Improve compatibility with Hexagon hardware (#1785)

The customization done via BENCHMARK_OS_QURT works just fine with the Hexagon simulator, but on at least some Hexagon hardware, both `qurt_timer_get_ticks()` and `std::chrono::now()` are broken and always return 0. This fixes the former by using the better-supported (and essentially identical `qurt_sysclock_get_hw_ticks()` call, and the latter by reading a 19.2MHz hardware counter (per suggestion from Qualcomm). Local testing seems to indicate these changes are just as robust under the simulator as before.

src/timers.cc

@@ -126,8 +126,12 @@ double ProcessCPUUsage() {
     return MakeTime(kernel_time, user_time);
   DiagnoseAndExit("GetProccessTimes() failed");
 #elif defined(BENCHMARK_OS_QURT)
+  // Note that qurt_timer_get_ticks() is no longer documented as of SDK 5.3.0,
+  // and doesn't appear to work on at least some devices (eg Samsung S22),
+  // so let's use the actually-documented and apparently-equivalent
+  // qurt_sysclock_get_hw_ticks() call instead.
   return static_cast<double>(
-             qurt_timer_timetick_to_us(qurt_timer_get_ticks())) *
+             qurt_timer_timetick_to_us(qurt_sysclock_get_hw_ticks())) *
          1.0e-6;
 #elif defined(BENCHMARK_OS_EMSCRIPTEN)
   // clock_gettime(CLOCK_PROCESS_CPUTIME_ID, ...) returns 0 on Emscripten.
@@ -160,8 +164,12 @@ double ThreadCPUUsage() {
                  &user_time);
   return MakeTime(kernel_time, user_time);
 #elif defined(BENCHMARK_OS_QURT)
+  // Note that qurt_timer_get_ticks() is no longer documented as of SDK 5.3.0,
+  // and doesn't appear to work on at least some devices (eg Samsung S22),
+  // so let's use the actually-documented and apparently-equivalent
+  // qurt_sysclock_get_hw_ticks() call instead.
   return static_cast<double>(
-             qurt_timer_timetick_to_us(qurt_timer_get_ticks())) *
+             qurt_timer_timetick_to_us(qurt_sysclock_get_hw_ticks())) *
          1.0e-6;
 #elif defined(BENCHMARK_OS_MACOSX)
   // FIXME We want to use clock_gettime, but its not available in MacOS 10.11.

src/timers.h

@@ -15,6 +15,29 @@ double ChildrenCPUUsage();
 // Return the CPU usage of the current thread
 double ThreadCPUUsage();
 
+#if defined(BENCHMARK_OS_QURT)
+
+// std::chrono::now() can return 0 on some Hexagon devices;
+// this reads the value of a 56-bit, 19.2MHz hardware counter
+// and converts it to seconds. Unlike std::chrono, this doesn't
+// return an absolute time, but since ChronoClockNow() is only used
+// to compute elapsed time, this shouldn't matter.
+struct QuRTClock {
+  typedef uint64_t rep;
+  typedef std::ratio<1, 19200000> period;
+  typedef std::chrono::duration<rep, period> duration;
+  typedef std::chrono::time_point<QuRTClock> time_point;
+  static const bool is_steady = false;
+
+  static time_point now() {
+    unsigned long long count;
+    asm volatile(" %0 = c31:30 " : "=r"(count));
+    return time_point(static_cast<duration>(count));
+  }
+};
+
+#else
+
 #if defined(HAVE_STEADY_CLOCK)
 template <bool HighResIsSteady = std::chrono::high_resolution_clock::is_steady>
 struct ChooseSteadyClock {
@@ -25,10 +48,14 @@ template <>
 struct ChooseSteadyClock<false> {
   typedef std::chrono::steady_clock type;
 };
+#endif  // HAVE_STEADY_CLOCK
+
 #endif
 
 struct ChooseClockType {
-#if defined(HAVE_STEADY_CLOCK)
+#if defined(BENCHMARK_OS_QURT)
+  typedef QuRTClock type;
+#elif defined(HAVE_STEADY_CLOCK)
   typedef ChooseSteadyClock<>::type type;
 #else
   typedef std::chrono::high_resolution_clock type;