Declare dependency on version.txt in Makefile

It's bogus since we mess with oldestVersionFullPrecision in addWrites

ConflictSet.cpp

@@ -2001,10 +2001,149 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
 
   assert(!(begin == -1 && end == 256));
 
+  switch (n->getType()) {
+  case Type_Node0:
+    return true;
+  case Type_Node3: {
+    auto *self = static_cast<Node3 *>(n);
+
+    ++begin;
+
+    const unsigned shiftUpperBound = end - begin;
+    const unsigned shiftAmount = begin;
+    auto inBounds = [&](unsigned c) {
+      return c - shiftAmount < shiftUpperBound;
+    };
+
+    uint32_t mask = 0;
+    for (int i = 0; i < Node3::kMaxNodes; ++i) {
+      mask |= inBounds(self->index[i]) << i;
+    }
+    mask &= (1 << self->numChildren) - 1;
+    if (!mask) {
+      return true;
+    }
+    auto *child = self->children[__builtin_ctz(mask)];
+    const bool firstRangeOk =
+        !child->entryPresent || child->entry.rangeVersion <= readVersion;
+    uint32_t compared = 0;
+    for (int i = 0; i < Node3::kMaxNodes; ++i) {
+      compared |= (self->childMaxVersion[i] > readVersion) << i;
+    }
+
+    return !(compared & mask) && firstRangeOk;
+  }
+  case Type_Node16: {
+    auto *self = static_cast<Node16 *>(n);
+
+    ++begin;
+
+    assert(begin <= end);
+    assert(end - begin < 256);
+
+#ifdef HAS_ARM_NEON
+
+    uint8x16_t indices;
+    memcpy(&indices, self->index, 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);
+
+    mask &= self->numChildren == 16
+                ? uint64_t(-1)
+                : (uint64_t(1) << (self->numChildren << 2)) - 1;
+    if (!mask) {
+      return true;
+    }
+    auto *child = self->children[__builtin_ctzll(mask) >> 2];
+    const bool firstRangeOk =
+        !child->entryPresent || child->entry.rangeVersion <= readVersion;
+
+    uint32x4_t w4[4];
+    memcpy(w4, self->childMaxVersion, sizeof(w4));
+    uint32_t rv;
+    memcpy(&rv, &readVersion, sizeof(rv));
+    const auto rvVec = vdupq_n_u32(rv);
+
+    int32x4_t z;
+    memset(&z, 0, sizeof(z));
+
+    uint16x4_t conflicting[4];
+    for (int i = 0; i < 4; ++i) {
+      conflicting[i] = vmovn_u32(
+          vcgtq_s32(vreinterpretq_s32_u32(vsubq_u32(w4[i], rvVec)), z));
+    }
+    auto combined =
+        vcombine_u8(vmovn_u16(vcombine_u16(conflicting[0], conflicting[1])),
+                    vmovn_u16(vcombine_u16(conflicting[2], conflicting[3])));
+
+    uint64_t compared = vget_lane_u64(
+        vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(combined), 4)), 0);
+
+    return !(compared & mask) && firstRangeOk;
+
+#elif defined(HAS_AVX)
+
+    __m128i indices;
+    memcpy(&indices, self->index, 16);
+    indices = _mm_sub_epi8(indices, _mm_set1_epi8(begin));
+    uint32_t mask =
+        0xffff &
+        ~_mm_movemask_epi8(_mm_cmpeq_epi8(
+            indices, _mm_max_epu8(indices, _mm_set1_epi8(end - begin))));
+    mask &= (1 << self->numChildren) - 1;
+    if (!mask) {
+      return true;
+    }
+    auto *child = self->children[__builtin_ctz(mask)];
+    const bool firstRangeOk =
+        !child->entryPresent || child->entry.rangeVersion <= readVersion;
+
+    uint32_t compared = 0;
+    if constexpr (kAVX512) {
+      compared = compare16_32bit_avx512(self->childMaxVersion, readVersion);
+    } else {
+      compared = compare16_32bit(self->childMaxVersion, readVersion);
+    }
+    return !(compared & mask) && firstRangeOk;
+
+#else
+
+    const unsigned shiftUpperBound = end - begin;
+    const unsigned shiftAmount = begin;
+    auto inBounds = [&](unsigned c) {
+      return c - shiftAmount < shiftUpperBound;
+    };
+
+    uint32_t mask = 0;
+    for (int i = 0; i < 16; ++i) {
+      mask |= inBounds(self->index[i]) << i;
+    }
+    mask &= (1 << self->numChildren) - 1;
+    if (!mask) {
+      return true;
+    }
+    auto *child = self->children[__builtin_ctz(mask)];
+    const bool firstRangeOk =
+        !child->entryPresent || child->entry.rangeVersion <= readVersion;
+    uint32_t compared = 0;
+    for (int i = 0; i < 16; ++i) {
+      compared |= (self->childMaxVersion[i] > readVersion) << i;
+    }
+    return !(compared & mask) && firstRangeOk;
+
+#endif
+  }
+  case Type_Node48: {
+    auto *self = static_cast<Node48 *>(n);
+
     {
-    int c = getChildGeq(n, begin + 1);
+      int c = self->bitSet.firstSetGeq(begin + 1);
       if (c >= 0 && c < end) {
-      auto *child = getChildExists(n, c);
+        auto *child = self->children[self->index[c]];
         if (child->entryPresent) {
           if (!(child->entry.rangeVersion <= readVersion)) {
             return false;
@@ -2014,42 +2153,11 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
       } else {
         return true;
       }
-  }
-
-  // [begin, end) is now the half-open interval of children we're interested in.
+      // [begin, end) is now the half-open interval of children we're interested
+      // in.
       assert(begin < end);
-
-  switch (n->getType()) {
-  case Type_Node0: // GCOVR_EXCL_LINE
-    // We would have returned above, after not finding a child
-    __builtin_unreachable(); // GCOVR_EXCL_LINE
-  case Type_Node3: {
-    auto *self = static_cast<Node3 *>(n);
-
-    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 < Node3::kMaxNodes; ++i) {
-      compared |= (self->childMaxVersion[i] > readVersion) << i;
     }
-    uint32_t mask = 0;
-    for (int i = 0; i < Node3::kMaxNodes; ++i) {
-      mask |= inBounds(self->index[i]) << i;
-    }
-    return !(compared & mask);
-  }
-  case Type_Node16: {
-    auto *self = static_cast<Node16 *>(n);
 
-    return scan16<kAVX512>(self->childMaxVersion, self->index, begin, end,
-                           readVersion);
-  }
-  case Type_Node48: {
-    auto *self = static_cast<Node48 *>(n);
     // Check all pages
     static_assert(Node48::kMaxOfMaxPageSize == 16);
     for (int i = 0; i < Node48::kMaxOfMaxTotalPages; ++i) {
@@ -2067,6 +2175,25 @@ bool checkMaxBetweenExclusive(Node *n, int begin, int end,
   case Type_Node256: {
     static_assert(Node256::kMaxOfMaxTotalPages == 16);
     auto *self = static_cast<Node256 *>(n);
+
+    {
+      int c = self->bitSet.firstSetGeq(begin + 1);
+      if (c >= 0 && c < end) {
+        auto *child = self->children[c];
+        if (child->entryPresent) {
+          if (!(child->entry.rangeVersion <= readVersion)) {
+            return false;
+          };
+        }
+        begin = c;
+      } else {
+        return true;
+      }
+      // [begin, end) is now the half-open interval of children we're interested
+      // in.
+      assert(begin < end);
+    }
+
     const int firstPage = begin >> Node256::kMaxOfMaxShift;
     const int lastPage = (end - 1) >> Node256::kMaxOfMaxShift;
     // Check the only page if there's only one
@@ -2884,11 +3011,10 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
       const auto &r = reads[i];
       auto begin = std::span<const uint8_t>(r.begin.p, r.begin.len);
       auto end = std::span<const uint8_t>(r.end.p, r.end.len);
+      assert(oldestVersionFullPrecision >=
+             newestVersionFullPrecision - kNominalVersionWindow);
       result[i] =
-          reads[i].readVersion < oldestVersionFullPrecision ||
-                  reads[i].readVersion <
-                      newestVersionFullPrecision - kNominalVersionWindow
-              ? TooOld
+          reads[i].readVersion < oldestVersionFullPrecision ? TooOld
           : (end.size() > 0
                  ? checkRangeRead(root, begin, end,
                                   InternalVersionT(reads[i].readVersion), this)
@@ -3155,7 +3281,6 @@ void internal_addWrites(ConflictSet::Impl *impl,
 
 void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
   mallocBytesDelta = 0;
-  assert(oldestVersion >= impl->oldestVersionFullPrecision);
   impl->setOldestVersion(oldestVersion);
   impl->totalBytes += mallocBytesDelta;
 #if SHOW_MEMORY

paper/Makefile

@@ -2,7 +2,7 @@
 
 all: paper.pdf
 
-paper.pdf: paper.tex $(wildcard *.tikz)
+paper.pdf: paper.tex $(wildcard *.tikz) version.txt
 	latexmk -pdf
 
 clean: