Ignore itlb in grafana instead

Events in the same group should be associated with the same set of
instructions

ServerBench.cpp

@@ -6,8 +6,10 @@
 #include <string.h>
 #include <string>
 #include <string_view>
+#include <sys/ioctl.h>
 #include <sys/resource.h>
 #include <sys/socket.h>
+#include <sys/types.h>
 #include <sys/uio.h>
 #include <thread>
 #include <unistd.h>
@@ -166,18 +168,11 @@ double toSeconds(timeval t) {
   return double(t.tv_sec) + double(t.tv_usec) * 1e-6;
 }
 
-#include <linux/perf_event.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <sys/ioctl.h>
-#include <sys/syscall.h>
-#include <sys/types.h>
-#include <unistd.h>
-
 #ifdef __linux__
+#include <linux/perf_event.h>
 struct PerfCounter {
-  PerfCounter(int type, int config, const std::string &labels = {})
+  PerfCounter(int type, int config, const std::string &labels = {},
+              int groupLeaderFd = -1)
       : labels(labels) {
     struct perf_event_attr pe;
 
@@ -189,7 +184,10 @@ struct PerfCounter {
     pe.exclude_kernel = 1;
     pe.exclude_hv = 1;
 
-    fd = perf_event_open(&pe, 0, -1, -1, 0);
+    fd = perf_event_open(&pe, 0, -1, groupLeaderFd, 0);
+    if (fd < 0 && errno != ENOENT && errno != EINVAL) {
+      perror(labels.c_str());
+    }
   }
 
   int64_t total() const {
@@ -217,6 +215,7 @@ struct PerfCounter {
 
   bool ok() const { return fd >= 0; }
   const std::string &getLabels() const { return labels; }
+  int getFd() const { return fd; }
 
 private:
   int fd;
@@ -249,7 +248,8 @@ int main(int argc, char **argv) {
     cs.getMetricsV1(&metrics, &metricsCount);
 
     PerfCounter instructions{PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS};
-    PerfCounter cycles{PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES};
+    PerfCounter cycles{PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES, "",
+                       instructions.getFd()};
 
     std::vector<PerfCounter> cacheCounters;
     for (auto [id, idStr] : std::initializer_list<std::pair<int, std::string>>{
@@ -257,8 +257,7 @@ int main(int argc, char **argv) {
              {PERF_COUNT_HW_CACHE_L1I, "l1i"},
              {PERF_COUNT_HW_CACHE_LL, "ll"},
              {PERF_COUNT_HW_CACHE_DTLB, "dtlb"},
-             // Somehow was showing a miss rate > 1 /shrug
-             //  {PERF_COUNT_HW_CACHE_ITLB, "itlb"},
+             {PERF_COUNT_HW_CACHE_ITLB, "itlb"},
              {PERF_COUNT_HW_CACHE_BPU, "bpu"},
              {PERF_COUNT_HW_CACHE_NODE, "node"},
          }) {
@@ -268,6 +267,7 @@ int main(int argc, char **argv) {
                {PERF_COUNT_HW_CACHE_OP_WRITE, "write"},
                {PERF_COUNT_HW_CACHE_OP_PREFETCH, "prefetch"},
            }) {
+        int groupLeaderFd = -1;
         for (auto [result, resultStr] :
              std::initializer_list<std::pair<int, std::string>>{
                  {PERF_COUNT_HW_CACHE_RESULT_MISS, "miss"},
@@ -276,10 +276,14 @@ int main(int argc, char **argv) {
           auto labels = "{id=\"" + idStr + "\", op=\"" + opStr +
                         "\", result=\"" + resultStr + "\"}";
           cacheCounters.emplace_back(PERF_TYPE_HW_CACHE,
-                                     id | (op << 8) | (result << 16), labels);
+                                     id | (op << 8) | (result << 16), labels,
+                                     groupLeaderFd);
           if (!cacheCounters.back().ok()) {
-            fprintf(stderr, "Could not open cache event: %s\n", labels.c_str());
             cacheCounters.pop_back();
+          } else {
+            if (groupLeaderFd == -1) {
+              groupLeaderFd = cacheCounters.back().getFd();
+            }
           }
         }
       }