Add optional microbenchmarks for rezero/scan avx512

CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.18)
 project(
   conflict-set
-  VERSION 0.0.13
+  VERSION 0.0.14
   DESCRIPTION
     "A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys."
   HOMEPAGE_URL "https://git.weaselab.dev/weaselab/conflict-set"

ConflictSet.cpp

@@ -4073,6 +4073,62 @@ struct __attribute__((visibility("default"))) PeakPrinter {
 
 #ifdef ENABLE_MAIN
 
+#define ANKERL_NANOBENCH_IMPLEMENT
+#include "third_party/nanobench.h"
+
+template <int kN> void benchRezero() {
+  static_assert(kN % 16 == 0);
+  ankerl::nanobench::Bench bench;
+  InternalVersionT vs[kN];
+  InternalVersionT zero;
+  bench.run("rezero" + std::to_string(kN), [&]() {
+    bench.doNotOptimizeAway(vs);
+    bench.doNotOptimizeAway(zero);
+    for (int i = 0; i < kN; i += 16) {
+      rezero16(vs + i, zero);
+    }
+  });
+}
+
+template <int kN> void benchScan1() {
+  static_assert(kN % 16 == 0);
+  ankerl::nanobench::Bench bench;
+  InternalVersionT vs[kN];
+  uint8_t is[kN];
+  uint8_t begin;
+  uint8_t end;
+  InternalVersionT v;
+  bench.run("scan" + std::to_string(kN), [&]() {
+    bench.doNotOptimizeAway(vs);
+    bench.doNotOptimizeAway(is);
+    bench.doNotOptimizeAway(begin);
+    bench.doNotOptimizeAway(end);
+    bench.doNotOptimizeAway(v);
+    for (int i = 0; i < kN; i += 16) {
+      scan16</*kAVX512=*/true>(vs + i, is + i, begin, end, v);
+    }
+  });
+}
+
+template <int kN> void benchScan2() {
+  static_assert(kN % 16 == 0);
+  ankerl::nanobench::Bench bench;
+  InternalVersionT vs[kN];
+  uint8_t is[kN];
+  uint8_t begin;
+  uint8_t end;
+  InternalVersionT v;
+  bench.run("scan" + std::to_string(kN), [&]() {
+    bench.doNotOptimizeAway(vs);
+    bench.doNotOptimizeAway(begin);
+    bench.doNotOptimizeAway(end);
+    bench.doNotOptimizeAway(v);
+    for (int i = 0; i < kN; i += 16) {
+      scan16</*kAVX512=*/true>(vs + i, begin, end, v);
+    }
+  });
+}
+
 void printTree() {
   int64_t writeVersion = 0;
   ConflictSet::Impl cs{writeVersion};
@@ -4094,7 +4150,7 @@ void printTree() {
   debugPrintDot(stdout, cs.root, &cs);
 }
 
-int main(void) { printTree(); }
+int main(void) { benchScan1<16>(); }
 #endif
 
 #ifdef ENABLE_FUZZ