Skip to content

Vectorize SPHENIX force interactions #78

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
TimGoTheCreator wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Vectorize SPHENIX force interactions #78

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
112 changes: 111 additions & 1 deletion src/cache.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,6 +80,13 @@ struct cache {
/* Particle sound speed. */
float *restrict soundspeed SWIFT_CACHE_ALIGN;

#if defined(SPHENIX_SPH)
/* Extra SPHENIX fields required by its vectorized force interaction. */
float *restrict viscosity_alpha SWIFT_CACHE_ALIGN;
float *restrict diffusion_alpha SWIFT_CACHE_ALIGN;
float *restrict u SWIFT_CACHE_ALIGN;
#endif

/* Cache size. */
int count;
};
Expand Down Expand Up @@ -146,6 +153,11 @@ __attribute__((always_inline)) INLINE void cache_init(struct cache *c,
free(c->pOrho2);
free(c->balsara);
free(c->soundspeed);
#if defined(SPHENIX_SPH)
free(c->viscosity_alpha);
free(c->diffusion_alpha);
free(c->u);
#endif
}

error += posix_memalign((void **)&c->x, SWIFT_CACHE_ALIGNMENT, sizeBytes);
Expand All @@ -167,6 +179,13 @@ __attribute__((always_inline)) INLINE void cache_init(struct cache *c,
posix_memalign((void **)&c->balsara, SWIFT_CACHE_ALIGNMENT, sizeBytes);
error +=
posix_memalign((void **)&c->soundspeed, SWIFT_CACHE_ALIGNMENT, sizeBytes);
#if defined(SPHENIX_SPH)
error += posix_memalign((void **)&c->viscosity_alpha, SWIFT_CACHE_ALIGNMENT,
sizeBytes);
error += posix_memalign((void **)&c->diffusion_alpha, SWIFT_CACHE_ALIGNMENT,
sizeBytes);
error += posix_memalign((void **)&c->u, SWIFT_CACHE_ALIGNMENT, sizeBytes);
#endif

if (error != 0)
error("Couldn't allocate cache, no. of particles: %d", (int)count);
Expand Down Expand Up @@ -495,7 +514,7 @@ __attribute__((always_inline)) INLINE int cache_read_force_particles(
const struct cell *restrict const ci,
struct cache *restrict const ci_cache) {

#if defined(GADGET2_SPH)
#if defined(GADGET2_SPH) || defined(SPHENIX_SPH)

/* Let the compiler know that the data is aligned and create pointers to the
* arrays inside the cache. */
Expand All @@ -516,6 +535,13 @@ __attribute__((always_inline)) INLINE int cache_read_force_particles(
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, soundspeed, ci_cache->soundspeed,
SWIFT_CACHE_ALIGNMENT);
#if defined(SPHENIX_SPH)
swift_declare_aligned_ptr(float, viscosity_alpha, ci_cache->viscosity_alpha,
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, diffusion_alpha, ci_cache->diffusion_alpha,
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, u, ci_cache->u, SWIFT_CACHE_ALIGNMENT);
#endif

const int count = ci->hydro.count;
const struct part *restrict parts = ci->hydro.parts;
Expand All @@ -541,6 +567,15 @@ __attribute__((always_inline)) INLINE int cache_read_force_particles(
pOrho2[i] = 1.f;
balsara[i] = 1.f;
soundspeed[i] = 1.f;
m[i] = 1.f;
vx[i] = 0.f;
vy[i] = 0.f;
vz[i] = 0.f;
#if defined(SPHENIX_SPH)
viscosity_alpha[i] = 0.f;
diffusion_alpha[i] = 0.f;
u[i] = 0.f;
#endif

continue;
}
Expand All @@ -555,7 +590,14 @@ __attribute__((always_inline)) INLINE int cache_read_force_particles(
vz[i] = parts[i].v[2];
rho[i] = parts[i].rho;
grad_h[i] = parts[i].force.f;
#if defined(GADGET2_SPH)
pOrho2[i] = parts[i].force.P_over_rho2;
#else
pOrho2[i] = parts[i].force.pressure;
viscosity_alpha[i] = parts[i].viscosity.alpha;
diffusion_alpha[i] = parts[i].diffusion.alpha;
u[i] = parts[i].u;
#endif
balsara[i] = parts[i].force.balsara;
soundspeed[i] = parts[i].force.soundspeed;
}
Expand All @@ -578,6 +620,15 @@ __attribute__((always_inline)) INLINE int cache_read_force_particles(
pOrho2[i] = 1.f;
balsara[i] = 1.f;
soundspeed[i] = 1.f;
m[i] = 1.f;
vx[i] = 0.f;
vy[i] = 0.f;
vz[i] = 0.f;
#if defined(SPHENIX_SPH)
viscosity_alpha[i] = 0.f;
diffusion_alpha[i] = 0.f;
u[i] = 0.f;
#endif
}
}

Expand Down Expand Up @@ -915,6 +966,13 @@ cache_read_two_partial_cells_sorted_force(
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, soundspeed, ci_cache->soundspeed,
SWIFT_CACHE_ALIGNMENT);
#if defined(SPHENIX_SPH)
swift_declare_aligned_ptr(float, viscosity_alpha, ci_cache->viscosity_alpha,
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, diffusion_alpha, ci_cache->diffusion_alpha,
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, u, ci_cache->u, SWIFT_CACHE_ALIGNMENT);
#endif

int ci_cache_count = ci->hydro.count - first_pi_align;
const double max_dx = max(ci->hydro.dx_max_part, cj->hydro.dx_max_part);
Expand Down Expand Up @@ -944,6 +1002,11 @@ cache_read_two_partial_cells_sorted_force(
pOrho2[i] = 1.f;
balsara[i] = 1.f;
soundspeed[i] = 1.f;
#if defined(SPHENIX_SPH)
viscosity_alpha[i] = 0.f;
diffusion_alpha[i] = 0.f;
u[i] = 0.f;
#endif

continue;
}
Expand All @@ -962,6 +1025,16 @@ cache_read_two_partial_cells_sorted_force(
pOrho2[i] = parts_i[idx].force.P_over_rho2;
balsara[i] = parts_i[idx].force.balsara;
soundspeed[i] = parts_i[idx].force.soundspeed;
#elif defined(SPHENIX_SPH)
m[i] = parts_i[idx].mass;
rho[i] = parts_i[idx].rho;
grad_h[i] = parts_i[idx].force.f;
pOrho2[i] = parts_i[idx].force.pressure;
balsara[i] = parts_i[idx].force.balsara;
soundspeed[i] = parts_i[idx].force.soundspeed;
viscosity_alpha[i] = parts_i[idx].viscosity.alpha;
diffusion_alpha[i] = parts_i[idx].diffusion.alpha;
u[i] = parts_i[idx].u;
#endif
}

Expand All @@ -983,6 +1056,11 @@ cache_read_two_partial_cells_sorted_force(
pOrho2[i] = 1.f;
balsara[i] = 1.f;
soundspeed[i] = 1.f;
#if defined(SPHENIX_SPH)
viscosity_alpha[i] = 0.f;
diffusion_alpha[i] = 0.f;
u[i] = 0.f;
#endif
}

/* Let the compiler know that the data is aligned and create pointers to the
Expand All @@ -1004,6 +1082,13 @@ cache_read_two_partial_cells_sorted_force(
SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, soundspeedj, cj_cache->soundspeed,
SWIFT_CACHE_ALIGNMENT);
#if defined(SPHENIX_SPH)
swift_declare_aligned_ptr(float, viscosity_alphaj,
cj_cache->viscosity_alpha, SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, diffusion_alphaj,
cj_cache->diffusion_alpha, SWIFT_CACHE_ALIGNMENT);
swift_declare_aligned_ptr(float, uj, cj_cache->u, SWIFT_CACHE_ALIGNMENT);
#endif

const float pos_padded_j[3] = {-(2. * cj->width[0] + max_dx),
-(2. * cj->width[1] + max_dx),
Expand All @@ -1028,6 +1113,11 @@ cache_read_two_partial_cells_sorted_force(
pOrho2j[i] = 1.f;
balsaraj[i] = 1.f;
soundspeedj[i] = 1.f;
#if defined(SPHENIX_SPH)
viscosity_alphaj[i] = 0.f;
diffusion_alphaj[i] = 0.f;
uj[i] = 0.f;
#endif

continue;
}
Expand All @@ -1046,6 +1136,16 @@ cache_read_two_partial_cells_sorted_force(
pOrho2j[i] = parts_j[idx].force.P_over_rho2;
balsaraj[i] = parts_j[idx].force.balsara;
soundspeedj[i] = parts_j[idx].force.soundspeed;
#elif defined(SPHENIX_SPH)
mj[i] = parts_j[idx].mass;
rhoj[i] = parts_j[idx].rho;
grad_hj[i] = parts_j[idx].force.f;
pOrho2j[i] = parts_j[idx].force.pressure;
balsaraj[i] = parts_j[idx].force.balsara;
soundspeedj[i] = parts_j[idx].force.soundspeed;
viscosity_alphaj[i] = parts_j[idx].viscosity.alpha;
diffusion_alphaj[i] = parts_j[idx].diffusion.alpha;
uj[i] = parts_j[idx].u;
#endif
}

Expand All @@ -1066,6 +1166,11 @@ cache_read_two_partial_cells_sorted_force(
pOrho2j[i] = 1.f;
balsaraj[i] = 1.f;
soundspeedj[i] = 1.f;
#if defined(SPHENIX_SPH)
viscosity_alphaj[i] = 0.f;
diffusion_alphaj[i] = 0.f;
uj[i] = 0.f;
#endif
}
}

Expand All @@ -1090,6 +1195,11 @@ static INLINE void cache_clean(struct cache *c) {
free(c->pOrho2);
free(c->balsara);
free(c->soundspeed);
#if defined(SPHENIX_SPH)
free(c->viscosity_alpha);
free(c->diffusion_alpha);
free(c->u);
#endif
}
c->count = 0;
}
Expand Down
Loading