PetscLogGpuTimeBegin#
Start timer for device
Synopsis#
#include "petscsys.h"
PetscErrorCode PetscLogGpuTimeBegin(void)
Notes#
When GPU is enabled, the timer is run on the GPU, it is a separate logging of time devoted to GPU computations (excluding kernel launch times).
When GPU is not available, the timer is run on the CPU, it is a separate logging of time devoted to GPU computations (including kernel launch times).
There is no need to call WaitForCUDA() or WaitForHIP() between PetscLogGpuTimeBegin() and
PetscLogGpuTimeEnd()
This timer should NOT include times for data transfers between the GPU and CPU, nor setup actions such as allocating space.
The regular logging captures the time for data transfers and any CPU activities during the event. It is used to compute the flop rate on the GPU as it is actively engaged in running a kernel.
Developer Notes#
The GPU event timer captures the execution time of all the kernels launched in the default
stream by the CPU between PetscLogGpuTimeBegin() and PetsLogGpuTimeEnd().
PetscLogGpuTimeBegin() and PetsLogGpuTimeEnd() insert the begin and end events into the
default stream (stream 0). The device will record a time stamp for the event when it reaches
that event in the stream. The function xxxEventSynchronize() is called in
PetsLogGpuTimeEnd() to block CPU execution, but not continued GPU execution, until the
timer event is recorded.
See Also#
Profiling, PetscLogView(), PetscLogGpuFlops(), PetscLogGpuTimeEnd(), PetscLogGpuTime()
Level#
intermediate
Location#
Index of all Profiling routines
Table of Contents for all manual pages
Index of all manual pages