Class CudaEventTimer

• Defined in File CudaEventTimer.h

Class Documentation

class CudaEventTimer

A timer based on GPU events

Public Types

enum class State

State of the CudaEventTimer, which can be either READY, RECODING or STEPPED

• READY: the timer is ready for starting a recording (cannot stop the timer)

• RECODING: the timer is recording (can not start a new recording or getting the elapsed time)

• STOPPED: the timer has stopped, but hasn’t sync with the GPU to get the elapsed time yet (cannot stop again or start a new recording)

When more than one device is included, the elapsed time will be the maximum time among all devices.

Values:

enumerator READY

enumerator RECORDING

enumerator STOPPED

Public Functions

inline explicit CudaEventTimer(const std::vector<Index> &deviceIndices)

Initialize a timer with the specified set of devices through their indices

CudaEventTimer(const CudaEventTimer&) = delete

CudaEventTimer &operator=(const CudaEventTimer&) = delete

inline CudaEventTimer(CudaEventTimer &&other) noexcept

inline CudaEventTimer &operator=(CudaEventTimer &&other) noexcept

inline ~CudaEventTimer()

inline State currentState() const

Get the current state of the timer

See also

mass::CudaEventTimer::State

inline void startTimer()

Starts the timer

The state must be READY, will transit to RECORDING

inline void stopTimer()

Stops the timer

The state must be RECORDING, will transit to STOPPED

inline float getElapsedTime()

Gets the elapsed time in milliseconds

The state must be either STOPPED or READY, will transit to READY.

• If the state is STOPPED, will first synchronize with the devices to get the elapsed time.

• If the state is READY, will return the elapsed time get from previous record