add more comments, undo loop unrolling.

makefile

@@ -24,7 +24,7 @@ lint:
 	clang-tidy squiggle_more.c -- -lm
 
 profile:
-	OMP_NUM_THREADS=16 sudo perf record ./examples/more/12_time_to_botec_parallel/example
+	sudo perf record ./examples/more/12_time_to_botec_parallel/example
 	sudo perf report
 	rm perf.data
 

squiggle_more.c

@@ -64,42 +64,16 @@ void sampler_parallel(double (*sampler)(uint64_t* seed), double* results, int n_
     {
 #pragma omp for
         for (i = 0; i < n_threads; i++) {
-            // Simple version 
-            /*
             int quotient = n_samples / n_threads;
             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(&(cache_box[i].seed));
-                // Could also result in inefficient cache stuff, but hopefully not too often
-            }
-            */
-            
-            // Version with loop unrolling
-            int quotient = n_samples / n_threads;
-            int lower_bound_inclusive = i * quotient;
-            int upper_bound_not_inclusive = ((i + 1) * quotient); // note the < in the for loop below,
-            int delta = quotient;
-            int eighth_of_deltas = delta/8; // why 8? a double in 8 bytes, 8 doubles is the size of a cache line
-            int k;
-            // to do: simplify these variables. Maybe divide by n_threads * 8 directly
-            for(int j=0; j<eighth_of_deltas; j++){
-                k = lower_bound_inclusive + j*8;
-                results[k+0] = sampler(&(cache_box[i].seed));
-                results[k+1] = sampler(&(cache_box[i].seed));
-                results[k+2] = sampler(&(cache_box[i].seed));
-                results[k+3] = sampler(&(cache_box[i].seed));
-                results[k+4] = sampler(&(cache_box[i].seed));
-                results[k+5] = sampler(&(cache_box[i].seed));
-                results[k+6] = sampler(&(cache_box[i].seed));
-                results[k+7] = sampler(&(cache_box[i].seed));
-                // these all fit one single cache line
-                // name of the techique: loop unrolling.
-            }
-            for(int k=eighth_of_deltas*8; k<upper_bound_not_inclusive; k++){
-                results[k] = sampler(&(cache_box[i].seed));
-
-            }
+                // In principle, these results[j] could also result in two threads competing for the same cache line. 
+                // In practice, though,
+                // a) this would happen infrequently
+                // b)            
+                }
             
         }
     }