GPU (Nvidia): detect GPU core usage

src/detection/gpu/gpu_driver_specific.h

@@ -42,6 +42,7 @@ typedef struct FFGpuDriverResult
     uint32_t* coreCount;
     FFGPUType* type;
     uint32_t* frequency;
+    uint8_t* usage;
 } FFGpuDriverResult;
 
 const char* ffDetectNvidiaGpuInfo(const FFGpuDriverCondition* cond, FFGpuDriverResult result, const char* soName);

src/detection/gpu/gpu_nvidia.c

@@ -12,6 +12,7 @@ struct FFNvmlData {
     FF_LIBRARY_SYMBOL(nvmlDeviceGetMemoryInfo_v2)
     FF_LIBRARY_SYMBOL(nvmlDeviceGetNumGpuCores)
     FF_LIBRARY_SYMBOL(nvmlDeviceGetMaxClockInfo)
+    FF_LIBRARY_SYMBOL(nvmlDeviceGetUtilizationRates)
     FF_LIBRARY_SYMBOL(nvmlDeviceGetBrand)
 
     bool inited;
@@ -35,6 +36,7 @@ const char* ffDetectNvidiaGpuInfo(const FFGpuDriverCondition* cond, FFGpuDriverR
         FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetMemoryInfo_v2)
         FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetNumGpuCores)
         FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetMaxClockInfo)
+        FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetUtilizationRates)
         FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetBrand)
 
         if (ffnvmlInit_v2() != NVML_SUCCESS)
@@ -124,6 +126,13 @@ const char* ffDetectNvidiaGpuInfo(const FFGpuDriverCondition* cond, FFGpuDriverR
     if (result.frequency)
         nvmlData.ffnvmlDeviceGetMaxClockInfo(device, NVML_CLOCK_GRAPHICS, result.frequency);
 
+    if (result.usage)
+    {
+        nvmlUtilization_t rates;
+        if (nvmlData.ffnvmlDeviceGetUtilizationRates(device, &rates) == NVML_SUCCESS)
+            *result.usage = (uint8_t) rates.gpu;
+    }
+
     return NULL;
 
 #else

src/detection/gpu/nvml.h

@@ -94,6 +94,16 @@ typedef enum {
     NVML_BRAND_COUNT,
 } nvmlBrandType_t;
 
+// https://docs.nvidia.com/deploy/nvml-api/structnvmlUtilization__t.html#structnvmlUtilization__t
+// Utilization information for a device.
+typedef struct
+{
+    // Percent of time over the past second during which one or more kernels was executing on the GPU
+    unsigned int gpu;
+    // Percent of time over the past second during which global (device) memory was being read or written
+    unsigned int memory;
+} nvmlUtilization_t;
+
 // https://docs.nvidia.com/deploy/nvml-api/group__nvmlInitializationAndCleanup.html#group__nvmlInitializationAndCleanup
 // Initialize NVML, but don't initialize any GPUs yet
 nvmlReturn_t nvmlInit_v2(void);
@@ -119,3 +129,5 @@ extern nvmlReturn_t nvmlDeviceGetNumGpuCores(nvmlDevice_t device, unsigned int*
 extern nvmlReturn_t nvmlDeviceGetMaxClockInfo(nvmlDevice_t device, nvmlClockType_t type, unsigned int* clock);
 // Retrieves the brand of this device
 extern nvmlReturn_t nvmlDeviceGetBrand(nvmlDevice_t device, nvmlBrandType_t* type);
+// Retrieves the current utilization rates for the device
+extern nvmlReturn_t nvmlDeviceGetUtilizationRates(nvmlDevice_t device, nvmlUtilization_t *utilization);