diff options
Diffstat (limited to 'Source/Parallelization')
| -rw-r--r-- | Source/Parallelization/WarpXComm.cpp | 100 | ||||
| -rw-r--r-- | Source/Parallelization/WarpXComm_K.H | 83 |
2 files changed, 62 insertions, 121 deletions
diff --git a/Source/Parallelization/WarpXComm.cpp b/Source/Parallelization/WarpXComm.cpp index 9c6d3e9a4..ee12a9d68 100644 --- a/Source/Parallelization/WarpXComm.cpp +++ b/Source/Parallelization/WarpXComm.cpp @@ -113,52 +113,36 @@ WarpX::UpdateAuxilaryDataStagToNodal () // for out-of-bound accesses due to large-stencil operations) Box bx = mfi.growntilebox(); - if (maxwell_solver_id == MaxwellSolverAlgo::PSATD) { + // Order of finite-order centering of fields + const int fg_nox = WarpX::field_centering_nox; + const int fg_noy = WarpX::field_centering_noy; + const int fg_noz = WarpX::field_centering_noz; -#ifdef WARPX_USE_PSATD + // Device vectors of stencil coefficients used for finite-order centering of fields + amrex::Real const * stencil_coeffs_x = WarpX::device_field_centering_stencil_coeffs_x.data(); + amrex::Real const * stencil_coeffs_y = WarpX::device_field_centering_stencil_coeffs_y.data(); + amrex::Real const * stencil_coeffs_z = WarpX::device_field_centering_stencil_coeffs_z.data(); - // Order of finite-order centering of fields - const int fg_nox = WarpX::field_centering_nox; - const int fg_noy = WarpX::field_centering_noy; - const int fg_noz = WarpX::field_centering_noz; - - // Device vectors of stencil coefficients used for finite-order centering of fields - amrex::Real const * stencil_coeffs_x = WarpX::device_field_centering_stencil_coeffs_x.data(); - amrex::Real const * stencil_coeffs_y = WarpX::device_field_centering_stencil_coeffs_y.data(); - amrex::Real const * stencil_coeffs_z = WarpX::device_field_centering_stencil_coeffs_z.data(); - - amrex::ParallelFor(bx, [=] AMREX_GPU_DEVICE (int j, int k, int l) noexcept - { - warpx_interp<true>(j, k, l, bx_aux, bx_fp, dst_stag, Bx_stag, fg_nox, fg_noy, fg_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + amrex::ParallelFor(bx, [=] AMREX_GPU_DEVICE (int j, int k, int l) noexcept + { + warpx_interp(j, k, l, bx_aux, bx_fp, dst_stag, Bx_stag, fg_nox, fg_noy, fg_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - warpx_interp<true>(j, k, l, by_aux, by_fp, dst_stag, By_stag, fg_nox, fg_noy, fg_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + warpx_interp(j, k, l, by_aux, by_fp, dst_stag, By_stag, fg_nox, fg_noy, fg_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - warpx_interp<true>(j, k, l, bz_aux, bz_fp, dst_stag, Bz_stag, fg_nox, fg_noy, fg_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + warpx_interp(j, k, l, bz_aux, bz_fp, dst_stag, Bz_stag, fg_nox, fg_noy, fg_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - warpx_interp<true>(j, k, l, ex_aux, ex_fp, dst_stag, Ex_stag, fg_nox, fg_noy, fg_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + warpx_interp(j, k, l, ex_aux, ex_fp, dst_stag, Ex_stag, fg_nox, fg_noy, fg_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - warpx_interp<true>(j, k, l, ey_aux, ey_fp, dst_stag, Ey_stag, fg_nox, fg_noy, fg_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + warpx_interp(j, k, l, ey_aux, ey_fp, dst_stag, Ey_stag, fg_nox, fg_noy, fg_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - warpx_interp<true>(j, k, l, ez_aux, ez_fp, dst_stag, Ez_stag, fg_nox, fg_noy, fg_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - }); -#endif - } else { // FDTD - amrex::ParallelFor(bx, [=] AMREX_GPU_DEVICE (int j, int k, int l) noexcept - { - warpx_interp(j, k, l, bx_aux, bx_fp, dst_stag, Bx_stag); - warpx_interp(j, k, l, by_aux, by_fp, dst_stag, By_stag); - warpx_interp(j, k, l, bz_aux, bz_fp, dst_stag, Bz_stag); - warpx_interp(j, k, l, ex_aux, ex_fp, dst_stag, Ex_stag); - warpx_interp(j, k, l, ey_aux, ey_fp, dst_stag, Ey_stag); - warpx_interp(j, k, l, ez_aux, ez_fp, dst_stag, Ez_stag); - }); - } + warpx_interp(j, k, l, ez_aux, ez_fp, dst_stag, Ez_stag, fg_nox, fg_noy, fg_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + }); } // NOTE: high-order interpolation is not implemented for mesh refinement @@ -453,35 +437,21 @@ void WarpX::UpdateCurrentNodalToStag (amrex::MultiFab& dst, amrex::MultiFab cons amrex::Array4<amrex::Real const> const& src_arr = src.const_array(mfi); amrex::Array4<amrex::Real> const& dst_arr = dst.array(mfi); - if (maxwell_solver_id == MaxwellSolverAlgo::PSATD) - { -#ifdef WARPX_USE_PSATD + // Order of finite-order centering of currents + const int cc_nox = WarpX::current_centering_nox; + const int cc_noy = WarpX::current_centering_noy; + const int cc_noz = WarpX::current_centering_noz; - // Order of finite-order centering of currents - const int cc_nox = WarpX::current_centering_nox; - const int cc_noy = WarpX::current_centering_noy; - const int cc_noz = WarpX::current_centering_noz; + // Device vectors of stencil coefficients used for finite-order centering of currents + amrex::Real const * stencil_coeffs_x = WarpX::device_current_centering_stencil_coeffs_x.data(); + amrex::Real const * stencil_coeffs_y = WarpX::device_current_centering_stencil_coeffs_y.data(); + amrex::Real const * stencil_coeffs_z = WarpX::device_current_centering_stencil_coeffs_z.data(); - // Device vectors of stencil coefficients used for finite-order centering of currents - amrex::Real const * stencil_coeffs_x = WarpX::device_current_centering_stencil_coeffs_x.data(); - amrex::Real const * stencil_coeffs_y = WarpX::device_current_centering_stencil_coeffs_y.data(); - amrex::Real const * stencil_coeffs_z = WarpX::device_current_centering_stencil_coeffs_z.data(); - - amrex::ParallelFor(bx, [=] AMREX_GPU_DEVICE (int j, int k, int l) noexcept - { - warpx_interp<true>(j, k, l, dst_arr, src_arr, dst_stag, src_stag, cc_nox, cc_noy, cc_noz, - stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); - }); -#endif - } - - else // FDTD + amrex::ParallelFor(bx, [=] AMREX_GPU_DEVICE (int j, int k, int l) noexcept { - amrex::ParallelFor(bx, [=] AMREX_GPU_DEVICE (int j, int k, int l) noexcept - { - warpx_interp<false>(j, k, l, dst_arr, src_arr, dst_stag, src_stag); - }); - } + warpx_interp(j, k, l, dst_arr, src_arr, dst_stag, src_stag, cc_nox, cc_noy, cc_noz, + stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); + }); } } diff --git a/Source/Parallelization/WarpXComm_K.H b/Source/Parallelization/WarpXComm_K.H index 01d90ff66..f3b39bc3e 100644 --- a/Source/Parallelization/WarpXComm_K.H +++ b/Source/Parallelization/WarpXComm_K.H @@ -610,8 +610,7 @@ void warpx_interp_nd_efield_z (int j, int k, int l, /** * \brief Arbitrary-order interpolation function used to center a given MultiFab between two grids - * with different staggerings. With the FDTD solver, this performs simple linear interpolation. - * With the PSATD solver, this performs arbitrary-order interpolation based on the Fornberg coefficients. + * with different staggerings. The arbitrary-order interpolation is based on the Fornberg coefficients. * The result is stored in the output array \c dst_arr. * * \param[in] j index along x of the output array @@ -628,7 +627,6 @@ void warpx_interp_nd_efield_z (int j, int k, int l, * \param[in] stencil_coeffs_y array of ordered Fornberg coefficients for finite-order centering stencil along y * \param[in] stencil_coeffs_z array of ordered Fornberg coefficients for finite-order centering stencil along z */ -template< bool IS_PSATD = false > AMREX_GPU_DEVICE AMREX_FORCE_INLINE void warpx_interp (const int j, const int k, @@ -656,16 +654,10 @@ void warpx_interp (const int j, // Avoid compiler warnings #if defined(WARPX_DIM_1D_Z) amrex::ignore_unused(nox, noy, stencil_coeffs_x, stencil_coeffs_y); -#endif -#if defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) +#elif defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) amrex::ignore_unused(noy, stencil_coeffs_y); #endif - // Avoid compiler warnings -#ifndef WARPX_USE_PSATD - amrex::ignore_unused(stencil_coeffs_x, stencil_coeffs_y, stencil_coeffs_z); -#endif - // If dst_nodal = true , we are centering from a staggered grid to a nodal grid // If dst_nodal = false, we are centering from a nodal grid to a staggered grid const bool dst_nodal = (dst_stag == amrex::IntVect::TheNodeVector()); @@ -693,7 +685,7 @@ void warpx_interp (const int j, #if defined(WARPX_DIM_1D_Z) const int noj = noz; -#elif defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) +#elif defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) const int noj = nox; const int nok = noz; #elif defined(WARPX_DIM_3D) @@ -707,7 +699,7 @@ void warpx_interp (const int j, #if defined(WARPX_DIM_1D_Z) constexpr amrex::Real wk = 1.0_rt; constexpr amrex::Real wl = 1.0_rt; -#elif defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) +#elif defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) const amrex::Real wk = (interp_k) ? 0.5_rt : 1.0_rt; constexpr amrex::Real wl = 1.0_rt; #elif defined(WARPX_DIM_3D) @@ -720,7 +712,7 @@ void warpx_interp (const int j, const int jmax = (interp_j) ? j + noj/2 + shift - 1 : j; // Min and max for interpolation loop along k -#if defined(WARPX_DIM_1D_Z) +#if defined(WARPX_DIM_1D_Z) // k = 0 always const int kmin = k; const int kmax = k; @@ -730,7 +722,7 @@ void warpx_interp (const int j, #endif // Min and max for interpolation loop along l -#if (AMREX_SPACEDIM <= 2) +#if (AMREX_SPACEDIM <= 2) // l = 0 always const int lmin = l; const int lmax = l; @@ -785,59 +777,38 @@ void warpx_interp (const int j, amrex::Real res = 0.0_rt; - if( !IS_PSATD ) // FDTD (linear interpolation) - { - for (int ll = 0; ll <= nl; ll++) - { - for (int kk = 0; kk <= nk; kk++) - { - for (int jj = 0; jj <= nj; jj++) - { - res += src_arr_zeropad(jmin+jj,kmin+kk,lmin+ll); - } - } - } + amrex::Real cj = 1.0_rt; + amrex::Real ck = 1.0_rt; + amrex::Real cl = 1.0_rt; - } - - else // PSATD (finite-order interpolation) - { #if defined(WARPX_DIM_1D_Z) - amrex::ignore_unused(stencil_coeffs_z); - amrex::Abort("PSATD not implemented in 1D"); -#endif - -#if (AMREX_SPACEDIM >= 2) // 1D not implemented for PSATD - amrex::Real cj = 1.0_rt; - amrex::Real ck = 1.0_rt; - amrex::Real cl = 1.0_rt; - - amrex::Real const* scj = stencil_coeffs_x; -#if defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) - amrex::Real const* sck = stencil_coeffs_z; + amrex::Real const* scj = stencil_coeffs_z; +#elif defined(WARPX_DIM_XZ) || defined(WARPX_DIM_RZ) + amrex::Real const* scj = stencil_coeffs_x; + amrex::Real const* sck = stencil_coeffs_z; #elif defined(WARPX_DIM_3D) - amrex::Real const* sck = stencil_coeffs_y; - amrex::Real const* scl = stencil_coeffs_z; + amrex::Real const* scj = stencil_coeffs_x; + amrex::Real const* sck = stencil_coeffs_y; + amrex::Real const* scl = stencil_coeffs_z; #endif - for (int ll = 0; ll <= nl; ll++) - { + for (int ll = 0; ll <= nl; ll++) + { #if defined(WARPX_DIM_3D) - if (interp_l) cl = scl[ll]; + if (interp_l) cl = scl[ll]; +#endif + for (int kk = 0; kk <= nk; kk++) + { +#if (AMREX_SPACEDIM >= 2) + if (interp_k) ck = sck[kk]; #endif - for (int kk = 0; kk <= nk; kk++) + for (int jj = 0; jj <= nj; jj++) { - if (interp_k) ck = sck[kk]; - - for (int jj = 0; jj <= nj; jj++) - { - if (interp_j) cj = scj[jj]; + if (interp_j) cj = scj[jj]; - res += cj * ck * cl * src_arr_zeropad(jmin+jj,kmin+kk,lmin+ll); - } + res += cj * ck * cl * src_arr_zeropad(jmin+jj,kmin+kk,lmin+ll); } } -#endif //1D } dst_arr(j,k,l) = wj * wk * wl * res; |