ktt::KernelResult Class Reference

#include <KernelResult.h>

Public Member Functions
	KernelResult ()
	KernelResult (const std::string &kernelName, const KernelConfiguration &configuration, const std::string &timestamp)
	KernelResult (const std::string &kernelName, const KernelConfiguration &configuration, const std::vector< ComputationResult > &results, const std::string &timestamp)
void	SetStatus (const ResultStatus status)
void	SetExtraDuration (const Nanoseconds duration)
void	SetExtraOverhead (const Nanoseconds overhead)
void	SetDataMovementOverhead (const Nanoseconds overhead)
void	SetValidationOverhead (const Nanoseconds overhead)
void	SetSearcherOverhead (const Nanoseconds overhead)
void	SetFailedKernelOverhead (const Nanoseconds overhead)
void	SetProfilingRunsOverhead (const Nanoseconds overhead)
void	SetProfilingOverhead (const Nanoseconds overhead)
void	SetPreciseMeasurementOverhead (const Nanoseconds overhead)
const std::string &	GetKernelName () const
const std::vector< ComputationResult > &	GetResults () const
const std::string &	GetTimestamp () const
const KernelConfiguration &	GetConfiguration () const
ResultStatus	GetStatus () const
Nanoseconds	GetKernelDuration () const
Nanoseconds	GetKernelOverheadFromCompResults () const
Nanoseconds	GetKernelOverhead () const
Nanoseconds	GetKernelOverheadFirstPass () const
Nanoseconds	GetCompilationOverheadFromCompResults () const
Nanoseconds	GetExtraDuration () const
Nanoseconds	GetExtraDurationFirstPass () const
Nanoseconds	GetExtraOverhead () const
Nanoseconds	GetDataMovementOverhead () const
Nanoseconds	GetValidationOverhead () const
Nanoseconds	GetSearcherOverhead () const
Nanoseconds	GetFailedKernelOverhead () const
Nanoseconds	GetProfilingRunsOverhead () const
Nanoseconds	GetProfilingInfrastructureOverhead () const
Nanoseconds	GetProfilingOverheadFromCompResults () const
Nanoseconds	GetPreciseMeasurementOverheadFromCompResults () const
Nanoseconds	GetProfilingOverhead () const
Nanoseconds	GetPreciseMeasurementOverhead () const
Nanoseconds	GetProfilingTotalOverhead () const
Nanoseconds	GetCompilationOverhead () const
Nanoseconds	GetTotalDuration () const
Nanoseconds	GetTotalOverhead () const
bool	IsValid () const
bool	HasRemainingProfilingRuns () const
void	FuseProfilingTimes (const KernelResult &previousResult, bool first)
void	CopyProfilingTimes (const KernelResult &originalResult)
void	TransferPowerData (const KernelResult &previousResult)

Detailed Description

Class which holds aggregate result data from kernel computation such as individual computation results, configuration and status.

Constructor & Destructor Documentation

KernelResult() [1/3]

ktt::KernelResult::KernelResult

(

)

Constructor which creates empty invalid kernel result.

KernelResult() [2/3]

ktt::KernelResult::KernelResult ( const std::string &  kernelName, const KernelConfiguration &  configuration, const std::string &  timestamp  )

explicit

Constructor which creates kernel result for the specified kernel and configuration.

Parameters

kernelName	Name of a kernel tied to the result.
configuration	Configuration tied to the result.
timestamp	Timestamp tied to this kernel run.

KernelResult() [3/3]

ktt::KernelResult::KernelResult ( const std::string &  kernelName, const KernelConfiguration &  configuration, const std::vector< ComputationResult > &  results, const std::string &  timestamp  )

explicit

Constructor which creates kernel result for the specified kernel and configuration. Also fills partial results.

Parameters

kernelName	Name of a kernel tied to the result.
configuration	Configuration tied to the result.
results	Partial results from all kernel definition runs under the kernel launch.
timestamp	Timestamp tied to the kernel run.

Member Function Documentation

CopyProfilingTimes()

void ktt::KernelResult::CopyProfilingTimes

(

const KernelResult &

originalResult

)

Copy overhead times collected from another KernelResult instance

Parameters

originalResult

source of overhead times

FuseProfilingTimes()

void ktt::KernelResult::FuseProfilingTimes	(	const KernelResult &	previousResult,
		bool	first
	)

Fuse overhead times collected from multiple profiling executions

Parameters

previousResult	result of the previous run
first	sets whether the first instance of the kernel configuration was computed (we need to fuse overhead time, but do not count it as overhead of profiling).

GetCompilationOverhead()

Nanoseconds ktt::KernelResult::GetCompilationOverhead

(

)

const

Retrieves duration of kernels compilation.

Returns

Duration of kernels compilation.

GetCompilationOverheadFromCompResults()

Nanoseconds ktt::KernelResult::GetCompilationOverheadFromCompResults

(

)

const

Retrieves total kernel compilation overhead from the partial results.

Returns

Total kernel compilation overhead from the partial results.

GetConfiguration()

const KernelConfiguration & ktt::KernelResult::GetConfiguration

(

)

const

Retrieves kernel configuration tied to the result.

Returns

Kernel configuration tied to the result.

GetDataMovementOverhead()

Nanoseconds ktt::KernelResult::GetDataMovementOverhead

(

)

const

Retrieves duration of buffer transfers (e.g., between host and device memory).

Returns

Duration of buffer transfers.

GetExtraDuration()

Nanoseconds ktt::KernelResult::GetExtraDuration

(

)

const

Retrieves duration of a kernel launcher. The duration of buffer transfers performed within the launcher is not included.

Returns

Kernel launcher duration.

GetExtraDurationFirstPass()

Nanoseconds ktt::KernelResult::GetExtraDurationFirstPass

(

)

const

Retrieves extra duration of a kernel launcher from the first (non-profiling) pass.

Returns

Kernel launcher extra duration from the first pass.

GetExtraOverhead()

Nanoseconds ktt::KernelResult::GetExtraOverhead

(

)

const

Retrieves duration of buffer transfers (e.g., between host and device memory).

Deprecated:

Use GetDataMovementOverhead() instead.

Returns

Duration of buffer transfers.

GetFailedKernelOverhead()

Nanoseconds ktt::KernelResult::GetFailedKernelOverhead

(

)

const

Retrieves duration of failed (uncompilable) kernel creation.

Returns

Duration of KTT/compiler spent on preparation of kernel which cannot be executed.

GetKernelDuration()

Nanoseconds ktt::KernelResult::GetKernelDuration

(

)

const

Retrieves total kernel duration from the partial results.

Returns

Total kernel duration from the partial results.

GetKernelName()

const std::string & ktt::KernelResult::GetKernelName

(

)

const

Returns name of a kernel tied to the result.

Returns

Name of a kernel tied to the result.

GetKernelOverhead()

Nanoseconds ktt::KernelResult::GetKernelOverhead

(

)

const

Retrieves total kernel overhead.

Returns

Total kernel overhead.

GetKernelOverheadFirstPass()

Nanoseconds ktt::KernelResult::GetKernelOverheadFirstPass

(

)

const

Retrieves kernel overhead of the first pass under the same configuration.

Returns

Kernel overhead of the first pass under the same configuration.

GetKernelOverheadFromCompResults()

Nanoseconds ktt::KernelResult::GetKernelOverheadFromCompResults

(

)

const

Retrieves total kernel overhead from the partial results.

Returns

Total kernel overhead from the partial results.

GetPreciseMeasurementOverhead()

Nanoseconds ktt::KernelResult::GetPreciseMeasurementOverhead

(

)

const

Retrieves duration of the precise multi-run measurement loop.

Returns

Duration of the precise multi-run measurement loop.

GetPreciseMeasurementOverheadFromCompResults()

Nanoseconds ktt::KernelResult::GetPreciseMeasurementOverheadFromCompResults

(

)

const

Retrieves the sum of precise multi-run measurement overhead from the partial results.

Returns

Sum of precise measurement overhead across all partial results.

GetProfilingInfrastructureOverhead()

Nanoseconds ktt::KernelResult::GetProfilingInfrastructureOverhead

(

)

const

Retrieves duration of CUPTI initialization and data collection, but without data movement and extra duration, for all profiling passes.

Returns

Duration of profiling infrastructure, but without data movement and extra duration.

GetProfilingOverhead()

Nanoseconds ktt::KernelResult::GetProfilingOverhead

(

)

const

Retrieves duration of all non-kernel operations performed during collection performance counters (e.g., CUPTI infrastructure and data movements for extra kernel runs) for all profiling runs.

Returns

Duration operations different than kernel execution needed to collect performance counters.

GetProfilingOverheadFromCompResults()

Nanoseconds ktt::KernelResult::GetProfilingOverheadFromCompResults

(

)

const

Retrieves duration of all non-kernel operations performed during collection performance counters (e.g., data movements for extra kernel runs) from the partial results.

Returns

Duration of operations different than kernel execution needed to collect performance counters from the partial results.

GetProfilingRunsOverhead()

Nanoseconds ktt::KernelResult::GetProfilingRunsOverhead

(

)

const

Retrieves duration and overhead of kernels executed to collect performance counters.

Returns

Duration of kernels executed just to collect performance counters.

GetProfilingTotalOverhead()

Nanoseconds ktt::KernelResult::GetProfilingTotalOverhead

(

)

const

Retrieves duration and overhead of all operations needed to collect performance counters.

Returns

Duration of operations required just to collect performance counters.

GetResults()

const std::vector< ComputationResult > & ktt::KernelResult::GetResults

(

)

const

Retrieves partial results from computations performed as part of the kernel launch.

Returns

Partial results from computations. See ComputationResult for more information.

GetSearcherOverhead()

Nanoseconds ktt::KernelResult::GetSearcherOverhead

(

)

const

Retrieves duration of searcher finding the next configuration to run.

Returns

Duration of searcher finding the next configuration to run.

GetStatus()

ResultStatus ktt::KernelResult::GetStatus

(

)

const

Retrieves status of the kernel result.

Returns

Status of the kernel result. See ResultStatus for more information.

GetTimestamp()

const std::string & ktt::KernelResult::GetTimestamp

(

)

const

Returns the timestamp corresponding to the kernel run

Returns

Timestamp of the kernel run

GetTotalDuration()

Nanoseconds ktt::KernelResult::GetTotalDuration

(

)

const

Retrieves the sum of kernel duration and extra duration.

Returns

Sum of kernel duration and extra duration.

GetTotalOverhead()

Nanoseconds ktt::KernelResult::GetTotalOverhead

(

)

const

Retrieves the sum of kernel overhead, data movement, validation, searcher, precise measurement and profiling related overhead.

Returns

The sum of kernel overhead, data movement, validation, searcher, precise measurement and profiling related overhead.

GetValidationOverhead()

Nanoseconds ktt::KernelResult::GetValidationOverhead

(

)

const

Retrieves duration of kernel output validation.

Returns

Duration of kernel output validation.

HasRemainingProfilingRuns()

bool ktt::KernelResult::HasRemainingProfilingRuns

(

)

const

Checks whether more kernel runs under the corresponding configuration need to be performed before all of the partial results have profiling data with valid information.

Returns

True if more kernel runs need to be performed under the same configuration. False otherwise.

IsValid()

bool ktt::KernelResult::IsValid

(

)

const

Checks whether kernel result is valid. I.e., its status has value Ok.

Returns

True if kernel result is valid. False otherwise.

SetDataMovementOverhead()

void ktt::KernelResult::SetDataMovementOverhead

(

const Nanoseconds

overhead

)

Sets duration of buffer transfers (e.g., between host and device memory).

Parameters

overhead

Duration of buffer transfers.

SetExtraDuration()

void ktt::KernelResult::SetExtraDuration

(

const Nanoseconds

duration

)

Sets duration of a kernel launcher. The duration of buffer transfers performed within the launcher is not included.

Parameters

duration

Kernel launcher duration.

SetExtraOverhead()

void ktt::KernelResult::SetExtraOverhead

(

const Nanoseconds

overhead

)

Sets duration of buffer transfers (e.g., between host and device memory).

Deprecated:

Use SetDataMovementOverhead() instead.

Parameters

overhead

Duration of buffer transfers.

SetFailedKernelOverhead()

void ktt::KernelResult::SetFailedKernelOverhead

(

const Nanoseconds

overhead

)

Sets duration of failed (uncompilable) kernel creation.

Parameters

overhead

Duration of KTT/compiler spent on preparation of kernel which cannot be executed.

SetPreciseMeasurementOverhead()

void ktt::KernelResult::SetPreciseMeasurementOverhead

(

const Nanoseconds

overhead

)

Sets duration of the precise multi-run measurement loop.

Parameters

overhead

Duration of the precise multi-run measurement loop.

SetProfilingOverhead()

void ktt::KernelResult::SetProfilingOverhead

(

const Nanoseconds

overhead

)

Sets time of additional overhead (data movement and extra duration) to collect performance counters.

Parameters

overhead

Additional overhead of kernels executed just to collect performance counters.

SetProfilingRunsOverhead()

void ktt::KernelResult::SetProfilingRunsOverhead

(

const Nanoseconds

overhead

)

Sets time of kernels executed to collect performance counters.

Parameters

overhead

Duration of kernels executed just to collect performance counters.

SetSearcherOverhead()

void ktt::KernelResult::SetSearcherOverhead

(

const Nanoseconds

overhead

)

Sets duration of searcher finding the next configuration to run.

Parameters

overhead

Duration of searcher finding the next configuration to run.

SetStatus()

void ktt::KernelResult::SetStatus

(

const ResultStatus

status

)

Sets kernel results status.

Parameters

status

Result status. See ResultStatus for more information.

SetValidationOverhead()

void ktt::KernelResult::SetValidationOverhead

(

const Nanoseconds

overhead

)

Sets duration of kernel output validation.

Parameters

overhead

Duration of kernel output validation.

TransferPowerData()

void ktt::KernelResult::TransferPowerData

(

const KernelResult &

previousResult

)

Transfer power data collected from previous execution

Parameters

previousResult

result of the previous run

---

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

• Source/Api/Output/KernelResult.h