
+    ./stockfish compiler
 ```
 
 ## Understanding the code base and participating in the project
diff --git a/src/Makefile b/src/Makefile
index b0274504..74ef87b9 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -370,7 +370,7 @@ else
 endif
 endif
 
-ifeq ($(KERNEL),Darwin) 
+ifeq ($(KERNEL),Darwin)
 	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.14
 	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.14
 	XCRUN = xcrun
diff --git a/src/bitboard.cpp b/src/bitboard.cpp
index f531010c..80206b58 100644
--- a/src/bitboard.cpp
+++ b/src/bitboard.cpp
@@ -39,6 +39,16 @@ namespace {
   Bitboard BishopTable[0x1480]; // To store bishop attacks
 
   void init_magics(PieceType pt, Bitboard table[], Magic magics[]);
+
+}
+
+
+/// safe_destination() returns the bitboard of target square for the given step
+/// from the given square. If the step is off the board, returns empty bitboard.
+
+inline Bitboard safe_destination(Square s, int step) {
+    Square to = Square(s + step);
+    return is_ok(to) && distance(s, to) <= 2 ? square_bb(to) : Bitboard(0);
 }
 
 
@@ -110,7 +120,7 @@ namespace {
     Direction   RookDirections[4] = {NORTH, SOUTH, EAST, WEST};
     Direction BishopDirections[4] = {NORTH_EAST, SOUTH_EAST, SOUTH_WEST, NORTH_WEST};
 
-    for(Direction d : (pt == ROOK ? RookDirections : BishopDirections))
+    for (Direction d : (pt == ROOK ? RookDirections : BishopDirections))
     {
         Square s = sq;
         while(safe_destination(s, d) && !(occupied & s))
diff --git a/src/bitboard.h b/src/bitboard.h
index a899d879..29d8f66d 100644
--- a/src/bitboard.h
+++ b/src/bitboard.h
@@ -279,16 +279,6 @@ inline int edge_distance(File f) { return std::min(f, File(FILE_H - f)); }
 inline int edge_distance(Rank r) { return std::min(r, Rank(RANK_8 - r)); }
 
 
-/// safe_destination() returns the bitboard of target square for the given step
-/// from the given square. If the step is off the board, returns empty bitboard.
-
-inline Bitboard safe_destination(Square s, int step)
-{
-    Square to = Square(s + step);
-    return is_ok(to) && distance(s, to) <= 2 ? square_bb(to) : Bitboard(0);
-}
-
-
 /// attacks_bb(Square) returns the pseudo attacks of the give piece type
 /// assuming an empty board.
 
diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index c66938d6..ce92db9a 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -288,8 +288,8 @@ namespace {
     attackedBy2[Us] = dblAttackByPawn | (attackedBy[Us][KING] & attackedBy[Us][PAWN]);
 
     // Init our king safety tables
-    Square s = make_square(Utility::clamp(file_of(ksq), FILE_B, FILE_G),
-                           Utility::clamp(rank_of(ksq), RANK_2, RANK_7));
+    Square s = make_square(std::clamp(file_of(ksq), FILE_B, FILE_G),
+                           std::clamp(rank_of(ksq), RANK_2, RANK_7));
     kingRing[Us] = attacks_bb<KING>(s) | s;
 
     kingAttackersCount[Them] = popcount(kingRing[Us] & pe->pawn_attacks(Them));
@@ -686,8 +686,8 @@ namespace {
             Square blockSq = s + Up;
 
             // Adjust bonus based on the king's proximity
-            bonus += make_score(0, (  (king_proximity(Them, blockSq) * 19) / 4
-                                     - king_proximity(Us,   blockSq) *  2) * w);
+            bonus += make_score(0, (  king_proximity(Them, blockSq) * 19 / 4
+                                    - king_proximity(Us,   blockSq) *  2) * w);
 
             // If blockSq is not the queening square then consider also a second push
             if (r != RANK_7)
@@ -731,7 +731,7 @@ namespace {
 
 
   // Evaluation::space() computes a space evaluation for a given side, aiming to improve game
-  // play in the opening. It is based on the number of safe squares on the 4 central files
+  // play in the opening. It is based on the number of safe squares on the four central files
   // on ranks 2 to 4. Completely safe squares behind a friendly pawn are counted twice.
   // Finally, the space bonus is multiplied by a weight which decreases according to occupancy.
 
@@ -804,7 +804,7 @@ namespace {
     // Now apply the bonus: note that we find the attacking side by extracting the
     // sign of the midgame or endgame values, and that we carefully cap the bonus
     // so that the midgame and endgame scores do not change sign after the bonus.
-    int u = ((mg > 0) - (mg < 0)) * Utility::clamp(complexity + 50, -abs(mg), 0);
+    int u = ((mg > 0) - (mg < 0)) * std::clamp(complexity + 50, -abs(mg), 0);
     int v = ((eg > 0) - (eg < 0)) * std::max(complexity, -abs(eg));
 
     mg += u;
@@ -951,8 +951,8 @@ Value Eval::evaluate(const Position& pos) {
   // Damp down the evaluation linearly when shuffling
   v = v * (100 - pos.rule50_count()) / 100;
 
-  // Guarantee evalution outside of TB range
-  v = Utility::clamp(v, VALUE_TB_LOSS_IN_MAX_PLY + 1, VALUE_TB_WIN_IN_MAX_PLY - 1);
+  // Guarantee evaluation does not hit the tablebase range
+  v = std::clamp(v, VALUE_TB_LOSS_IN_MAX_PLY + 1, VALUE_TB_WIN_IN_MAX_PLY - 1);
 
   return v;
 }
@@ -1013,6 +1013,5 @@ std::string Eval::trace(const Position& pos) {
   v = pos.side_to_move() == WHITE ? v : -v;
   ss << "\nFinal evaluation:     " << to_cp(v) << " (white side)\n";
 
-
   return ss.str();
 }
diff --git a/src/material.cpp b/src/material.cpp
index 0ef9926f..870a5e11 100644
--- a/src/material.cpp
+++ b/src/material.cpp
@@ -130,7 +130,7 @@ Entry* probe(const Position& pos) {
 
   Value npm_w = pos.non_pawn_material(WHITE);
   Value npm_b = pos.non_pawn_material(BLACK);
-  Value npm   = Utility::clamp(npm_w + npm_b, EndgameLimit, MidgameLimit);
+  Value npm   = std::clamp(npm_w + npm_b, EndgameLimit, MidgameLimit);
 
   // Map total non-pawn material into [PHASE_ENDGAME, PHASE_MIDGAME]
   e->gamePhase = Phase(((npm - EndgameLimit) * PHASE_MIDGAME) / (MidgameLimit - EndgameLimit));
diff --git a/src/misc.cpp b/src/misc.cpp
index 56a3dcad..80c436ac 100644
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -328,16 +328,16 @@ void prefetch(void* addr) {
 
 #endif
 
-/// Wrappers for systems where the c++17 implementation doesn't guarantee the availability of aligned_alloc.
-/// Memory allocated with std_aligned_alloc must be freed with std_aligned_free.
-///
+
+/// std_aligned_alloc() is our wrapper for systems where the c++17 implementation
+/// does not guarantee the availability of aligned_alloc(). Memory allocated with
+/// std_aligned_alloc() must be freed with std_aligned_free().
 
 void* std_aligned_alloc(size_t alignment, size_t size) {
+
 #if defined(POSIXALIGNEDALLOC)
-  void *pointer;
-  if(posix_memalign(&pointer, alignment, size) == 0)
-      return pointer;
-  return nullptr;
+  void *mem;
+  return posix_memalign(&mem, alignment, size) ? nullptr : mem;
 #elif defined(_WIN32)
   return _mm_malloc(size, alignment);
 #else
@@ -346,6 +346,7 @@ void* std_aligned_alloc(size_t alignment, size_t size) {
 }
 
 void std_aligned_free(void* ptr) {
+
 #if defined(POSIXALIGNEDALLOC)
   free(ptr);
 #elif defined(_WIN32)
@@ -355,7 +356,7 @@ void std_aligned_free(void* ptr) {
 #endif
 }
 
-/// aligned_ttmem_alloc() will return suitably aligned memory, and if possible use large pages.
+/// aligned_ttmem_alloc() will return suitably aligned memory, if possible using large pages.
 /// The returned pointer is the aligned one, while the mem argument is the one that needs
 /// to be passed to free. With c++17 some of this functionality could be simplified.
 
diff --git a/src/misc.h b/src/misc.h
index eb4e05c0..8ad17b50 100644
--- a/src/misc.h
+++ b/src/misc.h
@@ -65,14 +65,6 @@ std::ostream& operator<<(std::ostream&, SyncCout);
 #define sync_cout std::cout << IO_LOCK
 #define sync_endl std::endl << IO_UNLOCK
 
-namespace Utility {
-
-/// Clamp a value between lo and hi. Available in c++17.
-template<class T> constexpr const T& clamp(const T& v, const T& lo, const T& hi) {
-  return v < lo ? lo : v > hi ? hi : v;
-}
-
-}
 
 /// xorshift64star Pseudo-Random Number Generator
 /// This class is based on original code written and dedicated
diff --git a/src/movegen.cpp b/src/movegen.cpp
index d74df4c3..3340f65c 100644
--- a/src/movegen.cpp
+++ b/src/movegen.cpp
@@ -248,7 +248,7 @@ namespace {
             *moveList++ = make_move(ksq, pop_lsb(&b));
 
         if ((Type != CAPTURES) && pos.can_castle(Us & ANY_CASTLING))
-            for(CastlingRights cr : { Us & KING_SIDE, Us & QUEEN_SIDE } )
+            for (CastlingRights cr : { Us & KING_SIDE, Us & QUEEN_SIDE } )
                 if (!pos.castling_impeded(cr) && pos.can_castle(cr))
                     *moveList++ = make<CASTLING>(ksq, pos.castling_rook_square(cr));
     }
diff --git a/src/movepick.cpp b/src/movepick.cpp
index 96a44449..153d323e 100644
--- a/src/movepick.cpp
+++ b/src/movepick.cpp
@@ -182,7 +182,7 @@ top:
           --endMoves;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case REFUTATION:
       if (select<Next>([&](){ return    *cur != MOVE_NONE
@@ -190,7 +190,7 @@ top:
                                     &&  pos.pseudo_legal(*cur); }))
           return *(cur - 1);
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QUIET_INIT:
       if (!skipQuiets)
@@ -203,7 +203,7 @@ top:
       }
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QUIET:
       if (   !skipQuiets
@@ -217,7 +217,7 @@ top:
       endMoves = endBadCaptures;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case BAD_CAPTURE:
       return select<Next>([](){ return true; });
@@ -228,7 +228,7 @@ top:
 
       score<EVASIONS>();
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case EVASION:
       return select<Best>([](){ return true; });
@@ -246,14 +246,14 @@ top:
           return MOVE_NONE;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QCHECK_INIT:
       cur = moves;
       endMoves = generate<QUIET_CHECKS>(pos, cur);
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QCHECK:
       return select<Next>([](){ return true; });
diff --git a/src/movepick.h b/src/movepick.h
index f080935a..97ea5bec 100644
--- a/src/movepick.h
+++ b/src/movepick.h
@@ -86,14 +86,14 @@ enum StatsType { NoCaptures, Captures };
 /// the move's from and to squares, see www.chessprogramming.org/Butterfly_Boards
 typedef Stats<int16_t, 10692, COLOR_NB, int(SQUARE_NB) * int(SQUARE_NB)> ButterflyHistory;
 
-/// At higher depths LowPlyHistory records successful quiet moves near the root and quiet
-/// moves which are/were in the PV (ttPv)
-/// It is cleared with each new search and filled during iterative deepening
+/// At higher depths LowPlyHistory records successful quiet moves near the root
+/// and quiet moves which are/were in the PV (ttPv). It is cleared with each new
+/// search and filled during iterative deepening.
 constexpr int MAX_LPH = 4;
 typedef Stats<int16_t, 10692, MAX_LPH, int(SQUARE_NB) * int(SQUARE_NB)> LowPlyHistory;
 
 /// CounterMoveHistory stores counter moves indexed by [piece][to] of the previous
-/// move, see www.chessprogramming.org/Countermove_Heuristic
+/// move, see www.chessprogramming.org/Countermove_Heuristic 
 typedef Stats<Move, NOT_USED, PIECE_NB, SQUARE_NB> CounterMoveHistory;
 
 /// CapturePieceToHistory is addressed by a move's [piece][to][captured piece type]
diff --git a/src/pawns.cpp b/src/pawns.cpp
index 868d0c8e..af0f6618 100644
--- a/src/pawns.cpp
+++ b/src/pawns.cpp
@@ -219,7 +219,7 @@ Score Entry::evaluate_shelter(const Position& pos, Square ksq) const {
 
   Score bonus = make_score(5, 5);
 
-  File center = Utility::clamp(file_of(ksq), FILE_B, FILE_G);
+  File center = std::clamp(file_of(ksq), FILE_B, FILE_G);
   for (File f = File(center - 1); f <= File(center + 1); ++f)
   {
       b = ourPawns & file_bb(f);
diff --git a/src/position.cpp b/src/position.cpp
index 46e5d78b..02898547 100644
--- a/src/position.cpp
+++ b/src/position.cpp
@@ -1145,8 +1145,8 @@ bool Position::see_ge(Move m, Value threshold) const {
 
       // Don't allow pinned pieces to attack (except the king) as long as
       // there are pinners on their original square.
-      if (st->pinners[~stm] & occupied)
-          stmAttackers &= ~st->blockersForKing[stm];
+      if (pinners(~stm) & occupied)
+          stmAttackers &= ~blockers_for_king(stm);
 
       if (!stmAttackers)
           break;
diff --git a/src/position.h b/src/position.h
index a77050eb..5ce17277 100644
--- a/src/position.h
+++ b/src/position.h
@@ -113,6 +113,7 @@ public:
   Bitboard checkers() const;
   Bitboard blockers_for_king(Color c) const;
   Bitboard check_squares(PieceType pt) const;
+  Bitboard pinners(Color c) const;
   bool is_discovery_check_on_king(Color c, Move m) const;
 
   // Attacks to/from a given square
@@ -309,6 +310,10 @@ inline Bitboard Position::blockers_for_king(Color c) const {
   return st->blockersForKing[c];
 }
 
+inline Bitboard Position::pinners(Color c) const {
+  return st->pinners[c];
+}
+
 inline Bitboard Position::check_squares(PieceType pt) const {
   return st->checkSquares[pt];
 }
diff --git a/src/search.cpp b/src/search.cpp
index ba13680c..82d8bb9d 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -335,7 +335,7 @@ void Thread::search() {
   // for match (TC 60+0.6) results spanning a wide range of k values.
   PRNG rng(now());
   double floatLevel = Options["UCI_LimitStrength"] ?
-                      Utility::clamp(std::pow((Options["UCI_Elo"] - 1346.6) / 143.4, 1 / 0.806), 0.0, 20.0) :
+                      std::clamp(std::pow((Options["UCI_Elo"] - 1346.6) / 143.4, 1 / 0.806), 0.0, 20.0) :
                         double(Options["Skill Level"]);
   int intLevel = int(floatLevel) +
                  ((floatLevel - int(floatLevel)) * 1024 > rng.rand<unsigned>() % 1024  ? 1 : 0);
@@ -508,7 +508,7 @@ void Thread::search() {
       {
           double fallingEval = (318 + 6 * (mainThread->bestPreviousScore - bestValue)
                                     + 6 * (mainThread->iterValue[iterIdx] - bestValue)) / 825.0;
-          fallingEval = Utility::clamp(fallingEval, 0.5, 1.5);
+          fallingEval = std::clamp(fallingEval, 0.5, 1.5);
 
           // If the bestMove is stable over several iterations, reduce time accordingly
           timeReduction = lastBestMoveDepth + 9 < completedDepth ? 1.92 : 0.95;
@@ -807,8 +807,9 @@ namespace {
         &&  eval <= alpha - RazorMargin)
         return qsearch<NT>(pos, ss, alpha, beta);
 
-    improving =  (ss-2)->staticEval == VALUE_NONE ? (ss->staticEval > (ss-4)->staticEval
-              || (ss-4)->staticEval == VALUE_NONE) : ss->staticEval > (ss-2)->staticEval;
+    improving =  (ss-2)->staticEval == VALUE_NONE
+               ? ss->staticEval > (ss-4)->staticEval || (ss-4)->staticEval == VALUE_NONE
+               : ss->staticEval > (ss-2)->staticEval;
 
     // Step 8. Futility pruning: child node (~50 Elo)
     if (   !PvNode
@@ -879,8 +880,8 @@ namespace {
         // there and in further interactions with transposition table cutoff depth is set to depth - 3
         // because probCut search has depth set to depth - 4 but we also do a move before it
         // so effective depth is equal to depth - 3
-        && !(   ttHit 
-             && tte->depth() >= depth - 3 
+        && !(   ttHit
+             && tte->depth() >= depth - 3
              && ttValue != VALUE_NONE
              && ttValue < probCutBeta))
     {
@@ -1238,7 +1239,7 @@ moves_loop: // When in check, search starts from here
                 r++;
           }
 
-          Depth d = Utility::clamp(newDepth - r, 1, newDepth);
+          Depth d = std::clamp(newDepth - r, 1, newDepth);
 
           value = -search<NonPV>(pos, ss+1, -(alpha+1), -alpha, d, true);
 
diff --git a/src/timeman.cpp b/src/timeman.cpp
index df4ba9b2..6d9c95ef 100644
--- a/src/timeman.cpp
+++ b/src/timeman.cpp
@@ -38,9 +38,9 @@ void TimeManagement::init(Search::LimitsType& limits, Color us, int ply) {
   TimePoint slowMover       = TimePoint(Options["Slow Mover"]);
   TimePoint npmsec          = TimePoint(Options["nodestime"]);
 
-  // opt_scale is a percentage of available time to use for the current move.
-  // max_scale is a multiplier applied to optimumTime.
-  double opt_scale, max_scale;
+  // optScale is a percentage of available time to use for the current move.
+  // maxScale is a multiplier applied to optimumTime.
+  double optScale, maxScale;
 
   // If we have to play in 'nodes as time' mode, then convert from time
   // to nodes, and use resulting values in time management formulas.
@@ -75,22 +75,22 @@ void TimeManagement::init(Search::LimitsType& limits, Color us, int ply) {
   // game time for the current move, so also cap to 20% of available game time.
   if (limits.movestogo == 0)
   {
-      opt_scale = std::min(0.008 + std::pow(ply + 3.0, 0.5) / 250.0,
+      optScale = std::min(0.008 + std::pow(ply + 3.0, 0.5) / 250.0,
                            0.2 * limits.time[us] / double(timeLeft));
-      max_scale = std::min(7.0, 4.0 + ply / 12.0);
+      maxScale = std::min(7.0, 4.0 + ply / 12.0);
   }
 
   // x moves in y seconds (+ z increment)
   else
   {
-      opt_scale = std::min((0.8 + ply / 128.0) / mtg,
+      optScale = std::min((0.8 + ply / 128.0) / mtg,
                             0.8 * limits.time[us] / double(timeLeft));
-      max_scale = std::min(6.3, 1.5 + 0.11 * mtg);
+      maxScale = std::min(6.3, 1.5 + 0.11 * mtg);
   }
 
   // Never use more than 80% of the available time for this move
-  optimumTime = TimePoint(opt_scale * timeLeft);
-  maximumTime = TimePoint(std::min(0.8 * limits.time[us] - moveOverhead, max_scale * optimumTime));
+  optimumTime = TimePoint(optScale * timeLeft);
+  maximumTime = TimePoint(std::min(0.8 * limits.time[us] - moveOverhead, maxScale * optimumTime));
 
   if (Options["Ponder"])
       optimumTime += optimumTime / 4;
diff --git a/src/uci.cpp b/src/uci.cpp
index d6486320..bc0ee0a0 100644
--- a/src/uci.cpp
+++ b/src/uci.cpp
@@ -211,7 +211,7 @@ namespace {
      double b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
 
      // Transform eval to centipawns with limited range
-     double x = Utility::clamp(double(100 * v) / PawnValueEg, -1000.0, 1000.0);
+     double x = std::clamp(double(100 * v) / PawnValueEg, -1000.0, 1000.0);
 
      // Return win rate in per mille (rounded to nearest)
      return int(0.5 + 1000 / (1 + std::exp((a - x) / b)));

From cc9d503ddea998890112efd08fae3705f2727e37 Mon Sep 17 00:00:00 2001
From: syzygy1 <3028851+syzygy1@users.noreply.github.com>
Date: Sat, 22 Aug 2020 13:36:34 +0200
Subject: [PATCH 11/52] Skip the alignment bug workaround for Clang

Clang-10.0.0 poses as gcc-4.2:

$ clang++ -E -dM - </dev/null | grep GNUC

This means that Clang is using the workaround for the alignment bug of gcc-8
even though it does not have the bug (as far as I know).

This patch should speed up AVX2 and AVX512 compiles on Windows (when using Clang),
because it disables (for Clang) the gcc workaround we had introduced in this commit:
https://github.com/official-stockfish/Stockfish/commit/875183b310a8249922c2155e82cb4cecfae2097e

closes https://github.com/official-stockfish/Stockfish/pull/3050

No functional change.
---
 src/nnue/nnue_common.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/nnue/nnue_common.h b/src/nnue/nnue_common.h
index 4c93e3d1..a9d8e5af 100644
--- a/src/nnue/nnue_common.h
+++ b/src/nnue/nnue_common.h
@@ -47,7 +47,7 @@
 //       compiled with older g++ crashes because the output memory is not aligned
 //       even though alignas is specified.
 #if defined(USE_AVX2)
-#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32)
+#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32) && !defined(__clang__)
 #define _mm256_loadA_si256  _mm256_loadu_si256
 #define _mm256_storeA_si256 _mm256_storeu_si256
 #else
@@ -57,7 +57,7 @@
 #endif
 
 #if defined(USE_AVX512)
-#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32)
+#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32) && !defined(__clang__)
 #define _mm512_loadA_si512   _mm512_loadu_si512
 #define _mm512_storeA_si512  _mm512_storeu_si512
 #else

From d5f86b63597fdb25b8c670e42ca468c99ca95a74 Mon Sep 17 00:00:00 2001
From: Vizvezdenec <Vizvezdenec@gmail.com>
Date: Sun, 23 Aug 2020 14:22:32 +0300
Subject: [PATCH 12/52] Introduce movecount pruning for qsearch()

If in quiescence search, we assume that me can prune late moves when:

a) the move ordering count of the move is : moveCount > abs(depth) + 2
b) we are not in check
c) the late move does not give check
d) the late move is not an advanced pawn push

Modification of an original idea by @VoyagerOne.

STC
https://tests.stockfishchess.org/tests/view/5f40581787a5c3c63d8f535f
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 132848 W: 14999 L: 14661 D: 103188
Ptnml(0-2): 684, 11242, 42309, 11430, 759

LTC
https://tests.stockfishchess.org/tests/view/5f4226da87a5c3c63d8f5412
LLR: 2.98 (-2.94,2.94) {0.25,1.25}
Total: 12008 W: 678 L: 551 D: 10779
Ptnml(0-2): 8, 485, 4899, 596, 16

closes https://github.com/official-stockfish/Stockfish/pull/3053

Bench: 3749974
---
 src/movepick.h | 2 +-
 src/search.cpp | 6 +++++-
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/src/movepick.h b/src/movepick.h
index 97ea5bec..4c0ad551 100644
--- a/src/movepick.h
+++ b/src/movepick.h
@@ -93,7 +93,7 @@ constexpr int MAX_LPH = 4;
 typedef Stats<int16_t, 10692, MAX_LPH, int(SQUARE_NB) * int(SQUARE_NB)> LowPlyHistory;
 
 /// CounterMoveHistory stores counter moves indexed by [piece][to] of the previous
-/// move, see www.chessprogramming.org/Countermove_Heuristic 
+/// move, see www.chessprogramming.org/Countermove_Heuristic
 typedef Stats<Move, NOT_USED, PIECE_NB, SQUARE_NB> CounterMoveHistory;
 
 /// CapturePieceToHistory is addressed by a move's [piece][to][captured piece type]
diff --git a/src/search.cpp b/src/search.cpp
index 82d8bb9d..266e2db3 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1531,6 +1531,10 @@ moves_loop: // When in check, search starts from here
       {
           assert(type_of(move) != ENPASSANT); // Due to !pos.advanced_pawn_push
 
+          // moveCount pruning
+          if (moveCount > abs(depth) + 2)
+              continue;
+
           futilityValue = futilityBase + PieceValue[EG][pos.piece_on(to_sq(move))];
 
           if (futilityValue <= alpha)
@@ -1547,7 +1551,7 @@ moves_loop: // When in check, search starts from here
       }
 
       // Do not search moves with negative SEE values
-      if (  !ss->inCheck && !pos.see_ge(move))
+      if (!ss->inCheck && !pos.see_ge(move))
           continue;
 
       // Speculative prefetch as early as possible

From e453f09f06f41680ef96f594f593f8de33e62b8f Mon Sep 17 00:00:00 2001
From: George Sobala <gsobala@gmail.com>
Date: Mon, 24 Aug 2020 06:37:42 +0100
Subject: [PATCH 13/52] armv8 AArch64 does not require -mfpu=neon

-mpfu is not required on AArch64 / armv8 architecture on Linux and throws an error if present.
This PR has been tested on gcc and clang on Gentoo-64 and Raspian-64 on a Raspberry Pi 4,
as well as with a cross from Ubuntu
(`make clean && make -j build ARCH=armv8         COMP=gcc COMPILER=aarch64-linux-gnu-g++`)

fixes https://github.com/official-stockfish/Stockfish/issues/3056
closes https://github.com/official-stockfish/Stockfish/pull/3059

No functional change
---
 src/Makefile | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/src/Makefile b/src/Makefile
index 74ef87b9..3e1b7c35 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -241,7 +241,7 @@ ifeq ($(ARCH),armv7-neon)
 endif
 
 ifeq ($(ARCH),armv8)
-	arch = armv8-a
+	arch = armv8
 	prefetch = yes
 	popcnt = yes
 	neon = yes
@@ -285,7 +285,7 @@ ifeq ($(COMP),gcc)
 	CXX=g++
 	CXXFLAGS += -pedantic -Wextra -Wshadow
 
-	ifeq ($(arch),$(filter $(arch),armv7 armv8-a))
+	ifeq ($(arch),$(filter $(arch),armv7 armv8))
 		ifeq ($(OS),Android)
 			CXXFLAGS += -m$(bits)
 			LDFLAGS += -m$(bits)
@@ -387,7 +387,7 @@ ifeq ($(COMP),ndk)
 		CXXFLAGS += -mthumb -march=armv7-a -mfloat-abi=softfp -mfpu=neon
 		STRIP=arm-linux-androideabi-strip
 	endif
-	ifeq ($(arch),armv8-a)
+	ifeq ($(arch),armv8)
 		comp=aarch64-linux-android21-clang
 		CXX=aarch64-linux-android21-clang++
 		STRIP=aarch64-linux-android-strip
@@ -476,7 +476,7 @@ endif
 
 ### 3.6 popcnt
 ifeq ($(popcnt),yes)
-	ifeq ($(arch),$(filter $(arch),ppc64 armv7 armv8-a arm64))
+	ifeq ($(arch),$(filter $(arch),ppc64 armv7 armv8 arm64))
 		CXXFLAGS += -DUSE_POPCNT
 	else ifeq ($(comp),icc)
 		CXXFLAGS += -msse3 -DUSE_POPCNT
@@ -539,9 +539,11 @@ ifeq ($(neon),yes)
 	CXXFLAGS += -DUSE_NEON
 	ifeq ($(KERNEL),Linux)
 	ifneq ($(COMP),ndk)
+	ifneq ($(arch),armv8)
 		CXXFLAGS += -mfpu=neon
 	endif
 	endif
+	endif
 endif
 
 ### 3.7 pext
@@ -780,7 +782,7 @@ config-sanity:
 	@test "$(optimize)" = "yes" || test "$(optimize)" = "no"
 	@test "$(arch)" = "any" || test "$(arch)" = "x86_64" || test "$(arch)" = "i386" || \
 	 test "$(arch)" = "ppc64" || test "$(arch)" = "ppc" || \
-	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8-a" || test "$(arch)" = "arm64"
+	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64"
 	@test "$(bits)" = "32" || test "$(bits)" = "64"
 	@test "$(prefetch)" = "yes" || test "$(prefetch)" = "no"
 	@test "$(popcnt)" = "yes" || test "$(popcnt)" = "no"

From 701b2427bd84d112376ce858b66befc5b66c4bb2 Mon Sep 17 00:00:00 2001
From: mstembera <MissingEmail@email>
Date: Thu, 20 Aug 2020 16:59:27 -0700
Subject: [PATCH 14/52] Support VNNI on 256bit vectors

due to downclocking on current chips (tested up to cascade lake)
supporting avx512 and vnni512, it is better to use avx2 or vnni256
in multithreaded (in particular hyperthreaded) engine use.
In single threaded use, the picture is different.

gcc compilation for vnni256 requires a toolchain for gcc >= 9.

closes https://github.com/official-stockfish/Stockfish/pull/3038

No functional change
---
 .travis.yml                        |  6 +++--
 src/Makefile                       | 39 ++++++++++++++++++++++++------
 src/nnue/layers/affine_transform.h |  8 +++++-
 3 files changed, 42 insertions(+), 11 deletions(-)

diff --git a/.travis.yml b/.travis.yml
index a029c4fc..c1e6d6df 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -77,8 +77,10 @@ script:
   # compile only for some more advanced architectures (might not run in travis)
   - make clean && make -j2 ARCH=x86-64-avx2 build
   - make clean && make -j2 ARCH=x86-64-bmi2 build
-  # needs gcc 10 to compile
-  - if [[ "$COMPILER" != "g++-8" ]]; then make clean && make -j2 ARCH=x86-64-avx512 build; fi
+  - make clean && make -j2 ARCH=x86-64-avx512 build
+  - make clean && make -j2 ARCH=x86-64-vnni512 build
+  # requires gcc 9 or higher
+  - if [[ "$COMPILER" != "g++-8" ]]; make clean && make -j2 ARCH=x86-64-vnni256 build; fi
 
   #
   # Check perft and reproducible search
diff --git a/src/Makefile b/src/Makefile
index 3e1b7c35..228ea851 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -75,7 +75,8 @@ endif
 # sse41 = yes/no      --- -msse4.1         --- Use Intel Streaming SIMD Extensions 4.1
 # avx2 = yes/no       --- -mavx2           --- Use Intel Advanced Vector Extensions 2
 # avx512 = yes/no     --- -mavx512bw       --- Use Intel Advanced Vector Extensions 512
-# vnni = yes/no       --- -mavx512vnni     --- Use Intel Vector Neural Network Instructions 512
+# vnni256 = yes/no    --- -mavx512vnni     --- Use Intel Vector Neural Network Instructions 256
+# vnni512 = yes/no    --- -mavx512vnni     --- Use Intel Vector Neural Network Instructions 512
 # neon = yes/no       --- -DUSE_NEON       --- Use ARM SIMD architecture
 #
 # Note that Makefile is space sensitive, so when adding new architectures
@@ -102,7 +103,8 @@ ssse3 = no
 sse41 = no
 avx2 = no
 avx512 = no
-vnni = no
+vnni256 = no
+vnni512 = no
 neon = no
 ARCH = x86-64-modern
 STRIP = strip
@@ -192,7 +194,18 @@ ifeq ($(findstring -avx512,$(ARCH)),-avx512)
 	avx512 = yes
 endif
 
-ifeq ($(findstring -vnni,$(ARCH)),-vnni)
+ifeq ($(findstring -vnni256,$(ARCH)),-vnni256)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	avx2 = yes
+	pext = yes
+	vnni256 = yes
+endif
+
+ifeq ($(findstring -vnni512,$(ARCH)),-vnni512)
 	popcnt = yes
 	sse = yes
 	sse2 = yes
@@ -201,7 +214,7 @@ ifeq ($(findstring -vnni,$(ARCH)),-vnni)
 	avx2 = yes
 	pext = yes
 	avx512 = yes
-	vnni = yes
+	vnni512 = yes
 endif
 
 ifeq ($(sse),yes)
@@ -500,7 +513,14 @@ ifeq ($(avx512),yes)
 	endif
 endif
 
-ifeq ($(vnni),yes)
+ifeq ($(vnni256),yes)
+	CXXFLAGS += -DUSE_VNNI
+	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
+		CXXFLAGS += -mavx512vnni -mavx512dq -mavx512vl -mprefer-vector-width=256
+	endif
+endif
+
+ifeq ($(vnni512),yes)
 	CXXFLAGS += -DUSE_VNNI
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
 		CXXFLAGS += -mavx512vnni -mavx512dq -mavx512vl
@@ -623,7 +643,8 @@ help:
 	@echo ""
 	@echo "Supported archs:"
 	@echo ""
-	@echo "x86-64-vnni             > x86 64-bit with vnni support"
+	@echo "x86-64-vnni512          > x86 64-bit with vnni support 512bit wide"
+	@echo "x86-64-vnni256          > x86 64-bit with vnni support 256bit wide"
 	@echo "x86-64-avx512           > x86 64-bit with avx512 support"
 	@echo "x86-64-bmi2             > x86 64-bit with bmi2 support"
 	@echo "x86-64-avx2             > x86 64-bit with avx2 support"
@@ -767,7 +788,8 @@ config-sanity:
 	@echo "sse41: '$(sse41)'"
 	@echo "avx2: '$(avx2)'"
 	@echo "avx512: '$(avx512)'"
-	@echo "vnni: '$(vnni)'"
+	@echo "vnni256: '$(vnni256)'"
+	@echo "vnni512: '$(vnni512)'"
 	@echo "neon: '$(neon)'"
 	@echo ""
 	@echo "Flags:"
@@ -794,7 +816,8 @@ config-sanity:
 	@test "$(sse41)" = "yes" || test "$(sse41)" = "no"
 	@test "$(avx2)" = "yes" || test "$(avx2)" = "no"
 	@test "$(avx512)" = "yes" || test "$(avx512)" = "no"
-	@test "$(vnni)" = "yes" || test "$(vnni)" = "no"
+	@test "$(vnni256)" = "yes" || test "$(vnni256)" = "no"
+	@test "$(vnni512)" = "yes" || test "$(vnni512)" = "no"
 	@test "$(neon)" = "yes" || test "$(neon)" = "no"
 	@test "$(comp)" = "gcc" || test "$(comp)" = "icc" || test "$(comp)" = "mingw" || test "$(comp)" = "clang" \
 	|| test "$(comp)" = "armv7a-linux-androideabi16-clang"  || test "$(comp)" = "aarch64-linux-android21-clang"
diff --git a/src/nnue/layers/affine_transform.h b/src/nnue/layers/affine_transform.h
index 7ac5a1c0..94d0b5a9 100644
--- a/src/nnue/layers/affine_transform.h
+++ b/src/nnue/layers/affine_transform.h
@@ -85,8 +85,10 @@ namespace Eval::NNUE::Layers {
 
   #elif defined(USE_AVX2)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const __m256i kOnes = _mm256_set1_epi16(1);
       const auto input_vector = reinterpret_cast<const __m256i*>(input);
+  #if !defined(USE_VNNI)
+      const __m256i kOnes = _mm256_set1_epi16(1);
+  #endif
 
   #elif defined(USE_SSE2)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
@@ -145,9 +147,13 @@ namespace Eval::NNUE::Layers {
         __m256i sum = _mm256_setzero_si256();
         const auto row = reinterpret_cast<const __m256i*>(&weights_[offset]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
+  #if defined(USE_VNNI)
+          sum = _mm256_dpbusd_epi32(sum, _mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
+  #else
           __m256i product = _mm256_maddubs_epi16(_mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
           product = _mm256_madd_epi16(product, kOnes);
           sum = _mm256_add_epi32(sum, product);
+  #endif
         }
         __m128i sum128 = _mm_add_epi32(_mm256_castsi256_si128(sum), _mm256_extracti128_si256(sum, 1));
         sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_BADC));

From f7b3f0e8426bbf7414d139ed9d1cfa7a98d7314d Mon Sep 17 00:00:00 2001
From: Sami Kiminki <skiminki@users.noreply.github.com>
Date: Fri, 21 Aug 2020 12:12:39 +0300
Subject: [PATCH 15/52] Allow TT entries with key16==0 to be fetched

Fix the issue where a TT entry with key16==0 would always be reported
as a miss. Instead, we'll use depth8 to detect whether the TT entry is
occupied. In order to do that, we'll change DEPTH_OFFSET to -7
(depth8==0) to distinguish between an unoccupied entry and the
otherwise lowest possible depth, i.e., DEPTH_NONE (depth8==1).

To prevent a performance regression, we'll reorder the TT entry fields
by the access order of TranspositionTable::probe(). Memory in general
works fastest when accessed in sequential order. We'll also match the
store order in TTEntry::save() with the entry field order, and
re-order the 'if-or' expressions in TTEntry::save() from the cheapest
to the most expensive.

Finally, as we now have a proper TT entry occupancy test, we'll fix a
minor corner case with hashfull reporting. To reproduce:
- Use a big hash
- Either:
  a. Start 31 very quick searches (this wraparounds generation to 0); or
  b. Force generation of the first search to 0.
- go depth infinite

Before the fix, hashfull would incorrectly report nearly full hash
immediately after the search start, since
TranspositionTable::hashfull() used to consider only the entry
generation and not whether the entry was actually occupied.

STC:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 36848 W: 4091 L: 3898 D: 28859
Ptnml(0-2): 158, 2996, 11972, 3091, 207
https://tests.stockfishchess.org/tests/view/5f3f98d5dc02a01a0c2881f7

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 32280 W: 1828 L: 1653 D: 28799
Ptnml(0-2): 34, 1428, 13051, 1583, 44
https://tests.stockfishchess.org/tests/view/5f3fe77a87a5c3c63d8f5332

closes https://github.com/official-stockfish/Stockfish/pull/3048

Bench: 3760677
---
 src/tt.cpp  | 21 +++++++++++----------
 src/tt.h    | 12 ++++++------
 src/types.h |  3 ++-
 3 files changed, 19 insertions(+), 17 deletions(-)

diff --git a/src/tt.cpp b/src/tt.cpp
index d494c27d..60a3a5f1 100644
--- a/src/tt.cpp
+++ b/src/tt.cpp
@@ -37,18 +37,19 @@ void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev)
   if (m || (uint16_t)k != key16)
       move16 = (uint16_t)m;
 
-  // Overwrite less valuable entries
-  if ((uint16_t)k != key16
-      || d - DEPTH_OFFSET > depth8 - 4
-      || b == BOUND_EXACT)
+  // Overwrite less valuable entries (cheapest checks first)
+  if (b == BOUND_EXACT
+      || (uint16_t)k != key16
+      || d - DEPTH_OFFSET > depth8 - 4)
   {
-      assert(d >= DEPTH_OFFSET);
+      assert(d > DEPTH_OFFSET);
+      assert(d < 256 + DEPTH_OFFSET);
 
       key16     = (uint16_t)k;
+      depth8    = (uint8_t)(d - DEPTH_OFFSET);
+      genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
       value16   = (int16_t)v;
       eval16    = (int16_t)ev;
-      genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
-      depth8    = (uint8_t)(d - DEPTH_OFFSET);
   }
 }
 
@@ -119,11 +120,11 @@ TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
   const uint16_t key16 = (uint16_t)key;  // Use the low 16 bits as key inside the cluster
 
   for (int i = 0; i < ClusterSize; ++i)
-      if (!tte[i].key16 || tte[i].key16 == key16)
+      if (tte[i].key16 == key16 || !tte[i].depth8)
       {
           tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & 0x7)); // Refresh
 
-          return found = (bool)tte[i].key16, &tte[i];
+          return found = (bool)tte[i].depth8, &tte[i];
       }
 
   // Find an entry to be replaced according to the replacement strategy
@@ -149,7 +150,7 @@ int TranspositionTable::hashfull() const {
   int cnt = 0;
   for (int i = 0; i < 1000; ++i)
       for (int j = 0; j < ClusterSize; ++j)
-          cnt += (table[i].entry[j].genBound8 & 0xF8) == generation8;
+          cnt += table[i].entry[j].depth8 && (table[i].entry[j].genBound8 & 0xF8) == generation8;
 
   return cnt / ClusterSize;
 }
diff --git a/src/tt.h b/src/tt.h
index c177ca52..fdfd6769 100644
--- a/src/tt.h
+++ b/src/tt.h
@@ -25,13 +25,13 @@
 /// TTEntry struct is the 10 bytes transposition table entry, defined as below:
 ///
 /// key        16 bit
-/// move       16 bit
-/// value      16 bit
-/// eval value 16 bit
+/// depth       8 bit
 /// generation  5 bit
 /// pv node     1 bit
 /// bound type  2 bit
-/// depth       8 bit
+/// move       16 bit
+/// value      16 bit
+/// eval value 16 bit
 
 struct TTEntry {
 
@@ -47,11 +47,11 @@ private:
   friend class TranspositionTable;
 
   uint16_t key16;
+  uint8_t  depth8;
+  uint8_t  genBound8;
   uint16_t move16;
   int16_t  value16;
   int16_t  eval16;
-  uint8_t  genBound8;
-  uint8_t  depth8;
 };
 
 
diff --git a/src/types.h b/src/types.h
index 73da41e2..1cd711b6 100644
--- a/src/types.h
+++ b/src/types.h
@@ -232,7 +232,8 @@ enum : int {
   DEPTH_QS_RECAPTURES = -5,
 
   DEPTH_NONE   = -6,
-  DEPTH_OFFSET = DEPTH_NONE
+
+  DEPTH_OFFSET = -7 // value used only for TT entry occupancy check
 };
 
 enum Square : int {

From 843a961a8c10d5949e04718b829e3b3d5adeedb4 Mon Sep 17 00:00:00 2001
From: Vizvezdenec <Vizvezdenec@gmail.com>
Date: Mon, 24 Aug 2020 08:04:16 +0300
Subject: [PATCH 16/52] Introduce countermove based pruning for qsearch

This patch continues work of previous patch in introducing pruning heuristics in qsearch by analogy to main search, now with countermove based pruning.
Idea is that if move is late enough and is quite check (we do generate them in qsearch) and has bad enough countermove history - prune it.

passed STC
https://tests.stockfishchess.org/tests/view/5f41220287a5c3c63d8f53c5
LLR: 2.93 (-2.94,2.94) {-0.25,1.25}
Total: 35944 W: 4127 L: 3929 D: 27888
Ptnml(0-2): 196, 2970, 11459, 3134, 213

passed LTC
https://tests.stockfishchess.org/tests/view/5f41862f87a5c3c63d8f53e8
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 138448 W: 7655 L: 7252 D: 123541
Ptnml(0-2): 145, 6247, 56043, 6638, 151

closes https://github.com/official-stockfish/Stockfish/pull/3058

Bench: 3610676
---
 src/search.cpp | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/src/search.cpp b/src/search.cpp
index 266e2db3..2ca64a01 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1570,6 +1570,12 @@ moves_loop: // When in check, search starts from here
                                                                 [pos.moved_piece(move)]
                                                                 [to_sq(move)];
 
+      if (  !captureOrPromotion
+          && moveCount >= abs(depth) + 1
+          && (*contHist[0])[pos.moved_piece(move)][to_sq(move)] < CounterMovePruneThreshold
+          && (*contHist[1])[pos.moved_piece(move)][to_sq(move)] < CounterMovePruneThreshold)
+          continue;
+
       // Make and search the move
       pos.do_move(move, st, givesCheck);
       value = -qsearch<NT>(pos, ss+1, -beta, -alpha, depth - 1);

From 530fccbf272ffe424ae53a464b91db148cc968ae Mon Sep 17 00:00:00 2001
From: mstembera <MissingEmail@email>
Date: Mon, 24 Aug 2020 03:38:01 -0700
Subject: [PATCH 17/52] Allow for VNNI256 compilation with g++-8

explicitly pass needed -mavx512f -mavx512bw flags

closes https://github.com/official-stockfish/Stockfish/pull/3061

No functional change
---
 .travis.yml  | 3 +--
 src/Makefile | 2 +-
 2 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/.travis.yml b/.travis.yml
index c1e6d6df..092c7f53 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -79,8 +79,7 @@ script:
   - make clean && make -j2 ARCH=x86-64-bmi2 build
   - make clean && make -j2 ARCH=x86-64-avx512 build
   - make clean && make -j2 ARCH=x86-64-vnni512 build
-  # requires gcc 9 or higher
-  - if [[ "$COMPILER" != "g++-8" ]]; make clean && make -j2 ARCH=x86-64-vnni256 build; fi
+  - make clean && make -j2 ARCH=x86-64-vnni256 build
 
   #
   # Check perft and reproducible search
diff --git a/src/Makefile b/src/Makefile
index 228ea851..2e85a144 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -516,7 +516,7 @@ endif
 ifeq ($(vnni256),yes)
 	CXXFLAGS += -DUSE_VNNI
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
-		CXXFLAGS += -mavx512vnni -mavx512dq -mavx512vl -mprefer-vector-width=256
+		CXXFLAGS += -mavx512f -mavx512bw -mavx512vnni -mavx512dq -mavx512vl -mprefer-vector-width=256
 	endif
 endif
 

From b0b4ca17db49ed03057b5fa4ee4a12dab0e9c9e6 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Mon, 24 Aug 2020 21:32:04 +0200
Subject: [PATCH 18/52] Check ARCH=.... variable

to prevent user errors or generating untested code,
check explicitly that the ARCH variable is equivalent to a supported architecture
as listed in `make help`.

To nevertheless compile for an untested target the user can override the internal
variable, passing the undocumented `SUPPORTED_ARCH=true` to make.

closes https://github.com/official-stockfish/Stockfish/pull/3062

No functional change.
---
 src/Makefile | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/src/Makefile b/src/Makefile
index 2e85a144..703aa230 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -85,8 +85,15 @@ endif
 
 ### 2.1. General and architecture defaults
 
-ifeq ($(ARCH),)
-    empty_arch = yes
+# explicitly check for the list of supported architectures (as listed with make help),
+# the user can override with `make ARCH=x86-32-vnni256 SUPPORTED_ARCH=true`
+ifeq ($(ARCH),$(filter $(ARCH),x86-64-vnni512 x86-64-vnni256 x86-64-avx512 x86-64-bmi2 x86-64-avx2 \
+                               x86-64-sse41-popcnt x86-64-modern x86-64-ssse3 x86-64-sse3-popcnt \
+                               x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-32 \
+                               armv7 armv7-neon armv8 apple-silicon general-64 general-32))
+   SUPPORTED_ARCH=true
+else
+   SUPPORTED_ARCH=false
 endif
 
 optimize = yes
@@ -625,6 +632,7 @@ endif
 ### Section 4. Public Targets
 ### ==========================================================================
 
+
 help:
 	@echo ""
 	@echo "To compile stockfish, type: "
@@ -684,10 +692,12 @@ help:
 	@echo "make -j profile-build ARCH=x86-64-bmi2 COMP=gcc COMPCXX=g++-9.0"
 	@echo "make -j build ARCH=x86-64-ssse3 COMP=clang"
 	@echo ""
-ifneq ($(empty_arch), yes)
 	@echo "-------------------------------"
+ifeq ($(SUPPORTED_ARCH), true)
 	@echo "The selected architecture $(ARCH) will enable the following configuration: "
 	@$(MAKE) ARCH=$(ARCH) COMP=$(COMP) config-sanity
+else
+	@echo "Specify a supported architecture with the ARCH option for more details"
 endif
 
 
@@ -802,6 +812,7 @@ config-sanity:
 	@test "$(debug)" = "yes" || test "$(debug)" = "no"
 	@test "$(sanitize)" = "undefined" || test "$(sanitize)" = "thread" || test "$(sanitize)" = "address" || test "$(sanitize)" = "no"
 	@test "$(optimize)" = "yes" || test "$(optimize)" = "no"
+	@test "$(SUPPORTED_ARCH)" = "true"
 	@test "$(arch)" = "any" || test "$(arch)" = "x86_64" || test "$(arch)" = "i386" || \
 	 test "$(arch)" = "ppc64" || test "$(arch)" = "ppc" || \
 	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64"

From 9b4967071e2fb116673820127522bc43d01d2257 Mon Sep 17 00:00:00 2001
From: syzygy1 <3028851+syzygy1@users.noreply.github.com>
Date: Mon, 24 Aug 2020 02:29:38 +0200
Subject: [PATCH 19/52] Remove EvalList

This patch removes the EvalList structure from the Position object and generally simplifies the interface between do_move() and the NNUE code.

The NNUE evaluation function first calculates the "accumulator". The accumulator consists of two halves: one for white's perspective, one for black's perspective.

If the "friendly king" has moved or the accumulator for the parent position is not available, the accumulator for this half has to be calculated from scratch. To do this, the NNUE node needs to know the positions and types of all non-king pieces and the position of the friendly king. This information can easily be obtained from the Position object.

If the "friendly king" has not moved, its half of the accumulator can be calculated by incrementally updating the accumulator for the previous position. For this, the NNUE code needs to know which pieces have been added to which squares and which pieces have been removed from which squares. In principle this information can be derived from the Position object and StateInfo struct (in the same way as undo_move() does this). However, it is probably a bit faster to prepare this information in do_move(), so I have kept the DirtyPiece struct. Since the DirtyPiece struct now stores the squares rather than "PieceSquare" indices, there are now at most three "dirty pieces" (previously two). A promotion move that captures a piece removes the capturing pawn and the captured piece from the board (to SQ_NONE) and moves the promoted piece to the promotion square (from SQ_NONE).

An STC test has confirmed a small speedup:

https://tests.stockfishchess.org/tests/view/5f43f06b5089a564a10d850a
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 87704 W: 9763 L: 9500 D: 68441
Ptnml(0-2): 426, 6950, 28845, 7197, 434

closes https://github.com/official-stockfish/Stockfish/pull/3068

No functional change
---
 src/nnue/evaluate_nnue.cpp      |  43 ++++++------
 src/nnue/features/feature_set.h |   3 +-
 src/nnue/features/half_kp.cpp   |  58 ++++++----------
 src/nnue/features/half_kp.h     |  10 +--
 src/nnue/nnue_common.h          |  21 ++++++
 src/position.cpp                |  95 +++++++-------------------
 src/position.h                  |  23 -------
 src/types.h                     | 116 +++-----------------------------
 8 files changed, 96 insertions(+), 273 deletions(-)

diff --git a/src/nnue/evaluate_nnue.cpp b/src/nnue/evaluate_nnue.cpp
index dfbb1ac2..e6619089 100644
--- a/src/nnue/evaluate_nnue.cpp
+++ b/src/nnue/evaluate_nnue.cpp
@@ -29,30 +29,29 @@
 
 #include "evaluate_nnue.h"
 
-ExtPieceSquare kpp_board_index[PIECE_NB] = {
- // convention: W - us, B - them
- // viewed from other side, W and B are reversed
-    { PS_NONE,     PS_NONE     },
-    { PS_W_PAWN,   PS_B_PAWN   },
-    { PS_W_KNIGHT, PS_B_KNIGHT },
-    { PS_W_BISHOP, PS_B_BISHOP },
-    { PS_W_ROOK,   PS_B_ROOK   },
-    { PS_W_QUEEN,  PS_B_QUEEN  },
-    { PS_W_KING,   PS_B_KING   },
-    { PS_NONE,     PS_NONE     },
-    { PS_NONE,     PS_NONE     },
-    { PS_B_PAWN,   PS_W_PAWN   },
-    { PS_B_KNIGHT, PS_W_KNIGHT },
-    { PS_B_BISHOP, PS_W_BISHOP },
-    { PS_B_ROOK,   PS_W_ROOK   },
-    { PS_B_QUEEN,  PS_W_QUEEN  },
-    { PS_B_KING,   PS_W_KING   },
-    { PS_NONE,     PS_NONE     }
-};
-
-
 namespace Eval::NNUE {
 
+  uint32_t kpp_board_index[PIECE_NB][COLOR_NB] = {
+   // convention: W - us, B - them
+   // viewed from other side, W and B are reversed
+      { PS_NONE,     PS_NONE     },
+      { PS_W_PAWN,   PS_B_PAWN   },
+      { PS_W_KNIGHT, PS_B_KNIGHT },
+      { PS_W_BISHOP, PS_B_BISHOP },
+      { PS_W_ROOK,   PS_B_ROOK   },
+      { PS_W_QUEEN,  PS_B_QUEEN  },
+      { PS_W_KING,   PS_B_KING   },
+      { PS_NONE,     PS_NONE     },
+      { PS_NONE,     PS_NONE     },
+      { PS_B_PAWN,   PS_W_PAWN   },
+      { PS_B_KNIGHT, PS_W_KNIGHT },
+      { PS_B_BISHOP, PS_W_BISHOP },
+      { PS_B_ROOK,   PS_W_ROOK   },
+      { PS_B_QUEEN,  PS_W_QUEEN  },
+      { PS_B_KING,   PS_W_KING   },
+      { PS_NONE,     PS_NONE     }
+  };
+
   // Input feature converter
   AlignedPtr<FeatureTransformer> feature_transformer;
 
diff --git a/src/nnue/features/feature_set.h b/src/nnue/features/feature_set.h
index 79ca83ae..558a6b22 100644
--- a/src/nnue/features/feature_set.h
+++ b/src/nnue/features/feature_set.h
@@ -68,8 +68,7 @@ namespace Eval::NNUE::Features {
         reset[perspective] = false;
         switch (trigger) {
           case TriggerEvent::kFriendKingMoved:
-            reset[perspective] =
-                dp.pieceId[0] == PIECE_ID_KING + perspective;
+            reset[perspective] = dp.piece[0] == make_piece(perspective, KING);
             break;
           default:
             assert(false);
diff --git a/src/nnue/features/half_kp.cpp b/src/nnue/features/half_kp.cpp
index 628add6e..88e384a3 100644
--- a/src/nnue/features/half_kp.cpp
+++ b/src/nnue/features/half_kp.cpp
@@ -23,25 +23,17 @@
 
 namespace Eval::NNUE::Features {
 
-  // Find the index of the feature quantity from the king position and PieceSquare
-  template <Side AssociatedKing>
-  inline IndexType HalfKP<AssociatedKing>::MakeIndex(Square sq_k, PieceSquare p) {
-    return static_cast<IndexType>(PS_END) * static_cast<IndexType>(sq_k) + p;
+  // Orient a square according to perspective (rotates by 180 for black)
+  inline Square orient(Color perspective, Square s) {
+    return Square(int(s) ^ (bool(perspective) * 63));
   }
 
-  // Get pieces information
+  // Find the index of the feature quantity from the king position and PieceSquare
   template <Side AssociatedKing>
-  inline void HalfKP<AssociatedKing>::GetPieces(
-      const Position& pos, Color perspective,
-      PieceSquare** pieces, Square* sq_target_k) {
+  inline IndexType HalfKP<AssociatedKing>::MakeIndex(
+      Color perspective, Square s, Piece pc, Square ksq) {
 
-    *pieces = (perspective == BLACK) ?
-        pos.eval_list()->piece_list_fb() :
-        pos.eval_list()->piece_list_fw();
-    const PieceId target = (AssociatedKing == Side::kFriend) ?
-        static_cast<PieceId>(PIECE_ID_KING + perspective) :
-        static_cast<PieceId>(PIECE_ID_KING + ~perspective);
-    *sq_target_k = static_cast<Square>(((*pieces)[target] - PS_W_KING) % SQUARE_NB);
+    return IndexType(orient(perspective, s) + kpp_board_index[pc][perspective] + PS_END * ksq);
   }
 
   // Get a list of indices for active features
@@ -49,16 +41,11 @@ namespace Eval::NNUE::Features {
   void HalfKP<AssociatedKing>::AppendActiveIndices(
       const Position& pos, Color perspective, IndexList* active) {
 
-    // Do nothing if array size is small to avoid compiler warning
-    if (RawFeatures::kMaxActiveDimensions < kMaxActiveDimensions) return;
-
-    PieceSquare* pieces;
-    Square sq_target_k;
-    GetPieces(pos, perspective, &pieces, &sq_target_k);
-    for (PieceId i = PIECE_ID_ZERO; i < PIECE_ID_KING; ++i) {
-      if (pieces[i] != PS_NONE) {
-        active->push_back(MakeIndex(sq_target_k, pieces[i]));
-      }
+    Square ksq = orient(perspective, pos.square<KING>(perspective));
+    Bitboard bb = pos.pieces() & ~pos.pieces(KING);
+    while (bb) {
+      Square s = pop_lsb(&bb);
+      active->push_back(MakeIndex(perspective, s, pos.piece_on(s), ksq));
     }
   }
 
@@ -68,22 +55,15 @@ namespace Eval::NNUE::Features {
       const Position& pos, Color perspective,
       IndexList* removed, IndexList* added) {
 
-    PieceSquare* pieces;
-    Square sq_target_k;
-    GetPieces(pos, perspective, &pieces, &sq_target_k);
+    Square ksq = orient(perspective, pos.square<KING>(perspective));
     const auto& dp = pos.state()->dirtyPiece;
     for (int i = 0; i < dp.dirty_num; ++i) {
-      if (dp.pieceId[i] >= PIECE_ID_KING) continue;
-      const auto old_p = static_cast<PieceSquare>(
-          dp.old_piece[i].from[perspective]);
-      if (old_p != PS_NONE) {
-        removed->push_back(MakeIndex(sq_target_k, old_p));
-      }
-      const auto new_p = static_cast<PieceSquare>(
-          dp.new_piece[i].from[perspective]);
-      if (new_p != PS_NONE) {
-        added->push_back(MakeIndex(sq_target_k, new_p));
-      }
+      Piece pc = dp.piece[i];
+      if (type_of(pc) == KING) continue;
+      if (dp.from[i] != SQ_NONE)
+        removed->push_back(MakeIndex(perspective, dp.from[i], pc, ksq));
+      if (dp.to[i] != SQ_NONE)
+        added->push_back(MakeIndex(perspective, dp.to[i], pc, ksq));
     }
   }
 
diff --git a/src/nnue/features/half_kp.h b/src/nnue/features/half_kp.h
index 99842eea..ee6a8df3 100644
--- a/src/nnue/features/half_kp.h
+++ b/src/nnue/features/half_kp.h
@@ -41,7 +41,7 @@ namespace Eval::NNUE::Features {
     static constexpr IndexType kDimensions =
         static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_END);
     // Maximum number of simultaneously active features
-    static constexpr IndexType kMaxActiveDimensions = PIECE_ID_KING;
+    static constexpr IndexType kMaxActiveDimensions = 30; // Kings don't count
     // Trigger for full calculation instead of difference calculation
     static constexpr TriggerEvent kRefreshTrigger = TriggerEvent::kFriendKingMoved;
 
@@ -53,13 +53,9 @@ namespace Eval::NNUE::Features {
     static void AppendChangedIndices(const Position& pos, Color perspective,
                                      IndexList* removed, IndexList* added);
 
-    // Index of a feature for a given king position and another piece on some square
-    static IndexType MakeIndex(Square sq_k, PieceSquare p);
-
    private:
-    // Get pieces information
-    static void GetPieces(const Position& pos, Color perspective,
-                          PieceSquare** pieces, Square* sq_target_k);
+    // Index of a feature for a given king position and another piece on some square
+    static IndexType MakeIndex(Color perspective, Square s, Piece pc, Square sq_k);
   };
 
 }  // namespace Eval::NNUE::Features
diff --git a/src/nnue/nnue_common.h b/src/nnue/nnue_common.h
index a9d8e5af..7bc905dc 100644
--- a/src/nnue/nnue_common.h
+++ b/src/nnue/nnue_common.h
@@ -94,6 +94,27 @@ namespace Eval::NNUE {
 
   constexpr std::size_t kMaxSimdWidth = 32;
 
+  // unique number for each piece type on each square
+  enum {
+    PS_NONE     =  0,
+    PS_W_PAWN   =  1,
+    PS_B_PAWN   =  1 * SQUARE_NB + 1,
+    PS_W_KNIGHT =  2 * SQUARE_NB + 1,
+    PS_B_KNIGHT =  3 * SQUARE_NB + 1,
+    PS_W_BISHOP =  4 * SQUARE_NB + 1,
+    PS_B_BISHOP =  5 * SQUARE_NB + 1,
+    PS_W_ROOK   =  6 * SQUARE_NB + 1,
+    PS_B_ROOK   =  7 * SQUARE_NB + 1,
+    PS_W_QUEEN  =  8 * SQUARE_NB + 1,
+    PS_B_QUEEN  =  9 * SQUARE_NB + 1,
+    PS_W_KING   = 10 * SQUARE_NB + 1,
+    PS_END      = PS_W_KING, // pieces without kings (pawns included)
+    PS_B_KING   = 11 * SQUARE_NB + 1,
+    PS_END2     = 12 * SQUARE_NB + 1
+  };
+
+  extern uint32_t kpp_board_index[PIECE_NB][COLOR_NB];
+
   // Type of input feature after conversion
   using TransformedFeatureType = std::uint8_t;
   using IndexType = std::uint32_t;
diff --git a/src/position.cpp b/src/position.cpp
index 02898547..fe89b753 100644
--- a/src/position.cpp
+++ b/src/position.cpp
@@ -198,9 +198,6 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
   std::fill_n(&pieceList[0][0], sizeof(pieceList) / sizeof(Square), SQ_NONE);
   st = si;
 
-  // Each piece on board gets a unique ID used to track the piece later
-  PieceId piece_id, next_piece_id = PIECE_ID_ZERO;
-
   ss >> std::noskipws;
 
   // 1. Piece placement
@@ -212,21 +209,8 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
       else if (token == '/')
           sq += 2 * SOUTH;
 
-      else if ((idx = PieceToChar.find(token)) != string::npos)
-      {
-          auto pc = Piece(idx);
-          put_piece(pc, sq);
-
-          if (Eval::useNNUE)
-          {
-              // Kings get a fixed ID, other pieces get ID in order of placement
-              piece_id =
-                (idx == W_KING) ? PIECE_ID_WKING :
-                (idx == B_KING) ? PIECE_ID_BKING :
-                next_piece_id++;
-              evalList.put_piece(piece_id, sq, pc);
-          }
-
+      else if ((idx = PieceToChar.find(token)) != string::npos) {
+          put_piece(Piece(idx), sq);
           ++sq;
       }
   }
@@ -721,8 +705,6 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   // Used by NNUE
   st->accumulator.computed_accumulation = false;
   st->accumulator.computed_score = false;
-  PieceId dp0 = PIECE_ID_NONE;
-  PieceId dp1 = PIECE_ID_NONE;
   auto& dp = st->dirtyPiece;
   dp.dirty_num = 1;
 
@@ -775,12 +757,10 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
 
       if (Eval::useNNUE)
       {
-          dp.dirty_num = 2; // 2 pieces moved
-          dp1 = piece_id_on(capsq);
-          dp.pieceId[1] = dp1;
-          dp.old_piece[1] = evalList.piece_with_id(dp1);
-          evalList.put_piece(dp1, capsq, NO_PIECE);
-          dp.new_piece[1] = evalList.piece_with_id(dp1);
+          dp.dirty_num = 2;  // 1 piece moved, 1 piece captured
+          dp.piece[1] = captured;
+          dp.from[1] = capsq;
+          dp.to[1] = SQ_NONE;
       }
 
       // Update board and piece lists
@@ -821,11 +801,9 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   {
       if (Eval::useNNUE)
       {
-          dp0 = piece_id_on(from);
-          dp.pieceId[0] = dp0;
-          dp.old_piece[0] = evalList.piece_with_id(dp0);
-          evalList.put_piece(dp0, to, pc);
-          dp.new_piece[0] = evalList.piece_with_id(dp0);
+          dp.piece[0] = pc;
+          dp.from[0] = from;
+          dp.to[0] = to;
       }
 
       move_piece(from, to);
@@ -854,9 +832,12 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
 
           if (Eval::useNNUE)
           {
-              dp0 = piece_id_on(to);
-              evalList.put_piece(dp0, to, promotion);
-              dp.new_piece[0] = evalList.piece_with_id(dp0);
+              // Promoting pawn to SQ_NONE, promoted piece from SQ_NONE
+              dp.to[0] = SQ_NONE;
+              dp.piece[dp.dirty_num] = promotion;
+              dp.from[dp.dirty_num] = SQ_NONE;
+              dp.to[dp.dirty_num] = to;
+              dp.dirty_num++;
           }
 
           // Update hash keys
@@ -950,12 +931,6 @@ void Position::undo_move(Move m) {
   {
       move_piece(to, from); // Put the piece back at the source square
 
-      if (Eval::useNNUE)
-      {
-          PieceId dp0 = st->dirtyPiece.pieceId[0];
-          evalList.put_piece(dp0, from, pc);
-      }
-
       if (st->capturedPiece)
       {
           Square capsq = to;
@@ -972,14 +947,6 @@ void Position::undo_move(Move m) {
           }
 
           put_piece(st->capturedPiece, capsq); // Restore the captured piece
-
-          if (Eval::useNNUE)
-          {
-              PieceId dp1 = st->dirtyPiece.pieceId[1];
-              assert(evalList.piece_with_id(dp1).from[WHITE] == PS_NONE);
-              assert(evalList.piece_with_id(dp1).from[BLACK] == PS_NONE);
-              evalList.put_piece(dp1, capsq, st->capturedPiece);
-          }
       }
   }
 
@@ -1001,32 +968,16 @@ void Position::do_castling(Color us, Square from, Square& to, Square& rfrom, Squ
   rto = relative_square(us, kingSide ? SQ_F1 : SQ_D1);
   to = relative_square(us, kingSide ? SQ_G1 : SQ_C1);
 
-  if (Eval::useNNUE)
+  if (Do && Eval::useNNUE)
   {
-      PieceId dp0, dp1;
       auto& dp = st->dirtyPiece;
-      dp.dirty_num = 2; // 2 pieces moved
-
-      if (Do)
-      {
-          dp0 = piece_id_on(from);
-          dp1 = piece_id_on(rfrom);
-          dp.pieceId[0] = dp0;
-          dp.old_piece[0] = evalList.piece_with_id(dp0);
-          evalList.put_piece(dp0, to, make_piece(us, KING));
-          dp.new_piece[0] = evalList.piece_with_id(dp0);
-          dp.pieceId[1] = dp1;
-          dp.old_piece[1] = evalList.piece_with_id(dp1);
-          evalList.put_piece(dp1, rto, make_piece(us, ROOK));
-          dp.new_piece[1] = evalList.piece_with_id(dp1);
-      }
-      else
-      {
-          dp0 = piece_id_on(to);
-          dp1 = piece_id_on(rto);
-          evalList.put_piece(dp0, from, make_piece(us, KING));
-          evalList.put_piece(dp1, rfrom, make_piece(us, ROOK));
-      }
+      dp.piece[0] = make_piece(us, KING);
+      dp.from[0] = from;
+      dp.to[0] = to;
+      dp.piece[1] = make_piece(us, ROOK);
+      dp.from[1] = rfrom;
+      dp.to[1] = rto;
+      dp.dirty_num = 2;
   }
 
   // Remove both pieces first since squares could overlap in Chess960
diff --git a/src/position.h b/src/position.h
index 5ce17277..d6f5c9fd 100644
--- a/src/position.h
+++ b/src/position.h
@@ -171,7 +171,6 @@ public:
 
   // Used by NNUE
   StateInfo* state() const;
-  const EvalList* eval_list() const;
 
 private:
   // Initialization helpers (used while setting up a position)
@@ -186,9 +185,6 @@ private:
   template<bool Do>
   void do_castling(Color us, Square from, Square& to, Square& rfrom, Square& rto);
 
-  // ID of a piece on a given square
-  PieceId piece_id_on(Square sq) const;
-
   // Data members
   Piece board[SQUARE_NB];
   Bitboard byTypeBB[PIECE_TYPE_NB];
@@ -205,9 +201,6 @@ private:
   Thread* thisThread;
   StateInfo* st;
   bool chess960;
-
-  // List of pieces used in NNUE evaluation function
-  EvalList evalList;
 };
 
 namespace PSQT {
@@ -451,20 +444,4 @@ inline StateInfo* Position::state() const {
   return st;
 }
 
-inline const EvalList* Position::eval_list() const {
-
-  return &evalList;
-}
-
-inline PieceId Position::piece_id_on(Square sq) const
-{
-
-  assert(piece_on(sq) != NO_PIECE);
-
-  PieceId pid = evalList.piece_id_list[sq];
-  assert(is_ok(pid));
-
-  return pid;
-}
-
 #endif // #ifndef POSITION_H_INCLUDED
diff --git a/src/types.h b/src/types.h
index 1cd711b6..5873c698 100644
--- a/src/types.h
+++ b/src/types.h
@@ -201,22 +201,6 @@ enum Piece {
   PIECE_NB = 16
 };
 
-// An ID used to track the pieces. Max. 32 pieces on board.
-enum PieceId {
-  PIECE_ID_ZERO   = 0,
-  PIECE_ID_KING   = 30,
-  PIECE_ID_WKING  = 30,
-  PIECE_ID_BKING  = 31,
-  PIECE_ID_NONE   = 32
-};
-
-inline PieceId operator++(PieceId& d, int) {
-
-  PieceId x = d;
-  d = PieceId(int(d) + 1);
-  return x;
-}
-
 constexpr Value PieceValue[PHASE_NB][PIECE_NB] = {
   { VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO,
     VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO },
@@ -271,93 +255,20 @@ enum Rank : int {
   RANK_1, RANK_2, RANK_3, RANK_4, RANK_5, RANK_6, RANK_7, RANK_8, RANK_NB
 };
 
-// unique number for each piece type on each square
-enum PieceSquare : uint32_t {
-  PS_NONE     =  0,
-  PS_W_PAWN   =  1,
-  PS_B_PAWN   =  1 * SQUARE_NB + 1,
-  PS_W_KNIGHT =  2 * SQUARE_NB + 1,
-  PS_B_KNIGHT =  3 * SQUARE_NB + 1,
-  PS_W_BISHOP =  4 * SQUARE_NB + 1,
-  PS_B_BISHOP =  5 * SQUARE_NB + 1,
-  PS_W_ROOK   =  6 * SQUARE_NB + 1,
-  PS_B_ROOK   =  7 * SQUARE_NB + 1,
-  PS_W_QUEEN  =  8 * SQUARE_NB + 1,
-  PS_B_QUEEN  =  9 * SQUARE_NB + 1,
-  PS_W_KING   = 10 * SQUARE_NB + 1,
-  PS_END      = PS_W_KING, // pieces without kings (pawns included)
-  PS_B_KING   = 11 * SQUARE_NB + 1,
-  PS_END2     = 12 * SQUARE_NB + 1
-};
-
-struct ExtPieceSquare {
-  PieceSquare from[COLOR_NB];
-};
-
-// Array for finding the PieceSquare corresponding to the piece on the board
-extern ExtPieceSquare kpp_board_index[PIECE_NB];
-
-constexpr bool is_ok(PieceId pid);
-constexpr Square rotate180(Square sq);
-
-// Structure holding which tracked piece (PieceId) is where (PieceSquare)
-class EvalList {
-
-public:
-  // Max. number of pieces without kings is 30 but must be a multiple of 4 in AVX2
-  static const int MAX_LENGTH = 32;
-
-  // Array that holds the piece id for the pieces on the board
-  PieceId piece_id_list[SQUARE_NB];
-
-  // List of pieces, separate from White and Black POV
-  PieceSquare* piece_list_fw() const { return const_cast<PieceSquare*>(pieceListFw); }
-  PieceSquare* piece_list_fb() const { return const_cast<PieceSquare*>(pieceListFb); }
-
-  // Place the piece pc with piece_id on the square sq on the board
-  void put_piece(PieceId piece_id, Square sq, Piece pc)
-  {
-      assert(is_ok(piece_id));
-      if (pc != NO_PIECE)
-      {
-          pieceListFw[piece_id] = PieceSquare(kpp_board_index[pc].from[WHITE] + sq);
-          pieceListFb[piece_id] = PieceSquare(kpp_board_index[pc].from[BLACK] + rotate180(sq));
-          piece_id_list[sq] = piece_id;
-      }
-      else
-      {
-          pieceListFw[piece_id] = PS_NONE;
-          pieceListFb[piece_id] = PS_NONE;
-          piece_id_list[sq] = piece_id;
-      }
-  }
-
-  // Convert the specified piece_id piece to ExtPieceSquare type and return it
-  ExtPieceSquare piece_with_id(PieceId piece_id) const
-  {
-      ExtPieceSquare eps;
-      eps.from[WHITE] = pieceListFw[piece_id];
-      eps.from[BLACK] = pieceListFb[piece_id];
-      return eps;
-  }
-
-private:
-  PieceSquare pieceListFw[MAX_LENGTH];
-  PieceSquare pieceListFb[MAX_LENGTH];
-};
-
-// For differential evaluation of pieces that changed since last turn
+// Keep track of what a move changes on the board (used by NNUE)
 struct DirtyPiece {
 
   // Number of changed pieces
   int dirty_num;
 
-  // The ids of changed pieces, max. 2 pieces can change in one move
-  PieceId pieceId[2];
+  // Max 3 pieces can change in one move. A promotion with capture moves
+  // both the pawn and the captured piece to SQ_NONE and the piece promoted
+  // to from SQ_NONE to the capture square.
+  Piece piece[3];
 
-  // What changed from the piece with that piece number
-  ExtPieceSquare old_piece[2];
-  ExtPieceSquare new_piece[2];
+  // From and to squares, which may be SQ_NONE
+  Square from[3];
+  Square to[3];
 };
 
 /// Score enum stores a middlegame and an endgame value in a single integer (enum).
@@ -407,8 +318,6 @@ ENABLE_FULL_OPERATORS_ON(Value)
 ENABLE_FULL_OPERATORS_ON(Direction)
 
 ENABLE_INCR_OPERATORS_ON(Piece)
-ENABLE_INCR_OPERATORS_ON(PieceSquare)
-ENABLE_INCR_OPERATORS_ON(PieceId)
 ENABLE_INCR_OPERATORS_ON(PieceType)
 ENABLE_INCR_OPERATORS_ON(Square)
 ENABLE_INCR_OPERATORS_ON(File)
@@ -497,10 +406,6 @@ inline Color color_of(Piece pc) {
   return Color(pc >> 3);
 }
 
-constexpr bool is_ok(PieceId pid) {
-  return pid < PIECE_ID_NONE;
-}
-
 constexpr bool is_ok(Square s) {
   return s >= SQ_A1 && s <= SQ_H8;
 }
@@ -537,11 +442,6 @@ constexpr Square to_sq(Move m) {
   return Square(m & 0x3F);
 }
 
-// Return relative square when turning the board 180 degrees
-constexpr Square rotate180(Square sq) {
-  return (Square)(sq ^ 0x3F);
-}
-
 constexpr int from_to(Move m) {
  return m & 0xFFF;
 }

From 95b8f3f8005598fc3ad07a7a8f6440d828cabc29 Mon Sep 17 00:00:00 2001
From: VoyagerOne <excelgeek@gmail.com>
Date: Sun, 23 Aug 2020 12:04:50 -0400
Subject: [PATCH 20/52] Remove Reduce Depth

Remove Reduce Depth at PV nodes.

STC:
LLR: 2.94 (-2.94,2.94) {-1.25,0.25}
Total: 56760 W: 6299 L: 6236 D: 44225
Ptnml(0-2): 286, 4843, 18076, 4872, 303
https://tests.stockfishchess.org/tests/view/5f41356087a5c3c63d8f53c9

LTC:
LLR: 2.95 (-2.94,2.94) {-0.75,0.25}
Total: 17496 W: 954 L: 865 D: 15677
Ptnml(0-2): 13, 768, 7098, 855, 14
https://tests.stockfishchess.org/tests/view/5f41bb7687a5c3c63d8f53f9

closes https://github.com/official-stockfish/Stockfish/pull/3055

Bench: 3555051
---
 src/search.cpp | 24 +++++++++---------------
 1 file changed, 9 insertions(+), 15 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index 2ca64a01..d6b611a3 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -940,12 +940,6 @@ namespace {
             }
     }
 
-    // Step 11. If the position is not in TT, decrease depth by 2
-    if (   PvNode
-        && depth >= 6
-        && !ttMove)
-        depth -= 2;
-
 moves_loop: // When in check, search starts from here
 
     const PieceToHistory* contHist[] = { (ss-1)->continuationHistory, (ss-2)->continuationHistory,
@@ -969,7 +963,7 @@ moves_loop: // When in check, search starts from here
     // Mark this node as being searched
     ThreadHolding th(thisThread, posKey, ss->ply);
 
-    // Step 12. Loop through all pseudo-legal moves until no moves remain
+    // Step 11. Loop through all pseudo-legal moves until no moves remain
     // or a beta cutoff occurs.
     while ((move = mp.next_move(moveCountPruning)) != MOVE_NONE)
     {
@@ -1007,7 +1001,7 @@ moves_loop: // When in check, search starts from here
       // Calculate new depth for this move
       newDepth = depth - 1;
 
-      // Step 13. Pruning at shallow depth (~200 Elo)
+      // Step 12. Pruning at shallow depth (~200 Elo)
       if (  !rootNode
           && pos.non_pawn_material(us)
           && bestValue > VALUE_TB_LOSS_IN_MAX_PLY)
@@ -1065,7 +1059,7 @@ moves_loop: // When in check, search starts from here
           }
       }
 
-      // Step 14. Extensions (~75 Elo)
+      // Step 13. Extensions (~75 Elo)
 
       // Singular extension search (~70 Elo). If all moves but one fail low on a
       // search of (alpha-s, beta-s), and just one fails high on (alpha, beta),
@@ -1148,10 +1142,10 @@ moves_loop: // When in check, search starts from here
                                                                 [movedPiece]
                                                                 [to_sq(move)];
 
-      // Step 15. Make the move
+      // Step 14. Make the move
       pos.do_move(move, st, givesCheck);
 
-      // Step 16. Reduced depth search (LMR, ~200 Elo). If the move fails high it will be
+      // Step 15. Reduced depth search (LMR, ~200 Elo). If the move fails high it will be
       // re-searched at full depth.
       if (    depth >= 3
           &&  moveCount > 1 + 2 * rootNode + 2 * (PvNode && abs(bestValue) < 2)
@@ -1254,7 +1248,7 @@ moves_loop: // When in check, search starts from here
           didLMR = false;
       }
 
-      // Step 17. Full depth search when LMR is skipped or fails high
+      // Step 16. Full depth search when LMR is skipped or fails high
       if (doFullDepthSearch)
       {
           value = -search<NonPV>(pos, ss+1, -(alpha+1), -alpha, newDepth, !cutNode);
@@ -1282,12 +1276,12 @@ moves_loop: // When in check, search starts from here
           value = -search<PV>(pos, ss+1, -beta, -alpha, newDepth, false);
       }
 
-      // Step 18. Undo move
+      // Step 17. Undo move
       pos.undo_move(move);
 
       assert(value > -VALUE_INFINITE && value < VALUE_INFINITE);
 
-      // Step 19. Check for a new best move
+      // Step 18. Check for a new best move
       // Finished searching the move. If a stop occurred, the return value of
       // the search cannot be trusted, and we return immediately without
       // updating best move, PV and TT.
@@ -1364,7 +1358,7 @@ moves_loop: // When in check, search starts from here
         return VALUE_DRAW;
     */
 
-    // Step 20. Check for mate and stalemate
+    // Step 19. Check for mate and stalemate
     // All legal moves have been searched and if there are no legal moves, it
     // must be a mate or a stalemate. If we are in a singular extension search then
     // return a fail low score.

From 242a7d9fead561488ca176a4687deef8859918f2 Mon Sep 17 00:00:00 2001
From: VoyagerOne <excelgeek@gmail.com>
Date: Tue, 25 Aug 2020 09:10:47 -0400
Subject: [PATCH 21/52] Simplify MCP in QS

Simplify moveCount pruning in QS by removing depth dependency.

STC
LLR: 2.94 (-2.94,2.94) {-1.25,0.25}
Total: 42960 W: 4741 L: 4661 D: 33558
Ptnml(0-2): 218, 3574, 13804, 3678, 206
https://tests.stockfishchess.org/tests/view/5f42e3f75089a564a10d8493

LTC
LLR: 2.94 (-2.94,2.94) {-0.75,0.25}
Total: 66672 W: 3563 L: 3508 D: 59601
Ptnml(0-2): 71, 3064, 26996, 3149, 56
https://tests.stockfishchess.org/tests/view/5f4353285089a564a10d84d0

closes https://github.com/official-stockfish/Stockfish/pull/3067

Bench: 4074430
---
 src/search.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/search.cpp b/src/search.cpp
index d6b611a3..cae8a684 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1526,7 +1526,7 @@ moves_loop: // When in check, search starts from here
           assert(type_of(move) != ENPASSANT); // Due to !pos.advanced_pawn_push
 
           // moveCount pruning
-          if (moveCount > abs(depth) + 2)
+          if (moveCount > 2)
               continue;
 
           futilityValue = futilityBase + PieceValue[EG][pos.piece_on(to_sq(move))];

From 406979ea12ee7828e079871b0f9f3dc8f127a741 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ste=CC=81phane=20Nicolet?= <cassio@free.fr>
Date: Sun, 23 Aug 2020 13:43:38 +0200
Subject: [PATCH 22/52] Embed default net, and simplify using non-default nets

covers the most important cases from the user perspective:

It embeds the default net in the binary, so a download of that binary will result
in a working engine with the default net. The engine will be functional in the default mode
without any additional user action.

It allows non-default nets to be used, which will be looked for in up to
three directories (working directory, location of the binary, and optionally a specific default directory).
This mechanism is also kept for those developers that use MSVC,
the one compiler that doesn't have an easy mechanism for embedding data.

It is possible to disable embedding, and instead specify a specific directory, e.g. linux distros might want to use
CXXFLAGS="-DNNUE_EMBEDDING_OFF -DDEFAULT_NNUE_DIRECTORY=/usr/share/games/stockfish/" make -j ARCH=x86-64 profile-build

passed STC non-regression:
https://tests.stockfishchess.org/tests/view/5f4a581c150f0aef5f8ae03a
LLR: 2.95 (-2.94,2.94) {-1.25,-0.25}
Total: 66928 W: 7202 L: 7147 D: 52579
Ptnml(0-2): 291, 5309, 22211, 5360, 293

closes https://github.com/official-stockfish/Stockfish/pull/3070

fixes https://github.com/official-stockfish/Stockfish/issues/3030

No functional change.
---
 AUTHORS                    |   3 +-
 README.md                  |  18 +-
 appveyor.yml               |   2 +-
 src/Makefile               |  10 +-
 src/evaluate.cpp           | 103 +++++++++--
 src/evaluate.h             |   7 +-
 src/incbin/UNLICENCE       |  26 +++
 src/incbin/incbin.h        | 368 +++++++++++++++++++++++++++++++++++++
 src/main.cpp               |   1 +
 src/misc.cpp               |  59 ++++++
 src/misc.h                 |   9 +-
 src/nnue/evaluate_nnue.cpp |  14 +-
 src/ucioption.cpp          |   5 +-
 13 files changed, 582 insertions(+), 43 deletions(-)
 create mode 100644 src/incbin/UNLICENCE
 create mode 100755 src/incbin/incbin.h

diff --git a/AUTHORS b/AUTHORS
index c96f870a..c00ab657 100644
--- a/AUTHORS
+++ b/AUTHORS
@@ -36,10 +36,11 @@ Bryan Cross (crossbr)
 candirufish
 Chess13234
 Chris Cain (ceebo)
+Dale Weiler (graphitemaster)
 Dan Schmidt (dfannius)
 Daniel Axtens (daxtens)
 Daniel Dugovic (ddugovic)
-Dariusz Orzechowski
+Dariusz Orzechowski (dorzechowski)
 David Zar
 Daylen Yang (daylen)
 DiscanX
diff --git a/README.md b/README.md
index 2cc88bf4..96a495ae 100644
--- a/README.md
+++ b/README.md
@@ -12,9 +12,9 @@ about how to use Stockfish with it.
 
 The Stockfish engine features two evaluation functions for chess, the classical
 evaluation based on handcrafted terms, and the NNUE evaluation based on efficiently
-updateable neural networks. The classical evaluation runs efficiently on most 64bit
-CPU architectures, while the NNUE evaluation benefits strongly from the vector
-intrinsics available on modern CPUs (avx2 or similar).
+updateable neural networks. The classical evaluation runs efficiently on almost all
+CPU architectures, while the NNUE evaluation benefits from the vector
+intrinsics available on most CPUs (sse2, avx2, neon, or similar).
 
 
 ## Files
@@ -29,10 +29,11 @@ This distribution of Stockfish consists of the following files:
     that can be used to compile Stockfish on Unix-like systems.
 
   * a file with the .nnue extension, storing the neural network for the NNUE 
-    evaluation.
+    evaluation. Binary distributions will have this file embedded.
 
 Note: to use the NNUE evaluation, the additional data file with neural network parameters
-needs to be downloaded. The filename for the default net can be found as the default
+needs to be available. Normally, this file is already embedded in the binary or it can be downloaded.
+The filename for the default (recommended) net can be found as the default
 value of the `EvalFile` UCI option, with the format `nn-[SHA256 first 12 digits].nnue`
 (for instance, `nn-c157e0a5755b.nnue`). This file can be downloaded from
 ```
@@ -61,11 +62,14 @@ Currently, Stockfish has the following UCI options:
 
   * #### Use NNUE
     Toggle between the NNUE and classical evaluation functions. If set to "true",
-    the network parameters must be available to load from file (see also EvalFile).
+    the network parameters must be available to load from file (see also EvalFile),
+    if they are not embedded in the binary.
 
   * #### EvalFile
     The name of the file of the NNUE evaluation parameters. Depending on the GUI the
-    filename should include the full path to the folder/directory that contains the file.
+    filename might have to include the full path to the folder/directory that contains the file.
+    Other locations, such as the directory that contains the binary and the working directory,
+    are also searched.
 
   * #### UCI_AnalyseMode
     An option handled by your GUI.
diff --git a/appveyor.yml b/appveyor.yml
index a3732a23..ab608409 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -63,7 +63,7 @@ build_script:
   - cmake --build . --config %CONFIGURATION% -- /verbosity:minimal
   - ps: |
       # Download default NNUE net from fishtest
-      $nnuenet = Get-Content -Path src\ucioption.cpp | Select-String -CaseSensitive -Pattern "Option" | Select-String -CaseSensitive -Pattern "nn-[a-z0-9]{12}.nnue"
+      $nnuenet = Get-Content -Path src\evaluate.h | Select-String -CaseSensitive -Pattern "EvalFileDefaultName" | Select-String -CaseSensitive -Pattern "nn-[a-z0-9]{12}.nnue"
       $dummy = $nnuenet -match "(?<nnuenet>nn-[a-z0-9]{12}.nnue)"
       $nnuenet = $Matches.nnuenet
       Write-Host "Default net:" $nnuenet
diff --git a/src/Makefile b/src/Makefile
index 703aa230..5f363f02 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -614,10 +614,12 @@ ifeq ($(debug), no)
 # So, only enable it for a cross from Linux by default.
 	else ifeq ($(comp),mingw)
 	ifeq ($(KERNEL),Linux)
+	ifneq ($(arch),i386)
 		CXXFLAGS += -flto
 		LDFLAGS += $(CXXFLAGS) -flto=jobserver
 	endif
 	endif
+	endif
 endif
 endif
 
@@ -705,7 +707,7 @@ endif
         config-sanity icc-profile-use icc-profile-make gcc-profile-use gcc-profile-make \
         clang-profile-use clang-profile-make
 
-build: config-sanity
+build: config-sanity net
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) all
 
 profile-build: net config-sanity objclean profileclean
@@ -731,12 +733,13 @@ install:
 	-cp $(EXE) $(BINDIR)
 	-strip $(BINDIR)/$(EXE)
 
-#clean all
+# clean all
 clean: objclean profileclean
 	@rm -f .depend *~ core
 
+# evaluation network (nnue)
 net:
-	$(eval nnuenet := $(shell grep EvalFile ucioption.cpp | grep Option | sed 's/.*\(nn-[a-z0-9]\{12\}.nnue\).*/\1/'))
+	$(eval nnuenet := $(shell grep EvalFileDefaultName evaluate.h | grep define | sed 's/.*\(nn-[a-z0-9]\{12\}.nnue\).*/\1/'))
 	@echo "Default net: $(nnuenet)"
 	$(eval nnuedownloadurl := https://tests.stockfishchess.org/api/nn/$(nnuenet))
 	$(eval curl_or_wget := $(shell if hash curl 2>/dev/null; then echo "curl -skL"; elif hash wget 2>/dev/null; then echo "wget -qO-"; fi))
@@ -758,7 +761,6 @@ net:
             echo "shasum / sha256sum not found, skipping net validation"; \
         fi
 
-
 # clean binaries and objects
 objclean:
 	@rm -f $(EXE) *.o ./syzygy/*.o ./nnue/*.o ./nnue/features/*.o
diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index ce92db9a..67154751 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -20,51 +20,126 @@
 #include <cassert>
 #include <cstdlib>
 #include <cstring>   // For std::memset
+#include <fstream>
 #include <iomanip>
 #include <sstream>
 #include <iostream>
+#include <streambuf>
+#include <vector>
 
 #include "bitboard.h"
 #include "evaluate.h"
 #include "material.h"
+#include "misc.h"
 #include "pawns.h"
 #include "thread.h"
 #include "uci.h"
+#include "incbin/incbin.h"
+
+
+// Macro to embed the default NNUE file data in the engine binary (using incbin.h, by Dale Weiler).
+// This macro invocation will declare the following three variables
+//     const unsigned char        gEmbeddedNNUEData[];  // a pointer to the embedded data
+//     const unsigned char *const gEmbeddedNNUEEnd;     // a marker to the end
+//     const unsigned int         gEmbeddedNNUESize;    // the size of the embedded file
+// Note that this does not work in Microsof Visual Studio.
+#if !defined(_MSC_VER) && !defined(NNUE_EMBEDDING_OFF)
+  INCBIN(EmbeddedNNUE, EvalFileDefaultName);
+#else
+  const unsigned char        gEmbeddedNNUEData[1] = {0x0};
+  const unsigned char *const gEmbeddedNNUEEnd = &gEmbeddedNNUEData[1];
+  const unsigned int         gEmbeddedNNUESize = 1;
+#endif
+
+
+using namespace std;
+using namespace Eval::NNUE;
 
 namespace Eval {
 
   bool useNNUE;
-  std::string eval_file_loaded="None";
+  string eval_file_loaded = "None";
+
+  /// init_NNUE() tries to load a nnue network at startup time, or when the engine
+  /// receives a UCI command "setoption name EvalFile value nn-[a-z0-9]{12}.nnue"
+  /// The name of the nnue network is always retrieved from the EvalFile option.
+  /// We search the given network in three locations: internally (the default
+  /// network may be embedded in the binary), in the active working directory and
+  /// in the engine directory. Distro packagers may define the DEFAULT_NNUE_DIRECTORY
+  /// variable to have the engine search in a special directory in their distro.
 
   void init_NNUE() {
 
     useNNUE = Options["Use NNUE"];
-    std::string eval_file = std::string(Options["EvalFile"]);
-    if (useNNUE && eval_file_loaded != eval_file)
-        if (Eval::NNUE::load_eval_file(eval_file))
-            eval_file_loaded = eval_file;
+    if (!useNNUE)
+        return;
+
+    string eval_file = string(Options["EvalFile"]);
+
+    #if defined(DEFAULT_NNUE_DIRECTORY)
+    #define stringify2(x) #x
+    #define stringify(x) stringify2(x)
+    vector<string> dirs = { "<internal>" , "" , CommandLine::binaryDirectory , stringify(DEFAULT_NNUE_DIRECTORY) };
+    #else
+    vector<string> dirs = { "<internal>" , "" , CommandLine::binaryDirectory };
+    #endif
+
+    for (string directory : dirs)
+        if (eval_file_loaded != eval_file)
+        {
+            if (directory != "<internal>")
+            {
+                ifstream stream(directory + eval_file, ios::binary);
+                if (load_eval(eval_file, stream))
+                    eval_file_loaded = eval_file;
+            }
+
+            if (directory == "<internal>" && eval_file == EvalFileDefaultName)
+            {
+                // C++ way to prepare a buffer for a memory stream
+                class MemoryBuffer : public basic_streambuf<char> {
+                    public: MemoryBuffer(char* p, size_t n) { setg(p, p, p + n); setp(p, p + n); }
+                };
+
+                MemoryBuffer buffer(const_cast<char*>(reinterpret_cast<const char*>(gEmbeddedNNUEData)),
+                                    size_t(gEmbeddedNNUESize));
+
+                istream stream(&buffer);
+                if (load_eval(eval_file, stream))
+                    eval_file_loaded = eval_file;
+            }
+        }
   }
 
+  /// verify_NNUE() verifies that the last net used was loaded successfully
   void verify_NNUE() {
 
-    std::string eval_file = std::string(Options["EvalFile"]);
+    string eval_file = string(Options["EvalFile"]);
+
     if (useNNUE && eval_file_loaded != eval_file)
     {
         UCI::OptionsMap defaults;
         UCI::init(defaults);
 
-        sync_cout << "info string ERROR: NNUE evaluation used, but the network file " << eval_file << " was not loaded successfully." << sync_endl;
-        sync_cout << "info string ERROR: The UCI option EvalFile might need to specify the full path, including the directory/folder name, to the file." << sync_endl;
-        sync_cout << "info string ERROR: The default net can be downloaded from: https://tests.stockfishchess.org/api/nn/"+std::string(defaults["EvalFile"]) << sync_endl;
-        sync_cout << "info string ERROR: If the UCI option Use NNUE is set to true, network evaluation parameters compatible with the program must be available." << sync_endl;
-        sync_cout << "info string ERROR: The engine will be terminated now." << sync_endl;
-        std::exit(EXIT_FAILURE);
+        string msg1 = "If the UCI option \"Use NNUE\" is set to true, network evaluation parameters compatible with the engine must be available.";
+        string msg2 = "The option is set to true, but the network file " + eval_file + " was not loaded successfully.";
+        string msg3 = "The UCI option EvalFile might need to specify the full path, including the directory name, to the network file.";
+        string msg4 = "The default net can be downloaded from: https://tests.stockfishchess.org/api/nn/" + string(defaults["EvalFile"]);
+        string msg5 = "The engine will be terminated now.";
+
+        sync_cout << "info string ERROR: " << msg1 << sync_endl;
+        sync_cout << "info string ERROR: " << msg2 << sync_endl;
+        sync_cout << "info string ERROR: " << msg3 << sync_endl;
+        sync_cout << "info string ERROR: " << msg4 << sync_endl;
+        sync_cout << "info string ERROR: " << msg5 << sync_endl;
+
+        exit(EXIT_FAILURE);
     }
 
     if (useNNUE)
-        sync_cout << "info string NNUE evaluation using " << eval_file << " enabled." << sync_endl;
+        sync_cout << "info string NNUE evaluation using " << eval_file << " enabled" << sync_endl;
     else
-        sync_cout << "info string classical evaluation enabled." << sync_endl;
+        sync_cout << "info string classical evaluation enabled" << sync_endl;
   }
 }
 
diff --git a/src/evaluate.h b/src/evaluate.h
index e808068d..d701f5a7 100644
--- a/src/evaluate.h
+++ b/src/evaluate.h
@@ -35,12 +35,17 @@ namespace Eval {
   void init_NNUE();
   void verify_NNUE();
 
+  // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
+  // for the build process (profile-build and fishtest) to work. Do not change the
+  // name of the macro, as it is used in the Makefile.
+  #define EvalFileDefaultName   "nn-82215d0fd0df.nnue"
+
   namespace NNUE {
 
     Value evaluate(const Position& pos);
     Value compute_eval(const Position& pos);
     void  update_eval(const Position& pos);
-    bool  load_eval_file(const std::string& evalFile);
+    bool  load_eval(std::string streamName, std::istream& stream);
 
   } // namespace NNUE
 
diff --git a/src/incbin/UNLICENCE b/src/incbin/UNLICENCE
new file mode 100644
index 00000000..32484ab5
--- /dev/null
+++ b/src/incbin/UNLICENCE
@@ -0,0 +1,26 @@
+The file "incbin.h" is free and unencumbered software released into
+the public domain by Dale Weiler, see:
+   <https://github.com/graphitemaster/incbin>
+
+Anyone is free to copy, modify, publish, use, compile, sell, or
+distribute this software, either in source code form or as a compiled
+binary, for any purpose, commercial or non-commercial, and by any
+means.
+
+In jurisdictions that recognize copyright laws, the author or authors
+of this software dedicate any and all copyright interest in the
+software to the public domain. We make this dedication for the benefit
+of the public at large and to the detriment of our heirs and
+successors. We intend this dedication to be an overt act of
+relinquishment in perpetuity of all present and future rights to this
+software under copyright law.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
+
+For more information, please refer to <http://unlicense.org/>
diff --git a/src/incbin/incbin.h b/src/incbin/incbin.h
new file mode 100755
index 00000000..c19684d7
--- /dev/null
+++ b/src/incbin/incbin.h
@@ -0,0 +1,368 @@
+/**
+ * @file incbin.h
+ * @author Dale Weiler
+ * @brief Utility for including binary files
+ *
+ * Facilities for including binary files into the current translation unit and
+ * making use from them externally in other translation units.
+ */
+#ifndef INCBIN_HDR
+#define INCBIN_HDR
+#include <limits.h>
+#if   defined(__AVX512BW__) || \
+      defined(__AVX512CD__) || \
+      defined(__AVX512DQ__) || \
+      defined(__AVX512ER__) || \
+      defined(__AVX512PF__) || \
+      defined(__AVX512VL__) || \
+      defined(__AVX512F__)
+# define INCBIN_ALIGNMENT_INDEX 6
+#elif defined(__AVX__)      || \
+      defined(__AVX2__)
+# define INCBIN_ALIGNMENT_INDEX 5
+#elif defined(__SSE__)      || \
+      defined(__SSE2__)     || \
+      defined(__SSE3__)     || \
+      defined(__SSSE3__)    || \
+      defined(__SSE4_1__)   || \
+      defined(__SSE4_2__)   || \
+      defined(__neon__)
+# define INCBIN_ALIGNMENT_INDEX 4
+#elif ULONG_MAX != 0xffffffffu
+# define INCBIN_ALIGNMENT_INDEX 3
+# else
+# define INCBIN_ALIGNMENT_INDEX 2
+#endif
+
+/* Lookup table of (1 << n) where `n' is `INCBIN_ALIGNMENT_INDEX' */
+#define INCBIN_ALIGN_SHIFT_0 1
+#define INCBIN_ALIGN_SHIFT_1 2
+#define INCBIN_ALIGN_SHIFT_2 4
+#define INCBIN_ALIGN_SHIFT_3 8
+#define INCBIN_ALIGN_SHIFT_4 16
+#define INCBIN_ALIGN_SHIFT_5 32
+#define INCBIN_ALIGN_SHIFT_6 64
+
+/* Actual alignment value */
+#define INCBIN_ALIGNMENT \
+    INCBIN_CONCATENATE( \
+        INCBIN_CONCATENATE(INCBIN_ALIGN_SHIFT, _), \
+        INCBIN_ALIGNMENT_INDEX)
+
+/* Stringize */
+#define INCBIN_STR(X) \
+    #X
+#define INCBIN_STRINGIZE(X) \
+    INCBIN_STR(X)
+/* Concatenate */
+#define INCBIN_CAT(X, Y) \
+    X ## Y
+#define INCBIN_CONCATENATE(X, Y) \
+    INCBIN_CAT(X, Y)
+/* Deferred macro expansion */
+#define INCBIN_EVAL(X) \
+    X
+#define INCBIN_INVOKE(N, ...) \
+    INCBIN_EVAL(N(__VA_ARGS__))
+
+/* Green Hills uses a different directive for including binary data */
+#if defined(__ghs__)
+#  if (__ghs_asm == 2)
+#    define INCBIN_MACRO ".file"
+/* Or consider the ".myrawdata" entry in the ld file */
+#  else
+#    define INCBIN_MACRO "\tINCBIN"
+#  endif
+#else
+#  define INCBIN_MACRO ".incbin"
+#endif
+
+#ifndef _MSC_VER
+#  define INCBIN_ALIGN \
+    __attribute__((aligned(INCBIN_ALIGNMENT)))
+#else
+#  define INCBIN_ALIGN __declspec(align(INCBIN_ALIGNMENT))
+#endif
+
+#if defined(__arm__) || /* GNU C and RealView */ \
+    defined(__arm) || /* Diab */ \
+    defined(_ARM) /* ImageCraft */
+#  define INCBIN_ARM
+#endif
+
+#ifdef __GNUC__
+/* Utilize .balign where supported */
+#  define INCBIN_ALIGN_HOST ".balign " INCBIN_STRINGIZE(INCBIN_ALIGNMENT) "\n"
+#  define INCBIN_ALIGN_BYTE ".balign 1\n"
+#elif defined(INCBIN_ARM)
+/*
+ * On arm assemblers, the alignment value is calculated as (1 << n) where `n' is
+ * the shift count. This is the value passed to `.align'
+ */
+#  define INCBIN_ALIGN_HOST ".align " INCBIN_STRINGIZE(INCBIN_ALIGNMENT_INDEX) "\n"
+#  define INCBIN_ALIGN_BYTE ".align 0\n"
+#else
+/* We assume other inline assembler's treat `.align' as `.balign' */
+#  define INCBIN_ALIGN_HOST ".align " INCBIN_STRINGIZE(INCBIN_ALIGNMENT) "\n"
+#  define INCBIN_ALIGN_BYTE ".align 1\n"
+#endif
+
+/* INCBIN_CONST is used by incbin.c generated files */
+#if defined(__cplusplus)
+#  define INCBIN_EXTERNAL extern "C"
+#  define INCBIN_CONST    extern const
+#else
+#  define INCBIN_EXTERNAL extern
+#  define INCBIN_CONST    const
+#endif
+
+/**
+ * @brief Optionally override the linker section into which data is emitted.
+ *
+ * @warning If you use this facility, you'll have to deal with platform-specific linker output
+ * section naming on your own
+ *
+ * Overriding the default linker output section, e.g for esp8266/Arduino:
+ * @code
+ * #define INCBIN_OUTPUT_SECTION ".irom.text"
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ * // Data is emitted into program memory that never gets copied to RAM
+ * @endcode
+ */
+#if !defined(INCBIN_OUTPUT_SECTION)
+#  if defined(__APPLE__)
+#    define INCBIN_OUTPUT_SECTION         ".const_data"
+#  else
+#    define INCBIN_OUTPUT_SECTION         ".rodata"
+#  endif
+#endif
+
+#if defined(__APPLE__)
+/* The directives are different for Apple branded compilers */
+#  define INCBIN_SECTION         INCBIN_OUTPUT_SECTION "\n"
+#  define INCBIN_GLOBAL(NAME)    ".globl " INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME "\n"
+#  define INCBIN_INT             ".long "
+#  define INCBIN_MANGLE          "_"
+#  define INCBIN_BYTE            ".byte "
+#  define INCBIN_TYPE(...)
+#else
+#  define INCBIN_SECTION         ".section " INCBIN_OUTPUT_SECTION "\n"
+#  define INCBIN_GLOBAL(NAME)    ".global " INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME "\n"
+#  if defined(__ghs__)
+#    define INCBIN_INT           ".word "
+#  else
+#    define INCBIN_INT           ".int "
+#  endif
+#  if defined(__USER_LABEL_PREFIX__)
+#    define INCBIN_MANGLE        INCBIN_STRINGIZE(__USER_LABEL_PREFIX__)
+#  else
+#    define INCBIN_MANGLE        ""
+#  endif
+#  if defined(INCBIN_ARM)
+/* On arm assemblers, `@' is used as a line comment token */
+#    define INCBIN_TYPE(NAME)    ".type " INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME ", %object\n"
+#  elif defined(__MINGW32__) || defined(__MINGW64__)
+/* Mingw doesn't support this directive either */
+#    define INCBIN_TYPE(NAME)
+#  else
+/* It's safe to use `@' on other architectures */
+#    define INCBIN_TYPE(NAME)    ".type " INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME ", @object\n"
+#  endif
+#  define INCBIN_BYTE            ".byte "
+#endif
+
+/* List of style types used for symbol names */
+#define INCBIN_STYLE_CAMEL 0
+#define INCBIN_STYLE_SNAKE 1
+
+/**
+ * @brief Specify the prefix to use for symbol names.
+ *
+ * By default this is `g', producing symbols of the form:
+ * @code
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char gFooData[];
+ * // const unsigned char *const gFooEnd;
+ * // const unsigned int gFooSize;
+ * @endcode
+ *
+ * If however you specify a prefix before including: e.g:
+ * @code
+ * #define INCBIN_PREFIX incbin
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ *
+ * // Now you have the following symbols instead:
+ * // const unsigned char incbinFooData[];
+ * // const unsigned char *const incbinFooEnd;
+ * // const unsigned int incbinFooSize;
+ * @endcode
+ */
+#if !defined(INCBIN_PREFIX)
+#  define INCBIN_PREFIX g
+#endif
+
+/**
+ * @brief Specify the style used for symbol names.
+ *
+ * Possible options are
+ * - INCBIN_STYLE_CAMEL "CamelCase"
+ * - INCBIN_STYLE_SNAKE "snake_case"
+ *
+ * Default option is *INCBIN_STYLE_CAMEL* producing symbols of the form:
+ * @code
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char <prefix>FooData[];
+ * // const unsigned char *const <prefix>FooEnd;
+ * // const unsigned int <prefix>FooSize;
+ * @endcode
+ *
+ * If however you specify a style before including: e.g:
+ * @code
+ * #define INCBIN_STYLE INCBIN_STYLE_SNAKE
+ * #include "incbin.h"
+ * INCBIN(foo, "foo.txt");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char <prefix>foo_data[];
+ * // const unsigned char *const <prefix>foo_end;
+ * // const unsigned int <prefix>foo_size;
+ * @endcode
+ */
+#if !defined(INCBIN_STYLE)
+#  define INCBIN_STYLE INCBIN_STYLE_CAMEL
+#endif
+
+/* Style lookup tables */
+#define INCBIN_STYLE_0_DATA Data
+#define INCBIN_STYLE_0_END End
+#define INCBIN_STYLE_0_SIZE Size
+#define INCBIN_STYLE_1_DATA _data
+#define INCBIN_STYLE_1_END _end
+#define INCBIN_STYLE_1_SIZE _size
+
+/* Style lookup: returning identifier */
+#define INCBIN_STYLE_IDENT(TYPE) \
+    INCBIN_CONCATENATE( \
+        INCBIN_STYLE_, \
+        INCBIN_CONCATENATE( \
+            INCBIN_EVAL(INCBIN_STYLE), \
+            INCBIN_CONCATENATE(_, TYPE)))
+
+/* Style lookup: returning string literal */
+#define INCBIN_STYLE_STRING(TYPE) \
+    INCBIN_STRINGIZE( \
+        INCBIN_STYLE_IDENT(TYPE)) \
+
+/* Generate the global labels by indirectly invoking the macro with our style
+ * type and concatenating the name against them. */
+#define INCBIN_GLOBAL_LABELS(NAME, TYPE) \
+    INCBIN_INVOKE( \
+        INCBIN_GLOBAL, \
+        INCBIN_CONCATENATE( \
+            NAME, \
+            INCBIN_INVOKE( \
+                INCBIN_STYLE_IDENT, \
+                TYPE))) \
+    INCBIN_INVOKE( \
+        INCBIN_TYPE, \
+        INCBIN_CONCATENATE( \
+            NAME, \
+            INCBIN_INVOKE( \
+                INCBIN_STYLE_IDENT, \
+                TYPE)))
+
+/**
+ * @brief Externally reference binary data included in another translation unit.
+ *
+ * Produces three external symbols that reference the binary data included in
+ * another translation unit.
+ *
+ * The symbol names are a concatenation of `INCBIN_PREFIX' before *NAME*; with
+ * "Data", as well as "End" and "Size" after. An example is provided below.
+ *
+ * @param NAME The name given for the binary data
+ *
+ * @code
+ * INCBIN_EXTERN(Foo);
+ *
+ * // Now you have the following symbols:
+ * // extern const unsigned char <prefix>FooData[];
+ * // extern const unsigned char *const <prefix>FooEnd;
+ * // extern const unsigned int <prefix>FooSize;
+ * @endcode
+ */
+#define INCBIN_EXTERN(NAME) \
+    INCBIN_EXTERNAL const INCBIN_ALIGN unsigned char \
+        INCBIN_CONCATENATE( \
+            INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
+            INCBIN_STYLE_IDENT(DATA))[]; \
+    INCBIN_EXTERNAL const INCBIN_ALIGN unsigned char *const \
+    INCBIN_CONCATENATE( \
+        INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
+        INCBIN_STYLE_IDENT(END)); \
+    INCBIN_EXTERNAL const unsigned int \
+        INCBIN_CONCATENATE( \
+            INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
+            INCBIN_STYLE_IDENT(SIZE))
+
+/**
+ * @brief Include a binary file into the current translation unit.
+ *
+ * Includes a binary file into the current translation unit, producing three symbols
+ * for objects that encode the data and size respectively.
+ *
+ * The symbol names are a concatenation of `INCBIN_PREFIX' before *NAME*; with
+ * "Data", as well as "End" and "Size" after. An example is provided below.
+ *
+ * @param NAME The name to associate with this binary data (as an identifier.)
+ * @param FILENAME The file to include (as a string literal.)
+ *
+ * @code
+ * INCBIN(Icon, "icon.png");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char <prefix>IconData[];
+ * // const unsigned char *const <prefix>IconEnd;
+ * // const unsigned int <prefix>IconSize;
+ * @endcode
+ *
+ * @warning This must be used in global scope
+ * @warning The identifiers may be different if INCBIN_STYLE is not default
+ *
+ * To externally reference the data included by this in another translation unit
+ * please @see INCBIN_EXTERN.
+ */
+#ifdef _MSC_VER
+#define INCBIN(NAME, FILENAME) \
+    INCBIN_EXTERN(NAME)
+#else
+#define INCBIN(NAME, FILENAME) \
+    __asm__(INCBIN_SECTION \
+            INCBIN_GLOBAL_LABELS(NAME, DATA) \
+            INCBIN_ALIGN_HOST \
+            INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(DATA) ":\n" \
+            INCBIN_MACRO " \"" FILENAME "\"\n" \
+            INCBIN_GLOBAL_LABELS(NAME, END) \
+            INCBIN_ALIGN_BYTE \
+            INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(END) ":\n" \
+                INCBIN_BYTE "1\n" \
+            INCBIN_GLOBAL_LABELS(NAME, SIZE) \
+            INCBIN_ALIGN_HOST \
+            INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(SIZE) ":\n" \
+                INCBIN_INT INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(END) " - " \
+                           INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(DATA) "\n" \
+            INCBIN_ALIGN_HOST \
+            ".text\n" \
+    ); \
+    INCBIN_EXTERN(NAME)
+
+#endif
+#endif
diff --git a/src/main.cpp b/src/main.cpp
index fbad6622..f95db1c2 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -35,6 +35,7 @@ int main(int argc, char* argv[]) {
 
   std::cout << engine_info() << std::endl;
 
+  CommandLine::init(argc, argv);
   UCI::init(Options);
   Tune::init();
   PSQT::init();
diff --git a/src/misc.cpp b/src/misc.cpp
index 80c436ac..3fbdea35 100644
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -132,6 +132,7 @@ public:
 
 } // namespace
 
+
 /// engine_info() returns the full name of the current Stockfish version. This
 /// will be either "Stockfish <Tag> DD-MM-YY" (where DD-MM-YY is the date when
 /// the program was compiled) or "Stockfish <Version>", depending on whether
@@ -589,3 +590,61 @@ void bindThisThread(size_t idx) {
 #endif
 
 } // namespace WinProcGroup
+
+#ifdef _WIN32
+#include <direct.h>
+#define GETCWD _getcwd
+#else
+#include <unistd.h>
+#define GETCWD getcwd
+#endif
+
+namespace CommandLine {
+
+string argv0;            // path+name of the executable binary, as given by argv[0]
+string binaryDirectory;  // path of the executable directory
+string workingDirectory; // path of the working directory
+string pathSeparator;    // Separator for our current OS
+
+void init(int argc, char* argv[]) {
+    (void)argc;
+    string separator;
+
+    // extract the path+name of the executable binary
+    argv0 = argv[0];
+
+#ifdef _WIN32
+    pathSeparator = "\\";
+  #ifdef _MSC_VER
+    // Under windows argv[0] may not have the extension. Also _get_pgmptr() had
+    // issues in some windows 10 versions, so check returned values carefully.
+    char* pgmptr = nullptr;
+    if (!_get_pgmptr(&pgmptr) && pgmptr != nullptr && *pgmptr)
+        argv0 = pgmptr;
+  #endif
+#else
+    pathSeparator = "/";
+#endif
+
+    // extract the working directory
+    workingDirectory = "";
+    char buff[40000];
+    char* cwd = GETCWD(buff, 40000);
+    if (cwd)
+        workingDirectory = cwd;
+
+    // extract the binary directory path from argv0
+    binaryDirectory = argv0;
+    size_t pos = binaryDirectory.find_last_of("\\/");
+    if (pos == std::string::npos)
+        binaryDirectory = "." + pathSeparator;
+    else
+        binaryDirectory.resize(pos + 1);
+
+    // pattern replacement: "./" at the start of path is replaced by the working directory
+    if (binaryDirectory.find("." + pathSeparator) == 0)
+        binaryDirectory.replace(0, 1, workingDirectory);
+}
+
+
+} // namespace CommandLine
diff --git a/src/misc.h b/src/misc.h
index 8ad17b50..68b9c884 100644
--- a/src/misc.h
+++ b/src/misc.h
@@ -42,9 +42,7 @@ void dbg_mean_of(int v);
 void dbg_print();
 
 typedef std::chrono::milliseconds::rep TimePoint; // A value in milliseconds
-
 static_assert(sizeof(TimePoint) == sizeof(int64_t), "TimePoint should be 64 bits");
-
 inline TimePoint now() {
   return std::chrono::duration_cast<std::chrono::milliseconds>
         (std::chrono::steady_clock::now().time_since_epoch()).count();
@@ -126,4 +124,11 @@ namespace WinProcGroup {
   void bindThisThread(size_t idx);
 }
 
+namespace CommandLine {
+  void init(int argc, char* argv[]);
+
+  extern std::string binaryDirectory;  // path of the executable directory
+  extern std::string workingDirectory; // path of the working directory
+}
+
 #endif // #ifndef MISC_H_INCLUDED
diff --git a/src/nnue/evaluate_nnue.cpp b/src/nnue/evaluate_nnue.cpp
index e6619089..d6ac9894 100644
--- a/src/nnue/evaluate_nnue.cpp
+++ b/src/nnue/evaluate_nnue.cpp
@@ -18,7 +18,6 @@
 
 // Code for calculating NNUE evaluation function
 
-#include <fstream>
 #include <iostream>
 #include <set>
 
@@ -143,17 +142,12 @@ namespace Eval::NNUE {
     return accumulator.score;
   }
 
-  // Load the evaluation function file
-  bool load_eval_file(const std::string& evalFile) {
+  // Load eval, from a file stream or a memory stream
+  bool load_eval(std::string streamName, std::istream& stream) {
 
     Initialize();
-    fileName = evalFile;
-
-    std::ifstream stream(evalFile, std::ios::binary);
-
-    const bool result = ReadParameters(stream);
-
-    return result;
+    fileName = streamName;
+    return ReadParameters(stream);
   }
 
   // Evaluation function. Perform differential calculation.
diff --git a/src/ucioption.cpp b/src/ucioption.cpp
index ec83c7c8..5e747a7f 100644
--- a/src/ucioption.cpp
+++ b/src/ucioption.cpp
@@ -21,6 +21,7 @@
 #include <ostream>
 #include <sstream>
 
+#include "evaluate.h"
 #include "misc.h"
 #include "search.h"
 #include "thread.h"
@@ -79,9 +80,7 @@ void init(OptionsMap& o) {
   o["Syzygy50MoveRule"]      << Option(true);
   o["SyzygyProbeLimit"]      << Option(7, 0, 7);
   o["Use NNUE"]              << Option(true, on_use_NNUE);
-  // The default must follow the format nn-[SHA256 first 12 digits].nnue
-  // for the build process (profile-build and fishtest) to work.
-  o["EvalFile"]              << Option("nn-82215d0fd0df.nnue", on_eval_file);
+  o["EvalFile"]              << Option(EvalFileDefaultName, on_eval_file);
 }
 
 

From e4ed7d3dd7b8895ce523180cb3da3ec2714050fc Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Wed, 26 Aug 2020 18:00:54 +0200
Subject: [PATCH 23/52] Cleaner make help

do not print details if ARCH is an empty string. Follow up for b0b4ca17db49ed03057b5fa4ee4a12dab0e9c9e6

https://github.com/official-stockfish/Stockfish/pull/3071

No functional change
---
 src/Makefile | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

diff --git a/src/Makefile b/src/Makefile
index 5f363f02..9ae5a51c 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -85,12 +85,17 @@ endif
 
 ### 2.1. General and architecture defaults
 
+ifeq ($(ARCH),)
+   ARCH = x86-64-modern
+   help_skip_sanity = yes
+endif
 # explicitly check for the list of supported architectures (as listed with make help),
 # the user can override with `make ARCH=x86-32-vnni256 SUPPORTED_ARCH=true`
-ifeq ($(ARCH),$(filter $(ARCH),x86-64-vnni512 x86-64-vnni256 x86-64-avx512 x86-64-bmi2 x86-64-avx2 \
-                               x86-64-sse41-popcnt x86-64-modern x86-64-ssse3 x86-64-sse3-popcnt \
-                               x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-32 \
-                               armv7 armv7-neon armv8 apple-silicon general-64 general-32))
+ifeq ($(ARCH), $(filter $(ARCH), \
+                 x86-64-vnni512 x86-64-vnni256 x86-64-avx512 x86-64-bmi2 x86-64-avx2 \
+                 x86-64-sse41-popcnt x86-64-modern x86-64-ssse3 x86-64-sse3-popcnt \
+                 x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-32 \
+                 armv7 armv7-neon armv8 apple-silicon general-64 general-32))
    SUPPORTED_ARCH=true
 else
    SUPPORTED_ARCH=false
@@ -113,7 +118,6 @@ avx512 = no
 vnni256 = no
 vnni512 = no
 neon = no
-ARCH = x86-64-modern
 STRIP = strip
 
 ### 2.2 Architecture specific
@@ -695,11 +699,12 @@ help:
 	@echo "make -j build ARCH=x86-64-ssse3 COMP=clang"
 	@echo ""
 	@echo "-------------------------------"
-ifeq ($(SUPPORTED_ARCH), true)
+ifeq ($(SUPPORTED_ARCH)$(help_skip_sanity), true)
 	@echo "The selected architecture $(ARCH) will enable the following configuration: "
 	@$(MAKE) ARCH=$(ARCH) COMP=$(COMP) config-sanity
 else
 	@echo "Specify a supported architecture with the ARCH option for more details"
+	@echo ""
 endif
 
 

From d90d893b5eeea020b2d59d1372f5aa0a20b45412 Mon Sep 17 00:00:00 2001
From: Unai Corzo <corzounai@gmail.com>
Date: Fri, 28 Aug 2020 09:27:15 +0200
Subject: [PATCH 24/52] Reintroduce depth reduction

Reintroduce depth reduction if the position is not in TT.

STC https://tests.stockfishchess.org/tests/view/5f4652e85089a564a10d868c
LLR: 2.97 (-2.94,2.94) {-0.25,1.25}
Total: 40240 W: 4535 L: 4331 D: 31374
Ptnml(0-2): 215, 3276, 12969, 3410, 250

LTC https://tests.stockfishchess.org/tests/view/5f46ca5e5089a564a10d86f3
LLR: 2.93 (-2.94,2.94) {0.25,1.25}
Total: 63096 W: 3426 L: 3188 D: 56482
Ptnml(0-2): 51, 2798, 25645, 2970, 84

closes https://github.com/official-stockfish/Stockfish/pull/3072

bench: 3611906
---
 src/search.cpp | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index cae8a684..77447043 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -939,6 +939,12 @@ namespace {
                 }
             }
     }
+    
+    // Step 11. If the position is not in TT, decrease depth by 2
+    if (   PvNode
+        && depth >= 6
+        && !ttMove)
+        depth -= 2;
 
 moves_loop: // When in check, search starts from here
 
@@ -963,7 +969,7 @@ moves_loop: // When in check, search starts from here
     // Mark this node as being searched
     ThreadHolding th(thisThread, posKey, ss->ply);
 
-    // Step 11. Loop through all pseudo-legal moves until no moves remain
+    // Step 12. Loop through all pseudo-legal moves until no moves remain
     // or a beta cutoff occurs.
     while ((move = mp.next_move(moveCountPruning)) != MOVE_NONE)
     {
@@ -1001,7 +1007,7 @@ moves_loop: // When in check, search starts from here
       // Calculate new depth for this move
       newDepth = depth - 1;
 
-      // Step 12. Pruning at shallow depth (~200 Elo)
+      // Step 13. Pruning at shallow depth (~200 Elo)
       if (  !rootNode
           && pos.non_pawn_material(us)
           && bestValue > VALUE_TB_LOSS_IN_MAX_PLY)
@@ -1059,7 +1065,7 @@ moves_loop: // When in check, search starts from here
           }
       }
 
-      // Step 13. Extensions (~75 Elo)
+      // Step 14. Extensions (~75 Elo)
 
       // Singular extension search (~70 Elo). If all moves but one fail low on a
       // search of (alpha-s, beta-s), and just one fails high on (alpha, beta),
@@ -1142,10 +1148,10 @@ moves_loop: // When in check, search starts from here
                                                                 [movedPiece]
                                                                 [to_sq(move)];
 
-      // Step 14. Make the move
+      // Step 15. Make the move
       pos.do_move(move, st, givesCheck);
 
-      // Step 15. Reduced depth search (LMR, ~200 Elo). If the move fails high it will be
+      // Step 16. Reduced depth search (LMR, ~200 Elo). If the move fails high it will be
       // re-searched at full depth.
       if (    depth >= 3
           &&  moveCount > 1 + 2 * rootNode + 2 * (PvNode && abs(bestValue) < 2)
@@ -1248,7 +1254,7 @@ moves_loop: // When in check, search starts from here
           didLMR = false;
       }
 
-      // Step 16. Full depth search when LMR is skipped or fails high
+      // Step 17. Full depth search when LMR is skipped or fails high
       if (doFullDepthSearch)
       {
           value = -search<NonPV>(pos, ss+1, -(alpha+1), -alpha, newDepth, !cutNode);
@@ -1276,12 +1282,12 @@ moves_loop: // When in check, search starts from here
           value = -search<PV>(pos, ss+1, -beta, -alpha, newDepth, false);
       }
 
-      // Step 17. Undo move
+      // Step 18. Undo move
       pos.undo_move(move);
 
       assert(value > -VALUE_INFINITE && value < VALUE_INFINITE);
 
-      // Step 18. Check for a new best move
+      // Step 19. Check for a new best move
       // Finished searching the move. If a stop occurred, the return value of
       // the search cannot be trusted, and we return immediately without
       // updating best move, PV and TT.
@@ -1358,7 +1364,7 @@ moves_loop: // When in check, search starts from here
         return VALUE_DRAW;
     */
 
-    // Step 19. Check for mate and stalemate
+    // Step 20. Check for mate and stalemate
     // All legal moves have been searched and if there are no legal moves, it
     // must be a mate or a stalemate. If we are in a singular extension search then
     // return a fail low score.

From c02b3a4c7a339d212d5c6f75b3b89c926d33a800 Mon Sep 17 00:00:00 2001
From: MJZ1977 <37274752+MJZ1977@users.noreply.github.com>
Date: Fri, 28 Aug 2020 12:06:36 +0200
Subject: [PATCH 25/52] Add / remove leaves from search tree ttPv

add if previous leaf is in search tree and we didn't find a counter move
else remove the position if the leaf is the last one in search tree.

STC : https://tests.stockfishchess.org/tests/view/5f49203c3def640786115314
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 29968 W: 3381 L: 3195 D: 23392
Ptnml(0-2): 146, 2432, 9671, 2560, 175

LTC : https://tests.stockfishchess.org/tests/view/5f494bea3def640786115336
LLR: 2.96 (-2.94,2.94) {0.25,1.25}
Total: 84952 W: 4619 L: 4333 D: 76000
Ptnml(0-2): 86, 3765, 34481, 4065, 79

closes https://github.com/official-stockfish/Stockfish/pull/3075

Bench 3527337
---
 src/search.cpp | 34 ++++++++++++++++++++++++----------
 src/search.h   |  1 +
 2 files changed, 25 insertions(+), 10 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index 77447043..a2342a3c 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -597,7 +597,7 @@ namespace {
     Move ttMove, move, excludedMove, bestMove;
     Depth extension, newDepth;
     Value bestValue, value, ttValue, eval, maxValue, probCutBeta;
-    bool ttHit, ttPv, formerPv, givesCheck, improving, didLMR, priorCapture;
+    bool ttHit, formerPv, givesCheck, improving, didLMR, priorCapture;
     bool captureOrPromotion, doFullDepthSearch, moveCountPruning,
          ttCapture, singularQuietLMR;
     Piece movedPiece;
@@ -644,6 +644,7 @@ namespace {
     assert(0 <= ss->ply && ss->ply < MAX_PLY);
 
     (ss+1)->ply = ss->ply + 1;
+    (ss+1)->ttPv = false;
     (ss+1)->excludedMove = bestMove = MOVE_NONE;
     (ss+2)->killers[0] = (ss+2)->killers[1] = MOVE_NONE;
     Square prevSq = to_sq((ss-1)->currentMove);
@@ -667,10 +668,11 @@ namespace {
     ttValue = ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
     ttMove =  rootNode ? thisThread->rootMoves[thisThread->pvIdx].pv[0]
             : ttHit    ? tte->move() : MOVE_NONE;
-    ttPv = PvNode || (ttHit && tte->is_pv());
-    formerPv = ttPv && !PvNode;
+    if (!excludedMove)
+        ss->ttPv = PvNode || (ttHit && tte->is_pv());
+    formerPv = ss->ttPv && !PvNode;
 
-    if (   ttPv
+    if (   ss->ttPv
         && depth > 12
         && ss->ply - 1 < MAX_LPH
         && !priorCapture
@@ -748,7 +750,7 @@ namespace {
                 if (    b == BOUND_EXACT
                     || (b == BOUND_LOWER ? value >= beta : value <= alpha))
                 {
-                    tte->save(posKey, value_to_tt(value, ss->ply), ttPv, b,
+                    tte->save(posKey, value_to_tt(value, ss->ply), ss->ttPv, b,
                               std::min(MAX_PLY - 1, depth + 6),
                               MOVE_NONE, VALUE_NONE);
 
@@ -798,7 +800,7 @@ namespace {
         else
             ss->staticEval = eval = -(ss-1)->staticEval + 2 * Tempo;
 
-        tte->save(posKey, VALUE_NONE, ttPv, BOUND_NONE, DEPTH_NONE, MOVE_NONE, eval);
+        tte->save(posKey, VALUE_NONE, ss->ttPv, BOUND_NONE, DEPTH_NONE, MOVE_NONE, eval);
     }
 
     // Step 7. Razoring (~1 Elo)
@@ -824,7 +826,7 @@ namespace {
         && (ss-1)->statScore < 22977
         &&  eval >= beta
         &&  eval >= ss->staticEval
-        &&  ss->staticEval >= beta - 30 * depth - 28 * improving + 84 * ttPv + 182
+        &&  ss->staticEval >= beta - 30 * depth - 28 * improving + 84 * ss->ttPv + 182
         && !excludedMove
         &&  pos.non_pawn_material(us)
         && (ss->ply >= thisThread->nmpMinPly || us != thisThread->nmpColor))
@@ -898,6 +900,8 @@ namespace {
         assert(probCutBeta < VALUE_INFINITE);
         MovePicker mp(pos, ttMove, probCutBeta - ss->staticEval, &captureHistory);
         int probCutCount = 0;
+        bool ttPv = ss->ttPv;
+        ss->ttPv = false;
 
         while (   (move = mp.next_move()) != MOVE_NONE
                && probCutCount < 2 + 2 * cutNode)
@@ -938,6 +942,7 @@ namespace {
                     return value;
                 }
             }
+         ss->ttPv = ttPv;
     }
     
     // Step 11. If the position is not in TT, decrease depth by 2
@@ -1180,7 +1185,7 @@ moves_loop: // When in check, search starts from here
               r++;
 
           // Decrease reduction if position is or has been on the PV (~10 Elo)
-          if (ttPv)
+          if (ss->ttPv)
               r -= 2;
 
           if (moveCountPruning && !formerPv)
@@ -1209,7 +1214,7 @@ moves_loop: // When in check, search starts from here
               // hence break make_move(). (~2 Elo)
               else if (    type_of(move) == NORMAL
                        && !pos.see_ge(reverse_move(move)))
-                  r -= 2 + ttPv - (type_of(movedPiece) == PAWN);
+                  r -= 2 + ss->ttPv - (type_of(movedPiece) == PAWN);
 
               ss->statScore =  thisThread->mainHistory[us][from_to(move)]
                              + (*contHist[0])[movedPiece][to_sq(move)]
@@ -1387,8 +1392,17 @@ moves_loop: // When in check, search starts from here
     if (PvNode)
         bestValue = std::min(bestValue, maxValue);
 
+    // If no good move is found and the previous position was ttPv, then the previous
+    // opponent move is probably good and the new position is added to the search tree.
+    if (bestValue <= alpha)
+        ss->ttPv = ss->ttPv || ((ss-1)->ttPv && depth > 3);
+    // Otherwise, a counter move has been found and if the position is the last leaf
+    // in the search tree, remove the position from the search tree.
+    else if (depth > 3)
+        ss->ttPv = ss->ttPv && (ss+1)->ttPv;
+
     if (!excludedMove && !(rootNode && thisThread->pvIdx))
-        tte->save(posKey, value_to_tt(bestValue, ss->ply), ttPv,
+        tte->save(posKey, value_to_tt(bestValue, ss->ply), ss->ttPv,
                   bestValue >= beta ? BOUND_LOWER :
                   PvNode && bestMove ? BOUND_EXACT : BOUND_UPPER,
                   depth, bestMove, ss->staticEval);
diff --git a/src/search.h b/src/search.h
index 2554f3fb..79085189 100644
--- a/src/search.h
+++ b/src/search.h
@@ -48,6 +48,7 @@ struct Stack {
   int statScore;
   int moveCount;
   bool inCheck;
+  bool ttPv;
 };
 
 

From 9b5b9ec9a6a3a0e46ac00f58976887560948a7e2 Mon Sep 17 00:00:00 2001
From: VoyagerOne <excelgeek@gmail.com>
Date: Sat, 29 Aug 2020 21:13:05 -0400
Subject: [PATCH 26/52] QS Pruning Simplification

Remove depth dependence in QS pruning

STC:
LLR: 2.95 (-2.94,2.94) {-1.25,0.25}
Total: 40536 W: 4442 L: 4358 D: 31736
Ptnml(0-2): 209, 3330, 13118, 3390, 221
https://tests.stockfishchess.org/tests/view/5f49035b3def6407861152f9

LTC:
LLR: 2.95 (-2.94,2.94) {-0.75,0.25}
Total: 97104 W: 5164 L: 5130 D: 86810
Ptnml(0-2): 103, 4478, 39377, 4470, 124
https://tests.stockfishchess.org/tests/view/5f4939d53def640786115322

closes https://github.com/official-stockfish/Stockfish/pull/3077

Bench: 3865238
---
 src/search.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/search.cpp b/src/search.cpp
index a2342a3c..b319dff5 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1585,7 +1585,7 @@ moves_loop: // When in check, search starts from here
                                                                 [to_sq(move)];
 
       if (  !captureOrPromotion
-          && moveCount >= abs(depth) + 1
+          && moveCount
           && (*contHist[0])[pos.moved_piece(move)][to_sq(move)] < CounterMovePruneThreshold
           && (*contHist[1])[pos.moved_piece(move)][to_sq(move)] < CounterMovePruneThreshold)
           continue;

From e0bafa1911ede61b9268e0b461a5d8856d6cd6be Mon Sep 17 00:00:00 2001
From: FauziAkram <fauzi.dabat@hotmail.com>
Date: Sun, 30 Aug 2020 13:58:05 +0300
Subject: [PATCH 27/52] Update parametes in classical evaluation.

Passed STC (NNUE=False):
https://tests.stockfishchess.org/tests/view/5f42edfe5089a564a10d84a0
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 13840 W: 2591 L: 2336 D: 8913
Ptnml(0-2): 194, 1453, 3387, 1676, 210

Passed LTC (NNUE=False):
https://tests.stockfishchess.org/tests/view/5f4369795089a564a10d84d8
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 159744 W: 19430 L: 18850 D: 121464
Ptnml(0-2): 960, 14185, 49030, 14709, 988

closes https://github.com/official-stockfish/Stockfish/pull/3080

bench: 3736029
---
 src/evaluate.cpp | 10 +++++-----
 src/search.cpp   |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index 67154751..09f36513 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -227,26 +227,26 @@ namespace {
 
   // Outpost[knight/bishop] contains bonuses for each knight or bishop occupying a
   // pawn protected square on rank 4 to 6 which is also safe from a pawn attack.
-  constexpr Score Outpost[] = { S(56, 36), S(30, 23) };
+  constexpr Score Outpost[] = { S(56, 34), S(31, 23) };
 
   // PassedRank[Rank] contains a bonus according to the rank of a passed pawn
   constexpr Score PassedRank[RANK_NB] = {
-    S(0, 0), S(10, 28), S(17, 33), S(15, 41), S(62, 72), S(168, 177), S(276, 260)
+    S(0, 0), S(9, 28), S(15, 31), S(17, 39), S(64, 70), S(171, 177), S(277, 260)
   };
 
   // RookOnFile[semiopen/open] contains bonuses for each rook when there is
   // no (friendly) pawn on the rook file.
-  constexpr Score RookOnFile[] = { S(19, 7), S(48, 29) };
+  constexpr Score RookOnFile[] = { S(19, 7), S(48, 27) };
 
   // ThreatByMinor/ByRook[attacked PieceType] contains bonuses according to
   // which piece type attacks which one. Attacks on lesser pieces which are
   // pawn-defended are not considered.
   constexpr Score ThreatByMinor[PIECE_TYPE_NB] = {
-    S(0, 0), S(5, 32), S(57, 41), S(77, 56), S(88, 119), S(79, 161)
+    S(0, 0), S(5, 32), S(55, 41), S(77, 56), S(89, 119), S(79, 162)
   };
 
   constexpr Score ThreatByRook[PIECE_TYPE_NB] = {
-    S(0, 0), S(3, 46), S(37, 68), S(42, 60), S(0, 38), S(58, 41)
+    S(0, 0), S(3, 44), S(37, 68), S(42, 60), S(0, 39), S(58, 43)
   };
 
   // Assorted bonuses and penalties
diff --git a/src/search.cpp b/src/search.cpp
index b319dff5..e6e53e7c 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -944,7 +944,7 @@ namespace {
             }
          ss->ttPv = ttPv;
     }
-    
+
     // Step 11. If the position is not in TT, decrease depth by 2
     if (   PvNode
         && depth >= 6

From a0afe32d16554ff3b5c74f34ae56400f35759edf Mon Sep 17 00:00:00 2001
From: mstembera <MissingEmail@email>
Date: Sun, 30 Aug 2020 18:40:49 -0700
Subject: [PATCH 28/52] Use stable sort to make sure bench with TB yields same
 results everywhere.

std::sort() is not stable so different implementations can produce different results:
use the stable version instead.

Observed for '8/6k1/5r2/8/8/8/1K6/Q7 w - - 0 1' yielding different bench results for gcc and MSVC
and 3-4-5 syzygy TB prior to this patch.

closes https://github.com/official-stockfish/Stockfish/pull/3083

No functional change.
---
 src/search.cpp         | 2 +-
 src/syzygy/tbprobe.cpp | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index e6e53e7c..c676bd6d 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1964,7 +1964,7 @@ void Tablebases::rank_root_moves(Position& pos, Search::RootMoves& rootMoves) {
     if (RootInTB)
     {
         // Sort moves according to TB rank
-        std::sort(rootMoves.begin(), rootMoves.end(),
+        std::stable_sort(rootMoves.begin(), rootMoves.end(),
                   [](const RootMove &a, const RootMove &b) { return a.tbRank > b.tbRank; } );
 
         // Probe during search only if DTZ is not available and we are winning
diff --git a/src/syzygy/tbprobe.cpp b/src/syzygy/tbprobe.cpp
index 20215b96..3dfe3e3e 100644
--- a/src/syzygy/tbprobe.cpp
+++ b/src/syzygy/tbprobe.cpp
@@ -758,7 +758,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu
     if (entry->hasPawns) {
         idx = LeadPawnIdx[leadPawnsCnt][squares[0]];
 
-        std::sort(squares + 1, squares + leadPawnsCnt, pawns_comp);
+        std::stable_sort(squares + 1, squares + leadPawnsCnt, pawns_comp);
 
         for (int i = 1; i < leadPawnsCnt; ++i)
             idx += Binomial[i][MapPawns[squares[i]]];
@@ -859,7 +859,7 @@ encode_remaining:
 
     while (d->groupLen[++next])
     {
-        std::sort(groupSq, groupSq + d->groupLen[next]);
+        std::stable_sort(groupSq, groupSq + d->groupLen[next]);
         uint64_t n = 0;
 
         // Map down a square if "comes later" than a square in the previous

From a057f170c6920fc4d1abdae619c5259e9d80703c Mon Sep 17 00:00:00 2001
From: mstembera <MissingEmail@email>
Date: Sun, 30 Aug 2020 20:48:10 -0700
Subject: [PATCH 29/52] Use llvm linker with clang on windows for LTO.

other linkers might fail to link during the LTO phase.

The linker might have to be installed using
`pacman -Syu mingw-w64-x86_64-lld`

closes https://github.com/official-stockfish/Stockfish/pull/3084

No functional change.
---
 src/Makefile | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/Makefile b/src/Makefile
index 9ae5a51c..340b3008 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -595,6 +595,11 @@ ifeq ($(debug), no)
 		LDFLAGS += $(CXXFLAGS)
 	else ifeq ($(comp),clang)
 		CXXFLAGS += -flto=thin
+		ifneq ($(findstring MINGW,$(KERNEL)),)
+			CXXFLAGS += -fuse-ld=lld
+		else ifneq ($(findstring MSYS,$(KERNEL)),)
+			CXXFLAGS += -fuse-ld=lld
+		endif
 		LDFLAGS += $(CXXFLAGS)
 
 # GCC and CLANG use different methods for parallelizing LTO and CLANG pretends to be

From 61381372ec896ae6b0f139555e6e3f816d8aa570 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Mon, 31 Aug 2020 22:53:20 +0200
Subject: [PATCH 30/52] Always print an info line before a bestmove

if very few nodes are being searched before a bestmove is reported,
an info line might be missing.

fixes https://github.com/official-stockfish/Stockfish/issues/2757

closes https://github.com/official-stockfish/Stockfish/pull/3088

no functional change
---
 src/search.cpp | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index c676bd6d..c15cd753 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1861,12 +1861,15 @@ string UCI::pv(const Position& pos, Depth depth, Value alpha, Value beta) {
   {
       bool updated = rootMoves[i].score != -VALUE_INFINITE;
 
-      if (depth == 1 && !updated)
+      if (depth == 1 && !updated && i > 0)
           continue;
 
-      Depth d = updated ? depth : depth - 1;
+      Depth d = updated ? depth : std::max(1, depth - 1);
       Value v = updated ? rootMoves[i].score : rootMoves[i].previousScore;
 
+      if (v == -VALUE_INFINITE)
+          v = VALUE_ZERO;
+
       bool tb = TB::RootInTB && abs(v) < VALUE_MATE_IN_MAX_PLY;
       v = tb ? rootMoves[i].tbScore : v;
 

From a8bbaa17954471cf3fd8d168f1cafe3f2034730e Mon Sep 17 00:00:00 2001
From: VoyagerOne <excelgeek@gmail.com>
Date: Sun, 30 Aug 2020 13:57:57 -0400
Subject: [PATCH 31/52] LMR Root Node Simplification

Simplify LMR at Root node

STC:
LLR: 2.94 (-2.94,2.94) {-1.25,0.25}
Total: 71520 W: 7649 L: 7614 D: 56257
Ptnml(0-2): 346, 5845, 23349, 5868, 352
https://tests.stockfishchess.org/tests/view/5f4be8c0ba100690c5cc5cbb

LTC:
LLR: 2.95 (-2.94,2.94) {-0.75,0.25}
Total: 74832 W: 3997 L: 3948 D: 66887
Ptnml(0-2): 77, 3422, 30362, 3485, 70
https://tests.stockfishchess.org/tests/view/5f4c603eba100690c5cc5d0e

closes https://github.com/official-stockfish/Stockfish/pull/3091

Bench: 3624569
---
 src/search.cpp |  1 -
 src/thread.cpp | 11 -----------
 src/thread.h   |  1 -
 3 files changed, 13 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index c15cd753..b79fa6be 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1160,7 +1160,6 @@ moves_loop: // When in check, search starts from here
       // re-searched at full depth.
       if (    depth >= 3
           &&  moveCount > 1 + 2 * rootNode + 2 * (PvNode && abs(bestValue) < 2)
-          && (!rootNode || thisThread->best_move_count(move) == 0)
           && (  !captureOrPromotion
               || moveCountPruning
               || ss->staticEval + PieceValue[EG][pos.captured_piece()] <= alpha
diff --git a/src/thread.cpp b/src/thread.cpp
index 1aa66a81..b46fce5e 100644
--- a/src/thread.cpp
+++ b/src/thread.cpp
@@ -51,17 +51,6 @@ Thread::~Thread() {
 }
 
 
-/// Thread::bestMoveCount(Move move) return best move counter for the given root move
-
-int Thread::best_move_count(Move move) const {
-
-  auto rm = std::find(rootMoves.begin() + pvIdx,
-                      rootMoves.begin() + pvLast, move);
-
-  return rm != rootMoves.begin() + pvLast ? rm->bestMoveCount : 0;
-}
-
-
 /// Thread::clear() reset histories, usually before a new game
 
 void Thread::clear() {
diff --git a/src/thread.h b/src/thread.h
index 042bc2e9..34b99015 100644
--- a/src/thread.h
+++ b/src/thread.h
@@ -54,7 +54,6 @@ public:
   void idle_loop();
   void start_searching();
   void wait_for_search_finished();
-  int best_move_count(Move move) const;
 
   Pawns::Table pawnsTable;
   Material::Table materialTable;

From be87517734e2a1b222d1a35e98764382b4176732 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Tue, 1 Sep 2020 12:22:47 +0200
Subject: [PATCH 32/52] Only use MADV_RANDOM if defined

needed to compile on Haiku.

fixes https://github.com/official-stockfish/Stockfish/issues/3093

closes https://github.com/official-stockfish/Stockfish/pull/3094

No functional change
---
 src/syzygy/tbprobe.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/syzygy/tbprobe.cpp b/src/syzygy/tbprobe.cpp
index 3dfe3e3e..4d682f1a 100644
--- a/src/syzygy/tbprobe.cpp
+++ b/src/syzygy/tbprobe.cpp
@@ -223,7 +223,9 @@ public:
 
         *mapping = statbuf.st_size;
         *baseAddress = mmap(nullptr, statbuf.st_size, PROT_READ, MAP_SHARED, fd, 0);
+#if defined(MADV_RANDOM)
         madvise(*baseAddress, statbuf.st_size, MADV_RANDOM);
+#endif
         ::close(fd);
 
         if (*baseAddress == MAP_FAILED)

From aa2de712302a2379d8aa26127d86455ad276f512 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Wed, 2 Sep 2020 08:05:08 +0200
Subject: [PATCH 33/52] Update CPU contributors list

with fishtest data of Sept. 2 2020

closes https://github.com/official-stockfish/Stockfish/pull/3095

No functional change
---
 Top CPU Contributors.txt | 319 +++++++++++++++++++++------------------
 1 file changed, 169 insertions(+), 150 deletions(-)

diff --git a/Top CPU Contributors.txt b/Top CPU Contributors.txt
index 0ea5ac72..482e9000 100644
--- a/Top CPU Contributors.txt	
+++ b/Top CPU Contributors.txt	
@@ -1,154 +1,173 @@
-Contributors with >10,000 CPU hours as of January 7, 2020
+Contributors with >10,000 CPU hours as of Sept 2, 2020
 Thank you!
 
 Username                  CPU Hours   Games played
 --------------------------------------------------
-noobpwnftw                  9305707      695548021
-mlang                        780050       61648867
-dew                          621626       43921547
-mibere                       524702       42238645
-crunchy                      354587       27344275
-cw                           354495       27274181
-fastgm                       332801       22804359
-JojoM                        295750       20437451
-CSU_Dynasty                  262015       21828122
-Fisherman                    232181       18939229
-ctoks                        218866       17622052
-glinscott                    201989       13780820
-tvijlbrief                   201204       15337115
-velislav                     188630       14348485
-gvreuls                      187164       15149976
-bking_US                     180289       11876016
-nordlandia                   172076       13467830
-leszek                       157152       11443978
-Thanar                       148021       12365359
-spams                        141975       10319326
-drabel                       138073       11121749
-vdv                          137850        9394330
-mgrabiak                     133578       10454324
-TueRens                      132485       10878471
-bcross                       129683       11557084
-marrco                       126078        9356740
-sqrt2                        125830        9724586
-robal                        122873        9593418
-vdbergh                      120766        8926915
-malala                       115926        8002293
-CoffeeOne                    114241        5004100
-dsmith                       113189        7570238
-BrunoBanani                  104644        7436849
-Data                          92328        8220352
-mhoram                        89333        6695109
-davar                         87924        7009424
-xoto                          81094        6869316
-ElbertoOne                    80899        7023771
-grandphish2                   78067        6160199
-brabos                        77212        6186135
-psk                           75733        5984901
-BRAVONE                       73875        5054681
-sunu                          70771        5597972
-sterni1971                    70605        5590573
-MaZePallas                    66886        5188978
-Vizvezdenec                   63708        4967313
-nssy                          63462        5259388
-jromang                       61634        4940891
-teddybaer                     61231        5407666
-Pking_cda                     60099        5293873
-solarlight                    57469        5028306
-dv8silencer                   56913        3883992
-tinker                        54936        4086118
-renouve                       49732        3501516
-Freja                         49543        3733019
-robnjr                        46972        4053117
-rap                           46563        3219146
-Bobo1239                      46036        3817196
-ttruscott                     45304        3649765
-racerschmacer                 44881        3975413
-finfish                       44764        3370515
-eva42                         41783        3599691
-biffhero                      40263        3111352
-bigpen0r                      39817        3291647
-mhunt                         38871        2691355
-ronaldjerum                   38820        3240695
-Antihistamine                 38785        2761312
-pb00067                       38038        3086320
-speedycpu                     37591        3003273
-rkl                           37207        3289580
-VoyagerOne                    37050        3441673
-jbwiebe                       35320        2805433
-cuistot                       34191        2146279
-homyur                        33927        2850481
-manap                         32873        2327384
-gri                           32538        2515779
-oryx                          31267        2899051
-EthanOConnor                  30959        2090311
-SC                            30832        2730764
-csnodgrass                    29505        2688994
-jmdana                        29458        2205261
-strelock                      28219        2067805
-jkiiski                       27832        1904470
-Pyafue                        27533        1902349
-Garf                          27515        2747562
-eastorwest                    27421        2317535
-slakovv                       26903        2021889
-Prcuvu                        24835        2170122
-anst                          24714        2190091
-hyperbolic.tom                24319        2017394
-Patrick_G                     23687        1801617
-Sharaf_DG                     22896        1786697
-nabildanial                   22195        1519409
-chriswk                       21931        1868317
-achambord                     21665        1767323
-Zirie                         20887        1472937
-team-oh                       20217        1636708
-Isidor                        20096        1680691
-ncfish1                       19931        1520927
-nesoneg                       19875        1463031
-Spprtr                        19853        1548165
-JanErik                       19849        1703875
-agg177                        19478        1395014
-SFTUser                       19231        1567999
-xor12                         19017        1680165
-sg4032                        18431        1641865
-rstoesser                     18118        1293588
-MazeOfGalious                 17917        1629593
-j3corre                       17743         941444
-cisco2015                     17725        1690126
-ianh2105                      17706        1632562
-dex                           17678        1467203
-jundery                       17194        1115855
-iisiraider                    17019        1101015
-horst.prack                   17012        1465656
-Adrian.Schmidt123             16563        1281436
-purplefishies                 16342        1092533
-wei                           16274        1745989
-ville                         16144        1384026
-eudhan                        15712        1283717
-OuaisBla                      15581         972000
-DragonLord                    15559        1162790
-dju                           14716         875569
-chris                         14479        1487385
-0xB00B1ES                     14079        1001120
-OssumOpossum                  13776        1007129
-enedene                       13460         905279
-bpfliegel                     13346         884523
-Ente                          13198        1156722
-IgorLeMasson                  13087        1147232
-jpulman                       13000         870599
-ako027ako                     12775        1173203
-Nikolay.IT                    12352        1068349
-Andrew Grant                  12327         895539
-joster                        12008         950160
-AdrianSA                      11996         804972
-Nesa92                        11455        1111993
-fatmurphy                     11345         853210
-Dark_wizzie                   11108        1007152
-modolief                      10869         896470
-mschmidt                      10757         803401
-infinity                      10594         727027
-mabichito                     10524         749391
-Thomas A. Anderson            10474         732094
-thijsk                        10431         719357
-Flopzee                       10339         894821
-crocogoat                     10104        1013854
-SapphireBrand                 10104         969604
-stocky                        10017         699440
+noobpwnftw                 19352969     1231459677
+mlang                        957168       61657446
+dew                          949885       56893432
+mibere                       703817       46865007
+crunchy                      427035       27344275
+cw                           416006       27521077
+JojoM                        415904       24479564
+fastgm                       404873       23953472
+CSU_Dynasty                  335774       22850550
+tvijlbrief                   335199       21871270
+Fisherman                    325053       21786603
+gvreuls                      311480       20751516
+ctoks                        275877       18710423
+velislav                     241267       15596372
+glinscott                    217799       13780820
+nordlandia                   211692       13484886
+bcross                       206213       14934233
+bking_US                     198894       11876016
+leszek                       189170       11446821
+mgrabiak                     183896       11778092
+drabel                       181408       12489478
+TueRens                      181349       12192000
+Thanar                       179852       12365359
+vdv                          175171        9881246
+robal                        166948       10702862
+spams                        157128       10319326
+marrco                       149947        9376421
+sqrt2                        147963        9724586
+vdbergh                      137041        8926915
+CoffeeOne                    136294        5004100
+malala                       136182        8002293
+mhoram                       128934        8177193
+davar                        122092        7960001
+dsmith                       122059        7570238
+xoto                         119696        8222144
+grandphish2                  116481        7582197
+Data                         113305        8220352
+BrunoBanani                  112960        7436849
+ElbertoOne                    99028        7023771
+MaZePallas                    98571        6362619
+brabos                        92118        6186135
+psk                           89957        5984901
+sunu                          88463        6007033
+sterni1971                    86948        5613788
+Vizvezdenec                   83752        5343724
+BRAVONE                       81239        5054681
+nssy                          76497        5259388
+teddybaer                     75125        5407666
+Pking_cda                     73776        5293873
+jromang                       70695        4940891
+solarlight                    70517        5028306
+dv8silencer                   70287        3883992
+Bobo1239                      68515        4652287
+racerschmacer                 67468        4935996
+manap                         66273        4121774
+tinker                        63458        4213726
+linrock                       59082        4516053
+robnjr                        57262        4053117
+Freja                         56938        3733019
+ttruscott                     56005        3679485
+renouve                       53811        3501516
+cuistot                       52532        3014920
+finfish                       51360        3370515
+eva42                         51272        3599691
+rkl                           50759        3840947
+rap                           49985        3219146
+pb00067                       49727        3298270
+ronaldjerum                   47654        3240695
+bigpen0r                      47278        3291647
+biffhero                      46564        3111352
+VoyagerOne                    45386        3445881
+speedycpu                     43842        3003273
+jbwiebe                       43305        2805433
+Antihistamine                 41788        2761312
+mhunt                         41735        2691355
+eastorwest                    40387        2812173
+homyur                        39893        2850481
+gri                           39871        2515779
+oryx                          38228        2941656
+0x3C33                        37773        2529097
+SC                            37290        2731014
+csnodgrass                    36207        2688994
+jmdana                        36108        2205261
+strelock                      34716        2074055
+Garf                          33800        2747562
+EthanOConnor                  33370        2090311
+slakovv                       32915        2021889
+Spprtr                        32591        2139601
+Prcuvu                        30377        2170122
+anst                          30301        2190091
+jkiiski                       30136        1904470
+hyperbolic.tom                29840        2017394
+Pyafue                        29650        1902349
+OuaisBla                      27629        1578000
+chriswk                       26902        1868317
+achambord                     26582        1767323
+Patrick_G                     26276        1801617
+yorkman                       26193        1992080
+SFTUser                       25182        1675689
+nabildanial                   24942        1519409
+Sharaf_DG                     24765        1786697
+ncfish1                       24411        1520927
+agg177                        23890        1395014
+JanErik                       23408        1703875
+Isidor                        23388        1680691
+Norabor                       22976        1587862
+cisco2015                     22880        1759669
+Zirie                         22542        1472937
+team-oh                       22272        1636708
+MazeOfGalious                 21978        1629593
+sg4032                        21945        1643065
+ianh2105                      21725        1632562
+xor12                         21628        1680365
+dex                           21612        1467203
+nesoneg                       21494        1463031
+horst.prack                   20878        1465656
+0xB00B1ES                     20590        1208666
+j3corre                       20405         941444
+Adrian.Schmidt123             20316        1281436
+wei                           19973        1745989
+rstoesser                     19569        1293588
+eudhan                        19274        1283717
+Ente                          19070        1373058
+jundery                       18445        1115855
+iisiraider                    18247        1101015
+ville                         17883        1384026
+chris                         17698        1487385
+purplefishies                 17595        1092533
+DragonLord                    17014        1162790
+dju                           16515         929427
+IgorLeMasson                  16064        1147232
+ako027ako                     15671        1173203
+Nikolay.IT                    15154        1068349
+Andrew Grant                  15114         895539
+yurikvelo                     15027        1165616
+OssumOpossum                  14857        1007129
+enedene                       14476         905279
+bpfliegel                     14298         884523
+jpulman                       13982         870599
+joster                        13794         950160
+Nesa92                        13786        1114691
+Dark_wizzie                   13422        1007152
+Hjax                          13350         900887
+Fifis                         13313         965473
+mabichito                     12903         749391
+thijsk                        12886         722107
+crocogoat                     12876        1048802
+AdrianSA                      12860         804972
+Flopzee                       12698         894821
+fatmurphy                     12547         853210
+SapphireBrand                 12416         969604
+modolief                      12386         896470
+scuzzi                        12362         833465
+pgontarz                      12151         848794
+stocky                        11954         699440
+mschmidt                      11941         803401
+infinity                      11470         727027
+torbjo                        11387         728873
+Thomas A. Anderson            11372         732094
+snicolet                      11106         869170
+amicic                        10779         733593
+rpngn                         10712         688203
+d64                           10680         771144
+basepi                        10637         744851
+jjoshua2                      10559         670905
+dzjp                          10343         732529
+ols                           10259         570669
+lbraesch                      10252         647825

From c306d838697011da0a960758dde3f7ede6849060 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Wed, 2 Sep 2020 09:12:04 +0200
Subject: [PATCH 34/52] Stockfish 12

Official release version of Stockfish 12

Bench: 3624569

-----------------------

It is our pleasure to release Stockfish 12 to users world-wide

Downloads will be freely available at

https://stockfishchess.org/download/

This version 12 of Stockfish plays significantly stronger than
any of its predecessors. In a match against Stockfish 11,
Stockfish 12 will typically win at least ten times more game pairs
than it loses.

This jump in strength, visible in regular progression tests during
development[1], results from the introduction of an efficiently
updatable neural network (NNUE) for the evaluation in Stockfish[2],
and associated tuning of the engine as a whole. The concept of the
NNUE evaluation was first introduced in shogi, and ported to
Stockfish afterward. Stockfish remains a CPU-only engine, since the
NNUE networks can be very efficiently evaluated on CPUs. The
recommended parameters of the NNUE network are embedded in
distributed binaries, and Stockfish will use NNUE by default.

Both the NNUE and the classical evaluations are available, and
can be used to assign values to positions that are later used in
alpha-beta (PVS) search to find the best move. The classical
evaluation computes this value as a function of various chess
concepts, handcrafted by experts, tested and tuned using fishtest.
The NNUE evaluation computes this value with a neural network based
on basic inputs. The network is optimized and trained on the
evaluations of millions of positions.

The Stockfish project builds on a thriving community of enthusiasts
that contribute their expertise, time, and resources to build a free
and open source chess engine that is robust, widely available, and
very strong. We invite chess fans to join the fishtest testing
framework and programmers to contribute on github[3].

Stay safe and enjoy chess!

The Stockfish team

[1] https://github.com/glinscott/fishtest/wiki/Regression-Tests
[2] https://github.com/official-stockfish/Stockfish/commit/84f3e867903f62480c33243dd0ecbffd342796fc
[3] https://stockfishchess.org/get-involved/
---
 src/misc.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/misc.cpp b/src/misc.cpp
index 3fbdea35..22070f0e 100644
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -65,7 +65,7 @@ namespace {
 
 /// Version number. If Version is left empty, then compile date in the format
 /// DD-MM-YY and show in engine_info.
-const string Version = "";
+const string Version = "12";
 
 /// Our fancy logging facility. The trick here is to replace cin.rdbuf() and
 /// cout.rdbuf() with two Tie objects that tie cin and cout to a file stream. We

From 571c2d6d8daf70de884c493b40cf0279e9b48c61 Mon Sep 17 00:00:00 2001
From: Joost VandeVondele <Joost.VandeVondele@gmail.com>
Date: Fri, 4 Sep 2020 07:46:06 +0200
Subject: [PATCH 35/52] Restore development version

have fun!

No functional change
---
 src/misc.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/misc.cpp b/src/misc.cpp
index 22070f0e..3fbdea35 100644
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -65,7 +65,7 @@ namespace {
 
 /// Version number. If Version is left empty, then compile date in the format
 /// DD-MM-YY and show in engine_info.
-const string Version = "12";
+const string Version = "";
 
 /// Our fancy logging facility. The trick here is to replace cin.rdbuf() and
 /// cout.rdbuf() with two Tie objects that tie cin and cout to a file stream. We

From 0e1f734b05ee5c67e9a17ae0e2045a64209dee05 Mon Sep 17 00:00:00 2001
From: Unai Corzo <corzounai@gmail.com>
Date: Wed, 2 Sep 2020 16:45:49 +0200
Subject: [PATCH 36/52] Less pruning in qsearch

do not prune moves that give discovery checks, even if with negative SSE.

STC https://tests.stockfishchess.org/tests/view/5f4cb5e8ba100690c5cc5d25
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 91328 W: 9940 L: 9667 D: 71721
Ptnml(0-2): 491, 7345, 29693, 7670, 465

LTC https://tests.stockfishchess.org/tests/view/5f4dbc2eba100690c5cc5dac
LLR: 2.97 (-2.94,2.94) {0.25,1.25}
Total: 52448 W: 2799 L: 2586 D: 47063
Ptnml(0-2): 53, 2220, 21459, 2445, 47

closes https://github.com/official-stockfish/Stockfish/pull/3098

bench: 4031192
---
 src/search.cpp | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/search.cpp b/src/search.cpp
index b79fa6be..0d823c8e 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1564,7 +1564,9 @@ moves_loop: // When in check, search starts from here
       }
 
       // Do not search moves with negative SEE values
-      if (!ss->inCheck && !pos.see_ge(move))
+      if (   !ss->inCheck
+          && !(givesCheck && pos.is_discovery_check_on_king(~pos.side_to_move(), move))
+          && !pos.see_ge(move))
           continue;
 
       // Speculative prefetch as early as possible

From d6530f7d49ef45e38dacafd8a3a838130113265c Mon Sep 17 00:00:00 2001
From: Unai Corzo <corzounai@gmail.com>
Date: Thu, 3 Sep 2020 12:18:42 +0200
Subject: [PATCH 37/52] Simplify singularQuietLMR

remove formerPV dependence

STC https://tests.stockfishchess.org/tests/view/5f4cb922ba100690c5cc5d35
LLR: 2.96 (-2.94,2.94) {-1.25,0.25}
Total: 113672 W: 12347 L: 12368 D: 88957
Ptnml(0-2): 566, 9537, 36699, 9420, 614

LTC https://tests.stockfishchess.org/tests/view/5f4e8474ba100690c5cc5e12
LLR: 2.93 (-2.94,2.94) {-0.75,0.25}
Total: 43032 W: 2298 L: 2227 D: 38507
Ptnml(0-2): 45, 1940, 17475, 2011, 45

closes https://github.com/official-stockfish/Stockfish/pull/3102

bench: 3290084
---
 src/search.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/search.cpp b/src/search.cpp
index 0d823c8e..b5e190c8 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1196,7 +1196,7 @@ moves_loop: // When in check, search starts from here
 
           // Decrease reduction if ttMove has been singularly extended (~3 Elo)
           if (singularQuietLMR)
-              r -= 1 + formerPv;
+              r--;
 
           if (!captureOrPromotion)
           {

From 2a696115094882b7dc5c024a97ed7dc2bdc98642 Mon Sep 17 00:00:00 2001
From: VoyagerOne <excelgeek@gmail.com>
Date: Wed, 2 Sep 2020 16:58:44 -0400
Subject: [PATCH 38/52] LMR Simplification

remove reduction at non-check cut nodes for second move at low depths

STC:
LLR: 2.95 (-2.94,2.94) {-1.25,0.25}
Total: 61712 W: 6594 L: 6543 D: 48575
Ptnml(0-2): 293, 5085, 20082, 5070, 326
https://tests.stockfishchess.org/tests/view/5f5007d6ba100690c5cc5ea9

LTC:
LLR: 2.94 (-2.94,2.94) {-0.75,0.25}
Total: 57544 W: 2983 L: 2925 D: 51636
Ptnml(0-2): 47, 2568, 23495, 2604, 58
https://tests.stockfishchess.org/tests/view/5f50c597ba100690c5cc5ef7

closes https://github.com/official-stockfish/Stockfish/pull/3103

Bench: 3952302
---
 src/search.cpp | 7 -------
 1 file changed, 7 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index b5e190c8..a7692841 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1168,13 +1168,6 @@ moves_loop: // When in check, search starts from here
       {
           Depth r = reduction(improving, depth, moveCount);
 
-          // Decrease reduction at non-check cut nodes for second move at low depths
-          if (   cutNode
-              && depth <= 10
-              && moveCount <= 2
-              && !ss->inCheck)
-              r--;
-
           // Decrease reduction if the ttHit running average is large
           if (thisThread->ttHitAverage > 509 * TtHitAverageResolution * TtHitAverageWindow / 1024)
               r--;

From 9cc482c7889cbc5f6d92e1b69ccd28d422a44a32 Mon Sep 17 00:00:00 2001
From: Sergio Vieri <sergio.vieri.hp@gmail.com>
Date: Thu, 3 Sep 2020 20:22:51 +0800
Subject: [PATCH 39/52] Update default net to nn-308d71810dff.nnue

equivalent to 20200903-1739

Net trained from scratch, so it has quite different features extracted compared to the previous net (82215d0fd0df).

STC:
LLR: 2.98 (-2.94,2.94) {-0.25,1.25}
Total: 108328 W: 14048 L: 13719 D: 80561
Ptnml(0-2): 842, 10039, 32062, 10390, 831
https://tests.stockfishchess.org/tests/view/5f50e053ba100690c5cc5f00

LTC:
LLR: 2.96 (-2.94,2.94) {0.25,1.25}
Total: 13872 W: 1059 L: 890 D: 11923
Ptnml(0-2): 30, 724, 5270, 871, 41
https://tests.stockfishchess.org/tests/view/5f51821fba100690c5cc5f36

closes https://github.com/official-stockfish/Stockfish/pull/3104

Bench: 3832716
---
 src/evaluate.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/evaluate.h b/src/evaluate.h
index d701f5a7..3da6a9fe 100644
--- a/src/evaluate.h
+++ b/src/evaluate.h
@@ -38,7 +38,7 @@ namespace Eval {
   // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
   // for the build process (profile-build and fishtest) to work. Do not change the
   // name of the macro, as it is used in the Makefile.
-  #define EvalFileDefaultName   "nn-82215d0fd0df.nnue"
+  #define EvalFileDefaultName   "nn-308d71810dff.nnue"
 
   namespace NNUE {
 

From 9a063fc3cbc8f522215392db232eeb0e04e71b2c Mon Sep 17 00:00:00 2001
From: Vizvezdenec <Vizvezdenec@gmail.com>
Date: Fri, 4 Sep 2020 15:53:59 +0300
Subject: [PATCH 40/52] Adjust penalty on refuted early quiet moves

This patch changes how previous early moves are penalized in case
search finds a best move. Here, the first quiet move that was not
a transposition table move is penalized.

passed STC
https://tests.stockfishchess.org/tests/view/5f51d839ba100690c5cc5f69
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 10088 W: 1150 L: 997 D: 7941
Ptnml(0-2): 41, 772, 3278, 899, 54

passed LTC
https://tests.stockfishchess.org/tests/view/5f51e435ba100690c5cc5f76
LLR: 2.93 (-2.94,2.94) {0.25,1.25}
Total: 30808 W: 1564 L: 1405 D: 27839
Ptnml(0-2): 19, 1245, 12717, 1404, 19

closes https://github.com/official-stockfish/Stockfish/pull/3106

bench 3983758
---
 src/search.cpp | 42 +++++++++++++++++++++---------------------
 src/search.h   |  1 +
 2 files changed, 22 insertions(+), 21 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index a7692841..4aeadc28 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -597,7 +597,7 @@ namespace {
     Move ttMove, move, excludedMove, bestMove;
     Depth extension, newDepth;
     Value bestValue, value, ttValue, eval, maxValue, probCutBeta;
-    bool ttHit, formerPv, givesCheck, improving, didLMR, priorCapture;
+    bool formerPv, givesCheck, improving, didLMR, priorCapture;
     bool captureOrPromotion, doFullDepthSearch, moveCountPruning,
          ttCapture, singularQuietLMR;
     Piece movedPiece;
@@ -664,12 +664,12 @@ namespace {
     // position key in case of an excluded move.
     excludedMove = ss->excludedMove;
     posKey = excludedMove == MOVE_NONE ? pos.key() : pos.key() ^ make_key(excludedMove);
-    tte = TT.probe(posKey, ttHit);
-    ttValue = ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
+    tte = TT.probe(posKey, ss->ttHit);
+    ttValue = ss->ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
     ttMove =  rootNode ? thisThread->rootMoves[thisThread->pvIdx].pv[0]
-            : ttHit    ? tte->move() : MOVE_NONE;
+            : ss->ttHit    ? tte->move() : MOVE_NONE;
     if (!excludedMove)
-        ss->ttPv = PvNode || (ttHit && tte->is_pv());
+        ss->ttPv = PvNode || (ss->ttHit && tte->is_pv());
     formerPv = ss->ttPv && !PvNode;
 
     if (   ss->ttPv
@@ -681,11 +681,11 @@ namespace {
 
     // thisThread->ttHitAverage can be used to approximate the running average of ttHit
     thisThread->ttHitAverage =   (TtHitAverageWindow - 1) * thisThread->ttHitAverage / TtHitAverageWindow
-                                + TtHitAverageResolution * ttHit;
+                                + TtHitAverageResolution * ss->ttHit;
 
     // At non-PV nodes we check for an early TT cutoff
     if (  !PvNode
-        && ttHit
+        && ss->ttHit
         && tte->depth() >= depth
         && ttValue != VALUE_NONE // Possible in case of TT access race
         && (ttValue >= beta ? (tte->bound() & BOUND_LOWER)
@@ -778,7 +778,7 @@ namespace {
         improving = false;
         goto moves_loop;
     }
-    else if (ttHit)
+    else if (ss->ttHit)
     {
         // Never assume anything about values stored in TT
         ss->staticEval = eval = tte->eval();
@@ -882,14 +882,14 @@ namespace {
         // there and in further interactions with transposition table cutoff depth is set to depth - 3
         // because probCut search has depth set to depth - 4 but we also do a move before it
         // so effective depth is equal to depth - 3
-        && !(   ttHit
+        && !(   ss->ttHit
              && tte->depth() >= depth - 3
              && ttValue != VALUE_NONE
              && ttValue < probCutBeta))
     {
         // if ttMove is a capture and value from transposition table is good enough produce probCut
         // cutoff without digging into actual probCut search
-        if (   ttHit
+        if (   ss->ttHit
             && tte->depth() >= depth - 3
             && ttValue != VALUE_NONE
             && ttValue >= probCutBeta
@@ -933,7 +933,7 @@ namespace {
                 if (value >= probCutBeta)
                 {
                     // if transposition table doesn't have equal or more deep info write probCut data into it
-                    if ( !(ttHit
+                    if ( !(ss->ttHit
                        && tte->depth() >= depth - 3
                        && ttValue != VALUE_NONE))
                         tte->save(posKey, value_to_tt(value, ss->ply), ttPv,
@@ -1423,7 +1423,7 @@ moves_loop: // When in check, search starts from here
     Move ttMove, move, bestMove;
     Depth ttDepth;
     Value bestValue, value, ttValue, futilityValue, futilityBase, oldAlpha;
-    bool ttHit, pvHit, givesCheck, captureOrPromotion;
+    bool pvHit, givesCheck, captureOrPromotion;
     int moveCount;
 
     if (PvNode)
@@ -1453,13 +1453,13 @@ moves_loop: // When in check, search starts from here
                                                   : DEPTH_QS_NO_CHECKS;
     // Transposition table lookup
     posKey = pos.key();
-    tte = TT.probe(posKey, ttHit);
-    ttValue = ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
-    ttMove = ttHit ? tte->move() : MOVE_NONE;
-    pvHit = ttHit && tte->is_pv();
+    tte = TT.probe(posKey, ss->ttHit);
+    ttValue = ss->ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
+    ttMove = ss->ttHit ? tte->move() : MOVE_NONE;
+    pvHit = ss->ttHit && tte->is_pv();
 
     if (  !PvNode
-        && ttHit
+        && ss->ttHit
         && tte->depth() >= ttDepth
         && ttValue != VALUE_NONE // Only in case of TT access race
         && (ttValue >= beta ? (tte->bound() & BOUND_LOWER)
@@ -1474,7 +1474,7 @@ moves_loop: // When in check, search starts from here
     }
     else
     {
-        if (ttHit)
+        if (ss->ttHit)
         {
             // Never assume anything about values stored in TT
             if ((ss->staticEval = bestValue = tte->eval()) == VALUE_NONE)
@@ -1493,7 +1493,7 @@ moves_loop: // When in check, search starts from here
         // Stand pat. Return immediately if static value is at least beta
         if (bestValue >= beta)
         {
-            if (!ttHit)
+            if (!ss->ttHit)
                 tte->save(posKey, value_to_tt(bestValue, ss->ply), false, BOUND_LOWER,
                           DEPTH_NONE, MOVE_NONE, ss->staticEval);
 
@@ -1711,8 +1711,8 @@ moves_loop: // When in check, search starts from here
     else
         captureHistory[moved_piece][to_sq(bestMove)][captured] << bonus1;
 
-    // Extra penalty for a quiet TT or main killer move in previous ply when it gets refuted
-    if (   ((ss-1)->moveCount == 1 || ((ss-1)->currentMove == (ss-1)->killers[0]))
+    // Extra penalty for a quiet early move that was not a TT move or main killer move in previous ply when it gets refuted
+    if (   ((ss-1)->moveCount == 1 + (ss-1)->ttHit || ((ss-1)->currentMove == (ss-1)->killers[0]))
         && !pos.captured_piece())
             update_continuation_histories(ss-1, pos.piece_on(prevSq), prevSq, -bonus1);
 
diff --git a/src/search.h b/src/search.h
index 79085189..f60da4a5 100644
--- a/src/search.h
+++ b/src/search.h
@@ -49,6 +49,7 @@ struct Stack {
   int moveCount;
   bool inCheck;
   bool ttPv;
+  bool ttHit;
 };
 
 

From d539da19d2b13d70a81ab863f54046add0bc3b38 Mon Sep 17 00:00:00 2001
From: SFisGOD <jonathandumale@gmail.com>
Date: Fri, 4 Sep 2020 17:14:50 +0800
Subject: [PATCH 41/52] Use classical eval more often

If there is a moderate imbalance, use classical eval with small probability (1/16),
as derived from the node counter.

STC:
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 32320 W: 3562 L: 3377 D: 25381
Ptnml(0-2): 144, 2609, 10478, 2776, 153
https://tests.stockfishchess.org/tests/view/5f520615ba100690c5cc5f80

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 21032 W: 1116 L: 974 D: 18942
Ptnml(0-2): 20, 837, 8664, 971, 24
https://tests.stockfishchess.org/tests/view/5f522eaaba100690c5cc5f8c

closes https://github.com/official-stockfish/Stockfish/pull/3107

Bench: 4109324
---
 src/evaluate.cpp | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index 09f36513..db8379da 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -1015,12 +1015,16 @@ make_v:
 
 Value Eval::evaluate(const Position& pos) {
 
+  bool useClassical = abs(eg_value(pos.psq_score())) * 16 > NNUEThreshold1 * (16 + pos.rule50_count());
   bool classical = !Eval::useNNUE
-                ||  abs(eg_value(pos.psq_score())) * 16 > NNUEThreshold1 * (16 + pos.rule50_count());
+                ||  useClassical
+                || (abs(eg_value(pos.psq_score())) > PawnValueMg / 8 && !(pos.this_thread()->nodes & 0xF));
   Value v = classical ? Evaluation<NO_TRACE>(pos).value()
                       : NNUE::evaluate(pos) * 5 / 4 + Tempo;
 
-  if (classical && Eval::useNNUE && abs(v) * 16 < NNUEThreshold2 * (16 + pos.rule50_count()))
+  if (   useClassical 
+      && Eval::useNNUE 
+      && abs(v) * 16 < NNUEThreshold2 * (16 + pos.rule50_count()))
       v = NNUE::evaluate(pos) * 5 / 4 + Tempo;
 
   // Damp down the evaluation linearly when shuffling

From fc27d158c012341593518a05abf51903ecbcb495 Mon Sep 17 00:00:00 2001
From: syzygy1 <3028851+syzygy1@users.noreply.github.com>
Date: Sun, 6 Sep 2020 17:29:12 +0200
Subject: [PATCH 42/52] Bug fix in do_null_move() and NNUE simplification.

This fixes #3108 and removes some NNUE code that is currently not used.

At the moment, do_null_move() copies the accumulator from the previous
state into the new state, which is correct. It then clears the "computed_score"
flag because the side to move has changed, and with the other side to move
NNUE will return a completely different evaluation (normally with changed
sign but also with different NNUE-internal tempo bonus).

The problem is that do_null_move() clears the wrong flag. It clears the
computed_score flag of the old state, not of the new state. It turns out
that this almost never affects the search. For example, fixing it does not
change the current bench (but it does change the previous bench). This is
because the search code usually avoids calling evaluate() after a null move.

This PR corrects do_null_move() by removing the computed_score flag altogether.
The flag is not needed because nnue_evaluate() is never called twice on a position.

This PR also removes some unnecessary {}s and inserts a few blank lines
in the modified NNUE files in line with SF coding style.

Resulf ot STC non-regression test:
LLR: 2.95 (-2.94,2.94) {-1.25,0.25}
Total: 26328 W: 3118 L: 3012 D: 20198
Ptnml(0-2): 126, 2208, 8397, 2300, 133
https://tests.stockfishchess.org/tests/view/5f553ccc2d02727c56b36db1

closes https://github.com/official-stockfish/Stockfish/pull/3109

bench: 4109324
---
 src/nnue/evaluate_nnue.cpp          | 38 ++-----------------
 src/nnue/nnue_accumulator.h         |  2 -
 src/nnue/nnue_feature_transformer.h | 58 +++++++++++++----------------
 src/position.cpp                    |  2 -
 4 files changed, 29 insertions(+), 71 deletions(-)

diff --git a/src/nnue/evaluate_nnue.cpp b/src/nnue/evaluate_nnue.cpp
index d6ac9894..ed138881 100644
--- a/src/nnue/evaluate_nnue.cpp
+++ b/src/nnue/evaluate_nnue.cpp
@@ -115,31 +115,16 @@ namespace Eval::NNUE {
     return stream && stream.peek() == std::ios::traits_type::eof();
   }
 
-  // Proceed with the difference calculation if possible
-  static void UpdateAccumulatorIfPossible(const Position& pos) {
-
-    feature_transformer->UpdateAccumulatorIfPossible(pos);
-  }
-
-  // Calculate the evaluation value
-  static Value ComputeScore(const Position& pos, bool refresh) {
-
-    auto& accumulator = pos.state()->accumulator;
-    if (!refresh && accumulator.computed_score) {
-      return accumulator.score;
-    }
+  // Evaluation function. Perform differential calculation.
+  Value evaluate(const Position& pos) {
 
     alignas(kCacheLineSize) TransformedFeatureType
         transformed_features[FeatureTransformer::kBufferSize];
-    feature_transformer->Transform(pos, transformed_features, refresh);
+    feature_transformer->Transform(pos, transformed_features);
     alignas(kCacheLineSize) char buffer[Network::kBufferSize];
     const auto output = network->Propagate(transformed_features, buffer);
 
-    auto score = static_cast<Value>(output[0] / FV_SCALE);
-
-    accumulator.score = score;
-    accumulator.computed_score = true;
-    return accumulator.score;
+    return static_cast<Value>(output[0] / FV_SCALE);
   }
 
   // Load eval, from a file stream or a memory stream
@@ -150,19 +135,4 @@ namespace Eval::NNUE {
     return ReadParameters(stream);
   }
 
-  // Evaluation function. Perform differential calculation.
-  Value evaluate(const Position& pos) {
-    return ComputeScore(pos, false);
-  }
-
-  // Evaluation function. Perform full calculation.
-  Value compute_eval(const Position& pos) {
-    return ComputeScore(pos, true);
-  }
-
-  // Proceed with the difference calculation if possible
-  void update_eval(const Position& pos) {
-    UpdateAccumulatorIfPossible(pos);
-  }
-
 } // namespace Eval::NNUE
diff --git a/src/nnue/nnue_accumulator.h b/src/nnue/nnue_accumulator.h
index 69dfaad2..26370710 100644
--- a/src/nnue/nnue_accumulator.h
+++ b/src/nnue/nnue_accumulator.h
@@ -29,9 +29,7 @@ namespace Eval::NNUE {
   struct alignas(kCacheLineSize) Accumulator {
     std::int16_t
         accumulation[2][kRefreshTriggers.size()][kTransformedFeatureDimensions];
-    Value score;
     bool computed_accumulation;
-    bool computed_score;
   };
 
 }  // namespace Eval::NNUE
diff --git a/src/nnue/nnue_feature_transformer.h b/src/nnue/nnue_feature_transformer.h
index 43707610..2b6259c3 100644
--- a/src/nnue/nnue_feature_transformer.h
+++ b/src/nnue/nnue_feature_transformer.h
@@ -50,11 +50,13 @@ namespace Eval::NNUE {
 
     // Hash value embedded in the evaluation file
     static constexpr std::uint32_t GetHashValue() {
+
       return RawFeatures::kHashValue ^ kOutputDimensions;
     }
 
     // Read network parameters
     bool ReadParameters(std::istream& stream) {
+
       for (std::size_t i = 0; i < kHalfDimensions; ++i)
         biases_[i] = read_little_endian<BiasType>(stream);
       for (std::size_t i = 0; i < kHalfDimensions * kInputDimensions; ++i)
@@ -64,23 +66,26 @@ namespace Eval::NNUE {
 
     // Proceed with the difference calculation if possible
     bool UpdateAccumulatorIfPossible(const Position& pos) const {
+
       const auto now = pos.state();
-      if (now->accumulator.computed_accumulation) {
+      if (now->accumulator.computed_accumulation)
         return true;
-      }
+
       const auto prev = now->previous;
       if (prev && prev->accumulator.computed_accumulation) {
         UpdateAccumulator(pos);
         return true;
       }
+
       return false;
     }
 
     // Convert input features
-    void Transform(const Position& pos, OutputType* output, bool refresh) const {
-      if (refresh || !UpdateAccumulatorIfPossible(pos)) {
+    void Transform(const Position& pos, OutputType* output) const {
+
+      if (!UpdateAccumulatorIfPossible(pos))
         RefreshAccumulator(pos);
-      }
+
       const auto& accumulation = pos.state()->accumulator.accumulation;
 
   #if defined(USE_AVX2)
@@ -177,6 +182,7 @@ namespace Eval::NNUE {
    private:
     // Calculate cumulative value without using difference calculation
     void RefreshAccumulator(const Position& pos) const {
+
       auto& accumulator = pos.state()->accumulator;
       IndexType i = 0;
       Features::IndexList active_indices[2];
@@ -216,9 +222,8 @@ namespace Eval::NNUE {
               &accumulator.accumulation[perspective][i][0]);
           auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j) {
+          for (IndexType j = 0; j < kNumChunks; ++j)
             accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
-          }
 
   #elif defined(USE_NEON)
           auto accumulation = reinterpret_cast<int16x8_t*>(
@@ -240,11 +245,11 @@ namespace Eval::NNUE {
   #endif
 
       accumulator.computed_accumulation = true;
-      accumulator.computed_score = false;
     }
 
     // Calculate cumulative value using difference calculation
     void UpdateAccumulator(const Position& pos) const {
+
       const auto prev_accumulator = pos.state()->previous->accumulator;
       auto& accumulator = pos.state()->accumulator;
       IndexType i = 0;
@@ -288,33 +293,27 @@ namespace Eval::NNUE {
 
   #if defined(USE_AVX2)
             auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = _mm256_sub_epi16(accumulation[j], column[j]);
-            }
 
   #elif defined(USE_SSE2)
             auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = _mm_sub_epi16(accumulation[j], column[j]);
-            }
 
   #elif defined(USE_MMX)
             auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = _mm_sub_pi16(accumulation[j], column[j]);
-            }
 
   #elif defined(USE_NEON)
             auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = vsubq_s16(accumulation[j], column[j]);
-            }
 
   #else
-            for (IndexType j = 0; j < kHalfDimensions; ++j) {
-              accumulator.accumulation[perspective][i][j] -=
-                  weights_[offset + j];
-            }
+            for (IndexType j = 0; j < kHalfDimensions; ++j)
+              accumulator.accumulation[perspective][i][j] -= weights_[offset + j];
   #endif
 
           }
@@ -325,33 +324,27 @@ namespace Eval::NNUE {
 
   #if defined(USE_AVX2)
             auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = _mm256_add_epi16(accumulation[j], column[j]);
-            }
 
   #elif defined(USE_SSE2)
             auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
-            }
 
   #elif defined(USE_MMX)
             auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
-            }
 
   #elif defined(USE_NEON)
             auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
+            for (IndexType j = 0; j < kNumChunks; ++j)
               accumulation[j] = vaddq_s16(accumulation[j], column[j]);
-            }
 
   #else
-            for (IndexType j = 0; j < kHalfDimensions; ++j) {
-              accumulator.accumulation[perspective][i][j] +=
-                  weights_[offset + j];
-            }
+            for (IndexType j = 0; j < kHalfDimensions; ++j)
+              accumulator.accumulation[perspective][i][j] += weights_[offset + j];
   #endif
 
           }
@@ -362,7 +355,6 @@ namespace Eval::NNUE {
   #endif
 
       accumulator.computed_accumulation = true;
-      accumulator.computed_score = false;
     }
 
     using BiasType = std::int16_t;
diff --git a/src/position.cpp b/src/position.cpp
index fe89b753..e6a760d2 100644
--- a/src/position.cpp
+++ b/src/position.cpp
@@ -704,7 +704,6 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
 
   // Used by NNUE
   st->accumulator.computed_accumulation = false;
-  st->accumulator.computed_score = false;
   auto& dp = st->dirtyPiece;
   dp.dirty_num = 1;
 
@@ -1000,7 +999,6 @@ void Position::do_null_move(StateInfo& newSt) {
   if (Eval::useNNUE)
   {
       std::memcpy(&newSt, st, sizeof(StateInfo));
-      st->accumulator.computed_score = false;
   }
   else
       std::memcpy(&newSt, st, offsetof(StateInfo, accumulator));

From d2562cde12cdcc3df654279d6d632ae74c5f71af Mon Sep 17 00:00:00 2001
From: Gian-Carlo Pascutto <gcp@sjeng.org>
Date: Tue, 8 Sep 2020 15:37:53 +0200
Subject: [PATCH 43/52] Always re-enable NNUE after "bench".

Restore the default NNUE setting (enabled) after a bench command.
This also makes the resulting program settings independent of the
number of FENs that are being benched.

Fixes issue #3112.

closes https://github.com/official-stockfish/Stockfish/pull/3113

No functional change.
---
 src/benchmark.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/benchmark.cpp b/src/benchmark.cpp
index 806e9840..ffb631a2 100644
--- a/src/benchmark.cpp
+++ b/src/benchmark.cpp
@@ -164,5 +164,7 @@ vector<string> setup_bench(const Position& current, istream& is) {
           ++posCounter;
       }
 
+  list.emplace_back("setoption name Use NNUE value true");
+
   return list;
 }

From 0405f3540366cc16245d51531881c55d3726c8b5 Mon Sep 17 00:00:00 2001
From: SFisGOD <jonathandumale@gmail.com>
Date: Mon, 7 Sep 2020 04:54:26 +0800
Subject: [PATCH 44/52] Double probability of using classical eval

This patch doubles the moderate imbalance threshold and probability of using classical eval.
So now if imbalance is greater than PawnValueMg / 4 then there is a 1/8 chance of using classical eval.

STC:
LLR: 2.93 (-2.94,2.94) {-0.25,1.25}
Total: 10984 W: 1303 L: 1140 D: 8541
Ptnml(0-2): 58, 867, 3489, 1010, 68
https://tests.stockfishchess.org/tests/view/5f554c9f97da2d5437d3813e

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 43064 W: 2476 L: 2276 D: 38312
Ptnml(0-2): 37, 1985, 17308, 2145, 57
https://tests.stockfishchess.org/tests/view/5f55690a00a0aa2ca79f0a43

closes https://github.com/official-stockfish/Stockfish/pull/3114

Bench: 4161067
---
 src/evaluate.cpp | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/src/evaluate.cpp b/src/evaluate.cpp
index db8379da..faf71d27 100644
--- a/src/evaluate.cpp
+++ b/src/evaluate.cpp
@@ -1015,10 +1015,13 @@ make_v:
 
 Value Eval::evaluate(const Position& pos) {
 
+  // Use classical eval if there is a large imbalance
+  // If there is a moderate imbalance, use classical eval with probability (1/8),
+  // as derived from the node counter.
   bool useClassical = abs(eg_value(pos.psq_score())) * 16 > NNUEThreshold1 * (16 + pos.rule50_count());
   bool classical = !Eval::useNNUE
                 ||  useClassical
-                || (abs(eg_value(pos.psq_score())) > PawnValueMg / 8 && !(pos.this_thread()->nodes & 0xF));
+                || (abs(eg_value(pos.psq_score())) > PawnValueMg / 4 && !(pos.this_thread()->nodes & 0xB));
   Value v = classical ? Evaluation<NO_TRACE>(pos).value()
                       : NNUE::evaluate(pos) * 5 / 4 + Tempo;
 

From 35ab8254b70f62a4e0138c475fad0c77dcc0af2d Mon Sep 17 00:00:00 2001
From: mckx00 <mckx00@gmail.com>
Date: Sun, 13 Sep 2020 19:28:32 -0700
Subject: [PATCH 45/52] Simplify StatSCore Initialization

No need to initialize StatScore at rootNode. Current Logic is redundant because at subsequent levels the grandchildren statScore is initialized to zero.

closes https://github.com/official-stockfish/Stockfish/pull/3122

Non functional change.
---
 src/search.cpp | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index 4aeadc28..07c491b6 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -654,9 +654,7 @@ namespace {
     // starts with statScore = 0. Later grandchildren start with the last calculated
     // statScore of the previous grandchild. This influences the reduction rules in
     // LMR which are based on the statScore of parent position.
-    if (rootNode)
-        (ss+4)->statScore = 0;
-    else
+    if (!rootNode)
         (ss+2)->statScore = 0;
 
     // Step 4. Transposition table lookup. We don't want the score of a partial

From 7135678f71b7f6ee32e92b8dbef2b16b403d8ea9 Mon Sep 17 00:00:00 2001
From: Sergio Vieri <sergio.vieri.hp@gmail.com>
Date: Mon, 14 Sep 2020 17:24:05 +0800
Subject: [PATCH 46/52] Update default net to nn-03744f8d56d8.nnue

Equivalent to 20200914-1520

closes https://github.com/official-stockfish/Stockfish/pull/3123

Bench: 4222126
---
 src/evaluate.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/evaluate.h b/src/evaluate.h
index 3da6a9fe..c723bd8f 100644
--- a/src/evaluate.h
+++ b/src/evaluate.h
@@ -38,7 +38,7 @@ namespace Eval {
   // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
   // for the build process (profile-build and fishtest) to work. Do not change the
   // name of the macro, as it is used in the Makefile.
-  #define EvalFileDefaultName   "nn-308d71810dff.nnue"
+  #define EvalFileDefaultName   "nn-03744f8d56d8.nnue"
 
   namespace NNUE {
 

From 5f426d8667feda65eaf1eca699f629d31e170d43 Mon Sep 17 00:00:00 2001
From: xoto10 <me@example.com>
Date: Thu, 10 Sep 2020 21:10:57 +0100
Subject: [PATCH 47/52] Use 2 * bestMoveChanges.

NNUE appears to provide a more stable eval than the classic eval,
so the time use dependencies on bestMoveChanges, fallingEval,
etc may need to change to make the best use of available time.
This change doubles the effect of totBestMoveChanges when giving
more time because the choice of best move is unstable.

STC:
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 101928 W: 11995 L: 11698 D: 78235 Elo +0.78
Ptnml(0-2): 592, 8707, 32103, 8936, 626
https://tests.stockfishchess.org/tests/view/5f538a462d02727c56b36cec

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 186392 W: 10383 L: 9877 D: 166132 Elo +0.81
Ptnml(0-2): 207, 8370, 75539, 8870, 210
https://tests.stockfishchess.org/tests/view/5f54a9712d02727c56b36d5a

closes https://github.com/official-stockfish/Stockfish/pull/3119

Bench 4222126
---
 src/search.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/search.cpp b/src/search.cpp
index 07c491b6..c7d2efd4 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -520,7 +520,7 @@ void Thread::search() {
               totBestMoveChanges += th->bestMoveChanges;
               th->bestMoveChanges = 0;
           }
-          double bestMoveInstability = 1 + totBestMoveChanges / Threads.size();
+          double bestMoveInstability = 1 + 2 * totBestMoveChanges / Threads.size();
 
           double totalTime = rootMoves.size() == 1 ? 0 :
                              Time.optimum() * fallingEval * reduction * bestMoveInstability;

From d86663af141f1256bfc32ab95891e944d84e8755 Mon Sep 17 00:00:00 2001
From: syzygy1 <3028851+syzygy1@users.noreply.github.com>
Date: Sun, 13 Sep 2020 20:16:52 +0200
Subject: [PATCH 48/52] Improve NDK section in Makefile

This PR sets the "comp" variable simply to "clang",
which seems to be more consistent and allows a small simplification.

The PR also moves the section that sets "profile_make" and "profile_use" to after the NDK section,
which ensures that these variables are now set correctly for NDK/clang.

closes https://github.com/official-stockfish/Stockfish/pull/3121

No functional change
---
 src/Makefile | 32 +++++++++++++-------------------
 1 file changed, 13 insertions(+), 19 deletions(-)

diff --git a/src/Makefile b/src/Makefile
index 340b3008..54868b39 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -381,19 +381,6 @@ ifeq ($(COMP),clang)
 	endif
 endif
 
-ifeq ($(comp),icc)
-	profile_make = icc-profile-make
-	profile_use = icc-profile-use
-else
-ifeq ($(comp),clang)
-	profile_make = clang-profile-make
-	profile_use = clang-profile-use
-else
-	profile_make = gcc-profile-make
-	profile_use = gcc-profile-use
-endif
-endif
-
 ifeq ($(KERNEL),Darwin)
 	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.14
 	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.14
@@ -405,20 +392,30 @@ endif
 # Currently we don't know how to make PGO builds with the NDK yet.
 ifeq ($(COMP),ndk)
 	CXXFLAGS += -stdlib=libc++ -fPIE
+	comp=clang
 	ifeq ($(arch),armv7)
-		comp=armv7a-linux-androideabi16-clang
 		CXX=armv7a-linux-androideabi16-clang++
 		CXXFLAGS += -mthumb -march=armv7-a -mfloat-abi=softfp -mfpu=neon
 		STRIP=arm-linux-androideabi-strip
 	endif
 	ifeq ($(arch),armv8)
-		comp=aarch64-linux-android21-clang
 		CXX=aarch64-linux-android21-clang++
 		STRIP=aarch64-linux-android-strip
 	endif
 	LDFLAGS += -static-libstdc++ -pie -lm -latomic
 endif
 
+ifeq ($(comp),icc)
+	profile_make = icc-profile-make
+	profile_use = icc-profile-use
+else ifeq ($(comp),clang)
+	profile_make = clang-profile-make
+	profile_use = clang-profile-use
+else
+	profile_make = gcc-profile-make
+	profile_use = gcc-profile-use
+endif
+
 ### Travis CI script uses COMPILER to overwrite CXX
 ifdef COMPILER
 	COMPCXX=$(COMPILER)
@@ -590,10 +587,7 @@ endif
 ### needs access to the optimization flags.
 ifeq ($(optimize),yes)
 ifeq ($(debug), no)
-	ifeq ($(COMP),ndk)
-		CXXFLAGS += -flto=thin
-		LDFLAGS += $(CXXFLAGS)
-	else ifeq ($(comp),clang)
+	ifeq ($(comp),clang)
 		CXXFLAGS += -flto=thin
 		ifneq ($(findstring MINGW,$(KERNEL)),)
 			CXXFLAGS += -fuse-ld=lld

From df43805953b241f95c246ff3e96aece76b518590 Mon Sep 17 00:00:00 2001
From: GoldenRare <deshawnmohan@hotmail.com>
Date: Thu, 10 Sep 2020 00:24:40 -0400
Subject: [PATCH 49/52] Added FEN string to bench output

fixes https://github.com/official-stockfish/Stockfish/pull/3117

closes https://github.com/official-stockfish/Stockfish/pull/3118

No functional change
---
 AUTHORS     | 1 +
 src/uci.cpp | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/AUTHORS b/AUTHORS
index c00ab657..198dfa5a 100644
--- a/AUTHORS
+++ b/AUTHORS
@@ -63,6 +63,7 @@ Gary Heckman (gheckman)
 George Sobala (gsobala)
 gguliash
 Gian-Carlo Pascutto (gcp)
+Deshawn Mohan-Smith (GoldenRare)
 Gontran Lemaire (gonlem)
 Goodkov Vasiliy Aleksandrovich (goodkov)
 Gregor Cramer
diff --git a/src/uci.cpp b/src/uci.cpp
index bc0ee0a0..3f3cc458 100644
--- a/src/uci.cpp
+++ b/src/uci.cpp
@@ -170,7 +170,7 @@ namespace {
 
         if (token == "go" || token == "eval")
         {
-            cerr << "\nPosition: " << cnt++ << '/' << num << endl;
+            cerr << "\nPosition: " << cnt++ << '/' << num << " (" << pos.fen() << ")" << endl;
             if (token == "go")
             {
                go(pos, is, states);

From 0ca93c5b94b820a41e2850ede084096120128a28 Mon Sep 17 00:00:00 2001
From: Unai Corzo <corzounai@gmail.com>
Date: Wed, 16 Sep 2020 19:14:32 +0200
Subject: [PATCH 50/52] Remove castling extension

STC https://tests.stockfishchess.org/tests/view/5f5fa5348fbc1c8a3f476eca
LLR: 2.94 (-2.94,2.94) {-1.25,0.25}
Total: 38520 W: 4713 L: 4610 D: 29197
Ptnml(0-2): 233, 3486, 11734, 3559, 248

LTC https://tests.stockfishchess.org/tests/view/5f62166a912c15f19854b806
LLR: 2.93 (-2.94,2.94) {-0.75,0.25}
Total: 48024 W: 2673 L: 2600 D: 42751
Ptnml(0-2): 64, 2247, 19316, 2322, 63

closes https://github.com/official-stockfish/Stockfish/pull/3128

bench: 3818400
---
 src/search.cpp | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/src/search.cpp b/src/search.cpp
index c7d2efd4..17cd0a73 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1127,11 +1127,6 @@ moves_loop: // When in check, search starts from here
                && pos.non_pawn_material() <= 2 * RookValueMg)
           extension = 1;
 
-      // Castling extension
-      if (   type_of(move) == CASTLING
-          && popcount(pos.pieces(us) & ~pos.pieces(PAWN) & (to_sq(move) & KingSide ? KingSide : QueenSide)) <= 2)
-          extension = 1;
-
       // Late irreversible move extension
       if (   move == ttMove
           && pos.rule50_count() > 80

From 64a63464d7bc72a3aac33aa680cd2b2b240ff903 Mon Sep 17 00:00:00 2001
From: Unai Corzo <corzounai@gmail.com>
Date: Wed, 16 Sep 2020 20:42:38 +0200
Subject: [PATCH 51/52] Simplify futility pruning for captures

STC https://tests.stockfishchess.org/tests/view/5f61f0e4b91f2ec371e429c2
LLR: 2.94 (-2.94,2.94) {-1.25,0.25}
Total: 75512 W: 8747 L: 8704 D: 58061
Ptnml(0-2): 440, 6589, 23683, 6576, 468

LTC https://tests.stockfishchess.org/tests/view/5f6215d3912c15f19854b801
LLR: 2.95 (-2.94,2.94) {-0.75,0.25}
Total: 92912 W: 5030 L: 4992 D: 82890
Ptnml(0-2): 88, 4363, 37532, 4369, 104

closes https://github.com/official-stockfish/Stockfish/pull/3129

bench: 3856086
---
 src/search.cpp | 1 -
 1 file changed, 1 deletion(-)

diff --git a/src/search.cpp b/src/search.cpp
index 17cd0a73..9c5fb58b 100644
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -1056,7 +1056,6 @@ moves_loop: // When in check, search starts from here
               if (   !givesCheck
                   && lmrDepth < 6
                   && !(PvNode && abs(bestValue) < 2)
-                  && PieceValue[MG][type_of(movedPiece)] >= PieceValue[MG][type_of(pos.piece_on(to_sq(move)))]
                   && !ss->inCheck
                   && ss->staticEval + 169 + 244 * lmrDepth
                      + PieceValue[MG][type_of(pos.piece_on(to_sq(move)))] <= alpha)

From 8b8a510fd6a1a17b39b2d4b166f60ac7be0dab23 Mon Sep 17 00:00:00 2001
From: syzygy1 <3028851+syzygy1@users.noreply.github.com>
Date: Wed, 16 Sep 2020 17:39:11 +0200
Subject: [PATCH 52/52] Use tiling to speed up accumulator refreshes and
 updates

Perform the update and refresh operations tile by tile in a local
array of vectors. By selecting the array size carefully, we
achieve that the compiler keeps the whole array in vector registers.

Idea and original implementation by @sf-x.

STC: https://tests.stockfishchess.org/tests/view/5f623eec912c15f19854b855
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 4872 W: 623 L: 477 D: 3772
Ptnml(0-2): 14, 350, 1585, 450, 37

LTC: https://tests.stockfishchess.org/tests/view/5f62434e912c15f19854b860
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 25808 W: 1565 L: 1401 D: 22842
Ptnml(0-2): 23, 1186, 10332, 1330, 33

closes https://github.com/official-stockfish/Stockfish/pull/3130

No functional change
---
 src/nnue/nnue_feature_transformer.h | 233 +++++++++++++++-------------
 1 file changed, 125 insertions(+), 108 deletions(-)

diff --git a/src/nnue/nnue_feature_transformer.h b/src/nnue/nnue_feature_transformer.h
index 2b6259c3..e71ee60d 100644
--- a/src/nnue/nnue_feature_transformer.h
+++ b/src/nnue/nnue_feature_transformer.h
@@ -29,6 +29,56 @@
 
 namespace Eval::NNUE {
 
+  // If vector instructions are enabled, we update and refresh the
+  // accumulator tile by tile such that each tile fits in the CPU's
+  // vector registers.
+  #define TILING
+
+  #ifdef USE_AVX512
+  typedef __m512i vec_t;
+  #define vec_load(a) _mm512_loadA_si512(a)
+  #define vec_store(a,b) _mm512_storeA_si512(a,b)
+  #define vec_add_16(a,b) _mm512_add_epi16(a,b)
+  #define vec_sub_16(a,b) _mm512_sub_epi16(a,b)
+  static constexpr IndexType kNumRegs = 8; // only 8 are needed
+
+  #elif USE_AVX2
+  typedef __m256i vec_t;
+  #define vec_load(a) _mm256_loadA_si256(a)
+  #define vec_store(a,b) _mm256_storeA_si256(a,b)
+  #define vec_add_16(a,b) _mm256_add_epi16(a,b)
+  #define vec_sub_16(a,b) _mm256_sub_epi16(a,b)
+  static constexpr IndexType kNumRegs = 16;
+
+  #elif USE_SSE2
+  typedef __m128i vec_t;
+  #define vec_load(a) (*(a))
+  #define vec_store(a,b) *(a)=(b)
+  #define vec_add_16(a,b) _mm_add_epi16(a,b)
+  #define vec_sub_16(a,b) _mm_sub_epi16(a,b)
+  static constexpr IndexType kNumRegs = Is64Bit ? 16 : 8;
+
+  #elif USE_MMX
+  typedef __m64 vec_t;
+  #define vec_load(a) (*(a))
+  #define vec_store(a,b) *(a)=(b)
+  #define vec_add_16(a,b) _mm_add_pi16(a,b)
+  #define vec_sub_16(a,b) _mm_sub_pi16(a,b)
+  static constexpr IndexType kNumRegs = 8;
+
+  #elif USE_NEON
+  typedef int16x8_t vec_t;
+  #define vec_load(a) (*(a))
+  #define vec_store(a,b) *(a)=(b)
+  #define vec_add_16(a,b) vaddq_s16(a,b)
+  #define vec_sub_16(a,b) vsubq_s16(a,b)
+  static constexpr IndexType kNumRegs = 16;
+
+  #else
+  #undef TILING
+
+  #endif
+
   // Input feature converter
   class FeatureTransformer {
 
@@ -36,6 +86,11 @@ namespace Eval::NNUE {
     // Number of output dimensions for one side
     static constexpr IndexType kHalfDimensions = kTransformedFeatureDimensions;
 
+    #ifdef TILING
+    static constexpr IndexType kTileHeight = kNumRegs * sizeof(vec_t) / 2;
+    static_assert(kHalfDimensions % kTileHeight == 0, "kTileHeight must divide kHalfDimensions");
+    #endif
+
    public:
     // Output type
     using OutputType = TransformedFeatureType;
@@ -189,57 +244,41 @@ namespace Eval::NNUE {
       RawFeatures::AppendActiveIndices(pos, kRefreshTriggers[i],
                                        active_indices);
       for (Color perspective : { WHITE, BLACK }) {
+  #ifdef TILING
+        for (unsigned j = 0; j < kHalfDimensions / kTileHeight; ++j) {
+          auto biasesTile = reinterpret_cast<const vec_t*>(
+              &biases_[j * kTileHeight]);
+          auto accTile = reinterpret_cast<vec_t*>(
+              &accumulator.accumulation[perspective][i][j * kTileHeight]);
+          vec_t acc[kNumRegs];
+
+          for (unsigned k = 0; k < kNumRegs; ++k)
+            acc[k] = biasesTile[k];
+
+          for (const auto index : active_indices[perspective]) {
+            const IndexType offset = kHalfDimensions * index + j * kTileHeight;
+            auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
+
+            for (unsigned k = 0; k < kNumRegs; ++k)
+              acc[k] = vec_add_16(acc[k], column[k]);
+          }
+
+          for (unsigned k = 0; k < kNumRegs; k++)
+            vec_store(&accTile[k], acc[k]);
+        }
+  #else
         std::memcpy(accumulator.accumulation[perspective][i], biases_,
-                   kHalfDimensions * sizeof(BiasType));
+            kHalfDimensions * sizeof(BiasType));
+
         for (const auto index : active_indices[perspective]) {
           const IndexType offset = kHalfDimensions * index;
-  #if defined(USE_AVX512)
-          auto accumulation = reinterpret_cast<__m512i*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m512i*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            _mm512_storeA_si512(&accumulation[j], _mm512_add_epi16(_mm512_loadA_si512(&accumulation[j]), column[j]));
 
-  #elif defined(USE_AVX2)
-          auto accumulation = reinterpret_cast<__m256i*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            _mm256_storeA_si256(&accumulation[j], _mm256_add_epi16(_mm256_loadA_si256(&accumulation[j]), column[j]));
-
-  #elif defined(USE_SSE2)
-          auto accumulation = reinterpret_cast<__m128i*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
-
-  #elif defined(USE_MMX)
-          auto accumulation = reinterpret_cast<__m64*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
-
-  #elif defined(USE_NEON)
-          auto accumulation = reinterpret_cast<int16x8_t*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            accumulation[j] = vaddq_s16(accumulation[j], column[j]);
-
-  #else
           for (IndexType j = 0; j < kHalfDimensions; ++j)
             accumulator.accumulation[perspective][i][j] += weights_[offset + j];
-  #endif
-
         }
+  #endif
       }
+
   #if defined(USE_MMX)
       _mm_empty();
   #endif
@@ -257,29 +296,55 @@ namespace Eval::NNUE {
       bool reset[2];
       RawFeatures::AppendChangedIndices(pos, kRefreshTriggers[i],
                                         removed_indices, added_indices, reset);
-      for (Color perspective : { WHITE, BLACK }) {
 
-  #if defined(USE_AVX2)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<__m256i*>(
-            &accumulator.accumulation[perspective][i][0]);
+  #ifdef TILING
+      for (IndexType j = 0; j < kHalfDimensions / kTileHeight; ++j) {
+        for (Color perspective : { WHITE, BLACK }) {
+          auto accTile = reinterpret_cast<vec_t*>(
+              &accumulator.accumulation[perspective][i][j * kTileHeight]);
+          vec_t acc[kNumRegs];
 
-  #elif defined(USE_SSE2)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<__m128i*>(
-            &accumulator.accumulation[perspective][i][0]);
+          if (reset[perspective]) {
+            auto biasesTile = reinterpret_cast<const vec_t*>(
+                &biases_[j * kTileHeight]);
+            for (unsigned k = 0; k < kNumRegs; ++k)
+              acc[k] = biasesTile[k];
+          } else {
+            auto prevAccTile = reinterpret_cast<const vec_t*>(
+                &prev_accumulator.accumulation[perspective][i][j * kTileHeight]);
+            for (IndexType k = 0; k < kNumRegs; ++k)
+              acc[k] = vec_load(&prevAccTile[k]);
 
-  #elif defined(USE_MMX)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<__m64*>(
-            &accumulator.accumulation[perspective][i][0]);
+            // Difference calculation for the deactivated features
+            for (const auto index : removed_indices[perspective]) {
+              const IndexType offset = kHalfDimensions * index + j * kTileHeight;
+              auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
 
-  #elif defined(USE_NEON)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<int16x8_t*>(
-            &accumulator.accumulation[perspective][i][0]);
+              for (IndexType k = 0; k < kNumRegs; ++k)
+                acc[k] = vec_sub_16(acc[k], column[k]);
+            }
+          }
+          { // Difference calculation for the activated features
+            for (const auto index : added_indices[perspective]) {
+              const IndexType offset = kHalfDimensions * index + j * kTileHeight;
+              auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
+
+              for (IndexType k = 0; k < kNumRegs; ++k)
+                acc[k] = vec_add_16(acc[k], column[k]);
+            }
+          }
+
+          for (IndexType k = 0; k < kNumRegs; ++k)
+            vec_store(&accTile[k], acc[k]);
+        }
+      }
+  #if defined(USE_MMX)
+      _mm_empty();
   #endif
 
+  #else
+      for (Color perspective : { WHITE, BLACK }) {
+
         if (reset[perspective]) {
           std::memcpy(accumulator.accumulation[perspective][i], biases_,
                       kHalfDimensions * sizeof(BiasType));
@@ -291,67 +356,19 @@ namespace Eval::NNUE {
           for (const auto index : removed_indices[perspective]) {
             const IndexType offset = kHalfDimensions * index;
 
-  #if defined(USE_AVX2)
-            auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = _mm256_sub_epi16(accumulation[j], column[j]);
-
-  #elif defined(USE_SSE2)
-            auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = _mm_sub_epi16(accumulation[j], column[j]);
-
-  #elif defined(USE_MMX)
-            auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = _mm_sub_pi16(accumulation[j], column[j]);
-
-  #elif defined(USE_NEON)
-            auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = vsubq_s16(accumulation[j], column[j]);
-
-  #else
             for (IndexType j = 0; j < kHalfDimensions; ++j)
               accumulator.accumulation[perspective][i][j] -= weights_[offset + j];
-  #endif
-
           }
         }
         { // Difference calculation for the activated features
           for (const auto index : added_indices[perspective]) {
             const IndexType offset = kHalfDimensions * index;
 
-  #if defined(USE_AVX2)
-            auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = _mm256_add_epi16(accumulation[j], column[j]);
-
-  #elif defined(USE_SSE2)
-            auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
-
-  #elif defined(USE_MMX)
-            auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
-
-  #elif defined(USE_NEON)
-            auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j)
-              accumulation[j] = vaddq_s16(accumulation[j], column[j]);
-
-  #else
             for (IndexType j = 0; j < kHalfDimensions; ++j)
               accumulator.accumulation[perspective][i][j] += weights_[offset + j];
-  #endif
-
           }
         }
       }
-  #if defined(USE_MMX)
-      _mm_empty();
   #endif
 
       accumulator.computed_accumulation = true;