starpu-max/doc/doxygen/chapters/api/performance_model.doxy

/*
 * This file is part of the StarPU Handbook.
 * Copyright (C) 2009--2011  Universit@'e de Bordeaux
 * Copyright (C) 2010, 2011, 2012, 2013, 2014  Centre National de la Recherche Scientifique
 * Copyright (C) 2011, 2012 Institut National de Recherche en Informatique et Automatique
 * See the file version.doxy for copying conditions.
 */

/*! \defgroup API_Performance_Model Performance Model

\enum starpu_perfmodel_archtype
\ingroup API_Performance_Model
Enumerates the various types of architectures.

it is possible that we have multiple versions of the same kind of
workers, for instance multiple GPUs or even different CPUs within
the same machine so we do not use the archtype enum type directly
for performance models.

<ul>
<li> CPU types range within ::STARPU_CPU_DEFAULT (1 CPU),
::STARPU_CPU_DEFAULT+1 (2 CPUs), ... ::STARPU_CPU_DEFAULT +
STARPU_MAXCPUS - 1 (STARPU_MAXCPUS CPUs).
</li>
<li> CUDA types range within ::STARPU_CUDA_DEFAULT (GPU number 0),
::STARPU_CUDA_DEFAULT + 1 (GPU number 1), ..., ::STARPU_CUDA_DEFAULT +
STARPU_MAXCUDADEVS - 1 (GPU number STARPU_MAXCUDADEVS - 1).
</li>
<li> OpenCL types range within ::STARPU_OPENCL_DEFAULT (GPU number
0), ::STARPU_OPENCL_DEFAULT + 1 (GPU number 1), ...,
::STARPU_OPENCL_DEFAULT + STARPU_MAXOPENCLDEVS - 1 (GPU number
STARPU_MAXOPENCLDEVS - 1).
</ul>
\var starpu_perfmodel_archtype::STARPU_CPU_DEFAULT
\ingroup API_Performance_Model
CPU combined workers between 0 and STARPU_MAXCPUS-1
\var starpu_perfmodel_archtype::STARPU_CUDA_DEFAULT
\ingroup API_Performance_Model
CUDA workers
\var starpu_perfmodel_archtype::STARPU_OPENCL_DEFAULT
\ingroup API_Performance_Model
OpenCL workers
\var starpu_perfmodel_archtype::STARPU_MIC_DEFAULT
\ingroup API_Performance_Model
MIC workers
\var starpu_perfmodel_archtype::STARPU_SCC_DEFAULT
\ingroup API_Performance_Model
SCC workers

\enum starpu_perfmodel_type
\ingroup API_Performance_Model
TODO
\var starpu_perfmodel_type::STARPU_PER_ARCH
\ingroup API_Performance_Model
Application-provided per-arch cost model function
\var starpu_perfmodel_type::STARPU_COMMON
\ingroup API_Performance_Model
Application-provided common cost model function, with per-arch factor
\var starpu_perfmodel_type::STARPU_HISTORY_BASED
\ingroup API_Performance_Model
Automatic history-based cost model
\var starpu_perfmodel_type::STARPU_REGRESSION_BASED
\ingroup API_Performance_Model
Automatic linear regression-based cost model  (alpha * size ^ beta)
\var starpu_perfmodel_type::STARPU_NL_REGRESSION_BASED
\ingroup API_Performance_Model
Automatic non-linear regression-based cost model (a * size ^ b + c)

\struct starpu_perfmodel
Contains all information about a performance model. At least the
type and symbol fields have to be filled when defining a performance
model for a codelet. For compatibility, make sure to initialize the
whole structure to zero, either by using explicit memset, or by
letting the compiler implicitly do it in e.g. static storage case. If
not provided, other fields have to be zero.
\ingroup API_Performance_Model
\var starpu_perfmodel::type
is the type of performance model
<ul>
<li>::STARPU_HISTORY_BASED, ::STARPU_REGRESSION_BASED,
::STARPU_NL_REGRESSION_BASED: No other fields needs to be provided,
this is purely history-based.
</li>
<li> ::STARPU_PER_ARCH: field starpu_perfmodel::per_arch has to be
filled with functions which return the cost in micro-seconds.
</li>
<li> ::STARPU_COMMON: field starpu_perfmodel::cost_function has to be
filled with a function that returns the cost in micro-seconds on a
CPU, timing on other archs will be determined by multiplying by an
arch-specific factor.
</li>
</ul>
\var starpu_perfmodel::symbol
is the symbol name for the performance model, which will be used as
file name to store the model. It must be set otherwise the model will
be ignored.
This field is deprecated. Use instead the field starpu_perfmodel::cost_function field.
\var starpu_perfmodel::cost_function
Used by ::STARPU_COMMON: takes a task and implementation number, and
must return a task duration estimation in micro-seconds.
\var starpu_perfmodel::size_base
Used by ::STARPU_HISTORY_BASED, ::STARPU_REGRESSION_BASED and
::STARPU_NL_REGRESSION_BASED. If not NULL, takes a task and
implementation number, and returns the size to be used as index to distinguish
histories and as a base for regressions.
\var starpu_perfmodel::footprint
Used by ::STARPU_HISTORY_BASED. If not NULL, takes a task and returns the
footprint to be used as index to distinguish histories. The default is to use
the starpu_task_data_footprint function.
\var starpu_perfmodel::per_arch
Used by ::STARPU_PER_ARCH: array of structures starpu_per_arch_perfmodel
\var starpu_perfmodel::is_loaded
\private
Whether the performance model is already loaded from the disk.
\var starpu_perfmodel::benchmarking
\private
Whether the performance model is still being calibrated.
\var starpu_perfmodel::model_rwlock
\private
Lock to protect concurrency between loading from disk (W), updating
the values (W), and making a performance estimation (R).

\struct starpu_perfmodel_regression_model
...
\ingroup API_Performance_Model
\var starpu_perfmodel_regression_model::sumlny
sum of ln(measured)
\var starpu_perfmodel_regression_model::sumlnx
sum of ln(size)
\var starpu_perfmodel_regression_model::sumlnx2
sum of ln(size)^2
\var starpu_perfmodel_regression_model::minx
minimum size
\var starpu_perfmodel_regression_model::maxx
maximum size
\var starpu_perfmodel_regression_model::sumlnxlny
sum of ln(size)*ln(measured)
\var starpu_perfmodel_regression_model::alpha
estimated = alpha * size ^ beta
\var starpu_perfmodel_regression_model::beta
estimated = alpha * size ^ beta
\var starpu_perfmodel_regression_model::valid
whether the linear regression model is valid (i.e. enough measures)
\var starpu_perfmodel_regression_model::a
estimated = a size ^b + c
\var starpu_perfmodel_regression_model::b
estimated = a size ^b + c
\var starpu_perfmodel_regression_model::c
estimated = a size ^b + c
\var starpu_perfmodel_regression_model::nl_valid
whether the non-linear regression model is valid (i.e. enough measures)
\var starpu_perfmodel_regression_model::nsample
number of sample values for non-linear regression

\struct starpu_perfmodel_per_arch
contains information about the performance model of a given
arch.
\ingroup API_Performance_Model
This field is deprecated. Use instead the field
starpu_perfmodel_per_arch::cost_function.
\var starpu_perfmodel_per_arch::cost_function
Used by ::STARPU_PER_ARCH, must point to functions which take a task,
the target arch and implementation number (as mere conveniency, since
the array is already indexed by these), and must return a task
duration estimation in micro-seconds.
\var starpu_perfmodel_per_arch::size_base
Same as in structure starpu_perfmodel, but per-arch, in case it
depends on the architecture-specific implementation.
\var starpu_perfmodel_per_arch::history
\private
The history of performance measurements.
\var starpu_perfmodel_per_arch::list
\private
Used by ::STARPU_HISTORY_BASED and ::STARPU_NL_REGRESSION_BASED,
records all execution history measures.
\var starpu_perfmodel_per_arch::regression
\private
Used by ::STARPU_REGRESSION_BASED and
::STARPU_NL_REGRESSION_BASED, contains the estimated factors of the
regression.

\struct starpu_perfmodel_history_list
todo
\ingroup API_Performance_Model
\var starpu_perfmodel_history_list::next
todo
\var starpu_perfmodel_history_list::entry
todo

\struct starpu_perfmodel_history_entry
todo
\ingroup API_Performance_Model
\var starpu_perfmodel_history_entry::mean
mean_n = 1/n sum
\var starpu_perfmodel_history_entry::deviation
n dev_n = sum2 - 1/n (sum)^2
\var starpu_perfmodel_history_entry::sum
sum of samples (in µs)
\var starpu_perfmodel_history_entry::sum2
sum of samples^2
\var starpu_perfmodel_history_entry::nsample
number of samples
\var starpu_perfmodel_history_entry::footprint
data footprint
\var starpu_perfmodel_history_entry::size
in bytes
\var starpu_perfmodel_history_entry::flops
Provided by the application

\fn void starpu_perfmodel_init(FILE *f, struct starpu_perfmodel *model)
\ingroup API_Performance_Model
todo

\fn int starpu_perfmodel_load_symbol(const char *symbol, struct starpu_perfmodel *model)
\ingroup API_Performance_Model
loads a given performance model. The model structure has to be
completely zero, and will be filled with the information saved in
<c>$STARPU_HOME/.starpu</c>. The function is intended to be used by
external tools that should read the performance model files.

\fn int starpu_perfmodel_unload_model(struct starpu_perfmodel *model)
\ingroup API_Performance_Model
unloads the given model which has been previously loaded
through the function starpu_perfmodel_load_symbol()

\fn void starpu_perfmodel_debugfilepath(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, char *path, size_t maxlen, unsigned nimpl)
\ingroup API_Performance_Model
returns the path to the debugging information for the performance model.

\fn char* starpu_perfmodel_get_archtype_name(enum starpu_worker_archtype archtype)
\ingroup API_Performance_Model
todo

\fn void starpu_perfmodel_get_arch_name(struct starpu_perfmodel_arch *arch, char *archname, size_t maxlen, unsigned nimpl)
\ingroup API_Performance_Model
returns the architecture name for \p arch

\fn struct starpu_perfmodel_arch *starpu_worker_get_perf_archtype(int workerid)
\ingroup API_Performance_Model
returns the architecture type of a given worker.

\fn int starpu_perfmodel_list(FILE *output)
\ingroup API_Performance_Model
prints a list of all performance models on \p output

\fn void starpu_perfmodel_directory(FILE *output)
\ingroup API_Performance_Model
prints the directory name storing performance models on \p output

\fn void starpu_perfmodel_print(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, unsigned nimpl, char *parameter, uint32_t *footprint, FILE *output)
\ingroup API_Performance_Model
todo

\fn int starpu_perfmodel_print_all(struct starpu_perfmodel *model, char *arch, char *parameter, uint32_t *footprint, FILE *output)
\ingroup API_Performance_Model
todo

\fn void starpu_bus_print_bandwidth(FILE *f)
\ingroup API_Performance_Model
prints a matrix of bus bandwidths on \p f.

\fn void starpu_bus_print_affinity(FILE *f)
\ingroup API_Performance_Model
prints the affinity devices on \p f.

\fn void starpu_perfmodel_update_history(struct starpu_perfmodel *model, struct starpu_task *task, struct starpu_perfmodel_arch *arch, unsigned cpuid, unsigned nimpl, double measured);
\ingroup API_Performance_Model
This feeds the performance model model with an explicit
measurement measured (in µs), in addition to measurements done by StarPU
itself. This can be useful when the application already has an
existing set of measurements done in good conditions, that StarPU
could benefit from instead of doing on-line measurements. And example
of use can be seen in \ref PerformanceModelExample.

\fn double starpu_transfer_bandwidth(unsigned src_node, unsigned dst_node)
\ingroup API_Performance_Model
Return the bandwidth of data transfer between two memory nodes

\fn double starpu_transfer_latency(unsigned src_node, unsigned dst_node)
\ingroup API_Performance_Model
Return the latency of data transfer between two memory nodes

\fn double starpu_transfer_predict(unsigned src_node, unsigned dst_node, size_t size)
\ingroup API_Performance_Model
Return the estimated time to transfer a given size between two memory nodes.

\fn double starpu_permodel_history_based_expected_perf(struct starpu_perfmodel *model, struct starpu_perfmodel_arch* arch, uint32_t footprint)
\ingroup API_Performance_Model
todo

*/