Update XGMI perf counter test to show utilization

Also:
* When destroying a counter, make sure to stop the counter first
* In the test, do not stop (disable) the counter before
  reading it.
* Clean up some whitespace in other tests
* Re-add manual pdf file

Change-Id: I0786ef3a994ca568299c77e44f092af8943ac33d

src/rocm_smi.cc

@@ -566,7 +566,7 @@ rsmi_init(uint64_t flags) {
       smi.Initialize(flags);
     } catch(...) {
       smi.Cleanup();
-      throw;
+      throw amd::smi::rsmi_exception(RSMI_STATUS_INIT_ERROR, __FUNCTION__);
     }
   }
   refGuard.Dismiss();
@@ -2663,14 +2663,17 @@ rsmi_dev_counter_destroy(rsmi_event_handle_t evnt_handle) {
     return RSMI_STATUS_INVALID_ARGS;
   }
 
+  uint32_t ret = 0;
   amd::smi::evt::Event *evt =
                         reinterpret_cast<amd::smi::evt::Event *>(evnt_handle);
   uint32_t dv_ind = evt->dev_ind();
   DEVICE_MUTEX
   REQUIRE_ROOT_ACCESS
 
+  ret = evt->stopCounter();
+
   delete evt;
-  return RSMI_STATUS_SUCCESS;
+  return errno_to_rsmi_status(ret);;
   CATCH
 }
 
@@ -2730,6 +2733,12 @@ rsmi_counter_read(rsmi_event_handle_t evt_handle,
 
   ret = evt->getValue(value);
 
+  // If value > 2^48, then an overflow has occurred. We need to discard this
+  // value and re-read:
+  if (ret == 0 && value->value > 0xFFFFFFFFFFFF) {
+    ret = evt->getValue(value);
+  }
+
   return errno_to_rsmi_status(ret);
   CATCH
 }

src/rocm_smi_counters.cc

@@ -315,6 +315,7 @@ amd::smi::evt::Event::openPerfHandle(void) {
 
   attr_.size = sizeof(struct perf_event_attr);
   attr_.config = get_perf_attr_config(&event_info_);
+  attr_.sample_type = PERF_SAMPLE_IDENTIFIER;
   attr_.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
                           PERF_FORMAT_TOTAL_TIME_RUNNING;
   attr_.disabled = 1;