Merge pull request #23 from winapiadmin/pr_forceinline

Uses FORCEINLINE for critical-performance functions, and optimized pawn move generation

Author: winapiadmin

.github/workflows/cmake-multi-platform.yml

@@ -61,11 +61,7 @@ jobs:
       run: cmake --build ${{ steps.vars.outputs.dir }} --config ${{ matrix.build_type }}
 
     - name: Test
-      working-directory: ${{ steps.vars.outputs.dir }}
-      shell: bash
-      run: |
-        if [[ "${{ matrix.os }}" == "windows-latest" ]]; then
-          ctest --build-config ${{ matrix.build_type }} --verbose
-        else
-          ctest --verbose
-        fi
+      working-directory: ${{ steps.strings.outputs.build-output-dir }}
+      # Execute tests defined by the CMake configuration. Note that --build-config is needed because the default Windows generator is a multi-config generator (Visual Studio generator).
+      # See https://cmake.org/cmake/help/latest/manual/ctest.1.html for more detail
+      run: ctest --build-config ${{ matrix.build_type }} --verbose

.gitignore

@@ -9,4 +9,4 @@ build/
 !.gitignore
 CMakePresets.json
 CMakeSettings.json
-*.s
+*.s

CMakeLists.txt

@@ -5,7 +5,7 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
 # --- Compiler tuning ---
-set(ASAN_ENABLED OFF CACHE BOOL "Whether to enable ASan or not")
+set(SAN_ENABLED OFF CACHE BOOL "Whether to enable ASan or not")
 if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
   if (CMAKE_CXX_COMPILER_FRONTEND_VARIANT STREQUAL "MSVC")
     # clang-cl: forward GCC/Clang style constexpr flags via /clang:
@@ -24,24 +24,12 @@ if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
       -march=native -mtune=native
       -ftemplate-backtrace-limit=0
     )
-    if (ASAN_ENABLED)
-    add_compile_options(-fsanitize=address)
-    add_link_options(-fsanitize=address)
-    endif()
   endif()
 elseif (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
   add_compile_options(-fconstexpr-ops-limit=2000000000 -fconstexpr-depth=1024 -march=native -mtune=native -ftemplate-backtrace-limit=0)
-  if (ASAN_ENABLED)
-    add_compile_options(-fsanitize=address)
-    add_link_options(-fsanitize=address)
-  endif()
 elseif (CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
   set(ARCH_FLAG "/arch:AVX2" CACHE STRING "MSVC architecture flag (/arch:SSE2, /arch:AVX, /arch:AVX2, /arch:AVX512)")
   add_compile_options(/constexpr:steps2000000000 /constexpr:depth1024 ${ARCH_FLAG})
-  if (ASAN_ENABLED)
-    add_compile_options(/fsanitize=address)
-    add_link_options(/fsanitize=address)
-  endif()
 endif()
 
 add_compile_definitions(GENERATE_AT_RUNTIME)

bitboard.h

@@ -4,7 +4,7 @@
 #if defined(_MSC_VER)
 #include <intrin.h>
 #endif
-
+#include <immintrin.h>
 namespace chess {
 // -------------------------------
 // constexpr fallbacks
@@ -43,7 +43,7 @@ constexpr int msb_constexpr(Bitboard x) noexcept {
 #if defined(__GNUG__) || defined(__clang__)
 [[gnu::const]]
 #endif
-inline constexpr int popcount(Bitboard x) noexcept {
+__FORCEINLINE constexpr int popcount(Bitboard x) noexcept {
 #if defined(__GNUG__) || defined(__clang__)
     if (!is_constant_evaluated())
         return __builtin_popcountll(x);
@@ -57,7 +57,7 @@ inline constexpr int popcount(Bitboard x) noexcept {
 #if defined(__GNUG__) || defined(__clang__)
 [[gnu::const]]
 #endif
-inline constexpr int lsb(Bitboard x) noexcept {
+__FORCEINLINE constexpr int lsb(Bitboard x) noexcept {
 #if defined(__GNUG__) || defined(__clang__)
     if (!is_constant_evaluated())
         return __builtin_ctzll(x);
@@ -74,7 +74,7 @@ inline constexpr int lsb(Bitboard x) noexcept {
 #if defined(__GNUG__) || defined(__clang__)
 [[gnu::const]]
 #endif
-inline constexpr int msb(Bitboard x) noexcept {
+__FORCEINLINE constexpr int msb(Bitboard x) noexcept {
 #if defined(__GNUG__) || defined(__clang__)
     if (!is_constant_evaluated())
         return 63 - __builtin_clzll(x);
@@ -91,13 +91,17 @@ inline constexpr int msb(Bitboard x) noexcept {
 // -------------------------------
 // destructive variants
 // -------------------------------
-inline int pop_lsb(Bitboard &b) noexcept {
+__FORCEINLINE int pop_lsb(Bitboard &b) noexcept {
     int c = lsb(b);
+#ifndef __BMI2__
     b &= b - 1;
+#else
+    b = _blsr_u64(b);
+#endif
     return c;
 }
 
-inline int pop_msb(Bitboard &b) noexcept {
+__FORCEINLINE int pop_msb(Bitboard &b) noexcept {
     int c = msb(b);
     b &= ~(1ULL << c);
     return c;

examples/perft_with_tt.cpp

@@ -113,18 +113,17 @@ template <int DEPTH, bool EnableDiv=false, typename T=EnginePiece> void benchmar
     double avgTime = totalTime / N_RUNS;
     double avgMnps = (nodes / avgTime) / 1'000'000.0;
 
-    std::cout << "Average: " << avgTime << " s, " << avgMnps << " Mnps\n";
+    std::cout << "Average: " << avgTime << " s, " << avgMnps << " Mnps (nodes="<<(nodes/N_RUNS)<<")\n";
 }
 
 
 int main() {
-    _Position<ContiguousMappingPiece> pos("8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w - - 0 1");
+    _Position<EnginePiece> pos;
 
 #ifdef TT_ENABLED
     tt.resize(1 << 28);
 #endif
-    benchmark<3, true, ContiguousMappingPiece>(pos);
-
+    benchmark<7, true, EnginePiece>(pos);
     // Movelist moves;
     // pos.template legals(moves);
     // for (const Move &m : moves) {