sham::gpu_core_timeline_profilier Class Reference

This class implement the GPU core timeline tool from the original algorithm of A. Richermoz, F. Neyret 2024. More...

#include <shambackends/include/shambackends/gpu_core_timeline.hpp>

Classes
struct	acc
struct	local_access_t

Public Member Functions
	gpu_core_timeline_profilier (const sham::DeviceScheduler_ptr &dev_sched, u32 max_event_count)
	CTOR.
bool	is_available_on_device ()
	Check if gpu_core_timeline_profilier is available on the device.
void	setFrameStartClock ()
	Recover the current device time in the frame_start_clock buffer.
u64	get_base_clock_value ()
acc	get_write_access (sham::EventList &deps)
	Get a write access to the timeline events and the event count.
void	complete_event_state (sycl::event e)

Detailed Description

This class implement the GPU core timeline tool from the original algorithm of A. Richermoz, F. Neyret 2024.

This is a utility to profile the execution of kernels on a GPU. It provides an interface to extract the execution timeline of each work-groups.

Note

see the test file gpu_core_timelineTest.cpp for example usage

Definition at line 77 of file gpu_core_timeline.hpp.

Constructor & Destructor Documentation

gpu_core_timeline_profilier()

sham::gpu_core_timeline_profilier::gpu_core_timeline_profilier ( const sham::DeviceScheduler_ptr &  dev_sched, u32  max_event_count  )

inline

CTOR.

Definition at line 86 of file gpu_core_timeline.hpp.

Here is the call graph for this function:

Member Function Documentation

complete_event_state()

void sham::gpu_core_timeline_profilier::complete_event_state ( sycl::event  e )

inline

Completes the event state of the timeline events and the event count. This function is necessary to ensure that all events are properly registered after a kernel.

Parameters

e	The event to wait on.

Definition at line 302 of file gpu_core_timeline.hpp.

Here is the call graph for this function:

get_base_clock_value()

u64 sham::gpu_core_timeline_profilier::get_base_clock_value ( )

inline

Definition at line 158 of file gpu_core_timeline.hpp.

get_write_access()

acc sham::gpu_core_timeline_profilier::get_write_access ( sham::EventList &  deps )

inline

Get a write access to the timeline events and the event count.

Parameters

deps	The event list to wait on.

Returns

A structure that contains a write access to the timeline events, the write access to the event count and the size of the timeline events.

Definition at line 288 of file gpu_core_timeline.hpp.

Here is the call graph for this function:

is_available_on_device()

bool sham::gpu_core_timeline_profilier::is_available_on_device ( )

inline

Check if gpu_core_timeline_profilier is available on the device.

This function checks if the current device supports both sham::intrisics::get_device_clock and sham::intrisics::get_sm_id. If not, a warning message is logged.

This function is lazy, it will only check if the function is available on the first call.

Returns

true if gpu_core_timeline_profilier is available, false otherwise

Definition at line 105 of file gpu_core_timeline.hpp.

Here is the call graph for this function:

setFrameStartClock()

void sham::gpu_core_timeline_profilier::setFrameStartClock ( )

inline

Recover the current device time in the frame_start_clock buffer.

Definition at line 143 of file gpu_core_timeline.hpp.

Here is the call graph for this function:

---

The documentation for this class was generated from the following file:

• shambackends/include/shambackends/gpu_core_timeline.hpp