Support for QuRT OS (Hexagon RTOS) (#1497)

include/benchmark/benchmark.h

@@ -1534,8 +1534,15 @@ class Fixture : public internal::Benchmark {
 #endif
 
 // Helper macro to create a main routine in a test that runs the benchmarks
+// Note the workaround for Hexagon simulator passing argc != 0, argv = NULL.
 #define BENCHMARK_MAIN()                                                \
   int main(int argc, char** argv) {                                     \
+    char arg0_default[] = "benchmark";                                  \
+    char* args_default = arg0_default;                                  \
+    if (!argv) {                                                        \
+      argc = 1;                                                         \
+      argv = &args_default;                                             \
+    }                                                                   \
     ::benchmark::Initialize(&argc, argv);                               \
     if (::benchmark::ReportUnrecognizedArguments(argc, argv)) return 1; \
     ::benchmark::RunSpecifiedBenchmarks();                              \

src/benchmark.cc

@@ -19,7 +19,7 @@
 #include "internal_macros.h"
 
 #ifndef BENCHMARK_OS_WINDOWS
-#ifndef BENCHMARK_OS_FUCHSIA
+#if !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 #include <sys/resource.h>
 #endif
 #include <sys/time.h>

src/benchmark_register.cc

@@ -15,7 +15,7 @@
 #include "benchmark_register.h"
 
 #ifndef BENCHMARK_OS_WINDOWS
-#ifndef BENCHMARK_OS_FUCHSIA
+#if !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 #include <sys/resource.h>
 #endif
 #include <sys/time.h>

src/benchmark_runner.cc

@@ -19,7 +19,7 @@
 #include "internal_macros.h"
 
 #ifndef BENCHMARK_OS_WINDOWS
-#ifndef BENCHMARK_OS_FUCHSIA
+#if !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 #include <sys/resource.h>
 #endif
 #include <sys/time.h>

src/cycleclock.h

@@ -212,6 +212,10 @@ inline BENCHMARK_ALWAYS_INLINE int64_t Now() {
   struct timeval tv;
   gettimeofday(&tv, nullptr);
   return static_cast<int64_t>(tv.tv_sec) * 1000000 + tv.tv_usec;
+#elif defined(__hexagon__)
+  uint64_t pcycle;
+  asm volatile("%0 = C15:14" : "=r"(pcycle));
+  return static_cast<double>(pcycle);
 #else
 // The soft failover to a generic implementation is automatic only for ARM.
 // For other platforms the developer is expected to make an attempt to create

src/internal_macros.h

@@ -85,6 +85,8 @@
 #define BENCHMARK_OS_QNX 1
 #elif defined(__MVS__)
 #define BENCHMARK_OS_ZOS 1
+#elif defined(__hexagon__)
+#define BENCHMARK_OS_QURT 1
 #endif
 
 #if defined(__ANDROID__) && defined(__GLIBCXX__)

src/reporter.cc

@@ -36,7 +36,11 @@ void BenchmarkReporter::PrintBasicContext(std::ostream *out,
   BM_CHECK(out) << "cannot be null";
   auto &Out = *out;
 
+#ifndef BENCHMARK_OS_QURT
+  // Date/time information is not available on QuRT.
+  // Attempting to get it via this call cause the binary to crash.
   Out << LocalDateTimeString() << "\n";
+#endif
 
   if (context.executable_name)
     Out << "Running " << context.executable_name << "\n";

src/sysinfo.cc

@@ -23,7 +23,7 @@
 #include <codecvt>
 #else
 #include <fcntl.h>
-#ifndef BENCHMARK_OS_FUCHSIA
+#if !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 #include <sys/resource.h>
 #endif
 #include <sys/time.h>
@@ -42,6 +42,9 @@
 #if defined(BENCHMARK_OS_QNX)
 #include <sys/syspage.h>
 #endif
+#if defined(BENCHMARK_OS_QURT)
+#include <qurt.h>
+#endif
 
 #include <algorithm>
 #include <array>
@@ -402,6 +405,8 @@ std::vector<CPUInfo::CacheInfo> GetCacheSizes() {
   return GetCacheSizesWindows();
 #elif defined(BENCHMARK_OS_QNX)
   return GetCacheSizesQNX();
+#elif defined(BENCHMARK_OS_QURT)
+  return std::vector<CPUInfo::CacheInfo>();
 #else
   return GetCacheSizesFromKVFS();
 #endif
@@ -432,7 +437,15 @@ std::string GetSystemName() {
   str = std::string(converted.data());
 #endif
   return str;
-#else  // defined(BENCHMARK_OS_WINDOWS)
+#elif defined(BENCHMARK_OS_QURT)
+  std::string str = "Hexagon DSP";
+  qurt_arch_version_t arch_version_struct;
+  if (qurt_sysenv_get_arch_version(&arch_version_struct) == QURT_EOK) {
+    str += " v";
+    str += std::to_string(arch_version_struct.arch_version);
+  }
+  return str;
+#else
 #ifndef HOST_NAME_MAX
 #ifdef BENCHMARK_HAS_SYSCTL  // BSD/Mac Doesnt have HOST_NAME_MAX defined
 #define HOST_NAME_MAX 64
@@ -479,6 +492,12 @@ int GetNumCPUs() {
   return num_cpu;
 #elif defined(BENCHMARK_OS_QNX)
   return static_cast<int>(_syspage_ptr->num_cpu);
+#elif defined(BENCHMARK_OS_QURT)
+  qurt_sysenv_max_hthreads_t hardware_threads;
+  if (qurt_sysenv_get_max_hw_threads(&hardware_threads) != QURT_EOK) {
+    hardware_threads.max_hthreads = 1;
+  }
+  return hardware_threads.max_hthreads;
 #else
   int num_cpus = 0;
   int max_id = -1;
@@ -678,6 +697,9 @@ double GetCPUCyclesPerSecond(CPUInfo::Scaling scaling) {
 #elif defined(BENCHMARK_OS_QNX)
   return static_cast<double>((int64_t)(SYSPAGE_ENTRY(cpuinfo)->speed) *
                              (int64_t)(1000 * 1000));
+#elif defined(BENCHMARK_OS_QURT)
+  // QuRT doesn't provide any API to query Hexagon frequency.
+  return 1000000000;
 #endif
   // If we've fallen through, attempt to roughly estimate the CPU clock rate.
   static constexpr int estimate_time_ms = 1000;

src/timers.cc

@@ -23,7 +23,7 @@
 #include <windows.h>
 #else
 #include <fcntl.h>
-#ifndef BENCHMARK_OS_FUCHSIA
+#if !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 #include <sys/resource.h>
 #endif
 #include <sys/time.h>
@@ -38,6 +38,9 @@
 #include <mach/mach_port.h>
 #include <mach/thread_act.h>
 #endif
+#if defined(BENCHMARK_OS_QURT)
+#include <qurt.h>
+#endif
 #endif
 
 #ifdef BENCHMARK_OS_EMSCRIPTEN
@@ -79,7 +82,7 @@ double MakeTime(FILETIME const& kernel_time, FILETIME const& user_time) {
           static_cast<double>(user.QuadPart)) *
          1e-7;
 }
-#elif !defined(BENCHMARK_OS_FUCHSIA)
+#elif !defined(BENCHMARK_OS_FUCHSIA) && !defined(BENCHMARK_OS_QURT)
 double MakeTime(struct rusage const& ru) {
   return (static_cast<double>(ru.ru_utime.tv_sec) +
           static_cast<double>(ru.ru_utime.tv_usec) * 1e-6 +
@@ -119,6 +122,10 @@ double ProcessCPUUsage() {
                       &user_time))
     return MakeTime(kernel_time, user_time);
   DiagnoseAndExit("GetProccessTimes() failed");
+#elif defined(BENCHMARK_OS_QURT)
+  return static_cast<double>(
+             qurt_timer_timetick_to_us(qurt_timer_get_ticks())) *
+         1.0e-6;
 #elif defined(BENCHMARK_OS_EMSCRIPTEN)
   // clock_gettime(CLOCK_PROCESS_CPUTIME_ID, ...) returns 0 on Emscripten.
   // Use Emscripten-specific API. Reported CPU time would be exactly the
@@ -149,6 +156,10 @@ double ThreadCPUUsage() {
   GetThreadTimes(this_thread, &creation_time, &exit_time, &kernel_time,
                  &user_time);
   return MakeTime(kernel_time, user_time);
+#elif defined(BENCHMARK_OS_QURT)
+  return static_cast<double>(
+             qurt_timer_timetick_to_us(qurt_timer_get_ticks())) *
+         1.0e-6;
 #elif defined(BENCHMARK_OS_MACOSX)
   // FIXME We want to use clock_gettime, but its not available in MacOS 10.11.
   // See https://github.com/google/benchmark/pull/292