Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F79275029
lal_buck_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
Sun, Aug 25, 14:51
Size
5 KB
Mime Type
text/x-c++
Expires
Tue, Aug 27, 14:51 (2 d)
Engine
blob
Format
Raw Data
Handle
19619062
Attached To
rLAMMPS lammps
lal_buck_coul_long.cpp
View Options
/***************************************************************************
buck_coul_long.cpp
-------------------
Trung Dac Nguyen (ORNL)
Class for acceleration of the buck/coul/long pair style.
__________________________________________________________________________
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
email : nguyentd@ornl.gov
***************************************************************************/
#if defined(USE_OPENCL)
#include "buck_coul_long_cl.h"
#elif defined(USE_CUDART)
const
char
*
buck_coul_long
=
0
;
#else
#include "buck_coul_long_cubin.h"
#endif
#include "lal_buck_coul_long.h"
#include <cassert>
using
namespace
LAMMPS_AL
;
#define BuckCoulLongT BuckCoulLong<numtyp, acctyp>
extern
Device
<
PRECISION
,
ACC_PRECISION
>
device
;
template
<
class
numtyp
,
class
acctyp
>
BuckCoulLongT
::
BuckCoulLong
()
:
BaseCharge
<
numtyp
,
acctyp
>
(),
_allocated
(
false
)
{
}
template
<
class
numtyp
,
class
acctyp
>
BuckCoulLongT
::~
BuckCoulLongT
()
{
clear
();
}
template
<
class
numtyp
,
class
acctyp
>
int
BuckCoulLongT
::
bytes_per_atom
(
const
int
max_nbors
)
const
{
return
this
->
bytes_per_atom_atomic
(
max_nbors
);
}
template
<
class
numtyp
,
class
acctyp
>
int
BuckCoulLongT
::
init
(
const
int
ntypes
,
double
**
host_cutsq
,
double
**
host_rhoinv
,
double
**
host_buck1
,
double
**
host_buck2
,
double
**
host_a
,
double
**
host_c
,
double
**
host_offset
,
double
*
host_special_lj
,
const
int
nlocal
,
const
int
nall
,
const
int
max_nbors
,
const
int
maxspecial
,
const
double
cell_size
,
const
double
gpu_split
,
FILE
*
_screen
,
double
**
host_cut_ljsq
,
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
,
buck_coul_long
,
"k_buck_coul_long"
);
if
(
success
!=
0
)
return
success
;
// If atom type constants fit in shared memory use fast kernel
int
lj_types
=
ntypes
;
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_ONLY
);
for
(
int
i
=
0
;
i
<
lj_types
*
lj_types
;
i
++
)
host_write
[
i
]
=
0.0
;
coeff1
.
alloc
(
lj_types
*
lj_types
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
this
->
atom
->
type_pack4
(
ntypes
,
lj_types
,
coeff1
,
host_write
,
host_rhoinv
,
host_buck1
,
host_buck2
,
host_cut_ljsq
);
coeff2
.
alloc
(
lj_types
*
lj_types
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
this
->
atom
->
type_pack4
(
ntypes
,
lj_types
,
coeff2
,
host_write
,
host_a
,
host_c
,
host_offset
);
cutsq
.
alloc
(
lj_types
*
lj_types
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
this
->
atom
->
type_pack1
(
ntypes
,
lj_types
,
cutsq
,
host_write
,
host_cutsq
);
sp_lj
.
alloc
(
8
,
*
(
this
->
ucl_device
),
UCL_READ_ONLY
);
for
(
int
i
=
0
;
i
<
4
;
i
++
)
{
host_write
[
i
]
=
host_special_lj
[
i
];
host_write
[
i
+
4
]
=
host_special_coul
[
i
];
}
ucl_copy
(
sp_lj
,
host_write
,
8
,
false
);
_cut_coulsq
=
host_cut_coulsq
;
_qqrd2e
=
qqrd2e
;
_g_ewald
=
g_ewald
;
_allocated
=
true
;
this
->
_max_bytes
=
coeff1
.
row_bytes
()
+
coeff2
.
row_bytes
()
+
sp_lj
.
row_bytes
();
return
0
;
}
template
<
class
numtyp
,
class
acctyp
>
void
BuckCoulLongT
::
clear
()
{
if
(
!
_allocated
)
return
;
_allocated
=
false
;
coeff1
.
clear
();
coeff2
.
clear
();
cutsq
.
clear
();
sp_lj
.
clear
();
this
->
clear_atomic
();
}
template
<
class
numtyp
,
class
acctyp
>
double
BuckCoulLongT
::
host_memory_usage
()
const
{
return
this
->
host_memory_usage_atomic
()
+
sizeof
(
BuckCoulLong
<
numtyp
,
acctyp
>
);
}
// ---------------------------------------------------------------------------
// Calculate energies, forces, and torques
// ---------------------------------------------------------------------------
template
<
class
numtyp
,
class
acctyp
>
void
BuckCoulLongT
::
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
->
x
,
&
coeff1
,
&
coeff2
,
&
sp_lj
,
&
this
->
nbor
->
dev_nbor
,
&
this
->
_nbor_data
->
begin
(),
&
this
->
ans
->
force
,
&
this
->
ans
->
engv
,
&
eflag
,
&
vflag
,
&
ainum
,
&
nbor_pitch
,
&
this
->
atom
->
q
,
&
cutsq
,
&
_cut_coulsq
,
&
_qqrd2e
,
&
_g_ewald
,
&
this
->
_threads_per_atom
);
}
else
{
this
->
k_pair
.
set_size
(
GX
,
BX
);
this
->
k_pair
.
run
(
&
this
->
atom
->
x
,
&
coeff1
,
&
coeff2
,
&
_lj_types
,
&
sp_lj
,
&
this
->
nbor
->
dev_nbor
,
&
this
->
_nbor_data
->
begin
(),
&
this
->
ans
->
force
,
&
this
->
ans
->
engv
,
&
eflag
,
&
vflag
,
&
ainum
,
&
nbor_pitch
,
&
this
->
atom
->
q
,
&
cutsq
,
&
_cut_coulsq
,
&
_qqrd2e
,
&
_g_ewald
,
&
this
->
_threads_per_atom
);
}
this
->
time_pair
.
stop
();
}
template
class
BuckCoulLong
<
PRECISION
,
ACC_PRECISION
>
;
Event Timeline
Log In to Comment