measure.h File Reference

Distributed measurement infrastructure to measure the runtime costs of tensor programs. These functions are responsible for building the tvm module, uploading it to remote devices, recording the running time costs, and checking the correctness of the output. More...

#include <tvm/auto_scheduler/loop_state.h>
#include <tvm/auto_scheduler/search_task.h>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <utility>

Include dependency graph for measure.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	tvm::auto_scheduler::MeasureInputNode
	Store the input of a measurement. More...
class	tvm::auto_scheduler::MeasureInput
	Managed reference to MeasureInputNode. More...
class	tvm::auto_scheduler::BuildResultNode
	Store the result of a build. More...
class	tvm::auto_scheduler::BuildResult
	Managed reference to BuildResultNode. More...
class	tvm::auto_scheduler::MeasureResultNode
	Store the results of a measurement. More...
class	tvm::auto_scheduler::MeasureResult
	Managed reference to MeasureResultNode. More...
class	tvm::auto_scheduler::MeasureCallbackNode
	Bass class of measurement callbacks. More...
class	tvm::auto_scheduler::MeasureCallback
	Managed reference to MeasureCallbackNode. More...
class	tvm::auto_scheduler::PythonBasedMeasureCallbackNode
	A wrapper for measure callback defined by python code This class will call functions defined in the python. More...
class	tvm::auto_scheduler::PythonBasedMeasureCallback
	Managed reference to PythonBasedMeasureCallbackNode. More...
class	tvm::auto_scheduler::ProgramBuilderNode
	ProgramBuilder that builds the programs. More...
class	tvm::auto_scheduler::ProgramBuilder
	Managed reference to ProgramBuilderNode. More...
class	tvm::auto_scheduler::ProgramRunnerNode
	ProgramRunner that runs the built programs and measure the time cost. More...
class	tvm::auto_scheduler::ProgramRunner
	Managed reference to ProgramRunnerNode. More...
class	tvm::auto_scheduler::LocalBuilderNode
	LocalBuilder use local CPU cores to build programs in parallel. More...
class	tvm::auto_scheduler::LocalBuilder
	Managed reference to LocalBuilderNode. More...
class	tvm::auto_scheduler::LocalRunnerNode
	LocalRunner that uses local CPU/GPU to measure the time cost of programs. More...
class	tvm::auto_scheduler::LocalRunner
	Managed reference to LocalRunnerNode. More...
class	tvm::auto_scheduler::RPCRunnerNode
	RPCRunner that uses RPC call to measures the time cost of programs on remote devices. Or sometime we may need to use RPC even in local running to insulate the thread environment. (e.g. running CUDA programs) More...
class	tvm::auto_scheduler::RPCRunner
	Managed reference to RPCRunnerNode. More...
class	tvm::auto_scheduler::ProgramMeasurerNode
	Measurer that measures the time costs of tvm programs This class combines ProgramBuilder and ProgramRunner, and provides a simpler API. More...
class	tvm::auto_scheduler::ProgramMeasurer
	Managed reference to ProgramMeasurerNode. More...

Namespaces
	tvm
	runtime implementation for LibTorch/TorchScript.
	tvm::auto_scheduler

Enumerations
enum	tvm::auto_scheduler::MeasureErrorNO : int {   tvm::auto_scheduler::MeasureErrorNO::kNoError = 0, tvm::auto_scheduler::MeasureErrorNO::kInstantiationError = 1, tvm::auto_scheduler::MeasureErrorNO::kCompileHostError = 2, tvm::auto_scheduler::MeasureErrorNO::kCompileDeviceError = 3,   tvm::auto_scheduler::MeasureErrorNO::kRuntimeDeviceError = 4, tvm::auto_scheduler::MeasureErrorNO::kWrongAnswerError = 5, tvm::auto_scheduler::MeasureErrorNO::kBuildTimeoutError = 6, tvm::auto_scheduler::MeasureErrorNO::kRunTimeoutError = 7,   tvm::auto_scheduler::MeasureErrorNO::kUnknownError = 8 }
	The error code of one measurement. More...

Detailed Description

Distributed measurement infrastructure to measure the runtime costs of tensor programs. These functions are responsible for building the tvm module, uploading it to remote devices, recording the running time costs, and checking the correctness of the output.

The measurement is separated into two steps: build and run. A builder builds the executable binary files and a runner runs the binary files to get the measurement results. The flow of data structures is

            `ProgramBuilder`                 `ProgramRunner`

MeasureInput --------------—> BuildResult -------------—> MeasureResult

The core functions is implemented in python to utilize python's multiprocessing and error handling (see also python/tvm/auto_scheduler/measure.py). This c++ file is just a wrapper for the python functions.