top: minimize the statistics overhead for numa support

A recent libnuma potential corruption problem solution
has caused me to reevaluate some associated numa logic
for efficiency. Here is a summary of the problems that
exist with current libnuma/user possible interactions:

. Whenever the numa library was present extra overhead
would always be incurred in maintaining the node stats
even when the '2' or '3' commands were not being used.

. As part of such overhead a separate loop was used to
reinitialize each cpu/node structure with each display
cycle so that prior accumulated totals were preserved.
Again, it didn't matter if numa data was really shown.

This commit attempts to refocus on the 'critical path'
costs in a running top by optimizing for the occasions
when numa node data is not being displayed. Under such
conditions, no extra overhead will be incurred whether
or not a distribution has the libnuma library present.

To achieve this goal, some additional overhead will be
incurred, but only when actually displaying numa data.
And all such new costs have been minimized in spite of
the gcc inclination to duplicate subscript resolution.

Reference(s):
commit 24bd950cb2e1722d459461f0f9c0c30a4b9ffdaa

Signed-off-by: Jim Warner <james.warner@comcast.net>

diff --git a/top/top.c b/top/top.c
index e619ddd3455a37662f8b94a31de90c4e8e4d8f3e..9d12693d3b59339544a035693581a6520671a534 100644 (file)
--- a/top/top.c
+++ b/top/top.c
@@ -2361,18 +2361,9 @@ static CPU_t *cpus_refresh (CPU_t *cpus) {
 #endif
 
 #ifndef NUMA_DISABLE
-   for (i = 0; i < Numa_node_tot; i++) {
-      node = sumSLOT + 1 + i;
-      // remember from last time around
-      memcpy(&cpus[node].sav, &cpus[node].cur, sizeof(CT_t));
-      // initialize current node statistics
-      memset(&cpus[node].cur, 0, sizeof(CT_t));
-#ifndef CPU_ZEROTICS
-      cpus[node].edge = cpus[sumSLOT].edge;
-      // this is for symmetry only, it's not currently required
-      cpus[node].cur.tot = cpus[sumSLOT].cur.tot;
-#endif
-   }
+   // forget all of the prior node statistics (maybe)
+   if (CHKw(Curwin, View_CPUNOD))
+      memset(&cpus[sumSLOT + 1], 0, Numa_node_tot * sizeof(CPU_t));
 #endif
 
    // now value each separate cpu's tics...
@@ -2400,21 +2391,30 @@ static CPU_t *cpus_refresh (CPU_t *cpus) {
       cpus[i].id = i;
 #endif
 #ifndef NUMA_DISABLE
-      if (Numa_node_tot
+      /* henceforth, with just a little more arithmetic we can avoid
+         maintaining *any* node stats unless they're actually needed */
+      if (CHKw(Curwin, View_CPUNOD)
+      && Numa_node_tot
       && -1 < (node = Numa_node_of_cpu(cpus[i].id))) {
+         // use our own pointer to avoid gcc subscript bloat
+         CPU_t *nod_ptr = &cpus[sumSLOT + 1 + node];
+         nod_ptr->cur.u += cpus[i].cur.u; nod_ptr->sav.u += cpus[i].sav.u;
+         nod_ptr->cur.n += cpus[i].cur.n; nod_ptr->sav.n += cpus[i].sav.n;
+         nod_ptr->cur.s += cpus[i].cur.s; nod_ptr->sav.s += cpus[i].sav.s;
+         nod_ptr->cur.i += cpus[i].cur.i; nod_ptr->sav.i += cpus[i].sav.i;
+         nod_ptr->cur.w += cpus[i].cur.w; nod_ptr->sav.w += cpus[i].sav.w;
+         nod_ptr->cur.x += cpus[i].cur.x; nod_ptr->sav.x += cpus[i].sav.x;
+         nod_ptr->cur.y += cpus[i].cur.y; nod_ptr->sav.y += cpus[i].sav.y;
+         nod_ptr->cur.z += cpus[i].cur.z; nod_ptr->sav.z += cpus[i].sav.z;
+#ifndef CPU_ZEROTICS
+         /* yep, we re-value this repeatedly for each cpu encountered, but we
+            can then avoid a prior loop to selectively initialize each node */
+         nod_ptr->edge = cpus[sumSLOT].edge;
+#endif
          cpus[i].node = node;
-         node += (sumSLOT + 1);
-         cpus[node].cur.u += cpus[i].cur.u;
-         cpus[node].cur.n += cpus[i].cur.n;
-         cpus[node].cur.s += cpus[i].cur.s;
-         cpus[node].cur.i += cpus[i].cur.i;
-         cpus[node].cur.w += cpus[i].cur.w;
-         cpus[node].cur.x += cpus[i].cur.x;
-         cpus[node].cur.y += cpus[i].cur.y;
-         cpus[node].cur.z += cpus[i].cur.z;
       }
 #endif
-   }
+   } // end: for each cpu
 
    Cpu_faux_tot = i;      // tolerate cpus taken offline