test i there might be some "false sharing" going on with openmp

squiggle_more.c

@@ -9,6 +9,12 @@
 #include <string.h> // memcpy
 
 /* Parallel sampler */
+#define CACHE_LINE_SIZE 64
+typedef struct padded_seed_t {
+    uint64_t* seed;
+    char padding[CACHE_LINE_SIZE - sizeof(uint64_t*)];
+} padded_seed;
+
 void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_threads, int n_samples)
 {
 
@@ -29,14 +35,15 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
     int quotient = n_samples / n_threads;
     int divisor_multiple = quotient * n_threads;
 
-    uint64_t** seeds = malloc((size_t)n_threads * sizeof(uint64_t*));
+    // uint64_t** seeds = malloc((size_t)n_threads * sizeof(uint64_t*));
+    padded_seed* seeds = (padded_seed*) malloc(sizeof(padded_seed) * (size_t)n_threads);
     srand(1);
     for (int i = 0; i < n_threads; i++) {
-        seeds[i] = malloc(sizeof(uint64_t));
+        seeds[i].seed = malloc(sizeof(uint64_t*));
         // Constraints:
         // - xorshift can't start with 0
         // - the seeds should be reasonably separated and not correlated
-        *seeds[i] = (uint64_t)rand() * (UINT64_MAX / RAND_MAX);
+        *(seeds[i].seed) = (uint64_t)rand() * (UINT64_MAX / RAND_MAX);
         // printf("#%ld: %lu\n",i, *seeds[i]);
 
         // Other initializations tried:
@@ -53,17 +60,17 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
             int lower_bound_inclusive = i * quotient;
             int upper_bound_not_inclusive = ((i + 1) * quotient); // note the < in the for loop below,
             for (int j = lower_bound_inclusive; j < upper_bound_not_inclusive; j++) {
-                results[j] = sampler(seeds[i]);
+                results[j] = sampler(seeds[i].seed);
             }
         }
     }
     for (int j = divisor_multiple; j < n_samples; j++) {
-        results[j] = sampler(seeds[0]);
+        results[j] = sampler(seeds[0].seed);
         // we can just reuse a seed, this isn't problematic because we are not doing multithreading
     }
 
     for (int i = 0; i < n_threads; i++) {
-        free(seeds[i]);
+        free(seeds[i].seed);
     }
     free(seeds);
 }
@@ -109,7 +116,7 @@ static double quickselect(int k, double xs[], int n)
 
     double *ys = malloc((size_t)n * sizeof(double));
     memcpy(ys, xs, (size_t)n * sizeof(double));
-    // ^: don't make this operation "destructive"
+    // ^: don't rearrange item order in the original array
 
     int low = 0;
     int high = n - 1;