xpu::timings Class Reference

Timing information collected via xpu::push_timer and xpu::pop_timer. More...

Public Member Functions
	timings ()=default
std::string_view	name () const
double	wall () const
double	copy (direction dir) const
double	memset () const
template<typename K >
kernel_timings	kernel () const
std::vector< kernel_timings >	kernels () const
double	kernel_time () const
std::vector< timings >	children () const
bool	has_details () const
double	throughput () const
double	throughput_kernels () const
double	throughput_copy (direction dir) const
double	throughput_memset () const
void	merge (const timings &other)
	timings (detail::timings t)

Detailed Description

Timing information collected via xpu::push_timer and xpu::pop_timer.

See also

xpu::push_timer, xpu::pop_timer, xpu::kernel_timings

Constructor & Destructor Documentation

timings() [1/2]

xpu::timings::timings ( )

default

timings() [2/2]

xpu::timings::timings ( detail::timings  t )

inlineexplicit

Member Function Documentation

children()

std::vector<timings> xpu::timings::children

(

)

const

Returns all child timers.

copy()

double xpu::timings::copy ( direction  dir ) const

inline

Time spent in memcpy operations.

Parameters

dir	Direction of the copy.

Note

Requires profiling to enabled when calling xpu::initialize.

has_details()

bool xpu::timings::has_details ( ) const

inline

Returns true if copy, memset and kernel timings were collected.

kernel()

template<typename K >

kernel_timings xpu::timings::kernel ( ) const

inline

Time spent in kernel executions.

Note

Requires profiling to enabled when calling xpu::initialize.

kernel_time()

double xpu::timings::kernel_time ( ) const

inline

Returns

The total time spent in kernels. [ms]

kernels()

std::vector<kernel_timings> xpu::timings::kernels

(

)

const

Returns all kernel timings for this timer.

Note

Requires profiling to enabled when calling xpu::initialize.

memset()

double xpu::timings::memset ( ) const

inline

Time spent in memset operations.

Note

Requires profiling to enabled when calling xpu::initialize.

merge()

void xpu::timings::merge ( const timings &  other )

inline

Merges the given timings into this one.

name()

std::string_view xpu::timings::name ( ) const

inline

Name of this timer.

throughput()

double xpu::timings::throughput

(

)

const

Returns the total throughput. [GB/s] Input size is calculated via xpu::t_add_bytes .

throughput_copy()

double xpu::timings::throughput_copy

(

direction

dir

)

const

Returns the throughput of copy operations in the given direction, [GB/s]

throughput_kernels()

double xpu::timings::throughput_kernels

(

)

const

Returns the throughput of kernel executions. [GB/s]

throughput_memset()

double xpu::timings::throughput_memset

(

)

const

Returns the throughput of memset operations. [GB/s]

wall()

double xpu::timings::wall ( ) const

inline

Total (wall) time spent in this timer.

Note

This time is always collected, regardless of profiling being enabled or not.

---

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

• host.h