device Module Reference

Device abstraction, common interface for various accelerators.

Data Types
interface	device_associate
	Associate a Fortran array to a (allocated) device pointer. More...
interface	device_associated
	Check if a Fortran array is assoicated with a device pointer. More...
interface	device_deassociate
	Deassociate a Fortran array from a device pointer. More...
interface	device_get_ptr
	Return the device pointer for an associated Fortran array. More...
interface	device_map
	Map a Fortran array to a device (allocate and associate) More...
interface	device_memcpy
	Copy data between host and device (or device and device) More...
interface	device_sync
	Synchronize a device or stream. More...
interface	device_unmap
	Unmap a Fortran array from a device (deassociate and free) More...

Functions/Subroutines
subroutine, public	device_init
subroutine, public	device_finalize
subroutine, public	device_name (name)
integer function, public	device_count ()
	Return the number of available devices.
integer function, public	device_mp_count ()
	Return the number of multiprocessors (SMs on NVIDIA, CUs on AMD, compute units on OpenCL, GPU cores on Apple Silicon) of the current device, or 0 if the backend cannot tell.
integer(kind=i8) function, public	device_total_mem ()
	Return the total memory of the current device in bytes, or 0 if the backend cannot tell.
subroutine, public	device_alloc (x_d, s)
	Allocate memory on the device.
subroutine, public	device_free (x_d)
	Deallocate memory on the device.
subroutine, public	device_memset (x_d, v, s, sync, strm)
	Set memory on the device to a value.
subroutine	device_memcpy_r1 (x, x_d, n, dir, sync, strm)
	Copy data between host and device (rank 1 arrays)
subroutine	device_memcpy_r2 (x, x_d, n, dir, sync, strm)
	Copy data between host and device (rank 2 arrays)
subroutine	device_memcpy_r3 (x, x_d, n, dir, sync, strm)
	Copy data between host and device (rank 3 arrays)
subroutine	device_memcpy_r4 (x, x_d, n, dir, sync, strm)
	Copy data between host and device (rank 4 arrays)
subroutine	device_memcpy_cptr (dst, src, s, dir, sync, strm)
	Copy data between host and device (or device and device) (c-pointers)
subroutine	device_memcpy_common (ptr_h, x_d, s, dir, sync_device, stream)
	Copy data between host and device.
subroutine	device_associate_r1 (x, x_d, n)
	Associate a Fortran rank 1 array to a (allocated) device pointer.
subroutine	device_associate_r2 (x, x_d, n)
	Associate a Fortran rank 2 array to a (allocated) device pointer.
subroutine	device_associate_r3 (x, x_d, n)
	Associate a Fortran rank 3 array to a (allocated) device pointer.
subroutine	device_associate_r4 (x, x_d, n)
	Associate a Fortran rank 4 array to a (allocated) device pointer.
subroutine	device_deassociate_r1 (x)
	Deassociate a Fortran rank 1 array from a device pointer.
subroutine	device_deassociate_r2 (x)
	Deassociate a Fortran rank 2 array from a device pointer.
subroutine	device_deassociate_r3 (x)
	Deassociate a Fortran rank 3 array from a device pointer.
subroutine	device_deassociate_r4 (x)
	Deassociate a Fortran rank 4 array from a device pointer.
subroutine	device_map_common (ptr_h, x_d, s)
	Allocate device memory backing a mapped host array.
subroutine	device_map_r1 (x, x_d, n)
	Map a Fortran rank 1 array to a device (allocate and associate)
subroutine	device_map_r2 (x, x_d, n)
	Map a Fortran rank 2 array to a device (allocate and associate)
subroutine	device_map_r3 (x, x_d, n)
	Map a Fortran rank 3 array to a device (allocate and associate)
subroutine	device_map_r4 (x, x_d, n)
	Map a Fortran rank 4 array to a device (allocate and associate)
subroutine	device_unmap_r1 (x, x_d)
	Unmap a Fortran rank 1 array from a device (deassociate and free)
subroutine	device_unmap_r2 (x, x_d)
	Unmap a Fortran rank 2 array from a device (deassociate and free)
subroutine	device_unmap_r3 (x, x_d)
	Unmap a Fortran rank 3 array from a device (deassociate and free)
subroutine	device_unmap_r4 (x, x_d)
	Unmap a Fortran rank 4 array from a device (deassociate and free)
logical function	device_associated_r1 (x)
	Check if a Fortran rank 1 array is assoicated with a device pointer.
logical function	device_associated_r2 (x)
	Check if a Fortran rank 2 array is assoicated with a device pointer.
logical function	device_associated_r3 (x)
	Check if a Fortran rank 3 array is assoicated with a device pointer.
logical function	device_associated_r4 (x)
	Check if a Fortran rank 4 array is assoicated with a device pointer.
type(c_ptr) function	device_get_ptr_r1 (x)
	Return the device pointer for an associated Fortran rank 1 array.
type(c_ptr) function	device_get_ptr_r2 (x)
	Return the device pointer for an associated Fortran rank 2 array.
type(c_ptr) function	device_get_ptr_r3 (x)
	Return the device pointer for an associated Fortran rank 3 array.
type(c_ptr) function	device_get_ptr_r4 (x)
	Return the device pointer for an associated Fortran rank 4 array.
subroutine	device_sync_device ()
	Synchronize the device.
subroutine, public	device_sync_stream (stream)
	Synchronize a device stream.
subroutine, public	device_stream_create (stream, flags)
	Create a device stream/command queue.
subroutine, public	device_stream_create_with_priority (stream, flags, prio)
	Create a device stream/command queue with priority.
subroutine, public	device_stream_destroy (stream)
	Destroy a device stream/command queue.
subroutine, public	device_stream_wait_event (stream, event, flags)
	Synchronize a device stream with an event.
subroutine, public	device_profiler_start ()
	Start device profiling.
subroutine, public	device_profiler_stop ()
	Stop device profiling.
subroutine, public	device_event_create (event, flags)
	Create a device event queue.
subroutine, public	device_event_destroy (event)
	Destroy a device event.
subroutine, public	device_event_record (event, stream)
	Record a device event.
subroutine, public	device_event_sync (event)
	Synchronize an event.

Variables
integer, parameter, public	host_to_device = 1
integer, parameter, public	device_to_host = 2
integer, parameter, public	device_to_device = 3
type(c_ptr), bind(C), public	glb_cmd_queue = C_NULL_PTR
	Global command queue.
type(c_ptr), bind(C), public	aux_cmd_queue = C_NULL_PTR
	Aux command queue.
type(c_ptr), bind(C), public	prf_cmd_queue = C_NULL_PTR
	Profiling command queue.
type(c_ptr), bind(C), public	glb_cmd_event
	Event for the global command queue.
integer, public	strm_high_prio
	High priority stream setting.
integer, public	strm_low_prio
	Low priority stream setting.
type(htable_cptr_t)	device_addrtbl
	Table of host to device address mappings.

Function/Subroutine Documentation

device_alloc()

subroutine, public device::device_alloc	(	type(c_ptr), intent(inout)	x_d,
		integer(c_size_t)	s
	)

Definition at line 249 of file device.F90.

Here is the caller graph for this function:

device_associate_r1()

subroutine device::device_associate_r1 ( class(*), dimension(:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in), optional  n  )

private

Definition at line 675 of file device.F90.

device_associate_r2()

subroutine device::device_associate_r2 ( class(*), dimension(:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in), optional  n  )

private

Definition at line 712 of file device.F90.

device_associate_r3()

subroutine device::device_associate_r3 ( class(*), dimension(:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in), optional  n  )

private

Definition at line 749 of file device.F90.

device_associate_r4()

subroutine device::device_associate_r4 ( class(*), dimension(:,:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in), optional  n  )

private

Definition at line 785 of file device.F90.

device_associated_r1()

logical function device::device_associated_r1 ( class(*), dimension(:), intent(inout), target  x )

private

Definition at line 1237 of file device.F90.

device_associated_r2()

logical function device::device_associated_r2 ( class(*), dimension(:,:), intent(inout), target  x )

private

Definition at line 1264 of file device.F90.

device_associated_r3()

logical function device::device_associated_r3 ( class(*), dimension(:,:,:), intent(inout), target  x )

private

Definition at line 1291 of file device.F90.

device_associated_r4()

logical function device::device_associated_r4 ( class(*), dimension(:,:,:,:), intent(inout), target  x )

private

Definition at line 1318 of file device.F90.

device_count()

integer function, public device::device_count

Definition at line 197 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_deassociate_r1()

subroutine device::device_deassociate_r1 ( class(*), dimension(:), intent(inout), target  x )

private

Definition at line 822 of file device.F90.

device_deassociate_r2()

subroutine device::device_deassociate_r2 ( class(*), dimension(:,:), intent(inout), target  x )

private

Definition at line 846 of file device.F90.

device_deassociate_r3()

subroutine device::device_deassociate_r3 ( class(*), dimension(:,:,:), intent(inout), target  x )

private

Definition at line 870 of file device.F90.

device_deassociate_r4()

subroutine device::device_deassociate_r4 ( class(*), dimension(:,:,:,:), intent(inout), target  x )

private

Definition at line 894 of file device.F90.

device_event_create()

subroutine, public device::device_event_create	(	type(c_ptr), intent(inout)	event,
		integer, optional	flags
	)

Definition at line 1626 of file device.F90.

Here is the caller graph for this function:

device_event_destroy()

subroutine, public device::device_event_destroy

(

type(c_ptr), intent(inout)

event

)

Definition at line 1660 of file device.F90.

Here is the caller graph for this function:

device_event_record()

subroutine, public device::device_event_record	(	type(c_ptr), intent(in), target	event,
		type(c_ptr), intent(in)	stream
	)

Definition at line 1681 of file device.F90.

Here is the caller graph for this function:

device_event_sync()

subroutine, public device::device_event_sync

(

type(c_ptr), intent(in), target

event

)

Definition at line 1704 of file device.F90.

Here is the caller graph for this function:

device_finalize()

subroutine, public device::device_finalize

Definition at line 163 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_free()

subroutine, public device::device_free

(

type(c_ptr), intent(inout)

x_d

)

Definition at line 280 of file device.F90.

Here is the caller graph for this function:

device_get_ptr_r1()

type(c_ptr) function device::device_get_ptr_r1 ( class(*), dimension(:), intent(in), target  x )

private

Definition at line 1345 of file device.F90.

Here is the caller graph for this function:

device_get_ptr_r2()

type(c_ptr) function device::device_get_ptr_r2 ( class(*), dimension(:,:), intent(in), target  x )

private

Definition at line 1373 of file device.F90.

Here is the caller graph for this function:

device_get_ptr_r3()

type(c_ptr) function device::device_get_ptr_r3 ( class(*), dimension(:,:,:), intent(in), target  x )

private

Definition at line 1401 of file device.F90.

Here is the caller graph for this function:

device_get_ptr_r4()

type(c_ptr) function device::device_get_ptr_r4 ( class(*), dimension(:,:,:,:), intent(in), target  x )

private

Definition at line 1429 of file device.F90.

Here is the caller graph for this function:

device_init()

subroutine, public device::device_init

Definition at line 138 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_map_common()

subroutine device::device_map_common ( type(c_ptr), intent(in)  ptr_h, type(c_ptr), intent(inout)  x_d, integer(c_size_t), intent(in)  s  )

private

Note

On unified memory architectures (Metal) the device pointer aliases the host array whenever possible, such that host and device share a single allocation instead of replicating data

Definition at line 921 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_map_r1()

subroutine device::device_map_r1 ( class(*), dimension(:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n  )

private

Definition at line 953 of file device.F90.

device_map_r2()

subroutine device::device_map_r2 ( class(*), dimension(:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n  )

private

Definition at line 987 of file device.F90.

device_map_r3()

subroutine device::device_map_r3 ( class(*), dimension(:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n  )

private

Definition at line 1021 of file device.F90.

device_map_r4()

subroutine device::device_map_r4 ( class(*), dimension(:,:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n  )

private

Definition at line 1055 of file device.F90.

device_memcpy_common()

subroutine device::device_memcpy_common ( type(c_ptr), intent(inout)  ptr_h, type(c_ptr), intent(inout)  x_d, integer(c_size_t), intent(in)  s, integer, intent(in), value  dir, logical, intent(in)  sync_device, type(c_ptr), intent(inout)  stream  )

private

Note

For device to device copies, ptr_h is assumed to be the dst device pointer

Definition at line 537 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_memcpy_cptr()

subroutine device::device_memcpy_cptr ( type(c_ptr), intent(inout)  dst, type(c_ptr), intent(inout)  src, integer(c_size_t), intent(in)  s, integer, intent(in), value  dir, logical, optional  sync, type(c_ptr), optional  strm  )

private

Note

For host-device copies dst is the host pointer and src is the device pointer (regardless of dir)

Definition at line 508 of file device.F90.

device_memcpy_r1()

subroutine device::device_memcpy_r1 ( class(*), dimension(:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n, integer, intent(in), value  dir, logical  sync, type(c_ptr), optional  strm  )

private

Definition at line 354 of file device.F90.

device_memcpy_r2()

subroutine device::device_memcpy_r2 ( class(*), dimension(:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n, integer, intent(in), value  dir, logical  sync, type(c_ptr), optional  strm  )

private

Definition at line 392 of file device.F90.

device_memcpy_r3()

subroutine device::device_memcpy_r3 ( class(*), dimension(:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n, integer, intent(in), value  dir, logical  sync, type(c_ptr), optional  strm  )

private

Definition at line 430 of file device.F90.

device_memcpy_r4()

subroutine device::device_memcpy_r4 ( class(*), dimension(:,:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d, integer, intent(in)  n, integer, intent(in), value  dir, logical  sync, type(c_ptr), optional  strm  )

private

Definition at line 468 of file device.F90.

device_memset()

subroutine, public device::device_memset	(	type(c_ptr), intent(inout)	x_d,
		integer(c_int), target, value	v,
		integer(c_size_t), intent(in)	s,
		logical, optional	sync,
		type(c_ptr), optional	strm
	)

Definition at line 305 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_mp_count()

integer function, public device::device_mp_count

Note

Used to size work decompositions so that every launch fills the device with complete waves of blocks.

Definition at line 216 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_name()

subroutine, public device::device_name

(

character(len=*), intent(inout)

name

)

Definition at line 182 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_profiler_start()

subroutine, public device::device_profiler_start

Definition at line 1608 of file device.F90.

Here is the caller graph for this function:

device_profiler_stop()

subroutine, public device::device_profiler_stop

Definition at line 1617 of file device.F90.

Here is the caller graph for this function:

device_stream_create()

subroutine, public device::device_stream_create	(	type(c_ptr), intent(inout)	stream,
		integer, optional	flags
	)

Definition at line 1500 of file device.F90.

device_stream_create_with_priority()

subroutine, public device::device_stream_create_with_priority	(	type(c_ptr), intent(inout)	stream,
		integer, intent(in)	flags,
		integer, intent(in)	prio
	)

Definition at line 1537 of file device.F90.

Here is the caller graph for this function:

device_stream_destroy()

subroutine, public device::device_stream_destroy

(

type(c_ptr), intent(inout)

stream

)

Definition at line 1559 of file device.F90.

Here is the caller graph for this function:

device_stream_wait_event()

subroutine, public device::device_stream_wait_event	(	type(c_ptr), intent(in)	stream,
		type(c_ptr), intent(in), target	event,
		integer	flags
	)

Definition at line 1581 of file device.F90.

Here is the caller graph for this function:

device_sync_device()

subroutine device::device_sync_device

private

Definition at line 1457 of file device.F90.

device_sync_stream()

subroutine, public device::device_sync_stream

(

type(c_ptr), intent(in)

stream

)

Definition at line 1478 of file device.F90.

Here is the caller graph for this function:

device_total_mem()

integer(kind=i8) function, public device::device_total_mem

Note

On Metal (unified memory) this is the recommended maximum working set rather than physical memory.

Definition at line 234 of file device.F90.

Here is the call graph for this function:

Here is the caller graph for this function:

device_unmap_r1()

subroutine device::device_unmap_r1 ( class(*), dimension(:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d  )

private

Definition at line 1089 of file device.F90.

device_unmap_r2()

subroutine device::device_unmap_r2 ( class(*), dimension(:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d  )

private

Definition at line 1126 of file device.F90.

device_unmap_r3()

subroutine device::device_unmap_r3 ( class(*), dimension(:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d  )

private

Definition at line 1163 of file device.F90.

device_unmap_r4()

subroutine device::device_unmap_r4 ( class(*), dimension(:,:,:,:), intent(inout), target  x, type(c_ptr), intent(inout)  x_d  )

private

Definition at line 1200 of file device.F90.

Variable Documentation

aux_cmd_queue

type(c_ptr), bind(C), public device::aux_cmd_queue = C_NULL_PTR

Definition at line 55 of file device.F90.

device_addrtbl

type(htable_cptr_t) device::device_addrtbl

private

Definition at line 124 of file device.F90.

device_to_device

integer, parameter, public device::device_to_device = 3

Definition at line 48 of file device.F90.

device_to_host

integer, parameter, public device::device_to_host = 2

Definition at line 48 of file device.F90.

glb_cmd_event

type(c_ptr), bind(C), public device::glb_cmd_event

Definition at line 63 of file device.F90.

glb_cmd_queue

type(c_ptr), bind(C), public device::glb_cmd_queue = C_NULL_PTR

Definition at line 52 of file device.F90.

host_to_device

integer, parameter, public device::host_to_device = 1

Definition at line 48 of file device.F90.

prf_cmd_queue

type(c_ptr), bind(C), public device::prf_cmd_queue = C_NULL_PTR

Definition at line 59 of file device.F90.

strm_high_prio

integer, public device::strm_high_prio

Definition at line 66 of file device.F90.

strm_low_prio

integer, public device::strm_low_prio

Definition at line 69 of file device.F90.