Remove unused return type from propagate()

[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..134b7d13e191ab8a148b4d458c24d9100540e877 100644 (file)
--- a/src/nnue/layers/affine_transform_sparse_input.h
+++ b/src/nnue/layers/affine_transform_sparse_input.h
@@ -34,58 +34,15 @@
 */
 
 namespace Stockfish::Eval::NNUE::Layers {
-#if defined(__GNUC__)  // GCC, Clang, ICC
-
-  static inline IndexType lsb_(std::uint32_t b) {
-    assert(b);
-    return IndexType(__builtin_ctzl(b));
-  }
-
-#elif defined(_MSC_VER)  // MSVC
-
-  static inline IndexType lsb_(std::uint32_t b) {
-    assert(b);
-    unsigned long idx;
-    _BitScanForward(&idx, b);
-    return (IndexType) idx;
-  }
-
-#else  // Compiler is neither GCC nor MSVC compatible
-
-#error "Compiler not supported."
-
-#endif
-
 
 #if defined(USE_SSSE3)
   alignas(CacheLineSize) static inline const std::array<std::array<std::uint16_t, 8>, 256> lookup_indices = [](){
     std::array<std::array<std::uint16_t, 8>, 256> v{};
-    for (int i = 0; i < 256; ++i)
+    for (unsigned i = 0; i < 256; ++i)
     {
-      int j = i;
-      int k = 0;
+      std::uint64_t j = i, k = 0;
       while(j)
-      {
-        const IndexType lsbIndex = lsb_(std::uint32_t(j));
-        j &= j - 1;
-        v[i][k] = lsbIndex;
-        ++k;
-      }
-    }
-    return v;
-  }();
-  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][k++] = pop_lsb(j);
     }
     return v;
   }();
@@ -98,10 +55,14 @@ 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()))
+    #if defined(USE_VNNI) && !defined(USE_AVXVNNI)
+        #define vec_nnz(a) _mm256_cmpgt_epi32_mask(a, _mm256_setzero_si256())
+    #else
+        #define vec_nnz(a) _mm256_movemask_ps(_mm256_castsi256_ps(_mm256_cmpgt_epi32(a, _mm256_setzero_si256())))
+    #endif
 #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)
@@ -112,8 +73,8 @@ namespace Stockfish::Eval::NNUE::Layers {
 
     const auto inputVector = reinterpret_cast<const vec_t*>(input);
     IndexType count = 0;
-    __m128i base = _mm_set1_epi16(0);
-    __m128i increment = _mm_set1_epi16(8);
+    __m128i base = _mm_setzero_si128();
+    const __m128i increment = _mm_set1_epi16(8);
     for (IndexType i = 0; i < NumChunks; ++i)
     {
       // bitmask of nonzero values in this chunk
@@ -128,7 +89,7 @@ namespace Stockfish::Eval::NNUE::Layers {
         const auto lookup = (nnz >> (j * 8)) & 0xFF;
         const auto offsets = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&lookup_indices[lookup]));
         _mm_storeu_si128(reinterpret_cast<__m128i*>(out + count), _mm_add_epi16(base, offsets));
-        count += lookup_count[lookup];
+        count += popcount(lookup);
         base = _mm_add_epi16(base, increment);
       }
     }
@@ -141,7 +102,6 @@ namespace Stockfish::Eval::NNUE::Layers {
   template <IndexType InDims, IndexType OutDims>
   class AffineTransformSparseInput {
    public:
-    // Input/output type
     // Input/output type
     using InputType = std::uint8_t;
     using OutputType = std::int32_t;
@@ -174,7 +134,7 @@ namespace Stockfish::Eval::NNUE::Layers {
       return hashValue;
     }
 
-    static IndexType get_weight_index_scrambled(IndexType i)
+    static constexpr IndexType get_weight_index_scrambled(IndexType i)
     {
       return
         (i / ChunkSize) % (PaddedInputDimensions / ChunkSize) * OutputDimensions * ChunkSize +
@@ -182,7 +142,7 @@ namespace Stockfish::Eval::NNUE::Layers {
         i % ChunkSize;
     }
 
-    static IndexType get_weight_index(IndexType i)
+    static constexpr IndexType get_weight_index(IndexType i)
     {
 #if defined (USE_SSSE3)
       return get_weight_index_scrambled(i);
@@ -210,7 +170,7 @@ namespace Stockfish::Eval::NNUE::Layers {
       return !stream.fail();
     }
     // Forward propagation
-    const OutputType* propagate(
+    void propagate(
         const InputType* input, OutputType* output) const {
 
 #if defined (USE_SSSE3)
@@ -269,8 +229,6 @@ namespace Stockfish::Eval::NNUE::Layers {
         PaddedInputDimensions,
         OutputDimensions>(output, weights, biases, input);
 #endif
-
-      return output;
     }
 
    private: