Attempt at avx512f 32bit compare

ConflictSet.cpp

@@ -1766,6 +1766,30 @@ downLeftSpine:
   }
 }
 
+#ifdef HAS_AVX
+__attribute__((target("default"))) uint32_t compare16_32bit(const uint32_t *vs,
+                                                            uint32_t rv) {
+  __m128i w[4];
+  memcpy(w, vs, sizeof(w));
+  const auto rvVec = _mm_set1_epi32(rv);
+  const auto zero = _mm_setzero_si128();
+  for (int i = 0; i < 4; ++i) {
+    compared |=
+        _mm_movemask_ps(_mm_cmpgt_epi32(_mm_sub_epi32(w[i], rvVec), zero))
+        << (i * 4);
+  }
+  return compared;
+}
+
+__attribute__((target("avx512f"))) uint32_t compare16_32bit(const uint32_t *vs,
+                                                            uint32_t rv) {
+  __mm512i w;
+  memcpy(&w, vs, sizeof(w));
+  return _mm512_cmpgt_epi32_mask(_mm512_sub_epi32(w, _mm512_set1_epi32(rv)),
+                                 _mm512_setzero());
+}
+#endif
+
 // Returns true if v[i] <= readVersion for all i such that begin <= is[i] < end
 // Preconditions: begin <= end, end - begin < 256
 bool scan16(const InternalVersionT *vs, const uint8_t *is, int begin, int end,
@@ -1824,17 +1848,7 @@ bool scan16(const InternalVersionT *vs, const uint8_t *is, int begin, int end,
 
   uint32_t compared = 0;
 #if INTERNAL_VERSION_32_BIT
-  __m128i w4[4];
+  compared = compare16_32bit(vs, readVersion);
-  memcpy(w4, vs, sizeof(w4));
-  uint32_t rv;
-  memcpy(&rv, &readVersion, sizeof(rv));
-  const auto rvVec = _mm_set1_epi32(rv);
-  const auto zero = _mm_setzero_si128();
-  for (int i = 0; i < 4; ++i) {
-    compared |=
-        _mm_movemask_ps(_mm_cmpgt_epi32(_mm_sub_epi32(w4[i], rvVec), zero))
-        << (i * 4);
-  }
 #else
   for (int i = 0; i < 16; ++i) {
     compared |= (vs[i] > readVersion) << i;