does not actually deallocate but puts it in store. make sure you have already called the destructor! e.g. with obj->~T(); pool.deallocate(obj);

detectRate finds the rate between the devices.

/////////////////////////////////////////////////////////////////////////// Time in msec. for a single run based on the total number of COMPLETED runs

getGpuCountFromHeader counts how many GPUs on the host by parsing first line of nvidia-smi topo command. Header example: GPU0 GPU1 CPU Affinit

getLogFilePrefixAndExt returns the filename part and extension part from the FileLogger's filename.

getLogLevel get the level of log

getNumaAffinityColumnIndex get the index of "NUMA Affinity" from the topology header.

getGpuNumaInformation return numa information by provided card index.

/////////////////////////////////////////////////////////////////////////// Time in msec. after start. If the stop watch is still running (i.e. there

Get xvpu limit info of the container

General GPU Device CUDA Initialization

This function returns the best GPU (with maximum GFLOPS)

handleStaleLogRun runs in a goroutine to manage post-rotation compression and removal of stale log files.

handleStaleLogRun performs compression and removal of stale log files.

init 函数在程序启动时初始化命令行参数

initContext returns a context for the request.

loadCollectorService 根据 XPU 类型加载相应的收集器服务