/* StarPU --- Runtime system for heterogeneous multicore architectures.
 *
 * Copyright (C) 2009-2021  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
 *
 * StarPU is free software; you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published by
 * the Free Software Foundation; either version 2.1 of the License, or (at
 * your option) any later version.
 *
 * StarPU is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
 *
 * See the GNU Lesser General Public License in COPYING.LGPL for more details.
 */

/*! \page TasksInStarPU Tasks In StarPU

\section TaskGranularity Task Granularity

Like any other runtime, StarPU has some overhead to manage tasks. Since
it does smart scheduling and data management, this overhead is not always
neglectable. The order of magnitude of the overhead is typically a couple of
microseconds, which is actually quite smaller than the CUDA overhead itself. The
amount of work that a task should do should thus be somewhat
bigger, to make sure that the overhead becomes neglectible. The offline
performance feedback can provide a measure of task length, which should thus be
checked if bad performance are observed. To get a grasp at the scalability
possibility according to task size, one can run
<c>tests/microbenchs/tasks_size_overhead.sh</c> which draws curves of the
speedup of independent tasks of very small sizes.
To determine what task size your application is actually using, one can use
<c>starpu_fxt_data_trace</c>, see \ref DataTrace .

The choice of scheduler also has impact over the overhead: for instance, the
 scheduler <c>dmda</c> takes time to make a decision, while <c>eager</c> does
not. <c>tasks_size_overhead.sh</c> can again be used to get a grasp at how much
impact that has on the target machine.

\section TaskSubmission Task Submission

To let StarPU make online optimizations, tasks should be submitted
asynchronously as much as possible. Ideally, all tasks should be
submitted, and mere calls to starpu_task_wait_for_all() or
starpu_data_unregister() be done to wait for
termination. StarPU will then be able to rework the whole schedule, overlap
computation with communication, manage accelerator local memory usage, etc.

\section TaskPriorities Task Priorities

By default, StarPU will consider the tasks in the order they are submitted by
the application. If the application programmer knows that some tasks should
be performed in priority (for instance because their output is needed by many
other tasks and may thus be a bottleneck if not executed early
enough), the field starpu_task::priority should be set to provide the
priority information to StarPU.

\section TaskDependencies Task Dependencies

\subsection SequentialConsistency Sequential Consistency

By default, task dependencies are inferred from data dependency (sequential
coherency) by StarPU. The application can however disable sequential coherency
for some data, and dependencies can be specifically expressed.

Setting (or unsetting) sequential consistency can be done at the data
level by calling starpu_data_set_sequential_consistency_flag() for a
specific data or starpu_data_set_default_sequential_consistency_flag()
for all datas.

Setting (or unsetting) sequential consistency can also be done at task
level by setting the field starpu_task::sequential_consistency to \c 0.

Sequential consistency can also be set (or unset) for each handle of a
specific task, this is done by using the field
starpu_task::handles_sequential_consistency. When set, its value
should be a array with the number of elements being the number of
handles for the task, each element of the array being the sequential
consistency for the \c i-th handle of the task. The field can easily be
set when calling starpu_task_insert() with the flag
::STARPU_HANDLES_SEQUENTIAL_CONSISTENCY

\code{.c}
char *seq_consistency = malloc(cl.nbuffers * sizeof(char));
seq_consistency[0] = 1;
seq_consistency[1] = 1;
seq_consistency[2] = 0;
ret = starpu_task_insert(&cl,
	STARPU_RW, handleA, STARPU_RW, handleB, STARPU_RW, handleC,
	STARPU_HANDLES_SEQUENTIAL_CONSISTENCY, seq_consistency,
	0);
free(seq_consistency);
\endcode

The internal algorithm used by StarPU to set up implicit dependency is
as follows:
\code{.c}
if (sequential_consistency(task) == 1)
    for(i=0 ; i<STARPU_TASK_GET_NBUFFERS(task) ; i++)
      if (sequential_consistency(i-th data, task) == 1)
        if (sequential_consistency(i-th data) == 1)
           create_implicit_dependency(...)
\endcode

\subsection TasksAndTagsDependencies Tasks And Tags Dependencies

One can explicitely set dependencies between tasks using
starpu_task_declare_deps() or starpu_task_declare_deps_array(). Dependencies between tasks can be
expressed through tags associated to a tag with the field
starpu_task::tag_id and using the function starpu_tag_declare_deps()
or starpu_tag_declare_deps_array().

The termination of a task can be delayed through the function
starpu_task_end_dep_add() which specifies the number of calls to the function
starpu_task_end_dep_release() needed to trigger the task termination. One can
also use starpu_task_declare_end_deps() or starpu_task_declare_end_deps_array()
to delay the termination of a task until the termination of other tasks.

\section SettingManyDataHandlesForATask Setting Many Data Handles For a Task

The maximum number of data a task can manage is fixed by the macro
\ref STARPU_NMAXBUFS which has a default value which can be changed
through the \c configure option \ref enable-maxbuffers "--enable-maxbuffers".

However, it is possible to define tasks managing more data by using
the field starpu_task::dyn_handles when defining a task and the field
starpu_codelet::dyn_modes when defining the corresponding codelet.

\code{.c}
enum starpu_data_access_mode modes[STARPU_NMAXBUFS+1] =
{
	STARPU_R, STARPU_R, ...
};

struct starpu_codelet dummy_big_cl =
{
	.cuda_funcs = { dummy_big_kernel },
	.opencl_funcs = { dummy_big_kernel },
	.cpu_funcs = { dummy_big_kernel },
	.cpu_funcs_name = { "dummy_big_kernel" },
	.nbuffers = STARPU_NMAXBUFS+1,
	.dyn_modes = modes
};

task = starpu_task_create();
task->cl = &dummy_big_cl;
task->dyn_handles = malloc(task->cl->nbuffers * sizeof(starpu_data_handle_t));
for(i=0 ; i<task->cl->nbuffers ; i++)
{
	task->dyn_handles[i] = handle;
}
starpu_task_submit(task);
\endcode

\code{.c}
starpu_data_handle_t *handles = malloc(dummy_big_cl.nbuffers * sizeof(starpu_data_handle_t));
for(i=0 ; i<dummy_big_cl.nbuffers ; i++)
{
	handles[i] = handle;
}
starpu_task_insert(&dummy_big_cl,
         	  STARPU_VALUE, &dummy_big_cl.nbuffers, sizeof(dummy_big_cl.nbuffers),
		  STARPU_DATA_ARRAY, handles, dummy_big_cl.nbuffers,
		  0);
\endcode

The whole code for this complex data interface is available in the
file <c>examples/basic_examples/dynamic_handles.c</c>.

\section SettingVariableDataHandlesForATask Setting a Variable Number Of Data Handles For a Task

Normally, the number of data handles given to a task is set with
starpu_codelet::nbuffers. This field can however be set to
\ref STARPU_VARIABLE_NBUFFERS, in which case starpu_task::nbuffers
must be set, and starpu_task::modes (or starpu_task::dyn_modes,
see \ref SettingManyDataHandlesForATask) should be used to specify the modes for
the handles.

\section UsingMultipleImplementationsOfACodelet Using Multiple Implementations Of A Codelet

One may want to write multiple implementations of a codelet for a single type of
device and let StarPU choose which one to run. As an example, we will show how
to use SSE to scale a vector. The codelet can be written as follows:

\code{.c}
#include <xmmintrin.h>

void scal_sse_func(void *buffers[], void *cl_arg)
{
    float *vector = (float *) STARPU_VECTOR_GET_PTR(buffers[0]);
    unsigned int n = STARPU_VECTOR_GET_NX(buffers[0]);
    unsigned int n_iterations = n/4;
    if (n % 4 != 0)
        n_iterations++;

    __m128 *VECTOR = (__m128*) vector;
    __m128 factor __attribute__((aligned(16)));
    factor = _mm_set1_ps(*(float *) cl_arg);

    unsigned int i;
    for (i = 0; i < n_iterations; i++)
        VECTOR[i] = _mm_mul_ps(factor, VECTOR[i]);
}
\endcode

\code{.c}
struct starpu_codelet cl =
{
    .cpu_funcs = { scal_cpu_func, scal_sse_func },
    .cpu_funcs_name = { "scal_cpu_func", "scal_sse_func" },
    .nbuffers = 1,
    .modes = { STARPU_RW }
};
\endcode

Schedulers which are multi-implementation aware (only <c>dmda</c> and
<c>pheft</c> for now) will use the performance models of all the
provided implementations, and pick the one which seems to be the fastest.

\section EnablingImplementationAccordingToCapabilities Enabling Implementation According To Capabilities

Some implementations may not run on some devices. For instance, some CUDA
devices do not support double floating point precision, and thus the kernel
execution would just fail; or the device may not have enough shared memory for
the implementation being used. The field starpu_codelet::can_execute
permits to express this. For instance:

\code{.c}
static int can_execute(unsigned workerid, struct starpu_task *task, unsigned nimpl)
{
  const struct cudaDeviceProp *props;
  if (starpu_worker_get_type(workerid) == STARPU_CPU_WORKER)
    return 1;
  /* Cuda device */
  props = starpu_cuda_get_device_properties(workerid);
  if (props->major >= 2 || props->minor >= 3)
    /* At least compute capability 1.3, supports doubles */
    return 1;
  /* Old card, does not support doubles */
  return 0;
}

struct starpu_codelet cl =
{
    .can_execute = can_execute,
    .cpu_funcs = { cpu_func },
    .cpu_funcs_name = { "cpu_func" },
    .cuda_funcs = { gpu_func }
    .nbuffers = 1,
    .modes = { STARPU_RW }
};
\endcode

This can be essential e.g. when running on a machine which mixes various models
of CUDA devices, to take benefit from the new models without crashing on old models.

Note: the function starpu_codelet::can_execute is called by the
scheduler each time it tries to match a task with a worker, and should
thus be very fast. The function starpu_cuda_get_device_properties()
provides a quick access to CUDA properties of CUDA devices to achieve
such efficiency.

Another example is to compile CUDA code for various compute capabilities,
resulting with two CUDA functions, e.g. <c>scal_gpu_13</c> for compute capability
1.3, and <c>scal_gpu_20</c> for compute capability 2.0. Both functions can be
provided to StarPU by using starpu_codelet::cuda_funcs, and
starpu_codelet::can_execute can then be used to rule out the
<c>scal_gpu_20</c> variant on a CUDA device which will not be able to execute it:

\code{.c}
static int can_execute(unsigned workerid, struct starpu_task *task, unsigned nimpl)
{
  const struct cudaDeviceProp *props;
  if (starpu_worker_get_type(workerid) == STARPU_CPU_WORKER)
    return 1;
  /* Cuda device */
  if (nimpl == 0)
    /* Trying to execute the 1.3 capability variant, we assume it is ok in all cases.  */
    return 1;
  /* Trying to execute the 2.0 capability variant, check that the card can do it.  */
  props = starpu_cuda_get_device_properties(workerid);
  if (props->major >= 2 || props->minor >= 0)
    /* At least compute capability 2.0, can run it */
    return 1;
  /* Old card, does not support 2.0, will not be able to execute the 2.0 variant.  */
  return 0;
}

struct starpu_codelet cl =
{
    .can_execute = can_execute,
    .cpu_funcs = { cpu_func },
    .cpu_funcs_name = { "cpu_func" },
    .cuda_funcs = { scal_gpu_13, scal_gpu_20 },
    .nbuffers = 1,
    .modes = { STARPU_RW }
};
\endcode

Another example is having specialized implementations for some given common
sizes, for instance here we have a specialized implementation for 1024x1024
matrices:

\code{.c}
static int can_execute(unsigned workerid, struct starpu_task *task, unsigned nimpl)
{
  const struct cudaDeviceProp *props;
  if (starpu_worker_get_type(workerid) == STARPU_CPU_WORKER)
    return 1;
  /* Cuda device */
  switch (nimpl)
  {
    case 0:
      /* Trying to execute the generic capability variant.  */
      return 1;
    case 1:
    {
      /* Trying to execute the size == 1024 specific variant.  */
      struct starpu_matrix_interface *interface = starpu_data_get_interface_on_node(task->handles[0]);
      return STARPU_MATRIX_GET_NX(interface) == 1024 && STARPU_MATRIX_GET_NY(interface == 1024);
    }
  }
}

struct starpu_codelet cl =
{
    .can_execute = can_execute,
    .cpu_funcs = { cpu_func },
    .cpu_funcs_name = { "cpu_func" },
    .cuda_funcs = { potrf_gpu_generic, potrf_gpu_1024 },
    .nbuffers = 1,
    .modes = { STARPU_RW }
};
\endcode

Note that the most generic variant should be provided first, as some schedulers are
not able to try the different variants.

\section InsertTaskUtility Insert Task Utility

StarPU provides the wrapper function starpu_task_insert() to ease
the creation and submission of tasks.

Here the implementation of a codelet:

\code{.c}
void func_cpu(void *descr[], void *_args)
{
        int *x0 = (int *)STARPU_VARIABLE_GET_PTR(descr[0]);
        float *x1 = (float *)STARPU_VARIABLE_GET_PTR(descr[1]);
        int ifactor;
        float ffactor;

        starpu_codelet_unpack_args(_args, &ifactor, &ffactor);
        *x0 = *x0 * ifactor;
        *x1 = *x1 * ffactor;
}

struct starpu_codelet mycodelet =
{
        .cpu_funcs = { func_cpu },
        .cpu_funcs_name = { "func_cpu" },
        .nbuffers = 2,
        .modes = { STARPU_RW, STARPU_RW }
};
\endcode

And the call to the function starpu_task_insert():

\code{.c}
starpu_task_insert(&mycodelet,
                   STARPU_VALUE, &ifactor, sizeof(ifactor),
                   STARPU_VALUE, &ffactor, sizeof(ffactor),
                   STARPU_RW, data_handles[0],
		   STARPU_RW, data_handles[1],
                   0);
\endcode

The call to starpu_task_insert() is equivalent to the following
code:

\code{.c}
struct starpu_task *task = starpu_task_create();
task->cl = &mycodelet;
task->handles[0] = data_handles[0];
task->handles[1] = data_handles[1];
char *arg_buffer;
size_t arg_buffer_size;
starpu_codelet_pack_args(&arg_buffer, &arg_buffer_size,
                    STARPU_VALUE, &ifactor, sizeof(ifactor),
                    STARPU_VALUE, &ffactor, sizeof(ffactor),
                    0);
task->cl_arg = arg_buffer;
task->cl_arg_size = arg_buffer_size;
int ret = starpu_task_submit(task);
\endcode

Here a similar call using ::STARPU_DATA_ARRAY.

\code{.c}
starpu_task_insert(&mycodelet,
                   STARPU_DATA_ARRAY, data_handles, 2,
                   STARPU_VALUE, &ifactor, sizeof(ifactor),
                   STARPU_VALUE, &ffactor, sizeof(ffactor),
                   0);
\endcode

If some part of the task insertion depends on the value of some computation,
the macro ::STARPU_DATA_ACQUIRE_CB can be very convenient. For
instance, assuming that the index variable <c>i</c> was registered as handle
<c>A_handle[i]</c>:

\code{.c}
/* Compute which portion we will work on, e.g. pivot */
starpu_task_insert(&which_index, STARPU_W, i_handle, 0);

/* And submit the corresponding task */
STARPU_DATA_ACQUIRE_CB(i_handle, STARPU_R,
                       starpu_task_insert(&work, STARPU_RW, A_handle[i], 0));
\endcode

The macro ::STARPU_DATA_ACQUIRE_CB submits an asynchronous request for
acquiring data <c>i</c> for the main application, and will execute the code
given as third parameter when it is acquired. In other words, as soon as the
value of <c>i</c> computed by the codelet <c>which_index</c> can be read, the
portion of code passed as third parameter of ::STARPU_DATA_ACQUIRE_CB will
be executed, and is allowed to read from <c>i</c> to use it e.g. as an
index. Note that this macro is only avaible when compiling StarPU with
the compiler <c>gcc</c>.

StarPU also provides a utility function starpu_codelet_unpack_args() to retrieve the ::STARPU_VALUE arguments passed to the task. There is several ways of calling this function starpu_codelet_unpack_args().

\code{.c}
void func_cpu(void *descr[], void *_args)
{
        int ifactor;
        float ffactor;

        starpu_codelet_unpack_args(_args, &ifactor, &ffactor);
}
\endcode

\code{.c}
void func_cpu(void *descr[], void *_args)
{
        int ifactor;
        float ffactor;

        starpu_codelet_unpack_args(_args, &ifactor, 0);
        starpu_codelet_unpack_args(_args, &ifactor, &ffactor);
}
\endcode

\code{.c}
void func_cpu(void *descr[], void *_args)
{
        int ifactor;
        float ffactor;
	char buffer[100];

        starpu_codelet_unpack_args_and_copyleft(_args, buffer, 100, &ifactor, 0);
        starpu_codelet_unpack_args(buffer, &ffactor);
}
\endcode

\section GettingTaskChildren Getting Task Children

It may be interesting to get the list of tasks which depend on a given task,
notably when using implicit dependencies, since this list is computed by StarPU.
starpu_task_get_task_succs() provides it. For instance:

\code{.c}
struct starpu_task *tasks[4];
ret = starpu_task_get_task_succs(task, sizeof(tasks)/sizeof(*tasks), tasks);
\endcode

\section ParallelTasks Parallel Tasks

StarPU can leverage existing parallel computation libraries by the means of
parallel tasks. A parallel task is a task which is run by a set of CPUs
(called a parallel or combined worker) at the same time, by using an existing
parallel CPU implementation of the computation to be achieved. This can also be
useful to improve the load balance between slow CPUs and fast GPUs: since CPUs
work collectively on a single task, the completion time of tasks on CPUs become
comparable to the completion time on GPUs, thus relieving from granularity
discrepancy concerns. <c>hwloc</c> support needs to be enabled to get
good performance, otherwise StarPU will not know how to better group
cores.

Two modes of execution exist to accomodate with existing usages.

\subsection Fork-modeParallelTasks Fork-mode Parallel Tasks

In the Fork mode, StarPU will call the codelet function on one
of the CPUs of the combined worker. The codelet function can use
starpu_combined_worker_get_size() to get the number of threads it is
allowed to start to achieve the computation. The CPU binding mask for the whole
set of CPUs is already enforced, so that threads created by the function will
inherit the mask, and thus execute where StarPU expected, the OS being in charge
of choosing how to schedule threads on the corresponding CPUs. The application
can also choose to bind threads by hand, using e.g. <c>sched_getaffinity</c> to know
the CPU binding mask that StarPU chose.

For instance, using OpenMP (full source is available in
<c>examples/openmp/vector_scal.c</c>):

\snippet forkmode.c To be included. You should update doxygen if you see this text.

Other examples include for instance calling a BLAS parallel CPU implementation
(see <c>examples/mult/xgemm.c</c>).

\subsection SPMD-modeParallelTasks SPMD-mode Parallel Tasks

In the SPMD mode, StarPU will call the codelet function on
each CPU of the combined worker. The codelet function can use
starpu_combined_worker_get_size() to get the total number of CPUs
involved in the combined worker, and thus the number of calls that are made in
parallel to the function, and starpu_combined_worker_get_rank() to get
the rank of the current CPU within the combined worker. For instance:

\code{.c}
static void func(void *buffers[], void *args)
{
    unsigned i;
    float *factor = _args;
    struct starpu_vector_interface *vector = buffers[0];
    unsigned n = STARPU_VECTOR_GET_NX(vector);
    float *val = (float *)STARPU_VECTOR_GET_PTR(vector);

    /* Compute slice to compute */
    unsigned m = starpu_combined_worker_get_size();
    unsigned j = starpu_combined_worker_get_rank();
    unsigned slice = (n+m-1)/m;

    for (i = j * slice; i < (j+1) * slice && i < n; i++)
        val[i] *= *factor;
}

static struct starpu_codelet cl =
{
    .modes = { STARPU_RW },
    .type = STARPU_SPMD,
    .max_parallelism = INT_MAX,
    .cpu_funcs = { func },
    .cpu_funcs_name = { "func" },
    .nbuffers = 1,
}
\endcode

Of course, this trivial example will not really benefit from parallel task
execution, and was only meant to be simple to understand.  The benefit comes
when the computation to be done is so that threads have to e.g. exchange
intermediate results, or write to the data in a complex but safe way in the same
buffer.

\subsection ParallelTasksPerformance Parallel Tasks Performance

To benefit from parallel tasks, a parallel-task-aware StarPU scheduler has to
be used. When exposed to codelets with a flag ::STARPU_FORKJOIN or
::STARPU_SPMD, the schedulers <c>pheft</c> (parallel-heft) and <c>peager</c>
(parallel eager) will indeed also try to execute tasks with
several CPUs. It will automatically try the various available combined
worker sizes (making several measurements for each worker size) and
thus be able to avoid choosing a large combined worker if the codelet
does not actually scale so much.

This is however for now only proof of concept, and has not really been optimized yet.

\subsection CombinedWorkers Combined Workers

By default, StarPU creates combined workers according to the architecture
structure as detected by <c>hwloc</c>. It means that for each object of the <c>hwloc</c>
topology (NUMA node, socket, cache, ...) a combined worker will be created. If
some nodes of the hierarchy have a big arity (e.g. many cores in a socket
without a hierarchy of shared caches), StarPU will create combined workers of
intermediate sizes. The variable \ref STARPU_SYNTHESIZE_ARITY_COMBINED_WORKER
permits to tune the maximum arity between levels of combined workers.

The combined workers actually produced can be seen in the output of the
tool <c>starpu_machine_display</c> (the environment variable
\ref STARPU_SCHED has to be set to a combined worker-aware scheduler such
as <c>pheft</c> or <c>peager</c>).

\subsection ConcurrentParallelTasks Concurrent Parallel Tasks

Unfortunately, many environments and librairies do not support concurrent
calls.

For instance, most OpenMP implementations (including the main ones) do not
support concurrent <c>pragma omp parallel</c> statements without nesting them in
another <c>pragma omp parallel</c> statement, but StarPU does not yet support
creating its CPU workers by using such pragma.

Other parallel libraries are also not safe when being invoked concurrently
from different threads, due to the use of global variables in their sequential
sections for instance.

The solution is then to use only one combined worker at a time.  This can be
done by setting the field starpu_conf::single_combined_worker to <c>1</c>, or
setting the environment variable \ref STARPU_SINGLE_COMBINED_WORKER
to <c>1</c>. StarPU will then run only 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 try varying combined worker
sizes to look for the most efficient ones.

\subsection SynchronizationTasks Synchronization Tasks

For the application conveniency, it may be useful to define tasks which do not
actually make any computation, but wear for instance dependencies between other
tasks or tags, or to be submitted in callbacks, etc.

The obvious way is of course to make kernel functions empty, but such task will
thus have to wait for a worker to become ready, transfer data, etc.

A much lighter way to define a synchronization task is to set its field starpu_task::cl
to <c>NULL</c>. The task will thus be a mere synchronization point,
without any data access or execution content: as soon as its dependencies become
available, it will terminate, call the callbacks, and release dependencies.

An intermediate solution is to define a codelet with its field
starpu_codelet::where set to \ref STARPU_NOWHERE, for instance:

\code{.c}
struct starpu_codelet cl =
{
	.where = STARPU_NOWHERE,
	.nbuffers = 1,
	.modes = { STARPU_R },
}

task = starpu_task_create();
task->cl = &cl;
task->handles[0] = handle;
starpu_task_submit(task);
\endcode

will create a task which simply waits for the value of <c>handle</c> to be
available for read. This task can then be depended on, etc.

*/