Induce a cache miss in interleaving test

InterleavingTest.cpp

@@ -1,16 +1,24 @@
 #include <alloca.h>
 #include <cassert>
+#ifdef __x86_64__
+#include <immintrin.h>
+#endif
 
 #define ANKERL_NANOBENCH_IMPLEMENT
 #include "third_party/nanobench.h"
 
 struct Job {
-  int input;
+  int *input;
 };
 
 bool stepJob(void *x) {
   auto *j = (Job *)x;
-  return --j->input == 0;
+  auto &val = *j->input;
+  auto result = --val == 0;
+#ifdef __x86_64__
+  _mm_clflush(j->input);
+#endif
+  return result;
 }
 
 // So we can look at the disassembly more easily
@@ -108,9 +116,11 @@ int main() {
   Job jobs[kNumJobs];
   Job jobsCopy[kNumJobs];
   int iters = 0;
+  int originalInput[kNumJobs];
   for (int i = 0; i < kNumJobs; ++i) {
-    jobs[i].input = rand() % 5 + 3;
+    originalInput[i] = rand() % 5 + 3;
-    iters += jobs[i].input;
+    jobs[i].input = new int{originalInput[i]};
+    iters += *jobs[i].input;
   }
   bench.batch(iters);
 
@@ -120,6 +130,9 @@ int main() {
         std::make_pair(interleaveBoundedCyclicList,
                        "interleaveBoundedCyclicList"),
         std::make_pair(interleaveCyclicList, "interleaveCyclicList")}) {
+    for (int i = 0; i < kNumJobs; ++i) {
+      *jobs[i].input = originalInput[i];
+    }
     memcpy(jobsCopy, jobs, sizeof(jobs));
     void *ps[kNumJobs];
     for (int i = 0; i < kNumJobs; ++i) {
@@ -127,13 +140,16 @@ int main() {
     }
     scheduler(ps, stepJob, kNumJobs);
     for (int i = 0; i < kNumJobs; ++i) {
-      if (jobsCopy[i].input != 0) {
+      if (*jobsCopy[i].input != 0) {
         fprintf(stderr, "%s failed\n", name);
         abort();
       }
     }
 
     bench.run(name, [&]() {
+      for (int i = 0; i < kNumJobs; ++i) {
+        *jobs[i].input = originalInput[i];
+      }
       memcpy(jobsCopy, jobs, sizeof(jobs));
       void *ps[kNumJobs];
       for (int i = 0; i < kNumJobs; ++i) {
@@ -142,4 +158,7 @@ int main() {
       scheduler(ps, stepJob, kNumJobs);
     });
   }
+  for (int i = 0; i < kNumJobs; ++i) {
+    delete jobs[i].input;
+  }
 }