Disable throttling when sync exec tag is used.

docs/cli_help.md

@@ -135,13 +135,16 @@
 
 * `--throttle-threshold <value>`
   * Set the GPU throttle threshold as percentage of the device's default clock rate.
-  * Default is 75%.
+  * Default is 75.
+  * Set to 0 to disable throttle detection entirely.
+  * Note that throttling is disabled when `nvbench::exec_tag::sync` is used.
   * Applies to the most recent `--benchmark`, or all benchmarks if specified
     before any `--benchmark` arguments.
 
 * `--throttle-recovery-delay <value>`
   * Set the GPU throttle recovery delay in seconds.
   * Default is 0.05 seconds.
+  * Note that throttling is disabled when `nvbench::exec_tag::sync` is used.
   * Applies to the most recent `--benchmark`, or all benchmarks if specified
     before any `--benchmark` arguments.
 

nvbench/detail/measure_cold.cu

@@ -43,6 +43,7 @@ measure_cold_base::measure_cold_base(state &exec_state)
         exec_state.get_stopping_criterion())}
     , m_disable_blocking_kernel{exec_state.get_disable_blocking_kernel()}
     , m_run_once{exec_state.get_run_once()}
+    , m_check_throttling(!exec_state.get_run_once() && exec_state.get_throttle_threshold() > 0.f)
     , m_min_samples{exec_state.get_min_samples()}
     , m_skip_time{exec_state.get_skip_time()}
     , m_timeout{exec_state.get_timeout()}
@@ -94,7 +95,7 @@ void measure_cold_base::run_trials_prologue() { m_walltime_timer.start(); }
 
 void measure_cold_base::record_measurements()
 {
-  if (!m_run_once)
+  if (m_check_throttling)
   {
     const auto current_clock_rate = m_gpu_frequency.get_clock_frequency();
     const auto default_clock_rate =

nvbench/detail/measure_cold.cuh

@@ -95,6 +95,7 @@ protected:
 
   bool m_disable_blocking_kernel{false};
   bool m_run_once{false};
+  bool m_check_throttling;
 
   nvbench::int64_t m_min_samples{};
 
@@ -142,7 +143,7 @@ struct measure_cold_base::kernel_launch_timer
     {
       m_measure.block_stream();
     }
-    if (!m_measure.m_run_once)
+    if (m_measure.m_check_throttling)
     {
       m_measure.gpu_frequency_start();
     }
@@ -161,7 +162,7 @@ struct measure_cold_base::kernel_launch_timer
       m_measure.m_cpu_timer.start();
       m_measure.unblock_stream();
     }
-    if (!m_measure.m_run_once)
+    if (m_measure.m_check_throttling)
     {
       m_measure.gpu_frequency_stop();
     }

nvbench/detail/state_exec.cuh

@@ -101,10 +101,12 @@ void state::exec(ExecTags tags, KernelLauncher &&kernel_launcher)
                              "`set_is_cpu_only(true)` is NOT called when defining the benchmark.");
   }
 
-  // Syncing will cause the blocking kernel pattern to deadlock:
   if constexpr (modifier_tags & sync)
   {
+    // Syncing will cause the blocking kernel pattern to deadlock:
     this->set_disable_blocking_kernel(true);
+    // Syncing will cause the throttle frequency measurements to be skewed heavily:
+    this->set_throttle_threshold(0.f);
   }
 
   if (this->is_skipped())