unsigned long long
getTotalEccErrors( nvmlDevice_t dev, nvmlEccBitType_t bits)
{
unsigned long long counts = 0;
nvmlReturn_t bad;
bad = nvmlDeviceGetTotalEccErrors( dev, bits, NVML_VOLATILE_ECC , &counts);
if ( NVML_SUCCESS != bad ) {
SUBDBG( "something went wrong %s\n", nvmlErrorString(bad));
}
return counts;
}
int readNvmlCounters(HSP *sp, SFLHost_gpu_nvml *nvml) {
unsigned int i;
if(sp->nvml.gpu_count == 0) {
return NO;
}
// pick up latest value of accumulators
nvml->gpu_time = sp->nvml.nvml_gpu_time;
nvml->mem_time = sp->nvml.nvml_mem_time;
nvml->energy = sp->nvml.nvml_energy;
// and fill in the rest of the counters/gauges too
nvml->device_count = sp->nvml.gpu_count;
// zero these, and sum across all GPUs
nvml->mem_total = 0;
nvml->mem_free = 0;
nvml->ecc_errors = 0;
nvml->processes = 0;
// use the max across all GPUs
nvml->temperature = 0;
nvml->fan_speed = 0;
for (i = 0; i < sp->nvml.gpu_count; ++i) {
unsigned long long eccErrors;
unsigned int temp;
nvmlDevice_t gpu;
unsigned int speed;
unsigned int procs;
nvmlMemory_t memInfo;
nvmlReturn_t result;
if (NVML_SUCCESS != nvmlDeviceGetHandleByIndex(i, &gpu)) {
return NO;
}
if (NVML_SUCCESS == nvmlDeviceGetMemoryInfo(gpu, &memInfo)) {
nvml->mem_total += memInfo.total;
nvml->mem_free += memInfo.free;
}
if (NVML_SUCCESS == nvmlDeviceGetTotalEccErrors(gpu, NVML_SINGLE_BIT_ECC, NVML_VOLATILE_ECC, &eccErrors)) {
nvml->ecc_errors += eccErrors;
}
if (NVML_SUCCESS == nvmlDeviceGetTotalEccErrors(gpu, NVML_DOUBLE_BIT_ECC, NVML_VOLATILE_ECC, &eccErrors)) {
nvml->ecc_errors += eccErrors;
}
if (NVML_SUCCESS == nvmlDeviceGetTemperature(gpu, NVML_TEMPERATURE_GPU, &temp)) {
if (nvml->temperature < temp) {
nvml->temperature = temp;
}
}
if (NVML_SUCCESS == nvmlDeviceGetFanSpeed(gpu, &speed)) {
if (nvml->fan_speed < speed) {
nvml->fan_speed = speed;
}
}
result = nvmlDeviceGetComputeRunningProcesses(gpu, &procs, NULL);
if (NVML_SUCCESS == result || NVML_ERROR_INSUFFICIENT_SIZE == result) {
nvml->processes += procs;
}
}
return YES;
}
