Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
SWIFTsim
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Deploy
Releases
Model registry
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
SWIFT
SWIFTsim
Commits
1b0030c6
Commit
1b0030c6
authored
8 years ago
by
James Willis
Browse files
Options
Downloads
Patches
Plain Diff
Added function to caclculate the nonsym density interaction with one set of vectors.
parent
ce5ab497
Branches
Branches containing commit
Tags
Tags containing commit
1 merge request
!320
Dopair1 vectorisation merge
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
src/hydro/Gadget2/hydro_iact.h
+103
-0
103 additions, 0 deletions
src/hydro/Gadget2/hydro_iact.h
with
103 additions
and
0 deletions
src/hydro/Gadget2/hydro_iact.h
+
103
−
0
View file @
1b0030c6
...
...
@@ -373,6 +373,109 @@ runner_iact_nonsym_vec_density(float *R2, float *Dx, float *Hi, float *Hj,
#endif
}
#ifdef WITH_VECTORIZATION
/**
* @brief Density interaction computed using 2 interleaved vectors
* (non-symmetric vectorized version).
*/
__attribute__
((
always_inline
))
INLINE
static
void
runner_iact_nonsym_1_vec_density
(
float
*
R2
,
float
*
Dx
,
float
*
Dy
,
float
*
Dz
,
vector
hi_inv
,
vector
vix
,
vector
viy
,
vector
viz
,
float
*
Vjx
,
float
*
Vjy
,
float
*
Vjz
,
float
*
Mj
,
vector
*
rhoSum
,
vector
*
rho_dhSum
,
vector
*
wcountSum
,
vector
*
wcount_dhSum
,
vector
*
div_vSum
,
vector
*
curlvxSum
,
vector
*
curlvySum
,
vector
*
curlvzSum
,
vector
mask
,
vector
mask2
,
int
knlMask
,
int
knlMask2
)
{
vector
r
,
ri
,
r2
,
xi
,
wi
,
wi_dx
;
vector
mj
;
vector
dx
,
dy
,
dz
,
dvx
,
dvy
,
dvz
;
vector
vjx
,
vjy
,
vjz
;
vector
dvdr
;
vector
curlvrx
,
curlvry
,
curlvrz
;
/* Fill the vectors. */
mj
.
v
=
vec_load
(
Mj
);
vjx
.
v
=
vec_load
(
Vjx
);
vjy
.
v
=
vec_load
(
Vjy
);
vjz
.
v
=
vec_load
(
Vjz
);
dx
.
v
=
vec_load
(
Dx
);
dy
.
v
=
vec_load
(
Dy
);
dz
.
v
=
vec_load
(
Dz
);
/* Get the radius and inverse radius. */
r2
.
v
=
vec_load
(
R2
);
ri
=
vec_reciprocal_sqrt
(
r2
);
r
.
v
=
vec_mul
(
r2
.
v
,
ri
.
v
);
xi
.
v
=
vec_mul
(
r
.
v
,
hi_inv
.
v
);
/* Calculate the kernel for two particles. */
kernel_deval_1_vec
(
&
xi
,
&
wi
,
&
wi_dx
);
/* Compute dv. */
dvx
.
v
=
vec_sub
(
vix
.
v
,
vjx
.
v
);
dvy
.
v
=
vec_sub
(
viy
.
v
,
vjy
.
v
);
dvz
.
v
=
vec_sub
(
viz
.
v
,
vjz
.
v
);
/* Compute dv dot r */
dvdr
.
v
=
vec_fma
(
dvx
.
v
,
dx
.
v
,
vec_fma
(
dvy
.
v
,
dy
.
v
,
vec_mul
(
dvz
.
v
,
dz
.
v
)));
dvdr
.
v
=
vec_mul
(
dvdr
.
v
,
ri
.
v
);
/* Compute dv cross r */
curlvrx
.
v
=
vec_fma
(
dvy
.
v
,
dz
.
v
,
vec_mul
(
vec_set1
(
-
1
.
0
f
),
vec_mul
(
dvz
.
v
,
dy
.
v
)));
curlvry
.
v
=
vec_fma
(
dvz
.
v
,
dx
.
v
,
vec_mul
(
vec_set1
(
-
1
.
0
f
),
vec_mul
(
dvx
.
v
,
dz
.
v
)));
curlvrz
.
v
=
vec_fma
(
dvx
.
v
,
dy
.
v
,
vec_mul
(
vec_set1
(
-
1
.
0
f
),
vec_mul
(
dvy
.
v
,
dx
.
v
)));
curlvrx
.
v
=
vec_mul
(
curlvrx
.
v
,
ri
.
v
);
curlvry
.
v
=
vec_mul
(
curlvry
.
v
,
ri
.
v
);
curlvrz
.
v
=
vec_mul
(
curlvrz
.
v
,
ri
.
v
);
/* Mask updates to intermediate vector sums for particle pi. */
#ifdef HAVE_AVX512_F
rhoSum
->
v
=
_mm512_mask_add_ps
(
rhoSum
->
v
,
knlMask
,
vec_mul
(
mj
.
v
,
wi
.
v
),
rhoSum
->
v
);
rho_dhSum
->
v
=
_mm512_mask_sub_ps
(
rho_dhSum
->
v
,
knlMask
,
rho_dhSum
->
v
,
vec_mul
(
mj
.
v
,
vec_fma
(
vec_set1
(
hydro_dimension
),
wi
.
v
,
vec_mul
(
xi
.
v
,
wi_dx
.
v
))));
wcountSum
->
v
=
_mm512_mask_add_ps
(
wcountSum
->
v
,
knlMask
,
wi
.
v
,
wcountSum
->
v
);
wcount_dhSum
->
v
=
_mm512_mask_sub_ps
(
wcount_dhSum
->
v
,
knlMask
,
wcount_dhSum
->
v
,
vec_mul
(
xi
.
v
,
wi_dx
.
v
));
div_vSum
->
v
=
_mm512_mask_sub_ps
(
div_vSum
->
v
,
knlMask
,
div_vSum
->
v
,
vec_mul
(
mj
.
v
,
vec_mul
(
dvdr
.
v
,
wi_dx
.
v
)));
curlvxSum
->
v
=
_mm512_mask_add_ps
(
curlvxSum
->
v
,
knlMask
,
vec_mul
(
mj
.
v
,
vec_mul
(
curlvrx
.
v
,
wi_dx
.
v
)),
curlvxSum
->
v
);
curlvySum
->
v
=
_mm512_mask_add_ps
(
curlvySum
->
v
,
knlMask
,
vec_mul
(
mj
.
v
,
vec_mul
(
curlvry
.
v
,
wi_dx
.
v
)),
curlvySum
->
v
);
curlvzSum
->
v
=
_mm512_mask_add_ps
(
curlvzSum
->
v
,
knlMask
,
vec_mul
(
mj
.
v
,
vec_mul
(
curlvrz
.
v
,
wi_dx
.
v
)),
curlvzSum
->
v
);
#else
rhoSum
->
v
+=
vec_and
(
vec_mul
(
mj
.
v
,
wi
.
v
),
mask
.
v
);
rho_dhSum
->
v
-=
vec_and
(
vec_mul
(
mj
.
v
,
vec_fma
(
vec_set1
(
hydro_dimension
),
wi
.
v
,
vec_mul
(
xi
.
v
,
wi_dx
.
v
))),
mask
.
v
);
wcountSum
->
v
+=
vec_and
(
wi
.
v
,
mask
.
v
);
wcount_dhSum
->
v
-=
vec_and
(
vec_mul
(
xi
.
v
,
wi_dx
.
v
),
mask
.
v
);
div_vSum
->
v
-=
vec_and
(
vec_mul
(
mj
.
v
,
vec_mul
(
dvdr
.
v
,
wi_dx
.
v
)),
mask
.
v
);
curlvxSum
->
v
+=
vec_and
(
vec_mul
(
mj
.
v
,
vec_mul
(
curlvrx
.
v
,
wi_dx
.
v
)),
mask
.
v
);
curlvySum
->
v
+=
vec_and
(
vec_mul
(
mj
.
v
,
vec_mul
(
curlvry
.
v
,
wi_dx
.
v
)),
mask
.
v
);
curlvzSum
->
v
+=
vec_and
(
vec_mul
(
mj
.
v
,
vec_mul
(
curlvrz
.
v
,
wi_dx
.
v
)),
mask
.
v
);
#endif
}
#endif
#ifdef WITH_VECTORIZATION
/**
* @brief Density interaction computed using 2 interleaved vectors
...
...
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment