Share "scan16" between Node16 and Node48

ConflictSet.cpp

@@ -293,7 +293,7 @@ struct Node48 : Node {
   Node *children[kMaxNodes];
   int64_t childMaxVersion[kMaxNodes];
   uint8_t reverseIndex[kMaxNodes];
-  constexpr static int kMaxOfMaxPageSize = 8;
+  constexpr static int kMaxOfMaxPageSize = 16;
   constexpr static int kMaxOfMaxShift =
       std::countr_zero(uint32_t(kMaxOfMaxPageSize));
   constexpr static int kMaxOfMaxTotalPages = kMaxNodes / kMaxOfMaxPageSize;
@@ -1709,6 +1709,47 @@ downLeftSpine:
   }
 }
 
+// Returns true if all in-bounds vs are <= readVersion
+bool scan16(const int64_t *vs, const uint8_t *is, int begin, int end,
+            int64_t readVersion) {
+
+#ifdef HAS_ARM_NEON
+
+  assert(end - begin < 256);
+  uint8x16_t indices;
+  memcpy(&indices, is, 16);
+  // 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(vs[i] > readVersion) << (i << 2));
+  }
+  return !(compared & mask);
+
+#else
+
+  const unsigned shiftUpperBound = end - begin;
+  const unsigned shiftAmount = begin;
+  auto inBounds = [&](unsigned c) { return c - shiftAmount < shiftUpperBound; };
+
+  uint32_t compared = 0;
+  for (int i = 0; i < 16; ++i) {
+    compared |= ((vs[i] > readVersion) << i);
+  }
+  uint32_t mask = 0;
+  for (int i = 0; i < 16; ++i) {
+    mask |= inBounds(is[i]) << i;
+  }
+  return !(compared & mask);
+
+#endif
+}
+
 // Return whether or not the max version among all keys starting with the search
 // path of n + [child], where child in (begin, end) is <= readVersion. Does not
 // account for the range version of firstGt(searchpath(n) + [end - 1])
@@ -1759,49 +1800,23 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
   case Type_Node16: {
     auto *self = static_cast<Node16 *>(n);
 
-#ifdef HAS_ARM_NEON
+    return scan16(self->childMaxVersion, self->index, begin, end, readVersion);
-    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);
     // Check all pages
-    bool result = true;
+    static_assert(Node48::kMaxOfMaxPageSize == 16);
     for (int i = 0; i < Node48::kMaxOfMaxTotalPages; ++i) {
       if (self->maxOfMax[i] > readVersion) {
-        for (int j = 0; j < Node48::kMaxOfMaxPageSize; ++j) {
+        if (!scan16(self->childMaxVersion + (i << Node48::kMaxOfMaxShift),
-          int k = (i << Node48::kMaxOfMaxShift) + j;
+                    self->reverseIndex + (i << Node48::kMaxOfMaxShift), begin,
-          result &= !((self->childMaxVersion[k] > readVersion) &
+                    end, readVersion)) {
-                      inBounds(self->reverseIndex[k]));
+          return false;
         }
       }
     }
-    return result;
+    return true;
   }
   case Type_Node256: {
     auto *self = static_cast<Node256 *>(n);