Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F87440987
coul_long.cpp
No One
Temporary
Actions
Download File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Award Token
Subscribers
None
File Metadata
Details
File Info
Storage
Attached
Created
Sat, Oct 12, 17:16
Size
5 KB
Mime Type
text/x-c++
Expires
Mon, Oct 14, 17:16 (2 d)
Engine
blob
Format
Raw Data
Handle
21595219
Attached To
rLAMMPS lammps
coul_long.cpp
View Options
/***************************************************************************
coul_long.cpp
-------------------
Axel Kohlmeyer (Temple)
Class for acceleration of the coul/long pair style.
__________________________________________________________________________
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin : July 2011
email : a.kohlmeyer@temple.edu
***************************************************************************/
#ifdef USE_OPENCL
#include "coul_long_cl.h"
#else
#include "coul_long_ptx.h"
#endif
#include "coul_long.h"
#include <cassert>
using
namespace
LAMMPS_AL
;
#define CoulLongT CoulLong<numtyp, acctyp>
extern
Device
<
PRECISION
,
ACC_PRECISION
>
pair_gpu_device
;
template
<
class
numtyp
,
class
acctyp
>
CoulLongT
::
CoulLong
()
:
BaseCharge
<
numtyp
,
acctyp
>
(),
_allocated
(
false
)
{
}
template
<
class
numtyp
,
class
acctyp
>
CoulLongT
::~
CoulLong
()
{
clear
();
}
template
<
class
numtyp
,
class
acctyp
>
int
CoulLongT
::
bytes_per_atom
(
const
int
max_nbors
)
const
{
return
this
->
bytes_per_atom_atomic
(
max_nbors
);
}
template
<
class
numtyp
,
class
acctyp
>
int
CoulLongT
::
init
(
const
int
nlocal
,
const
int
nall
,
const
int
max_nbors
,
const
int
maxspecial
,
const
double
cell_size
,
const
double
gpu_split
,
FILE
*
_screen
,
const
double
host_cut_coulsq
,
double
*
host_special_coul
,
const
double
qqrd2e
,
const
double
g_ewald
)
{
int
success
;
success
=
this
->
init_atomic
(
nlocal
,
nall
,
max_nbors
,
maxspecial
,
cell_size
,
gpu_split
,
_screen
,
coul_long
);
if
(
success
!=
0
)
return
success
;
// we don't have atom types for coulomb only,
// but go with the minimum so that we can use
// the same infrastructure as lj/cut/coul/long/gpu.
int
lj_types
=
1
;
shared_types
=
false
;
int
max_shared_types
=
this
->
device
->
max_shared_types
();
if
(
lj_types
<=
max_shared_types
&&
this
->
_block_size
>=
max_shared_types
)
{
lj_types
=
max_shared_types
;
shared_types
=
true
;
}
_lj_types
=
lj_types
;
// Allocate a host write buffer for data initialization
UCL_H_Vec
<
numtyp
>
host_write
(
lj_types
*
lj_types
*
32
,
*
(
this
->
ucl_device
),
UCL_WRITE_OPTIMIZED
);
for
(
int
i
=
0
;
i
<
lj_types
*
lj_types
;
i
++
)
host_write
[
i
]
=
0.0
;
lj1
.
alloc
(
lj_types
*
lj_types
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
lj3
.
alloc
(
lj_types
*
lj_types
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
sp_cl
.
alloc
(
4
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
for
(
int
i
=
0
;
i
<
4
;
i
++
)
{
host_write
[
i
]
=
host_special_coul
[
i
];
}
ucl_copy
(
sp_cl
,
host_write
,
4
,
false
);
_cut_coulsq
=
host_cut_coulsq
;
_qqrd2e
=
qqrd2e
;
_g_ewald
=
g_ewald
;
_allocated
=
true
;
this
->
_max_bytes
=
lj1
.
row_bytes
()
+
lj3
.
row_bytes
()
+
sp_cl
.
row_bytes
();
return
0
;
}
template
<
class
numtyp
,
class
acctyp
>
void
CoulLongT
::
clear
()
{
if
(
!
_allocated
)
return
;
_allocated
=
false
;
lj1
.
clear
();
lj3
.
clear
();
sp_cl
.
clear
();
this
->
clear_atomic
();
}
template
<
class
numtyp
,
class
acctyp
>
double
CoulLongT
::
host_memory_usage
()
const
{
return
this
->
host_memory_usage_atomic
()
+
sizeof
(
CoulLong
<
numtyp
,
acctyp
>
);
}
// ---------------------------------------------------------------------------
// Calculate energies, forces, and torques
// ---------------------------------------------------------------------------
template
<
class
numtyp
,
class
acctyp
>
void
CoulLongT
::
loop
(
const
bool
_eflag
,
const
bool
_vflag
)
{
// Compute the block size and grid size to keep all cores busy
const
int
BX
=
this
->
block_size
();
int
eflag
,
vflag
;
if
(
_eflag
)
eflag
=
1
;
else
eflag
=
0
;
if
(
_vflag
)
vflag
=
1
;
else
vflag
=
0
;
int
GX
=
static_cast
<
int
>
(
ceil
(
static_cast
<
double
>
(
this
->
ans
->
inum
())
/
(
BX
/
this
->
_threads_per_atom
)));
int
ainum
=
this
->
ans
->
inum
();
int
nbor_pitch
=
this
->
nbor
->
nbor_pitch
();
this
->
time_pair
.
start
();
if
(
shared_types
)
{
this
->
k_pair_fast
.
set_size
(
GX
,
BX
);
this
->
k_pair_fast
.
run
(
&
this
->
atom
->
dev_x
.
begin
(),
&
lj1
.
begin
(),
&
lj3
.
begin
(),
&
sp_cl
.
begin
(),
&
this
->
nbor
->
dev_nbor
.
begin
(),
&
this
->
_nbor_data
->
begin
(),
&
this
->
ans
->
dev_ans
.
begin
(),
&
this
->
ans
->
dev_engv
.
begin
(),
&
eflag
,
&
vflag
,
&
ainum
,
&
nbor_pitch
,
&
this
->
atom
->
dev_q
.
begin
(),
&
_cut_coulsq
,
&
_qqrd2e
,
&
_g_ewald
,
&
this
->
_threads_per_atom
);
}
else
{
this
->
k_pair
.
set_size
(
GX
,
BX
);
this
->
k_pair
.
run
(
&
this
->
atom
->
dev_x
.
begin
(),
&
lj1
.
begin
(),
&
lj3
.
begin
(),
&
_lj_types
,
&
sp_cl
.
begin
(),
&
this
->
nbor
->
dev_nbor
.
begin
(),
&
this
->
_nbor_data
->
begin
(),
&
this
->
ans
->
dev_ans
.
begin
(),
&
this
->
ans
->
dev_engv
.
begin
(),
&
eflag
,
&
vflag
,
&
ainum
,
&
nbor_pitch
,
&
this
->
atom
->
dev_q
.
begin
(),
&
_cut_coulsq
,
&
_qqrd2e
,
&
_g_ewald
,
&
this
->
_threads_per_atom
);
}
this
->
time_pair
.
stop
();
}
template
class
CoulLong
<
PRECISION
,
ACC_PRECISION
>
;
Event Timeline
Log In to Comment