Simd bounds checking for scan for Node16

ConflictSet.cpp

@@ -1719,6 +1719,9 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
   assume(-1 <= end);
   assume(end <= 256);
   assume(begin < end);
+
+  assert(!(begin == -1 && end == 256));
+
   {
     int c = getChildGeq(n, begin + 1);
     if (c >= 0 && c < end) {
@@ -1755,17 +1758,35 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
   } break;
   case Type_Node16: {
     auto *self = static_cast<Node16 *>(n);
+
+#ifdef HAS_ARM_NEON
+    assert(end - begin < 256);
+    uint8x16_t indices;
+    memcpy(&indices, self->index, sizeof(self->index));
+    // 0xff for each in bounds
+    auto results =
+        vcltq_u8(vsubq_u8(indices, vdupq_n_u8(begin)), vdupq_n_u8(end - begin));
+    // 0xf for each 0xff
+    uint64_t mask = vget_lane_u64(
+        vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(results), 4)), 0);
+
+    uint64_t compared = 0;
+    for (int i = 0; i < 16; ++i) {
+      compared |=
+          (uint64_t(self->childMaxVersion[i] > readVersion) << (i << 2));
+    }
+    return !(compared & mask);
+#else
     uint32_t compared = 0;
     for (int i = 0; i < 16; ++i) {
       compared |= ((self->childMaxVersion[i] > readVersion) << i);
     }
-
     uint32_t mask = 0;
     for (int i = 0; i < 16; ++i) {
       mask |= inBounds(self->index[i]) << i;
     }
-
     return !(compared & mask);
+#endif
   } break;
   case Type_Node48: {
     auto *self = static_cast<Node48 *>(n);