[OMPT] Second chunk of final OMPT 5.0 interface updates
The omp-tools.h file is generated from the OpenMP spec to ensure that the interface is implemented as specified. The other changes are necessary to update the interface implementation to the final version as published in 5.0. The omp-tools.h header was previously called ompt.h, currently a copy under this name is installed for legacy tools. Patch partially perpared by @sconvent Reviewers: AndreyChurbanov, hbae, Hahnfeld Reviewed By: hbae Tags: #openmp, #ompt Differential Revision: https://reviews.llvm.org/D55579 llvm-svn: 351197
This commit is contained in:
@@ -57,7 +57,7 @@ add_custom_command(TARGET omp POST_BUILD
|
||||
)
|
||||
if(${LIBOMP_OMPT_SUPPORT})
|
||||
add_custom_command(TARGET omp POST_BUILD
|
||||
COMMAND ${CMAKE_COMMAND} -E copy ompt.h ${LIBOMP_EXPORTS_CMN_DIR}
|
||||
COMMAND ${CMAKE_COMMAND} -E copy omp-tools.h ${LIBOMP_EXPORTS_CMN_DIR}
|
||||
)
|
||||
endif()
|
||||
if(${LIBOMP_FORTRAN_MODULES})
|
||||
|
||||
@@ -9,11 +9,11 @@
|
||||
#//===----------------------------------------------------------------------===//
|
||||
#
|
||||
|
||||
# Configure omp.h, kmp_config.h and ompt.h if necessary
|
||||
# Configure omp.h, kmp_config.h and omp-tools.h if necessary
|
||||
configure_file(${LIBOMP_INC_DIR}/omp.h.var omp.h @ONLY)
|
||||
configure_file(kmp_config.h.cmake kmp_config.h @ONLY)
|
||||
if(${LIBOMP_OMPT_SUPPORT})
|
||||
configure_file(${LIBOMP_INC_DIR}/ompt.h.var ompt.h @ONLY)
|
||||
configure_file(${LIBOMP_INC_DIR}/omp-tools.h.var omp-tools.h @ONLY)
|
||||
endif()
|
||||
|
||||
# Generate message catalog files: kmp_i18n_id.inc and kmp_i18n_default.inc
|
||||
@@ -319,7 +319,9 @@ install(
|
||||
DESTINATION ${LIBOMP_HEADERS_INSTALL_PATH}
|
||||
)
|
||||
if(${LIBOMP_OMPT_SUPPORT})
|
||||
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/ompt.h DESTINATION ${LIBOMP_HEADERS_INSTALL_PATH})
|
||||
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/omp-tools.h DESTINATION ${LIBOMP_HEADERS_INSTALL_PATH})
|
||||
# install under legacy name ompt.h
|
||||
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/omp-tools.h DESTINATION ${LIBOMP_HEADERS_INSTALL_PATH} RENAME ompt.h)
|
||||
endif()
|
||||
if(${LIBOMP_FORTRAN_MODULES})
|
||||
install(FILES
|
||||
|
||||
1083
openmp/runtime/src/include/50/omp-tools.h.var
Normal file
1083
openmp/runtime/src/include/50/omp-tools.h.var
Normal file
File diff suppressed because it is too large
Load Diff
@@ -1,735 +0,0 @@
|
||||
/*
|
||||
* include/50/ompt.h.var
|
||||
*/
|
||||
|
||||
//===----------------------------------------------------------------------===//
|
||||
//
|
||||
// The LLVM Compiler Infrastructure
|
||||
//
|
||||
// This file is dual licensed under the MIT and the University of Illinois Open
|
||||
// Source Licenses. See LICENSE.txt for details.
|
||||
//
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
#ifndef __OMPT__
|
||||
#define __OMPT__
|
||||
|
||||
/*****************************************************************************
|
||||
* system include files
|
||||
*****************************************************************************/
|
||||
|
||||
#include <stdint.h>
|
||||
#include <stddef.h>
|
||||
|
||||
|
||||
|
||||
/*****************************************************************************
|
||||
* iteration macros
|
||||
*****************************************************************************/
|
||||
|
||||
#define FOREACH_OMPT_INQUIRY_FN(macro) \
|
||||
macro (ompt_enumerate_states) \
|
||||
macro (ompt_enumerate_mutex_impls) \
|
||||
\
|
||||
macro (ompt_set_callback) \
|
||||
macro (ompt_get_callback) \
|
||||
\
|
||||
macro (ompt_get_state) \
|
||||
\
|
||||
macro (ompt_get_parallel_info) \
|
||||
macro (ompt_get_task_info) \
|
||||
macro (ompt_get_task_memory) \
|
||||
macro (ompt_get_thread_data) \
|
||||
macro (ompt_get_unique_id) \
|
||||
macro (ompt_finalize_tool) \
|
||||
\
|
||||
macro(ompt_get_num_procs) \
|
||||
macro(ompt_get_num_places) \
|
||||
macro(ompt_get_place_proc_ids) \
|
||||
macro(ompt_get_place_num) \
|
||||
macro(ompt_get_partition_place_nums) \
|
||||
macro(ompt_get_proc_id) \
|
||||
\
|
||||
macro(ompt_get_target_info) \
|
||||
macro(ompt_get_num_devices)
|
||||
|
||||
#define FOREACH_OMPT_STATE(macro) \
|
||||
\
|
||||
/* first available state */ \
|
||||
macro (ompt_state_undefined, 0x102) /* undefined thread state */ \
|
||||
\
|
||||
/* work states (0..15) */ \
|
||||
macro (ompt_state_work_serial, 0x000) /* working outside parallel */ \
|
||||
macro (ompt_state_work_parallel, 0x001) /* working within parallel */ \
|
||||
macro (ompt_state_work_reduction, 0x002) /* performing a reduction */ \
|
||||
\
|
||||
/* barrier wait states (16..31) */ \
|
||||
macro (ompt_state_wait_barrier, 0x010) /* waiting at a barrier */ \
|
||||
macro (ompt_state_wait_barrier_implicit_parallel, 0x011) \
|
||||
/* implicit barrier at the end of parallel region */\
|
||||
macro (ompt_state_wait_barrier_implicit_workshare, 0x012) \
|
||||
/* implicit barrier at the end of worksharing */ \
|
||||
macro (ompt_state_wait_barrier_implicit, 0x013) /* implicit barrier */ \
|
||||
macro (ompt_state_wait_barrier_explicit, 0x014) /* explicit barrier */ \
|
||||
\
|
||||
/* task wait states (32..63) */ \
|
||||
macro (ompt_state_wait_taskwait, 0x020) /* waiting at a taskwait */ \
|
||||
macro (ompt_state_wait_taskgroup, 0x021) /* waiting at a taskgroup */ \
|
||||
\
|
||||
/* mutex wait states (64..127) */ \
|
||||
macro (ompt_state_wait_mutex, 0x040) \
|
||||
macro (ompt_state_wait_lock, 0x041) /* waiting for lock */ \
|
||||
macro (ompt_state_wait_critical, 0x042) /* waiting for critical */ \
|
||||
macro (ompt_state_wait_atomic, 0x043) /* waiting for atomic */ \
|
||||
macro (ompt_state_wait_ordered, 0x044) /* waiting for ordered */ \
|
||||
\
|
||||
/* target wait states (128..255) */ \
|
||||
macro (ompt_state_wait_target, 0x080) /* waiting for target region */ \
|
||||
macro (ompt_state_wait_target_map, 0x081) /* waiting for target data mapping operation */ \
|
||||
macro (ompt_state_wait_target_update, 0x082) /* waiting for target update operation */ \
|
||||
\
|
||||
/* misc (256..511) */ \
|
||||
macro (ompt_state_idle, 0x100) /* waiting for work */ \
|
||||
macro (ompt_state_overhead, 0x101) /* overhead excluding wait states */ \
|
||||
\
|
||||
/* implementation-specific states (512..) */
|
||||
|
||||
|
||||
#define FOREACH_KMP_MUTEX_IMPL(macro) \
|
||||
macro (ompt_mutex_impl_none, 0) /* unknown implementation */ \
|
||||
macro (kmp_mutex_impl_spin, 1) /* based on spin */ \
|
||||
macro (kmp_mutex_impl_queuing, 2) /* based on some fair policy */ \
|
||||
macro (kmp_mutex_impl_speculative, 3) /* based on HW-supported speculation */
|
||||
|
||||
#define FOREACH_OMPT_EVENT(macro) \
|
||||
\
|
||||
/*--- Mandatory Events ---*/ \
|
||||
macro (ompt_callback_thread_begin, ompt_callback_thread_begin_t, 1) /* thread begin */ \
|
||||
macro (ompt_callback_thread_end, ompt_callback_thread_end_t, 2) /* thread end */ \
|
||||
\
|
||||
macro (ompt_callback_parallel_begin, ompt_callback_parallel_begin_t, 3) /* parallel begin */ \
|
||||
macro (ompt_callback_parallel_end, ompt_callback_parallel_end_t, 4) /* parallel end */ \
|
||||
\
|
||||
macro (ompt_callback_task_create, ompt_callback_task_create_t, 5) /* task begin */ \
|
||||
macro (ompt_callback_task_schedule, ompt_callback_task_schedule_t, 6) /* task schedule */ \
|
||||
macro (ompt_callback_implicit_task, ompt_callback_implicit_task_t, 7) /* implicit task */ \
|
||||
\
|
||||
macro (ompt_callback_target, ompt_callback_target_t, 8) /* target */ \
|
||||
macro (ompt_callback_target_data_op, ompt_callback_target_data_op_t, 9) /* target data op */ \
|
||||
macro (ompt_callback_target_submit, ompt_callback_target_submit_t, 10) /* target submit */ \
|
||||
\
|
||||
macro (ompt_callback_control_tool, ompt_callback_control_tool_t, 11) /* control tool */ \
|
||||
\
|
||||
macro (ompt_callback_device_initialize, ompt_callback_device_initialize_t, 12) /* device initialize */ \
|
||||
macro (ompt_callback_device_finalize, ompt_callback_device_finalize_t, 13) /* device finalize */ \
|
||||
\
|
||||
macro (ompt_callback_device_load, ompt_callback_device_load_t, 14) /* device load */ \
|
||||
macro (ompt_callback_device_unload, ompt_callback_device_unload_t, 15) /* device unload */ \
|
||||
\
|
||||
/* Optional Events */ \
|
||||
macro (ompt_callback_sync_region_wait, ompt_callback_sync_region_t, 16) /* sync region wait begin or end */ \
|
||||
\
|
||||
macro (ompt_callback_mutex_released, ompt_callback_mutex_t, 17) /* mutex released */ \
|
||||
\
|
||||
macro (ompt_callback_task_dependences, ompt_callback_task_dependences_t, 18) /* report task dependences */ \
|
||||
macro (ompt_callback_task_dependence, ompt_callback_task_dependence_t, 19) /* report task dependence */ \
|
||||
\
|
||||
macro (ompt_callback_work, ompt_callback_work_t, 20) /* task at work begin or end */ \
|
||||
\
|
||||
macro (ompt_callback_master, ompt_callback_master_t, 21) /* task at master begin or end */ \
|
||||
\
|
||||
macro (ompt_callback_target_map, ompt_callback_target_map_t, 22) /* target map */ \
|
||||
\
|
||||
macro (ompt_callback_sync_region, ompt_callback_sync_region_t, 23) /* sync region begin or end */ \
|
||||
\
|
||||
macro (ompt_callback_lock_init, ompt_callback_mutex_acquire_t, 24) /* lock init */ \
|
||||
macro (ompt_callback_lock_destroy, ompt_callback_mutex_t, 25) /* lock destroy */ \
|
||||
\
|
||||
macro (ompt_callback_mutex_acquire, ompt_callback_mutex_acquire_t, 26) /* mutex acquire */ \
|
||||
macro (ompt_callback_mutex_acquired, ompt_callback_mutex_t, 27) /* mutex acquired */ \
|
||||
\
|
||||
macro (ompt_callback_nest_lock, ompt_callback_nest_lock_t, 28) /* nest lock */ \
|
||||
\
|
||||
macro (ompt_callback_flush, ompt_callback_flush_t, 29) /* after executing flush */ \
|
||||
\
|
||||
macro (ompt_callback_cancel, ompt_callback_cancel_t, 30) /* cancel innermost binding region */ \
|
||||
\
|
||||
macro (ompt_callback_reduction, ompt_callback_sync_region_t, 31) /* reduction */ \
|
||||
\
|
||||
macro (ompt_callback_dispatch, ompt_callback_dispatch_t, 32) /* dispatch of work */
|
||||
|
||||
|
||||
|
||||
/*****************************************************************************
|
||||
* data types
|
||||
*****************************************************************************/
|
||||
|
||||
/*---------------------
|
||||
* identifiers
|
||||
*---------------------*/
|
||||
|
||||
typedef uint64_t ompt_id_t;
|
||||
#define ompt_id_none 0
|
||||
|
||||
typedef union ompt_data_t {
|
||||
uint64_t value; /* data initialized by runtime to unique id */
|
||||
void *ptr; /* pointer under tool control */
|
||||
} ompt_data_t;
|
||||
|
||||
static const ompt_data_t ompt_data_none = {0};
|
||||
|
||||
typedef uint64_t ompt_wait_id_t;
|
||||
static const ompt_wait_id_t omp_wait_id_none = 0;
|
||||
|
||||
typedef void ompt_device_t;
|
||||
|
||||
|
||||
/*---------------------
|
||||
* dependences types
|
||||
*---------------------*/
|
||||
|
||||
typedef enum ompt_task_dependence_type_t {
|
||||
// a two bit field for the dependence type
|
||||
ompt_task_dependence_type_in = 1,
|
||||
ompt_task_dependence_type_out = 2,
|
||||
ompt_task_dependence_type_inout = 3,
|
||||
ompt_task_dependence_type_mutexinoutset = 4
|
||||
} ompt_task_dependence_type_t;
|
||||
|
||||
typedef struct ompt_task_dependence_t {
|
||||
void *variable_addr;
|
||||
ompt_task_dependence_type_t dependence_type;
|
||||
} ompt_task_dependence_t;
|
||||
|
||||
|
||||
/*****************************************************************************
|
||||
* enumerations for thread states and runtime events
|
||||
*****************************************************************************/
|
||||
|
||||
/*---------------------
|
||||
* runtime states
|
||||
*---------------------*/
|
||||
|
||||
typedef enum {
|
||||
#define ompt_state_macro(state, code) state = code,
|
||||
FOREACH_OMPT_STATE(ompt_state_macro)
|
||||
#undef ompt_state_macro
|
||||
} ompt_state_t;
|
||||
|
||||
typedef enum ompt_frame_flag_t {
|
||||
ompt_frame_runtime = 0x00,
|
||||
ompt_frame_application = 0x01,
|
||||
ompt_frame_cfa = 0x10,
|
||||
ompt_frame_framepointer = 0x20,
|
||||
ompt_frame_stackaddress = 0x30
|
||||
} ompt_frame_flag_t;
|
||||
|
||||
|
||||
/*---------------------
|
||||
* runtime events
|
||||
*---------------------*/
|
||||
|
||||
typedef enum ompt_callbacks_e{
|
||||
#define ompt_event_macro(event, callback, eventid) event = eventid,
|
||||
FOREACH_OMPT_EVENT(ompt_event_macro)
|
||||
#undef ompt_event_macro
|
||||
} ompt_callbacks_t;
|
||||
|
||||
|
||||
/*---------------------
|
||||
* set callback results
|
||||
*---------------------*/
|
||||
typedef enum ompt_set_result_t {
|
||||
ompt_set_error = 0,
|
||||
ompt_set_never = 1,
|
||||
ompt_set_sometimes = 2,
|
||||
ompt_set_sometimes_paired = 3,
|
||||
ompt_set_always = 4
|
||||
} ompt_set_result_t;
|
||||
|
||||
|
||||
/*----------------------
|
||||
* mutex implementations
|
||||
*----------------------*/
|
||||
typedef enum kmp_mutex_impl_t {
|
||||
#define kmp_mutex_impl_macro(impl, code) impl = code,
|
||||
FOREACH_KMP_MUTEX_IMPL(kmp_mutex_impl_macro)
|
||||
#undef kmp_mutex_impl_macro
|
||||
} kmp_mutex_impl_t;
|
||||
|
||||
|
||||
/*****************************************************************************
|
||||
* callback signatures
|
||||
*****************************************************************************/
|
||||
|
||||
/* initialization */
|
||||
typedef void (*ompt_interface_fn_t)(void);
|
||||
|
||||
typedef ompt_interface_fn_t (*ompt_function_lookup_t)(
|
||||
const char * /* entry point to look up */
|
||||
);
|
||||
|
||||
/* threads */
|
||||
typedef enum ompt_thread_t {
|
||||
ompt_thread_initial = 1, // start the enumeration at 1
|
||||
ompt_thread_worker = 2,
|
||||
ompt_thread_other = 3,
|
||||
ompt_thread_unknown = 4
|
||||
} ompt_thread_t;
|
||||
|
||||
typedef struct ompt_frame_t {
|
||||
ompt_data_t exit_frame;
|
||||
ompt_data_t enter_frame;
|
||||
int exit_frame_flags;
|
||||
int enter_frame_flags;
|
||||
} ompt_frame_t;
|
||||
typedef enum ompt_parallel_flag_t {
|
||||
ompt_parallel_invoker_program = 0x00000001, /* program invokes master task */
|
||||
ompt_parallel_invoker_runtime = 0x00000002, /* runtime invokes master task */
|
||||
ompt_parallel_league = 0x40000000,
|
||||
ompt_parallel_team = 0x80000000
|
||||
} ompt_parallel_flag_t;
|
||||
|
||||
typedef void (*ompt_callback_thread_begin_t) (
|
||||
ompt_thread_t thread_type, /* type of thread */
|
||||
ompt_data_t *thread_data /* data of thread */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_thread_end_t) (
|
||||
ompt_data_t *thread_data /* data of thread */
|
||||
);
|
||||
|
||||
typedef void (*ompt_wait_callback_t) (
|
||||
ompt_wait_id_t wait_id /* wait data */
|
||||
);
|
||||
|
||||
/* parallel and workshares */
|
||||
typedef enum ompt_scope_endpoint_t {
|
||||
ompt_scope_begin = 1,
|
||||
ompt_scope_end = 2
|
||||
} ompt_scope_endpoint_t;
|
||||
|
||||
|
||||
/* implicit task */
|
||||
typedef void (*ompt_callback_implicit_task_t) (
|
||||
ompt_scope_endpoint_t endpoint, /* endpoint of implicit task */
|
||||
ompt_data_t *parallel_data, /* data of parallel region */
|
||||
ompt_data_t *task_data, /* data of implicit task */
|
||||
unsigned int actual_parallelism, /* team size */
|
||||
unsigned int index /* thread number of calling thread */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_parallel_begin_t) (
|
||||
ompt_data_t *encountering_task_data, /* data of encountering task */
|
||||
const ompt_frame_t *encountering_task_frame, /* frame data of encountering task */
|
||||
ompt_data_t *parallel_data, /* data of parallel region */
|
||||
unsigned int requested_team_size, /* requested number of threads in team */
|
||||
int flag, /* flag for additional information */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_parallel_end_t) (
|
||||
ompt_data_t *parallel_data, /* data of parallel region */
|
||||
ompt_data_t *encountering_task_data, /* data of encountering task */
|
||||
int flag, /* flag for additional information */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
/* tasks */
|
||||
typedef enum ompt_task_flag_t {
|
||||
ompt_task_initial = 0x1,
|
||||
ompt_task_implicit = 0x2,
|
||||
ompt_task_explicit = 0x4,
|
||||
ompt_task_target = 0x8,
|
||||
ompt_task_undeferred = 0x8000000,
|
||||
ompt_task_untied = 0x10000000,
|
||||
ompt_task_final = 0x20000000,
|
||||
ompt_task_mergeable = 0x40000000,
|
||||
ompt_task_merged = 0x80000000
|
||||
} ompt_task_flag_t;
|
||||
|
||||
typedef enum ompt_task_status_t {
|
||||
ompt_task_complete = 1,
|
||||
ompt_task_yield = 2,
|
||||
ompt_task_cancel = 3,
|
||||
ompt_task_switch = 4
|
||||
} ompt_task_status_t;
|
||||
|
||||
typedef void (*ompt_callback_task_schedule_t) (
|
||||
ompt_data_t *prior_task_data, /* data of prior task */
|
||||
ompt_task_status_t prior_task_status, /* status of prior task */
|
||||
ompt_data_t *next_task_data /* data of next task */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_task_create_t) (
|
||||
ompt_data_t *encountering_task_data, /* data of parent task */
|
||||
const ompt_frame_t *encountering_task_frame, /* frame data for parent task */
|
||||
ompt_data_t *new_task_data, /* data of created task */
|
||||
int flag, /* type of created task */
|
||||
int has_dependences, /* created task has dependences */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
/* task dependences */
|
||||
typedef void (*ompt_callback_task_dependences_t) (
|
||||
ompt_data_t *task_data, /* data of task */
|
||||
const ompt_task_dependence_t *deps, /* dependences of task */
|
||||
int ndeps /* dependences count of task */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_task_dependence_t) (
|
||||
ompt_data_t *src_task_data, /* data of source task */
|
||||
ompt_data_t *sink_task_data /* data of sink task */
|
||||
);
|
||||
|
||||
/* target and device */
|
||||
typedef enum ompt_target_t {
|
||||
ompt_target = 1,
|
||||
ompt_target_enter_data = 2,
|
||||
ompt_target_exit_data = 3,
|
||||
ompt_target_update = 4
|
||||
} ompt_target_t;
|
||||
|
||||
typedef void (*ompt_callback_target_t) (
|
||||
ompt_target_t kind,
|
||||
ompt_scope_endpoint_t endpoint,
|
||||
uint64_t device_num,
|
||||
ompt_data_t *task_data,
|
||||
ompt_id_t target_id,
|
||||
const void *codeptr_ra
|
||||
);
|
||||
|
||||
typedef enum ompt_target_data_op_t {
|
||||
ompt_target_data_alloc = 1,
|
||||
ompt_target_data_transfer_to_dev = 2,
|
||||
ompt_target_data_transfer_from_dev = 3,
|
||||
ompt_target_data_delete = 4
|
||||
} ompt_target_data_op_t;
|
||||
|
||||
typedef void (*ompt_callback_target_data_op_t) (
|
||||
ompt_id_t target_id,
|
||||
ompt_id_t host_op_id,
|
||||
ompt_target_data_op_t optype,
|
||||
void *src_addr,
|
||||
int src_device_num,
|
||||
void *dest_addr,
|
||||
int dest_device_num,
|
||||
size_t bytes,
|
||||
const void *codeptr_ra
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_target_submit_t) (
|
||||
ompt_id_t target_id,
|
||||
ompt_id_t host_op_id,
|
||||
unsigned int requested_num_teams
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_target_map_t) (
|
||||
ompt_id_t target_id,
|
||||
unsigned int nitems,
|
||||
void **host_addr,
|
||||
void **device_addr,
|
||||
size_t *bytes,
|
||||
unsigned int *mapping_flags,
|
||||
const void *codeptr_ra
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_device_initialize_t) (
|
||||
uint64_t device_num,
|
||||
const char *type,
|
||||
ompt_device_t *device,
|
||||
ompt_function_lookup_t lookup,
|
||||
const char *documentation
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_device_finalize_t) (
|
||||
uint64_t device_num
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_device_load_t) (
|
||||
uint64_t device_num,
|
||||
const char * filename,
|
||||
int64_t offset_in_file,
|
||||
void * vma_in_file,
|
||||
size_t bytes,
|
||||
void * host_addr,
|
||||
void * device_addr,
|
||||
uint64_t module_id
|
||||
);
|
||||
|
||||
#define ompt_addr_unknown ((void *) ~0)
|
||||
|
||||
typedef void (*ompt_callback_device_unload_t) (
|
||||
uint64_t device_num,
|
||||
uint64_t module_id
|
||||
);
|
||||
|
||||
/* control_tool */
|
||||
typedef int (*ompt_callback_control_tool_t) (
|
||||
uint64_t command, /* command of control call */
|
||||
uint64_t modifier, /* modifier of control call */
|
||||
void *arg, /* argument of control call */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef enum ompt_mutex_t {
|
||||
ompt_mutex_lock = 1,
|
||||
ompt_mutex_nest_lock = 2,
|
||||
ompt_mutex_critical = 3,
|
||||
ompt_mutex_atomic = 4,
|
||||
ompt_mutex_ordered = 5
|
||||
} ompt_mutex_t;
|
||||
|
||||
typedef void (*ompt_callback_mutex_acquire_t) (
|
||||
ompt_mutex_t kind, /* mutex kind */
|
||||
unsigned int hint, /* mutex hint */
|
||||
unsigned int impl, /* mutex implementation */
|
||||
ompt_wait_id_t wait_id, /* id of object being awaited */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_mutex_t) (
|
||||
ompt_mutex_t kind, /* mutex kind */
|
||||
ompt_wait_id_t wait_id, /* id of object being awaited */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_nest_lock_t) (
|
||||
ompt_scope_endpoint_t endpoint, /* endpoint of nested lock */
|
||||
ompt_wait_id_t wait_id, /* id of object being awaited */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_master_t) (
|
||||
ompt_scope_endpoint_t endpoint, /* endpoint of master region */
|
||||
ompt_data_t *parallel_data, /* data of parallel region */
|
||||
ompt_data_t *task_data, /* data of task */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef enum ompt_work_t {
|
||||
ompt_work_loop = 1,
|
||||
ompt_work_sections = 2,
|
||||
ompt_work_single_executor = 3,
|
||||
ompt_work_single_other = 4,
|
||||
ompt_work_workshare = 5,
|
||||
ompt_work_distribute = 6,
|
||||
ompt_work_taskloop = 7
|
||||
} ompt_work_t;
|
||||
|
||||
typedef void (*ompt_callback_work_t) (
|
||||
ompt_work_t wstype, /* type of work region */
|
||||
ompt_scope_endpoint_t endpoint, /* endpoint of work region */
|
||||
ompt_data_t *parallel_data, /* data of parallel region */
|
||||
ompt_data_t *task_data, /* data of task */
|
||||
uint64_t count, /* quantity of work */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef enum ompt_sync_region_t {
|
||||
ompt_sync_region_barrier = 1,
|
||||
ompt_sync_region_barrier_implicit = 2,
|
||||
ompt_sync_region_barrier_explicit = 3,
|
||||
ompt_sync_region_barrier_implementation = 4,
|
||||
ompt_sync_region_taskwait = 5,
|
||||
ompt_sync_region_taskgroup = 6,
|
||||
ompt_sync_region_reduction = 7
|
||||
} ompt_sync_region_t;
|
||||
|
||||
typedef void (*ompt_callback_sync_region_t) (
|
||||
ompt_sync_region_t kind, /* kind of sync region */
|
||||
ompt_scope_endpoint_t endpoint, /* endpoint of sync region */
|
||||
ompt_data_t *parallel_data, /* data of parallel region */
|
||||
ompt_data_t *task_data, /* data of task */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef enum ompt_cancel_flag_t {
|
||||
ompt_cancel_parallel = 0x01,
|
||||
ompt_cancel_sections = 0x02,
|
||||
ompt_cancel_loop = 0x04,
|
||||
ompt_cancel_taskgroup = 0x08,
|
||||
ompt_cancel_activated = 0x10,
|
||||
ompt_cancel_detected = 0x20,
|
||||
ompt_cancel_discarded_task = 0x40
|
||||
} ompt_cancel_flag_t;
|
||||
|
||||
typedef void (*ompt_callback_cancel_t) (
|
||||
ompt_data_t *task_data, /* data of task */
|
||||
int flags, /* cancel flags */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_flush_t) (
|
||||
ompt_data_t *thread_data, /* data of thread */
|
||||
const void *codeptr_ra /* return address of runtime call */
|
||||
);
|
||||
|
||||
typedef enum ompt_dispatch_t {
|
||||
ompt_dispatch_iteration = 1,
|
||||
ompt_dispatch_section = 2
|
||||
} ompt_dispatch_t;
|
||||
|
||||
typedef void (*ompt_callback_dispatch_t) (
|
||||
ompt_data_t *parallel_data,
|
||||
ompt_data_t *task_data,
|
||||
ompt_dispatch_t kind,
|
||||
ompt_data_t instance
|
||||
);
|
||||
|
||||
/****************************************************************************
|
||||
* ompt API
|
||||
***************************************************************************/
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#define OMPT_API_FNTYPE(fn) fn##_t
|
||||
|
||||
#define OMPT_API_FUNCTION(return_type, fn, args) \
|
||||
typedef return_type (*OMPT_API_FNTYPE(fn)) args
|
||||
|
||||
|
||||
|
||||
/****************************************************************************
|
||||
* INQUIRY FUNCTIONS
|
||||
***************************************************************************/
|
||||
|
||||
/* state */
|
||||
OMPT_API_FUNCTION(ompt_state_t, ompt_get_state, (
|
||||
ompt_wait_id_t *wait_id
|
||||
));
|
||||
|
||||
/* thread */
|
||||
OMPT_API_FUNCTION(ompt_data_t*, ompt_get_thread_data, (void));
|
||||
|
||||
/* parallel region */
|
||||
OMPT_API_FUNCTION(int, ompt_get_parallel_info, (
|
||||
int ancestor_level,
|
||||
ompt_data_t **parallel_data,
|
||||
int *team_size
|
||||
));
|
||||
|
||||
/* task */
|
||||
OMPT_API_FUNCTION(int, ompt_get_task_info, (
|
||||
int ancestor_level,
|
||||
int *type,
|
||||
ompt_data_t **task_data,
|
||||
ompt_frame_t **task_frame,
|
||||
ompt_data_t **parallel_data,
|
||||
int *thread_num
|
||||
));
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_task_memory, (
|
||||
void **addr,
|
||||
size_t *size,
|
||||
int block
|
||||
));
|
||||
|
||||
/* procs */
|
||||
OMPT_API_FUNCTION(int, ompt_get_num_procs, (void));
|
||||
|
||||
/* places */
|
||||
OMPT_API_FUNCTION(int, ompt_get_num_places, (void));
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_place_proc_ids, (
|
||||
int place_num,
|
||||
int ids_size,
|
||||
int *ids
|
||||
));
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_place_num, (void));
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_partition_place_nums, (
|
||||
int place_nums_size,
|
||||
int *place_nums
|
||||
));
|
||||
|
||||
/* proc_id */
|
||||
OMPT_API_FUNCTION(int, ompt_get_proc_id, (void));
|
||||
|
||||
|
||||
/****************************************************************************
|
||||
* INITIALIZATION FUNCTIONS
|
||||
***************************************************************************/
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_initialize, (
|
||||
ompt_function_lookup_t lookup,
|
||||
ompt_data_t *tool_data
|
||||
));
|
||||
|
||||
OMPT_API_FUNCTION(void, ompt_finalize, (
|
||||
ompt_data_t *tool_data
|
||||
));
|
||||
|
||||
typedef struct ompt_start_tool_result_t {
|
||||
ompt_initialize_t initialize;
|
||||
ompt_finalize_t finalize;
|
||||
ompt_data_t tool_data;
|
||||
} ompt_start_tool_result_t;
|
||||
|
||||
/* initialization interface to be defined by tool */
|
||||
#ifdef _WIN32
|
||||
__declspec(dllexport)
|
||||
#endif
|
||||
ompt_start_tool_result_t * ompt_start_tool(
|
||||
unsigned int omp_version,
|
||||
const char * runtime_version
|
||||
);
|
||||
|
||||
typedef void (*ompt_callback_t)(void);
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_set_callback, (
|
||||
ompt_callbacks_t which,
|
||||
ompt_callback_t callback
|
||||
));
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_callback, (
|
||||
ompt_callbacks_t which,
|
||||
ompt_callback_t *callback
|
||||
));
|
||||
|
||||
|
||||
|
||||
/****************************************************************************
|
||||
* MISCELLANEOUS FUNCTIONS
|
||||
***************************************************************************/
|
||||
|
||||
/* state enumeration */
|
||||
OMPT_API_FUNCTION(int, ompt_enumerate_states, (
|
||||
int current_state,
|
||||
int *next_state,
|
||||
const char **next_state_name
|
||||
));
|
||||
|
||||
/* mutex implementation enumeration */
|
||||
OMPT_API_FUNCTION(int, ompt_enumerate_mutex_impls, (
|
||||
int current_impl,
|
||||
int *next_impl,
|
||||
const char **next_impl_name
|
||||
));
|
||||
|
||||
/* get_unique_id */
|
||||
OMPT_API_FUNCTION(uint64_t, ompt_get_unique_id, (void));
|
||||
|
||||
/* finalize tool */
|
||||
OMPT_API_FUNCTION(void, ompt_finalize_tool, (void));
|
||||
|
||||
#ifdef __cplusplus
|
||||
};
|
||||
#endif
|
||||
|
||||
/****************************************************************************
|
||||
* TARGET
|
||||
***************************************************************************/
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_target_info, (
|
||||
uint64_t *device_num,
|
||||
ompt_id_t *target_id,
|
||||
ompt_id_t *host_op_id
|
||||
));
|
||||
|
||||
OMPT_API_FUNCTION(int, ompt_get_num_devices, (void));
|
||||
|
||||
#endif /* __OMPT__ */
|
||||
@@ -1910,7 +1910,7 @@ void __kmp_fork_barrier(int gtid, int tid) {
|
||||
#endif
|
||||
if (!KMP_MASTER_TID(ds_tid) && ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, task_data, 0, ds_tid);
|
||||
ompt_scope_end, NULL, task_data, 0, ds_tid, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
@@ -511,7 +511,7 @@ void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid) {
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, OMPT_CUR_TASK_DATA(this_thr), 1,
|
||||
OMPT_CUR_TASK_INFO(this_thr)->thread_num);
|
||||
OMPT_CUR_TASK_INFO(this_thr)->thread_num, ompt_task_implicit);
|
||||
}
|
||||
|
||||
// reset clear the task id only after unlinking the task
|
||||
@@ -1292,7 +1292,7 @@ __ompt_get_mutex_impl_type(void *user_lock, kmp_indirect_lock_t *ilock = 0) {
|
||||
return kmp_mutex_impl_speculative;
|
||||
#endif
|
||||
default:
|
||||
return ompt_mutex_impl_none;
|
||||
return kmp_mutex_impl_none;
|
||||
}
|
||||
ilock = KMP_LOOKUP_I_LOCK(user_lock);
|
||||
}
|
||||
@@ -1316,7 +1316,7 @@ __ompt_get_mutex_impl_type(void *user_lock, kmp_indirect_lock_t *ilock = 0) {
|
||||
case locktag_nested_drdpa:
|
||||
return kmp_mutex_impl_queuing;
|
||||
default:
|
||||
return ompt_mutex_impl_none;
|
||||
return kmp_mutex_impl_none;
|
||||
}
|
||||
}
|
||||
#else
|
||||
@@ -1339,7 +1339,7 @@ static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
|
||||
return kmp_mutex_impl_speculative;
|
||||
#endif
|
||||
default:
|
||||
return ompt_mutex_impl_none;
|
||||
return kmp_mutex_impl_none;
|
||||
}
|
||||
}
|
||||
#endif // KMP_USE_DYNAMIC_LOCK
|
||||
|
||||
@@ -400,7 +400,7 @@ static
|
||||
ompt_team_size = __kmp_team_from_gtid(gtid)->t.t_nproc;
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_begin, &(team_info->parallel_data),
|
||||
&(task_info->task_data), ompt_team_size, __kmp_tid_from_gtid(gtid));
|
||||
&(task_info->task_data), ompt_team_size, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
task_info->thread_num = __kmp_tid_from_gtid(gtid);
|
||||
}
|
||||
thr->th.ompt_thread_info.state = ompt_state_work_parallel;
|
||||
|
||||
@@ -1429,7 +1429,7 @@ void __kmp_serialized_parallel(ident_t *loc, kmp_int32 global_tid) {
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_begin, OMPT_CUR_TEAM_DATA(this_thr),
|
||||
OMPT_CUR_TASK_DATA(this_thr), 1, __kmp_tid_from_gtid(global_tid));
|
||||
OMPT_CUR_TASK_DATA(this_thr), 1, __kmp_tid_from_gtid(global_tid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
OMPT_CUR_TASK_INFO(this_thr)
|
||||
->thread_num = __kmp_tid_from_gtid(global_tid);
|
||||
}
|
||||
@@ -1595,7 +1595,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
|
||||
implicit_task_data, 1, __kmp_tid_from_gtid(gtid));
|
||||
implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
OMPT_CUR_TASK_INFO(master_th)
|
||||
->thread_num = __kmp_tid_from_gtid(gtid);
|
||||
}
|
||||
@@ -1625,7 +1625,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, implicit_task_data, 1,
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num);
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
}
|
||||
__ompt_lw_taskteam_unlink(master_th);
|
||||
|
||||
@@ -1807,7 +1807,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
|
||||
&(task_info->task_data), 1, __kmp_tid_from_gtid(gtid));
|
||||
&(task_info->task_data), 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
OMPT_CUR_TASK_INFO(master_th)
|
||||
->thread_num = __kmp_tid_from_gtid(gtid);
|
||||
}
|
||||
@@ -1837,7 +1837,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, &(task_info->task_data), 1,
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num);
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
}
|
||||
|
||||
__ompt_lw_taskteam_unlink(master_th);
|
||||
@@ -1908,7 +1908,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
|
||||
implicit_task_data, 1, __kmp_tid_from_gtid(gtid));
|
||||
implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
OMPT_CUR_TASK_INFO(master_th)
|
||||
->thread_num = __kmp_tid_from_gtid(gtid);
|
||||
}
|
||||
@@ -1937,7 +1937,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, &(task_info->task_data), 1,
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num);
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
}
|
||||
|
||||
ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
|
||||
@@ -2541,7 +2541,7 @@ void __kmp_join_call(ident_t *loc, int gtid
|
||||
int ompt_team_size = team->t.t_nproc;
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size,
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num);
|
||||
OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
}
|
||||
|
||||
task_info->frame.exit_frame = ompt_data_none;
|
||||
@@ -7041,7 +7041,7 @@ int __kmp_invoke_task_func(int gtid) {
|
||||
ompt_team_size = team->t.t_nproc;
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_begin, my_parallel_data, my_task_data, ompt_team_size,
|
||||
__kmp_tid_from_gtid(gtid));
|
||||
__kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
OMPT_CUR_TASK_INFO(this_thr)->thread_num = __kmp_tid_from_gtid(gtid);
|
||||
}
|
||||
#endif
|
||||
@@ -7291,7 +7291,7 @@ void __kmp_internal_join(ident_t *id, int gtid, kmp_team_t *team) {
|
||||
#endif
|
||||
if (!KMP_MASTER_TID(ds_tid) && ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, task_data, 0, ds_tid);
|
||||
ompt_scope_end, NULL, task_data, 0, ds_tid, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
@@ -485,43 +485,43 @@ kmp_int32 __kmpc_omp_task_with_deps(ident_t *loc_ref, kmp_int32 gtid,
|
||||
#if OMPT_OPTIONAL
|
||||
/* OMPT grab all dependences if requested by the tool */
|
||||
if (ndeps + ndeps_noalias > 0 &&
|
||||
ompt_enabled.ompt_callback_task_dependences) {
|
||||
ompt_enabled.ompt_callback_dependences) {
|
||||
kmp_int32 i;
|
||||
|
||||
new_taskdata->ompt_task_info.ndeps = ndeps + ndeps_noalias;
|
||||
new_taskdata->ompt_task_info.deps =
|
||||
(ompt_task_dependence_t *)KMP_OMPT_DEPS_ALLOC(
|
||||
thread, (ndeps + ndeps_noalias) * sizeof(ompt_task_dependence_t));
|
||||
(ompt_dependence_t *)KMP_OMPT_DEPS_ALLOC(
|
||||
thread, (ndeps + ndeps_noalias) * sizeof(ompt_dependence_t));
|
||||
|
||||
KMP_ASSERT(new_taskdata->ompt_task_info.deps != NULL);
|
||||
|
||||
for (i = 0; i < ndeps; i++) {
|
||||
new_taskdata->ompt_task_info.deps[i].variable_addr =
|
||||
new_taskdata->ompt_task_info.deps[i].variable.ptr =
|
||||
(void *)dep_list[i].base_addr;
|
||||
if (dep_list[i].flags.in && dep_list[i].flags.out)
|
||||
new_taskdata->ompt_task_info.deps[i].dependence_type =
|
||||
ompt_task_dependence_type_inout;
|
||||
ompt_dependence_type_inout;
|
||||
else if (dep_list[i].flags.out)
|
||||
new_taskdata->ompt_task_info.deps[i].dependence_type =
|
||||
ompt_task_dependence_type_out;
|
||||
ompt_dependence_type_out;
|
||||
else if (dep_list[i].flags.in)
|
||||
new_taskdata->ompt_task_info.deps[i].dependence_type =
|
||||
ompt_task_dependence_type_in;
|
||||
ompt_dependence_type_in;
|
||||
}
|
||||
for (i = 0; i < ndeps_noalias; i++) {
|
||||
new_taskdata->ompt_task_info.deps[ndeps + i].variable_addr =
|
||||
new_taskdata->ompt_task_info.deps[ndeps + i].variable.ptr =
|
||||
(void *)noalias_dep_list[i].base_addr;
|
||||
if (noalias_dep_list[i].flags.in && noalias_dep_list[i].flags.out)
|
||||
new_taskdata->ompt_task_info.deps[ndeps + i].dependence_type =
|
||||
ompt_task_dependence_type_inout;
|
||||
ompt_dependence_type_inout;
|
||||
else if (noalias_dep_list[i].flags.out)
|
||||
new_taskdata->ompt_task_info.deps[ndeps + i].dependence_type =
|
||||
ompt_task_dependence_type_out;
|
||||
ompt_dependence_type_out;
|
||||
else if (noalias_dep_list[i].flags.in)
|
||||
new_taskdata->ompt_task_info.deps[ndeps + i].dependence_type =
|
||||
ompt_task_dependence_type_in;
|
||||
ompt_dependence_type_in;
|
||||
}
|
||||
ompt_callbacks.ompt_callback(ompt_callback_task_dependences)(
|
||||
ompt_callbacks.ompt_callback(ompt_callback_dependences)(
|
||||
&(new_taskdata->ompt_task_info.task_data),
|
||||
new_taskdata->ompt_task_info.deps, new_taskdata->ompt_task_info.ndeps);
|
||||
/* We can now free the allocated memory for the dependencies */
|
||||
|
||||
@@ -140,7 +140,7 @@ static inline void __ompt_implicit_task_end(kmp_info_t *this_thr,
|
||||
if (!KMP_MASTER_TID(ds_tid)) {
|
||||
if (ompt_enabled.ompt_callback_implicit_task) {
|
||||
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
|
||||
ompt_scope_end, NULL, tId, 0, ds_tid);
|
||||
ompt_scope_end, NULL, tId, 0, ds_tid, ompt_task_implicit);
|
||||
}
|
||||
// return to idle state
|
||||
this_thr->th.ompt_thread_info.state = ompt_state_idle;
|
||||
|
||||
@@ -78,11 +78,11 @@
|
||||
#define ompt_callback_mutex_released_implemented ompt_event_MAY_ALWAYS_OPTIONAL
|
||||
|
||||
#if OMP_40_ENABLED
|
||||
#define ompt_callback_task_dependences_implemented \
|
||||
#define ompt_callback_dependences_implemented \
|
||||
ompt_event_MAY_ALWAYS_OPTIONAL
|
||||
#define ompt_callback_task_dependence_implemented ompt_event_MAY_ALWAYS_OPTIONAL
|
||||
#else
|
||||
#define ompt_callback_task_dependences_implemented ompt_event_UNIMPLEMENTED
|
||||
#define ompt_callback_dependences_implemented ompt_event_UNIMPLEMENTED
|
||||
#define ompt_callback_task_dependence_implemented ompt_event_UNIMPLEMENTED
|
||||
#endif /* OMP_40_ENABLED */
|
||||
|
||||
|
||||
@@ -327,6 +327,8 @@ void ompt_pre_init() {
|
||||
#endif
|
||||
}
|
||||
|
||||
extern "C" int omp_get_initial_device(void);
|
||||
|
||||
void ompt_post_init() {
|
||||
//--------------------------------------------------
|
||||
// Execute the post-initialization logic only once.
|
||||
@@ -343,7 +345,7 @@ void ompt_post_init() {
|
||||
//--------------------------------------------------
|
||||
if (ompt_start_tool_result) {
|
||||
ompt_enabled.enabled = !!ompt_start_tool_result->initialize(
|
||||
ompt_fn_lookup, &(ompt_start_tool_result->tool_data));
|
||||
ompt_fn_lookup, omp_get_initial_device(), &(ompt_start_tool_result->tool_data));
|
||||
|
||||
if (!ompt_enabled.enabled) {
|
||||
// tool not enabled, zero out the bitmap, and done
|
||||
@@ -422,7 +424,7 @@ OMPT_API_ROUTINE int ompt_enumerate_mutex_impls(int current_impl,
|
||||
* callbacks
|
||||
****************************************************************************/
|
||||
|
||||
OMPT_API_ROUTINE int ompt_set_callback(ompt_callbacks_t which,
|
||||
OMPT_API_ROUTINE ompt_set_result_t ompt_set_callback(ompt_callbacks_t which,
|
||||
ompt_callback_t callback) {
|
||||
switch (which) {
|
||||
|
||||
@@ -482,8 +484,8 @@ OMPT_API_ROUTINE int ompt_get_parallel_info(int ancestor_level,
|
||||
team_size);
|
||||
}
|
||||
|
||||
OMPT_API_ROUTINE ompt_state_t ompt_get_state(ompt_wait_id_t *wait_id) {
|
||||
ompt_state_t thread_state = __ompt_get_state_internal(wait_id);
|
||||
OMPT_API_ROUTINE int ompt_get_state(ompt_wait_id_t *wait_id) {
|
||||
int thread_state = __ompt_get_state_internal(wait_id);
|
||||
|
||||
if (thread_state == ompt_state_undefined) {
|
||||
thread_state = ompt_state_work_serial;
|
||||
|
||||
@@ -15,7 +15,7 @@
|
||||
#define __OMPT_INTERNAL_H__
|
||||
|
||||
#include "ompt-event-specific.h"
|
||||
#include "ompt.h"
|
||||
#include "omp-tools.h"
|
||||
|
||||
#define OMPT_VERSION 1
|
||||
|
||||
@@ -60,7 +60,7 @@ typedef struct {
|
||||
int thread_num;
|
||||
#if OMP_40_ENABLED
|
||||
int ndeps;
|
||||
ompt_task_dependence_t *deps;
|
||||
ompt_dependence_t *deps;
|
||||
#endif /* OMP_40_ENABLED */
|
||||
} ompt_task_info_t;
|
||||
|
||||
|
||||
@@ -214,7 +214,7 @@ void __ompt_thread_assign_wait_id(void *variable) {
|
||||
ti->th.ompt_thread_info.wait_id = (ompt_wait_id_t)variable;
|
||||
}
|
||||
|
||||
ompt_state_t __ompt_get_state_internal(ompt_wait_id_t *omp_wait_id) {
|
||||
int __ompt_get_state_internal(ompt_wait_id_t *omp_wait_id) {
|
||||
kmp_info_t *ti = ompt_get_thread();
|
||||
|
||||
if (ti) {
|
||||
|
||||
@@ -127,7 +127,7 @@ config.substitutions.append(("%flags", config.test_flags))
|
||||
config.substitutions.append(("%python", '"%s"' % (sys.executable)))
|
||||
|
||||
if config.has_ompt:
|
||||
config.substitutions.append(("FileCheck", config.test_filecheck))
|
||||
config.substitutions.append(("FileCheck", "tee %%s.out | %s" % config.test_filecheck))
|
||||
config.substitutions.append(("%sort-threads", "sort -n -s"))
|
||||
if config.operating_system == 'Windows':
|
||||
# No such environment variable on Windows.
|
||||
|
||||
@@ -8,7 +8,7 @@
|
||||
#endif
|
||||
#include <inttypes.h>
|
||||
#include <omp.h>
|
||||
#include <ompt.h>
|
||||
#include <omp-tools.h>
|
||||
#include "ompt-signal.h"
|
||||
|
||||
// Used to detect architecture
|
||||
@@ -23,10 +23,13 @@ static const char* ompt_thread_t_values[] = {
|
||||
|
||||
static const char* ompt_task_status_t_values[] = {
|
||||
NULL,
|
||||
"ompt_task_complete",
|
||||
"ompt_task_yield",
|
||||
"ompt_task_cancel",
|
||||
"ompt_task_others"
|
||||
"ompt_task_complete", // 1
|
||||
"ompt_task_yield", // 2
|
||||
"ompt_task_cancel", // 3
|
||||
"ompt_task_detach", // 4
|
||||
"ompt_task_early_fulfill", // 5
|
||||
"ompt_task_late_fulfill", // 6
|
||||
"ompt_task_switch" // 7
|
||||
};
|
||||
static const char* ompt_cancel_flag_t_values[] = {
|
||||
"ompt_cancel_parallel",
|
||||
@@ -439,7 +442,8 @@ on_ompt_callback_implicit_task(
|
||||
ompt_data_t *parallel_data,
|
||||
ompt_data_t *task_data,
|
||||
unsigned int team_size,
|
||||
unsigned int thread_num)
|
||||
unsigned int thread_num,
|
||||
int flags)
|
||||
{
|
||||
switch(endpoint)
|
||||
{
|
||||
@@ -651,9 +655,9 @@ on_ompt_callback_task_schedule(
|
||||
}
|
||||
|
||||
static void
|
||||
on_ompt_callback_task_dependences(
|
||||
on_ompt_callback_dependences(
|
||||
ompt_data_t *task_data,
|
||||
const ompt_task_dependence_t *deps,
|
||||
const ompt_dependence_t *deps,
|
||||
int ndeps)
|
||||
{
|
||||
printf("%" PRIu64 ": ompt_event_task_dependences: task_id=%" PRIu64 ", deps=%p, ndeps=%d\n", ompt_get_thread_data()->value, task_data->value, (void *)deps, ndeps);
|
||||
@@ -710,6 +714,7 @@ do{ \
|
||||
|
||||
int ompt_initialize(
|
||||
ompt_function_lookup_t lookup,
|
||||
int initial_device_num,
|
||||
ompt_data_t *tool_data)
|
||||
{
|
||||
ompt_set_callback = (ompt_set_callback_t) lookup("ompt_set_callback");
|
||||
@@ -747,7 +752,7 @@ int ompt_initialize(
|
||||
register_callback(ompt_callback_parallel_end);
|
||||
register_callback(ompt_callback_task_create);
|
||||
register_callback(ompt_callback_task_schedule);
|
||||
register_callback(ompt_callback_task_dependences);
|
||||
register_callback(ompt_callback_dependences);
|
||||
register_callback(ompt_callback_task_dependence);
|
||||
register_callback(ompt_callback_thread_begin);
|
||||
register_callback(ompt_callback_thread_end);
|
||||
@@ -760,6 +765,9 @@ void ompt_finalize(ompt_data_t *tool_data)
|
||||
printf("0: ompt_event_runtime_shutdown\n");
|
||||
}
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
ompt_start_tool_result_t* ompt_start_tool(
|
||||
unsigned int omp_version,
|
||||
const char *runtime_version)
|
||||
@@ -767,3 +775,6 @@ ompt_start_tool_result_t* ompt_start_tool(
|
||||
static ompt_start_tool_result_t ompt_start_tool_result = {&ompt_initialize,&ompt_finalize, 0};
|
||||
return &ompt_start_tool_result;
|
||||
}
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
||||
@@ -75,7 +75,7 @@ int main()
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID]], parent_task_frame.exit={{0x[0-f]*}}, parent_task_frame.reenter={{0x[0-f]*}}, new_task_id=[[THIRD_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}, task_type=ompt_task_explicit=4, has_dependences=no
|
||||
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID]], parent_task_frame.exit={{0x[0-f]*}}, parent_task_frame.reenter={{0x[0-f]*}}, new_task_id=[[CANCEL_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}, task_type=ompt_task_explicit|ompt_task_undeferred=134217732, has_dependences=no
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[PARENT_TASK_ID]], second_task_id=[[CANCEL_TASK_ID]], prior_task_status=ompt_task_others=4
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[PARENT_TASK_ID]], second_task_id=[[CANCEL_TASK_ID]], prior_task_status=ompt_task_switch=7
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_cancel: task_data=[[CANCEL_TASK_ID]], flags=ompt_cancel_taskgroup|ompt_cancel_activated=24, codeptr_ra={{0x[0-f]*}}
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[CANCEL_TASK_ID]], second_task_id=[[PARENT_TASK_ID]], prior_task_status=ompt_task_cancel=3
|
||||
|
||||
|
||||
@@ -49,7 +49,7 @@ int main()
|
||||
#ifdef TOOL
|
||||
|
||||
#include <stdio.h>
|
||||
#include <ompt.h>
|
||||
#include <omp-tools.h>
|
||||
|
||||
int ompt_initialize(
|
||||
ompt_function_lookup_t lookup,
|
||||
|
||||
@@ -18,7 +18,7 @@
|
||||
#ifdef CODE
|
||||
#include "stdio.h"
|
||||
#include "omp.h"
|
||||
#include "ompt.h"
|
||||
#include "omp-tools.h"
|
||||
|
||||
int main()
|
||||
{
|
||||
@@ -52,7 +52,7 @@ int main()
|
||||
|
||||
#ifdef TOOL
|
||||
|
||||
#include <ompt.h>
|
||||
#include <omp-tools.h>
|
||||
#include "stdio.h"
|
||||
|
||||
#ifdef SECOND_TOOL
|
||||
|
||||
@@ -28,7 +28,7 @@
|
||||
#ifdef CODE
|
||||
#include "stdio.h"
|
||||
#include "omp.h"
|
||||
#include "ompt.h"
|
||||
#include "omp-tools.h"
|
||||
|
||||
int main()
|
||||
{
|
||||
@@ -56,7 +56,7 @@ int main()
|
||||
|
||||
#ifdef TOOL
|
||||
|
||||
#include <ompt.h>
|
||||
#include <omp-tools.h>
|
||||
#include "stdio.h"
|
||||
|
||||
ompt_start_tool_result_t* ompt_start_tool(
|
||||
|
||||
@@ -57,7 +57,7 @@ int main()
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame={{0x[0-f]+}}
|
||||
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id=[[EXPLICIT_TASK_ID:[0-9]+]]
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[NESTED_IMPLICIT_TASK_ID]], second_task_id=[[EXPLICIT_TASK_ID]], prior_task_status=ompt_task_others=4
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[NESTED_IMPLICIT_TASK_ID]], second_task_id=[[EXPLICIT_TASK_ID]], prior_task_status=ompt_task_switch=7
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[EXPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
|
||||
|
||||
@@ -57,7 +57,7 @@ int main()
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame={{0x[0-f]+}}
|
||||
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id=[[EXPLICIT_TASK_ID:[0-9]+]]
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[NESTED_IMPLICIT_TASK_ID]], second_task_id=[[EXPLICIT_TASK_ID]], prior_task_status=ompt_task_others=4
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[NESTED_IMPLICIT_TASK_ID]], second_task_id=[[EXPLICIT_TASK_ID]], prior_task_status=ompt_task_switch=7
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[EXPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
|
||||
// CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
|
||||
|
||||
@@ -12,7 +12,7 @@
|
||||
#include <unistd.h>
|
||||
#include <inttypes.h>
|
||||
#include <omp.h>
|
||||
#include <ompt.h>
|
||||
#include <omp-tools.h>
|
||||
|
||||
static const char* ompt_thread_t_values[] = {
|
||||
NULL,
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
|
||||
// RUN: %libomp-compile-and-run | %sort-threads | tee %s.out | FileCheck %s
|
||||
// REQUIRES: ompt
|
||||
// UNSUPPORTED: gcc-4, gcc-5, gcc-6, gcc-7
|
||||
#define TEST_NEED_PRINT_FRAME_FROM_OUTLINED_FN
|
||||
|
||||
@@ -51,7 +51,7 @@ int main()
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[IMPLICIT_TASK_ID]], second_task_id=[[MAIN_TASK]], prior_task_status=ompt_task_yield=2
|
||||
// CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[MAIN_TASK]], second_task_id=[[IMPLICIT_TASK_ID]], prior_task_status=ompt_task_complete=1
|
||||
|
||||
// CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_task_schedule: first_task_id={{[0-9]+}}, second_task_id=[[WORKER_TASK]], prior_task_status=ompt_task_others=4
|
||||
// CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_task_schedule: first_task_id={{[0-9]+}}, second_task_id=[[WORKER_TASK]], prior_task_status=ompt_task_switch=7
|
||||
// CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[WORKER_TASK]], second_task_id={{[0-9]+}}, prior_task_status=ompt_task_complete=1
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user