Switch to CLOCK_BOOTTIME for HSA system clock.

This is consistent with KFD and has significantly better latency.
KFD is taking this as the definition of the SystemClockCounter.

Change-Id: I4c1b3bc58c738206265c55ebefd41356c013bfe5


[ROCm/ROCR-Runtime commit: 0ee82742a7]

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/runtime.cpp

@@ -580,9 +580,7 @@ hsa_status_t Runtime::GetSystemInfo(hsa_system_info_t attribute, void* value) {
       *((uint16_t*)value) = HSA_VERSION_MINOR;
       break;
     case HSA_SYSTEM_INFO_TIMESTAMP: {
-      HsaClockCounters clocks;
-      hsaKmtGetClockCounters(0, &clocks);
-      *((uint64_t*)value) = clocks.SystemClockCounter;
+      *((uint64_t*)value) = os::ReadSystemClock();
       break;
     }
     case HSA_SYSTEM_INFO_TIMESTAMP_FREQUENCY: {
@@ -1349,10 +1347,8 @@ hsa_status_t Runtime::Load() {
 
   // Setup system clock frequency for the first time.
   if (sys_clock_freq_ == 0) {
-    // Cache system clock frequency
-    HsaClockCounters clocks;
-    hsaKmtGetClockCounters(0, &clocks);
-    sys_clock_freq_ = clocks.SystemClockFrequencyHz;
+    sys_clock_freq_ = os::SystemClockFrequency();
+    if (sys_clock_freq_ < 100000) debug_warning("System clock resolution is low.");
   }
 
   BindVmFaultHandler();

projects/rocr-runtime/runtime/hsa-runtime/core/util/lnx/os_linux.cpp

@@ -515,6 +515,25 @@ void DestroySharedMutex(SharedMutex lock) {
   delete *(pthread_rwlock_t**)&lock;
 }
 
+static uint64_t sys_clock_period_ = 0;
+
+uint64_t ReadSystemClock() {
+  struct timespec ts;
+  clock_gettime(CLOCK_BOOTTIME, &ts);
+  uint64_t time = (uint64_t(ts.tv_sec) * 1000000000 + uint64_t(ts.tv_nsec));
+  if (sys_clock_period_ != 1)
+    return time / sys_clock_period_;
+  else
+    return time;
+}
+
+uint64_t SystemClockFrequency() {
+  struct timespec ts;
+  clock_getres(CLOCK_BOOTTIME, &ts);
+  sys_clock_period_ = (uint64_t(ts.tv_sec) * 1000000000 + uint64_t(ts.tv_nsec));
+  return 1000000000 / sys_clock_period_;
+}
+
 }   //  namespace os
 }   //  namespace rocr
 

projects/rocr-runtime/runtime/hsa-runtime/core/util/os.h

@@ -266,6 +266,14 @@ uint64_t ReadAccurateClock();
 /// seconds.  This frequency does not change at runtime.
 /// @return returns the frequency
 uint64_t AccurateClockFrequency();
+
+/// @brief read the system clock which serves as the HSA system clock
+/// counter in KFD.
+uint64_t ReadSystemClock();
+
+/// @brief read the system clock frequency
+uint64_t SystemClockFrequency();
+
 }   //  namespace os
 }   //  namespace rocr
 

projects/rocr-runtime/runtime/hsa-runtime/core/util/win/os_win.cpp

@@ -270,6 +270,16 @@ void DestroySharedMutex(SharedMutex lock) {
   abort();
 }
 
+uint64_t ReadSystemClock() {
+  assert(false && "Not implemented.");
+  abort();
+}
+
+uint64_t SystemClockFrequency() {
+  assert(false && "Not implemented.");
+  abort();
+}
+
 }   //  namespace os
 }   //  namespace rocr