/* StarPU --- Runtime system for heterogeneous multicore architectures. * * Copyright (C) 2009-2021 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria * Copyright (C) 2021 Federal University of Rio Grande do Sul (UFRGS) * * 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. */ #ifndef __STARPU_DATA_H__ #define __STARPU_DATA_H__ #include #ifdef __cplusplus extern "C" { #endif /** @defgroup API_Data_Management Data Management @brief Data management facilities provided by StarPU. We show how to use existing data interfaces in \ref API_Data_Interfaces, but developers can design their own data interfaces if required. @{ */ struct _starpu_data_state; /** StarPU uses ::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 ::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. */ typedef struct _starpu_data_state* starpu_data_handle_t; /** Describe a StarPU data access mode Note: when adding a flag here, update _starpu_detect_implicit_data_deps_with_handle Note: other STARPU_* values in include/starpu_task_util.h */ enum starpu_data_access_mode { STARPU_NONE=0, /**< todo */ STARPU_R=(1<<0), /**< read-only mode */ STARPU_W=(1<<1), /**< write-only mode */ STARPU_RW=(STARPU_R|STARPU_W), /**< read-write mode. Equivalent to ::STARPU_R|::STARPU_W */ STARPU_SCRATCH=(1<<2), /**< 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 ::STARPU_R and ::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 ::STARPU_SCRATCH mode should be registered with node -1 and a NULL pointer, since the value of the provided buffer is simply ignored for now. */ STARPU_REDUX=(1<<3), /**< todo */ STARPU_COMMUTE=(1<<4), /**< ::STARPU_COMMUTE can be passed along ::STARPU_W or ::STARPU_RW to express that StarPU can let tasks commute, which is useful e.g. when bringing a contribution into some data, which can be done in any order (but still require sequential consistency against reads or non-commutative writes). */ STARPU_SSEND=(1<<5), /**< used in starpu_mpi_insert_task() to specify the data has to be sent using a synchronous and non-blocking mode (see starpu_mpi_issend()) */ STARPU_LOCALITY=(1<<6), /**< used to tell the scheduler which data is the most important for the task, and should thus be used to try to group tasks on the same core or cache, etc. For now only the ws and lws schedulers take this flag into account, and only when rebuild with \c USE_LOCALITY flag defined in the src/sched_policies/work_stealing_policy.c source code. */ STARPU_MPI_REDUX=(1<<7), /**< Inter-node reduction only. Codelets contributing to these reductions should be registered with ::STARPU_RW | ::STARPU_COMMUTE access modes. When inserting these tasks through the MPI layer however, the access mode needs to be ::STARPU_MPI_REDUX. */ STARPU_NOPLAN=(1<<8), /**< Disable automatic submission of asynchronous partitioning/unpartitioning */ STARPU_ACCESS_MODE_MAX=(1<<9) /**< The purpose of ::STARPU_ACCESS_MODE_MAX is to be the maximum of this enum. */ }; struct starpu_data_interface_ops; /** Set the name of the data, to be shown in various profiling tools. */ void starpu_data_set_name(starpu_data_handle_t handle, const char *name); /** Set the coordinates of the data, to be shown in various profiling tools. \p dimensions is the size of the \p dims array. This can be for instance the tile coordinates within a big matrix. */ void starpu_data_set_coordinates_array(starpu_data_handle_t handle, unsigned dimensions, int dims[]); /** Set the coordinates of the data, to be shown in various profiling tools. \p dimensions is the number of subsequent \c int parameters. This can be for instance the tile coordinates within a big matrix. */ void starpu_data_set_coordinates(starpu_data_handle_t handle, unsigned dimensions, ...); /** Get the coordinates of the data, as set by a previous call to starpu_data_set_coordinates_array() or starpu_data_set_coordinates() \p dimensions is the size of the \p dims array. This returns the actual number of returned coordinates. */ unsigned starpu_data_get_coordinates_array(starpu_data_handle_t handle, unsigned dimensions, int dims[]); /** Unregister a data \p 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. In case we do not need to update the value of the data in the home node, we can use the function starpu_data_unregister_no_coherency() instead. */ void starpu_data_unregister(starpu_data_handle_t handle); /** Similar to starpu_data_unregister(), except that StarPU does not put back a valid copy into the home node, in the buffer that was initially registered. */ void starpu_data_unregister_no_coherency(starpu_data_handle_t handle); /** Destroy the data \p handle once it is no longer needed by any submitted task. No coherency is provided. This is not safe to call starpu_data_unregister_submit() on a handle that comes from the registration of a non-NULL application home buffer, since the moment when the unregistration will happen is unknown to the application. Only calling starpu_shutdown() allows to be sure that the data was really unregistered. */ void starpu_data_unregister_submit(starpu_data_handle_t handle); /** Destroy all replicates of the data \p handle immediately. After data invalidation, the first access to \p handle must be performed in ::STARPU_W mode. Accessing an invalidated data in ::STARPU_R mode results in undefined behaviour. */ void starpu_data_invalidate(starpu_data_handle_t handle); /** Submit invalidation of the data \p handle after completion of previously submitted tasks. */ void starpu_data_invalidate_submit(starpu_data_handle_t handle); /** Specify that the data \p handle can be discarded without impacting the application. */ void starpu_data_advise_as_important(starpu_data_handle_t handle, unsigned is_important); /** @name Access registered data from the application @{ */ /** This macro can be used to acquire data, but not require it to be available on a given node, only enforce R/W dependencies. This can for instance be used to wait for tasks which produce the data, but without requesting a fetch to the main memory. */ #define STARPU_ACQUIRE_NO_NODE -1 /** Similar to ::STARPU_ACQUIRE_NO_NODE, but will lock the data on all nodes, preventing them from being evicted for instance. This is mostly useful inside StarPU only. */ #define STARPU_ACQUIRE_NO_NODE_LOCK_ALL -2 /** 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 \p handle 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 with \p mode. starpu_data_release() must be called once the application no longer needs to access the piece of data. Note that implicit data dependencies are also enforced by starpu_data_acquire(), i.e. starpu_data_acquire() will wait for all tasks scheduled to work on the data, unless they have been disabled explictly by calling starpu_data_set_default_sequential_consistency_flag() or starpu_data_set_sequential_consistency_flag(). starpu_data_acquire() is a blocking call, so that it cannot be called from tasks or from their callbacks (in that case, starpu_data_acquire() returns -EDEADLK). Upon successful completion, this function returns 0. */ int starpu_data_acquire(starpu_data_handle_t handle, enum starpu_data_access_mode mode); /** Similar to starpu_data_acquire(), except that the data will be available on the given memory node instead of main memory. ::STARPU_ACQUIRE_NO_NODE and ::STARPU_ACQUIRE_NO_NODE_LOCK_ALL can be used instead of an explicit node number. */ int starpu_data_acquire_on_node(starpu_data_handle_t handle, int node, enum starpu_data_access_mode mode); /** Asynchronous equivalent of starpu_data_acquire(). When the data specified in \p handle is available in the access \p mode, the \p callback function is executed. The application may access the requested data during the execution of \p callback. The \p callback function must call starpu_data_release() once the application no longer needs to access the piece of data. Note that implicit data dependencies are also enforced by starpu_data_acquire_cb() in case they are not disabled. Contrary to starpu_data_acquire(), this function is non-blocking and may be called from task callbacks. Upon successful completion, this function returns 0. */ int starpu_data_acquire_cb(starpu_data_handle_t handle, enum starpu_data_access_mode mode, void (*callback)(void *), void *arg); /** Similar to starpu_data_acquire_cb(), except that the data will be available on the given memory node instead of main memory. ::STARPU_ACQUIRE_NO_NODE and ::STARPU_ACQUIRE_NO_NODE_LOCK_ALL can be used instead of an explicit node number. */ int starpu_data_acquire_on_node_cb(starpu_data_handle_t handle, int node, enum starpu_data_access_mode mode, void (*callback)(void *), void *arg); /** Similar to starpu_data_acquire_cb() with the possibility of enabling or disabling data dependencies. When the data specified in \p handle is available in the access \p mode, the \p callback function is executed. The application may access the requested data during the execution of this \p callback. The \p callback function must call starpu_data_release() once the application no longer needs to access the piece of data. Note that implicit data dependencies are also enforced by starpu_data_acquire_cb_sequential_consistency() in case they are not disabled specifically for the given \p handle or by the parameter \p sequential_consistency. Similarly to starpu_data_acquire_cb(), this function is non-blocking and may be called from task callbacks. Upon successful completion, this function returns 0. */ int starpu_data_acquire_cb_sequential_consistency(starpu_data_handle_t handle, enum starpu_data_access_mode mode, void (*callback)(void *), void *arg, int sequential_consistency); /** Similar to starpu_data_acquire_cb_sequential_consistency(), except that the data will be available on the given memory node instead of main memory. ::STARPU_ACQUIRE_NO_NODE and ::STARPU_ACQUIRE_NO_NODE_LOCK_ALL can be used instead of an explicit node number. */ int starpu_data_acquire_on_node_cb_sequential_consistency(starpu_data_handle_t handle, int node, enum starpu_data_access_mode mode, void (*callback)(void *), void *arg, int sequential_consistency); int starpu_data_acquire_on_node_cb_sequential_consistency_quick(starpu_data_handle_t handle, int node, enum starpu_data_access_mode mode, void (*callback)(void *), void *arg, int sequential_consistency, int quick); /** Similar to starpu_data_acquire_on_node_cb_sequential_consistency(), except that the \e pre_sync_jobid and \e post_sync_jobid parameters can be used to retrieve the jobid of the synchronization tasks. \e pre_sync_jobid happens just before the acquisition, and \e post_sync_jobid happens just after the release. callback_acquired is called when the data is acquired in terms of semantic, but the data is not fetched yet. It is given a pointer to the node, which it can modify if it wishes so. This is a very internal interface, subject to changes, do not use this. */ int starpu_data_acquire_on_node_cb_sequential_consistency_sync_jobids(starpu_data_handle_t handle, int node, enum starpu_data_access_mode mode, void (*callback_acquired)(void *arg, int *node, enum starpu_data_access_mode mode), void (*callback)(void *arg), void *arg, int sequential_consistency, int quick, long *pre_sync_jobid, long *post_sync_jobid, int prio); /** The application can call this function instead of starpu_data_acquire() so as to acquire the data like starpu_data_acquire(), but only if all previously-submitted tasks have completed, in which case starpu_data_acquire_try() returns 0. StarPU will have ensured that the application will get an up-to-date copy of \p handle in main memory located where the data was originally registered. starpu_data_release() must be called once the application no longer needs to access the piece of data. */ int starpu_data_acquire_try(starpu_data_handle_t handle, enum starpu_data_access_mode mode); /** Similar to starpu_data_acquire_try(), except that the data will be available on the given memory node instead of main memory. ::STARPU_ACQUIRE_NO_NODE and ::STARPU_ACQUIRE_NO_NODE_LOCK_ALL can be used instead of an explicit node number. */ int starpu_data_acquire_on_node_try(starpu_data_handle_t handle, int node, enum starpu_data_access_mode mode); #ifdef __GCC__ /** STARPU_DATA_ACQUIRE_CB() is the same as starpu_data_acquire_cb(), except that the code to be executed in a callback is directly provided as a macro parameter, and the data \p 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. */ # define STARPU_DATA_ACQUIRE_CB(handle, mode, code) do \ { \ \ void callback(void *arg) \ { \ code; \ starpu_data_release(handle); \ } \ starpu_data_acquire_cb(handle, mode, callback, NULL); \ } \ while(0) #endif /** Release the piece of data acquired by the application either by starpu_data_acquire() or by starpu_data_acquire_cb(). */ void starpu_data_release(starpu_data_handle_t handle); /** Similar to starpu_data_release(), except that the data was made available on the given memory \p node instead of main memory. The \p node parameter must be exactly the same as the corresponding \c starpu_data_acquire_on_node* call. */ void starpu_data_release_on_node(starpu_data_handle_t handle, int node); /** Partly release the piece of data acquired by the application either by starpu_data_acquire() or by starpu_data_acquire_cb(), switching the acquisition down to \p down_to_mode. For now, only releasing from STARPU_RW or STARPU_W acquisition down to STARPU_R is supported, or down to the same acquisition. STARPU_NONE can also be passed as \p down_to_mode, in which case this is equivalent to calling starpu_data_release(). */ void starpu_data_release_to(starpu_data_handle_t handle, enum starpu_data_access_mode down_to_mode); /** Similar to starpu_data_release_to(), except that the data was made available on the given memory \p node instead of main memory. The \p node parameter must be exactly the same as the corresponding \c starpu_data_acquire_on_node* call. */ void starpu_data_release_to_on_node(starpu_data_handle_t handle, enum starpu_data_access_mode down_to_mode, int node); /** @} */ /** This is an arbiter, which implements an advanced but centralized management of concurrent data accesses, see \ref ConcurrentDataAccess for the details. */ typedef struct starpu_arbiter *starpu_arbiter_t; /** Create a data access arbiter, see \ref ConcurrentDataAccess for the details */ starpu_arbiter_t starpu_arbiter_create(void) STARPU_ATTRIBUTE_MALLOC; /** Make access to \p handle managed by \p arbiter */ void starpu_data_assign_arbiter(starpu_data_handle_t handle, starpu_arbiter_t arbiter); /** Destroy the \p arbiter . This must only be called after all data assigned to it have been unregistered. */ void starpu_arbiter_destroy(starpu_arbiter_t arbiter); /** Explicitly ask StarPU to allocate room for a piece of data on the specified memory \p node. */ int starpu_data_request_allocation(starpu_data_handle_t handle, unsigned node); /** Prefetch levels Data requests are ordered by priorities, but also by prefetching level, between data that a task wants now, and data that we will probably want "soon". */ enum starpu_is_prefetch { /** A task really needs it now! */ STARPU_FETCH = 0, /** A task will need it soon */ STARPU_TASK_PREFETCH = 1, /** It is a good idea to have it asap */ STARPU_PREFETCH = 2, /** Get this here when you have time to */ STARPU_IDLEFETCH = 3, STARPU_NFETCH }; /** Issue a fetch request for the data \p handle to \p node, i.e. requests that the data be replicated to the given node as soon as possible, so that it is available there for tasks. If \p async is 0, the call will block until the transfer is achieved, else the call will return immediately, after having just queued the request. In the latter case, the request will asynchronously wait for the completion of any task writing on the data. */ int starpu_data_fetch_on_node(starpu_data_handle_t handle, unsigned node, unsigned async); /** Issue a prefetch request for the data \p handle to \p node, i.e. requests that the data be replicated to \p node when there is room for it, so that it is available there for tasks. If \p async is 0, the call will block until the transfer is achieved, else the call will return immediately, after having just queued the request. In the latter case, the request will asynchronously wait for the completion of any task writing on the data. */ int starpu_data_prefetch_on_node(starpu_data_handle_t handle, unsigned node, unsigned async); int starpu_data_prefetch_on_node_prio(starpu_data_handle_t handle, unsigned node, unsigned async, int prio); /** Issue an idle prefetch request for the data \p handle to \p node, i.e. requests that the data be replicated to \p node, so that it is available there for tasks, but only when the bus is really idle. If \p async is 0, the call will block until the transfer is achieved, else the call will return immediately, after having just queued the request. In the latter case, the request will asynchronously wait for the completion of any task writing on the data. */ int starpu_data_idle_prefetch_on_node(starpu_data_handle_t handle, unsigned node, unsigned async); int starpu_data_idle_prefetch_on_node_prio(starpu_data_handle_t handle, unsigned node, unsigned async, int prio); /** Check whether a valid copy of \p handle is currently available on memory node \p node (or a transfer request for getting so is ongoing). */ unsigned starpu_data_is_on_node(starpu_data_handle_t handle, unsigned node); /** Advise StarPU that \p handle will not be used in the close future, and is thus a good candidate for eviction from GPUs. StarPU will thus write its value back to its home node when the bus is idle, and select this data in priority for eviction when memory gets low. */ void starpu_data_wont_use(starpu_data_handle_t handle); /** Advise StarPU to evict \p handle from the memory node \p node StarPU will thus write its value back to its home node, before evicting it. This may however fail if e.g. some task is still working on it. If the eviction was successful, 0 is returned ; -1 is returned otherwise. */ int starpu_data_evict_from_node(starpu_data_handle_t handle, unsigned node); /** Set the write-through mask of the data \p handle (and its children), 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. When the data is modified, it is automatically transfered into those memory nodes. For instance a 1<<0 write-through mask means that the CUDA workers will commit their changes in main memory (node 0). */ void starpu_data_set_wt_mask(starpu_data_handle_t handle, uint32_t wt_mask); /** @name 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. @{ */ /** Set 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 starpu_data_set_default_sequential_consistency_flag(). */ void starpu_data_set_sequential_consistency_flag(starpu_data_handle_t handle, unsigned flag); /** Get the data consistency mode associated to the data handle \p handle */ unsigned starpu_data_get_sequential_consistency_flag(starpu_data_handle_t handle); /** Return the default sequential consistency flag */ unsigned starpu_data_get_default_sequential_consistency_flag(void); /** 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 function starpu_data_set_sequential_consistency_flag(). */ void starpu_data_set_default_sequential_consistency_flag(unsigned flag); /** @} */ /** Set whether this data should be elligible to be evicted to disk storage (1) or not (0). The default is 1. */ void starpu_data_set_ooc_flag(starpu_data_handle_t handle, unsigned flag); /** Get whether this data was set to be elligible to be evicted to disk storage (1) or not (0). */ unsigned starpu_data_get_ooc_flag(starpu_data_handle_t handle); /** Query the status of \p handle on the specified \p memory_node. \p is_allocated tells whether memory was allocated there for the data. \p is_valid tells whether the actual value is available there. \p is_loading tells whether the actual value is getting loaded there. \p is_requested tells whether the actual value is requested to be loaded there by some fetch/prefetch/idlefetch request. */ void starpu_data_query_status2(starpu_data_handle_t handle, int memory_node, int *is_allocated, int *is_valid, int *is_loading, int *is_requested); /** Same as starpu_data_query_status2(), but without the is_loading parameter. */ void starpu_data_query_status(starpu_data_handle_t handle, int memory_node, int *is_allocated, int *is_valid, int *is_requested); struct starpu_codelet; /** Set the codelets to be used for \p handle when it is accessed in the mode ::STARPU_REDUX. Per-worker buffers will be initialized with the codelet \p init_cl (which has to take one handle with STARPU_W), and reduction between per-worker buffers will be done with the codelet \p redux_cl (which has to take a first accumulation handle with STARPU_RW|STARPU_COMMUTE, and a second contribution handle with STARPU_R). */ void starpu_data_set_reduction_methods(starpu_data_handle_t handle, struct starpu_codelet *redux_cl, struct starpu_codelet *init_cl); struct starpu_data_interface_ops* starpu_data_get_interface_ops(starpu_data_handle_t handle); unsigned starpu_data_test_if_allocated_on_node(starpu_data_handle_t handle, unsigned memory_node); void starpu_memchunk_tidy(unsigned memory_node); /** Set the field \c user_data for the \p handle to \p user_data . It can then be retrieved with starpu_data_get_user_data(). \p user_data can be any application-defined value, for instance a pointer to an object-oriented container for the data. */ void starpu_data_set_user_data(starpu_data_handle_t handle, void* user_data); /** Retrieve the field \c user_data previously set for the \p handle. */ void *starpu_data_get_user_data(starpu_data_handle_t handle); /** Set the field \c sched_data for the \p handle to \p sched_data . It can then be retrieved with starpu_data_get_sched_data(). \p sched_data can be any scheduler-defined value. */ void starpu_data_set_sched_data(starpu_data_handle_t handle, void* sched_data); /** Retrieve the field \c sched_data previously set for the \p handle. */ void *starpu_data_get_sched_data(starpu_data_handle_t handle); /** Check whether data \p handle can be evicted now from node \p node */ int starpu_data_can_evict(starpu_data_handle_t handle, unsigned node, enum starpu_is_prefetch is_prefetch); /** @} */ #ifdef __cplusplus } #endif #endif /* __STARPU_DATA_H__ */