Fix shared memory Table size and accesses reporting in GUPS benchmark (#60)

* Fix shared memory Table size and accesses reporting in GUPS benchmark

Based on feedback from rkarim2 in issue #56:
- For shared memory tests, Table size now shows the actual total shared memory used
- Display shared memory allocation details (bytes per block × number of blocks)
- Use correct accesses_per_elem_sh for shared memory tests
- Report size in MB for shared memory vs GB for global memory

This fixes the misleading Table size output that showed irrelevant global memory
sizes (e.g., 4.2 GB) when running shared memory tests.

* Remove comment per reviewer feedback

Removed the comment on line 497 as requested by rkarim2 in PR review.

Author: dmikushin

posts/gups/gups.cu

@@ -484,14 +484,29 @@ int main(int argc, char* argv[])
   }
   size_t total_num_thread = thread * grid;
 
-  printf(
-      "Table size = %zu (%lf GB.)\nTotal number of threads %zu\nEach thread "
-      "access %d locations.\nNumber of iterations = %d\n",
-      working_set,
-      working_set * sizeof(benchtype) / 1e9,
-      total_num_thread,
-      accesses_per_elem,
-      repeats);
+  if (!shared_mem) {
+    printf(
+        "Table size = %zu (%lf GB.)\nTotal number of threads %zu\nEach thread "
+        "access %d locations.\nNumber of iterations = %d\n",
+        working_set,
+        working_set * sizeof(benchtype) / 1e9,
+        total_num_thread,
+        accesses_per_elem,
+        repeats);
+  } else {
+    size_t total_shmem = grid * n_shmem * sizeof(benchtype);
+    printf(
+        "Table size = %zu (%lf MB.) [shared memory: %zu bytes per block x %zu blocks]\n"
+        "Total number of threads %zu\nEach thread "
+        "access %d locations.\nNumber of iterations = %d\n",
+        total_shmem / sizeof(benchtype),
+        total_shmem / 1e6,
+        n_shmem * sizeof(benchtype),
+        grid,
+        total_num_thread,
+        accesses_per_elem_sh,
+        repeats);
+  }
 
   benchtype* d_t;
   if (!shared_mem)