support older CUDA versions

SC-SGS · Dec 3, 2024 · cadcd4b · cadcd4b
1 parent bac9573
commit cadcd4b
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 0 deletions.
diff --git a/include/hws/gpu_nvidia/utility.hpp b/include/hws/gpu_nvidia/utility.hpp
@@ -52,13 +52,17 @@ namespace hws::detail {
     #define HWS_CUDA_ERROR_CHECK(cuda_func) cuda_func;
 #endif
 
+#if CUDA_VERSION >= 12000
+
 /**
  * @brief Convert the clock throttle reason event bitmask to a string representation. If the provided bitmask represents multiple reasons, they are split using "|".
  * @param[in] clocks_event_reasons the bitmask to convert to a string
  * @return all event throttle reasons (`[[nodiscard]]`)
  */
 [[nodiscard]] std::string throttle_event_reason_to_string(unsigned long long clocks_event_reasons);
 
+#endif
+
 }  // namespace hws::detail
 
 #endif  // HWS_GPU_NVIDIA_UTILITY_HPP_
diff --git a/src/hws/gpu_nvidia/hardware_sampler.cpp b/src/hws/gpu_nvidia/hardware_sampler.cpp
@@ -267,11 +267,13 @@ void gpu_nvidia_hardware_sampler::sampling_loop() {
             clock_samples_.memory_clock_frequency_ = decltype(clock_samples_.memory_clock_frequency_)::value_type{ static_cast<decltype(clock_samples_.memory_clock_frequency_)::value_type::value_type>(clock_mem) };
         }
 
+#if CUDA_VERSION >= 12000
         decltype(clock_samples_.throttle_reason_)::value_type::value_type clock_throttle_reason{};
         if (nvmlDeviceGetCurrentClocksEventReasons(device, &clock_throttle_reason) == NVML_SUCCESS) {
             clock_samples_.throttle_reason_ = decltype(clock_samples_.throttle_reason_)::value_type{ clock_throttle_reason };
             clock_samples_.throttle_reason_string_ = decltype(clock_samples_.throttle_reason_string_)::value_type{ detail::throttle_event_reason_to_string(clock_throttle_reason) };
         }
+#endif
 
         nvmlEnableState_t mode{};
         nvmlEnableState_t default_mode{};
@@ -464,12 +466,14 @@ void gpu_nvidia_hardware_sampler::sampling_loop() {
                     clock_samples_.memory_clock_frequency_->push_back(static_cast<decltype(clock_samples_.memory_clock_frequency_)::value_type::value_type>(value));
                 }
 
+#if CUDA_VERSION >= 12000
                 if (clock_samples_.throttle_reason_string_.has_value()) {
                     decltype(clock_samples_.throttle_reason_)::value_type::value_type value{};
                     HWS_NVML_ERROR_CHECK(nvmlDeviceGetCurrentClocksEventReasons(device, &value))
                     clock_samples_.throttle_reason_->push_back(value);
                     clock_samples_.throttle_reason_string_->push_back(detail::throttle_event_reason_to_string(value));
                 }
+#endif
 
                 if (clock_samples_.auto_boosted_clock_.has_value()) {
                     nvmlEnableState_t mode{};

diff --git a/src/hws/gpu_nvidia/utility.cpp b/src/hws/gpu_nvidia/utility.cpp
@@ -16,6 +16,8 @@
 
 namespace hws::detail {
 
+#if CUDA_VERSION >= 12000
+
 std::string throttle_event_reason_to_string(const unsigned long long clocks_event_reasons) {
     if (clocks_event_reasons == 0ull) {
         return "None";
@@ -52,4 +54,6 @@ std::string throttle_event_reason_to_string(const unsigned long long clocks_even
     }
 }
 
+#endif
+
 }  // namespace hws::detail