Add error handling case for lockhammer -o option

We can clearly observe 10% difference in lock access latency for 2
different pinning order (default vs optimized) like the following:

./build/lh_osq_lock -t 5 -a 1000000
5000000 lock loops
20784292691 ns scheduled
4156990971 ns elapsed (~4.999841 cores)
4156.858538 ns per access (scheduled)
4156.836709 ns per access (real)
831.398194 ns access rate
1.000000 average depth
5, 4.999841, 4156.858538, 4156.836709, 831.398194, 1.000000

./build/lh_osq_lock -t 5 -a 1000000 -o 0,32,4,36,8
5000000 lock loops
18483098623 ns scheduled
3696729985 ns elapsed (~4.999850 cores)
3696.619725 ns per access (scheduled)
3696.614593 ns per access (real)
739.345997 ns access rate
1.000000 average depth
5, 4.999850, 3696.619725, 3696.614593, 739.345997, 1.000000

Author: Yangzheng Bai

benchmarks/lockhammer/src/lockhammer.c

@@ -179,8 +179,11 @@ int main(int argc, char** argv)
                 return 1;
             }
             csv = strtok(optarg, ",");
-            for (int i = 0; i < num_cores && csv != NULL; csv = strtok(NULL, ","), ++i)
-                *(args.pinorder + i) = strtol(csv, (char **) NULL, 10);
+            for (int i = 0; i < num_cores && csv != NULL; csv = strtok(NULL, ","), ++i) {
+                optval = strtol(csv, (char **) NULL, 10);
+                if (optval >= 0 && optval < num_cores) *(args.pinorder + i) = optval;
+                else fprintf(stderr, "WARNING: core number %ld is out of range.\n", optval);
+            }
             break;
           case 's':
             args.safemode = 1;