CartesianRender.cpp

Go to the documentation of this file.

// -------------------------------------------------------//
//
// SHAMROCK code for hydrodynamics
// Copyright (c) 2021-2026 Timothée David--Cléris <tim.shamrock@proton.me>
// SPDX-License-Identifier: CeCILL Free Software License Agreement v2.1
// Shamrock is licensed under the CeCILL 2.1 License, see LICENSE for more information
//
// -------------------------------------------------------//
 
#include "shambase/exception.hpp"
#include "shambackends/kernel_call.hpp"
#include "shammath/AABB.hpp"
#include "shammodels/sph/math/density.hpp"
#include "shammodels/sph/modules/render/CartesianRender.hpp"
#include "shammodels/sph/modules/render/RenderFieldGetter.hpp"
#include "shamrock/scheduler/SchedulerUtility.hpp"
 
namespace shammodels::sph::modules {
 
    template<class Tvec>
    sham::DeviceBuffer<Tvec> pixel_to_positions(
        Tvec center, Tvec delta_x, Tvec delta_y, u32 nx, u32 ny) {
 
        sham::DeviceBuffer<Tvec> ret{nx * ny, shamsys::instance::get_compute_scheduler_ptr()};
 
        sham::DeviceQueue &q = shamsys::instance::get_compute_scheduler().get_queue();
 
        sham::kernel_call(
            q, sham::MultiRef{}, sham::MultiRef{ret}, nx * ny, [=](u32 gid, Tvec *position) {
                u32 ix        = gid % nx;
                u32 iy        = gid / nx;
                f64 fx        = ((f64(ix) + 0.5) / nx) - 0.5;
                f64 fy        = ((f64(iy) + 0.5) / ny) - 0.5;
                position[gid] = center + delta_x * fx + delta_y * fy;
            });
 
        return ret;
    }
 
    template<class Tvec>
    sham::DeviceBuffer<shammath::Ray<Tvec>> pixel_to_orthographic_rays(
        Tvec center, Tvec delta_x, Tvec delta_y, u32 nx, u32 ny) {
 
        using Tscal = shambase::VecComponent<Tvec>;
 
        sham::DeviceBuffer<shammath::Ray<Tvec>> ret{
            nx * ny, shamsys::instance::get_compute_scheduler_ptr()};
 
        sham::DeviceQueue &q = shamsys::instance::get_compute_scheduler().get_queue();
 
        Tvec e_z  = sycl::cross(delta_x, delta_y);
        Tscal len = sycl::length(e_z);
        if (!(len > 0)) {
            throw shambase::make_except_with_loc<std::invalid_argument>(shambase::format(
                "The cross product of delta_x and delta_y is zero\n"
                "  args :"
                "    center  = {}\n"
                "    delta_x = {}\n"
                "    delta_y = {}\n"
                "    nx      = {}\n"
                "    ny      = {}\n"
                "  -> e_z = {}\n",
                center,
                delta_x,
                delta_y,
                nx,
                ny,
                e_z));
        }
        e_z /= len;
 
        sham::kernel_call(
            q,
            sham::MultiRef{},
            sham::MultiRef{ret},
            nx * ny,
            [=](u32 gid, shammath::Ray<Tvec> *ray) {
                u32 ix          = gid % nx;
                u32 iy          = gid / nx;
                f64 fx          = ((f64(ix) + 0.5) / nx) - 0.5;
                f64 fy          = ((f64(iy) + 0.5) / ny) - 0.5;
                Tvec pos_render = center + delta_x * fx + delta_y * fy;
 
                ray[gid] = shammath::Ray<Tvec>(pos_render, e_z);
            });
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_slice(
        std::string field_name,
        const sham::DeviceBuffer<Tvec> &positions,
        std::optional<std::function<py::array_t<Tfield>(size_t, pybind11::dict &)>> custom_getter)
        -> sham::DeviceBuffer<Tfield> {
 
        if (shamcomm::world_rank() == 0) {
            logger::info_ln(
                "sph::CartesianRender",
                shambase::format(
                    "compute_slice field_name: {}, positions count: {}",
                    field_name,
                    positions.get_size()));
        }
 
        shambase::Timer t;
        t.start();
 
        auto ret = RenderFieldGetter<Tvec, Tfield, SPHKernel>(context, solver_config, storage)
                       .runner_function(
                           field_name,
                           [&](auto field_getter) -> sham::DeviceBuffer<Tfield> {
                               return compute_slice(field_getter, positions);
                           },
                           custom_getter);
 
        t.end();
        if (shamcomm::world_rank() == 0) {
            logger::info_ln(
                "sph::CartesianRender",
                shambase::format("compute_slice took {}", t.get_time_str()));
        }
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_column_integ(
        std::string field_name,
        const sham::DeviceBuffer<shammath::Ray<Tvec>> &rays,
        std::optional<std::function<py::array_t<Tfield>(size_t, pybind11::dict &)>> custom_getter)
        -> sham::DeviceBuffer<Tfield> {
 
        if (shamcomm::world_rank() == 0) {
            logger::info_ln(
                "sph::CartesianRender",
                shambase::format(
                    "compute_column_integ field_name: {}, rays count: {}",
                    field_name,
                    rays.get_size()));
        }
 
        shambase::Timer t;
        t.start();
 
        auto ret = RenderFieldGetter<Tvec, Tfield, SPHKernel>(context, solver_config, storage)
                       .runner_function(
                           field_name,
                           [&](auto field_getter) -> sham::DeviceBuffer<Tfield> {
                               return compute_column_integ(field_getter, rays);
                           },
                           custom_getter);
 
        t.end();
        if (shamcomm::world_rank() == 0) {
            logger::info_ln(
                "sph::CartesianRender",
                shambase::format("compute_column_integ took {}", t.get_time_str()));
        }
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_azymuthal_integ(
        std::string field_name,
        const sham::DeviceBuffer<shammath::RingRay<Tvec>> &ring_rays,
        std::optional<std::function<py::array_t<Tfield>(size_t, pybind11::dict &)>> custom_getter)
        -> sham::DeviceBuffer<Tfield> {
 
        if (shamcomm::world_rank() == 0) {
            logger::info_ln(
                "sph::CartesianRender",
                shambase::format(
                    "compute_azymuthal_integ field_name: {}, ring_rays count: {}",
                    field_name,
                    ring_rays.get_size()));
        }
 
        shambase::Timer t;
        t.start();
 
        auto ret = RenderFieldGetter<Tvec, Tfield, SPHKernel>(context, solver_config, storage)
                       .runner_function(
                           field_name,
                           [&](auto field_getter) -> sham::DeviceBuffer<Tfield> {
                               return compute_azymuthal_integ(field_getter, ring_rays);
                           },
                           custom_getter);
 
        t.end();
        if (shamcomm::world_rank() == 0) {
            logger::info_ln(
                "sph::CartesianRender",
                shambase::format("compute_azymuthal_integ took {}", t.get_time_str()));
        }
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_slice(
        std::function<field_getter_t> field_getter, const sham::DeviceBuffer<Tvec> &positions)
        -> sham::DeviceBuffer<Tfield> {
 
        sham::DeviceBuffer<Tfield> ret{
            positions.get_size(), shamsys::instance::get_compute_scheduler_ptr()};
        ret.fill(sham::VectorProperties<Tfield>::get_zero());
 
        using u_morton = u32;
        using RTree    = RadixTree<u_morton, Tvec>;
 
        shamrock::patch::PatchCoordTransform<Tvec> transf
            = scheduler().get_sim_box().template get_patch_transform<Tvec>();
 
        scheduler().for_each_patchdata_nonempty([&](const shamrock::patch::Patch cur_p,
                                                    shamrock::patch::PatchDataLayer &pdat) {
            shammath::CoordRange<Tvec> box = transf.to_obj_coord(cur_p);
 
            PatchDataField<Tvec> &main_field = pdat.get_field<Tvec>(0);
 
            auto &buf_xyz = pdat.get_field<Tvec>(0).get_buf();
            auto &buf_hpart
                = pdat.get_field<Tscal>(pdat.pdl().get_field_idx<Tscal>("hpart")).get_buf();
 
            auto &buf_field_to_render = field_getter(cur_p, pdat);
 
            u32 obj_cnt = main_field.get_obj_cnt();
 
            RTree tree(
                shamsys::instance::get_compute_scheduler_ptr(),
                {box.lower, box.upper},
                buf_xyz,
                obj_cnt,
                solver_config.tree_reduction_level);
 
            tree.compute_cell_ibounding_box(shamsys::instance::get_compute_queue());
            tree.convert_bounding_box(shamsys::instance::get_compute_queue());
 
            RadixTreeField<Tscal> hmax_tree = tree.compute_int_boxes(
                shamsys::instance::get_compute_queue(),
                pdat.get_field<Tscal>(pdat.pdl().get_field_idx<Tscal>("hpart")).get_buf(),
                1);
 
            sham::DeviceQueue &q = shamsys::instance::get_compute_scheduler().get_queue();
 
            sham::EventList depends_list;
            Tfield *render_field = ret.get_write_access(depends_list);
 
            const Tvec *pixel_positions = positions.get_read_access(depends_list);
 
            auto xyz      = buf_xyz.get_read_access(depends_list);
            auto hpart    = buf_hpart.get_read_access(depends_list);
            auto torender = buf_field_to_render.get_read_access(depends_list);
 
            sycl::event e2 = q.submit(depends_list, [&, render_field](sycl::handler &cgh) {
                shamrock::tree::ObjectIterator particle_looper(tree, cgh);
 
                sycl::accessor hmax{
                    shambase::get_check_ref(hmax_tree.radix_tree_field_buf), cgh, sycl::read_only};
 
                constexpr Tscal Rker2 = Kernel::Rkern * Kernel::Rkern;
 
                Tscal partmass = solver_config.gpart_mass;
 
                shambase::parallel_for(
                    cgh, positions.get_size(), "compute slice render", [=](u32 gid) {
                        Tvec pos_render = pixel_positions[gid];
 
                        Tfield ret = sham::VectorProperties<Tfield>::get_zero();
 
                        particle_looper.rtree_for(
                            [&](u32 node_id, Tvec bmin, Tvec bmax) -> bool {
                                Tscal rint_cell = hmax[node_id] * Kernel::Rkern;
 
                                auto interbox
                                    = shammath::CoordRange<Tvec>{bmin, bmax}.expand_all(rint_cell);
 
                                return interbox.contain_pos(pos_render);
                            },
                            [&](u32 id_b) {
                                Tvec dr    = pos_render - xyz[id_b];
                                Tscal rab2 = sycl::dot(dr, dr);
                                Tscal h_b  = hpart[id_b];
 
                                if (rab2 > h_b * h_b * Rker2) {
                                    return;
                                }
 
                                Tscal rab = sycl::sqrt(rab2);
 
                                Tfield val = torender[id_b];
 
                                Tscal rho_b = shamrock::sph::rho_h(partmass, h_b, Kernel::hfactd);
 
                                ret += partmass * val * Kernel::W_3d(rab, h_b) / rho_b;
                            });
 
                        render_field[gid] += ret;
                    });
            });
 
            buf_xyz.complete_event_state(e2);
            buf_hpart.complete_event_state(e2);
            buf_field_to_render.complete_event_state(e2);
            ret.complete_event_state(e2);
            positions.complete_event_state(e2);
        });
 
        shamalgs::collective::reduce_buffer_in_place_sum(ret, MPI_COMM_WORLD);
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_column_integ(
        std::function<field_getter_t> field_getter,
        const sham::DeviceBuffer<shammath::Ray<Tvec>> &rays) -> sham::DeviceBuffer<Tfield> {
 
        sham::DeviceBuffer<Tfield> ret{
            rays.get_size(), shamsys::instance::get_compute_scheduler_ptr()};
        ret.fill(sham::VectorProperties<Tfield>::get_zero());
 
        using u_morton = u32;
        using RTree    = RadixTree<u_morton, Tvec>;
 
        shamrock::patch::PatchCoordTransform<Tvec> transf
            = scheduler().get_sim_box().template get_patch_transform<Tvec>();
 
        scheduler().for_each_patchdata_nonempty([&](const shamrock::patch::Patch cur_p,
                                                    shamrock::patch::PatchDataLayer &pdat) {
            shammath::CoordRange<Tvec> box = transf.to_obj_coord(cur_p);
 
            PatchDataField<Tvec> &main_field = pdat.get_field<Tvec>(0);
 
            auto &buf_xyz = pdat.get_field<Tvec>(0).get_buf();
            auto &buf_hpart
                = pdat.get_field<Tscal>(pdat.pdl().get_field_idx<Tscal>("hpart")).get_buf();
 
            auto &buf_field_to_render = field_getter(cur_p, pdat);
 
            u32 obj_cnt = main_field.get_obj_cnt();
 
            RTree tree(
                shamsys::instance::get_compute_scheduler_ptr(),
                {box.lower, box.upper},
                buf_xyz,
                obj_cnt,
                solver_config.tree_reduction_level);
 
            tree.compute_cell_ibounding_box(shamsys::instance::get_compute_queue());
            tree.convert_bounding_box(shamsys::instance::get_compute_queue());
 
            RadixTreeField<Tscal> hmax_tree = tree.compute_int_boxes(
                shamsys::instance::get_compute_queue(),
                pdat.get_field<Tscal>(pdat.pdl().get_field_idx<Tscal>("hpart")).get_buf(),
                1);
 
            sham::DeviceQueue &q = shamsys::instance::get_compute_scheduler().get_queue();
 
            sham::EventList depends_list;
            Tfield *render_field = ret.get_write_access(depends_list);
 
            const shammath::Ray<Tvec> *image_rays = rays.get_read_access(depends_list);
 
            auto xyz      = buf_xyz.get_read_access(depends_list);
            auto hpart    = buf_hpart.get_read_access(depends_list);
            auto torender = buf_field_to_render.get_read_access(depends_list);
 
            sycl::event e2 = q.submit(depends_list, [&, render_field](sycl::handler &cgh) {
                shamrock::tree::ObjectIterator particle_looper(tree, cgh);
 
                sycl::accessor hmax{
                    shambase::get_check_ref(hmax_tree.radix_tree_field_buf), cgh, sycl::read_only};
 
                constexpr Tscal Rker2 = Kernel::Rkern * Kernel::Rkern;
 
                Tscal partmass = solver_config.gpart_mass;
 
                shambase::parallel_for(cgh, rays.get_size(), "compute slice render", [=](u32 gid) {
                    Tfield ret = sham::VectorProperties<Tfield>::get_zero();
 
                    shammath::Ray<Tvec> ray = image_rays[gid];
 
                    particle_looper.rtree_for(
                        [&](u32 node_id, Tvec bmin, Tvec bmax) -> bool {
                            Tscal rint_cell = hmax[node_id] * Kernel::Rkern;
 
                            auto interbox = shammath::AABB<Tvec>{bmin, bmax}.expand_all(rint_cell);
 
                            return interbox.intersect_ray(ray);
                        },
                        [&](u32 id_b) {
                            Tvec dr = ray.origin - xyz[id_b];
 
                            dr -= ray.direction * sycl::dot(dr, ray.direction);
 
                            Tscal rab2 = sycl::dot(dr, dr);
                            Tscal h_b  = hpart[id_b];
 
                            if (rab2 > h_b * h_b * Rker2) {
                                return;
                            }
 
                            Tscal rab = sycl::sqrt(rab2);
 
                            Tfield val = torender[id_b];
 
                            Tscal rho_b = shamrock::sph::rho_h(partmass, h_b, Kernel::hfactd);
 
                            ret += partmass * val * Kernel::Y_3d(rab, h_b, 4) / rho_b;
                        });
 
                    render_field[gid] += ret;
                });
            });
 
            buf_xyz.complete_event_state(e2);
            buf_hpart.complete_event_state(e2);
            buf_field_to_render.complete_event_state(e2);
            ret.complete_event_state(e2);
            rays.complete_event_state(e2);
        });
 
        shamalgs::collective::reduce_buffer_in_place_sum(ret, MPI_COMM_WORLD);
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_azymuthal_integ(
        std::function<field_getter_t> field_getter,
        const sham::DeviceBuffer<shammath::RingRay<Tvec>> &ring_rays)
        -> sham::DeviceBuffer<Tfield> {
 
        sham::DeviceBuffer<Tfield> ret{
            ring_rays.get_size(), shamsys::instance::get_compute_scheduler_ptr()};
        ret.fill(sham::VectorProperties<Tfield>::get_zero());
 
        using u_morton = u32;
        using RTree    = RadixTree<u_morton, Tvec>;
 
        shamrock::patch::PatchCoordTransform<Tvec> transf
            = scheduler().get_sim_box().template get_patch_transform<Tvec>();
 
        scheduler().for_each_patchdata_nonempty([&](const shamrock::patch::Patch cur_p,
                                                    shamrock::patch::PatchDataLayer &pdat) {
            shammath::CoordRange<Tvec> box = transf.to_obj_coord(cur_p);
 
            PatchDataField<Tvec> &main_field = pdat.get_field<Tvec>(0);
 
            auto &buf_xyz = pdat.get_field<Tvec>(0).get_buf();
            auto &buf_hpart
                = pdat.get_field<Tscal>(pdat.pdl().get_field_idx<Tscal>("hpart")).get_buf();
 
            auto &buf_field_to_render = field_getter(cur_p, pdat);
 
            u32 obj_cnt = main_field.get_obj_cnt();
 
            RTree tree(
                shamsys::instance::get_compute_scheduler_ptr(),
                {box.lower, box.upper},
                buf_xyz,
                obj_cnt,
                solver_config.tree_reduction_level);
 
            tree.compute_cell_ibounding_box(shamsys::instance::get_compute_queue());
            tree.convert_bounding_box(shamsys::instance::get_compute_queue());
 
            RadixTreeField<Tscal> hmax_tree = tree.compute_int_boxes(
                shamsys::instance::get_compute_queue(),
                pdat.get_field<Tscal>(pdat.pdl().get_field_idx<Tscal>("hpart")).get_buf(),
                1);
 
            sham::DeviceQueue &q = shamsys::instance::get_compute_scheduler().get_queue();
 
            sham::EventList depends_list;
            Tfield *render_field = ret.get_write_access(depends_list);
 
            const shammath::RingRay<Tvec> *ring_rays_ptr = ring_rays.get_read_access(depends_list);
 
            auto xyz      = buf_xyz.get_read_access(depends_list);
            auto hpart    = buf_hpart.get_read_access(depends_list);
            auto torender = buf_field_to_render.get_read_access(depends_list);
 
            sycl::event e2 = q.submit(depends_list, [&, render_field](sycl::handler &cgh) {
                shamrock::tree::ObjectIterator particle_looper(tree, cgh);
 
                sycl::accessor hmax{
                    shambase::get_check_ref(hmax_tree.radix_tree_field_buf), cgh, sycl::read_only};
 
                constexpr Tscal Rker2 = Kernel::Rkern * Kernel::Rkern;
 
                Tscal partmass = solver_config.gpart_mass;
 
                shambase::parallel_for(
                    cgh, ring_rays.get_size(), "compute slice render", [=](u32 gid) {
                        Tfield ret = sham::VectorProperties<Tfield>::get_zero();
 
                        shammath::RingRay<Tvec> ring_ray = ring_rays_ptr[gid];
                        Tvec ez                          = ring_ray.get_ez();
 
                        particle_looper.rtree_for(
                            [&](u32 node_id, Tvec bmin, Tvec bmax) -> bool {
                                Tscal rint_cell = hmax[node_id] * Kernel::Rkern;
 
                                auto interbox
                                    = shammath::AABB<Tvec>{bmin, bmax}.expand_all(rint_cell);
 
                                return interbox.intersect_ring_ray_approx(ring_ray);
                            },
                            [&](u32 id_b) {
                                Tvec r_center = ring_ray.center - xyz[id_b];
 
                                Tscal z_val = sycl::dot(r_center, ez);
                                Tscal x_val = sycl::dot(r_center, ring_ray.e_x);
                                Tscal y_val = sycl::dot(r_center, ring_ray.e_y);
                                Tscal r_val = sycl::sqrt(x_val * x_val + y_val * y_val);
 
                                Tscal delta_r = r_val - ring_ray.radius;
 
                                Tscal rab2_ring = z_val * z_val + delta_r * delta_r;
                                Tscal h_b       = hpart[id_b];
 
                                if (rab2_ring > h_b * h_b * Rker2) {
                                    return;
                                }
 
                                Tscal rab = sycl::sqrt(rab2_ring);
 
                                Tfield val = torender[id_b];
 
                                Tscal rho_b = shamrock::sph::rho_h(partmass, h_b, Kernel::hfactd);
 
                                // TODO: account for curvature
                                ret += partmass * val * Kernel::Y_3d(rab, h_b, 4) / rho_b;
                            });
 
                        render_field[gid] += ret;
                    });
            });
 
            buf_xyz.complete_event_state(e2);
            buf_hpart.complete_event_state(e2);
            buf_field_to_render.complete_event_state(e2);
            ret.complete_event_state(e2);
            ring_rays.complete_event_state(e2);
        });
 
        shamalgs::collective::reduce_buffer_in_place_sum(ret, MPI_COMM_WORLD);
 
        return ret;
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_slice(
        std::function<field_getter_t> field_getter,
        Tvec center,
        Tvec delta_x,
        Tvec delta_y,
        u32 nx,
        u32 ny) -> sham::DeviceBuffer<Tfield> {
 
        auto positions = pixel_to_positions(center, delta_x, delta_y, nx, ny);
 
        return compute_slice(field_getter, positions);
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_column_integ(
        std::function<field_getter_t> field_getter,
        Tvec center,
        Tvec delta_x,
        Tvec delta_y,
        u32 nx,
        u32 ny) -> sham::DeviceBuffer<Tfield> {
 
        auto rays = pixel_to_orthographic_rays(center, delta_x, delta_y, nx, ny);
 
        return compute_column_integ(field_getter, rays);
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_slice(
        std::string field_name,
        Tvec center,
        Tvec delta_x,
        Tvec delta_y,
        u32 nx,
        u32 ny,
        std::optional<std::function<pybind11::array_t<Tfield>(size_t, pybind11::dict &)>>
            custom_getter) -> sham::DeviceBuffer<Tfield> {
        auto positions = pixel_to_positions(center, delta_x, delta_y, nx, ny);
        return compute_slice(field_name, positions, custom_getter);
    }
 
    template<class Tvec, class Tfield, template<class> class SPHKernel>
    auto CartesianRender<Tvec, Tfield, SPHKernel>::compute_column_integ(
        std::string field_name,
        Tvec center,
        Tvec delta_x,
        Tvec delta_y,
        u32 nx,
        u32 ny,
        std::optional<std::function<pybind11::array_t<Tfield>(size_t, pybind11::dict &)>>
            custom_getter) -> sham::DeviceBuffer<Tfield> {
        auto rays = pixel_to_orthographic_rays(center, delta_x, delta_y, nx, ny);
        return compute_column_integ(field_name, rays, custom_getter);
    }
 
} // namespace shammodels::sph::modules
 
using namespace shammath;
template class shammodels::sph::modules::CartesianRender<f64_3, f64, M4>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64, M6>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64, M8>;
 
template class shammodels::sph::modules::CartesianRender<f64_3, f64, C2>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64, C4>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64, C6>;
 
template class shammodels::sph::modules::CartesianRender<f64_3, f64_3, M4>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64_3, M6>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64_3, M8>;
 
template class shammodels::sph::modules::CartesianRender<f64_3, f64_3, C2>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64_3, C4>;
template class shammodels::sph::modules::CartesianRender<f64_3, f64_3, C6>;