Remove handcrafted MMX code

too small a benefit to maintain this old target

closes https://github.com/official-stockfish/Stockfish/pull/4804

No functional change

src/Makefile

@@ -674,7 +674,6 @@ ifeq ($(sse2),yes)
 endif
 
 ifeq ($(mmx),yes)
-	CXXFLAGS += -DUSE_MMX
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw icx))
 		CXXFLAGS += -mmmx
 	endif
@@ -794,11 +793,11 @@ help:
 	@echo "x86-64-sse41-popcnt     > x86 64-bit with sse41 and popcnt support"
 	@echo "x86-64-modern           > deprecated, currently x86-64-sse41-popcnt"
 	@echo "x86-64-ssse3            > x86 64-bit with ssse3 support"
-	@echo "x86-64-sse3-popcnt      > x86 64-bit with sse3 and popcnt support"
+	@echo "x86-64-sse3-popcnt      > x86 64-bit with sse3 compile and popcnt support"
 	@echo "x86-64                  > x86 64-bit generic (with sse2 support)"
 	@echo "x86-32-sse41-popcnt     > x86 32-bit with sse41 and popcnt support"
 	@echo "x86-32-sse2             > x86 32-bit with sse2 support"
-	@echo "x86-32                  > x86 32-bit generic (with mmx and sse support)"
+	@echo "x86-32                  > x86 32-bit generic (with mmx compile support)"
 	@echo "ppc-64                  > PPC 64-bit"
 	@echo "ppc-32                  > PPC 32-bit"
 	@echo "armv7                   > ARMv7 32-bit"

src/misc.cpp

@@ -282,9 +282,6 @@ std::string compiler_info() {
     compiler += " SSE2";
   #endif
   compiler += (HasPopCnt ? " POPCNT" : "");
-  #if defined(USE_MMX)
-    compiler += " MMX";
-  #endif
   #if defined(USE_NEON_DOTPROD)
     compiler += " NEON_DOTPROD";
   #elif defined(USE_NEON)

src/nnue/layers/affine_transform.h

@@ -45,18 +45,13 @@ namespace Stockfish::Eval::NNUE::Layers {
   template <IndexType InputDimensions, IndexType PaddedInputDimensions, IndexType OutputDimensions>
   static void affine_transform_non_ssse3(std::int32_t* output, const std::int8_t* weights, const std::int32_t* biases, const std::uint8_t* input)
   {
-# if defined(USE_SSE2) || defined(USE_MMX) || defined(USE_NEON_DOTPROD) || defined(USE_NEON)
+# if defined(USE_SSE2) || defined(USE_NEON_DOTPROD) || defined(USE_NEON)
 # if defined(USE_SSE2)
     // At least a multiple of 16, with SSE2.
     constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
     const __m128i Zeros = _mm_setzero_si128();
     const auto inputVector = reinterpret_cast<const __m128i*>(input);
 
-# elif defined(USE_MMX)
-    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / 8;
-    const __m64 Zeros = _mm_setzero_si64();
-    const auto inputVector = reinterpret_cast<const __m64*>(input);
-
 # elif defined(USE_NEON_DOTPROD)
     constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
     const auto inputVector = reinterpret_cast<const int8x16_t*>(input);
@@ -92,26 +87,6 @@ namespace Stockfish::Eval::NNUE::Layers {
       sum = _mm_add_epi32(sum, sum_second_32);
       output[i] = _mm_cvtsi128_si32(sum);
 
-# elif defined(USE_MMX)
-      __m64 sumLo = _mm_cvtsi32_si64(biases[i]);
-      __m64 sumHi = Zeros;
-      const auto row = reinterpret_cast<const __m64*>(&weights[offset]);
-      for (IndexType j = 0; j < NumChunks; ++j) {
-        __m64 row_j = row[j];
-        __m64 input_j = inputVector[j];
-        __m64 extendedRowLo = _mm_srai_pi16(_mm_unpacklo_pi8(row_j, row_j), 8);
-        __m64 extendedRowHi = _mm_srai_pi16(_mm_unpackhi_pi8(row_j, row_j), 8);
-        __m64 extendedInputLo = _mm_unpacklo_pi8(input_j, Zeros);
-        __m64 extendedInputHi = _mm_unpackhi_pi8(input_j, Zeros);
-        __m64 productLo = _mm_madd_pi16(extendedRowLo, extendedInputLo);
-        __m64 productHi = _mm_madd_pi16(extendedRowHi, extendedInputHi);
-        sumLo = _mm_add_pi32(sumLo, productLo);
-        sumHi = _mm_add_pi32(sumHi, productHi);
-      }
-      __m64 sum = _mm_add_pi32(sumLo, sumHi);
-      sum = _mm_add_pi32(sum, _mm_unpackhi_pi32(sum, sum));
-      output[i] = _mm_cvtsi64_si32(sum);
-
 # elif defined(USE_NEON_DOTPROD)
       int32x4_t sum = {biases[i]};
       const auto row = reinterpret_cast<const int8x16_t*>(&weights[offset]);
@@ -132,11 +107,6 @@ namespace Stockfish::Eval::NNUE::Layers {
 
 # endif
     }
-
-# if defined(USE_MMX)
-    _mm_empty();
-# endif
-
 # else
   std::memcpy(output, biases, sizeof(std::int32_t) * OutputDimensions);
 

src/nnue/layers/clipped_relu.h

@@ -135,24 +135,6 @@ namespace Stockfish::Eval::NNUE::Layers {
       }
       constexpr IndexType Start = NumChunks * SimdWidth;
 
-  #elif defined(USE_MMX)
-      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
-      const __m64 k0x80s = _mm_set1_pi8(-128);
-      const auto in = reinterpret_cast<const __m64*>(input);
-      const auto out = reinterpret_cast<__m64*>(output);
-      for (IndexType i = 0; i < NumChunks; ++i) {
-        const __m64 words0 = _mm_srai_pi16(
-            _mm_packs_pi32(in[i * 4 + 0], in[i * 4 + 1]),
-            WeightScaleBits);
-        const __m64 words1 = _mm_srai_pi16(
-            _mm_packs_pi32(in[i * 4 + 2], in[i * 4 + 3]),
-            WeightScaleBits);
-        const __m64 packedbytes = _mm_packs_pi16(words0, words1);
-        out[i] = _mm_subs_pi8(_mm_adds_pi8(packedbytes, k0x80s), k0x80s);
-      }
-      _mm_empty();
-      constexpr IndexType Start = NumChunks * SimdWidth;
-
   #elif defined(USE_NEON)
       constexpr IndexType NumChunks = InputDimensions / (SimdWidth / 2);
       const int8x8_t Zero = {0};

src/nnue/layers/simd.h

@@ -31,9 +31,6 @@
 #elif defined(USE_SSE2)
 # include <emmintrin.h>
 
-#elif defined(USE_MMX)
-# include <mmintrin.h>
-
 #elif defined(USE_NEON)
 # include <arm_neon.h>
 #endif

src/nnue/nnue_common.h

@@ -42,9 +42,6 @@
 #elif defined(USE_SSE2)
 #include <emmintrin.h>
 
-#elif defined(USE_MMX)
-#include <mmintrin.h>
-
 #elif defined(USE_NEON)
 #include <arm_neon.h>
 #endif
@@ -71,9 +68,6 @@ namespace Stockfish::Eval::NNUE {
   #elif defined(USE_SSE2)
   constexpr std::size_t SimdWidth = 16;
 
-  #elif defined(USE_MMX)
-  constexpr std::size_t SimdWidth = 8;
-
   #elif defined(USE_NEON)
   constexpr std::size_t SimdWidth = 16;
   #endif

src/nnue/nnue_feature_transformer.h

@@ -117,34 +117,6 @@ namespace Stockfish::Eval::NNUE {
   #define NumRegistersSIMD (Is64Bit ? 16 : 8)
   #define MaxChunkSize 16
 
-  #elif USE_MMX
-  using vec_t = __m64;
-  using psqt_vec_t = __m64;
-  #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)
-  #define vec_mul_16(a,b) _mm_mullo_pi16(a,b)
-  #define vec_zero() _mm_setzero_si64()
-  #define vec_set_16(a) _mm_set1_pi16(a)
-  inline vec_t vec_max_16(vec_t a,vec_t b){
-    vec_t comparison = _mm_cmpgt_pi16(a,b);
-    return _mm_or_si64(_mm_and_si64(comparison, a), _mm_andnot_si64(comparison, b));
-  }
-  inline vec_t vec_min_16(vec_t a,vec_t b){
-    vec_t comparison = _mm_cmpgt_pi16(a,b);
-    return _mm_or_si64(_mm_and_si64(comparison, b), _mm_andnot_si64(comparison, a));
-  }
-  #define vec_msb_pack_16(a,b) _mm_packs_pi16(_mm_srli_pi16(a,7),_mm_srli_pi16(b,7))
-  #define vec_load_psqt(a) (*(a))
-  #define vec_store_psqt(a,b) *(a)=(b)
-  #define vec_add_psqt_32(a,b) _mm_add_pi32(a,b)
-  #define vec_sub_psqt_32(a,b) _mm_sub_pi32(a,b)
-  #define vec_zero_psqt() _mm_setzero_si64()
-  #define vec_cleanup() _mm_empty()
-  #define NumRegistersSIMD 8
-  #define MaxChunkSize 8
-
   #elif USE_NEON
   using vec_t = int16x8_t;
   using psqt_vec_t = int32x4_t;
@@ -335,10 +307,6 @@ namespace Stockfish::Eval::NNUE {
 #endif
       }
 
-#if defined(vec_cleanup)
-      vec_cleanup();
-#endif
-
       return psqt;
     } // end of function transform()
 
@@ -529,10 +497,6 @@ namespace Stockfish::Eval::NNUE {
         }
       }
 #endif
-
-  #if defined(USE_MMX)
-      _mm_empty();
-  #endif
     }
 
     template<Color Perspective>
@@ -613,10 +577,6 @@ namespace Stockfish::Eval::NNUE {
           accumulator.psqtAccumulation[Perspective][k] += psqtWeights[index * PSQTBuckets + k];
       }
 #endif
-
-  #if defined(USE_MMX)
-      _mm_empty();
-  #endif
     }
 
     template<Color Perspective>