Skip to content
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

Jump to bottom

Fix GPU RegTest #1171

Merged
merged 1 commit into from
Jul 19, 2023
Merged

Fix GPU RegTest #1171

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
85 changes: 8 additions & 77 deletions Source/TimeIntegration/ERF_MRI.H
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,50 +210,15 @@ public:
/**********************************************/
/* RK3 Integration with Acoustic Sub-stepping */
/**********************************************/

// Start with S_new (aka S_stage) holding S_old
#ifdef _OPENMP
#pragma omp parallel if (amrex::Gpu::notInLaunchRegion())
#endif
Vector<int> num_vars = {Cons::NumVars, 1, 1, 1};
for (int i(0); i<n_data; ++i)
{
for ( MFIter mfi(S_old[IntVar::cons],TilingIfNotGPU()); mfi.isValid(); ++mfi) {
const Box gbx = mfi.tilebox().grow(S_old[IntVar::cons].nGrowVect());
const Box gtbx = mfi.nodaltilebox(0).grow(S_old[IntVar::xmom].nGrowVect());
const Box gtby = mfi.nodaltilebox(1).grow(S_old[IntVar::ymom].nGrowVect());
const Box gtbz = mfi.nodaltilebox(2).grow(S_old[IntVar::zmom].nGrowVect());

Vector<Array4<Real> > sold_h(n_data);
Vector<Array4<Real> > snew_h(n_data);

for (int i = 0; i < n_data; ++i) {
sold_h[i] = S_old[i].array(mfi);
snew_h[i] = S_new[i].array(mfi);
}

Gpu::AsyncVector<Array4<Real> > sold_d(n_data);
Gpu::AsyncVector<Array4<Real> > snew_d(n_data);

Gpu::copy(Gpu::hostToDevice, sold_h.begin(), sold_h.end(), sold_d.begin());
Gpu::copy(Gpu::hostToDevice, snew_h.begin(), snew_h.end(), snew_d.begin());

Array4<Real>* sold = sold_d.dataPtr();
Array4<Real>* snew = snew_d.dataPtr();

ParallelFor(gbx, static_cast<int>(Cons::NumVars),
[=] AMREX_GPU_DEVICE (int i, int j, int k, int n) {
snew[IntVar::cons](i,j,k,n) = sold[IntVar::cons](i,j,k,n);
});

ParallelFor(gtbx, gtby, gtbz,
[=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
snew[IntVar::xmom](i,j,k) = sold[IntVar::xmom](i,j,k);
},
[=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
snew[IntVar::ymom](i,j,k) = sold[IntVar::ymom](i,j,k);
},
[=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
snew[IntVar::zmom](i,j,k) = sold[IntVar::zmom](i,j,k);
});
// Copy old -> new
MultiFab::Copy(S_new[i],S_old[i],0,0,num_vars[i],S_old[i].nGrowVect());

// Copy old momentum -> scratch momentum
if (i>=1) {
MultiFab::Copy((*S_scratch)[i],S_old[i],0,0,num_vars[i],S_old[i].nGrowVect());
}
}

Expand All @@ -263,40 +228,6 @@ public:
// How many timesteps taken by the fast integrator
int nsubsteps;

// We copy (rho theta) and the velocities here -- the velocities
// will be over-written in slow_rhs on all valid faces but we
// use this copy to fill in the ghost locations which will
// be needed for metric terms
#ifdef _OPENMP
#pragma omp parallel if (amrex::Gpu::notInLaunchRegion())
#endif
{
for ( MFIter mfi(S_old[IntVar::cons],TilingIfNotGPU()); mfi.isValid(); ++mfi) {
const Box gtbx = mfi.nodaltilebox(0).grow(S_old[IntVar::xmom].nGrowVect());
const Box gtby = mfi.nodaltilebox(1).grow(S_old[IntVar::ymom].nGrowVect());
const Box gtbz = mfi.nodaltilebox(2).grow(S_old[IntVar::zmom].nGrowVect());

const Array4<Real>& scrh_xmom = (*S_scratch)[IntVar::xmom].array(mfi);
const Array4<Real>& scrh_ymom = (*S_scratch)[IntVar::ymom].array(mfi);
const Array4<Real>& scrh_zmom = (*S_scratch)[IntVar::zmom].array(mfi);

const Array4<Real>& sold_xmom = S_old[IntVar::xmom].array(mfi);
const Array4<Real>& sold_ymom = S_old[IntVar::ymom].array(mfi);
const Array4<Real>& sold_zmom = S_old[IntVar::zmom].array(mfi);

ParallelFor(gtbx, gtby, gtbz,
[=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
scrh_xmom(i,j,k) = sold_xmom(i,j,k);
},
[=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
scrh_ymom(i,j,k) = sold_ymom(i,j,k);
},
[=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
scrh_zmom(i,j,k) = sold_zmom(i,j,k);
});
}
}

// This is the final time of the full timestep (also the 3rd RK stage)
// Real new_time = time + timestep;

Expand Down
Loading