Simplify lookup_count and clean up pieces().

[stockfish] / src / nnue / layers / affine_transform_sparse_input.h

diff --git a/src/nnue/layers/affine_transform_sparse_input.h b/src/nnue/layers/affine_transform_sparse_input.h
index 00b17c19f129b868050492371a8752c54efdf5ce..18c166cd9d7897e8c221a8eafb61ce68c1c52ad4 100644 (file)
--- a/src/nnue/layers/affine_transform_sparse_input.h
+++ b/src/nnue/layers/affine_transform_sparse_input.h
@@ -24,6 +24,7 @@
 #include <iostream>
 #include <algorithm>
 #include <array>
+#include <bitset>
 #include <type_traits>
 #include "../nnue_common.h"
 #include "affine_transform.h"
@@ -77,16 +78,7 @@ namespace Stockfish::Eval::NNUE::Layers {
   alignas(CacheLineSize) static inline const std::array<unsigned, 256> lookup_count = [](){
     std::array<unsigned, 256> v;
     for (int i = 0; i < 256; ++i)
-    {
-      int j = i;
-      int k = 0;
-      while(j)
-      {
-        j &= j - 1;
-        ++k;
-      }
-      v[i] = k;
-    }
+      v[i] = unsigned(std::bitset<8>(i).count());
     return v;
   }();
 
@@ -98,10 +90,10 @@ namespace Stockfish::Eval::NNUE::Layers {
     #define vec_nnz(a) _mm512_cmpgt_epi32_mask(a, _mm512_setzero_si512())
 #elif defined (USE_AVX2)
     using vec_t = __m256i;
-    #define vec_nnz(a) _mm256_movemask_ps((__m256)_mm256_cmpgt_epi32(a, _mm256_setzero_si256()))
+    #define vec_nnz(a) _mm256_movemask_ps(_mm256_castsi256_ps(_mm256_cmpgt_epi32(a, _mm256_setzero_si256())))
 #elif defined (USE_SSSE3)
     using vec_t = __m128i;
-    #define vec_nnz(a) _mm_movemask_ps((__m128)_mm_cmpgt_epi32(a, _mm_setzero_si128()))
+    #define vec_nnz(a) _mm_movemask_ps(_mm_castsi128_ps(_mm_cmpgt_epi32(a, _mm_setzero_si128())))
 #endif
     constexpr IndexType InputSimdWidth = sizeof(vec_t) / sizeof(std::int32_t);
     // Inputs are processed InputSimdWidth at a time and outputs are processed 8 at a time so we process in chunks of max(InputSimdWidth, 8)