starpu-max/doc/chapters/api.texi

@c -*-texinfo-*-

@c This file is part of the StarPU Handbook.
@c Copyright (C) 2009--2011  Universit@'e de Bordeaux 1
@c Copyright (C) 2010, 2011, 2012, 2013  Centre National de la Recherche Scientifique
@c Copyright (C) 2011, 2012 Institut National de Recherche en Informatique et Automatique
@c See the file starpu.texi for copying conditions.

@menu
* Versioning::
* Initialization and Termination::
* Standard memory library::
* Workers' Properties::
* Data Management::
* Data Interfaces::
* Data Partition::
* Multiformat Data Interface::
* Codelets and Tasks::
* Insert Task::
* Explicit Dependencies::
* Implicit Data Dependencies::
* Performance Model API::
* Profiling API::
* Theoretical lower bound on execution time API::
* CUDA extensions::
* OpenCL extensions::
* MIC extensions::
* SCC extensions::
* Miscellaneous helpers::
* FXT Support::
* FFT Support::
* MPI::
* Task Bundles::
* Task Lists::
* Using Parallel Tasks::
* Scheduling Contexts::
* Scheduling Policy::
* Running drivers::
* Expert mode::
@end menu

@node Versioning
@section Versioning

@defmac STARPU_MAJOR_VERSION
Define the major version of StarPU
@end defmac

@defmac STARPU_MINOR_VERSION
Define the minor version of StarPU
@end defmac

@defmac STARPU_RELEASE_VERSION
Define the release version of StarPU
@end defmac

@deftypefun void starpu_get_version (int *@var{major}, int *@var{minor}, int *@var{release})
Return as 3 integers the release version of StarPU.
@end deftypefun

@node Initialization and Termination
@section Initialization and Termination

@deftp {Data Type} {struct starpu_driver}
@table @asis
@item @code{enum starpu_worker_archtype type}
The type of the driver. Only STARPU_CPU_DRIVER, STARPU_CUDA_DRIVER and
STARPU_OPENCL_DRIVER are currently supported.
@item @code{union id} Anonymous union
@table @asis
@item @code{unsigned cpu_id}
Should only be used if type is STARPU_CPU_WORKER.
@item @code{unsigned cuda_id}
Should only be used if type is STARPU_CUDA_WORKER.
@item @code{cl_device_id opencl_id}
Should only be used if type is STARPU_OPENCL_WORKER.
@end table
@end table
@end deftp

@deftp {Data Type} {struct starpu_conf}
This structure is passed to the @code{starpu_init} function in order
to configure StarPU. It has to be initialized with @code{starpu_conf_init}.
When the default value is used, StarPU automatically selects the number of
processing units and takes the default scheduling policy. The environment
variables overwrite the equivalent parameters.

@table @asis
@item @code{const char *sched_policy_name} (default = NULL)
This is the name of the scheduling policy. This can also be specified
with the @code{STARPU_SCHED} environment variable.

@item @code{struct starpu_sched_policy *sched_policy} (default = NULL)
This is the definition of the scheduling policy. This field is ignored
if @code{sched_policy_name} is set.

@item @code{int ncpus} (default = -1)
This is the number of CPU cores that StarPU can use. This can also be
specified with the @code{STARPU_NCPU} environment variable.

@item @code{int ncuda} (default = -1)
This is the number of CUDA devices that StarPU can use. This can also
be specified with the @code{STARPU_NCUDA} environment variable.

@item @code{int nopencl} (default = -1)
This is the number of OpenCL devices that StarPU can use. This can
also be specified with the @code{STARPU_NOPENCL} environment variable.

@item @code{int nmic} (default = -1)
This is the number of MIC devices that StarPU can use. This can
also be specified with the @code{STARPU_NMIC} environment variable.

@item @code{int nscc} (default = -1)
This is the number of SCC devices that StarPU can use. This can
also be specified with the @code{STARPU_NSCC} environment variable.

@item @code{unsigned use_explicit_workers_bindid} (default = 0)
If this flag is set, the @code{workers_bindid} array indicates where the
different workers are bound, otherwise StarPU automatically selects where to
bind the different workers. This can also be specified with the
@code{STARPU_WORKERS_CPUID} environment variable.

@item @code{unsigned workers_bindid[STARPU_NMAXWORKERS]}
If the @code{use_explicit_workers_bindid} flag is set, this array
indicates where to bind the different workers. The i-th entry of the
@code{workers_bindid} indicates the logical identifier of the
processor which should execute the i-th worker. Note that the logical
ordering of the CPUs is either determined by the OS, or provided by
the @code{hwloc} library in case it is available.

@item @code{unsigned use_explicit_workers_cuda_gpuid} (default = 0)
If this flag is set, the CUDA workers will be attached to the CUDA devices
specified in the @code{workers_cuda_gpuid} array. Otherwise, StarPU affects the
CUDA devices in a round-robin fashion. This can also be specified with the
@code{STARPU_WORKERS_CUDAID} environment variable.

@item @code{unsigned workers_cuda_gpuid[STARPU_NMAXWORKERS]}
If the @code{use_explicit_workers_cuda_gpuid} flag is set, this array
contains the logical identifiers of the CUDA devices (as used by
@code{cudaGetDevice}).

@item @code{unsigned use_explicit_workers_opencl_gpuid} (default = 0)
If this flag is set, the OpenCL workers will be attached to the OpenCL devices
specified in the @code{workers_opencl_gpuid} array. Otherwise, StarPU affects
the OpenCL devices in a round-robin fashion. This can also be specified with
the @code{STARPU_WORKERS_OPENCLID} environment variable.

@item @code{unsigned workers_opencl_gpuid[STARPU_NMAXWORKERS]}
If the @code{use_explicit_workers_opencl_gpuid} flag is set, this array
contains the logical identifiers of the OpenCL devices to be used.

@item @code{unsigned use_explicit_workers_mic_gpuid} (default = 0)
If this flag is set, the MIC workers will be attached to the MIC devices
specified in the @code{workers_mic_gpuid} array. Otherwise, StarPU affects
the MIC devices in a round-robin fashion. This can also be specified with
the @code{STARPU_WORKERS_MICID} environment variable.

@item @code{unsigned workers_mic_gpuid[STARPU_NMAXWORKERS]}
If the @code{use_explicit_workers_mic_gpuid} flag is set, this array
contains the logical identifiers of the MIC devices to be used.

@item @code{unsigned use_explicit_workers_scc_gpuid} (default = 0)
If this flag is set, the SCC workers will be attached to the SCC devices
specified in the @code{workers_scc_gpuid} array. Otherwise, StarPU affects
the SCC devices in a round-robin fashion. This can also be specified with
the @code{STARPU_WORKERS_SCCID} environment variable.

@item @code{unsigned workers_scc_gpuid[STARPU_NMAXWORKERS]}
If the @code{use_explicit_workers_scc_gpuid} flag is set, this array
contains the logical identifiers of the SCC devices to be used.

@item @code{int calibrate} (default = 0)
If this flag is set, StarPU will calibrate the performance models when
executing tasks. If this value is equal to @code{-1}, the default value is
used. If the value is equal to @code{1}, it will force continuing
calibration. If the value is equal to @code{2}, the existing performance
models will be overwritten. This can also be specified with the
@code{STARPU_CALIBRATE} environment variable.

@item @code{int bus_calibrate} (default = 0)
If this flag is set, StarPU will recalibrate the bus.  If this value is equal
to @code{-1}, the default value is used.  This can also be specified with the
@code{STARPU_BUS_CALIBRATE} environment variable.

@item @code{int single_combined_worker} (default = 0)
By default, StarPU executes parallel tasks concurrently.
Some parallel libraries (e.g. most OpenMP implementations) however do
not support concurrent calls to parallel code. In such case, setting this flag
makes StarPU only start one parallel task at a time (but other
CPU and GPU tasks are not affected and can be run concurrently). The parallel
task scheduler will however still however still try varying combined worker
sizes to look for the most efficient ones.
This can also be specified with the @code{STARPU_SINGLE_COMBINED_WORKER} environment variable.

@item @code{mic_sink_program_path} (default = NULL)
Path to the program to execute on the MIC device, compiled for MIC
architecture. When set to NULL, StarPU automatically looks next to the host
program location.

@item @code{int disable_asynchronous_copy} (default = 0)
This flag should be set to 1 to disable asynchronous copies between
CPUs and all accelerators. This can also be specified with the
@code{STARPU_DISABLE_ASYNCHRONOUS_COPY} environment variable.
The AMD implementation of OpenCL is known to
fail when copying data asynchronously. When using this implementation,
it is therefore necessary to disable asynchronous data transfers.
This can also be specified at compilation time by giving to the
configure script the option @code{--disable-asynchronous-copy}.

@item @code{int disable_asynchronous_cuda_copy} (default = 0)
This flag should be set to 1 to disable asynchronous copies between
CPUs and CUDA accelerators. This can also be specified with the
@code{STARPU_DISABLE_ASYNCHRONOUS_CUDA_COPY} environment variable.
This can also be specified at compilation time by giving to the
configure script the option @code{--disable-asynchronous-cuda-copy}.

@item @code{int disable_asynchronous_opencl_copy} (default = 0)
This flag should be set to 1 to disable asynchronous copies between
CPUs and OpenCL accelerators. This can also be specified with the
@code{STARPU_DISABLE_ASYNCHRONOUS_OPENCL_COPY} environment variable.
The AMD implementation of OpenCL is known to
fail when copying data asynchronously. When using this implementation,
it is therefore necessary to disable asynchronous data transfers.
This can also be specified at compilation time by giving to the
configure script the option @code{--disable-asynchronous-opencl-copy}.

@item @code{int *cuda_opengl_interoperability} (default = NULL)
This can be set to an array of CUDA device identifiers for which
@code{cudaGLSetGLDevice} should be called instead of @code{cudaSetDevice}. Its
size is specified by the @code{n_cuda_opengl_interoperability} field below

@item @code{int *n_cuda_opengl_interoperability} (default = 0)
This has to be set to the size of the array pointed to by the
@code{cuda_opengl_interoperability} field.

@item @code{struct starpu_driver *not_launched_drivers}
The drivers that should not be launched by StarPU.

@item @code{unsigned n_not_launched_drivers}
The number of StarPU drivers that should not be launched by StarPU.

@item @code{trace_buffer_size}
Specifies the buffer size used for FxT tracing. Starting from FxT version
0.2.12, the buffer will automatically be flushed when it fills in, but it may
still be interesting to specify a bigger value to avoid any flushing (which
would disturb the trace).

@end table
@end deftp

@deftypefun int starpu_init ({struct starpu_conf *}@var{conf})
This is StarPU initialization method, which must be called prior to any other
StarPU call.  It is possible to specify StarPU's configuration (e.g. scheduling
policy, number of cores, ...) by passing a non-null argument. Default
configuration is used if the passed argument is @code{NULL}.

Upon successful completion, this function returns 0. Otherwise, @code{-ENODEV}
indicates that no worker was available (so that StarPU was not initialized).
@end deftypefun

@deftypefun int starpu_initialize ({struct starpu_conf *}@var{conf}, int @var{argc}, {char ***}@var{argv})
This is the same as @code{starpu_init}, but also takes the @code{argc} and
@code{argv} as gotten by the application. This is needed for MIC and SCC
execution so that instances of StarPU can know whether they are slaves or masters.
@end deftypefun

@deftypefun int starpu_conf_init ({struct starpu_conf *}@var{conf})
This function initializes the @var{conf} structure passed as argument
with the default values. In case some configuration parameters are already
specified through environment variables, @code{starpu_conf_init} initializes
the fields of the structure according to the environment variables. For
instance if @code{STARPU_CALIBRATE} is set, its value is put in the
@code{.calibrate} field of the structure passed as argument.

Upon successful completion, this function returns 0. Otherwise, @code{-EINVAL}
indicates that the argument was NULL.
@end deftypefun

@deftypefun void starpu_shutdown (void)
This is StarPU termination method. It must be called at the end of the
application: statistics and other post-mortem debugging information are not
guaranteed to be available until this method has been called.
@end deftypefun

@deftypefun int starpu_asynchronous_copy_disabled (void)
Return 1 if asynchronous data transfers between CPU and accelerators
are disabled.
@end deftypefun

@deftypefun int starpu_asynchronous_cuda_copy_disabled (void)
Return 1 if asynchronous data transfers between CPU and CUDA accelerators
are disabled.
@end deftypefun

@deftypefun int starpu_asynchronous_opencl_copy_disabled (void)
Return 1 if asynchronous data transfers between CPU and OpenCL accelerators
are disabled.
@end deftypefun

@node Standard memory library
@section Standard memory library

@defmac STARPU_MALLOC_PINNED
Value passed to the function @code{starpu_malloc_flags} to
indicate the memory allocation should be pinned.
@end defmac

@defmac STARPU_MALLOC_COUNT
Value passed to the function @code{starpu_malloc_flags} to
indicate the memory allocation should be in the limit defined by
the environment variables @code{STARPU_LIMIT_CUDA_devid_MEM},
@code{STARPU_LIMIT_CUDA_MEM}, @code{STARPU_LIMIT_OPENCL_devid_MEM},
@code{STARPU_LIMIT_OPENCL_MEM} and @code{STARPU_LIMIT_CPU_MEM}
(@pxref{Limit memory}). If no memory is available, it tries to reclaim
memory from StarPU. Memory allocated this way needs to be freed by
calling the @code{starpu_free_flags} function with the same flag.
@end defmac

@deftypefun int starpu_malloc_flags (void **@var{A}, size_t @var{dim}, int @var{flags})
Performs a memory allocation based on the constraints defined by the
given @var{flag}.
@end deftypefun

@deftypefun void starpu_malloc_set_align (size_t @var{align})
This functions sets an alignment constraints for @code{starpu_malloc}
allocations. @var{align} must be a power of two. This is for instance called
automatically by the OpenCL driver to specify its own alignment constraints.
@end deftypefun

@deftypefun int starpu_malloc (void **@var{A}, size_t @var{dim})
This function allocates data of the given size in main memory. It will also try to pin it in
CUDA or OpenCL, so that data transfers from this buffer can be asynchronous, and
thus permit data transfer and computation overlapping. The allocated buffer must
be freed thanks to the @code{starpu_free} function.
@end deftypefun

@deftypefun int starpu_free (void *@var{A})
This function frees memory which has previously been allocated with
@code{starpu_malloc}.
@end deftypefun

@deftypefun int starpu_free_flags (void *@var{A}, size_t @var{dim}, int @var{flags})
This function frees memory by specifying its size. The given
@var{flags} should be consistent with the ones given to
@code{starpu_malloc_flags} when allocating the memory.
@end deftypefun

@deftypefun ssize_t starpu_memory_get_available (unsigned @var{node})
If a memory limit is defined on the given node (@pxref{Limit memory}),
return the amount of available memory on the node. Otherwise return
@code{-1}.
@end deftypefun

@node Workers' Properties
@section Workers' Properties

@deftp {Data Type} {enum starpu_worker_archtype}
The different values are:
@table @asis
@item @code{STARPU_CPU_WORKER}
@item @code{STARPU_CUDA_WORKER}
@item @code{STARPU_OPENCL_WORKER}
@item @code{STARPU_MIC_WORKER}
@item @code{STARPU_SCC_WORKER}
@end table
@end deftp

@deftypefun unsigned starpu_worker_get_count (void)
This function returns the number of workers (i.e. processing units executing
StarPU tasks). The returned value should be at most @code{STARPU_NMAXWORKERS}.
@end deftypefun

@deftypefun int starpu_worker_get_count_by_type ({enum starpu_worker_archtype} @var{type})
Returns the number of workers of the given @var{type}. A positive
(or @code{NULL}) value is returned in case of success, @code{-EINVAL} indicates that
the type is not valid otherwise.
@end deftypefun

@deftypefun unsigned starpu_cpu_worker_get_count (void)
This function returns the number of CPUs controlled by StarPU. The returned
value should be at most @code{STARPU_MAXCPUS}.
@end deftypefun

@deftypefun unsigned starpu_cuda_worker_get_count (void)
This function returns the number of CUDA devices controlled by StarPU. The returned
value should be at most @code{STARPU_MAXCUDADEVS}.
@end deftypefun

@deftypefun unsigned starpu_opencl_worker_get_count (void)
This function returns the number of OpenCL devices controlled by StarPU. The returned
value should be at most @code{STARPU_MAXOPENCLDEVS}.
@end deftypefun

@deftypefun unsigned starpu_mic_worker_get_count (void)
This function returns the number of MIC workers controlled by StarPU.
@end deftypefun

@deftypefun unsigned starpu_mic_device_get_count (void)
This function returns the number of MIC devices controlled by StarPU. The returned
value should be at most @code{STARPU_MAXMICDEVS}.
@end deftypefun

@deftypefun unsigned starpu_scc_worker_get_count (void)
This function returns the number of SCC devices controlled by StarPU. The returned
value should be at most @code{STARPU_MAXSCCDEVS}.
@end deftypefun

@deftypefun int starpu_worker_get_id (void)
This function returns the identifier of the current worker, i.e the one associated to the calling
thread. The returned value is either -1 if the current context is not a StarPU
worker (i.e. when called from the application outside a task or a callback), or
an integer between 0 and @code{starpu_worker_get_count() - 1}.
@end deftypefun

@deftypefun int starpu_worker_get_ids_by_type ({enum starpu_worker_archtype} @var{type}, int *@var{workerids}, int @var{maxsize})
This function gets the list of identifiers of workers with the given
type. It fills the workerids array with the identifiers of the workers that have the type
indicated in the first argument. The maxsize argument indicates the size of the
workids array. The returned value gives the number of identifiers that were put
in the array. @code{-ERANGE} is returned is maxsize is lower than the number of
workers with the appropriate type: in that case, the array is filled with the
maxsize first elements. To avoid such overflows, the value of maxsize can be
chosen by the means of the @code{starpu_worker_get_count_by_type} function, or
by passing a value greater or equal to @code{STARPU_NMAXWORKERS}.
@end deftypefun

@deftypefun int starpu_worker_get_by_type ({enum starpu_worker_archtype} @var{type}, int @var{num})
This returns the identifier of the @var{num}-th worker that has the specified type
@var{type}. If there are no such worker, -1 is returned.
@end deftypefun

@deftypefun int starpu_worker_get_by_devid ({enum starpu_worker_archtype} @var{type}, int @var{devid})
This returns the identifier of the worker that has the specified type
@var{type} and devid @var{devid} (which may not be the n-th, if some devices are
skipped for instance). If there are no such worker, -1 is returned.
@end deftypefun

@deftypefun int starpu_worker_get_devid (int @var{id})
This functions returns the device id of the given worker. The worker
should be identified with the value returned by the @code{starpu_worker_get_id} function. In the case of a
CUDA worker, this device identifier is the logical device identifier exposed by
CUDA (used by the @code{cudaGetDevice} function for instance). The device
identifier of a CPU worker is the logical identifier of the core on which the
worker was bound; this identifier is either provided by the OS or by the
@code{hwloc} library in case it is available.
@end deftypefun

@deftypefun {enum starpu_worker_archtype} starpu_worker_get_type (int @var{id})
This function returns the type of processing unit associated to a
worker. The worker identifier is a value returned by the
@code{starpu_worker_get_id} function). The returned value
indicates the architecture of the worker: @code{STARPU_CPU_WORKER} for a CPU
core, @code{STARPU_CUDA_WORKER} for a CUDA device, and
@code{STARPU_OPENCL_WORKER} for a OpenCL device. The value returned for an invalid
identifier is unspecified.
@end deftypefun

@deftypefun void starpu_worker_get_name (int @var{id}, char *@var{dst}, size_t @var{maxlen})
This function allows to get the name of a given worker.
StarPU associates a unique human readable string to each processing unit. This
function copies at most the @var{maxlen} first bytes of the unique string
associated to a worker identified by its identifier @var{id} into the
@var{dst} buffer. The caller is responsible for ensuring that the @var{dst}
is a valid pointer to a buffer of @var{maxlen} bytes at least. Calling this
function on an invalid identifier results in an unspecified behaviour.
@end deftypefun

@deftypefun unsigned starpu_worker_get_memory_node (unsigned @var{workerid})
This function returns the identifier of the memory node associated to the
worker identified by @var{workerid}.
@end deftypefun

@deftp {Data Type} {enum starpu_node_kind}
todo
@table @asis
@item @code{STARPU_UNUSED}
@item @code{STARPU_CPU_RAM}
@item @code{STARPU_CUDA_RAM}
@item @code{STARPU_OPENCL_RAM}
@item @code{STARPU_MIC_RAM}
@item @code{STARPU_SCC_RAM}
@item @code{STARPU_SCC_SHM}
@end table
@end deftp

@deftypefun {enum starpu_node_kind} starpu_node_get_kind (unsigned @var{node})
Returns the type of the given node as defined by @code{enum
starpu_node_kind}. For example, when defining a new data interface,
this function should be used in the allocation function to determine
on which device the memory needs to be allocated.
@end deftypefun

@node Data Management
@section Data Management

@menu
* Introduction to Data Management::
* Basic Data Management API::
* Access registered data from the application::
@end menu

This section describes the data management facilities provided by StarPU.

We show how to use existing data interfaces in @ref{Data Interfaces}, but developers can
design their own data interfaces if required.

@node Introduction to Data Management
@subsection Introduction
Data management is done at a high-level in StarPU: rather than accessing a mere
list of contiguous buffers, the tasks may manipulate data that are described by
a high-level construct which we call data interface.

An example of data interface is the "vector" interface which describes a
contiguous data array on a spefic memory node. This interface is a simple
structure containing the number of elements in the array, the size of the
elements, and the address of the array in the appropriate address space (this
address may be invalid if there is no valid copy of the array in the memory
node). More informations on the data interfaces provided by StarPU are
given in @ref{Data Interfaces}.

When a piece of data managed by StarPU is used by a task, the task
implementation is given a pointer to an interface describing a valid copy of
the data that is accessible from the current processing unit.

Every worker is associated to a memory node which is a logical abstraction of
the address space from which the processing unit gets its data. For instance,
the memory node associated to the different CPU workers represents main memory
(RAM), the memory node associated to a GPU is DRAM embedded on the device.
Every memory node is identified by a logical index which is accessible from the
@code{starpu_worker_get_memory_node} function. When registering a piece of data
to StarPU, the specified memory node indicates where the piece of data
initially resides (we also call this memory node the home node of a piece of
data).

@node Basic Data Management API
@subsection Basic Data Management API

@deftp {Data Type} {enum starpu_data_access_mode}
This datatype describes a data access mode. The different available modes are:
@table @asis
@item @code{STARPU_R}: read-only mode.
@item @code{STARPU_W}: write-only mode.
@item @code{STARPU_RW}: read-write mode.
 This is equivalent to @code{STARPU_R|STARPU_W}.
@item @code{STARPU_SCRATCH}: scratch memory.
A temporary buffer is allocated for the task, but StarPU does not
enforce data consistency---i.e. each device has its own buffer,
independently from each other (even for CPUs), and no data transfer is
ever performed.  This is useful for temporary variables to avoid
allocating/freeing buffers inside each task.

Currently, no behavior is defined concerning the relation with the
@code{STARPU_R} and @code{STARPU_W} modes and the value provided at
registration---i.e., the value of the scratch buffer is undefined at
entry of the codelet function.  It is being considered for future
extensions at least to define the initial value.  For now, data to be
used in @code{SCRATCH} mode should be registered with node @code{-1} and
a @code{NULL} pointer, since the value of the provided buffer is simply
ignored for now.
@item @code{STARPU_REDUX}: reduction mode. TODO!
@end table
@end deftp

@deftp {Data Type} {starpu_data_handle_t}
StarPU uses @code{starpu_data_handle_t} as an opaque handle to manage a piece of
data. Once a piece of data has been registered to StarPU, it is associated to a
@code{starpu_data_handle_t} which keeps track of the state of the piece of data
over the entire machine, so that we can maintain data consistency and locate
data replicates for instance.
@end deftp

@deftypefun void starpu_data_register (starpu_data_handle_t *@var{handleptr}, unsigned @var{home_node}, void *@var{data_interface}, {struct starpu_data_interface_ops} *@var{ops})
Register a piece of data into the handle located at the @var{handleptr}
address. The @var{data_interface} buffer contains the initial description of the
data in the home node. The @var{ops} argument is a pointer to a structure
describing the different methods used to manipulate this type of interface. See
@ref{struct starpu_data_interface_ops} for more details on this structure.

If @code{home_node} is -1, StarPU will automatically
allocate the memory when it is used for the
first time in write-only mode. Once such data handle has been automatically
allocated, it is possible to access it using any access mode.

Note that StarPU supplies a set of predefined types of interface (e.g. vector or
matrix) which can be registered by the means of helper functions (e.g.
@code{starpu_vector_data_register} or @code{starpu_matrix_data_register}).
@end deftypefun

@deftypefun void starpu_data_register_same ({starpu_data_handle_t *}@var{handledst}, starpu_data_handle_t @var{handlesrc})
Register a new piece of data into the handle @var{handledst} with the
same interface as the handle @var{handlesrc}.
@end deftypefun

@deftypefun void starpu_data_unregister (starpu_data_handle_t @var{handle})
This function unregisters a data handle from StarPU. If the data was
automatically allocated by StarPU because the home node was -1, all
automatically allocated buffers are freed. Otherwise, a valid copy of the data
is put back into the home node in the buffer that was initially registered.
Using a data handle that has been unregistered from StarPU results in an
undefined behaviour.
@end deftypefun

@deftypefun void starpu_data_unregister_no_coherency (starpu_data_handle_t @var{handle})
This is the same as starpu_data_unregister, except that StarPU does not put back
a valid copy into the home node, in the buffer that was initially registered.
@end deftypefun

@deftypefun void starpu_data_unregister_submit (starpu_data_handle_t @var{handle})
Destroy the data handle once it is not needed anymore by any submitted
task. No coherency is assumed.
@end deftypefun

@deftypefun void starpu_data_invalidate (starpu_data_handle_t @var{handle})
Destroy all replicates of the data handle immediately. After data invalidation,
the first access to the handle must be performed in write-only mode.
Accessing an invalidated data in read-mode results in undefined
behaviour.
@end deftypefun

@deftypefun void starpu_data_invalidate_submit (starpu_data_handle_t @var{handle})
Submits invalidation of the data handle after completion of previously submitted tasks.
@end deftypefun

@c TODO create a specific sections about user interaction with the DSM ?

@deftypefun void starpu_data_set_wt_mask (starpu_data_handle_t @var{handle}, uint32_t @var{wt_mask})
This function sets the write-through mask of a given data, i.e. a bitmask of
nodes where the data should be always replicated after modification. It also
prevents the data from being evicted from these nodes when memory gets scarse.
@end deftypefun

@deftypefun int starpu_data_prefetch_on_node (starpu_data_handle_t @var{handle}, unsigned @var{node}, unsigned @var{async})
Issue a prefetch request for a given data to a given node, i.e.
requests that the data be replicated to the given node, so that it is available
there for tasks. If the @var{async} parameter is 0, the call will block until
the transfer is achieved, else the call will return as soon as the request is
scheduled (which may however have to wait for a task completion).
@end deftypefun

@deftypefun starpu_data_handle_t starpu_data_lookup ({const void *}@var{ptr})
Return the handle corresponding to the data pointed to by the @var{ptr}
host pointer.
@end deftypefun

@deftypefun int starpu_data_request_allocation (starpu_data_handle_t @var{handle}, unsigned @var{node})
Explicitly ask StarPU to allocate room for a piece of data on the specified
memory node.
@end deftypefun

@deftypefun void starpu_data_query_status (starpu_data_handle_t @var{handle}, int @var{memory_node}, {int *}@var{is_allocated}, {int *}@var{is_valid}, {int *}@var{is_requested})
Query the status of the handle on the specified memory node.
@end deftypefun

@deftypefun void starpu_data_advise_as_important (starpu_data_handle_t @var{handle}, unsigned @var{is_important})
This function allows to specify that a piece of data can be discarded
without impacting the application.
@end deftypefun

@deftypefun void starpu_data_set_reduction_methods (starpu_data_handle_t @var{handle}, {struct starpu_codelet *}@var{redux_cl}, {struct starpu_codelet *}@var{init_cl})
This sets the codelets to be used for the @var{handle} when it is accessed in
REDUX mode. Per-worker buffers will be initialized with the @var{init_cl}
codelet, and reduction between per-worker buffers will be done with the
@var{redux_cl} codelet.
@end deftypefun

@deftypefun struct starpu_data_interface_ops* starpu_data_get_interface_ops (starpu_data_handle_t @var{handle})
Get a pointer to the structure describing the different methods used
to manipulate the given data. See @ref{struct starpu_data_interface_ops} for more details on this structure.
@end deftypefun

@deftypefun unsigned starpu_data_get_sequential_consistency_flag (starpu_data_handle_t @var{handle})
Return the sequential consistency flag of the given data.
@end deftypefun

@node Access registered data from the application
@subsection Access registered data from the application

@deftypefun int starpu_data_acquire (starpu_data_handle_t @var{handle}, {enum starpu_data_access_mode} @var{mode})
The application must call this function prior to accessing registered data from
main memory outside tasks. StarPU ensures that the application will get an
up-to-date copy of the data in main memory located where the data was
originally registered, and that all concurrent accesses (e.g. from tasks) will
be consistent with the access mode specified in the @var{mode} argument.
@code{starpu_data_release} must be called once the application does not need to
access the piece of data anymore.  Note that implicit data
dependencies are also enforced by @code{starpu_data_acquire}, i.e.
@code{starpu_data_acquire} will wait for all tasks scheduled to work on
the data, unless they have been disabled explictly by calling
@code{starpu_data_set_default_sequential_consistency_flag} or
@code{starpu_data_set_sequential_consistency_flag}.
@code{starpu_data_acquire} is a blocking call, so that it cannot be called from
tasks or from their callbacks (in that case, @code{starpu_data_acquire} returns
@code{-EDEADLK}). Upon successful completion, this function returns 0.
@end deftypefun


@deftypefun int starpu_data_acquire_cb (starpu_data_handle_t @var{handle}, {enum starpu_data_access_mode} @var{mode}, void (*@var{callback})(void *), void *@var{arg})
@code{starpu_data_acquire_cb} is the asynchronous equivalent of
@code{starpu_data_acquire}. When the data specified in the first argument is
available in the appropriate access mode, the callback function is executed.
The application may access the requested data during the execution of this
callback. The callback function must call @code{starpu_data_release} once the
application does not need to access the piece of data anymore.
Note that implicit data dependencies are also enforced by
@code{starpu_data_acquire_cb} in case they are not disabled.
 Contrary to @code{starpu_data_acquire}, this function is non-blocking and may
be called from task callbacks. Upon successful completion, this function
returns 0.
@end deftypefun

@deftypefun int starpu_data_acquire_on_node (starpu_data_handle_t @var{handle}, unsigned @var{node}, {enum starpu_data_access_mode} @var{mode})
This is the same as @code{starpu_data_acquire}, except that the data will be
available on the given memory node instead of main memory.
@end deftypefun

@deftypefun int starpu_data_acquire_on_node_cb (starpu_data_handle_t @var{handle}, unsigned @var{node}, {enum starpu_data_access_mode} @var{mode}, void (*@var{callback})(void *), void *@var{arg})
This is the same as @code{starpu_data_acquire_cb}, except that the data will be
available on the given memory node instead of main memory.
@end deftypefun

@defmac STARPU_DATA_ACQUIRE_CB (starpu_data_handle_t @var{handle}, {enum starpu_data_access_mode} @var{mode}, code)
@code{STARPU_DATA_ACQUIRE_CB} is the same as @code{starpu_data_acquire_cb},
except that the code to be executed in a callback is directly provided as a
macro parameter, and the data handle is automatically released after it. This
permits to easily execute code which depends on the value of some registered
data. This is non-blocking too and may be called from task callbacks.
@end defmac

@deftypefun void starpu_data_release (starpu_data_handle_t @var{handle})
This function releases the piece of data acquired by the application either by
@code{starpu_data_acquire} or by @code{starpu_data_acquire_cb}.
@end deftypefun

@deftypefun void starpu_data_release_on_node (starpu_data_handle_t @var{handle}, unsigned @var{node})
This is the same as @code{starpu_data_release}, except that the data will be
available on the given memory node instead of main memory.
@end deftypefun

@node Data Interfaces
@section Data Interfaces

@menu
* Registering Data::
* Accessing Data Interfaces::
* Defining Interface::
@end menu

@node Registering Data
@subsection Registering Data

There are several ways to register a memory region so that it can be managed by
StarPU.  The functions below allow the registration of vectors, 2D matrices, 3D
matrices as well as  BCSR and CSR sparse matrices.

@deftypefun void starpu_void_data_register ({starpu_data_handle_t *}@var{handle})
Register a void interface. There is no data really associated to that
interface, but it may be used as a synchronization mechanism. It also
permits to express an abstract piece of data that is managed by the
application internally: this makes it possible to forbid the
concurrent execution of different tasks accessing the same "void" data
in read-write concurrently.
@end deftypefun

@deftypefun void starpu_variable_data_register ({starpu_data_handle_t *}@var{handle}, unsigned @var{home_node}, uintptr_t @var{ptr}, size_t @var{size})
Register the @var{size}-byte element pointed to by @var{ptr}, which is
typically a scalar, and initialize @var{handle} to represent this data
item.

@cartouche
@smallexample
float var;
starpu_data_handle_t var_handle;
starpu_variable_data_register(&var_handle, 0, (uintptr_t)&var, sizeof(var));
@end smallexample
@end cartouche
@end deftypefun

@deftypefun void starpu_vector_data_register ({starpu_data_handle_t *}@var{handle}, unsigned @var{home_node}, uintptr_t @var{ptr}, uint32_t @var{nx}, size_t @var{elemsize})
Register the @var{nx} @var{elemsize}-byte elements pointed to by
@var{ptr} and initialize @var{handle} to represent it.

@cartouche
@smallexample
float vector[NX];
starpu_data_handle_t vector_handle;
starpu_vector_data_register(&vector_handle, 0, (uintptr_t)vector, NX,
                            sizeof(vector[0]));
@end smallexample
@end cartouche
@end deftypefun

@deftypefun void starpu_matrix_data_register ({starpu_data_handle_t *}@var{handle}, unsigned @var{home_node}, uintptr_t @var{ptr}, uint32_t @var{ld}, uint32_t @var{nx}, uint32_t @var{ny}, size_t @var{elemsize})
Register the @var{nx}x@var{ny} 2D matrix of @var{elemsize}-byte elements
pointed by @var{ptr} and initialize @var{handle} to represent it.
@var{ld} specifies the number of elements between rows.
a value greater than @var{nx} adds padding, which can be useful for
alignment purposes.

@cartouche
@smallexample
float *matrix;
starpu_data_handle_t matrix_handle;
matrix = (float*)malloc(width * height * sizeof(float));
starpu_matrix_data_register(&matrix_handle, 0, (uintptr_t)matrix,
                            width, width, height, sizeof(float));
@end smallexample
@end cartouche
@end deftypefun

@deftypefun void starpu_block_data_register ({starpu_data_handle_t *}@var{handle}, unsigned @var{home_node}, uintptr_t @var{ptr}, uint32_t @var{ldy}, uint32_t @var{ldz}, uint32_t @var{nx}, uint32_t @var{ny}, uint32_t @var{nz}, size_t @var{elemsize})
Register the @var{nx}x@var{ny}x@var{nz} 3D matrix of @var{elemsize}-byte
elements pointed by @var{ptr} and initialize @var{handle} to represent
it.  Again, @var{ldy} and @var{ldz} specify the number of elements
between rows and between z planes.

@cartouche
@smallexample
float *block;
starpu_data_handle_t block_handle;
block = (float*)malloc(nx*ny*nz*sizeof(float));
starpu_block_data_register(&block_handle, 0, (uintptr_t)block,
                           nx, nx*ny, nx, ny, nz, sizeof(float));
@end smallexample
@end cartouche
@end deftypefun

@deftypefun void starpu_bcsr_data_register (starpu_data_handle_t *@var{handle}, unsigned @var{home_node}, uint32_t @var{nnz}, uint32_t @var{nrow}, uintptr_t @var{nzval}, uint32_t *@var{colind}, uint32_t *@var{rowptr}, uint32_t @var{firstentry}, uint32_t @var{r}, uint32_t @var{c}, size_t @var{elemsize})
This variant of @code{starpu_data_register} uses the BCSR (Blocked
Compressed Sparse Row Representation) sparse matrix interface.
Register the sparse matrix made of @var{nnz} non-zero blocks of elements of size
@var{elemsize} stored in @var{nzval} and initializes @var{handle} to represent
it. Blocks have size @var{r} * @var{c}. @var{nrow} is the number of rows (in
terms of blocks), @code{colind[i]} is the block-column index for block @code{i}
in @code{nzval}, @code{rowptr[i]} is the block-index (in nzval) of the first block of row @code{i}.
@var{firstentry} is the index of the first entry of the given arrays (usually 0
or 1).
@end deftypefun

@deftypefun void starpu_csr_data_register (starpu_data_handle_t *@var{handle}, unsigned @var{home_node}, uint32_t @var{nnz}, uint32_t @var{nrow}, uintptr_t @var{nzval}, uint32_t *@var{colind}, uint32_t *@var{rowptr}, uint32_t @var{firstentry}, size_t @var{elemsize})
This variant of @code{starpu_data_register} uses the CSR (Compressed
Sparse Row Representation) sparse matrix interface.
TODO
@end deftypefun

@deftypefun void starpu_coo_data_register (starpu_data_handle_t *@var{handleptr}, unsigned @var{home_node}, uint32_t @var{nx}, uint32_t @var{ny}, uint32_t @var{n_values}, uint32_t *@var{columns}, uint32_t *@var{rows}, uintptr_t @var{values}, size_t @var{elemsize});
Register the @var{nx}x@var{ny} 2D matrix given in the COO format, using the
@var{columns}, @var{rows}, @var{values} arrays, which must have @var{n_values}
elements of size @var{elemsize}. Initialize @var{handleptr}.
@end deftypefun

@deftypefun {void *} starpu_data_get_interface_on_node (starpu_data_handle_t @var{handle}, unsigned @var{memory_node})
Return the interface associated with @var{handle} on @var{memory_node}.
@end deftypefun

@node Accessing Data Interfaces
@subsection Accessing Data Interfaces

Each data interface is provided with a set of field access functions.
The ones using a @code{void *} parameter aimed to be used in codelet
implementations (see for example the code in @ref{Vector Scaling Using StarPU's API}).

@deftp {Data Type} {enum starpu_data_interface_id}
The different values are:
@table @asis
@item @code{STARPU_MATRIX_INTERFACE_ID}
@item @code{STARPU_BLOCK_INTERFACE_ID}
@item @code{STARPU_VECTOR_INTERFACE_ID}
@item @code{STARPU_CSR_INTERFACE_ID}
@item @code{STARPU_BCSR_INTERFACE_ID}
@item @code{STARPU_VARIABLE_INTERFACE_ID}
@item @code{STARPU_VOID_INTERFACE_ID}
@item @code{STARPU_MULTIFORMAT_INTERFACE_ID}
@item @code{STARPU_COO_INTERCACE_ID}
@item @code{STARPU_NINTERFACES_ID}: number of data interfaces
@end table
@end deftp

@menu
* Accessing Handle::
* Accessing Variable Data Interfaces::
* Accessing Vector Data Interfaces::
* Accessing Matrix Data Interfaces::
* Accessing Block Data Interfaces::
* Accessing BCSR Data Interfaces::
* Accessing CSR Data Interfaces::
* Accessing COO Data Interfaces::
@end menu

@node Accessing Handle
@subsubsection Handle

@deftypefun {void *} starpu_data_handle_to_pointer (starpu_data_handle_t @var{handle}, unsigned @var{node})
Return the pointer associated with @var{handle} on node @var{node} or
@code{NULL} if @var{handle}'s interface does not support this
operation or data for this handle is not allocated on that node.
@end deftypefun

@deftypefun {void *} starpu_data_get_local_ptr (starpu_data_handle_t @var{handle})
Return the local pointer associated with @var{handle} or @code{NULL}
if @var{handle}'s interface does not have data allocated locally
@end deftypefun

@deftypefun {enum starpu_data_interface_id} starpu_data_get_interface_id (starpu_data_handle_t @var{handle})
Return the unique identifier of the interface associated with the given @var{handle}.
@end deftypefun

@deftypefun size_t starpu_data_get_size (starpu_data_handle_t @var{handle})
Return the size of the data associated with @var{handle}
@end deftypefun

@deftypefun int starpu_data_pack (starpu_data_handle_t @var{handle}, {void **}@var{ptr}, {starpu_ssize_t *}@var{count})
Execute the packing operation of the interface of the data registered
at @var{handle} (@pxref{struct starpu_data_interface_ops}). This
packing operation must allocate a buffer large enough at @var{ptr} and
copy into the newly allocated buffer the data associated to
@var{handle}. @var{count} will be set to the size of the allocated
buffer.
If @var{ptr} is @code{NULL}, the function should not copy the data in the
buffer but just set @var{count} to the size of the buffer which
would have been allocated. The special value @code{-1} indicates the
size is yet unknown.
@end deftypefun

@deftypefun int starpu_data_unpack (starpu_data_handle_t @var{handle}, {void *}@var{ptr}, size_t @var{count})
Unpack in @var{handle} the data located at @var{ptr} of size
@var{count} as described by the interface of the data. The interface
registered at @var{handle} must define a unpacking operation
(@pxref{struct starpu_data_interface_ops}). The memory at the address @code{ptr}
is freed after calling the data unpacking operation.
@end deftypefun

@node Accessing Variable Data Interfaces
@subsubsection Variable Data Interfaces

@deftypefun size_t starpu_variable_get_elemsize (starpu_data_handle_t @var{handle})
Return the size of the variable designated by @var{handle}.
@end deftypefun

@deftypefun uintptr_t starpu_variable_get_local_ptr (starpu_data_handle_t @var{handle})
Return a pointer to the variable designated by @var{handle}.
@end deftypefun

@defmac STARPU_VARIABLE_GET_PTR ({void *}@var{interface})
Return a pointer to the variable designated by @var{interface}.
@end defmac

@defmac STARPU_VARIABLE_GET_ELEMSIZE ({void *}@var{interface})
Return the size of the variable designated by @var{interface}.
@end defmac

@defmac STARPU_VARIABLE_GET_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the variable designated by @var{interface}, to be
used on OpenCL. The offset documented below has to be used in addition to this.
@end defmac

@defmac STARPU_VARIABLE_GET_OFFSET ({void *}@var{interface})
Return the offset in the variable designated by @var{interface}, to be used
with the device handle.
@end defmac

@node Accessing Vector Data Interfaces
@subsubsection Vector Data Interfaces

@deftypefun uint32_t starpu_vector_get_nx (starpu_data_handle_t @var{handle})
Return the number of elements registered into the array designated by @var{handle}.
@end deftypefun

@deftypefun size_t starpu_vector_get_elemsize (starpu_data_handle_t @var{handle})
Return the size of each element of the array designated by @var{handle}.
@end deftypefun

@deftypefun uintptr_t starpu_vector_get_local_ptr (starpu_data_handle_t @var{handle})
Return the local pointer associated with @var{handle}.
@end deftypefun

@defmac STARPU_VECTOR_GET_PTR ({void *}@var{interface})
Return a pointer to the array designated by @var{interface}, valid on CPUs and
CUDA only. For OpenCL, the device handle and offset need to be used instead.
@end defmac

@defmac STARPU_VECTOR_GET_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the array designated by @var{interface}, to be used on OpenCL. the offset
documented below has to be used in addition to this.
@end defmac

@defmac STARPU_VECTOR_GET_OFFSET ({void *}@var{interface})
Return the offset in the array designated by @var{interface}, to be used with the device handle.
@end defmac

@defmac STARPU_VECTOR_GET_NX ({void *}@var{interface})
Return the number of elements registered into the array designated by @var{interface}.
@end defmac

@defmac STARPU_VECTOR_GET_ELEMSIZE ({void *}@var{interface})
Return the size of each element of the array designated by @var{interface}.
@end defmac

@node Accessing Matrix Data Interfaces
@subsubsection Matrix Data Interfaces

@deftypefun uint32_t starpu_matrix_get_nx (starpu_data_handle_t @var{handle})
Return the number of elements on the x-axis of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_matrix_get_ny (starpu_data_handle_t @var{handle})
Return the number of elements on the y-axis of the matrix designated by
@var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_matrix_get_local_ld (starpu_data_handle_t @var{handle})
Return the number of elements between each row of the matrix designated by
@var{handle}. Maybe be equal to nx when there is no padding.
@end deftypefun

@deftypefun uintptr_t starpu_matrix_get_local_ptr (starpu_data_handle_t @var{handle})
Return the local pointer associated with @var{handle}.
@end deftypefun

@deftypefun size_t starpu_matrix_get_elemsize (starpu_data_handle_t @var{handle})
Return the size of the elements registered into the matrix designated by
@var{handle}.
@end deftypefun

@defmac STARPU_MATRIX_GET_PTR ({void *}@var{interface})
Return a pointer to the matrix designated by @var{interface}, valid on CPUs and
CUDA devices only. For OpenCL devices, the device handle and offset need to be
used instead.
@end defmac

@defmac STARPU_MATRIX_GET_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the matrix designated by @var{interface}, to be used
on OpenCL. The offset documented below has to be used in addition to this.
@end defmac

@defmac STARPU_MATRIX_GET_OFFSET ({void *}@var{interface})
Return the offset in the matrix designated by @var{interface}, to be used with
the device handle.
@end defmac

@defmac STARPU_MATRIX_GET_NX ({void *}@var{interface})
Return the number of elements on the x-axis of the matrix designated by
@var{interface}.
@end defmac

@defmac STARPU_MATRIX_GET_NY ({void *}@var{interface})
Return the number of elements on the y-axis of the matrix designated by
@var{interface}.
@end defmac

@defmac STARPU_MATRIX_GET_LD ({void *}@var{interface})
Return the number of elements between each row of the matrix designated by
@var{interface}. May be equal to nx when there is no padding.
@end defmac

@defmac STARPU_MATRIX_GET_ELEMSIZE ({void *}@var{interface})
Return the size of the elements registered into the matrix designated by
@var{interface}.
@end defmac

@node Accessing Block Data Interfaces
@subsubsection Block Data Interfaces

@deftypefun uint32_t starpu_block_get_nx (starpu_data_handle_t @var{handle})
Return the number of elements on the x-axis of the block designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_block_get_ny (starpu_data_handle_t @var{handle})
Return the number of elements on the y-axis of the block designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_block_get_nz (starpu_data_handle_t @var{handle})
Return the number of elements on the z-axis of the block designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_block_get_local_ldy (starpu_data_handle_t @var{handle})
Return the number of elements between each row of the block designated by
@var{handle}, in the format of the current memory node.
@end deftypefun

@deftypefun uint32_t starpu_block_get_local_ldz (starpu_data_handle_t @var{handle})
Return the number of elements between each z plane of the block designated by
@var{handle}, in the format of the current memory node.
@end deftypefun

@deftypefun uintptr_t starpu_block_get_local_ptr (starpu_data_handle_t @var{handle})
Return the local pointer associated with @var{handle}.
@end deftypefun

@deftypefun size_t starpu_block_get_elemsize (starpu_data_handle_t @var{handle})
Return the size of the elements of the block designated by @var{handle}.
@end deftypefun

@defmac STARPU_BLOCK_GET_PTR ({void *}@var{interface})
Return a pointer to the block designated by @var{interface}.
@end defmac

@defmac STARPU_BLOCK_GET_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the block designated by @var{interface}, to be used
on OpenCL. The offset document below has to be used in addition to this.
@end defmac

@defmac STARPU_BLOCK_GET_OFFSET ({void *}@var{interface})
Return the offset in the block designated by @var{interface}, to be used with
the device handle.
@end defmac

@defmac STARPU_BLOCK_GET_NX ({void *}@var{interface})
Return the number of elements on the x-axis of the block designated by @var{handle}.
@end defmac

@defmac STARPU_BLOCK_GET_NY ({void *}@var{interface})
Return the number of elements on the y-axis of the block designated by @var{handle}.
@end defmac

@defmac STARPU_BLOCK_GET_NZ ({void *}@var{interface})
Return the number of elements on the z-axis of the block designated by @var{handle}.
@end defmac

@defmac STARPU_BLOCK_GET_LDY ({void *}@var{interface})
Return the number of elements between each row of the block designated by
@var{interface}. May be equal to nx when there is no padding.
@end defmac

@defmac STARPU_BLOCK_GET_LDZ ({void *}@var{interface})
Return the number of elements between each z plane of the block designated by
@var{interface}. May be equal to nx*ny when there is no padding.
@end defmac

@defmac STARPU_BLOCK_GET_ELEMSIZE ({void *}@var{interface})
Return the size of the elements of the matrix designated by @var{interface}.
@end defmac

@node Accessing BCSR Data Interfaces
@subsubsection BCSR Data Interfaces

@deftypefun uint32_t starpu_bcsr_get_nnz (starpu_data_handle_t @var{handle})
Return the number of non-zero elements in the matrix designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_bcsr_get_nrow (starpu_data_handle_t @var{handle})
Return the number of rows (in terms of blocks of size r*c) in the matrix
designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_bcsr_get_firstentry (starpu_data_handle_t @var{handle})
Return the index at which all arrays (the column indexes, the row pointers...)
of the matrix desginated by @var{handle} start.
@end deftypefun

@deftypefun uintptr_t starpu_bcsr_get_local_nzval (starpu_data_handle_t @var{handle})
Return a pointer to the non-zero values of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun {uint32_t *} starpu_bcsr_get_local_colind (starpu_data_handle_t @var{handle})
Return a pointer to the column index, which holds the positions of the non-zero
entries in the matrix designated by @var{handle}.
@end deftypefun

@deftypefun {uint32_t *} starpu_bcsr_get_local_rowptr (starpu_data_handle_t @var{handle})
Return the row pointer array of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_bcsr_get_r (starpu_data_handle_t @var{handle})
Return the number of rows in a block.
@end deftypefun

@deftypefun uint32_t starpu_bcsr_get_c (starpu_data_handle_t @var{handle})
Return the numberof columns in a block.
@end deftypefun

@deftypefun size_t starpu_bcsr_get_elemsize (starpu_data_handle_t @var{handle})
Return the size of the elements in the matrix designated by @var{handle}.
@end deftypefun

@defmac STARPU_BCSR_GET_NNZ ({void *}@var{interface})
Return the number of non-zero values in the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_BCSR_GET_NZVAL ({void *}@var{interface})
Return a pointer to the non-zero values of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_BCSR_GET_NZVAL_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the array of non-zero values in the matrix designated
by @var{interface}. The offset documented below has to be used in addition to
this.
@end defmac

@defmac STARPU_BCSR_GET_COLIND ({void *}@var{interface})
Return a pointer to the column index of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_BCSR_GET_COLIND_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the column index of the matrix designated by
@var{interface}. The offset documented below has to be used in addition to
this.
@end defmac

@defmac STARPU_BCSR_GET_ROWPTR ({void *}@var{interface})
Return a pointer to the row pointer array of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_CSR_GET_ROWPTR_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the row pointer array of the matrix designated by
@var{interface}. The offset documented below has to be used in addition to
this.
@end defmac

@defmac STARPU_BCSR_GET_OFFSET ({void *}@var{interface})
Return the offset in the arrays (coling, rowptr, nzval) of the matrix
designated by @var{interface}, to be used with the device handles.
@end defmac

@node Accessing CSR Data Interfaces
@subsubsection CSR Data Interfaces

@deftypefun uint32_t starpu_csr_get_nnz (starpu_data_handle_t @var{handle})
Return the number of non-zero values in the matrix designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_csr_get_nrow (starpu_data_handle_t @var{handle})
Return the size of the row pointer array of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun uint32_t starpu_csr_get_firstentry (starpu_data_handle_t @var{handle})
Return the index at which all arrays (the column indexes, the row pointers...)
of the matrix designated by @var{handle} start.
@end deftypefun

@deftypefun uintptr_t starpu_csr_get_local_nzval (starpu_data_handle_t @var{handle})
Return a local pointer to the non-zero values of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun {uint32_t *} starpu_csr_get_local_colind (starpu_data_handle_t @var{handle})
Return a local pointer to the column index of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun {uint32_t *} starpu_csr_get_local_rowptr (starpu_data_handle_t @var{handle})
Return a local pointer to the row pointer array of the matrix designated by @var{handle}.
@end deftypefun

@deftypefun size_t starpu_csr_get_elemsize (starpu_data_handle_t @var{handle})
Return the size of the elements registered into the matrix designated by @var{handle}.
@end deftypefun

@defmac STARPU_CSR_GET_NNZ ({void *}@var{interface})
Return the number of non-zero values in the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_CSR_GET_NROW ({void *}@var{interface})
Return the size of the row pointer array of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_CSR_GET_NZVAL ({void *}@var{interface})
Return a pointer to the non-zero values of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_CSR_GET_NZVAL_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the array of non-zero values in the matrix designated
by @var{interface}. The offset documented below has to be used in addition to
this.
@end defmac

@defmac STARPU_CSR_GET_COLIND ({void *}@var{interface})
Return a pointer to the column index of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_CSR_GET_COLIND_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the column index of the matrix designated by
@var{interface}. The offset documented below has to be used in addition to
this.
@end defmac

@defmac STARPU_CSR_GET_ROWPTR ({void *}@var{interface})
Return a pointer to the row pointer array of the matrix designated by @var{interface}.
@end defmac

@defmac STARPU_CSR_GET_ROWPTR_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the row pointer array of the matrix designated by
@var{interface}. The offset documented below has to be used in addition to
this.
@end defmac

@defmac STARPU_CSR_GET_OFFSET ({void *}@var{interface})
Return the offset in the arrays (colind, rowptr, nzval) of the matrix
designated by @var{interface}, to be used with the device handles.
@end defmac

@defmac STARPU_CSR_GET_FIRSTENTRY ({void *}@var{interface})
Return the index at which all arrays (the column indexes, the row pointers...)
of the @var{interface} start.
@end defmac

@defmac STARPU_CSR_GET_ELEMSIZE ({void *}@var{interface})
Return the size of the elements registered into the matrix designated by @var{interface}.
@end defmac


@node Accessing COO Data Interfaces
@subsubsection COO Data Interfaces
@defmac STARPU_COO_GET_COLUMNS ({void *}@var{interface})
Return a pointer to the column array of the matrix designated by
@var{interface}.
@end defmac
@defmac STARPU_COO_GET_COLUMNS_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the column array of the matrix designated by
@var{interface}, to be used on OpenCL. The offset documented below has to be
used in addition to this.
@end defmac
@defmac STARPU_COO_GET_ROWS (interface)
Return a pointer to the rows array of the matrix designated by @var{interface}.
@end defmac
@defmac STARPU_COO_GET_ROWS_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the row array of the matrix designated by
@var{interface}, to be used on OpenCL. The offset documented below has to be
used in addition to this.
@end defmac
@defmac STARPU_COO_GET_VALUES (interface)
Return a pointer to the values array of the matrix designated by
@var{interface}.
@end defmac
@defmac STARPU_COO_GET_VALUES_DEV_HANDLE ({void *}@var{interface})
Return a device handle for the value array of the matrix designated by
@var{interface}, to be used on OpenCL. The offset documented below has to be
used in addition to this.
@end defmac
@defmac STARPU_COO_GET_OFFSET ({void *}@var{itnerface})
Return the offset in the arrays of the COO matrix designated by @var{interface}.
@end defmac
@defmac STARPU_COO_GET_NX (interface)
Return the number of elements on the x-axis of the matrix designated by
@var{interface}.
@end defmac
@defmac STARPU_COO_GET_NY (interface)
Return the number of elements on the y-axis of the matrix designated by
@var{interface}.
@end defmac
@defmac STARPU_COO_GET_NVALUES (interface)
Return the number of values registered in the matrix designated by
@var{interface}.
@end defmac
@defmac STARPU_COO_GET_ELEMSIZE (interface)
Return the size of the elements registered into the matrix designated by
@var{interface}.
@end defmac

@node Defining Interface
@subsection Defining Interface

Applications can provide their own interface as shown in
@pxref{Defining a New Data Interface}.

@deftypefun uintptr_t starpu_malloc_on_node (unsigned @var{dst_node}, size_t @var{size})
Allocate @var{size} bytes on node @var{dst_node}. This returns 0 if allocation
failed, the allocation method should then return -ENOMEM as allocated size.
@end deftypefun

@deftypefun void starpu_free_on_node (unsigned @var{dst_node}, uintptr_t @var{addr}, size_t @var{size})
Free @var{addr} of @var{size} bytes on node @var{dst_node}.
@end deftypefun

@deftp {Data Type} {struct starpu_data_interface_ops}
@anchor{struct starpu_data_interface_ops}
Per-interface data transfer methods.

@table @asis
@item @code{void (*register_data_handle)(starpu_data_handle_t handle, unsigned home_node, void *data_interface)}
Register an existing interface into a data handle.

@item @code{starpu_ssize_t (*allocate_data_on_node)(void *data_interface, unsigned node)}
Allocate data for the interface on a given node.

@item @code{ void (*free_data_on_node)(void *data_interface, unsigned node)}
Free data of the interface on a given node.

@item @code{ const struct starpu_data_copy_methods *copy_methods}
ram/cuda/opencl synchronous and asynchronous transfer methods.

@item @code{ void * (*handle_to_pointer)(starpu_data_handle_t handle, unsigned node)}
Return the current pointer (if any) for the handle on the given node.

@item @code{ size_t (*get_size)(starpu_data_handle_t handle)}
Return an estimation of the size of data, for performance models.

@item @code{ uint32_t (*footprint)(starpu_data_handle_t handle)}
Return a 32bit footprint which characterizes the data size.

@item @code{ int (*compare)(void *data_interface_a, void *data_interface_b)}
Compare the data size of two interfaces.

@item @code{ void (*display)(starpu_data_handle_t handle, FILE *f)}
Dump the sizes of a handle to a file.

@item @code{enum starpu_data_interface_id interfaceid}
An identifier that is unique to each interface.

@item @code{size_t interface_size}
The size of the interface data descriptor.

@item @code{int is_multiformat}
todo

@item @code{struct starpu_multiformat_data_interface_ops* (*get_mf_ops)(void *data_interface)}
todo

@item @code{int (*pack_data)(starpu_data_handle_t handle, unsigned node, void **ptr, ssize_t *count)}
Pack the data handle into a contiguous buffer at the address
@code{ptr} and set the size of the newly created buffer in
@code{count}. If @var{ptr} is @code{NULL}, the function should not copy the data in the
buffer but just set @var{count} to the size of the buffer which
would have been allocated. The special value @code{-1} indicates the
size is yet unknown.

@item @code{int (*unpack_data)(starpu_data_handle_t handle, unsigned node, void *ptr, size_t count)}
Unpack the data handle from the contiguous buffer at the address @code{ptr} of size @var{count}

@end table
@end deftp

@deftp {Data Type} {struct starpu_data_copy_methods}
Defines the per-interface methods. If the @code{any_to_any} method is provided,
it will be used by default if no more specific method is provided. It can still
be useful to provide more specific method in case of e.g. available particular
CUDA or OpenCL support.

@table @asis
@item @code{int (*@{ram,cuda,opencl,mic@}_to_@{ram,cuda,opencl,mic@})(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node)}
These 14 functions define how to copy data from the @var{src_interface}
interface on the @var{src_node} node to the @var{dst_interface} interface
on the @var{dst_node} node. They return 0 on success.

@item @code{int (*@{ram,cuda@}_to_@{ram,cuda@}_async)(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, cudaStream_t stream)}
These 3 functions (@code{ram_to_ram} is not among these) define how to copy
data from the @var{src_interface} interface on the @var{src_node} node to the
@var{dst_interface} interface on the @var{dst_node} node, using the given
@var{stream}. Must return 0 if the transfer was actually completed completely
synchronously, or -EAGAIN if at least some transfers are still ongoing and
should be awaited for by the core.

@item @code{int (*@{ram,opencl@}_to_@{ram,opencl@}_async)(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, /* cl_event * */ void *event)}
These 3 functions (@code{ram_to_ram} is not among them) define how to copy
data from the @var{src_interface} interface on the @var{src_node} node to the
@var{dst_interface} interface on the @var{dst_node} node, by recording in
@var{event}, a pointer to a cl_event, the event of the last submitted transfer.
Must return 0 if the transfer was actually completed completely synchronously,
or -EAGAIN if at least some transfers are still ongoing and should be awaited
for by the core.

@item @code{int (*@{ram,mic@}_to_@{ram,mic@}_async)(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node)}
These 2 functions (@code{ram_to_ram} and @code{mic_to_mic} are not among them) define how to copy
data from the @var{src_interface} interface on the @var{src_node} node to the
@var{dst_interface} interface on the @var{dst_node} node.
Must return 0 if the transfer was actually completed completely synchronously,
or -EAGAIN if at least some transfers are still ongoing and should be awaited
for by the core.

@item @code{int (*@{src,sink@}_to_@{src,sink@}_async)(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node)}
These 3 functions (@code{src_to_src} is not among them) define how to copy
data from the @var{src_interface} interface on the @var{src_node} node to the
@var{dst_interface} interface on the @var{dst_node} node.
Must return 0 if the transfer was actually completed completely synchronously,
or -EAGAIN if at least some transfers are still ongoing and should be awaited
for by the core.

@item @code{int (*any_to_any)(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data)}
Define how to copy data from the @var{src_interface} interface on the
@var{src_node} node to the @var{dst_interface} interface on the @var{dst_node}
node. This is meant to be implemented through the @var{starpu_interface_copy}
helper, to which @var{async_data} should be passed as such, and will be used to
manage asynchronicity. This must return -EAGAIN if any of the
@var{starpu_interface_copy} calls has returned -EAGAIN (i.e. at least some
transfer is still ongoing), and return 0 otherwise.

@end table
@end deftp

@deftypefun int starpu_interface_copy (uintptr_t @var{src}, size_t @var{src_offset}, unsigned @var{src_node}, uintptr_t @var{dst}, size_t @var{dst_offset}, unsigned @var{dst_node}, size_t @var{size}, {void *}@var{async_data})
Copy @var{size} bytes from byte offset @var{src_offset} of @var{src} on
@var{src_node} to byte offset @var{dst_offset} of @var{dst} on @var{dst_node}.
This is to be used in the @var{any_to_any} copy method, which is provided with
the @var{async_data} to be pased to @var{starpu_interface_copy}. this returns
-EAGAIN if the transfer is still ongoing, or 0 if the transfer is already
completed.
@end deftypefun


@deftypefun uint32_t starpu_hash_crc32c_be_n ({void *}@var{input}, size_t @var{n}, uint32_t @var{inputcrc})
Compute the CRC of a byte buffer seeded by the inputcrc "current
state". The return value should be considered as the new "current
state" for future CRC computation. This is used for computing data size
footprint.
@end deftypefun

@deftypefun uint32_t starpu_hash_crc32c_be (uint32_t @var{input}, uint32_t @var{inputcrc})
Compute the CRC of a 32bit number seeded by the inputcrc "current
state". The return value should be considered as the new "current
state" for future CRC computation. This is used for computing data size
footprint.
@end deftypefun

@deftypefun uint32_t starpu_hash_crc32c_string ({char *}@var{str}, uint32_t @var{inputcrc})
Compute the CRC of a string seeded by the inputcrc "current state".
The return value should be considered as the new "current state" for
future CRC computation. This is used for computing data size footprint.
@end deftypefun

@deftypefun int starpu_data_interface_get_next_id (void)
Returns the next available id for a newly created data interface
(@pxref{Defining a New Data Interface}).
@end deftypefun

@node Data Partition
@section Data Partition

@menu
* Basic API::
* Predefined filter functions::
@end menu

@node Basic API
@subsection Basic API

@deftp {Data Type} {struct starpu_data_filter}
The filter structure describes a data partitioning operation, to be given to the
@code{starpu_data_partition} function, see @ref{starpu_data_partition}
for an example. The different fields are:

@table @asis
@item @code{void (*filter_func)(void *father_interface, void* child_interface, struct starpu_data_filter *, unsigned id, unsigned nparts)}
This function fills the @code{child_interface} structure with interface
information for the @code{id}-th child of the parent @code{father_interface} (among @code{nparts}).

@item @code{unsigned nchildren}
This is the number of parts to partition the data into.

@item @code{unsigned (*get_nchildren)(struct starpu_data_filter *, starpu_data_handle_t initial_handle)}
This returns the number of children. This can be used instead of @code{nchildren} when the number of
children depends on the actual data (e.g. the number of blocks in a sparse
matrix).

@item @code{struct starpu_data_interface_ops *(*get_child_ops)(struct starpu_data_filter *, unsigned id)}
In case the resulting children use a different data interface, this function
returns which interface is used by child number @code{id}.

@item @code{unsigned filter_arg}
Allow to define an additional parameter for the filter function.

@item @code{void *filter_arg_ptr}
Allow to define an additional pointer parameter for the filter
function, such as the sizes of the different parts.
@end table
@end deftp

@deftypefun void starpu_data_partition (starpu_data_handle_t @var{initial_handle}, {struct starpu_data_filter *}@var{f})
@anchor{starpu_data_partition}
This requests partitioning one StarPU data @var{initial_handle} into several
subdata according to the filter @var{f}, as shown in the following example:

@cartouche
@smallexample
struct starpu_data_filter f = @{
    .filter_func = starpu_matrix_filter_block,
    .nchildren = nslicesx,
    .get_nchildren = NULL,
    .get_child_ops = NULL
@};
starpu_data_partition(A_handle, &f);
@end smallexample
@end cartouche
@end deftypefun

@deftypefun void starpu_data_unpartition (starpu_data_handle_t @var{root_data}, unsigned @var{gathering_node})
This unapplies one filter, thus unpartitioning the data. The pieces of data are
collected back into one big piece in the @var{gathering_node} (usually 0). Tasks
working on the partitioned data must be already finished when calling @code{starpu_data_unpartition}.
@cartouche
@smallexample
starpu_data_unpartition(A_handle, 0);
@end smallexample
@end cartouche
@end deftypefun

@deftypefun int starpu_data_get_nb_children (starpu_data_handle_t @var{handle})
This function returns the number of children.
@end deftypefun

@deftypefun starpu_data_handle_t starpu_data_get_child (starpu_data_handle_t @var{handle}, unsigned @var{i})
Return the @var{i}th child of the given @var{handle}, which must have been partitionned beforehand.
@end deftypefun

@deftypefun starpu_data_handle_t starpu_data_get_sub_data (starpu_data_handle_t @var{root_data}, unsigned @var{depth}, ... )
After partitioning a StarPU data by applying a filter,
@code{starpu_data_get_sub_data} can be used to get handles for each of
the data portions. @var{root_data} is the parent data that was
partitioned. @var{depth} is the number of filters to traverse (in
case several filters have been applied, to e.g. partition in row
blocks, and then in column blocks), and the subsequent
parameters are the indexes. The function returns a handle to the
subdata.
@cartouche
@smallexample
h = starpu_data_get_sub_data(A_handle, 1, taskx);
@end smallexample
@end cartouche
@end deftypefun

@deftypefun starpu_data_handle_t starpu_data_vget_sub_data (starpu_data_handle_t @var{root_data}, unsigned @var{depth}, va_list @var{pa})
This function is similar to @code{starpu_data_get_sub_data} but uses a
va_list for the parameter list.
@end deftypefun

@deftypefun void starpu_data_map_filters (starpu_data_handle_t @var{root_data}, unsigned @var{nfilters}, ...)
Applies @var{nfilters} filters to the handle designated by @var{root_handle}
recursively. @var{nfilters} pointers to variables of the type
starpu_data_filter should be given.
@end deftypefun

@deftypefun void starpu_data_vmap_filters (starpu_data_handle_t @var{root_data}, unsigned @var{nfilters}, va_list @var{pa})
Applies @var{nfilters} filters to the handle designated by @var{root_handle}
recursively. It uses a va_list of pointers to variables of the typer
starpu_data_filter.
@end deftypefun

@node Predefined filter functions
@subsection Predefined filter functions

@menu
* Partitioning Vector Data::
* Partitioning Matrix Data::
* Partitioning 3D Matrix Data::
* Partitioning BCSR Data::
@end menu

This section gives a partial list of the predefined partitioning functions.
Examples on how to use them are shown in @ref{Partitioning Data}. The complete
list can be found in @code{starpu_data_filters.h} .

@node Partitioning Vector Data
@subsubsection Partitioning Vector Data

@deftypefun void starpu_vector_filter_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
Return in @code{*@var{child_interface}} the @var{id}th element of the
vector represented by @var{father_interface} once partitioned in
@var{nparts} chunks of equal size.
@end deftypefun

@deftypefun void starpu_vector_filter_block_shadow (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
Return in @code{*@var{child_interface}} the @var{id}th element of the
vector represented by @var{father_interface} once partitioned in
@var{nparts} chunks of equal size with a shadow border @code{filter_arg_ptr}, thus getting a vector of size (n-2*shadow)/nparts+2*shadow

The @code{filter_arg_ptr} field must be the shadow size casted into @code{void*}.

IMPORTANT: This can only be used for read-only access, as no coherency is
enforced for the shadowed parts.

A usage example is available in examples/filters/shadow.c
@end deftypefun

@deftypefun void starpu_vector_filter_list (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
Return in @code{*@var{child_interface}} the @var{id}th element of the
vector represented by @var{father_interface} once partitioned into
@var{nparts} chunks according to the @code{filter_arg_ptr} field of
@code{*@var{f}}.

The @code{filter_arg_ptr} field must point to an array of @var{nparts}
@code{uint32_t} elements, each of which specifies the number of elements
in each chunk of the partition.
@end deftypefun

@deftypefun void starpu_vector_filter_divide_in_2 (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
Return in @code{*@var{child_interface}} the @var{id}th element of the
vector represented by @var{father_interface} once partitioned in two
chunks of equal size, ignoring @var{nparts}.  Thus, @var{id} must be
@code{0} or @code{1}.
@end deftypefun


@node Partitioning Matrix Data
@subsubsection Partitioning Matrix Data

@deftypefun void starpu_matrix_filter_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a dense Matrix along the x dimension, thus getting (x/nparts,y)
matrices. If nparts does not divide x, the last submatrix contains the
remainder.
@end deftypefun

@deftypefun void starpu_matrix_filter_block_shadow (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a dense Matrix along the x dimension, with a shadow border
@code{filter_arg_ptr}, thus getting ((x-2*shadow)/nparts+2*shadow,y)
matrices. If nparts does not divide x-2*shadow, the last submatrix contains the
remainder.

IMPORTANT: This can only be used for read-only access, as no coherency is
enforced for the shadowed parts.

A usage example is available in examples/filters/shadow2d.c
@end deftypefun

@deftypefun void starpu_matrix_filter_vertical_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a dense Matrix along the y dimension, thus getting (x,y/nparts)
matrices. If nparts does not divide y, the last submatrix contains the
remainder.
@end deftypefun

@deftypefun void starpu_matrix_filter_vertical_block_shadow (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a dense Matrix along the y dimension, with a shadow border
@code{filter_arg_ptr}, thus getting (x,(y-2*shadow)/nparts+2*shadow)
matrices. If nparts does not divide y-2*shadow, the last submatrix contains the
remainder.

IMPORTANT: This can only be used for read-only access, as no coherency is
enforced for the shadowed parts.

A usage example is available in examples/filters/shadow2d.c
@end deftypefun

@node Partitioning 3D Matrix Data
@subsubsection Partitioning 3D Matrix Data

A usage example is available in examples/filters/shadow3d.c

@deftypefun void starpu_block_filter_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a 3D matrix along the X dimension, thus getting (x/nparts,y,z)
3D matrices. If nparts does not divide x, the last submatrix contains the
remainder.
@end deftypefun

@deftypefun void starpu_block_filter_block_shadow (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a 3D matrix along the X dimension, with a shadow border
@code{filter_arg_ptr}, thus getting ((x-2*shadow)/nparts+2*shadow,y,z) 3D
matrices. If nparts does not divide x, the last submatrix contains the
remainder.

IMPORTANT: This can only be used for read-only access, as no coherency is
enforced for the shadowed parts.
@end deftypefun

@deftypefun void starpu_block_filter_vertical_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a 3D matrix along the Y dimension, thus getting (x,y/nparts,z)
3D matrices. If nparts does not divide y, the last submatrix contains the
remainder.
@end deftypefun

@deftypefun void starpu_block_filter_vertical_block_shadow (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a 3D matrix along the Y dimension, with a shadow border
@code{filter_arg_ptr}, thus getting (x,(y-2*shadow)/nparts+2*shadow,z) 3D
matrices. If nparts does not divide y, the last submatrix contains the
remainder.

IMPORTANT: This can only be used for read-only access, as no coherency is
enforced for the shadowed parts.
@end deftypefun

@deftypefun void starpu_block_filter_depth_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a 3D matrix along the Z dimension, thus getting (x,y,z/nparts)
3D matrices. If nparts does not divide z, the last submatrix contains the
remainder.
@end deftypefun

@deftypefun void starpu_block_filter_depth_block_shadow (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a 3D matrix along the Z dimension, with a shadow border
@code{filter_arg_ptr}, thus getting (x,y,(z-2*shadow)/nparts+2*shadow)
3D matrices. If nparts does not divide z, the last submatrix contains the
remainder.

IMPORTANT: This can only be used for read-only access, as no coherency is
enforced for the shadowed parts.
@end deftypefun

@node Partitioning BCSR Data
@subsubsection Partitioning BCSR Data

@deftypefun void starpu_bcsr_filter_canonical_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a block-sparse matrix into dense matrices.
@end deftypefun

@deftypefun void starpu_csr_filter_vertical_block (void *@var{father_interface}, void *@var{child_interface}, {struct starpu_data_filter} *@var{f}, unsigned @var{id}, unsigned @var{nparts})
This partitions a block-sparse matrix into vertical block-sparse matrices.
@end deftypefun

@node Multiformat Data Interface
@section Multiformat Data Interface

@deftp {Data Type} {struct starpu_multiformat_data_interface_ops}
The different fields are:
@table @asis
@item @code{size_t cpu_elemsize}
the size of each element on CPUs,

@item @code{size_t opencl_elemsize}
the size of each element on OpenCL devices,

@item @code{struct starpu_codelet *cpu_to_opencl_cl}
pointer to a codelet which converts from CPU to OpenCL

@item @code{struct starpu_codelet *opencl_to_cpu_cl}
pointer to a codelet which converts from OpenCL to CPU

@item @code{size_t cuda_elemsize}
the size of each element on CUDA devices,

@item @code{struct starpu_codelet *cpu_to_cuda_cl}
pointer to a codelet which converts from CPU to CUDA

@item @code{struct starpu_codelet *cuda_to_cpu_cl}
pointer to a codelet which converts from CUDA to CPU

@item @code{size_t mic_elemsize}
the size of each element on MIC devices,

@item @code{struct starpu_codelet *cpu_to_mic_cl}
pointer to a codelet which converts from CPU to MIC

@item @code{struct starpu_codelet *mic_to_cpu_cl}
pointer to a codelet which converts from MIC to CPU

@item @code{size_t scc_elemsize}
the size of each element on SCC devices,

@item @code{struct starpu_codelet *cpu_to_scc_cl}
pointer to a codelet which converts from CPU to SCC

@item @code{struct starpu_codelet *scc_to_cpu_cl}
pointer to a codelet which converts from SCC to CPU
@end table
@end deftp

@deftypefun void starpu_multiformat_data_register (starpu_data_handle_t *@var{handle}, unsigned @var{home_node}, void *@var{ptr}, uint32_t @var{nobjects}, struct starpu_multiformat_data_interface_ops *@var{format_ops})
Register a piece of data that can be represented in different ways, depending upon
the processing unit that manipulates it. It allows the programmer, for instance, to
use an array of structures when working on a CPU, and a structure of arrays when
working on a GPU.

@var{nobjects} is the number of elements in the data. @var{format_ops} describes
the format.
@end deftypefun

@defmac STARPU_MULTIFORMAT_GET_CPU_PTR ({void *}@var{interface})
returns the local pointer to the data with CPU format.
@end defmac

@defmac STARPU_MULTIFORMAT_GET_CUDA_PTR ({void *}@var{interface})
returns the local pointer to the data with CUDA format.
@end defmac

@defmac STARPU_MULTIFORMAT_GET_OPENCL_PTR ({void *}@var{interface})
returns the local pointer to the data with OpenCL format.
@end defmac

@defmac STARPU_MULTIFORMAT_GET_NX  ({void *}@var{interface})
returns the number of elements in the data.
@end defmac

@node Codelets and Tasks
@section Codelets and Tasks

This section describes the interface to manipulate codelets and tasks.

@deftp {Data Type} {enum starpu_codelet_type}
Describes the type of parallel task. The different values are:
@table @asis
@item @code{STARPU_SEQ} (default) for classical sequential tasks.
@item @code{STARPU_SPMD} for a parallel task whose threads are handled by
StarPU, the code has to use @code{starpu_combined_worker_get_size} and
@code{starpu_combined_worker_get_rank} to distribute the work
@item @code{STARPU_FORKJOIN} for a parallel task whose threads are started by
the codelet function, which has to use @code{starpu_combined_worker_get_size} to
determine how many threads should be started.
@end table
See @ref{Parallel Tasks} for details.
@end deftp

@defmac STARPU_CPU
This macro is used when setting the field @code{where} of a @code{struct
starpu_codelet} to specify the codelet may be executed on a CPU
processing unit.
@end defmac

@defmac STARPU_CUDA
This macro is used when setting the field @code{where} of a @code{struct
starpu_codelet} to specify the codelet may be executed on a CUDA
processing unit.
@end defmac

@defmac STARPU_OPENCL
This macro is used when setting the field @code{where} of a @code{struct
starpu_codelet} to specify the codelet may be executed on an OpenCL
processing unit.
@end defmac

@defmac STARPU_MIC
This macro is used when setting the field @code{where} of a @code{struct
starpu_codelet} to specify the codelet may be executed on a MIC
processing unit.
@end defmac

@defmac STARPU_SCC
This macro is used when setting the field @code{where} of a @code{struct
starpu_codelet} to specify the codelet may be executed on an SCC
processing unit.
@end defmac

@defmac STARPU_MULTIPLE_CPU_IMPLEMENTATIONS
Setting the field @code{cpu_func} of a @code{struct starpu_codelet}
with this macro indicates the codelet will have several
implementations. The use of this macro is deprecated. One should
always only define the field @code{cpu_funcs}.
@end defmac

@defmac STARPU_MULTIPLE_CUDA_IMPLEMENTATIONS
Setting the field @code{cuda_func} of a @code{struct starpu_codelet}
with this macro indicates the codelet will have several
implementations. The use of this macro is deprecated. One should
always only define the field @code{cuda_funcs}.
@end defmac

@defmac STARPU_MULTIPLE_OPENCL_IMPLEMENTATIONS
Setting the field @code{opencl_func} of a @code{struct starpu_codelet}
with this macro indicates the codelet will have several
implementations. The use of this macro is deprecated. One should
always only define the field @code{opencl_funcs}.
@end defmac

@deftp {Data Type} {struct starpu_codelet}
The codelet structure describes a kernel that is possibly implemented on various
targets. 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.

@table @asis
@item @code{uint32_t where} (optional)
Indicates which types of processing units are able to execute the
codelet. The different values
@code{STARPU_CPU}, @code{STARPU_CUDA},
@code{STARPU_OPENCL} can be combined to specify
on which types of processing units the codelet can be executed.
@code{STARPU_CPU|STARPU_CUDA} for instance indicates that the codelet is
implemented for both CPU cores and CUDA devices while @code{STARPU_OPENCL}
indicates that it is only available on OpenCL devices. If the field is
unset, its value will be automatically set based on the availability
of the @code{XXX_funcs} fields defined below.

@item @code{int (*can_execute)(unsigned workerid, struct starpu_task *task, unsigned nimpl)} (optional)
Defines a function which should return 1 if the worker designated by
@var{workerid} can execute the @var{nimpl}th implementation of the
given @var{task}, 0 otherwise.

@item @code{enum starpu_codelet_type type} (optional)
The default is @code{STARPU_SEQ}, i.e. usual sequential implementation. Other
values (@code{STARPU_SPMD} or @code{STARPU_FORKJOIN} declare that a parallel
implementation is also available. See @ref{Parallel Tasks} for details.

@item @code{int max_parallelism} (optional)
If a parallel implementation is available, this denotes the maximum combined
worker size that StarPU will use to execute parallel tasks for this codelet.

@item @code{starpu_cpu_func_t cpu_func} (optional)
This field has been made deprecated. One should use instead the
@code{cpu_funcs} field.

@item @code{starpu_cpu_func_t cpu_funcs[STARPU_MAXIMPLEMENTATIONS]} (optional)
Is an array of function pointers to the CPU implementations of the codelet.
It must be terminated by a NULL value.
The functions prototype must be: @code{void cpu_func(void *buffers[], void *cl_arg)}. The first
argument being the array of data managed by the data management library, and
the second argument is a pointer to the argument passed from the @code{cl_arg}
field of the @code{starpu_task} structure.
If the @code{where} field is set, then the @code{cpu_funcs} field is
ignored if @code{STARPU_CPU} does not appear in the @code{where}
field, it must be non-null otherwise.

@item @code{char * cpu_funcs_name[STARPU_MAXIMPLEMENTATIONS]} (optional)
Is an array of strings which provide the name of the CPU functions referenced in
the @code{cpu_funcs} array. This can be used when running on MIC devices or the
SCC platform, for StarPU to simply look up the MIC function implementation
through its name.

@item @code{starpu_cuda_func_t cuda_func} (optional)
This field has been made deprecated. One should use instead the
@code{cuda_funcs} field.

@item @code{starpu_cuda_func_t cuda_funcs[STARPU_MAXIMPLEMENTATIONS]} (optional)
Is an array of function pointers to the CUDA implementations of the codelet.
It must be terminated by a NULL value.
@emph{The functions must be host-functions written in the CUDA runtime
API}. Their prototype must
be: @code{void cuda_func(void *buffers[], void *cl_arg);}.
If the @code{where} field is set, then the @code{cuda_funcs}
field is ignored if @code{STARPU_CUDA} does not appear in the @code{where}
field, it must be non-null otherwise.

@item @code{starpu_opencl_func_t opencl_func} (optional)
This field has been made deprecated. One should use instead the
@code{opencl_funcs} field.

@item @code{starpu_opencl_func_t opencl_funcs[STARPU_MAXIMPLEMENTATIONS]} (optional)
Is an array of function pointers to the OpenCL implementations of the codelet.
It must be terminated by a NULL value.
The functions prototype must be:
@code{void opencl_func(void *buffers[], void *cl_arg);}.
If the @code{where} field is set, then the @code{opencl_funcs} field
is ignored if @code{STARPU_OPENCL} does not appear in the @code{where}
field, it must be non-null otherwise.

@item @code{starpu_mic_func_t mic_funcs[STARPU_MAXIMPLEMENTATIONS]} (optional)
Is an array of function pointers to a function which returns the MIC
implementation of the codelet.
It must be terminated by a NULL value.
The functions prototype must be:
@code{starpu_mic_kernel_t mic_func(struct starpu_codelet *cl, unsigned nimpl);}.
If the @code{where} field is set, then the @code{mic_funcs} field
is ignored if @code{STARPU_MIC} does not appear in the @code{where}
field. It can be null if @code{cpu_funcs_name} is non-NULL, in which case StarPU
will simply make a symbol lookup to get the implementation.

@item @code{starpu_scc_func_t scc_funcs[STARPU_MAXIMPLEMENTATIONS]} (optional)
Is an array of function pointers to a function which returns the SCC
implementation of the codelet.
It must be terminated by a NULL value.
The functions prototype must be:
@code{starpu_scc_kernel_t scc_func(struct starpu_codelet *cl, unsigned nimpl);}.
If the @code{where} field is set, then the @code{scc_funcs} field
is ignored if @code{STARPU_SCC} does not appear in the @code{where}
field. It can be null if @code{cpu_funcs_name} is non-NULL, in which case StarPU
will simply make a symbol lookup to get the implementation.

@item @code{unsigned nbuffers}
Specifies the number of arguments taken by the codelet. These arguments are
managed by the DSM and are accessed from the @code{void *buffers[]}
array. The constant argument passed with the @code{cl_arg} field of the
@code{starpu_task} structure is not counted in this number.  This value should
not be above @code{STARPU_NMAXBUFS}.

@item @code{enum starpu_data_access_mode modes[STARPU_NMAXBUFS]}
Is an array of @code{enum starpu_data_access_mode}. It describes the
required access modes to the data neeeded by the codelet (e.g.
@code{STARPU_RW}). The number of entries in this array must be
specified in the @code{nbuffers} field (defined above), and should not
exceed @code{STARPU_NMAXBUFS}.
If unsufficient, this value can be set with the @code{--enable-maxbuffers}
option when configuring StarPU.

@item @code{enum starpu_data_access_mode *dyn_modes}
Is an array of @code{enum starpu_data_access_mode}. It describes the
required access modes to the data neeeded by the codelet (e.g.
@code{STARPU_RW}). The number of entries in this array must be
specified in the @code{nbuffers} field (defined above).
This field should be used for codelets having a number of datas
greater than @code{STARPU_NMAXBUFS} (@pxref{Setting the Data Handles
for a Task}).
When defining a codelet, one should either define this field or the
field @code{modes} defined above. 

@item @code{struct starpu_perfmodel *model} (optional)
This is a pointer to the task duration performance model associated to this
codelet. This optional field is ignored when set to @code{NULL} or
when its @code{symbol} field is not set.

@item @code{struct starpu_perfmodel *power_model} (optional)
This is a pointer to the task power consumption performance model associated
to this codelet. This optional field is ignored when set to
@code{NULL} or when its @code{symbol} field is not set.
In the case of parallel codelets, this has to account for all processing units
involved in the parallel execution.

@item @code{unsigned long per_worker_stats[STARPU_NMAXWORKERS]} (optional)
Statistics collected at runtime: this is filled by StarPU and should not be
accessed directly, but for example by calling the
@code{starpu_codelet_display_stats} function (See
@ref{starpu_codelet_display_stats} for details).

@item @code{const char *name} (optional)
Define the name of the codelet. This can be useful for debugging purposes.

@end table
@end deftp

@deftypefun void starpu_codelet_init ({struct starpu_codelet} *@var{cl})
@anchor{starpu_codelet_init}
Initialize @var{cl} with default values. Codelets should preferably be
initialized statically as shown in @ref{Defining a Codelet}. However
such a initialisation is not always possible, e.g. when using C++.
@end deftypefun

@deftp {Data Type} {enum starpu_task_status}
State of a task, can be either of
@table @asis
@item @code{STARPU_TASK_INVALID} The task has just been initialized.
@item @code{STARPU_TASK_BLOCKED} The task has just been submitted, and its dependencies has not been checked yet.
@item @code{STARPU_TASK_READY} The task is ready for execution.
@item @code{STARPU_TASK_RUNNING} The task is running on some worker.
@item @code{STARPU_TASK_FINISHED} The task is finished executing.
@item @code{STARPU_TASK_BLOCKED_ON_TAG} The task is waiting for a tag.
@item @code{STARPU_TASK_BLOCKED_ON_TASK} The task is waiting for a task.
@item @code{STARPU_TASK_BLOCKED_ON_DATA} The task is waiting for some data.
@end table
@end deftp

@deftp {Data Type} {struct starpu_data_descr}
This type is used to describe a data handle along with an
access mode.
@table @asis
@item @code{starpu_data_handle_t handle} describes a data,
@item @code{enum starpu_data_access_mode mode} describes its access mode
@end table
@end deftp


@deftp {Data Type} {struct starpu_task}
The @code{starpu_task} structure describes a task that can be offloaded on the various
processing units managed by StarPU. It instantiates a codelet. It can either be
allocated dynamically with the @code{starpu_task_create} method, or declared
statically. In the latter case, the programmer has to zero the
@code{starpu_task} structure and to fill the different fields properly. The
indicated default values correspond to the configuration of a task allocated
with @code{starpu_task_create}.

@table @asis
@item @code{struct starpu_codelet *cl}
Is a pointer to the corresponding @code{struct starpu_codelet} data structure. This
describes where the kernel should be executed, and supplies the appropriate
implementations. When set to @code{NULL}, no code is executed during the tasks,
such empty tasks can be useful for synchronization purposes.

@item @code{struct starpu_data_descr buffers[STARPU_NMAXBUFS]}
This field has been made deprecated. One should use instead the
@code{handles} field to specify the handles to the data accessed by
the task. The access modes are now defined in the @code{mode} field of
the @code{struct starpu_codelet cl} field defined above.

@item @code{starpu_data_handle_t handles[STARPU_NMAXBUFS]}
Is an array of @code{starpu_data_handle_t}. It specifies the handles
to the different pieces of data accessed by the task. The number
of entries in this array must be specified in the @code{nbuffers} field of the
@code{struct starpu_codelet} structure, and should not exceed
@code{STARPU_NMAXBUFS}.
If unsufficient, this value can be set with the @code{--enable-maxbuffers}
option when configuring StarPU.

@item @code{starpu_data_handle_t *dyn_handles}
Is an array of @code{starpu_data_handle_t}. It specifies the handles
to the different pieces of data accessed by the task. The number
of entries in this array must be specified in the @code{nbuffers} field of the
@code{struct starpu_codelet} structure.
This field should be used for tasks having a number of datas
greater than @code{STARPU_NMAXBUFS} (@pxref{Setting the Data Handles
for a Task}).
When defining a task, one should either define this field or the
field @code{handles} defined above.

@item @code{void *interfaces[STARPU_NMAXBUFS]}
The actual data pointers to the memory node where execution will happen, managed
by the DSM.

@item @code{void **dyn_interfaces}
The actual data pointers to the memory node where execution will happen, managed
by the DSM. Is used when the field @code{dyn_handles} is defined.

@item @code{void *cl_arg} (optional; default: @code{NULL})
This pointer is passed to the codelet through the second argument
of the codelet implementation (e.g. @code{cpu_func} or @code{cuda_func}).

@item @code{size_t cl_arg_size} (optional)
For some specific drivers, the @code{cl_arg} pointer cannot not be directly
given to the driver function. A buffer of size @code{cl_arg_size}
needs to be allocated on the driver. This buffer is then filled with
the @code{cl_arg_size} bytes starting at address @code{cl_arg}. In
this case, the argument given to the codelet is therefore not the
@code{cl_arg} pointer, but the address of the buffer in local store
(LS) instead.
This field is ignored for CPU, CUDA and OpenCL codelets, where the
@code{cl_arg} pointer is given as such.

@item @code{unsigned cl_arg_free} (optional)
In case @code{cl_arg} was allocated by the application through @code{malloc},
setting @code{cl_arg_free} to 1 makes StarPU automatically call
@code{free(cl_arg)} when destroying the task. This saves the user from
defining a callback just for that. This is mostly useful when targetting MIC or
SCC, where the codelet does not execute in the same memory space as the main
thread.

@item @code{void (*callback_func)(void *)} (optional) (default: @code{NULL})
This is a function pointer of prototype @code{void (*f)(void *)} which
specifies a possible callback. If this pointer is non-null, the callback
function is executed @emph{on the host} after the execution of the task. Tasks
which depend on it might already be executing. The callback is passed the
value contained in the @code{callback_arg} field. No callback is executed if the
field is set to @code{NULL}.

@item @code{void *callback_arg} (optional) (default: @code{NULL})
This is the pointer passed to the callback function. This field is ignored if
the @code{callback_func} is set to @code{NULL}.

@item @code{unsigned use_tag} (optional) (default: @code{0})
If set, this flag indicates that the task should be associated with the tag
contained in the @code{tag_id} field. Tag allow the application to synchronize
with the task and to express task dependencies easily.

@item @code{starpu_tag_t tag_id}
This field contains the tag associated to the task if the @code{use_tag} field
was set, it is ignored otherwise.

@item @code{unsigned sequential_consistency}
If this flag is set (which is the default), sequential consistency is enforced
for the data parameters of this task for which sequential consistency is
enabled. Clearing this flag permits to disable sequential consistency for this
task, even if data have it enabled.

@item @code{unsigned synchronous}
If this flag is set, the @code{starpu_task_submit} function is blocking and
returns only when the task has been executed (or if no worker is able to
process the task). Otherwise, @code{starpu_task_submit} returns immediately.

@item @code{int priority} (optional) (default: @code{STARPU_DEFAULT_PRIO})
This field indicates a level of priority for the task. This is an integer value
that must be set between the return values of the
@code{starpu_sched_get_min_priority} function for the least important tasks,
and that of the @code{starpu_sched_get_max_priority} for the most important
tasks (included). The @code{STARPU_MIN_PRIO} and @code{STARPU_MAX_PRIO} macros
are provided for convenience and respectively returns value of
@code{starpu_sched_get_min_priority} and @code{starpu_sched_get_max_priority}.
Default priority is @code{STARPU_DEFAULT_PRIO}, which is always defined as 0 in
order to allow static task initialization.  Scheduling strategies that take
priorities into account can use this parameter to take better scheduling
decisions, but the scheduling policy may also ignore it.

@item @code{unsigned execute_on_a_specific_worker} (default: @code{0})
If this flag is set, StarPU will bypass the scheduler and directly affect this
task to the worker specified by the @code{workerid} field.

@item @code{unsigned workerid} (optional)
If the @code{execute_on_a_specific_worker} field is set, this field indicates
which is the identifier of the worker that should process this task (as
returned by @code{starpu_worker_get_id}). This field is ignored if
@code{execute_on_a_specific_worker} field is set to 0.

@item @code{starpu_task_bundle_t bundle} (optional)
The bundle that includes this task. If no bundle is used, this should be NULL.

@item @code{int detach} (optional) (default: @code{1})
If this flag is set, it is not possible to synchronize with the task
by the means of @code{starpu_task_wait} later on. Internal data structures
are only guaranteed to be freed once @code{starpu_task_wait} is called if the
flag is not set.

@item @code{int destroy} (optional) (default: @code{0} for starpu_task_init, @code{1} for starpu_task_create)
If this flag is set, the task structure will automatically be freed, either
after the execution of the callback if the task is detached, or during
@code{starpu_task_wait} otherwise. If this flag is not set, dynamically
allocated data structures will not be freed until @code{starpu_task_destroy} is
called explicitly. Setting this flag for a statically allocated task structure
will result in undefined behaviour. The flag is set to 1 when the task is
created by calling @code{starpu_task_create()}. Note that
@code{starpu_task_wait_for_all} will not free any task.

@item @code{int regenerate} (optional)
If this flag is set, the task will be re-submitted to StarPU once it has been
executed. This flag must not be set if the destroy flag is set too.

@item @code{enum starpu_task_status status} (optional)
Current state of the task.

@item @code{struct starpu_profiling_task_info *profiling_info} (optional)
Profiling information for the task.

@item @code{double predicted} (output field)
Predicted duration of the task. This field is only set if the scheduling
strategy used performance models.

@item @code{double predicted_transfer} (optional)
Predicted data transfer duration for the task in microseconds. This field is
only valid if the scheduling strategy uses performance models.

@item @code{struct starpu_task *prev}
A pointer to the previous task. This should only be used by StarPU.

@item @code{struct starpu_task *next}
A pointer to the next task. This should only be used by StarPU.

@item @code{unsigned int mf_skip}
This is only used for tasks that use multiformat handle. This should only be
used by StarPU.

@item @code{double flops}
This can be set to the number of floating points operations that the task
will have to achieve. This is useful for easily getting GFlops curves from
@code{starpu_perfmodel_plot}, and for the hypervisor load balancing.

@item @code{void *starpu_private}
This is private to StarPU, do not modify. If the task is allocated by hand
(without starpu_task_create), this field should be set to NULL.

@item @code{int magic}
This field is set when initializing a task. It prevents a task from being
submitted if it has not been properly initialized.
@end table

@end deftp

@deftypefun void starpu_task_init ({struct starpu_task} *@var{task})
Initialize @var{task} with default values. This function is implicitly
called by @code{starpu_task_create}. By default, tasks initialized with
@code{starpu_task_init} must be deinitialized explicitly with
@code{starpu_task_clean}. Tasks can also be initialized statically,
using @code{STARPU_TASK_INITIALIZER} defined below.
@end deftypefun

@defmac STARPU_TASK_INITIALIZER
It is possible to initialize statically allocated tasks with this
value. This is equivalent to initializing a starpu_task structure with
the @code{starpu_task_init} function defined above.
@end defmac

@defmac STARPU_TASK_GET_HANDLE ({struct starpu_task} *@var{task}, int @var{i})
Return the i-th data handle of the given task. If the task is defined
with a static or dynamic number of handles, will either return the
i-th element of the field @code{handles} or the i-th element of the field
@code{dyn_handles} (@pxref{Setting the Data Handles for a Task})
@end defmac

@defmac STARPU_TASK_SET_HANDLE ({struct starpu_task} *@var{task}, starpu_data_handle_t @var{handle}, int @var{i})
Set the i-th data handle of the given task with the given dat handle.
If the task is defined with a static or dynamic number of handles,
will either set the i-th element of the field @code{handles} or the
i-th element of the field @code{dyn_handles} (@pxref{Setting the Data
Handles for a Task})
@end defmac

@defmac STARPU_CODELET_GET_MODE ({struct starpu_codelet *}@var{codelet}, int @var{i})
Return the access mode of the i-th data handle of the given codelet.
If the codelet is defined with a static or dynamic number of handles,
will either return the i-th element of the field @code{modes} or the
i-th element of the field @code{dyn_modes} (@pxref{Setting the Data
Handles for a Task})
@end defmac

@defmac STARPU_CODELET_SET_MODE ({struct starpu_codelet *}@var{codelet}, {enum starpu_data_access_mode} @var{mode}, int @var{i})
Set the access mode of the i-th data handle of the given codelet.
If the codelet is defined with a static or dynamic number of handles,
will either set the i-th element of the field @code{modes} or the
i-th element of the field @code{dyn_modes} (@pxref{Setting the Data
Handles for a Task})
@end defmac

@deftypefun {struct starpu_task *} starpu_task_create (void)
Allocate a task structure and initialize it with default values. Tasks
allocated dynamically with @code{starpu_task_create} are automatically freed when the
task is terminated. This means that the task pointer can not be used any more
once the task is submitted, since it can be executed at any time (unless
dependencies make it wait) and thus freed at any time.
If the destroy flag is explicitly unset, the resources used
by the task have to be freed by calling
@code{starpu_task_destroy}.
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_dup ({struct starpu_task *}@var{task})
Allocate a task structure which is the exact duplicate of the given task.
@end deftypefun

@deftypefun void starpu_task_clean ({struct starpu_task} *@var{task})
Release all the structures automatically allocated to execute @var{task}, but
not the task structure itself and values set by the user remain unchanged.
It is thus useful for statically allocated tasks for instance.
It is also useful when the user wants to execute the same operation several
times with as least overhead as possible.
It is called automatically by @code{starpu_task_destroy}.
It has to be called only after explicitly waiting for the task or after
@code{starpu_shutdown} (waiting for the callback is not enough, since starpu
still manipulates the task after calling the callback).
@end deftypefun

@deftypefun void starpu_task_destroy ({struct starpu_task} *@var{task})
Free the resource allocated during @code{starpu_task_create} and
associated with @var{task}. This function is already called automatically
after the execution of a task when the @code{destroy} flag of the
@code{starpu_task} structure is set, which is the default for tasks created by
@code{starpu_task_create}.  Calling this function on a statically allocated task
results in an undefined behaviour.
@end deftypefun

@deftypefun int starpu_task_wait ({struct starpu_task} *@var{task})
This function blocks until @var{task} has been executed. It is not possible to
synchronize with a task more than once. It is not possible to wait for
synchronous or detached tasks.

Upon successful completion, this function returns 0. Otherwise, @code{-EINVAL}
indicates that the specified task was either synchronous or detached.
@end deftypefun

@deftypefun int starpu_task_submit ({struct starpu_task} *@var{task})
This function submits @var{task} to StarPU. Calling this function does
not mean that the task will be executed immediately as there can be data or task
(tag) dependencies that are not fulfilled yet: StarPU will take care of
scheduling this task with respect to such dependencies.
This function returns immediately if the @code{synchronous} field of the
@code{starpu_task} structure was set to 0, and block until the termination of
the task otherwise. It is also possible to synchronize the application with
asynchronous tasks by the means of tags, using the @code{starpu_tag_wait}
function for instance.

In case of success, this function returns 0, a return value of @code{-ENODEV}
means that there is no worker able to process this task (e.g. there is no GPU
available and this task is only implemented for CUDA devices).

starpu_task_submit() can be called from anywhere, including codelet
functions and callbacks, provided that the @code{synchronous} field of the
@code{starpu_task} structure is left to 0.
@end deftypefun

@deftypefun int starpu_task_wait_for_all (void)
This function blocks until all the tasks that were submitted are terminated. It
does not destroy these tasks.
@end deftypefun

@deftypefun int starpu_task_nready (void)
@end deftypefun

@deftypefun int starpu_task_nsubmitted (void)
Return the number of submitted tasks which have not completed yet.
@end deftypefun

@deftypefun int starpu_task_nready (void)
Return the number of submitted tasks which are ready for execution are already
executing. It thus does not include tasks waiting for dependencies.
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_get_current (void)
This function returns the task currently executed by the worker, or
NULL if it is called either from a thread that is not a task or simply
because there is no task being executed at the moment.
This function must be called from the callback (not from the codelet).
@end deftypefun

@deftypefun void starpu_codelet_display_stats ({struct starpu_codelet} *@var{cl})
@anchor{starpu_codelet_display_stats}
Output on @code{stderr} some statistics on the codelet @var{cl}.
@end deftypefun

@deftypefun int starpu_task_wait_for_no_ready (void)
This function waits until there is no more ready task.
@end deftypefun

@deftypefun void starpu_task_set_implementation ({struct starpu_task *}@var{task}, unsigned @var{impl})
This function should be called by schedulers to specify the codelet
implementation to be executed when executing the task.
@end deftypefun

@deftypefun unsigned starpu_task_get_implementation ({struct starpu_task *}@var{task})
This function return the codelet implementation to be executed when
executing the task.
@end deftypefun

@c Callbacks: what can we put in callbacks ?

@node Insert Task
@section Insert Task

@deftypefun int starpu_insert_task (struct starpu_codelet *@var{cl}, ...)
Create and submit a task corresponding to @var{cl} with the following
arguments.  The argument list must be zero-terminated.

The arguments following the codelets can be of the following types:

@itemize
@item
@code{STARPU_R}, @code{STARPU_W}, @code{STARPU_RW}, @code{STARPU_SCRATCH}, @code{STARPU_REDUX} an access mode followed by a data handle;
@item
@code{STARPU_DATA_ARRAY} followed by an array of data handles and its number of elements;
@item
the specific values @code{STARPU_VALUE}, @code{STARPU_CALLBACK},
@code{STARPU_CALLBACK_ARG}, @code{STARPU_CALLBACK_WITH_ARG},
@code{STARPU_PRIORITY}, @code{STARPU_TAG}, @code{STARPU_FLOPS}, @code{STARPU_SCHED_CTX} followed by the appropriated objects
as defined below.
@end itemize

When using @code{STARPU_DATA_ARRAY}, the access mode of the data
handles is not defined.

Parameters to be passed to the codelet implementation are defined
through the type @code{STARPU_VALUE}. The function
@code{starpu_codelet_unpack_args} must be called within the codelet
implementation to retrieve them.
@end deftypefun

@defmac STARPU_VALUE
this macro is used when calling @code{starpu_insert_task}, and must be
followed by a pointer to a constant value and the size of the constant
@end defmac

@defmac STARPU_CALLBACK
this macro is used when calling @code{starpu_insert_task}, and must be
followed by a pointer to a callback function
@end defmac

@defmac STARPU_CALLBACK_ARG
this macro is used when calling @code{starpu_insert_task}, and must be
followed by a pointer to be given as an argument to the callback
function
@end defmac

@defmac  STARPU_CALLBACK_WITH_ARG
this macro is used when calling @code{starpu_insert_task}, and must be
followed by two pointers: one to a callback function, and the other to
be given as an argument to the callback function; this is equivalent
to using both @code{STARPU_CALLBACK} and
@code{STARPU_CALLBACK_WITH_ARG}
@end defmac

@defmac STARPU_PRIORITY
this macro is used when calling @code{starpu_insert_task}, and must be
followed by a integer defining a priority level
@end defmac

@defmac STARPU_TAG
this macro is used when calling @code{starpu_insert_task}, and must be
followed by a tag.
@end defmac

@defmac STARPU_FLOPS
this macro is used when calling @code{starpu_insert_task}, and must be followed
by an amount of floating point operations, as a double. The user may have to
explicitly cast into double, otherwise parameter passing will not work.
@end defmac

@defmac STARPU_SCHED_CTX
this macro is used when calling @code{starpu_insert_task}, and must be followed
by the id of the scheduling context to which we want to submit the task.
@end defmac

@deftypefun void starpu_codelet_pack_args ({void **}@var{arg_buffer}, {size_t *}@var{arg_buffer_size}, ...)
Pack arguments of type @code{STARPU_VALUE} into a buffer which can be
given to a codelet and later unpacked with the function
@code{starpu_codelet_unpack_args} defined below.
@end deftypefun

@deftypefun void starpu_codelet_unpack_args ({void *}@var{cl_arg}, ...)
Retrieve the arguments of type @code{STARPU_VALUE} associated to a
task automatically created using the function
@code{starpu_insert_task} defined above.
@end deftypefun

@node Explicit Dependencies
@section Explicit Dependencies

@deftypefun void starpu_task_declare_deps_array ({struct starpu_task} *@var{task}, unsigned @var{ndeps}, {struct starpu_task} *@var{task_array}[])
Declare task dependencies between a @var{task} and an array of tasks of length
@var{ndeps}. This function must be called prior to the submission of the task,
but it may called after the submission or the execution of the tasks in the
array, provided the tasks are still valid (ie. they were not automatically
destroyed). Calling this function on a task that was already submitted or with
an entry of @var{task_array} that is not a valid task anymore results in an
undefined behaviour. If @var{ndeps} is null, no dependency is added. It is
possible to call @code{starpu_task_declare_deps_array} multiple times on the
same task, in this case, the dependencies are added. It is possible to have
redundancy in the task dependencies.
@end deftypefun

@deftp {Data Type} {starpu_tag_t}
This type defines a task logical identifer. It is possible to associate a task with a unique ``tag'' chosen by the application, and to express
dependencies between tasks by the means of those tags. To do so, fill the
@code{tag_id} field of the @code{starpu_task} structure with a tag number (can
be arbitrary) and set the @code{use_tag} field to 1.

If @code{starpu_tag_declare_deps} is called with this tag number, the task will
not be started until the tasks which holds the declared dependency tags are
completed.
@end deftp

@deftypefun void starpu_tag_declare_deps (starpu_tag_t @var{id}, unsigned @var{ndeps}, ...)
Specify the dependencies of the task identified by tag @var{id}. The first
argument specifies the tag which is configured, the second argument gives the
number of tag(s) on which @var{id} depends. The following arguments are the
tags which have to be terminated to unlock the task.

This function must be called before the associated task is submitted to StarPU
with @code{starpu_task_submit}.

Because of the variable arity of @code{starpu_tag_declare_deps}, note that the
last arguments @emph{must} be of type @code{starpu_tag_t}: constant values
typically need to be explicitly casted. Using the
@code{starpu_tag_declare_deps_array} function avoids this hazard.

@cartouche
@smallexample
/*  Tag 0x1 depends on tags 0x32 and 0x52 */
starpu_tag_declare_deps((starpu_tag_t)0x1,
        2, (starpu_tag_t)0x32, (starpu_tag_t)0x52);
@end smallexample
@end cartouche
@end deftypefun

@deftypefun void starpu_tag_declare_deps_array (starpu_tag_t @var{id}, unsigned @var{ndeps}, {starpu_tag_t *}@var{array})
This function is similar to @code{starpu_tag_declare_deps}, except
that its does not take a variable number of arguments but an array of
tags of size @var{ndeps}.
@cartouche
@smallexample
/*  Tag 0x1 depends on tags 0x32 and 0x52 */
starpu_tag_t tag_array[2] = @{0x32, 0x52@};
starpu_tag_declare_deps_array((starpu_tag_t)0x1, 2, tag_array);
@end smallexample
@end cartouche
@end deftypefun

@deftypefun int starpu_tag_wait (starpu_tag_t @var{id})
This function blocks until the task associated to tag @var{id} has been
executed. This is a blocking call which must therefore not be called within
tasks or callbacks, but only from the application directly.  It is possible to
synchronize with the same tag multiple times, as long as the
@code{starpu_tag_remove} function is not called.  Note that it is still
possible to synchronize with a tag associated to a task which @code{starpu_task}
data structure was freed (e.g. if the @code{destroy} flag of the
@code{starpu_task} was enabled).
@end deftypefun

@deftypefun int starpu_tag_wait_array (unsigned @var{ntags}, starpu_tag_t *@var{id})
This function is similar to @code{starpu_tag_wait} except that it blocks until
@emph{all} the @var{ntags} tags contained in the @var{id} array are
terminated.
@end deftypefun

@deftypefun void starpu_tag_restart (starpu_tag_t @var{id})
This function can be used to clear the "already notified" status
of a tag which is not associated with a task. Before that, calling
@code{starpu_tag_notify_from_apps} again will not notify the successors. After
that, the next call to @code{starpu_tag_notify_from_apps} will notify the
successors.
@end deftypefun

@deftypefun void starpu_tag_remove (starpu_tag_t @var{id})
This function releases the resources associated to tag @var{id}. It can be
called once the corresponding task has been executed and when there is
no other tag that depend on this tag anymore.
@end deftypefun

@deftypefun void starpu_tag_notify_from_apps (starpu_tag_t @var{id})
This function explicitly unlocks tag @var{id}. It may be useful in the
case of applications which execute part of their computation outside StarPU
tasks (e.g. third-party libraries).  It is also provided as a
convenient tool for the programmer, for instance to entirely construct the task
DAG before actually giving StarPU the opportunity to execute the tasks. When
called several times on the same tag, notification will be done only on first
call, thus implementing "OR" dependencies, until the tag is restarted using
@code{starpu_tag_restart}.
@end deftypefun

@node Implicit Data Dependencies
@section Implicit Data Dependencies

In this section, we describe how StarPU makes it possible to insert implicit
task dependencies in order to enforce sequential data consistency. When this
data consistency is enabled on a specific data handle, any data access will
appear as sequentially consistent from the application. For instance, if the
application submits two tasks that access the same piece of data in read-only
mode, and then a third task that access it in write mode, dependencies will be
added between the two first tasks and the third one. Implicit data dependencies
are also inserted in the case of data accesses from the application.

@deftypefun void starpu_data_set_default_sequential_consistency_flag (unsigned @var{flag})
Set the default sequential consistency flag. If a non-zero value is passed, a
sequential data consistency will be enforced for all handles registered after
this function call, otherwise it is disabled. By default, StarPU enables
sequential data consistency. It is also possible to select the data consistency
mode of a specific data handle with the
@code{starpu_data_set_sequential_consistency_flag} function.
@end deftypefun

@deftypefun unsigned starpu_data_get_default_sequential_consistency_flag (void)
Return the default sequential consistency flag
@end deftypefun

@deftypefun void starpu_data_set_sequential_consistency_flag (starpu_data_handle_t @var{handle}, unsigned @var{flag})
Sets the data consistency mode associated to a data handle. The consistency
mode set using this function has the priority over the default mode which can
be set with @code{starpu_data_set_default_sequential_consistency_flag}.
@end deftypefun

@node Performance Model API
@section Performance Model API

@deftp {Data Type} {enum starpu_perfmodel_archtype}
Enumerates the various types of architectures.
CPU types range within STARPU_CPU_DEFAULT (1 CPU), STARPU_CPU_DEFAULT+1 (2 CPUs), ... STARPU_CPU_DEFAULT + STARPU_MAXCPUS - 1 (STARPU_MAXCPUS CPUs).
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).
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).
@table @asis
@item @code{STARPU_CPU_DEFAULT}
@item @code{STARPU_CUDA_DEFAULT}
@item @code{STARPU_OPENCL_DEFAULT}
@end table
@end deftp

@deftp {Data Type} {enum starpu_perfmodel_type}
The possible values are:
@table @asis
@item @code{STARPU_PER_ARCH} for application-provided per-arch cost model functions.
@item @code{STARPU_COMMON} for application-provided common cost model function, with per-arch factor.
@item @code{STARPU_HISTORY_BASED} for automatic history-based cost model.
@item @code{STARPU_REGRESSION_BASED} for automatic linear regression-based cost model (alpha * size ^ beta).
@item @code{STARPU_NL_REGRESSION_BASED} for automatic non-linear regression-based cost mode (a * size ^ b + c).
@end table
@end deftp

@deftp {Data Type} {struct starpu_perfmodel}
@anchor{struct starpu_perfmodel}
contains all information about a performance model. At least the
@code{type} and @code{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.

@table @asis
@item @code{type}
is the type of performance model @code{enum starpu_perfmodel_type}:
@code{STARPU_HISTORY_BASED},
@code{STARPU_REGRESSION_BASED}, @code{STARPU_NL_REGRESSION_BASED}: No
other fields needs to be provided, this is purely history-based. @code{STARPU_PER_ARCH}:
@code{per_arch} has to be filled with functions which return the cost in
micro-seconds. @code{STARPU_COMMON}: @code{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.

@item @code{const char *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.

@item @code{double (*cost_model)(struct starpu_data_descr *)}
This field is deprecated. Use instead the @code{cost_function} field.

@item @code{double (*cost_function)(struct starpu_task *, unsigned nimpl)}
Used by @code{STARPU_COMMON}: takes a task and
implementation number, and must return a task duration estimation in micro-seconds.

@item @code{size_t (*size_base)(struct starpu_task *, unsigned nimpl)}
Used by @code{STARPU_HISTORY_BASED} and
@code{STARPU_*REGRESSION_BASED}. If not NULL, takes a task and
implementation number, and returns the size to be used as index for
history and regression.

@item @code{struct starpu_perfmodel_per_arch per_arch[STARPU_NARCH_VARIATIONS][STARPU_MAXIMPLEMENTATIONS]}
Used by @code{STARPU_PER_ARCH}: array of @code{struct
starpu_per_arch_perfmodel} structures.

@item @code{unsigned is_loaded}
Whether the performance model is already loaded from the disk.

@item @code{unsigned benchmarking}
Whether the performance model is still being calibrated.

@item @code{pthread_rwlock_t model_rwlock}
Lock to protect concurrency between loading from disk (W), updating the values
(W), and making a performance estimation (R).

@end table
@end deftp

@deftp {Data Type} {struct starpu_perfmodel_regression_model}
@table @asis
@item @code{double sumlny} sum of ln(measured)
@item @code{double sumlnx} sum of ln(size)
@item @code{double sumlnx2} sum of ln(size)^2
@item @code{unsigned long minx} minimum size
@item @code{unsigned long maxx} maximum size
@item @code{double sumlnxlny} sum of ln(size)*ln(measured)
@item @code{double alpha} 	 estimated = alpha * size ^ beta
@item @code{double beta}
@item @code{unsigned valid} whether the linear regression model is valid (i.e. enough measures)
@item @code{double a, b, c} estimaed = a size ^b + c
@item @code{unsigned nl_valid} whether the non-linear regression model is valid (i.e. enough measures)
@item @code{unsigned nsample} number of sample values for non-linear regression
@end table
@end deftp

@deftp {Data Type} {struct starpu_perfmodel_per_arch}
contains information about the performance model of a given arch.

@table @asis
@item @code{double (*cost_model)(struct starpu_data_descr *t)}
This field is deprecated. Use instead the @code{cost_function} field.

@item @code{double (*cost_function)(struct starpu_task *task, enum starpu_perfmodel_archtype arch, unsigned nimpl)}
Used by @code{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.

@item @code{size_t (*size_base)(struct starpu_task *, enum starpu_perfmodel_archtype arch, unsigned nimpl)}
Same as in @ref{struct starpu_perfmodel}, but per-arch, in
case it depends on the architecture-specific implementation.

@item @code{struct starpu_htbl32_node *history}
The history of performance measurements.

@item @code{struct starpu_perfmodel_history_list *list}
Used by @code{STARPU_HISTORY_BASED} and @code{STARPU_NL_REGRESSION_BASED},
records all execution history measures.

@item @code{struct starpu_perfmodel_regression_model regression}
Used by @code{STARPU_HISTORY_REGRESION_BASED} and
@code{STARPU_NL_REGRESSION_BASED}, contains the estimated factors of the
regression.

@end table
@end deftp

@deftypefun int starpu_perfmodel_load_symbol ({const char} *@var{symbol}, {struct starpu_perfmodel} *@var{model})
loads a given performance model. The @var{model} structure has to be completely zero, and will be filled with the information saved in @code{$STARPU_HOME/.starpu}.
@end deftypefun

@deftypefun int starpu_perfmodel_unload_model ({struct starpu_perfmodel} *@var{model})
unloads the given model which has been previously loaded through the function @code{starpu_perfmodel_load_symbol}
@end deftypefun

@deftypefun void starpu_perfmodel_debugfilepath ({struct starpu_perfmodel} *@var{model}, {enum starpu_perfmodel_archtype} @var{arch}, char *@var{path}, size_t @var{maxlen}, unsigned nimpl)
returns the path to the debugging information for the performance model.
@end deftypefun

@deftypefun void starpu_perfmodel_get_arch_name ({enum starpu_perfmodel_archtype} @var{arch}, char *@var{archname}, size_t @var{maxlen}, unsigned nimpl)
returns the architecture name for @var{arch}.
@end deftypefun

@deftypefun {enum starpu_perfmodel_archtype} starpu_worker_get_perf_archtype (int @var{workerid})
returns the architecture type of a given worker.
@end deftypefun

@deftypefun int starpu_perfmodel_list ({FILE *}@var{output})
prints a list of all performance models on @var{output}.
@end deftypefun

@deftypefun void starpu_perfmodel_print ({struct starpu_perfmodel *}@var{model}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned @var{nimpl}, {char *}@var{parameter}, {uint32_t *}footprint, {FILE *}@var{output})
todo
@end deftypefun

@deftypefun int starpu_perfmodel_print_all ({struct starpu_perfmodel *}@var{model}, {char *}@var{arch}, @var{char *}parameter, {uint32_t *}@var{footprint}, {FILE *}@var{output})
todo
@end deftypefun

@deftypefun void starpu_bus_print_bandwidth ({FILE *}@var{f})
prints a matrix of bus bandwidths on @var{f}.
@end deftypefun

@deftypefun void starpu_bus_print_affinity ({FILE *}@var{f})
prints the affinity devices on @var{f}.
@end deftypefun

@deftypefun void starpu_topology_print ({FILE *}@var{f})
prints a description of the topology on @var{f}.
@end deftypefun

@deftypefun void starpu_perfmodel_update_history ({struct starpu_perfmodel *}@var{model}, {struct starpu_task *}@var{task}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned @var{cpuid}, unsigned @var{nimpl}, double @var{measured});
This feeds the performance model @var{model} with an explicit measurement
@var{measured}, 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 see in @ref{Performance model example}.
@end deftypefun

@node Profiling API
@section Profiling API

@deftypefun int starpu_profiling_status_set (int @var{status})
Thie function sets the profiling status. Profiling is activated by passing
@code{STARPU_PROFILING_ENABLE} in @var{status}. Passing
@code{STARPU_PROFILING_DISABLE} disables profiling. Calling this function
resets all profiling measurements. When profiling is enabled, the
@code{profiling_info} field of the @code{struct starpu_task} structure points
to a valid @code{struct starpu_profiling_task_info} structure containing
information about the execution of the task.

Negative return values indicate an error, otherwise the previous status is
returned.
@end deftypefun

@deftypefun int starpu_profiling_status_get (void)
Return the current profiling status or a negative value in case there was an error.
@end deftypefun

@deftypefun void starpu_profiling_set_id (int @var{new_id})
This function sets the ID used for profiling trace filename. It needs to be
called before starpu_init.
@end deftypefun

@deftp {Data Type} {struct starpu_profiling_task_info}
This structure contains information about the execution of a task. It is
accessible from the @code{.profiling_info} field of the @code{starpu_task}
structure if profiling was enabled. The different fields are:

@table @asis
@item @code{struct timespec submit_time}
Date of task submission (relative to the initialization of StarPU).

@item @code{struct timespec push_start_time}
Time when the task was submitted to the scheduler.

@item @code{struct timespec push_end_time}
Time when the scheduler finished with the task submission.

@item @code{struct timespec pop_start_time}
Time when the scheduler started to be requested for a task, and eventually gave
that task.

@item @code{struct timespec pop_end_time}
Time when the scheduler finished providing the task for execution.

@item @code{struct timespec acquire_data_start_time}
Time when the worker started fetching input data.

@item @code{struct timespec acquire_data_end_time}
Time when the worker finished fetching input data.

@item @code{struct timespec start_time}
Date of task execution beginning (relative to the initialization of StarPU).

@item @code{struct timespec end_time}
Date of task execution termination (relative to the initialization of StarPU).

@item @code{struct timespec release_data_start_time}
Time when the worker started releasing data.

@item @code{struct timespec release_data_end_time}
Time when the worker finished releasing data.

@item @code{struct timespec callback_start_time}
Time when the worker started the application callback for the task.

@item @code{struct timespec callback_end_time}
Time when the worker finished the application callback for the task.

@item @code{workerid}
Identifier of the worker which has executed the task.

@item @code{uint64_t used_cycles}
Number of cycles used by the task, only available in the MoviSim

@item @code{uint64_t stall_cycles}
Number of cycles stalled within the task, only available in the MoviSim

@item @code{double power_consumed}
Power consumed by the task, only available in the MoviSim

@end table
@end deftp

@deftp {Data Type} {struct starpu_profiling_worker_info}
This structure contains the profiling information associated to a
worker. The different fields are:

@table @asis
@item @code{struct timespec start_time}
Starting date for the reported profiling measurements.

@item @code{struct timespec total_time}
Duration of the profiling measurement interval.

@item @code{struct timespec executing_time}
Time spent by the worker to execute tasks during the profiling measurement interval.

@item @code{struct timespec sleeping_time}
Time spent idling by the worker during the profiling measurement interval.

@item @code{int executed_tasks}
Number of tasks executed by the worker during the profiling measurement interval.

@item @code{uint64_t used_cycles}
Number of cycles used by the worker, only available in the MoviSim

@item @code{uint64_t stall_cycles}
Number of cycles stalled within the worker, only available in the MoviSim

@item @code{double power_consumed}
Power consumed by the worker, only available in the MoviSim

@end table
@end deftp

@deftypefun int starpu_profiling_worker_get_info (int @var{workerid}, {struct starpu_profiling_worker_info *}@var{worker_info})
Get the profiling info associated to the worker identified by @var{workerid},
and reset the profiling measurements. If the @var{worker_info} argument is
NULL, only reset the counters associated to worker @var{workerid}.

Upon successful completion, this function returns 0. Otherwise, a negative
value is returned.
@end deftypefun

@deftp {Data Type} {struct starpu_profiling_bus_info}
The different fields are:
@table @asis
@item @code{struct timespec start_time}
Time of bus profiling startup.

@item @code{struct timespec total_time}
Total time of bus profiling.

@item @code{int long long transferred_bytes}
Number of bytes transferred during profiling.

@item @code{int transfer_count}
Number of transfers during profiling.
@end table
@end deftp

@deftypefun int starpu_bus_get_profiling_info (int @var{busid}, {struct starpu_profiling_bus_info *}@var{bus_info})
Get the profiling info associated to the worker designated by @var{workerid},
and reset the profiling measurements. If worker_info is NULL, only reset the
counters.
@end deftypefun

@deftypefun int starpu_bus_get_count (void)
Return the number of buses in the machine.
@end deftypefun

@deftypefun int starpu_bus_get_id (int @var{src}, int @var{dst})
Return the identifier of the bus between @var{src} and @var{dst}
@end deftypefun

@deftypefun int starpu_bus_get_src (int @var{busid})
Return the source point of bus @var{busid}
@end deftypefun

@deftypefun int starpu_bus_get_dst (int @var{busid})
Return the destination point of bus @var{busid}
@end deftypefun

@deftypefun double starpu_timing_timespec_delay_us ({struct timespec} *@var{start}, {struct timespec} *@var{end})
Returns the time elapsed between @var{start} and @var{end} in microseconds.
@end deftypefun

@deftypefun double starpu_timing_timespec_to_us ({struct timespec} *@var{ts})
Converts the given timespec @var{ts} into microseconds.
@end deftypefun

@deftypefun void starpu_profiling_bus_helper_display_summary (void)
Displays statistics about the bus on stderr. if the  environment
variable @code{STARPU_BUS_STATS} is defined. The function is called
automatically by @code{starpu_shutdown()}.
@end deftypefun

@deftypefun void starpu_profiling_worker_helper_display_summary (void)
Displays statistics about the workers on stderr if the environment
variable @code{STARPU_WORKER_STATS} is defined. The function is called
automatically by @code{starpu_shutdown()}.
@end deftypefun

@deftypefun void starpu_data_display_memory_stats ()
Display statistics about the current data handles registered within
StarPU. StarPU must have been configured with the option
@code{----enable-memory-stats} (@pxref{Memory feedback}).
@end deftypefun

@node Theoretical lower bound on execution time API
@section Theoretical lower bound on execution time

@deftypefun void starpu_bound_start (int @var{deps}, int @var{prio})
Start recording tasks (resets stats).  @var{deps} tells whether
dependencies should be recorded too (this is quite expensive)
@end deftypefun

@deftypefun void starpu_bound_stop (void)
Stop recording tasks
@end deftypefun

@deftypefun void starpu_bound_print_dot ({FILE *}@var{output})
Print the DAG that was recorded
@end deftypefun

@deftypefun void starpu_bound_compute ({double *}@var{res}, {double *}@var{integer_res}, int @var{integer})
Get theoretical upper bound (in ms) (needs glpk support detected by @code{configure} script). It returns 0 if some performance models are not calibrated.
@end deftypefun

@deftypefun void starpu_bound_print_lp ({FILE *}@var{output})
Emit the Linear Programming system on @var{output} for the recorded tasks, in
the lp format
@end deftypefun

@deftypefun void starpu_bound_print_mps ({FILE *}@var{output})
Emit the Linear Programming system on @var{output} for the recorded tasks, in
the mps format
@end deftypefun

@deftypefun void starpu_bound_print ({FILE *}@var{output}, int @var{integer})
Emit statistics of actual execution vs theoretical upper bound. @var{integer}
permits to choose between integer solving (which takes a long time but is
correct), and relaxed solving (which provides an approximate solution).
@end deftypefun

@node CUDA extensions
@section CUDA extensions

@defmac STARPU_USE_CUDA
This macro is defined when StarPU has been installed with CUDA
support. It should be used in your code to detect the availability of
CUDA as shown in @ref{Full source code for the 'Scaling a Vector' example}.
@end defmac

@deftypefun cudaStream_t starpu_cuda_get_local_stream (void)
This function gets the current worker's CUDA stream.
StarPU provides a stream for every CUDA device controlled by StarPU. This
function is only provided for convenience so that programmers can easily use
asynchronous operations within codelets without having to create a stream by
hand. Note that the application is not forced to use the stream provided by
@code{starpu_cuda_get_local_stream} and may also create its own streams.
Synchronizing with @code{cudaThreadSynchronize()} is allowed, but will reduce
the likelihood of having all transfers overlapped.
@end deftypefun

@deftypefun {const struct cudaDeviceProp *} starpu_cuda_get_device_properties (unsigned @var{workerid})
This function returns a pointer to device properties for worker @var{workerid}
(assumed to be a CUDA worker).
@end deftypefun

@deftypefun void starpu_cuda_report_error ({const char *}@var{func}, {const char *}@var{file}, int @var{line}, cudaError_t @var{status})
Report a CUDA error.
@end deftypefun

@defmac STARPU_CUDA_REPORT_ERROR (cudaError_t @var{status})
Calls starpu_cuda_report_error, passing the current function, file and line
position.
@end defmac

@deftypefun int starpu_cuda_copy_async_sync ({void *}@var{src_ptr}, unsigned @var{src_node}, {void *}@var{dst_ptr}, unsigned @var{dst_node}, size_t @var{ssize}, cudaStream_t @var{stream}, {enum cudaMemcpyKind} @var{kind})
Copy @var{ssize} bytes from the pointer @var{src_ptr} on
@var{src_node} to the pointer @var{dst_ptr} on @var{dst_node}.
The function first tries to copy the data asynchronous (unless
@var{stream} is @code{NULL}. If the asynchronous copy fails or if
@var{stream} is @code{NULL}, it copies the data synchronously.
The function returns @code{-EAGAIN} if the asynchronous launch was
successfull. It returns 0 if the synchronous copy was successful, or
fails otherwise.
@end deftypefun

@deftypefun void starpu_cuda_set_device (unsigned @var{devid})
Calls @code{cudaSetDevice(devid)} or @code{cudaGLSetGLDevice(devid)}, according to
whether @code{devid} is among the @code{cuda_opengl_interoperability} field of
the @code{starpu_conf} structure.
@end deftypefun

@deftypefun void starpu_cublas_init (void)
This function initializes CUBLAS on every CUDA device.
The CUBLAS library must be initialized prior to any CUBLAS call. Calling
@code{starpu_cublas_init} will initialize CUBLAS on every CUDA device
controlled by StarPU. This call blocks until CUBLAS has been properly
initialized on every device.
@end deftypefun

@deftypefun void starpu_cublas_shutdown (void)
This function synchronously deinitializes the CUBLAS library on every CUDA device.
@end deftypefun

@deftypefun void starpu_cublas_report_error ({const char *}@var{func}, {const char *}@var{file}, int @var{line}, cublasStatus @var{status})
Report a cublas error.
@end deftypefun

@defmac STARPU_CUBLAS_REPORT_ERROR (cublasStatus @var{status})
Calls starpu_cublas_report_error, passing the current function, file and line
position.
@end defmac

@node OpenCL extensions
@section OpenCL extensions

@menu
* Writing OpenCL kernels::      Writing OpenCL kernels
* Compiling OpenCL kernels::    Compiling OpenCL kernels
* Loading OpenCL kernels::      Loading OpenCL kernels
* OpenCL statistics::           Collecting statistics from OpenCL
* OpenCL utilities::            Utilities for OpenCL
@end menu

@defmac STARPU_USE_OPENCL
This macro is defined when StarPU has been installed with OpenCL
support. It should be used in your code to detect the availability of
OpenCL as shown in @ref{Full source code for the 'Scaling a Vector' example}.
@end defmac

@node Writing OpenCL kernels
@subsection Writing OpenCL kernels

@deftypefun void starpu_opencl_get_context (int @var{devid}, {cl_context *}@var{context})
Places the OpenCL context of the device designated by @var{devid} into @var{context}.
@end deftypefun

@deftypefun void starpu_opencl_get_device (int @var{devid}, {cl_device_id *}@var{device})
Places the cl_device_id corresponding to @var{devid} in @var{device}.
@end deftypefun

@deftypefun void starpu_opencl_get_queue (int @var{devid}, {cl_command_queue *}@var{queue})
Places the command queue of the the device designated by @var{devid} into @var{queue}.
@end deftypefun

@deftypefun void starpu_opencl_get_current_context ({cl_context *}@var{context})
Return the context of the current worker.
@end deftypefun

@deftypefun void starpu_opencl_get_current_queue ({cl_command_queue *}@var{queue})
Return the computation kernel command queue of the current worker.
@end deftypefun

@deftypefun int starpu_opencl_set_kernel_args ({cl_int *}@var{err}, {cl_kernel *}@var{kernel}, ...)
Sets the arguments of a given kernel. The list of arguments must be given as
(size_t @var{size_of_the_argument}, cl_mem * @var{pointer_to_the_argument}).
The last argument must be 0. Returns the number of arguments that were
successfully set. In case of failure, returns the id of the argument
that could not be set and @var{err} is set to the error returned by
OpenCL. Otherwise, returns the number of arguments that were set.

@cartouche
@smallexample
int n;
cl_int err;
cl_kernel kernel;
n = starpu_opencl_set_kernel_args(&err, 2, &kernel,
                                  sizeof(foo), &foo,
                                  sizeof(bar), &bar,
                                  0);
if (n != 2)
   fprintf(stderr, "Error : %d\n", err);
@end smallexample
@end cartouche
@end deftypefun

@node Compiling OpenCL kernels
@subsection Compiling OpenCL kernels

Source codes for OpenCL kernels can be stored in a file or in a
string. StarPU provides functions to build the program executable for
each available OpenCL device as a @code{cl_program} object. This
program executable can then be loaded within a specific queue as
explained in the next section. These are only helpers, Applications
can also fill a @code{starpu_opencl_program} array by hand for more advanced
use (e.g. different programs on the different OpenCL devices, for
relocation purpose for instance).

@deftp {Data Type} {struct starpu_opencl_program}
Stores the OpenCL programs as compiled for the different OpenCL
devices. The different fields are:
@table @asis
@item @code{cl_program programs[STARPU_MAXOPENCLDEVS]}
Stores each program for each OpenCL device.
@end table
@end deftp

@deftypefun int starpu_opencl_load_opencl_from_file ({const char} *@var{source_file_name}, {struct starpu_opencl_program} *@var{opencl_programs}, {const char}* @var{build_options})
@anchor{starpu_opencl_load_opencl_from_file}
This function compiles an OpenCL source code stored in a file.
@end deftypefun

@deftypefun int starpu_opencl_load_opencl_from_string ({const char} *@var{opencl_program_source}, {struct starpu_opencl_program} *@var{opencl_programs}, {const char}* @var{build_options})
This function compiles an OpenCL source code stored in a string.
@end deftypefun

@deftypefun int starpu_opencl_unload_opencl ({struct starpu_opencl_program} *@var{opencl_programs})
This function unloads an OpenCL compiled code.
@end deftypefun

@deftypefun void starpu_opencl_load_program_source ({const char *}@var{source_file_name}, char *@var{located_file_name}, char *@var{located_dir_name}, char *@var{opencl_program_source})
@anchor{starpu_opencl_load_program_source}
Store the contents of the file @var{source_file_name} in the buffer
@var{opencl_program_source}. The file @var{source_file_name} can be
located in the current directory, or in the directory specified by the
environment variable @code{STARPU_OPENCL_PROGRAM_DIR} (@pxref{STARPU_OPENCL_PROGRAM_DIR}), or in the
directory @code{share/starpu/opencl} of the installation directory of
StarPU, or in the source directory of StarPU.
When the file is found, @code{located_file_name} is the full name of
the file as it has been located on the system, @code{located_dir_name}
the directory where it has been located. Otherwise, they are both set
to the empty string.
@end deftypefun

@deftypefun int starpu_opencl_compile_opencl_from_file ({const char *}@var{source_file_name}, {const char *} @var{build_options})
Compile the OpenCL kernel stored in the file @code{source_file_name}
with the given options @code{build_options} and stores the result in
the directory @code{$STARPU_HOME/.starpu/opencl} with the same
filename as @code{source_file_name}. The compilation is done for every
OpenCL device, and the filename is suffixed with the vendor id and the
device id of the OpenCL device.
@end deftypefun

@deftypefun int starpu_opencl_compile_opencl_from_string ({const char *}@var{opencl_program_source}, {const char *}@var{file_name}, {const char* }@var{build_options})
Compile the OpenCL kernel in the string @code{opencl_program_source}
with the given options @code{build_options} and stores the result in
the directory @code{$STARPU_HOME/.starpu/opencl}
with the filename
@code{file_name}. The compilation is done for every
OpenCL device, and the filename is suffixed with the vendor id and the
device id of the OpenCL device.
@end deftypefun

@deftypefun int starpu_opencl_load_binary_opencl ({const char *}@var{kernel_id}, {struct starpu_opencl_program *}@var{opencl_programs})
Compile the binary OpenCL kernel identified with @var{id}. For every
OpenCL device, the binary OpenCL kernel will be loaded from the file
@code{$STARPU_HOME/.starpu/opencl/<kernel_id>.<device_type>.vendor_id_<vendor_id>_device_id_<device_id>}.
@end deftypefun

@node Loading OpenCL kernels
@subsection Loading OpenCL kernels

@deftypefun int starpu_opencl_load_kernel (cl_kernel *@var{kernel}, cl_command_queue *@var{queue}, {struct starpu_opencl_program} *@var{opencl_programs}, {const char} *@var{kernel_name}, int @var{devid})
Create a kernel @var{kernel} for device @var{devid}, on its computation command
queue returned in @var{queue}, using program @var{opencl_programs} and name
@var{kernel_name}
@end deftypefun

@deftypefun int starpu_opencl_release_kernel (cl_kernel @var{kernel})
Release the given @var{kernel}, to be called after kernel execution.
@end deftypefun

@node OpenCL statistics
@subsection OpenCL statistics

@deftypefun int starpu_opencl_collect_stats (cl_event @var{event})
This function allows to collect statistics on a kernel execution.
After termination of the kernels, the OpenCL codelet should call this function
to pass it the even returned by @code{clEnqueueNDRangeKernel}, to let StarPU
collect statistics about the kernel execution (used cycles, consumed power).
@end deftypefun

@node OpenCL utilities
@subsection OpenCL utilities

@deftypefun {const char *} starpu_opencl_error_string (cl_int @var{status})
Return the error message in English corresponding to @var{status}, an
OpenCL error code.
@end deftypefun

@deftypefun void starpu_opencl_display_error ({const char *}@var{func}, {const char *}@var{file}, int @var{line}, {const char *}@var{msg}, cl_int @var{status})
Given a valid error @var{status}, prints the corresponding error message on
stdout, along with the given function name @var{func}, the given filename
@var{file}, the given line number @var{line} and the given message @var{msg}.
@end deftypefun

@defmac STARPU_OPENCL_DISPLAY_ERROR (cl_int @var{status})
Call the function @code{starpu_opencl_display_error} with the given
error @var{status}, the current function name, current file and line
number, and a empty message.
@end defmac

@deftypefun void starpu_opencl_report_error ({const char *}@var{func}, {const char *}@var{file}, int @var{line}, {const char *}@var{msg}, cl_int @var{status})
Call the function @code{starpu_opencl_display_error} and abort.
@end deftypefun

@defmac STARPU_OPENCL_REPORT_ERROR (cl_int @var{status})
Call the function @code{starpu_opencl_report_error} with the given
error @var{status}, with the current function name, current file and
line number, and a empty message.
@end defmac

@defmac STARPU_OPENCL_REPORT_ERROR_WITH_MSG ({const char *}@var{msg}, cl_int @var{status})
Call the function @code{starpu_opencl_report_error} with the given
message and the given error @var{status}, with the current function
name, current file and line number.
@end defmac

@deftypefun cl_int starpu_opencl_allocate_memory ({cl_mem *}@var{addr}, size_t @var{size}, cl_mem_flags @var{flags})
Allocate @var{size} bytes of memory, stored in @var{addr}. @var{flags} must be a
valid combination of cl_mem_flags values.
@end deftypefun

@deftypefun cl_int starpu_opencl_copy_ram_to_opencl ({void *}@var{ptr}, unsigned @var{src_node}, cl_mem @var{buffer}, unsigned @var{dst_node}, size_t @var{size}, size_t @var{offset}, {cl_event *}@var{event}, {int *}@var{ret})
Copy @var{size} bytes from the given @var{ptr} on
RAM @var{src_node} to the given @var{buffer} on OpenCL @var{dst_node}.
@var{offset} is the offset, in bytes, in @var{buffer}.
if @var{event} is NULL, the copy is synchronous, i.e the queue is
synchronised before returning. If non NULL, @var{event} can be used
after the call to wait for this particular copy to complete.
This function returns CL_SUCCESS if the copy was successful, or a valid OpenCL error code
otherwise. The integer pointed to by @var{ret} is set to -EAGAIN if the asynchronous launch
was successful, or to 0 if event was NULL.
@end deftypefun

@deftypefun cl_int starpu_opencl_copy_opencl_to_ram (cl_mem @var{buffer}, unsigned @var{src_node}, void *@var{ptr}, unsigned @var{dst_node}, size_t @var{size}, size_t @var{offset}, {cl_event *}@var{event}, {int *}@var{ret})
Copy @var{size} bytes asynchronously from the given @var{buffer} on
OpenCL @var{src_node} to the given @var{ptr} on RAM @var{dst_node}.
@var{offset} is the offset, in bytes, in @var{buffer}.
if @var{event} is NULL, the copy is synchronous, i.e the queue is
synchronised before returning. If non NULL, @var{event} can be used
after the call to wait for this particular copy to complete.
This function returns CL_SUCCESS if the copy was successful, or a valid OpenCL error code
otherwise. The integer pointed to by @var{ret} is set to -EAGAIN if the asynchronous launch
was successful, or to 0 if event was NULL.
@end deftypefun

@deftypefun cl_int starpu_opencl_copy_opencl_to_opencl (cl_mem @var{src}, unsigned @var{src_node}, size_t @var{src_offset}, cl_mem @var{dst}, unsigned @var{dst_node}, size_t @var{dst_offset}, size_t @var{size}, {cl_event *}@var{event}, {int *}@var{ret})
Copy @var{size} bytes asynchronously from byte offset @var{src_offset} of
@var{src} on OpenCL @var{src_node} to byte offset @var{dst_offset} of @var{dst} on
OpenCL @var{dst_node}.
if @var{event} is NULL, the copy is synchronous, i.e the queue is
synchronised before returning. If non NULL, @var{event} can be used
after the call to wait for this particular copy to complete.
This function returns CL_SUCCESS if the copy was successful, or a valid OpenCL error code
otherwise. The integer pointed to by @var{ret} is set to -EAGAIN if the asynchronous launch
was successful, or to 0 if event was NULL.
@end deftypefun

@deftypefun cl_int starpu_opencl_copy_async_sync (uintptr_t @var{src}, size_t @var{src_offset}, unsigned @var{src_node}, uintptr_t @var{dst}, size_t @var{dst_offset}, unsigned @var{dst_node}, size_t @var{size}, {cl_event *}@var{event})
Copy @var{size} bytes from byte offset @var{src_offset} of
@var{src} on @var{src_node} to byte offset @var{dst_offset} of @var{dst} on
@var{dst_node}. if @var{event} is NULL, the copy is synchronous, i.e the queue is
synchronised before returning. If non NULL, @var{event} can be used
after the call to wait for this particular copy to complete.
The function returns @code{-EAGAIN} if the asynchronous launch was
successfull. It returns 0 if the synchronous copy was successful, or
fails otherwise.
@end deftypefun

@node MIC extensions
@section MIC extensions

@defmac STARPU_USE_MIC
This macro is defined when StarPU has been installed with MIC
support. It should be used in your code to detect the availability of
MIC.
@end defmac

@deftypefun int starpu_mic_register_kernel({starpu_mic_func_symbol_t *}@var{symbol}, {const char *}@var{func_name})
Initiate a lookup on each MIC device to find the adress of the function
named FUNC_NAME, store them in the global array kernels and return
the index in the array through SYMBOL.
@end deftypefun

@deftypefun starpu_mic_kernel_t starpu_mic_get_kernel(starpu_mic_func_symbol_t @var{symbol})
If success, return the pointer to the function defined by SYMBOL on the
device linked to the called device. This can for instance be used in a
@code{starpu_mic_func_t} implementation.
@end deftypefun

@node SCC extensions
@section SCC extensions

@defmac STARPU_USE_SCC
This macro is defined when StarPU has been installed with SCC
support. It should be used in your code to detect the availability of
SCC.
@end defmac

@deftypefun int starpu_scc_register_kernel({starpu_scc_func_symbol_t *}@var{symbol}, {const char *}@var{func_name})
Initiate a lookup on each SCC device to find the adress of the function
named FUNC_NAME, store them in the global array kernels and return
the index in the array through SYMBOL.
@end deftypefun

@deftypefun starpu_scc_kernel_t starpu_scc_get_kernel(starpu_scc_func_symbol_t @var{symbol})
If success, return the pointer to the function defined by SYMBOL on the
device linked to the called device. This can for instance be used in a
@code{starpu_scc_func_t} implementation.
@end deftypefun

@node Miscellaneous helpers
@section Miscellaneous helpers

@deftypefun int starpu_data_cpy (starpu_data_handle_t @var{dst_handle}, starpu_data_handle_t @var{src_handle}, int @var{asynchronous}, void (*@var{callback_func})(void*), void *@var{callback_arg})
Copy the content of the @var{src_handle} into the @var{dst_handle} handle.
The @var{asynchronous} parameter indicates whether the function should
block or not. In the case of an asynchronous call, it is possible to
synchronize with the termination of this operation either by the means of
implicit dependencies (if enabled) or by calling
@code{starpu_task_wait_for_all()}. If @var{callback_func} is not @code{NULL},
this callback function is executed after the handle has been copied, and it is
given the @var{callback_arg} pointer as argument.
@end deftypefun

@deftypefun void starpu_execute_on_each_worker (void (*@var{func})(void *), void *@var{arg}, uint32_t @var{where})
This function executes the given function on a subset of workers.
When calling this method, the offloaded function specified by the first argument is
executed by every StarPU worker that may execute the function.
The second argument is passed to the offloaded function.
The last argument specifies on which types of processing units the function
should be executed. Similarly to the @var{where} field of the
@code{struct starpu_codelet} structure, it is possible to specify that the function
should be executed on every CUDA device and every CPU by passing
@code{STARPU_CPU|STARPU_CUDA}.
This function blocks until the function has been executed on every appropriate
processing units, so that it may not be called from a callback function for
instance.
@end deftypefun

@node FXT Support
@section FXT Support

@deftypefun void starpu_fxt_start_profiling (void)
Start recording the trace. The trace is by default started from
@code{starpu_init()} call, but can be paused by using
@code{starpu_fxt_stop_profiling}, in which case
@code{starpu_fxt_start_profiling} should be called to specify when to resume
recording events.
@end deftypefun

@deftypefun void starpu_fxt_stop_profiling (void)
Stop recording the trace. The trace is by default stopped at
@code{starpu_shutdown()} call. @code{starpu_fxt_stop_profiling} can however be
used to stop it earlier. @code{starpu_fxt_start_profiling} can then be called to
start recording it again, etc.
@end deftypefun

@node FFT Support
@section FFT Support

@deftypefun {void *} starpufft_malloc (size_t @var{n})
Allocates memory for @var{n} bytes. This is preferred over @code{malloc}, since
it allocates pinned memory, which allows overlapped transfers.
@end deftypefun

@deftypefun {void *} starpufft_free (void *@var{p})
Release memory previously allocated.
@end deftypefun

@deftypefun {struct starpufft_plan *} starpufft_plan_dft_1d (int @var{n}, int @var{sign}, unsigned @var{flags})
Initializes a plan for 1D FFT of size @var{n}. @var{sign} can be
@code{STARPUFFT_FORWARD} or @code{STARPUFFT_INVERSE}. @var{flags} must be 0.
@end deftypefun

@deftypefun {struct starpufft_plan *} starpufft_plan_dft_2d (int @var{n}, int @var{m}, int @var{sign}, unsigned @var{flags})
Initializes a plan for 2D FFT of size (@var{n}, @var{m}). @var{sign} can be
@code{STARPUFFT_FORWARD} or @code{STARPUFFT_INVERSE}. @var{flags} must be 0.
@end deftypefun

@deftypefun {struct starpu_task *} starpufft_start (starpufft_plan @var{p}, void *@var{in}, void *@var{out})
Start an FFT previously planned as @var{p}, using @var{in} and @var{out} as
input and output. This only submits the task and does not wait for it.
The application should call @code{starpufft_cleanup} to unregister the data.
@end deftypefun

@deftypefun {struct starpu_task *} starpufft_start_handle (starpufft_plan @var{p}, starpu_data_handle_t @var{in}, starpu_data_handle_t @var{out})
Start an FFT previously planned as @var{p}, using data handles @var{in} and
@var{out} as input and output (assumed to be vectors of elements of the expected
types). This only submits the task and does not wait for it.
@end deftypefun

@deftypefun void starpufft_execute (starpufft_plan @var{p}, void *@var{in}, void *@var{out})
Execute an FFT previously planned as @var{p}, using @var{in} and @var{out} as
input and output. This submits and waits for the task.
@end deftypefun

@deftypefun void starpufft_execute_handle (starpufft_plan @var{p}, starpu_data_handle_t @var{in}, starpu_data_handle_t @var{out})
Execute an FFT previously planned as @var{p}, using data handles @var{in} and
@var{out} as input and output (assumed to be vectors of elements of the expected
types). This submits and waits for the task.
@end deftypefun

@deftypefun void starpufft_cleanup (starpufft_plan @var{p})
Releases data for plan @var{p}, in the @code{starpufft_start} case.
@end deftypefun

@deftypefun void starpufft_destroy_plan (starpufft_plan @var{p})
Destroys plan @var{p}, i.e. release all CPU (fftw) and GPU (cufft) resources.
@end deftypefun

@node MPI
@section MPI

@menu
* Initialisation::
* Communication::
* Communication Cache::
* MPI Insert Task::
* Collective Operations::
@end menu

@node Initialisation
@subsection Initialisation

@deftypefun int starpu_mpi_init (int *@var{argc}, char ***@var{argv}, int initialize_mpi)
Initializes the starpumpi library. @code{initialize_mpi} indicates if
MPI should be initialized or not by StarPU. If the value is not @code{0},
MPI will be initialized by calling @code{MPI_Init_Thread(argc, argv,
MPI_THREAD_SERIALIZED, ...)}.
@end deftypefun

@deftypefun int starpu_mpi_initialize (void)
This function has been made deprecated. One should use instead the
function @code{starpu_mpi_init()} defined above.
This function does not call @code{MPI_Init}, it should be called beforehand.
@end deftypefun

@deftypefun int starpu_mpi_initialize_extended (int *@var{rank}, int *@var{world_size})
This function has been made deprecated. One should use instead the
function @code{starpu_mpi_init()} defined above.
MPI will be initialized by starpumpi by calling @code{MPI_Init_Thread(argc, argv,
MPI_THREAD_SERIALIZED, ...)}.
@end deftypefun

@deftypefun int starpu_mpi_shutdown (void)
Cleans the starpumpi library. This must be called between calling
@code{starpu_mpi} functions and @code{starpu_shutdown()}.
@code{MPI_Finalize()} will be called if StarPU-MPI has been initialized
by @code{starpu_mpi_init()}.
@end deftypefun

@deftypefun void starpu_mpi_comm_amounts_retrieve (size_t *@var{comm_amounts})
Retrieve the current amount of communications from the current node in
the array @code{comm_amounts} which must have a size greater or equal
to the world size. Communications statistics must be enabled
(@pxref{STARPU_COMM_STATS}).
@end deftypefun

@deftypefun void starpu_mpi_set_communication_tag (int @var{tag})
@anchor{starpu_mpi_set_communication_tag}
Tell StarPU-MPI which MPI tag to use for all its communications.
@end deftypefun

@deftypefun int starpu_mpi_get_communication_tag (void)
@anchor{starpu_mpi_get_communication_tag}
Returns the MPI tag which will be used for all StarPU-MPI communications.
@end deftypefun

@node Communication
@subsection Communication

@deftypefun int starpu_mpi_send (starpu_data_handle_t @var{data_handle}, int @var{dest}, int @var{mpi_tag}, MPI_Comm @var{comm})
Performs a standard-mode, blocking send of @var{data_handle} to the
node @var{dest} using the message tag @code{mpi_tag} within the
communicator @var{comm}.
@end deftypefun

@deftypefun int starpu_mpi_recv (starpu_data_handle_t @var{data_handle}, int @var{source}, int @var{mpi_tag}, MPI_Comm @var{comm}, MPI_Status *@var{status})
Performs a standard-mode, blocking receive in @var{data_handle} from the
node @var{source} using the message tag @code{mpi_tag} within the
communicator @var{comm}.
@end deftypefun

@deftypefun int starpu_mpi_isend (starpu_data_handle_t @var{data_handle}, starpu_mpi_req *@var{req}, int @var{dest}, int @var{mpi_tag}, MPI_Comm @var{comm})
Posts a standard-mode, non blocking send of @var{data_handle} to the
node @var{dest} using the message tag @code{mpi_tag} within the
communicator @var{comm}. After the call, the pointer to the request
@var{req} can be used to test or to wait for the completion of the communication.
@end deftypefun

@deftypefun int starpu_mpi_irecv (starpu_data_handle_t @var{data_handle}, starpu_mpi_req *@var{req}, int @var{source}, int @var{mpi_tag}, MPI_Comm @var{comm})
Posts a nonblocking receive in @var{data_handle} from the
node @var{source} using the message tag @code{mpi_tag} within the
communicator @var{comm}. After the call, the pointer to the request
@var{req} can be used to test or to wait for the completion of the communication.
@end deftypefun

@deftypefun int starpu_mpi_isend_detached (starpu_data_handle_t @var{data_handle}, int @var{dest}, int @var{mpi_tag}, MPI_Comm @var{comm}, void (*@var{callback})(void *), void *@var{arg})
Posts a standard-mode, non blocking send of @var{data_handle} to the
node @var{dest} using the message tag @code{mpi_tag} within the
communicator @var{comm}. On completion, the @var{callback} function is
called with the argument @var{arg}. Similarly to the pthread detached
functionality, when a detached communication completes, its resources
are automatically released back to the system, there is no need to
test or to wait for the completion of the request.
@end deftypefun

@deftypefun int starpu_mpi_irecv_detached (starpu_data_handle_t @var{data_handle}, int @var{source}, int @var{mpi_tag}, MPI_Comm @var{comm}, void (*@var{callback})(void *), void *@var{arg})
Posts a nonblocking receive in @var{data_handle} from the
node @var{source} using the message tag @code{mpi_tag} within the
communicator @var{comm}. On completion, the @var{callback} function is
called with the argument @var{arg}. Similarly to the pthread detached
functionality, when a detached communication completes, its resources
are automatically released back to the system, there is no need to
test or to wait for the completion of the request.
@end deftypefun

@deftypefun int starpu_mpi_wait (starpu_mpi_req *@var{req}, MPI_Status *@var{status})
Returns when the operation identified by request @var{req} is complete.
@end deftypefun

@deftypefun int starpu_mpi_test (starpu_mpi_req *@var{req}, int *@var{flag}, MPI_Status *@var{status})
If the operation identified by @var{req} is complete, set @var{flag}
to 1. The @var{status} object is set to contain information on the
completed operation.
@end deftypefun

@deftypefun int starpu_mpi_barrier (MPI_Comm @var{comm})
Blocks the caller until all group members of the communicator
@var{comm} have called it.
@end deftypefun

@deftypefun int starpu_mpi_isend_detached_unlock_tag (starpu_data_handle_t @var{data_handle}, int @var{dest}, int @var{mpi_tag}, MPI_Comm @var{comm}, starpu_tag_t @var{tag})
Posts a standard-mode, non blocking send of @var{data_handle} to the
node @var{dest} using the message tag @code{mpi_tag} within the
communicator @var{comm}. On completion, @var{tag} is unlocked.
@end deftypefun

@deftypefun int starpu_mpi_irecv_detached_unlock_tag (starpu_data_handle_t @var{data_handle}, int @var{source}, int @var{mpi_tag}, MPI_Comm @var{comm}, starpu_tag_t @var{tag})
Posts a nonblocking receive in @var{data_handle} from the
node @var{source} using the message tag @code{mpi_tag} within the
communicator @var{comm}. On completion, @var{tag} is unlocked.
@end deftypefun

@deftypefun int starpu_mpi_isend_array_detached_unlock_tag (unsigned @var{array_size}, starpu_data_handle_t *@var{data_handle}, int *@var{dest}, int *@var{mpi_tag}, MPI_Comm *@var{comm}, starpu_tag_t @var{tag})
Posts @var{array_size} standard-mode, non blocking send. Each post
sends the n-th data of the array @var{data_handle} to the n-th node of
the array @var{dest}
using the n-th message tag of the array @code{mpi_tag} within the n-th
communicator of the array
@var{comm}. On completion of the all the requests, @var{tag} is unlocked.
@end deftypefun

@deftypefun int starpu_mpi_irecv_array_detached_unlock_tag (unsigned @var{array_size}, starpu_data_handle_t *@var{data_handle}, int *@var{source}, int *@var{mpi_tag}, MPI_Comm *@var{comm}, starpu_tag_t @var{tag})
Posts @var{array_size} nonblocking receive. Each post receives in the
n-th data of the array @var{data_handle} from the n-th
node of the array @var{source} using the n-th message tag of the array
@code{mpi_tag} within the n-th communicator of the array @var{comm}.
On completion of the all the requests, @var{tag} is unlocked.
@end deftypefun

@node Communication Cache
@subsection Communication Cache

@deftypefun void starpu_mpi_cache_flush (MPI_Comm @var{comm}, starpu_data_handle_t @var{data_handle})
Clear the send and receive communication cache for the data
@var{data_handle}. The function has to be called synchronously by all
the MPI nodes.
The function does nothing if the cache mechanism is disabled (@pxref{STARPU_MPI_CACHE}).
@end deftypefun

@deftypefun void starpu_mpi_cache_flush_all_data (MPI_Comm @var{comm})
Clear the send and receive communication cache for all data. The
function has to be called synchronously by all the MPI nodes.
The function does nothing if the cache mechanism is disabled (@pxref{STARPU_MPI_CACHE}).
@end deftypefun

@node MPI Insert Task
@subsection MPI Insert Task

@deftypefun int starpu_data_set_tag (starpu_data_handle_t @var{handle}, int @var{tag})
Tell StarPU-MPI which MPI tag to use when exchanging the data.
@end deftypefun

@deftypefun int starpu_data_get_tag (starpu_data_handle_t @var{handle})
Returns the MPI tag to be used when exchanging the data.
@end deftypefun

@deftypefun int starpu_data_set_rank (starpu_data_handle_t @var{handle}, int @var{rank})
Tell StarPU-MPI which MPI node "owns" a given data, that is, the node which will
always keep an up-to-date value, and will by default execute tasks which write
to it.
@end deftypefun

@deftypefun int starpu_data_get_rank (starpu_data_handle_t @var{handle})
Returns the last value set by @code{starpu_data_set_rank}.
@end deftypefun

@deftypefun starpu_data_handle_t starpu_data_get_data_handle_from_tag (int @var{tag})
Returns the data handle associated to the MPI tag, or NULL if there is not.
@end deftypefun

@defmac STARPU_EXECUTE_ON_NODE
this macro is used when calling @code{starpu_mpi_insert_task}, and
must be followed by a integer value which specified the node on which
to execute the codelet.
@end defmac

@defmac STARPU_EXECUTE_ON_DATA
this macro is used when calling @code{starpu_mpi_insert_task}, and
must be followed by a data handle to specify that the node owning the
given data will execute the codelet.
@end defmac

@deftypefun int starpu_mpi_insert_task (MPI_Comm @var{comm}, struct starpu_codelet *@var{codelet}, ...)
Create and submit a task corresponding to @var{codelet} with the following
arguments.  The argument list must be zero-terminated.

The arguments following the codelets are the same types as for the
function @code{starpu_insert_task} defined in @ref{Insert Task
Utility}. The extra argument @code{STARPU_EXECUTE_ON_NODE} followed by an
integer allows to specify the MPI node to execute the codelet. It is also
possible to specify that the node owning a specific data will execute
the codelet, by using @code{STARPU_EXECUTE_ON_DATA} followed by a data
handle.

The internal algorithm is as follows:
@enumerate
@item Find out which MPI node is going to execute the codelet.
      @enumerate
      @item If there is only one node owning data in W mode, it will
      be selected;
      @item If there is several nodes owning data in W node, the one
      selected will be the one having the least data in R mode so as
      to minimize the amount of data to be transfered;
      @item The argument @code{STARPU_EXECUTE_ON_NODE} followed by an
      integer can be used to specify the node;
      @item The argument @code{STARPU_EXECUTE_ON_DATA} followed by a
      data handle can be used to specify that the node owing the given
      data will execute the codelet.
      @end enumerate
@item Send and receive data as requested. Nodes owning data which need to be
read by the task are sending them to the MPI node which will execute it. The
latter receives them.
@item Execute the codelet. This is done by the MPI node selected in the
1st step of the algorithm.
@item If several MPI nodes own data to be written to, send written
data back to their owners.
@end enumerate

The algorithm also includes a communication cache mechanism that
allows not to send data twice to the same MPI node, unless the data
has been modified. The cache can be disabled
(@pxref{STARPU_MPI_CACHE}).
@c todo parler plus du cache

@end deftypefun

@deftypefun void starpu_mpi_get_data_on_node (MPI_Comm @var{comm}, starpu_data_handle_t @var{data_handle}, int @var{node})
Transfer data @var{data_handle} to MPI node @var{node}, sending it from its
owner if needed. At least the target node and the owner have to call the
function.
@end deftypefun

@deftypefun void starpu_mpi_get_data_on_node_detached (MPI_Comm @var{comm}, starpu_data_handle_t @var{data_handle}, int @var{node}, {void (*}@var{callback})(void*), {void *}@var{arg})
Transfer data @var{data_handle} to MPI node @var{node}, sending it from its
owner if needed. At least the target node and the owner have to call the
function. On reception, the @var{callback} function is called with the
argument @var{arg}.
@end deftypefun

@node Collective Operations
@subsection Collective Operations

@deftypefun void starpu_mpi_redux_data (MPI_Comm @var{comm}, starpu_data_handle_t @var{data_handle})
Perform a reduction on the given data. All nodes send the data to its
owner node which will perform a reduction.
@end deftypefun

@deftypefun int starpu_mpi_scatter_detached (starpu_data_handle_t *@var{data_handles}, int @var{count}, int @var{root}, MPI_Comm @var{comm}, {void (*}@var{scallback})(void *), {void *}@var{sarg}, {void (*}@var{rcallback})(void *), {void *}@var{rarg})
Scatter data among processes of the communicator based on the ownership of
the data. For each data of the array @var{data_handles}, the
process @var{root} sends the data to the process owning this data.
Processes receiving data must have valid data handles to receive them.
On completion of the collective communication, the @var{scallback} function is
called with the argument @var{sarg} on the process @var{root}, the @var{rcallback} function is
called with the argument @var{rarg} on any other process.
@end deftypefun

@deftypefun int starpu_mpi_gather_detached (starpu_data_handle_t *@var{data_handles}, int @var{count}, int @var{root}, MPI_Comm @var{comm}, {void (*}@var{scallback})(void *), {void *}@var{sarg}, {void (*}@var{rcallback})(void *), {void *}@var{rarg})
Gather data from the different processes of the communicator onto the
process @var{root}. Each process owning data handle in the array
@var{data_handles} will send them to the process @var{root}. The
process @var{root} must have valid data handles to receive the data.
On completion of the collective communication, the @var{rcallback} function is
called with the argument @var{rarg} on the process @var{root}, the @var{scallback} function is
called with the argument @var{sarg} on any other process.
@end deftypefun

@node Task Bundles
@section Task Bundles

@deftp {Data Type} {starpu_task_bundle_t}
Opaque structure describing a list of tasks that should be scheduled
on the same worker whenever it's possible. It must be considered as a
hint given to the scheduler as there is no guarantee that they will be
executed on the same worker.
@end deftp

@deftypefun void starpu_task_bundle_create ({starpu_task_bundle_t *}@var{bundle})
Factory function creating and initializing @var{bundle}, when the call returns, memory needed is allocated and @var{bundle} is ready to use.
@end deftypefun

@deftypefun int starpu_task_bundle_insert (starpu_task_bundle_t @var{bundle}, {struct starpu_task *}@var{task})
Insert @var{task} in @var{bundle}. Until @var{task} is removed from @var{bundle} its expected length and data transfer time will be considered along those of the other tasks of @var{bundle}.
This function mustn't be called if @var{bundle} is already closed and/or @var{task} is already submitted.
@end deftypefun

@deftypefun int starpu_task_bundle_remove (starpu_task_bundle_t @var{bundle}, {struct starpu_task *}@var{task})
Remove @var{task} from @var{bundle}.
Of course @var{task} must have been previously inserted @var{bundle}.
This function mustn't be called if @var{bundle} is already closed and/or @var{task} is already submitted. Doing so would result in undefined behaviour.
@end deftypefun

@deftypefun void starpu_task_bundle_close (starpu_task_bundle_t @var{bundle})
Inform the runtime that the user won't modify @var{bundle} anymore, it means no more inserting or removing task. Thus the runtime can destroy it when possible.
@end deftypefun

@deftypefun double starpu_task_bundle_expected_length (starpu_task_bundle_t @var{bundle}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned @var{nimpl})
Return the expected duration of the entire task bundle in µs.
@end deftypefun

@deftypefun double starpu_task_bundle_expected_power (starpu_task_bundle_t @var{bundle}, enum starpu_perfmodel_archtype @var{arch}, unsigned @var{nimpl})
Return the expected power consumption of the entire task bundle in J.
@end deftypefun

@deftypefun double starpu_task_bundle_expected_data_transfer_time (starpu_task_bundle_t @var{bundle}, unsigned @var{memory_node})
Return the time (in µs) expected to transfer all data used within the bundle.
@end deftypefun

@node Task Lists
@section Task Lists

@deftp {Data Type} {struct starpu_task_list}
Stores a double-chained list of tasks
@end deftp

@deftypefun void starpu_task_list_init ({struct starpu_task_list *}@var{list})
Initialize a list structure
@end deftypefun

@deftypefun void starpu_task_list_push_front ({struct starpu_task_list *}@var{list}, {struct starpu_task *}@var{task})
Push a task at the front of a list
@end deftypefun

@deftypefun void starpu_task_list_push_back ({struct starpu_task_list *}@var{list}, {struct starpu_task *}@var{task})
Push a task at the back of a list
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_front ({struct starpu_task_list *}@var{list})
Get the front of the list (without removing it)
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_back ({struct starpu_task_list *}@var{list})
Get the back of the list (without removing it)
@end deftypefun

@deftypefun int starpu_task_list_empty ({struct starpu_task_list *}@var{list})
Test if a list is empty
@end deftypefun

@deftypefun void starpu_task_list_erase ({struct starpu_task_list *}@var{list}, {struct starpu_task *}@var{task})
Remove an element from the list
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_pop_front ({struct starpu_task_list *}@var{list})
Remove the element at the front of the list
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_pop_back ({struct starpu_task_list *}@var{list})
Remove the element at the back of the list
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_begin ({struct starpu_task_list *}@var{list})
Get the first task of the list.
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_end ({struct starpu_task_list *}@var{list})
Get the end of the list.
@end deftypefun

@deftypefun {struct starpu_task *} starpu_task_list_next ({struct starpu_task *}@var{task})
Get the next task of the list. This is not erase-safe.
@end deftypefun

@node Using Parallel Tasks
@section Using Parallel Tasks

These are used by parallel tasks:

@deftypefun int starpu_combined_worker_get_size (void)
Return the size of the current combined worker, i.e. the total number of cpus
running the same task in the case of SPMD parallel tasks, or the total number
of threads that the task is allowed to start in the case of FORKJOIN parallel
tasks.
@end deftypefun

@deftypefun int starpu_combined_worker_get_rank (void)
Return the rank of the current thread within the combined worker. Can only be
used in FORKJOIN parallel tasks, to know which part of the task to work on.
@end deftypefun

Most of these are used for schedulers which support parallel tasks.

@deftypefun unsigned starpu_combined_worker_get_count (void)
Return the number of different combined workers.
@end deftypefun

@deftypefun int starpu_combined_worker_get_id (void)
Return the identifier of the current combined worker.
@end deftypefun

@deftypefun int starpu_combined_worker_assign_workerid (int @var{nworkers}, int @var{workerid_array}[])
Register a new combined worker and get its identifier
@end deftypefun

@deftypefun int starpu_combined_worker_get_description (int @var{workerid}, {int *}@var{worker_size}, {int **}@var{combined_workerid})
Get the description of a combined worker
@end deftypefun

@deftypefun int starpu_combined_worker_can_execute_task (unsigned @var{workerid}, {struct starpu_task *}@var{task}, unsigned @var{nimpl})
Variant of starpu_worker_can_execute_task compatible with combined workers
@end deftypefun

@deftypefun void starpu_parallel_task_barrier_init ({struct starpu_task* }@var{task}, int @var{workerid})
Initialise the barrier for the parallel task, and dispatch the task
between the different combined workers
@end deftypefun

@deftp {Data Type} {struct starpu_machine_topology}
@table @asis
@item @code{unsigned nworkers}
Total number of workers.

@item @code{unsigned ncombinedworkers}
Total number of combined workers.

@item @code{hwloc_topology_t hwtopology}
Topology as detected by hwloc.

To maintain ABI compatibility when hwloc is not available, the field
is replaced with @code{void *dummy}

@item @code{unsigned nhwcpus}
Total number of CPUs, as detected by the topology code. May be different from
the actual number of CPU workers.

@item @code{unsigned nhwcudagpus}
Total number of CUDA devices, as detected. May be different from the actual
number of CUDA workers.

@item @code{unsigned nhwopenclgpus}
Total number of OpenCL devices, as detected. May be different from the actual
number of OpenCL workers.

@item @code{unsigned nhscc}
Total number of SCC cores, as detected. May be different from the actual
number of core workers.

@item @code{unsigned ncpus}
Actual number of CPU workers used by StarPU.

@item @code{unsigned ncudagpus}
Actual number of CUDA workers used by StarPU.

@item @code{unsigned nopenclgpus}
Actual number of OpenCL workers used by StarPU.

@item @code{unsigned nsccdevices}
Actual number of SCC workers used by StarPU.

@item @code{unsigned workers_bindid[STARPU_NMAXWORKERS]}
Indicates the successive cpu identifier that should be used to bind the
workers. It is either filled according to the user's explicit
parameters (from starpu_conf) or according to the STARPU_WORKERS_CPUID env.
variable. Otherwise, a round-robin policy is used to distributed the workers
over the cpus.

@item @code{unsigned workers_cuda_gpuid[STARPU_NMAXWORKERS]}
Indicates the successive CUDA identifier that should be used by the CUDA
driver.  It is either filled according to the user's explicit parameters (from
starpu_conf) or according to the STARPU_WORKERS_CUDAID env. variable. Otherwise,
they are taken in ID order.

@item @code{unsigned workers_opencl_gpuid[STARPU_NMAXWORKERS]}
Indicates the successive OpenCL identifier that should be used by the OpenCL
driver.  It is either filled according to the user's explicit parameters (from
starpu_conf) or according to the STARPU_WORKERS_OPENCLID env. variable. Otherwise,
they are taken in ID order.

@item @code{unsigned workers_mic_deviceid[STARPU_NMAXWORKERS]}
Indicates the successive MIC devices that should be used by the MIC
driver.  It is either filled according to the user's explicit parameters (from
starpu_conf) or according to the STARPU_WORKERS_MICID env. variable. Otherwise,
they are taken in ID order.

@item @code{unsigned workers_scc_deviceid[STARPU_NMAXWORKERS]}
Indicates the successive SCC devices that should be used by the SCC
driver.  It is either filled according to the user's explicit parameters (from
starpu_conf) or according to the STARPU_WORKERS_SCCID env. variable. Otherwise,
they are taken in ID order.

@end table
@end deftp

@node Scheduling Contexts
@section Scheduling Contexts

StarPU permits on one hand grouping workers in combined workers in order to execute a parallel task and on the other hand grouping tasks in bundles that will be executed by a single specified worker.
In contrast when we group workers in scheduling contexts we submit starpu tasks to them and we schedule them with the policy assigned to the context.
Scheduling contexts can be created, deleted and modified dynamically.

@deftypefun unsigned starpu_sched_ctx_create (const char *@var{policy_name}, int *@var{workerids_ctx}, int @var{nworkers_ctx}, const char *@var{sched_ctx_name})
This function creates a scheduling context which uses the scheduling policy indicated in the first argument and assigns the workers indicated in the second argument to execute the tasks submitted to it.
The return value represents the identifier of the context that has just been created. It will be further used to indicate the context the tasks will be submitted to. The return value should be at most @code{STARPU_NMAX_SCHED_CTXS}.
@end deftypefun

@deftypefun void starpu_sched_ctx_delete (unsigned @var{sched_ctx_id})
Delete scheduling context @var{sched_ctx_id} and transfer remaining workers to the inheritor scheduling context.
@end deftypefun

@deftypefun void starpu_sched_ctx_add_workers ({int *}@var{workerids_ctx}, int @var{nworkers_ctx}, unsigned @var{sched_ctx_id})
This function adds dynamically the workers indicated in the first argument to the context indicated in the last argument. The last argument cannot be greater than  @code{STARPU_NMAX_SCHED_CTXS}.
@end deftypefun

@deftypefun void starpu_sched_ctx_remove_workers ({int *}@var{workerids_ctx}, int @var{nworkers_ctx}, unsigned @var{sched_ctx_id})
This function removes the workers indicated in the first argument from the context indicated in the last argument. The last argument cannot be greater than  @code{STARPU_NMAX_SCHED_CTXS}.
@end deftypefun

A scheduling context manages a collection of workers that can be memorized using different data structures. Thus, a generic structure is available in order to simplify the choice of its type.
Only the list data structure is available but further data structures(like tree) implementations are foreseen.

@deftp {Data Type} {struct starpu_worker_collection}
@table @asis
@item @code{void *workerids}
The workerids managed by the collection
@item @code{unsigned nworkers}
The number of workerids
@item @code{pthread_key_t cursor_key} (optional)
The cursor needed to iterate the collection (depending on the data structure)
@item @code{int type}
The type of structure (currently STARPU_WORKER_LIST is the only one available)
@item @code{unsigned (*has_next)(struct starpu_worker_collection *workers)}
Checks if there is a next worker
@item @code{int (*get_next)(struct starpu_worker_collection *workers)}
Gets the next worker
@item @code{int (*add)(struct starpu_worker_collection *workers, int worker)}
Adds a worker to the collection
@item @code{int (*remove)(struct starpu_worker_collection *workers, int worker)}
Removes a worker from the collection
@item @code{void* (*init)(struct starpu_worker_collection *workers)}
Initialize the collection
@item @code{void (*deinit)(struct starpu_worker_collection *workers)}
Deinitialize the colection
@item @code{void (*init_cursor)(struct starpu_worker_collection *workers)} (optional)
Initialize the cursor if there is one
@item @code{void (*deinit_cursor)(struct starpu_worker_collection *workers)} (optional)
Deinitialize the cursor if there is one

@end table
@end deftp

@deftypefun struct starpu_worker_collection* starpu_sched_ctx_create_worker_collection (unsigned @var{sched_ctx_id}, int @var{type})
Create a worker collection of the type indicated by the last parameter for the context specified through the first parameter.
@end deftypefun

@deftypefun void starpu_sched_ctx_delete_worker_collection (unsigned @var{sched_ctx_id})
Delete the worker collection of the specified scheduling context
@end deftypefun

@deftypefun struct starpu_worker_collection* starpu_sched_ctx_get_worker_collection (unsigned @var{sched_ctx_id})
Return the worker collection managed by the indicated context
@end deftypefun

@deftypefun void starpu_sched_ctx_set_context (unsigned *@var{sched_ctx_id})
Set the scheduling context the subsequent tasks will be submitted to
@end deftypefun

@deftypefun unsigned starpu_sched_ctx_get_context (void)
Return the scheduling context the tasks are currently submitted to
@end deftypefun

@deftypefun unsigned starpu_sched_ctx_get_nworkers (unsigned @var{sched_ctx_id})
Return the number of workers managed by the specified contexts
(Usually needed to verify if it manages any workers or if it should be blocked)
@end deftypefun

@deftypefun unsigned starpu_sched_ctx_get_nshared_workers (unsigned @var{sched_ctx_id}, unsigned @var{sched_ctx_id2})
Return the number of workers shared by two contexts
@end deftypefun

@deftypefun int starpu_sched_ctx_set_min_priority (unsigned @var{sched_ctx_id}, int @var{min_prio})
Defines the minimum task priority level supported by the scheduling
policy of the given scheduler context. The
default minimum priority level is the same as the default priority level which
is 0 by convention.  The application may access that value by calling the
@code{starpu_sched_ctx_get_min_priority} function. This function should only be
called from the initialization method of the scheduling policy, and should not
be used directly from the application.
@end deftypefun

@deftypefun int starpu_sched_ctx_set_max_priority (unsigned @var{sched_ctx_id}, int @var{max_prio})
Defines the maximum priority level supported by the scheduling policy of the given scheduler context. The
default maximum priority level is 1.  The application may access that value by
calling the @code{starpu_sched_ctx_get_max_priority} function. This function should
only be called from the initialization method of the scheduling policy, and
should not be used directly from the application.
@end deftypefun

@deftypefun int starpu_sched_ctx_get_min_priority (unsigned @var{sched_ctx_id})
Returns the current minimum priority level supported by the
scheduling policy of the given scheduler context.
@end deftypefun

@deftypefun int starpu_sched_ctx_get_max_priority (unsigned @var{sched_ctx_id})
Returns the current maximum priority level supported by the
scheduling policy of the given scheduler context.
@end deftypefun

@node Scheduling Policy
@section Scheduling Policy

TODO

While StarPU comes with a variety of scheduling policies (@pxref{Task
scheduling policy}), it may sometimes be desirable to implement custom
policies to address specific problems.  The API described below allows
users to write their own scheduling policy.

@deftp {Data Type} {struct starpu_sched_policy}
This structure contains all the methods that implement a scheduling policy.  An
application may specify which scheduling strategy in the @code{sched_policy}
field of the @code{starpu_conf} structure passed to the @code{starpu_init}
function. The different fields are:

@table @asis
@item @code{void (*init_sched)(unsigned sched_ctx_id)}
Initialize the scheduling policy.

@item @code{void (*deinit_sched)(unsigned sched_ctx_id)}
Cleanup the scheduling policy.

@item @code{int (*push_task)(struct starpu_task *)}
Insert a task into the scheduler.

@item @code{void (*push_task_notify)(struct starpu_task *, int workerid)}
Notify the scheduler that a task was pushed on a given worker. This method is
called when a task that was explicitely assigned to a worker becomes ready and
is about to be executed by the worker. This method therefore permits to keep
the state of of the scheduler coherent even when StarPU bypasses the scheduling
strategy.

@item @code{struct starpu_task *(*pop_task)(unsigned sched_ctx_id)} (optional)
Get a task from the scheduler. The mutex associated to the worker is already
taken when this method is called. If this method is defined as @code{NULL}, the
worker will only execute tasks from its local queue. In this case, the
@code{push_task} method should use the @code{starpu_push_local_task} method to
assign tasks to the different workers.

@item @code{struct starpu_task *(*pop_every_task)(unsigned sched_ctx_id)}
Remove all available tasks from the scheduler (tasks are chained by the means
of the prev and next fields of the starpu_task structure). The mutex associated
to the worker is already taken when this method is called. This is currently
not used.

@item @code{void (*pre_exec_hook)(struct starpu_task *)} (optional)
This method is called every time a task is starting.

@item @code{void (*post_exec_hook)(struct starpu_task *)} (optional)
This method is called every time a task has been executed.

@item @code{void (*add_workers)(unsigned sched_ctx_id, int *workerids, unsigned nworkers)}
Initialize scheduling structures corresponding to each worker used by the policy.

@item @code{void (*remove_workers)(unsigned sched_ctx_id, int *workerids, unsigned nworkers)}
Deinitialize scheduling structures corresponding to each worker used by the policy.

@item @code{const char *policy_name} (optional)
Name of the policy.

@item @code{const char *policy_description} (optional)
Description of the policy.
@end table
@end deftp

@deftypefun {struct starpu_sched_policy **} starpu_sched_get_predefined_policies ()
Return an NULL-terminated array of all the predefined scheduling policies.
@end deftypefun

@deftypefun void starpu_sched_ctx_set_policy_data (unsigned @var{sched_ctx_id}, {void *} @var{policy_data})
Each scheduling policy uses some specific data (queues, variables, additional condition variables).
It is memorize through a local structure. This function assigns it to a scheduling context.
@end deftypefun

@deftypefun void* starpu_sched_ctx_get_policy_data (unsigned @var{sched_ctx_id})
Returns the policy data previously assigned to a context
@end deftypefun

@deftypefun int starpu_sched_set_min_priority (int @var{min_prio})
Defines the minimum task priority level supported by the scheduling policy. The
default minimum priority level is the same as the default priority level which
is 0 by convention.  The application may access that value by calling the
@code{starpu_sched_get_min_priority} function. This function should only be
called from the initialization method of the scheduling policy, and should not
be used directly from the application.
@end deftypefun

@deftypefun int starpu_sched_set_max_priority (int @var{max_prio})
Defines the maximum priority level supported by the scheduling policy. The
default maximum priority level is 1.  The application may access that value by
calling the @code{starpu_sched_get_max_priority} function. This function should
only be called from the initialization method of the scheduling policy, and
should not be used directly from the application.
@end deftypefun

@deftypefun int starpu_sched_get_min_priority (void)
Returns the current minimum priority level supported by the
scheduling policy
@end deftypefun

@deftypefun int starpu_sched_get_max_priority (void)
Returns the current maximum priority level supported by the
scheduling policy
@end deftypefun

@deftypefun int starpu_push_local_task (int @var{workerid}, {struct starpu_task} *@var{task}, int @var{back})
The scheduling policy may put tasks directly into a worker's local queue so
that it is not always necessary to create its own queue when the local queue
is sufficient. If @var{back} not null, @var{task} is put at the back of the queue
where the worker will pop tasks first. Setting @var{back} to 0 therefore ensures
a FIFO ordering.
@end deftypefun

@deftypefun int starpu_push_task_end ({struct starpu_task} *@var{task})
This function must be called by a scheduler to notify that the given
task has just been pushed.
@end deftypefun

@deftypefun int starpu_worker_can_execute_task (unsigned @var{workerid}, {struct starpu_task *}@var{task}, unsigned {nimpl})
Check if the worker specified by workerid can execute the codelet. Schedulers need to call it before assigning a task to a worker, otherwise the task may fail to execute.
@end deftypefun

@deftypefun double starpu_timing_now (void)
Return the current date in µs
@end deftypefun

@deftypefun uint32_t starpu_task_footprint ({struct starpu_perfmodel *}@var{model}, {struct starpu_task *} @var{task}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned @var{nimpl})
Returns the footprint for a given task
@end deftypefun

@deftypefun double starpu_task_expected_length ({struct starpu_task *}@var{task}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned @var{nimpl})
Returns expected task duration in µs
@end deftypefun

@deftypefun double starpu_worker_get_relative_speedup ({enum starpu_perfmodel_archtype} @var{perf_archtype})
Returns an estimated speedup factor relative to CPU speed
@end deftypefun

@deftypefun double starpu_task_expected_data_transfer_time (unsigned @var{memory_node}, {struct starpu_task *}@var{task})
Returns expected data transfer time in µs
@end deftypefun

@deftypefun double starpu_data_expected_transfer_time (starpu_data_handle_t @var{handle}, unsigned @var{memory_node}, {enum starpu_data_access_mode} @var{mode})
Predict the transfer time (in µs) to move a handle to a memory node
@end deftypefun

@deftypefun double starpu_task_expected_power ({struct starpu_task *}@var{task}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned @var{nimpl})
Returns expected power consumption in J
@end deftypefun

@deftypefun double starpu_task_expected_conversion_time ({struct starpu_task *}@var{task}, {enum starpu_perfmodel_archtype} @var{arch}, unsigned {nimpl})
Returns expected conversion time in ms (multiformat interface only)
@end deftypefun

@node Running drivers
@section Running drivers

@deftypefun int starpu_driver_run ({struct starpu_driver *}@var{d})
Initialize the given driver, run it until it receives a request to terminate,
deinitialize it and return 0 on success. It returns -EINVAL if @code{d->type}
is not a valid StarPU device type (STARPU_CPU_WORKER, STARPU_CUDA_WORKER or
STARPU_OPENCL_WORKER). This is the same as using the following
functions: calling @code{starpu_driver_init()}, then calling
@code{starpu_driver_run_once()} in a loop, and eventually
@code{starpu_driver_deinit()}.
@end deftypefun

@deftypefun int starpu_driver_init (struct starpu_driver *@var{d})
Initialize the given driver. Returns 0 on success, -EINVAL if
@code{d->type} is not a valid StarPU device type (STARPU_CPU_WORKER,
STARPU_CUDA_WORKER or STARPU_OPENCL_WORKER).
@end deftypefun

@deftypefun int starpu_driver_run_once (struct starpu_driver *@var{d})
Run the driver once, then returns 0 on success, -EINVAL if
@code{d->type} is not a valid StarPU device type (STARPU_CPU_WORKER,
STARPU_CUDA_WORKER or STARPU_OPENCL_WORKER).
@end deftypefun

@deftypefun int starpu_driver_deinit (struct starpu_driver *@var{d})
Deinitialize the given driver. Returns 0 on success, -EINVAL if
@code{d->type} is not a valid StarPU device type (STARPU_CPU_WORKER,
STARPU_CUDA_WORKER or STARPU_OPENCL_WORKER).
@end deftypefun

@deftypefun void starpu_drivers_request_termination (void)
Notify all running drivers they should terminate.
@end deftypefun

@node Expert mode
@section Expert mode

@deftypefun void starpu_wake_all_blocked_workers (void)
Wake all the workers, so they can inspect data requests and task submissions
again.
@end deftypefun

@deftypefun int starpu_progression_hook_register (unsigned (*@var{func})(void *arg), void *@var{arg})
Register a progression hook, to be called when workers are idle.
@end deftypefun

@deftypefun void starpu_progression_hook_deregister (int @var{hook_id})
Unregister a given progression hook.
@end deftypefun