From 670c30e9eb273f83b519167b097fbce4785c56c8 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Mon, 24 Feb 2025 15:59:28 +0100
Subject: [PATCH 01/35] Add cpp implementations (untested)
---
src/horizontal/CMakeLists.txt | 2 +
src/horizontal/lib_divrot.cpp | 328 ++++++++++++++++++++++++++++++
src/horizontal/lib_divrot.hpp | 42 ++++
test/c/test_horizontal_divrot.cpp | 0
4 files changed, 372 insertions(+)
create mode 100644 src/horizontal/lib_divrot.cpp
create mode 100644 src/horizontal/lib_divrot.hpp
create mode 100644 test/c/test_horizontal_divrot.cpp
diff --git a/src/horizontal/CMakeLists.txt b/src/horizontal/CMakeLists.txt
index 078a14d..198488f 100644
--- a/src/horizontal/CMakeLists.txt
+++ b/src/horizontal/CMakeLists.txt
@@ -11,6 +11,7 @@
add_library(
iconmath-horizontal
+ lib_divrot.cpp
mo_lib_divrot.F90
mo_lib_laplace.F90
mo_lib_gradients.F90)
@@ -57,6 +58,7 @@ target_include_directories(
# Path to the internal C/C++ headers (for testing): Requires CMake 3.15+ for
# multiple compile languages
# https://cmake.org/cmake/help/latest/manual/cmake-generator-expressions.7.html
+ $<BUILD_INTERFACE:$<$<COMPILE_LANGUAGE:C,CXX>:${PROJECT_SOURCE_DIR}/src>>
$<BUILD_INTERFACE:$<$<COMPILE_LANGUAGE:C,CXX>:${CMAKE_CURRENT_SOURCE_DIR}>>
PRIVATE
# Path to config.h (for C and C++ only): Requires CMake 3.15+ for multiple
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
new file mode 100644
index 0000000..9dd698c
--- /dev/null
+++ b/src/horizontal/lib_divrot.cpp
@@ -0,0 +1,328 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#include <lib_divrot.hpp>
+#include <support/mo_lib_loopindices.hpp>
+#include <vector>
+
+template <typename T>
+void recon_lsq_cell_l_(const T *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const T *lsq_qtmat_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, bool l_consv, bool lacc,
+ bool acc_async, int nblks_c, int lsq_dim_unk,
+ int lsq_dim_c) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedT3D z_d(3);
+ UnmanagedT3D z_qt_times_d(2);
+
+ UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
+
+ UnmanagedConstT3D p_cc_view(p_cc);
+ UnmanagedT3D p_coeff_view(p_coeff);
+
+ UnmanagedConstT3D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
+ lsq_dim_c, nblks_c);
+ UnmanagedConstT3D lsq_rmat_rdiag_c_view(lsq_rmat_rdiag_c, nproma, lsq_dim_unk,
+ nblks_c);
+ UnmanagedConstT3D lsq_rmat_utri_c_view(
+ lsq_rmat_utri_c, nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2,
+ nblks_c);
+ UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
+
+ for (int jb = i_startblk; jb <= i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy(
+ {slev, i_startidx}, {elev + 1, i_endidx + 1});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_l_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_d(1) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_d(2) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_d(3) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+ // matrix multiplication Q^T d (partitioned into 2 dot products)
+ z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1) +
+ lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2) +
+ lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3);
+ z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1) +
+ lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2) +
+ lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3);
+
+ p_coeff_view(3, jc, jk, jb) =
+ lsq_rmat_rdiag_c_view(jc, 2, jb) * z_qt_times_d(2);
+ p_coeff_view(2, jc, jk, jb) =
+ lsq_rmat_rdiag_c_view(jc, 1, jb) *
+ (z_qt_times_d(1) -
+ lsq_rmat_utri_c_view(jc, 1, jb) * p_coeff_view(3, jc, jk, jb));
+ p_coeff_view(1, jc, jk, jb) = p_cc_view(jc, jk, jb);
+ });
+ if (l_consv) {
+ Kokkos::parallel_for(
+ "recon_lsq_cell_l_consv", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ p_coeff_view(1, jc, jk, jb) =
+ p_coeff_view(1, jc, jk, jb) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
+ p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2);
+ });
+ }
+ }
+
+ if (!acc_async)
+ Kokkos::fence();
+}
+
+template <typename T>
+void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const T *lsq_pseudoinv,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, bool l_consv,
+ bool lacc, bool acc_async, int nblks_c,
+ int lsq_dim_unk, int lsq_dim_c) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedT3D z_b(3);
+
+ UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
+
+ UnmanagedConstT3D p_cc_view(p_cc);
+ UnmanagedT3D p_coeff_view(p_coeff);
+
+ UnmanagedConstT3D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
+ lsq_dim_c, nblks_c);
+ UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
+
+ for (int jb = i_startblk; jb <= i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy(
+ {slev, i_startidx}, {elev + 1, i_endidx + 1});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_l_svd_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_b(1) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_b(2) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_b(3) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+
+ p_coeff_view(3, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 2, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 2, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 2, 3, jb) * z_b(3);
+ p_coeff_view(2, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 1, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 1, 3, jb) * z_b(3);
+ p_coeff_view(1, jc, jk, jb) = p_cc_view(jc, jk, jb);
+ });
+ if (l_consv) {
+ Kokkos::parallel_for(
+ "recon_lsq_cell_l_svd_consv", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ p_coeff_view(1, jc, jk, jb) =
+ p_coeff_view(1, jc, jk, jb) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
+ p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2);
+ });
+ }
+ }
+
+ if (!acc_async)
+ Kokkos::fence();
+}
+
+template <typename T>
+void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma,
+ int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedT3D z_d(lsq_high_set_dim_c, nproma, elev);
+ UnmanagedT3D z_qt_times_d(5);
+
+ UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
+ UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
+
+ UnmanagedConstT3D p_cc_view(p_cc);
+ UnmanagedT3D p_coeff_view(p_coeff);
+
+ UnmanagedConstT3D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
+ lsq_dim_c, nblks_c);
+ UnmanagedConstT3D ptr_rrdiag(lsq_rmat_rdiag_c, nproma, lsq_dim_unk, nblks_c);
+ UnmanagedConstT3D ptr_rutri(lsq_rmat_utri_c, nproma,
+ (lsq_dim_unk ^ 2 - lsq_dim_unk) / 2, nblks_c);
+ UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
+
+ if (patch_id > 1 || l_limited_area) {
+ Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
+ {1, i_startidx_in, slev, i_startblk},
+ {7, i_endidx_in + 1, elev + 1, i_endblk + 1});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_q_init", initPolicy,
+ KOKKOS_LAMBDA(const int z, const int jc, const int jk, const int jb) {
+ p_coeff_view(z, jc, jk, jb) = 0;
+ });
+ }
+
+ for (int jb = i_startblk; jb <= i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy(
+ {slev, i_startidx}, {elev + 1, i_endidx + 1});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_q_step1", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_d(1, jc, jk) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_d(2, jc, jk) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_d(3, jc, jk) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+ z_d(4, jc, jk) = p_cc_view(iidx(jc, jb, 4), jk, iblk(jc, jb, 4)) -
+ p_cc_view(jc, jk, jb);
+ z_d(5, jc, jk) = p_cc_view(iidx(jc, jb, 5), jk, iblk(jc, jb, 5)) -
+ p_cc_view(jc, jk, jb);
+ z_d(6, jc, jk) = p_cc_view(iidx(jc, jb, 6), jk, iblk(jc, jb, 6)) -
+ p_cc_view(jc, jk, jb);
+ z_d(7, jc, jk) = p_cc_view(iidx(jc, jb, 7), jk, iblk(jc, jb, 7)) -
+ p_cc_view(jc, jk, jb);
+ z_d(8, jc, jk) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
+ p_cc_view(jc, jk, jb);
+ z_d(9, jc, jk) = p_cc_view(iidx(jc, jb, 9), jk, iblk(jc, jb, 9)) -
+ p_cc_view(jc, jk, jb);
+ });
+ Kokkos::parallel_for(
+ "recon_lsq_cell_q_step2", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 8, jb) * z_d(8, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 9, jb) * z_d(9, jc, jk);
+ z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 8, jb) * z_d(8, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 9, jb) * z_d(9, jc, jk);
+ z_qt_times_d(3) = lsq_qtmat_c_view(jc, 3, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 8, jb) * z_d(8, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 9, jb) * z_d(9, jc, jk);
+ z_qt_times_d(4) = lsq_qtmat_c_view(jc, 4, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 8, jb) * z_d(8, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 9, jb) * z_d(9, jc, jk);
+ z_qt_times_d(5) = lsq_qtmat_c_view(jc, 5, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 8, jb) * z_d(8, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 9, jb) * z_d(9, jc, jk);
+
+ p_coeff_view(6, jc, jk, jb) = ptr_rrdiag(jc, 5, jb) * z_qt_times_d(5);
+ p_coeff_view(5, jc, jk, jb) =
+ ptr_rrdiag(jc, 4, jb) *
+ (z_qt_times_d(4) -
+ ptr_rutri(jc, 1, jb) * p_coeff_view(6, jc, jk, jb));
+ p_coeff_view(4, jc, jk, jb) =
+ ptr_rrdiag(jc, 3, jb) *
+ (z_qt_times_d(3) -
+ ptr_rutri(jc, 2, jb) * p_coeff_view(5, jc, jk, jb) -
+ ptr_rutri(jc, 3, jb) * p_coeff_view(6, jc, jk, jb));
+ p_coeff_view(3, jc, jk, jb) =
+ ptr_rrdiag(jc, 2, jb) *
+ (z_qt_times_d(2) -
+ ptr_rutri(jc, 4, jb) * p_coeff_view(4, jc, jk, jb) -
+ ptr_rutri(jc, 5, jb) * p_coeff_view(5, jc, jk, jb) -
+ ptr_rutri(jc, 6, jb) * p_coeff_view(6, jc, jk, jb));
+ p_coeff_view(2, jc, jk, jb) =
+ ptr_rrdiag(jc, 1, jb) *
+ (z_qt_times_d(1) -
+ ptr_rutri(jc, 7, jb) * p_coeff_view(3, jc, jk, jb) -
+ ptr_rutri(jc, 8, jb) * p_coeff_view(4, jc, jk, jb) -
+ ptr_rutri(jc, 9, jb) * p_coeff_view(5, jc, jk, jb) -
+ ptr_rutri(jc, 10, jb) * p_coeff_view(6, jc, jk, jb));
+ p_coeff_view(1, jc, jk, jb) =
+ p_cc(jc, jk, jb) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
+ p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) -
+ p_coeff_view(4, jc, jk, jb) * lsq_moments_view(jc, jb, 3) -
+ p_coeff_view(5, jc, jk, jb) * lsq_moments_view(jc, jb, 4) -
+ p_coeff_view(6, jc, jk, jb) * lsq_moments_view(jc, jb, 5);
+ });
+ }
+
+ Kokkos::fence();
+}
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
new file mode 100644
index 0000000..6977e5d
--- /dev/null
+++ b/src/horizontal/lib_divrot.hpp
@@ -0,0 +1,42 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#pragma once
+
+#include <Kokkos_Core.hpp>
+
+template <typename T>
+void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const T *lsq_qtmat_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, bool l_consv, bool lacc,
+ bool acc_async, int nblks_c, int lsq_dim_unk,
+ int lsq_dim_c);
+
+template <typename T>
+void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const T *lsq_pseudoinv,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, bool l_consv,
+ bool lacc, bool acc_async, int nblks_c,
+ int lsq_dim_unk, int lsq_dim_c);
+
+template <typename T>
+void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma,
+ int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c);
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
new file mode 100644
index 0000000..e69de29
--
GitLab
From f0cb7af9e59cd93ff5a3a920767430e1c033814b Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Mon, 24 Feb 2025 22:25:34 +0100
Subject: [PATCH 02/35] Fix indexes
---
src/horizontal/lib_divrot.cpp | 162 +++++++++++++++++-----------------
1 file changed, 81 insertions(+), 81 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 9dd698c..addd485 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -49,46 +49,46 @@ void recon_lsq_cell_l_(const T *p_cc, const int *cell_neighbor_idx,
nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
- for (int jb = i_startblk; jb <= i_endblk; ++jb) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
i_endblk, i_startidx, i_endidx);
- Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy(
- {slev, i_startidx}, {elev + 1, i_endidx + 1});
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
Kokkos::parallel_for(
"recon_lsq_cell_l_inner", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_d(0) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
z_d(1) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
p_cc_view(jc, jk, jb);
z_d(2) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
p_cc_view(jc, jk, jb);
- z_d(3) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
- p_cc_view(jc, jk, jb);
// matrix multiplication Q^T d (partitioned into 2 dot products)
- z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1) +
- lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2) +
- lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3);
- z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1) +
- lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2) +
- lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3);
+ z_qt_times_d(0) = lsq_qtmat_c_view(jc, 0, 0, jb) * z_d(0) +
+ lsq_qtmat_c_view(jc, 0, 1, jb) * z_d(1) +
+ lsq_qtmat_c_view(jc, 0, 2, jb) * z_d(2);
+ z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 0, jb) * z_d(0) +
+ lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1) +
+ lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2);
- p_coeff_view(3, jc, jk, jb) =
- lsq_rmat_rdiag_c_view(jc, 2, jb) * z_qt_times_d(2);
p_coeff_view(2, jc, jk, jb) =
- lsq_rmat_rdiag_c_view(jc, 1, jb) *
- (z_qt_times_d(1) -
- lsq_rmat_utri_c_view(jc, 1, jb) * p_coeff_view(3, jc, jk, jb));
- p_coeff_view(1, jc, jk, jb) = p_cc_view(jc, jk, jb);
+ lsq_rmat_rdiag_c_view(jc, 1, jb) * z_qt_times_d(1);
+ p_coeff_view(1, jc, jk, jb) =
+ lsq_rmat_rdiag_c_view(jc, 0, jb) *
+ (z_qt_times_d(0) -
+ lsq_rmat_utri_c_view(jc, 0, jb) * p_coeff_view(2, jc, jk, jb));
+ p_coeff_view(0, jc, jk, jb) = p_cc_view(jc, jk, jb);
});
if (l_consv) {
Kokkos::parallel_for(
"recon_lsq_cell_l_consv", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
- p_coeff_view(1, jc, jk, jb) =
- p_coeff_view(1, jc, jk, jb) -
- p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
- p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2);
+ p_coeff_view(0, jc, jk, jb) =
+ p_coeff_view(0, jc, jk, jb) -
+ p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1);
});
}
}
@@ -126,41 +126,41 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
lsq_dim_c, nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
- for (int jb = i_startblk; jb <= i_endblk; ++jb) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
i_endblk, i_startidx, i_endidx);
- Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy(
- {slev, i_startidx}, {elev + 1, i_endidx + 1});
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
Kokkos::parallel_for(
"recon_lsq_cell_l_svd_inner", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_b(0) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
z_b(1) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
p_cc_view(jc, jk, jb);
z_b(2) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
p_cc_view(jc, jk, jb);
- z_b(3) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
- p_cc_view(jc, jk, jb);
- p_coeff_view(3, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 2, 1, jb) * z_b(1) +
- lsq_pseudoinv_view(jc, 2, 2, jb) * z_b(2) +
- lsq_pseudoinv_view(jc, 2, 3, jb) * z_b(3);
p_coeff_view(2, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 1, 0, jb) * z_b(0) +
lsq_pseudoinv_view(jc, 1, 1, jb) * z_b(1) +
- lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2) +
- lsq_pseudoinv_view(jc, 1, 3, jb) * z_b(3);
- p_coeff_view(1, jc, jk, jb) = p_cc_view(jc, jk, jb);
+ lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2);
+ p_coeff_view(1, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 0, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 0, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 0, 2, jb) * z_b(2);
+ p_coeff_view(0, jc, jk, jb) = p_cc_view(jc, jk, jb);
});
if (l_consv) {
Kokkos::parallel_for(
"recon_lsq_cell_l_svd_consv", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
- p_coeff_view(1, jc, jk, jb) =
- p_coeff_view(1, jc, jk, jb) -
- p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
- p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2);
+ p_coeff_view(0, jc, jk, jb) =
+ p_coeff_view(0, jc, jk, jb) -
+ p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1);
});
}
}
@@ -199,13 +199,13 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
lsq_dim_c, nblks_c);
UnmanagedConstT3D ptr_rrdiag(lsq_rmat_rdiag_c, nproma, lsq_dim_unk, nblks_c);
UnmanagedConstT3D ptr_rutri(lsq_rmat_utri_c, nproma,
- (lsq_dim_unk ^ 2 - lsq_dim_unk) / 2, nblks_c);
+ (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2,
+ nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
if (patch_id > 1 || l_limited_area) {
Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
- {1, i_startidx_in, slev, i_startblk},
- {7, i_endidx_in + 1, elev + 1, i_endblk + 1});
+ {0, i_startidx_in, slev, i_startblk}, {6, i_endidx_in, elev, i_endblk});
Kokkos::parallel_for(
"recon_lsq_cell_q_init", initPolicy,
KOKKOS_LAMBDA(const int z, const int jc, const int jk, const int jb) {
@@ -213,16 +213,18 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
});
}
- for (int jb = i_startblk; jb <= i_endblk; ++jb) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
i_endblk, i_startidx, i_endidx);
- Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy(
- {slev, i_startidx}, {elev + 1, i_endidx + 1});
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
Kokkos::parallel_for(
"recon_lsq_cell_q_step1", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_d(0, jc, jk) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
z_d(1, jc, jk) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
p_cc_view(jc, jk, jb);
z_d(2, jc, jk) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
@@ -239,88 +241,86 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
p_cc_view(jc, jk, jb);
z_d(8, jc, jk) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
p_cc_view(jc, jk, jb);
- z_d(9, jc, jk) = p_cc_view(iidx(jc, jb, 9), jk, iblk(jc, jb, 9)) -
- p_cc_view(jc, jk, jb);
});
Kokkos::parallel_for(
"recon_lsq_cell_q_step2", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
- z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1, jc, jk) +
+ z_qt_times_d(0) = lsq_qtmat_c_view(jc, 0, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1, jc, jk) +
lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2, jc, jk) +
lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3, jc, jk) +
lsq_qtmat_c_view(jc, 1, 4, jb) * z_d(4, jc, jk) +
lsq_qtmat_c_view(jc, 1, 5, jb) * z_d(5, jc, jk) +
lsq_qtmat_c_view(jc, 1, 6, jb) * z_d(6, jc, jk) +
lsq_qtmat_c_view(jc, 1, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 8, jb) * z_d(8, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 9, jb) * z_d(9, jc, jk);
- z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1, jc, jk) +
lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2, jc, jk) +
lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3, jc, jk) +
lsq_qtmat_c_view(jc, 2, 4, jb) * z_d(4, jc, jk) +
lsq_qtmat_c_view(jc, 2, 5, jb) * z_d(5, jc, jk) +
lsq_qtmat_c_view(jc, 2, 6, jb) * z_d(6, jc, jk) +
lsq_qtmat_c_view(jc, 2, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 8, jb) * z_d(8, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 9, jb) * z_d(9, jc, jk);
- z_qt_times_d(3) = lsq_qtmat_c_view(jc, 3, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(3) = lsq_qtmat_c_view(jc, 3, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 1, jb) * z_d(1, jc, jk) +
lsq_qtmat_c_view(jc, 3, 2, jb) * z_d(2, jc, jk) +
lsq_qtmat_c_view(jc, 3, 3, jb) * z_d(3, jc, jk) +
lsq_qtmat_c_view(jc, 3, 4, jb) * z_d(4, jc, jk) +
lsq_qtmat_c_view(jc, 3, 5, jb) * z_d(5, jc, jk) +
lsq_qtmat_c_view(jc, 3, 6, jb) * z_d(6, jc, jk) +
lsq_qtmat_c_view(jc, 3, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 8, jb) * z_d(8, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 9, jb) * z_d(9, jc, jk);
- z_qt_times_d(4) = lsq_qtmat_c_view(jc, 4, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(4) = lsq_qtmat_c_view(jc, 4, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 1, jb) * z_d(1, jc, jk) +
lsq_qtmat_c_view(jc, 4, 2, jb) * z_d(2, jc, jk) +
lsq_qtmat_c_view(jc, 4, 3, jb) * z_d(3, jc, jk) +
lsq_qtmat_c_view(jc, 4, 4, jb) * z_d(4, jc, jk) +
lsq_qtmat_c_view(jc, 4, 5, jb) * z_d(5, jc, jk) +
lsq_qtmat_c_view(jc, 4, 6, jb) * z_d(6, jc, jk) +
lsq_qtmat_c_view(jc, 4, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 8, jb) * z_d(8, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 9, jb) * z_d(9, jc, jk);
- z_qt_times_d(5) = lsq_qtmat_c_view(jc, 5, 1, jb) * z_d(1, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 2, jb) * z_d(2, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 3, jb) * z_d(3, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 4, jb) * z_d(4, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 5, jb) * z_d(5, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 6, jb) * z_d(6, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 8, jb) * z_d(8, jc, jk) +
- lsq_qtmat_c_view(jc, 5, 9, jb) * z_d(9, jc, jk);
+ lsq_qtmat_c_view(jc, 4, 8, jb) * z_d(8, jc, jk);
- p_coeff_view(6, jc, jk, jb) = ptr_rrdiag(jc, 5, jb) * z_qt_times_d(5);
- p_coeff_view(5, jc, jk, jb) =
- ptr_rrdiag(jc, 4, jb) *
- (z_qt_times_d(4) -
- ptr_rutri(jc, 1, jb) * p_coeff_view(6, jc, jk, jb));
+ p_coeff_view(5, jc, jk, jb) = ptr_rrdiag(jc, 4, jb) * z_qt_times_d(4);
p_coeff_view(4, jc, jk, jb) =
ptr_rrdiag(jc, 3, jb) *
(z_qt_times_d(3) -
- ptr_rutri(jc, 2, jb) * p_coeff_view(5, jc, jk, jb) -
- ptr_rutri(jc, 3, jb) * p_coeff_view(6, jc, jk, jb));
+ ptr_rutri(jc, 0, jb) * p_coeff_view(5, jc, jk, jb));
p_coeff_view(3, jc, jk, jb) =
ptr_rrdiag(jc, 2, jb) *
(z_qt_times_d(2) -
- ptr_rutri(jc, 4, jb) * p_coeff_view(4, jc, jk, jb) -
- ptr_rutri(jc, 5, jb) * p_coeff_view(5, jc, jk, jb) -
- ptr_rutri(jc, 6, jb) * p_coeff_view(6, jc, jk, jb));
+ ptr_rutri(jc, 1, jb) * p_coeff_view(4, jc, jk, jb) -
+ ptr_rutri(jc, 2, jb) * p_coeff_view(5, jc, jk, jb));
p_coeff_view(2, jc, jk, jb) =
ptr_rrdiag(jc, 1, jb) *
(z_qt_times_d(1) -
+ ptr_rutri(jc, 3, jb) * p_coeff_view(3, jc, jk, jb) -
+ ptr_rutri(jc, 4, jb) * p_coeff_view(4, jc, jk, jb) -
+ ptr_rutri(jc, 5, jb) * p_coeff_view(5, jc, jk, jb));
+ p_coeff_view(1, jc, jk, jb) =
+ ptr_rrdiag(jc, 0, jb) *
+ (z_qt_times_d(0) -
+ ptr_rutri(jc, 6, jb) * p_coeff_view(2, jc, jk, jb) -
ptr_rutri(jc, 7, jb) * p_coeff_view(3, jc, jk, jb) -
ptr_rutri(jc, 8, jb) * p_coeff_view(4, jc, jk, jb) -
- ptr_rutri(jc, 9, jb) * p_coeff_view(5, jc, jk, jb) -
- ptr_rutri(jc, 10, jb) * p_coeff_view(6, jc, jk, jb));
- p_coeff_view(1, jc, jk, jb) =
+ ptr_rutri(jc, 9, jb) * p_coeff_view(5, jc, jk, jb));
+ p_coeff_view(0, jc, jk, jb) =
p_cc(jc, jk, jb) -
+ p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) -
p_coeff_view(4, jc, jk, jb) * lsq_moments_view(jc, jb, 3) -
- p_coeff_view(5, jc, jk, jb) * lsq_moments_view(jc, jb, 4) -
- p_coeff_view(6, jc, jk, jb) * lsq_moments_view(jc, jb, 5);
+ p_coeff_view(5, jc, jk, jb) * lsq_moments_view(jc, jb, 4);
});
}
--
GitLab
From 9d04113e48f8e7081adc57b1f7f890346785a58b Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Tue, 25 Feb 2025 14:18:33 +0100
Subject: [PATCH 03/35] Add cpp implementations (untested)
---
src/horizontal/lib_divrot.cpp | 617 +++++++++++++++++++++++++++++++++-
src/horizontal/lib_divrot.hpp | 29 ++
2 files changed, 630 insertions(+), 16 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index addd485..5b51d98 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -25,22 +25,27 @@ void recon_lsq_cell_l_(const T *p_cc, const int *cell_neighbor_idx,
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
- typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT1D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT3D z_d(3);
- UnmanagedT3D z_qt_times_d(2);
+ UnmanagedT1D z_d(3);
+ UnmanagedT1D z_qt_times_d(2);
UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT3D p_coeff_view(p_coeff);
+ UnmanagedT4D p_coeff_view(p_coeff);
- UnmanagedConstT3D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
+ UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
UnmanagedConstT3D lsq_rmat_rdiag_c_view(lsq_rmat_rdiag_c, nproma, lsq_dim_unk,
nblks_c);
@@ -108,21 +113,26 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
- typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT1D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT3D z_b(3);
+ UnmanagedT1D z_b(3);
UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT3D p_coeff_view(p_coeff);
+ UnmanagedT4D p_coeff_view(p_coeff);
- UnmanagedConstT3D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
+ UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
@@ -180,22 +190,29 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT1D;
typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT3D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
UnmanagedT3D z_d(lsq_high_set_dim_c, nproma, elev);
- UnmanagedT3D z_qt_times_d(5);
+ UnmanagedT1D z_qt_times_d(5);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT3D p_coeff_view(p_coeff);
+ UnmanagedT4D p_coeff_view(p_coeff);
- UnmanagedConstT3D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
+ UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
UnmanagedConstT3D ptr_rrdiag(lsq_rmat_rdiag_c, nproma, lsq_dim_unk, nblks_c);
UnmanagedConstT3D ptr_rutri(lsq_rmat_utri_c, nproma,
@@ -208,8 +225,8 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
{0, i_startidx_in, slev, i_startblk}, {6, i_endidx_in, elev, i_endblk});
Kokkos::parallel_for(
"recon_lsq_cell_q_init", initPolicy,
- KOKKOS_LAMBDA(const int z, const int jc, const int jk, const int jb) {
- p_coeff_view(z, jc, jk, jb) = 0;
+ KOKKOS_LAMBDA(const int ji, const int jc, const int jk, const int jb) {
+ p_coeff_view(ji, jc, jk, jb) = 0;
});
}
@@ -326,3 +343,571 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::fence();
}
+
+template <typename T>
+void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
+ const int *lsq_blk_c, const T *lsq_pseudoinv,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, int patch_id,
+ int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk,
+ int lsq_dim_c) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedT3D z_b(lsq_high_set_dim_c, nproma, elev);
+
+ UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
+ UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
+
+ UnmanagedConstT3D p_cc_view(p_cc);
+ UnmanagedT4D p_coeff_view(p_coeff);
+
+ UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
+ lsq_dim_c, nblks_c);
+ UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
+
+ if (patch_id > 1 || l_limited_area) {
+ Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
+ {0, i_startidx_in, slev, i_startblk}, {6, i_endidx_in, elev, i_endblk});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_q_svd_init", initPolicy,
+ KOKKOS_LAMBDA(const int ji, const int jc, const int jk, const int jb) {
+ p_coeff_view(ji, jc, jk, jb) = 0;
+ });
+ }
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_q_svd_step1", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_b(0, jc, jk) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
+ z_b(1, jc, jk) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_b(2, jc, jk) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_b(3, jc, jk) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+ z_b(4, jc, jk) = p_cc_view(iidx(jc, jb, 4), jk, iblk(jc, jb, 4)) -
+ p_cc_view(jc, jk, jb);
+ z_b(5, jc, jk) = p_cc_view(iidx(jc, jb, 5), jk, iblk(jc, jb, 5)) -
+ p_cc_view(jc, jk, jb);
+ z_b(6, jc, jk) = p_cc_view(iidx(jc, jb, 6), jk, iblk(jc, jb, 6)) -
+ p_cc_view(jc, jk, jb);
+ z_b(7, jc, jk) = p_cc_view(iidx(jc, jb, 7), jk, iblk(jc, jb, 7)) -
+ p_cc_view(jc, jk, jb);
+ z_b(8, jc, jk) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
+ p_cc_view(jc, jk, jb);
+ });
+ Kokkos::parallel_for(
+ "recon_lsq_cell_q_svd_step2", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ p_coeff_view(5, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 4, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(4, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 3, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(3, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 2, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(2, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 1, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(1, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 0, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(0, jc, jk, jb) =
+ p_cc_view(jc, jk, jb) -
+ p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
+ p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) -
+ p_coeff_view(4, jc, jk, jb) * lsq_moments_view(jc, jb, 3) -
+ p_coeff_view(5, jc, jk, jb) * lsq_moments_view(jc, jb, 4);
+ });
+ }
+
+ Kokkos::fence();
+}
+
+template <typename T>
+void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma,
+ int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT1D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedT3D z_d(lsq_high_set_dim_c, nproma, elev);
+ UnmanagedT1D z_qt_times_d(9);
+
+ UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
+ UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
+
+ UnmanagedConstT3D p_cc_view(p_cc);
+ UnmanagedT4D p_coeff_view(p_coeff);
+
+ UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
+ lsq_dim_c, nblks_c);
+ UnmanagedConstT3D ptr_rrdiag(lsq_rmat_rdiag_c, nproma, lsq_dim_unk, nblks_c);
+ UnmanagedConstT3D ptr_rutri(lsq_rmat_utri_c, nproma,
+ (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2,
+ nblks_c);
+ UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
+
+ if (patch_id > 1 || l_limited_area) {
+ Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
+ {0, i_startidx_in, slev, i_startblk}, {9, i_endidx_in, elev, i_endblk});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_c_init", initPolicy,
+ KOKKOS_LAMBDA(const int ji, const int jc, const int jk, const int jb) {
+ p_coeff_view(ji, jc, jk, jb) = 0;
+ });
+ }
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_c_step1", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_d(0, jc, jk) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
+ z_d(1, jc, jk) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_d(2, jc, jk) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_d(3, jc, jk) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+ z_d(4, jc, jk) = p_cc_view(iidx(jc, jb, 4), jk, iblk(jc, jb, 4)) -
+ p_cc_view(jc, jk, jb);
+ z_d(5, jc, jk) = p_cc_view(iidx(jc, jb, 5), jk, iblk(jc, jb, 5)) -
+ p_cc_view(jc, jk, jb);
+ z_d(6, jc, jk) = p_cc_view(iidx(jc, jb, 6), jk, iblk(jc, jb, 6)) -
+ p_cc_view(jc, jk, jb);
+ z_d(7, jc, jk) = p_cc_view(iidx(jc, jb, 7), jk, iblk(jc, jb, 7)) -
+ p_cc_view(jc, jk, jb);
+ z_d(8, jc, jk) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
+ p_cc_view(jc, jk, jb);
+ });
+ Kokkos::parallel_for(
+ "recon_lsq_cell_c_step2", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_qt_times_d(0) = lsq_qtmat_c_view(jc, 0, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(3) = lsq_qtmat_c_view(jc, 3, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(4) = lsq_qtmat_c_view(jc, 4, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(5) = lsq_qtmat_c_view(jc, 5, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 5, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(6) = lsq_qtmat_c_view(jc, 6, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 6, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(7) = lsq_qtmat_c_view(jc, 7, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 7, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(8) = lsq_qtmat_c_view(jc, 8, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 8, 8, jb) * z_d(8, jc, jk);
+
+ p_coeff_view(9, jc, jk, jb) = ptr_rrdiag(jc, 8, jb) * z_qt_times_d(8);
+ p_coeff_view(8, jc, jk, jb) =
+ ptr_rrdiag(jc, 7, jb) *
+ (z_qt_times_d(7) -
+ ptr_rutri(jc, 0, jb) * p_coeff_view(9, jc, jk, jb));
+ p_coeff_view(7, jc, jk, jb) =
+ ptr_rrdiag(jc, 6, jb) *
+ (z_qt_times_d(6) -
+ (ptr_rutri(jc, 1, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 2, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(6, jc, jk, jb) =
+ ptr_rrdiag(jc, 5, jb) *
+ (z_qt_times_d(5) -
+ (ptr_rutri(jc, 3, jb) * p_coeff_view(7, jc, jk, jb) +
+ ptr_rutri(jc, 4, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 5, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(5, jc, jk, jb) =
+ ptr_rrdiag(jc, 4, jb) *
+ (z_qt_times_d(4) -
+ (ptr_rutri(jc, 6, jb) * p_coeff_view(6, jc, jk, jb) +
+ ptr_rutri(jc, 7, jb) * p_coeff_view(7, jc, jk, jb) +
+ ptr_rutri(jc, 8, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 9, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(4, jc, jk, jb) =
+ ptr_rrdiag(jc, 3, jb) *
+ (z_qt_times_d(3) -
+ (ptr_rutri(jc, 10, jb) * p_coeff_view(5, jc, jk, jb) +
+ ptr_rutri(jc, 11, jb) * p_coeff_view(6, jc, jk, jb) +
+ ptr_rutri(jc, 12, jb) * p_coeff_view(7, jc, jk, jb) +
+ ptr_rutri(jc, 13, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 14, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(3, jc, jk, jb) =
+ ptr_rrdiag(jc, 2, jb) *
+ (z_qt_times_d(2) -
+ (ptr_rutri(jc, 15, jb) * p_coeff_view(4, jc, jk, jb) +
+ ptr_rutri(jc, 16, jb) * p_coeff_view(5, jc, jk, jb) +
+ ptr_rutri(jc, 17, jb) * p_coeff_view(6, jc, jk, jb) +
+ ptr_rutri(jc, 18, jb) * p_coeff_view(7, jc, jk, jb) +
+ ptr_rutri(jc, 19, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 20, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(2, jc, jk, jb) =
+ ptr_rrdiag(jc, 1, jb) *
+ (z_qt_times_d(1) -
+ (ptr_rutri(jc, 21, jb) * p_coeff_view(3, jc, jk, jb) +
+ ptr_rutri(jc, 22, jb) * p_coeff_view(4, jc, jk, jb) +
+ ptr_rutri(jc, 23, jb) * p_coeff_view(5, jc, jk, jb) +
+ ptr_rutri(jc, 24, jb) * p_coeff_view(6, jc, jk, jb) +
+ ptr_rutri(jc, 25, jb) * p_coeff_view(7, jc, jk, jb) +
+ ptr_rutri(jc, 26, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 27, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(1, jc, jk, jb) =
+ ptr_rrdiag(jc, 0, jb) *
+ (z_qt_times_d(0) -
+ (ptr_rutri(jc, 28, jb) * p_coeff_view(2, jc, jk, jb) +
+ ptr_rutri(jc, 29, jb) * p_coeff_view(3, jc, jk, jb) +
+ ptr_rutri(jc, 30, jb) * p_coeff_view(4, jc, jk, jb) +
+ ptr_rutri(jc, 31, jb) * p_coeff_view(5, jc, jk, jb) +
+ ptr_rutri(jc, 32, jb) * p_coeff_view(6, jc, jk, jb) +
+ ptr_rutri(jc, 33, jb) * p_coeff_view(7, jc, jk, jb) +
+ ptr_rutri(jc, 34, jb) * p_coeff_view(8, jc, jk, jb) +
+ ptr_rutri(jc, 35, jb) * p_coeff_view(9, jc, jk, jb)));
+ p_coeff_view(0, jc, jk, jb) =
+ p_cc(jc, jk, jb) -
+ (p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) +
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) +
+ p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) +
+ p_coeff_view(4, jc, jk, jb) * lsq_moments_view(jc, jb, 3) +
+ p_coeff_view(5, jc, jk, jb) * lsq_moments_view(jc, jb, 4) +
+ p_coeff_view(6, jc, jk, jb) * lsq_moments_view(jc, jb, 5) +
+ p_coeff_view(7, jc, jk, jb) * lsq_moments_view(jc, jb, 6) +
+ p_coeff_view(8, jc, jk, jb) * lsq_moments_view(jc, jb, 7) +
+ p_coeff_view(9, jc, jk, jb) * lsq_moments_view(jc, jb, 8));
+ });
+ }
+
+ Kokkos::fence();
+}
+
+template <typename T>
+void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
+ const int *lsq_blk_c, const T *lsq_pseudoinv,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, int patch_id,
+ int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk,
+ int lsq_dim_c) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT1D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedT1D z_b(9);
+
+ UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
+ UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
+
+ UnmanagedConstT3D p_cc_view(p_cc);
+ UnmanagedT4D p_coeff_view(p_coeff);
+
+ UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
+ lsq_dim_c, nblks_c);
+ UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
+
+ if (patch_id > 1 || l_limited_area) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<3>> initPolicy({slev, i_startidx, 0},
+ {elev, i_endidx, 9});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_c_svd_init", initPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc, const int ji) {
+ p_coeff_view(ji, jc, jk, jb) = 0;
+ });
+ }
+ }
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "recon_lsq_cell_c_svd_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ z_b(0, jc, jk) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
+ z_b(1, jc, jk) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_b(2, jc, jk) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_b(3, jc, jk) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+ z_b(4, jc, jk) = p_cc_view(iidx(jc, jb, 4), jk, iblk(jc, jb, 4)) -
+ p_cc_view(jc, jk, jb);
+ z_b(5, jc, jk) = p_cc_view(iidx(jc, jb, 5), jk, iblk(jc, jb, 5)) -
+ p_cc_view(jc, jk, jb);
+ z_b(6, jc, jk) = p_cc_view(iidx(jc, jb, 6), jk, iblk(jc, jb, 6)) -
+ p_cc_view(jc, jk, jb);
+ z_b(7, jc, jk) = p_cc_view(iidx(jc, jb, 7), jk, iblk(jc, jb, 7)) -
+ p_cc_view(jc, jk, jb);
+ z_b(8, jc, jk) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
+ p_cc_view(jc, jk, jb);
+
+ p_coeff_view(9, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 8, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 8, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(8, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 7, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 7, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(7, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 6, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 6, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(6, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 5, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 5, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(5, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 4, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 4, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(4, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 3, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 3, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(3, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 2, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 2, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(2, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 1, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 1, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(1, jc, jk, jb) =
+ lsq_pseudoinv_view(jc, 0, 0, jb) * z_b(0, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 1, jb) * z_b(1, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 2, jb) * z_b(2, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 3, jb) * z_b(3, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 4, jb) * z_b(4, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 5, jb) * z_b(5, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 6, jb) * z_b(6, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 7, jb) * z_b(7, jc, jk) +
+ lsq_pseudoinv_view(jc, 0, 8, jb) * z_b(8, jc, jk);
+ p_coeff_view(0, jc, jk, jb) =
+ p_cc_view(jc, jk, jb) -
+ p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
+ p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
+ p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) -
+ p_coeff_view(4, jc, jk, jb) * lsq_moments_view(jc, jb, 3) -
+ p_coeff_view(5, jc, jk, jb) * lsq_moments_view(jc, jb, 4) -
+ p_coeff_view(6, jc, jk, jb) * lsq_moments_view(jc, jb, 5) -
+ p_coeff_view(7, jc, jk, jb) * lsq_moments_view(jc, jb, 6) -
+ p_coeff_view(8, jc, jk, jb) * lsq_moments_view(jc, jb, 7) -
+ p_coeff_view(9, jc, jk, jb) * lsq_moments_view(jc, jb, 8);
+ });
+ }
+}
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
index 6977e5d..c32ee12 100644
--- a/src/horizontal/lib_divrot.hpp
+++ b/src/horizontal/lib_divrot.hpp
@@ -40,3 +40,32 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c);
+
+template <typename T>
+void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
+ const int *lsq_blk_c, const T *lsq_pseudoinv,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, int patch_id,
+ int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk,
+ int lsq_dim_c);
+
+template <typename T>
+void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma,
+ int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c);
+
+template <typename T>
+void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
+ const int *lsq_blk_c, const T *lsq_pseudoinv,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, int patch_id,
+ int lsq_high_set_dim_c, bool l_limited_area,
+ bool lacc, int nblks_c, int lsq_dim_unk,
+ int lsq_dim_c);
--
GitLab
From 642a9b166d2bdf4f04dadcb21674984e2876f86b Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Wed, 26 Feb 2025 11:03:35 +0100
Subject: [PATCH 04/35] Add cpp implementations (untested)
---
src/horizontal/lib_divrot.cpp | 442 +++++++++++++++++++++++++++++++++-
src/horizontal/lib_divrot.hpp | 52 +++-
2 files changed, 486 insertions(+), 8 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 5b51d98..3586b03 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -182,7 +182,7 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
template <typename T>
void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
@@ -347,7 +347,7 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
template <typename T>
void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T *p_coeff, int i_startblk,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
@@ -487,7 +487,7 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
template <typename T>
void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
@@ -730,7 +730,7 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
template <typename T>
void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T *p_coeff, int i_startblk,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
@@ -910,4 +910,438 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
p_coeff_view(9, jc, jk, jb) * lsq_moments_view(jc, jb, 8);
});
}
+
+ Kokkos::fence();
+}
+
+template <typename T>
+void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
+ const T *geofac_div, T *div_vec_c, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, int nlev, int nblks_c, int nblks_e) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedConstT3D vec_e_view(vec_e, nproma, nlev, nblks_e);
+
+ UnmanagedConstInt3D iidx(cell_edge_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iblk(cell_edge_blk, nproma, nblks_c, 3);
+
+ UnmanagedConstT3D geofac_div_view(geofac_div, nproma, 3, nblks_c);
+ UnmanagedT3D div_vec_c_view(div_vec_c, nproma, nlev, nblks_c);
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div3d_inner", innerPolicy, KOKKOS_LAMBDA(const int jk, const int jc) {
+ div_vec_c_view(jc, jk, jb) =
+ vec_e_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) *
+ geofac_div_view(jc, 0, jb) +
+ vec_e_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) *
+ geofac_div_view(jc, 1, jb) +
+ vec_e_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) *
+ geofac_div_view(jc, 2, jb);
+ });
+ }
+}
+
+template <typename T>
+void div3d_2field(const T *vec_e, const int *cell_edge_idx,
+ const int *cell_edge_blk, const T *geofac_div, T &div_vec_c,
+ const T *in2, T &out2, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev,
+ int nproma, bool lacc, int nlev, int nblks_c, int nblks_e) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedConstT3D vec_e_view(vec_e, nproma, nlev, nblks_e);
+
+ UnmanagedConstInt3D iidx(cell_edge_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iblk(cell_edge_blk, nproma, nblks_c, 3);
+
+ UnmanagedConstT3D geofac_div_view(geofac_div, nproma, 3, nblks_c);
+ UnmanagedT3D div_vec_c_view(div_vec_c, nproma, nlev, nblks_c);
+
+ UnmanagedConstT3D in2_view(in2, nproma, nlev, nblks_e);
+ UnmanagedT3D out2_view(out2, nproma, nlev, nblks_c);
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div3d_2field_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ div_vec_c_view(jc, jk, jb) =
+ vec_e_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) *
+ geofac_div_view(jc, 0, jb) +
+ vec_e_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) *
+ geofac_div_view(jc, 1, jb) +
+ vec_e_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) *
+ geofac_div_view(jc, 2, jb);
+
+ out2_view(jc, jk, jb) =
+ in2_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) *
+ geofac_div_view(jc, 0, jb) +
+ in2_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) *
+ geofac_div_view(jc, 1, jb) +
+ in2_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) *
+ geofac_div_view(jc, 2, jb);
+ });
+ }
+}
+
+template <typename T>
+void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
+ const T *geofac_div, const T *f4din, T &f4dout, int dim4d,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in,
+ const int *slev, const int *elev, int nproma, bool lacc, int nlev,
+ int nblks_c, int nblks_e) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstT4D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT4D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedConstInt3D iidx(cell_edge_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iblk(cell_edge_blk, nproma, nblks_c, 3);
+
+ UnmanagedConstT3D geofac_div_view(geofac_div, nproma, 3, nblks_c);
+
+ UnmanagedConstT4D f4din_view(f4din, nproma, nlev, nblks_e, dim4d);
+ UnmanagedT4D f4dout_view(f4dout, nproma, nlev, nblks_c, dim4d);
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ for (int ji = 0; ji < dim4d; ++ji) {
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev[ji], i_startidx},
+ {elev[ji], i_endidx});
+ Kokkos::parallel_for(
+ "div4d_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ f4dout_view(jc, jk, jb, ji) =
+ f4din_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0), ji) *
+ geofac_div_view(jc, 0, jb) +
+ f4din_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1), ji) *
+ geofac_div_view(jc, 1, jb) +
+ f4din_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2), ji) *
+ geofac_div_view(jc, 2, jb);
+ });
+ }
+ }
+}
+
+template <typename T>
+void div_avg(const T *vec_e, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const int *cell_edge_idx,
+ const int *cell_edge_blk, const T *geofac_div, const T *avg_coeff,
+ T &div_vec_c, const T *opt_in2, T &opt_out2,
+ const int *i_startblk_in, const int *i_endblk_in,
+ const int *i_startidx_in, const int *i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, bool l_limited_area,
+ bool l2fields, bool lacc, int nlev, int nblks_c, int nblks_e) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedConstT3D vec_e_view(vec_e, nproma, nlev, nblks_e);
+
+ UnmanagedConstInt3D inidx(cell_neighbor_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D inblk(cell_neighbor_blk, nproma, nblks_c, 3);
+ UnmanagedConstInt3D ieidx(cell_edge_idx, nproma, nblks_c, 3);
+ UnmanagedConstInt3D ieblk(cell_edge_blk, nproma, nblks_c, 3);
+
+ UnmanagedConstT3D geofac_div_view(geofac_div, nproma, 4, nblks_e);
+ UnmanagedConstT3D avg_coeff_view(avg_coeff, nproma, nlev, nblks_c);
+
+ UnmanagedT3D div_vec_c_view(div_vec_c, nproma, nlev, nblks_c);
+
+ UnmanagedConstT3D opt_in2_view(opt_in2, nproma, nlev, nblks_e);
+ UnmanagedT3D opt_out2_view(opt_out2, nproma, nlev, nblks_c);
+
+ UnmanagedT3D aux_c(nproma, nlev, nblks_c);
+ UnmanagedT3D aux_c2(nproma, nlev, nblks_c);
+
+ int i_startblk = i_startblk_in[0];
+ int i_endblk = i_endblk_in[0];
+
+ if (l2fields) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk, i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div_avg_step1", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ aux_c(jc, jk, jb) =
+ vec_e_view(ieidx(jc, jb, 0), jk, ieblk(jc, jb, 0)) *
+ geofac_div_view(jc, 0, jb) +
+ vec_e_view(ieidx(jc, jb, 1), jk, ieblk(jc, jb, 1)) *
+ geofac_div_view(jc, 1, jb) +
+ vec_e_view(ieidx(jc, jb, 2), jk, ieblk(jc, jb, 2)) *
+ geofac_div_view(jc, 2, jb);
+ aux_c2(jc, jk, jb) =
+ opt_in2(ieidx(jc, jb, 0), jk, ieblk(jc, jb, 0)) *
+ geofac_div_view(jc, 0, jb) +
+ opt_in2(ieidx(jc, jb, 1), jk, ieblk(jc, jb, 1)) *
+ geofac_div_view(jc, 1, jb) +
+ opt_in2(ieidx(jc, jb, 2), jk, ieblk(jc, jb, 2)) *
+ geofac_div_view(jc, 2, jb);
+ });
+ }
+ } else {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk, i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div_avg_step2", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ aux_c(jc, jk, jb) =
+ vec_e_view(ieidx(jc, jb, 0), jk, ieblk(jc, jb, 0)) *
+ geofac_div_view(jc, 0, jb) +
+ vec_e_view(ieidx(jc, jb, 1), jk, ieblk(jc, jb, 1)) *
+ geofac_div_view(jc, 1, jb) +
+ vec_e_view(ieidx(jc, jb, 2), jk, ieblk(jc, jb, 2)) *
+ geofac_div_view(jc, 2, jb);
+ });
+ }
+ }
+
+ if (patch_id > 1 || l_limited_area) {
+ i_startblk = i_startblk_in[1];
+ i_endblk = i_endblk_in[1];
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk, i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div_avg_step3", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ div_vec_c_view(jc, jk, jb) = aux_c(jc, jk, jb);
+ });
+ }
+
+ if (l2fields) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk, i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div_avg_step4", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ opt_out2_view(jc, jk, jb) = aux_c2(jc, jk, jb);
+ });
+ }
+ }
+ }
+
+ i_startblk = i_startblk_in[2];
+ i_endblk = i_endblk_in[2];
+
+ if (l2fields) {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk, i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div_avg_step5", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ div_vec_c_view(jc, jk, jb) =
+ aux_c(jc, jk, jb) * avg_coeff_view(jc, 0, jb) +
+ aux_c(inidx(jc, jb, 0), jk, inblk(jc, jb, 0)) *
+ avg_coeff_view(jc, 1, jb) +
+ aux_c(inidx(jc, jb, 1), jk, inblk(jc, jb, 1)) *
+ avg_coeff_view(jc, 2, jb) +
+ aux_c(inidx(jc, jb, 2), jk, inblk(jc, jb, 2)) *
+ avg_coeff_view(jc, 3, jb);
+ opt_out2_view(jc, jk, jb) =
+ aux_c2(jc, jk, jb) * avg_coeff_view(jc, 0, jb) +
+ aux_c2(inidx(jc, jb, 0), jk, inblk(jc, jb, 0)) *
+ avg_coeff_view(jc, 1, jb) +
+ aux_c2(inidx(jc, jb, 1), jk, inblk(jc, jb, 1)) *
+ avg_coeff_view(jc, 2, jb) +
+ aux_c2(inidx(jc, jb, 2), jk, inblk(jc, jb, 2)) *
+ avg_coeff_view(jc, 3, jb);
+ });
+ }
+ } else {
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk, i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "div_avg_step6", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jc) {
+ div_vec_c_view(jc, jk, jb) =
+ aux_c(jc, jk, jb) * avg_coeff_view(jc, 0, jb) +
+ aux_c(inidx(jc, jb, 0), jk, inblk(jc, jb, 0)) *
+ avg_coeff_view(jc, 1, jb) +
+ aux_c(inidx(jc, jb, 1), jk, inblk(jc, jb, 1)) *
+ avg_coeff_view(jc, 2, jb) +
+ aux_c(inidx(jc, jb, 2), jk, inblk(jc, jb, 2)) *
+ avg_coeff_view(jc, 3, jb);
+ });
+ }
+ }
+}
+
+template <typename T>
+void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
+ const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, int nlev, int nblks_e, int nblks_v) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedConstT3D vec_e_view(vec_e, nproma, nlev, nblks_e);
+
+ UnmanagedConstInt3D iidx(vert_edge_idx, nproma, nblks_v, 6);
+ UnmanagedConstInt3D iblk(vert_edge_blk, nproma, nblks_v, 6);
+
+ UnmanagedConstT3D geofac_rot_view(geofac_rot, nproma, 6, nblks_v);
+
+ UnmanagedT3D rot_vec_view(rot_vec, nproma, nlev, nblks_v);
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_v_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "rot_vertex_atmos_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jv) {
+ rot_vec_view(jv, jk, jb) =
+ vec_e_view(iidx(jv, jb, 0), jk, iblk(jv, jb, 0)) *
+ geofac_rot_view(jv, 0, jb) +
+ vec_e_view(iidx(jv, jb, 1), jk, iblk(jv, jb, 1)) *
+ geofac_rot_view(jv, 1, jb) +
+ vec_e_view(iidx(jv, jb, 2), jk, iblk(jv, jb, 2)) *
+ geofac_rot_view(jv, 2, jb) +
+ vec_e_view(iidx(jv, jb, 3), jk, iblk(jv, jb, 3)) *
+ geofac_rot_view(jv, 3, jb) +
+ vec_e_view(iidx(jv, jb, 4), jk, iblk(jv, jb, 4)) *
+ geofac_rot_view(jv, 4, jb) +
+ vec_e_view(iidx(jv, jb, 5), jk, iblk(jv, jb, 5)) *
+ geofac_rot_view(jv, 5, jb);
+ });
+ }
+}
+
+template <typename T>
+void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
+ const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma, bool lacc,
+ bool acc_async, int nlev, int nblks_e, int nblks_v) {
+ // Wrap raw pointers in unmanaged Kokkos Views.
+ typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedConstT3D;
+ typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
+ UnmanagedT3D;
+ typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
+ Kokkos::MemoryUnmanaged>
+ UnmanagedConstInt3D;
+
+ UnmanagedConstT3D vec_e_view(vec_e, nproma, nlev, nblks_e);
+
+ UnmanagedConstInt3D iidx(vert_edge_idx, nproma, nblks_v, 6);
+ UnmanagedConstInt3D iblk(vert_edge_blk, nproma, nblks_v, 6);
+
+ UnmanagedConstT3D geofac_rot_view(geofac_rot, nproma, 6, nblks_v);
+
+ UnmanagedT3D rot_vec_view(rot_vec, nproma, nlev, nblks_v);
+
+ for (int jb = i_startblk; jb < i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_v_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
+ i_endblk, i_startidx, i_endidx);
+
+ Kokkos::MDRangePolicy<Kokkos::Rank<2>> innerPolicy({slev, i_startidx},
+ {elev, i_endidx});
+ Kokkos::parallel_for(
+ "rot_vertex_atmos_inner", innerPolicy,
+ KOKKOS_LAMBDA(const int jk, const int jv) {
+ rot_vec_view(jv, jk, jb) =
+ vec_e_view(iidx(jv, jb, 0), jk, iblk(jv, jb, 0)) *
+ geofac_rot_view(jv, 0, jb) +
+ vec_e_view(iidx(jv, jb, 1), jk, iblk(jv, jb, 1)) *
+ geofac_rot_view(jv, 1, jb) +
+ vec_e_view(iidx(jv, jb, 2), jk, iblk(jv, jb, 2)) *
+ geofac_rot_view(jv, 2, jb) +
+ vec_e_view(iidx(jv, jb, 3), jk, iblk(jv, jb, 3)) *
+ geofac_rot_view(jv, 3, jb) +
+ vec_e_view(iidx(jv, jb, 4), jk, iblk(jv, jb, 4)) *
+ geofac_rot_view(jv, 4, jb) +
+ vec_e_view(iidx(jv, jb, 5), jk, iblk(jv, jb, 5)) *
+ geofac_rot_view(jv, 5, jb);
+ });
+ }
+
+ if (!acc_async)
+ Kokkos::fence();
}
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
index c32ee12..36ed138 100644
--- a/src/horizontal/lib_divrot.hpp
+++ b/src/horizontal/lib_divrot.hpp
@@ -35,7 +35,7 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
template <typename T>
void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
@@ -44,7 +44,7 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
template <typename T>
void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T *p_coeff, int i_startblk,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
@@ -54,7 +54,7 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
template <typename T>
void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
+ const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
@@ -63,9 +63,53 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
template <typename T>
void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T *p_coeff, int i_startblk,
+ const T *lsq_moments, T &p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
bool lacc, int nblks_c, int lsq_dim_unk,
int lsq_dim_c);
+
+template <typename T>
+void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
+ const T *geofac_div, T &div_vec_c, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, int nlev, int nblks_c, int nblks_e);
+
+template <typename T>
+void div3d_2field(const T *vec_e, const int *cell_edge_idx,
+ const int *cell_edge_blk, const T *geofac_div, T &div_vec_c,
+ const T *in2, T &out2, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev,
+ int nproma, bool lacc, int nlev, int nblks_c, int nblks_e);
+
+template <typename T>
+void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
+ const T *geofac_div, const T *f4din, T &f4dout, int dim4d,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in,
+ const int *slev, const int *elev, int nproma, bool lacc, int nlev,
+ int nblks_c, int nblks_e);
+
+template <typename T>
+void div_avg(const T *vec_e, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const int *cell_edge_idx,
+ const int *cell_edge_blk, const T *geofac_div, const T *avg_coeff,
+ T &div_vec_c, const T *opt_in2, T &opt_out2,
+ const int *i_startblk_in, const int *i_endblk_in,
+ const int *i_startidx_in, const int *i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, bool l_limited_area,
+ bool l2fields, bool lacc, int nlev, int nblks_c, int nblks_e);
+
+template <typename T>
+void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
+ const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, int nlev, int nblks_e, int nblks_v);
+
+template <typename T>
+void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
+ const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma, bool lacc,
+ bool acc_async, int nlev, int nblks_e, int nblks_v);
--
GitLab
From 9560ca72ebca7ac290d1e97fda36aa3a74dd2fea Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Wed, 26 Feb 2025 11:07:20 +0100
Subject: [PATCH 05/35] Fix typo
---
_typos.toml | 1 +
1 file changed, 1 insertion(+)
diff --git a/_typos.toml b/_typos.toml
index 4fe4968..58a18ef 100644
--- a/_typos.toml
+++ b/_typos.toml
@@ -1,6 +1,7 @@
[default]
extend-ignore-re = [
".*_pn",
+ "f4dout_*",
]
extend-ignore-words-re = [
"Comput",
--
GitLab
From ea4fc0b5e7bbe35b051fc732cbb875ee628e2105 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Thu, 27 Feb 2025 14:41:29 +0100
Subject: [PATCH 06/35] Fix bug and add first test
---
src/horizontal/lib_divrot.cpp | 76 +++++++++-----
src/horizontal/lib_divrot.hpp | 16 +--
test/c/CMakeLists.txt | 4 +-
test/c/test_horizontal_divrot.cpp | 159 ++++++++++++++++++++++++++++++
4 files changed, 220 insertions(+), 35 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 3586b03..c615a42 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -9,19 +9,20 @@
// SPDX-License-Identifier: BSD-3-Clause
// ---------------------------------------------------------------
+#include <iostream>
#include <lib_divrot.hpp>
#include <support/mo_lib_loopindices.hpp>
#include <vector>
template <typename T>
-void recon_lsq_cell_l_(const T *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const T *lsq_qtmat_c,
- const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, T &p_coeff, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- int slev, int elev, int nproma, bool l_consv, bool lacc,
- bool acc_async, int nblks_c, int lsq_dim_unk,
- int lsq_dim_c) {
+void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const T *lsq_qtmat_c,
+ const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ int slev, int elev, int nproma, bool l_consv, bool lacc,
+ bool acc_async, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
@@ -36,14 +37,14 @@ void recon_lsq_cell_l_(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT1D z_d(3);
- UnmanagedT1D z_qt_times_d(2);
+ Kokkos::View<T *> z_d("z_d", 3);
+ Kokkos::View<T *> z_qt_times_d("z_qt_times_d", 2);
UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
- UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT4D p_coeff_view(p_coeff);
+ UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -102,13 +103,32 @@ void recon_lsq_cell_l_(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::fence();
}
+template void
+recon_lsq_cell_l<float>(const float *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const float *lsq_qtmat_c,
+ const float *lsq_rmat_rdiag_c,
+ const float *lsq_rmat_utri_c, const float *lsq_moments,
+ float *p_coeff, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev,
+ int nproma, bool l_consv, bool lacc, bool acc_async,
+ int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c);
+
+template void recon_lsq_cell_l<double>(
+ const double *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const double *lsq_qtmat_c,
+ const double *lsq_rmat_rdiag_c, const double *lsq_rmat_utri_c,
+ const double *lsq_moments, double *p_coeff, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev,
+ int lsq_dim_unk, int lsq_dim_c);
+
template <typename T>
void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const T *lsq_pseudoinv,
const T *lsq_moments, T &p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, bool l_consv,
- bool lacc, bool acc_async, int nblks_c,
+ bool lacc, bool acc_async, int nblks_c, int nlev,
int lsq_dim_unk, int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
@@ -129,8 +149,8 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
- UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT4D p_coeff_view(p_coeff);
+ UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -186,7 +206,8 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c) {
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
@@ -209,8 +230,8 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
- UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT4D p_coeff_view(p_coeff);
+ UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -351,7 +372,7 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk,
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
@@ -372,8 +393,8 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
- UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT4D p_coeff_view(p_coeff);
+ UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -491,7 +512,8 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c) {
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
@@ -514,8 +536,8 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
- UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT4D p_coeff_view(p_coeff);
+ UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -734,7 +756,7 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk,
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
@@ -755,8 +777,8 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
- UnmanagedConstT3D p_cc_view(p_cc);
- UnmanagedT4D p_coeff_view(p_coeff);
+ UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
index 36ed138..db60b29 100644
--- a/src/horizontal/lib_divrot.hpp
+++ b/src/horizontal/lib_divrot.hpp
@@ -17,10 +17,10 @@ template <typename T>
void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const T *lsq_qtmat_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, T &p_coeff, int i_startblk,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, bool l_consv, bool lacc,
- bool acc_async, int nblks_c, int lsq_dim_unk,
+ bool acc_async, int nblks_c, int nlev, int lsq_dim_unk,
int lsq_dim_c);
template <typename T>
@@ -29,7 +29,7 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
const T *lsq_moments, T &p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, bool l_consv,
- bool lacc, bool acc_async, int nblks_c,
+ bool lacc, bool acc_async, int nblks_c, int nlev,
int lsq_dim_unk, int lsq_dim_c);
template <typename T>
@@ -39,7 +39,8 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c);
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
template <typename T>
void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
@@ -48,7 +49,7 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk,
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
int lsq_dim_c);
template <typename T>
@@ -58,7 +59,8 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk, int lsq_dim_c);
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
template <typename T>
void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
@@ -67,7 +69,7 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int lsq_dim_unk,
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
int lsq_dim_c);
template <typename T>
diff --git a/test/c/CMakeLists.txt b/test/c/CMakeLists.txt
index c9320cb..cf68892 100644
--- a/test/c/CMakeLists.txt
+++ b/test/c/CMakeLists.txt
@@ -27,8 +27,9 @@ endif()
set(SOURCES
main.cpp
+ test_horizontal_divrot.cpp
test_tdma_solver.cpp
- test_interpolation_vector.cpp
+ # test_interpolation_vector.cpp
test_intp_rbf.cpp
test_interpolation_scalar.cpp
)
@@ -40,6 +41,7 @@ target_link_libraries(iconmath_test_c
PUBLIC
iconmath-support
iconmath-interpolation
+ iconmath-horizontal
PRIVATE
gtest_main
Kokkos::kokkos
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index e69de29..1915fa4 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -0,0 +1,159 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#include <Kokkos_Core.hpp>
+#include <gtest/gtest.h>
+#include <horizontal/lib_divrot.hpp>
+#include <vector>
+
+// Template helpers for combining multiple dimension array sizes
+template <typename... Ts> size_t dim_combine(Ts... dims) { return 0; }
+template <typename T> size_t dim_combine(T dim) {
+ return static_cast<size_t>(dim);
+}
+template <typename T, typename... Ts> size_t dim_combine(T dim, Ts... dims) {
+ return static_cast<size_t>(dim) * dim_combine(dims...);
+}
+
+// Enum class for the reconstruction method
+enum class ReconstructionMethod {
+ linear,
+ quadratic,
+ cubic,
+};
+
+// Template function for LayoutLeft ID access in compile time
+template <class T, auto> using always_t = T;
+template <int... Dims> int At_impl(always_t<int, Dims>... ids) { return 0; }
+template <int LastDim> int At_impl(int prefix, int id) { return id * prefix; }
+template <int FirstDim, int... Dims>
+constexpr int At_impl(int prefix, int id, always_t<int, Dims>... ids) {
+ return id * prefix + At_impl<Dims...>(prefix * FirstDim, ids...);
+}
+template <int FirstDim, int... Dims>
+// At<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
+// LayoutLeft
+int At(int id, always_t<int, Dims>... ids) {
+ return id + At_impl<Dims...>(FirstDim, ids...);
+}
+
+typedef ::testing::Types<float, double> ValueTypes;
+
+template <typename ValueType>
+class HorizontalDivrotTest : public ::testing::Test {
+protected:
+ // [lsq_dim_c, lsq_dim_unk]
+ static constexpr std::tuple<int, int>
+ init_lsq_dim(ReconstructionMethod method) {
+ switch (method) {
+ case ReconstructionMethod::linear:
+ return std::make_tuple(3, 2);
+ case ReconstructionMethod::quadratic:
+ return std::make_tuple(9, 5);
+ case ReconstructionMethod::cubic:
+ return std::make_tuple(9, 9);
+ }
+ }
+
+ // Constant dimensions.
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 1; // number of vertical levels
+ static constexpr int nblks_c = 1; // number of cell blocks (for p_e_in)
+ static constexpr std::tuple<int, int> lsq_dim =
+ init_lsq_dim(ReconstructionMethod::linear);
+ static constexpr int lsq_dim_c = std::get<0>(lsq_dim);
+ static constexpr int lsq_dim_unk = std::get<1>(lsq_dim);
+
+ // Parameter values.
+ int i_startblk = 0;
+ int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
+ int i_startidx_in = 0;
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ int slev = 0;
+ int elev = nlev; // Full vertical range (0 .. nlev-1)
+ bool lacc = false; // Not using ACC-specific behavior.
+ bool acc_async = false; // No asynchronous execution.
+ bool l_consv = true; // No conservative correction
+
+ std::vector<ValueType> p_cc;
+ std::vector<int> cell_neighbor_idx;
+ std::vector<int> cell_neighbor_blk;
+ std::vector<ValueType> lsq_qtmat_c;
+ std::vector<ValueType> lsq_rmat_rdiag_c;
+ std::vector<ValueType> lsq_rmat_utri_c;
+ std::vector<ValueType> lsq_moments;
+ std::vector<ValueType> p_coeff;
+
+ HorizontalDivrotTest() {
+ p_cc.resize(dim_combine(nproma, nlev, nblks_c));
+ cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, 3));
+ cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, 3));
+ lsq_qtmat_c.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
+ lsq_rmat_rdiag_c.resize(dim_combine(nproma, lsq_dim_unk, nblks_c));
+ lsq_rmat_utri_c.resize(dim_combine(
+ nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c));
+ lsq_moments.resize(dim_combine(nproma, nblks_c, lsq_dim_unk));
+ p_coeff.resize(dim_combine(lsq_dim_c, nproma, nlev, nblks_c));
+ }
+};
+
+TYPED_TEST_SUITE(HorizontalDivrotTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
+ this->init_lsq_dim(ReconstructionMethod::linear);
+ constexpr int nproma = TestFixture::nproma;
+ constexpr int nlev = TestFixture::nlev;
+ constexpr int nblks_c = TestFixture::nblks_c;
+ constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
+ constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[At<nproma, nlev, nblks_c>(i, 0, 0)] = (TypeParam)(i + 1);
+
+ this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 1)] = i;
+ this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 2)] = i;
+ for (int j = 0; j < 3; ++j) {
+ this->cell_neighbor_blk[At<nproma, nblks_c, 3>(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_qtmat_c[At<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 0, j,
+ 0)] = 1.0;
+ this->lsq_qtmat_c[At<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 1, j,
+ 0)] = 0.5;
+ this->p_coeff[At<lsq_dim_c, nproma, nlev, nblks_c>(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_rmat_rdiag_c[At<nproma, lsq_dim_unk, nblks_c>(i, 0, 0)] = 2.0;
+ this->lsq_rmat_rdiag_c[At<nproma, lsq_dim_unk, nblks_c>(i, 1, 0)] = 2.0;
+ this->lsq_rmat_utri_c
+ [At<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>(
+ i, 0, 0)] = 0.1;
+
+ this->lsq_moments[At<nproma, nblks_c, lsq_dim_unk>(i, 0, 0)] = 0.2;
+ this->lsq_moments[At<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
+ }
+
+ recon_lsq_cell_l<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ EXPECT_NEAR(this->p_coeff[0], 0.34, 1e-6);
+ EXPECT_NEAR(this->p_coeff[1], 1.8, 1e-6);
+ EXPECT_NEAR(this->p_coeff[2], 1.0, 1e-6);
+}
--
GitLab
From a254c39e98c857e638e8213822ffabc5a91073d5 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Thu, 27 Feb 2025 14:42:51 +0100
Subject: [PATCH 07/35] Reverse commented file
---
test/c/CMakeLists.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/test/c/CMakeLists.txt b/test/c/CMakeLists.txt
index cf68892..18c2710 100644
--- a/test/c/CMakeLists.txt
+++ b/test/c/CMakeLists.txt
@@ -29,7 +29,7 @@ set(SOURCES
main.cpp
test_horizontal_divrot.cpp
test_tdma_solver.cpp
- # test_interpolation_vector.cpp
+ test_interpolation_vector.cpp
test_intp_rbf.cpp
test_interpolation_scalar.cpp
)
--
GitLab
From 79c31189b41cb725141020b0ffc4eb0d72ceac7d Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Thu, 27 Feb 2025 15:28:43 +0100
Subject: [PATCH 08/35] Add comments for templates
---
test/c/test_horizontal_divrot.cpp | 37 +++++++++++++++++++++++++++----
1 file changed, 33 insertions(+), 4 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 1915fa4..26ba118 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -14,11 +14,15 @@
#include <horizontal/lib_divrot.hpp>
#include <vector>
-// Template helpers for combining multiple dimension array sizes
+// Template helpers for combining multiple dimension array sizes.
+// The base function of dimension combine. Should not be used.
template <typename... Ts> size_t dim_combine(Ts... dims) { return 0; }
+// Template specialization of only one dimension, returns the dimension itself.
template <typename T> size_t dim_combine(T dim) {
return static_cast<size_t>(dim);
}
+// Template specialization of picking out the first dimension. The combined
+// dimension is the first dimension times the combined dimension of the rest.
template <typename T, typename... Ts> size_t dim_combine(T dim, Ts... dims) {
return static_cast<size_t>(dim) * dim_combine(dims...);
}
@@ -31,16 +35,41 @@ enum class ReconstructionMethod {
};
// Template function for LayoutLeft ID access in compile time
+// For example, a multi-dimensional array A of dimensions <2, 3, 4, 5> gets its
+// corresponding vector id (LayoutLeft) by
+// At<2, 3, 4, 5>(id1, id2, id3, id4).
+// The At_impl then adds the id from beginning to the end and pass the id prefix
+// to the next recursive At_impl function. In this example,
+// At<2, 3, 4, 5>(id1, id2, id3, id4) {
+// return id1 + At_impl<3, 4, 5>(2, id2, id3, id4);
+// }
+// At_impl<3, 4, 5>(2, id2, id3, id4) {
+// return id2 * 2 + At_impl<4, 5>(2 * 3, id3, id4);
+// }
+// At_impl<4, 5>(2 * 3, id3, id4) {
+// return id3 * 2 * 3 + At_impl<5>(2 * 3 * 4, id4);
+// }
+// At_impl<5>(2 * 3 * 4, id4) {
+// return id4 * 2 * 3 * 4;
+// }
+// Which gives
+// At<2, 3, 4, 5>(id1, id2, id3, id4) = id1 + id2 * 2 +
+// id3 * 2 * 3 + id4 * 2 * 3 * 4
+// Helper type converting integer numbers to int
template <class T, auto> using always_t = T;
+// Base function of At_impl. Should not be used.
template <int... Dims> int At_impl(always_t<int, Dims>... ids) { return 0; }
+// Template specialization of the last ID
template <int LastDim> int At_impl(int prefix, int id) { return id * prefix; }
+// Template specialization of At_impl, accumulate the return value using the
+// first id and pass the prefix to the next recursive At_impl function.
template <int FirstDim, int... Dims>
-constexpr int At_impl(int prefix, int id, always_t<int, Dims>... ids) {
+int At_impl(int prefix, int id, always_t<int, Dims>... ids) {
return id * prefix + At_impl<Dims...>(prefix * FirstDim, ids...);
}
-template <int FirstDim, int... Dims>
// At<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
-// LayoutLeft
+// LayoutLeft. Use this function instead of At_impl.
+template <int FirstDim, int... Dims>
int At(int id, always_t<int, Dims>... ids) {
return id + At_impl<Dims...>(FirstDim, ids...);
}
--
GitLab
From 35c5d8122cb9829ebbc46f27e0669cebca63289e Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Thu, 27 Feb 2025 16:04:35 +0100
Subject: [PATCH 09/35] Make reconstruction method templated
---
test/c/test_horizontal_divrot.cpp | 25 ++++++++++++++++++-------
1 file changed, 18 insertions(+), 7 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 26ba118..082afa3 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -74,11 +74,21 @@ int At(int id, always_t<int, Dims>... ids) {
return id + At_impl<Dims...>(FirstDim, ids...);
}
-typedef ::testing::Types<float, double> ValueTypes;
+// ValueType struct for compute precision and reconstruction method.
+template <typename ValueType, int ReconMethod> struct DivrotType {
+ using type = ValueType;
+ static constexpr int get_recon_method() { return ReconMethod; };
+};
+
+typedef ::testing::Types<
+ DivrotType<float, static_cast<int>(ReconstructionMethod::linear)>,
+ DivrotType<double, static_cast<int>(ReconstructionMethod::linear)>>
+ ValueTypes;
-template <typename ValueType>
+template <typename ValueTypes>
class HorizontalDivrotTest : public ::testing::Test {
protected:
+ using ValueType = typename ValueTypes::type;
// [lsq_dim_c, lsq_dim_unk]
static constexpr std::tuple<int, int>
init_lsq_dim(ReconstructionMethod method) {
@@ -96,8 +106,8 @@ protected:
static constexpr int nproma = 3; // inner loop length
static constexpr int nlev = 1; // number of vertical levels
static constexpr int nblks_c = 1; // number of cell blocks (for p_e_in)
- static constexpr std::tuple<int, int> lsq_dim =
- init_lsq_dim(ReconstructionMethod::linear);
+ static constexpr std::tuple<int, int> lsq_dim = init_lsq_dim(
+ static_cast<ReconstructionMethod>(ValueTypes::get_recon_method()));
static constexpr int lsq_dim_c = std::get<0>(lsq_dim);
static constexpr int lsq_dim_unk = std::get<1>(lsq_dim);
@@ -137,7 +147,8 @@ protected:
TYPED_TEST_SUITE(HorizontalDivrotTest, ValueTypes);
TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
- this->init_lsq_dim(ReconstructionMethod::linear);
+ using ValueType = typename TestFixture::ValueType;
+
constexpr int nproma = TestFixture::nproma;
constexpr int nlev = TestFixture::nlev;
constexpr int nblks_c = TestFixture::nblks_c;
@@ -146,7 +157,7 @@ TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
// Initialization
for (int i = 0; i < nproma; ++i) {
- this->p_cc[At<nproma, nlev, nblks_c>(i, 0, 0)] = (TypeParam)(i + 1);
+ this->p_cc[At<nproma, nlev, nblks_c>(i, 0, 0)] = (i + 1);
this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 0)] = (i + 1) % nproma;
this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 1)] = i;
@@ -173,7 +184,7 @@ TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
this->lsq_moments[At<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
}
- recon_lsq_cell_l<TypeParam>(
+ recon_lsq_cell_l<ValueType>(
this->p_cc.data(), this->cell_neighbor_idx.data(),
this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
--
GitLab
From b9e1e645b9a6cd43cdaf22c9978db393643cc823 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Fri, 28 Feb 2025 18:55:40 +0100
Subject: [PATCH 10/35] Instantiate and fix bugs
---
src/horizontal/lib_divrot.cpp | 437 +++++++++++++++++++++++-----------
1 file changed, 297 insertions(+), 140 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index c615a42..9dce2e4 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -29,8 +29,6 @@ void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstT4D;
- typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT1D;
typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
@@ -125,7 +123,7 @@ template void recon_lsq_cell_l<double>(
template <typename T>
void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const T *lsq_pseudoinv,
- const T *lsq_moments, T &p_coeff, int i_startblk,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, bool l_consv,
bool lacc, bool acc_async, int nblks_c, int nlev,
@@ -136,15 +134,13 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstT4D;
- typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT1D;
typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT1D z_b(3);
+ Kokkos::View<T *> z_b("z_b", 3);
UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
@@ -199,10 +195,26 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::fence();
}
+template void recon_lsq_cell_l_svd<float>(
+ const float *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const float *lsq_pseudoinv,
+ const float *lsq_moments, float *p_coeff, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev,
+ int lsq_dim_unk, int lsq_dim_c);
+
+template void recon_lsq_cell_l_svd<double>(
+ const double *p_cc, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const double *lsq_pseudoinv,
+ const double *lsq_moments, double *p_coeff, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev,
+ int lsq_dim_unk, int lsq_dim_c);
+
template <typename T>
void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
+ const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
@@ -214,18 +226,14 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstT4D;
- typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT1D;
- typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT3D;
typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT3D z_d(lsq_high_set_dim_c, nproma, elev);
- UnmanagedT1D z_qt_times_d(5);
+ Kokkos::View<T ***> z_d("z_d", lsq_high_set_dim_c, nproma, elev);
+ Kokkos::View<T *> z_qt_times_d("z_qt_times_d", 5);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
@@ -353,7 +361,7 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
ptr_rutri(jc, 8, jb) * p_coeff_view(4, jc, jk, jb) -
ptr_rutri(jc, 9, jb) * p_coeff_view(5, jc, jk, jb));
p_coeff_view(0, jc, jk, jb) =
- p_cc(jc, jk, jb) -
+ p_cc_view(jc, jk, jb) -
p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) -
p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) -
@@ -365,10 +373,28 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::fence();
}
+template void recon_lsq_cell_q<float>(
+ const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const float *lsq_rmat_rdiag_c, const float *lsq_rmat_utri_c,
+ const float *lsq_moments, const float *lsq_qtmat_c, float *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
+template void recon_lsq_cell_q<double>(
+ const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const double *lsq_rmat_rdiag_c, const double *lsq_rmat_utri_c,
+ const double *lsq_moments, const double *lsq_qtmat_c, double *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
template <typename T>
void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T &p_coeff, int i_startblk,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
@@ -380,15 +406,13 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstT4D;
- typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT3D;
typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT3D z_b(lsq_high_set_dim_c, nproma, elev);
+ Kokkos::View<T ***> z_b("z_b", lsq_high_set_dim_c, nproma, elev);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
@@ -505,10 +529,26 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
Kokkos::fence();
}
+template void recon_lsq_cell_q_svd<float>(
+ const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const float *lsq_pseudoinv, const float *lsq_moments, float *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
+template void recon_lsq_cell_q_svd<double>(
+ const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const double *lsq_pseudoinv, const double *lsq_moments, double *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
template <typename T>
void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
+ const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
@@ -520,18 +560,14 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstT4D;
- typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT1D;
- typedef Kokkos::View<T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT3D;
typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT3D z_d(lsq_high_set_dim_c, nproma, elev);
- UnmanagedT1D z_qt_times_d(9);
+ Kokkos::View<T ***> z_d("z_d", lsq_high_set_dim_c, nproma, elev);
+ Kokkos::View<T *> z_qt_times_d("z_qt_times_d", 9);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
@@ -733,7 +769,7 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
ptr_rutri(jc, 34, jb) * p_coeff_view(8, jc, jk, jb) +
ptr_rutri(jc, 35, jb) * p_coeff_view(9, jc, jk, jb)));
p_coeff_view(0, jc, jk, jb) =
- p_cc(jc, jk, jb) -
+ p_cc_view(jc, jk, jb) -
(p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) +
p_coeff_view(2, jc, jk, jb) * lsq_moments_view(jc, jb, 1) +
p_coeff_view(3, jc, jk, jb) * lsq_moments_view(jc, jb, 2) +
@@ -749,10 +785,28 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::fence();
}
+template void recon_lsq_cell_c<float>(
+ const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const float *lsq_rmat_rdiag_c, const float *lsq_rmat_utri_c,
+ const float *lsq_moments, const float *lsq_qtmat_c, float *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
+template void recon_lsq_cell_c<double>(
+ const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const double *lsq_rmat_rdiag_c, const double *lsq_rmat_utri_c,
+ const double *lsq_moments, const double *lsq_qtmat_c, double *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
template <typename T>
void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T &p_coeff, int i_startblk,
+ const T *lsq_moments, T *p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
int lsq_high_set_dim_c, bool l_limited_area,
@@ -764,15 +818,13 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
typedef Kokkos::View<const T ****, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstT4D;
- typedef Kokkos::View<T *, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
- UnmanagedT1D;
typedef Kokkos::View<T ****, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedT4D;
typedef Kokkos::View<const int ***, Kokkos::LayoutLeft,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- UnmanagedT1D z_b(9);
+ Kokkos::View<T *> z_b("z_b", 9);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
@@ -810,115 +862,115 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
Kokkos::parallel_for(
"recon_lsq_cell_c_svd_inner", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
- z_b(0, jc, jk) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
- p_cc_view(jc, jk, jb);
- z_b(1, jc, jk) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
- p_cc_view(jc, jk, jb);
- z_b(2, jc, jk) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
- p_cc_view(jc, jk, jb);
- z_b(3, jc, jk) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
- p_cc_view(jc, jk, jb);
- z_b(4, jc, jk) = p_cc_view(iidx(jc, jb, 4), jk, iblk(jc, jb, 4)) -
- p_cc_view(jc, jk, jb);
- z_b(5, jc, jk) = p_cc_view(iidx(jc, jb, 5), jk, iblk(jc, jb, 5)) -
- p_cc_view(jc, jk, jb);
- z_b(6, jc, jk) = p_cc_view(iidx(jc, jb, 6), jk, iblk(jc, jb, 6)) -
- p_cc_view(jc, jk, jb);
- z_b(7, jc, jk) = p_cc_view(iidx(jc, jb, 7), jk, iblk(jc, jb, 7)) -
- p_cc_view(jc, jk, jb);
- z_b(8, jc, jk) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
- p_cc_view(jc, jk, jb);
+ z_b(0) = p_cc_view(iidx(jc, jb, 0), jk, iblk(jc, jb, 0)) -
+ p_cc_view(jc, jk, jb);
+ z_b(1) = p_cc_view(iidx(jc, jb, 1), jk, iblk(jc, jb, 1)) -
+ p_cc_view(jc, jk, jb);
+ z_b(2) = p_cc_view(iidx(jc, jb, 2), jk, iblk(jc, jb, 2)) -
+ p_cc_view(jc, jk, jb);
+ z_b(3) = p_cc_view(iidx(jc, jb, 3), jk, iblk(jc, jb, 3)) -
+ p_cc_view(jc, jk, jb);
+ z_b(4) = p_cc_view(iidx(jc, jb, 4), jk, iblk(jc, jb, 4)) -
+ p_cc_view(jc, jk, jb);
+ z_b(5) = p_cc_view(iidx(jc, jb, 5), jk, iblk(jc, jb, 5)) -
+ p_cc_view(jc, jk, jb);
+ z_b(6) = p_cc_view(iidx(jc, jb, 6), jk, iblk(jc, jb, 6)) -
+ p_cc_view(jc, jk, jb);
+ z_b(7) = p_cc_view(iidx(jc, jb, 7), jk, iblk(jc, jb, 7)) -
+ p_cc_view(jc, jk, jb);
+ z_b(8) = p_cc_view(iidx(jc, jb, 8), jk, iblk(jc, jb, 8)) -
+ p_cc_view(jc, jk, jb);
p_coeff_view(9, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 8, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 8, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 8, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 8, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 8, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 8, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 8, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 8, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 8, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 8, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 8, 8, jb) * z_b(8);
p_coeff_view(8, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 7, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 7, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 7, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 7, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 7, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 7, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 7, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 7, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 7, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 7, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 7, 8, jb) * z_b(8);
p_coeff_view(7, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 6, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 6, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 6, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 6, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 6, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 6, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 6, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 6, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 6, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 6, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 6, 8, jb) * z_b(8);
p_coeff_view(6, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 5, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 5, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 5, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 5, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 5, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 5, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 5, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 5, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 5, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 5, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 5, 8, jb) * z_b(8);
p_coeff_view(5, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 4, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 4, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 4, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 4, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 4, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 4, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 4, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 4, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 4, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 4, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 4, 8, jb) * z_b(8);
p_coeff_view(4, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 3, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 3, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 3, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 3, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 3, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 3, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 3, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 3, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 3, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 3, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 3, 8, jb) * z_b(8);
p_coeff_view(3, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 2, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 2, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 2, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 2, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 2, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 2, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 2, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 2, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 2, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 2, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 2, 8, jb) * z_b(8);
p_coeff_view(2, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 1, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 1, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 1, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 1, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 1, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 1, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 1, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 1, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 1, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 1, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 1, 8, jb) * z_b(8);
p_coeff_view(1, jc, jk, jb) =
- lsq_pseudoinv_view(jc, 0, 0, jb) * z_b(0, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 1, jb) * z_b(1, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 2, jb) * z_b(2, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 3, jb) * z_b(3, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 4, jb) * z_b(4, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 5, jb) * z_b(5, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 6, jb) * z_b(6, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 7, jb) * z_b(7, jc, jk) +
- lsq_pseudoinv_view(jc, 0, 8, jb) * z_b(8, jc, jk);
+ lsq_pseudoinv_view(jc, 0, 0, jb) * z_b(0) +
+ lsq_pseudoinv_view(jc, 0, 1, jb) * z_b(1) +
+ lsq_pseudoinv_view(jc, 0, 2, jb) * z_b(2) +
+ lsq_pseudoinv_view(jc, 0, 3, jb) * z_b(3) +
+ lsq_pseudoinv_view(jc, 0, 4, jb) * z_b(4) +
+ lsq_pseudoinv_view(jc, 0, 5, jb) * z_b(5) +
+ lsq_pseudoinv_view(jc, 0, 6, jb) * z_b(6) +
+ lsq_pseudoinv_view(jc, 0, 7, jb) * z_b(7) +
+ lsq_pseudoinv_view(jc, 0, 8, jb) * z_b(8);
p_coeff_view(0, jc, jk, jb) =
p_cc_view(jc, jk, jb) -
p_coeff_view(1, jc, jk, jb) * lsq_moments_view(jc, jb, 0) -
@@ -936,6 +988,22 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
Kokkos::fence();
}
+template void recon_lsq_cell_c_svd<float>(
+ const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const float *lsq_pseudoinv, const float *lsq_moments, float *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
+template void recon_lsq_cell_c_svd<double>(
+ const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
+ const double *lsq_pseudoinv, const double *lsq_moments, double *p_coeff,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
+ int lsq_dim_c);
+
template <typename T>
void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
const T *geofac_div, T *div_vec_c, int i_startblk, int i_endblk,
@@ -978,10 +1046,24 @@ void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
}
}
+template void div3d<float>(const float *vec_e, const int *cell_edge_idx,
+ const int *cell_edge_blk, const float *geofac_div,
+ float *div_vec_c, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, bool lacc, int nlev,
+ int nblks_c, int nblks_e);
+
+template void div3d<double>(const double *vec_e, const int *cell_edge_idx,
+ const int *cell_edge_blk, const double *geofac_div,
+ double *div_vec_c, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, bool lacc, int nlev,
+ int nblks_c, int nblks_e);
+
template <typename T>
void div3d_2field(const T *vec_e, const int *cell_edge_idx,
- const int *cell_edge_blk, const T *geofac_div, T &div_vec_c,
- const T *in2, T &out2, int i_startblk, int i_endblk,
+ const int *cell_edge_blk, const T *geofac_div, T *div_vec_c,
+ const T *in2, T *out2, int i_startblk, int i_endblk,
int i_startidx_in, int i_endidx_in, int slev, int elev,
int nproma, bool lacc, int nlev, int nblks_c, int nblks_e) {
// Wrap raw pointers in unmanaged Kokkos Views.
@@ -1033,9 +1115,23 @@ void div3d_2field(const T *vec_e, const int *cell_edge_idx,
}
}
+template void div3d_2field<float>(
+ const float *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
+ const float *geofac_div, float *div_vec_c, const float *in2, float *out2,
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
+ int elev, int nproma, bool lacc, int nlev, int nblks_c, int nblks_e);
+
+template void
+div3d_2field<double>(const double *vec_e, const int *cell_edge_idx,
+ const int *cell_edge_blk, const double *geofac_div,
+ double *div_vec_c, const double *in2, double *out2,
+ int i_startblk, int i_endblk, int i_startidx_in,
+ int i_endidx_in, int slev, int elev, int nproma, bool lacc,
+ int nlev, int nblks_c, int nblks_e);
+
template <typename T>
void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
- const T *geofac_div, const T *f4din, T &f4dout, int dim4d,
+ const T *geofac_div, const T *f4din, T *f4dout, int dim4d,
int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in,
const int *slev, const int *elev, int nproma, bool lacc, int nlev,
int nblks_c, int nblks_e) {
@@ -1084,11 +1180,25 @@ void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
}
}
+template void div4d<float>(const int *cell_edge_idx, const int *cell_edge_blk,
+ const float *geofac_div, const float *f4din,
+ float *f4dout, int dim4d, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ const int *slev, const int *elev, int nproma,
+ bool lacc, int nlev, int nblks_c, int nblks_e);
+
+template void div4d<double>(const int *cell_edge_idx, const int *cell_edge_blk,
+ const double *geofac_div, const double *f4din,
+ double *f4dout, int dim4d, int i_startblk,
+ int i_endblk, int i_startidx_in, int i_endidx_in,
+ const int *slev, const int *elev, int nproma,
+ bool lacc, int nlev, int nblks_c, int nblks_e);
+
template <typename T>
void div_avg(const T *vec_e, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const int *cell_edge_idx,
const int *cell_edge_blk, const T *geofac_div, const T *avg_coeff,
- T &div_vec_c, const T *opt_in2, T &opt_out2,
+ T *div_vec_c, const T *opt_in2, T *opt_out2,
const int *i_startblk_in, const int *i_endblk_in,
const int *i_startidx_in, const int *i_endidx_in, int slev,
int elev, int nproma, int patch_id, bool l_limited_area,
@@ -1117,8 +1227,8 @@ void div_avg(const T *vec_e, const int *cell_neighbor_idx,
UnmanagedConstT3D opt_in2_view(opt_in2, nproma, nlev, nblks_e);
UnmanagedT3D opt_out2_view(opt_out2, nproma, nlev, nblks_c);
- UnmanagedT3D aux_c(nproma, nlev, nblks_c);
- UnmanagedT3D aux_c2(nproma, nlev, nblks_c);
+ Kokkos::View<T ***> aux_c("aux_c", nproma, nlev, nblks_c);
+ Kokkos::View<T ***> aux_c2("aux_c2", nproma, nlev, nblks_c);
int i_startblk = i_startblk_in[0];
int i_endblk = i_endblk_in[0];
@@ -1142,11 +1252,11 @@ void div_avg(const T *vec_e, const int *cell_neighbor_idx,
vec_e_view(ieidx(jc, jb, 2), jk, ieblk(jc, jb, 2)) *
geofac_div_view(jc, 2, jb);
aux_c2(jc, jk, jb) =
- opt_in2(ieidx(jc, jb, 0), jk, ieblk(jc, jb, 0)) *
+ opt_in2_view(ieidx(jc, jb, 0), jk, ieblk(jc, jb, 0)) *
geofac_div_view(jc, 0, jb) +
- opt_in2(ieidx(jc, jb, 1), jk, ieblk(jc, jb, 1)) *
+ opt_in2_view(ieidx(jc, jb, 1), jk, ieblk(jc, jb, 1)) *
geofac_div_view(jc, 1, jb) +
- opt_in2(ieidx(jc, jb, 2), jk, ieblk(jc, jb, 2)) *
+ opt_in2_view(ieidx(jc, jb, 2), jk, ieblk(jc, jb, 2)) *
geofac_div_view(jc, 2, jb);
});
}
@@ -1263,9 +1373,32 @@ void div_avg(const T *vec_e, const int *cell_neighbor_idx,
}
}
+template void div_avg<float>(const float *vec_e, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk,
+ const int *cell_edge_idx, const int *cell_edge_blk,
+ const float *geofac_div, const float *avg_coeff,
+ float *div_vec_c, const float *opt_in2,
+ float *opt_out2, const int *i_startblk_in,
+ const int *i_endblk_in, const int *i_startidx_in,
+ const int *i_endidx_in, int slev, int elev,
+ int nproma, int patch_id, bool l_limited_area,
+ bool l2fields, bool lacc, int nlev, int nblks_c,
+ int nblks_e);
+
+template void
+div_avg<double>(const double *vec_e, const int *cell_neighbor_idx,
+ const int *cell_neighbor_blk, const int *cell_edge_idx,
+ const int *cell_edge_blk, const double *geofac_div,
+ const double *avg_coeff, double *div_vec_c,
+ const double *opt_in2, double *opt_out2,
+ const int *i_startblk_in, const int *i_endblk_in,
+ const int *i_startidx_in, const int *i_endidx_in, int slev,
+ int elev, int nproma, int patch_id, bool l_limited_area,
+ bool l2fields, bool lacc, int nlev, int nblks_c, int nblks_e);
+
template <typename T>
void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
- const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
+ const int *vert_edge_blk, const T *geofac_rot, T *rot_vec,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma,
bool lacc, int nlev, int nblks_e, int nblks_v) {
@@ -1314,9 +1447,21 @@ void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
}
}
+template void rot_vertex_atmos<float>(
+ const float *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
+ const float *geofac_rot, float *rot_vec, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, int nlev, int nblks_e, int nblks_v);
+
+template void rot_vertex_atmos<double>(
+ const double *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
+ const double *geofac_rot, double *rot_vec, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, int nlev, int nblks_e, int nblks_v);
+
template <typename T>
void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
- const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
+ const int *vert_edge_blk, const T *geofac_rot, T *rot_vec,
int i_startblk, int i_endblk, int i_startidx_in,
int i_endidx_in, int slev, int elev, int nproma, bool lacc,
bool acc_async, int nlev, int nblks_e, int nblks_v) {
@@ -1367,3 +1512,15 @@ void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
if (!acc_async)
Kokkos::fence();
}
+
+template void rot_vertex_ri<float>(
+ const float *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
+ const float *geofac_rot, float *rot_vec, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, bool acc_async, int nlev, int nblks_e, int nblks_v);
+
+template void rot_vertex_ri<double>(
+ const double *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
+ const double *geofac_rot, double *rot_vec, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
+ bool lacc, bool acc_async, int nlev, int nblks_e, int nblks_v);
--
GitLab
From 5b6606842ed56a87c83b365eb79c6d1cdee3a08e Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Sat, 1 Mar 2025 10:47:43 +0100
Subject: [PATCH 11/35] Separate linear, quadratic, and cubic tests
---
test/c/test_horizontal_divrot.cpp | 65 ++++++++++++++++++++++---------
1 file changed, 46 insertions(+), 19 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 082afa3..8bd782c 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -74,21 +74,9 @@ int At(int id, always_t<int, Dims>... ids) {
return id + At_impl<Dims...>(FirstDim, ids...);
}
-// ValueType struct for compute precision and reconstruction method.
-template <typename ValueType, int ReconMethod> struct DivrotType {
- using type = ValueType;
- static constexpr int get_recon_method() { return ReconMethod; };
-};
-
-typedef ::testing::Types<
- DivrotType<float, static_cast<int>(ReconstructionMethod::linear)>,
- DivrotType<double, static_cast<int>(ReconstructionMethod::linear)>>
- ValueTypes;
-
-template <typename ValueTypes>
+template <typename ValueType, int ReconMethod>
class HorizontalDivrotTest : public ::testing::Test {
protected:
- using ValueType = typename ValueTypes::type;
// [lsq_dim_c, lsq_dim_unk]
static constexpr std::tuple<int, int>
init_lsq_dim(ReconstructionMethod method) {
@@ -106,8 +94,8 @@ protected:
static constexpr int nproma = 3; // inner loop length
static constexpr int nlev = 1; // number of vertical levels
static constexpr int nblks_c = 1; // number of cell blocks (for p_e_in)
- static constexpr std::tuple<int, int> lsq_dim = init_lsq_dim(
- static_cast<ReconstructionMethod>(ValueTypes::get_recon_method()));
+ static constexpr std::tuple<int, int> lsq_dim =
+ init_lsq_dim(static_cast<ReconstructionMethod>(ReconMethod));
static constexpr int lsq_dim_c = std::get<0>(lsq_dim);
static constexpr int lsq_dim_unk = std::get<1>(lsq_dim);
@@ -144,11 +132,32 @@ protected:
}
};
-TYPED_TEST_SUITE(HorizontalDivrotTest, ValueTypes);
+template <typename ValueType>
+class HorizontalDivrotLinearTest
+ : public HorizontalDivrotTest<
+ ValueType, static_cast<int>(ReconstructionMethod::linear)> {};
-TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
- using ValueType = typename TestFixture::ValueType;
+template <typename ValueType>
+class HorizontalDivrotQuadraticTest
+ : public HorizontalDivrotTest<
+ ValueType, static_cast<int>(ReconstructionMethod::quadratic)> {};
+
+template <typename ValueType>
+class HorizontalDivrotCubicTest
+ : public HorizontalDivrotTest<ValueType, static_cast<int>(
+ ReconstructionMethod::cubic)> {
+};
+
+typedef ::testing::Types<float, double> ValueTypes;
+
+TYPED_TEST_SUITE(HorizontalDivrotLinearTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivrotLinearTest, TestLsqDimensions) {
+ EXPECT_EQ(TestFixture::lsq_dim_c, 3);
+ EXPECT_EQ(TestFixture::lsq_dim_unk, 2);
+}
+TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
constexpr int nproma = TestFixture::nproma;
constexpr int nlev = TestFixture::nlev;
constexpr int nblks_c = TestFixture::nblks_c;
@@ -184,7 +193,7 @@ TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
this->lsq_moments[At<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
}
- recon_lsq_cell_l<ValueType>(
+ recon_lsq_cell_l<TypeParam>(
this->p_cc.data(), this->cell_neighbor_idx.data(),
this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
@@ -197,3 +206,21 @@ TYPED_TEST(HorizontalDivrotTest, TestReconLsqCellLinear) {
EXPECT_NEAR(this->p_coeff[1], 1.8, 1e-6);
EXPECT_NEAR(this->p_coeff[2], 1.0, 1e-6);
}
+
+typedef ::testing::Types<float, double> ValueTypes;
+
+TYPED_TEST_SUITE(HorizontalDivrotQuadraticTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivrotQuadraticTest, TestLsqDimensions) {
+ EXPECT_EQ(TestFixture::lsq_dim_c, 9);
+ EXPECT_EQ(TestFixture::lsq_dim_unk, 5);
+}
+
+typedef ::testing::Types<float, double> ValueTypes;
+
+TYPED_TEST_SUITE(HorizontalDivrotCubicTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
+ EXPECT_EQ(TestFixture::lsq_dim_c, 9);
+ EXPECT_EQ(TestFixture::lsq_dim_unk, 9);
+}
--
GitLab
From efc6beef573e980f39a9df286c50f9ba39542401 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Mon, 3 Mar 2025 09:48:25 +0100
Subject: [PATCH 12/35] Use snake case for at functions
---
test/c/test_horizontal_divrot.cpp | 79 ++++++++++++++++---------------
1 file changed, 41 insertions(+), 38 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 8bd782c..a99556f 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -37,41 +37,41 @@ enum class ReconstructionMethod {
// Template function for LayoutLeft ID access in compile time
// For example, a multi-dimensional array A of dimensions <2, 3, 4, 5> gets its
// corresponding vector id (LayoutLeft) by
-// At<2, 3, 4, 5>(id1, id2, id3, id4).
-// The At_impl then adds the id from beginning to the end and pass the id prefix
-// to the next recursive At_impl function. In this example,
-// At<2, 3, 4, 5>(id1, id2, id3, id4) {
-// return id1 + At_impl<3, 4, 5>(2, id2, id3, id4);
+// at<2, 3, 4, 5>(id1, id2, id3, id4).
+// The at_impl then adds the id from beginning to the end and pass the id prefix
+// to the next recursive at_impl function. In this example,
+// at<2, 3, 4, 5>(id1, id2, id3, id4) {
+// return id1 + at_impl<3, 4, 5>(2, id2, id3, id4);
// }
-// At_impl<3, 4, 5>(2, id2, id3, id4) {
-// return id2 * 2 + At_impl<4, 5>(2 * 3, id3, id4);
+// at_impl<3, 4, 5>(2, id2, id3, id4) {
+// return id2 * 2 + at_impl<4, 5>(2 * 3, id3, id4);
// }
-// At_impl<4, 5>(2 * 3, id3, id4) {
-// return id3 * 2 * 3 + At_impl<5>(2 * 3 * 4, id4);
+// at_impl<4, 5>(2 * 3, id3, id4) {
+// return id3 * 2 * 3 + at_impl<5>(2 * 3 * 4, id4);
// }
-// At_impl<5>(2 * 3 * 4, id4) {
+// at_impl<5>(2 * 3 * 4, id4) {
// return id4 * 2 * 3 * 4;
// }
// Which gives
-// At<2, 3, 4, 5>(id1, id2, id3, id4) = id1 + id2 * 2 +
+// at<2, 3, 4, 5>(id1, id2, id3, id4) = id1 + id2 * 2 +
// id3 * 2 * 3 + id4 * 2 * 3 * 4
// Helper type converting integer numbers to int
template <class T, auto> using always_t = T;
-// Base function of At_impl. Should not be used.
-template <int... Dims> int At_impl(always_t<int, Dims>... ids) { return 0; }
+// Base function of at_impl. Should not be used.
+template <int... Dims> int at_impl(always_t<int, Dims>... ids) { return 0; }
// Template specialization of the last ID
-template <int LastDim> int At_impl(int prefix, int id) { return id * prefix; }
-// Template specialization of At_impl, accumulate the return value using the
-// first id and pass the prefix to the next recursive At_impl function.
+template <int LastDim> int at_impl(int prefix, int id) { return id * prefix; }
+// Template specialization of at_impl, accumulate the return value using the
+// first id and pass the prefix to the next recursive at_impl function.
template <int FirstDim, int... Dims>
-int At_impl(int prefix, int id, always_t<int, Dims>... ids) {
- return id * prefix + At_impl<Dims...>(prefix * FirstDim, ids...);
+int at_impl(int prefix, int id, always_t<int, Dims>... ids) {
+ return id * prefix + at_impl<Dims...>(prefix * FirstDim, ids...);
}
-// At<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
-// LayoutLeft. Use this function instead of At_impl.
+// at<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
+// LayoutLeft. Use this function instead of at_impl.
template <int FirstDim, int... Dims>
-int At(int id, always_t<int, Dims>... ids) {
- return id + At_impl<Dims...>(FirstDim, ids...);
+int at(int id, always_t<int, Dims>... ids) {
+ return id + at_impl<Dims...>(FirstDim, ids...);
}
template <typename ValueType, int ReconMethod>
@@ -166,31 +166,31 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
// Initialization
for (int i = 0; i < nproma; ++i) {
- this->p_cc[At<nproma, nlev, nblks_c>(i, 0, 0)] = (i + 1);
+ this->p_cc[at<nproma, nlev, nblks_c>(i, 0, 0)] = (i + 1);
- this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 1)] = i;
- this->cell_neighbor_idx[At<nproma, nblks_c, 3>(i, 0, 2)] = i;
+ this->cell_neighbor_idx[at<nproma, nblks_c, 3>(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[at<nproma, nblks_c, 3>(i, 0, 1)] = i;
+ this->cell_neighbor_idx[at<nproma, nblks_c, 3>(i, 0, 2)] = i;
for (int j = 0; j < 3; ++j) {
- this->cell_neighbor_blk[At<nproma, nblks_c, 3>(i, 0, j)] = 0;
+ this->cell_neighbor_blk[at<nproma, nblks_c, 3>(i, 0, j)] = 0;
}
for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_qtmat_c[At<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 0, j,
+ this->lsq_qtmat_c[at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 0, j,
0)] = 1.0;
- this->lsq_qtmat_c[At<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 1, j,
+ this->lsq_qtmat_c[at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 1, j,
0)] = 0.5;
- this->p_coeff[At<lsq_dim_c, nproma, nlev, nblks_c>(j, i, 0, 0)] = 0.0;
+ this->p_coeff[at<lsq_dim_c, nproma, nlev, nblks_c>(j, i, 0, 0)] = 0.0;
}
- this->lsq_rmat_rdiag_c[At<nproma, lsq_dim_unk, nblks_c>(i, 0, 0)] = 2.0;
- this->lsq_rmat_rdiag_c[At<nproma, lsq_dim_unk, nblks_c>(i, 1, 0)] = 2.0;
+ this->lsq_rmat_rdiag_c[at<nproma, lsq_dim_unk, nblks_c>(i, 0, 0)] = 2.0;
+ this->lsq_rmat_rdiag_c[at<nproma, lsq_dim_unk, nblks_c>(i, 1, 0)] = 2.0;
this->lsq_rmat_utri_c
- [At<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>(
+ [at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>(
i, 0, 0)] = 0.1;
- this->lsq_moments[At<nproma, nblks_c, lsq_dim_unk>(i, 0, 0)] = 0.2;
- this->lsq_moments[At<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
+ this->lsq_moments[at<nproma, nblks_c, lsq_dim_unk>(i, 0, 0)] = 0.2;
+ this->lsq_moments[at<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
}
recon_lsq_cell_l<TypeParam>(
@@ -202,9 +202,12 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
- EXPECT_NEAR(this->p_coeff[0], 0.34, 1e-6);
- EXPECT_NEAR(this->p_coeff[1], 1.8, 1e-6);
- EXPECT_NEAR(this->p_coeff[2], 1.0, 1e-6);
+ EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.34, 1e-6);
+ EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.8, 1e-6);
+ EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 1.0, 1e-6);
}
typedef ::testing::Types<float, double> ValueTypes;
--
GitLab
From 4a18f708a16c5dcbfeccb5afa00c9205bda19e01 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Mon, 3 Mar 2025 11:29:28 +0100
Subject: [PATCH 13/35] Add exception for new typo version
---
_typos.toml | 1 +
1 file changed, 1 insertion(+)
diff --git a/_typos.toml b/_typos.toml
index 58a18ef..8de4a86 100644
--- a/_typos.toml
+++ b/_typos.toml
@@ -11,6 +11,7 @@ extend-ignore-words-re = [
Wirth = "Wirth" # author name
nin = "nin" # number of inputs
Pilar = "Pilar" # author name
+Comput = "Comput" # abbreviation for Computational
[default.extend-identifiers]
f4dout = "f4dout" # file name
--
GitLab
From 77fa7b0db81c08976dcba21b5129f3c6448efe39 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Mon, 3 Mar 2025 13:24:52 +0100
Subject: [PATCH 14/35] Define template and instantiate functions
---
src/horizontal/CMakeLists.txt | 1 -
src/horizontal/lib_divrot.cpp | 197 +++--------------------------
src/horizontal/lib_divrot.hpp | 198 ++++++++++++++++--------------
src/types.hpp | 16 +++
test/c/test_horizontal_divrot.cpp | 3 +-
5 files changed, 140 insertions(+), 275 deletions(-)
create mode 100644 src/types.hpp
diff --git a/src/horizontal/CMakeLists.txt b/src/horizontal/CMakeLists.txt
index 198488f..d403cb2 100644
--- a/src/horizontal/CMakeLists.txt
+++ b/src/horizontal/CMakeLists.txt
@@ -59,7 +59,6 @@ target_include_directories(
# multiple compile languages
# https://cmake.org/cmake/help/latest/manual/cmake-generator-expressions.7.html
$<BUILD_INTERFACE:$<$<COMPILE_LANGUAGE:C,CXX>:${PROJECT_SOURCE_DIR}/src>>
- $<BUILD_INTERFACE:$<$<COMPILE_LANGUAGE:C,CXX>:${CMAKE_CURRENT_SOURCE_DIR}>>
PRIVATE
# Path to config.h (for C and C++ only): Requires CMake 3.15+ for multiple
# compile languages
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 9dce2e4..be6d9da 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -10,10 +10,11 @@
// ---------------------------------------------------------------
#include <iostream>
-#include <lib_divrot.hpp>
-#include <support/mo_lib_loopindices.hpp>
#include <vector>
+#include <horizontal/lib_divrot.hpp>
+#include <support/mo_lib_loopindices.hpp>
+
template <typename T>
void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const T *lsq_qtmat_c,
@@ -101,24 +102,7 @@ void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::fence();
}
-template void
-recon_lsq_cell_l<float>(const float *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const float *lsq_qtmat_c,
- const float *lsq_rmat_rdiag_c,
- const float *lsq_rmat_utri_c, const float *lsq_moments,
- float *p_coeff, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev,
- int nproma, bool l_consv, bool lacc, bool acc_async,
- int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c);
-
-template void recon_lsq_cell_l<double>(
- const double *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const double *lsq_qtmat_c,
- const double *lsq_rmat_rdiag_c, const double *lsq_rmat_utri_c,
- const double *lsq_moments, double *p_coeff, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev,
- int lsq_dim_unk, int lsq_dim_c);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_L);
template <typename T>
void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
@@ -195,21 +179,7 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::fence();
}
-template void recon_lsq_cell_l_svd<float>(
- const float *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const float *lsq_pseudoinv,
- const float *lsq_moments, float *p_coeff, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev,
- int lsq_dim_unk, int lsq_dim_c);
-
-template void recon_lsq_cell_l_svd<double>(
- const double *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const double *lsq_pseudoinv,
- const double *lsq_moments, double *p_coeff, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev,
- int lsq_dim_unk, int lsq_dim_c);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_L_SVD);
template <typename T>
void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
@@ -373,23 +343,7 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::fence();
}
-template void recon_lsq_cell_q<float>(
- const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const float *lsq_rmat_rdiag_c, const float *lsq_rmat_utri_c,
- const float *lsq_moments, const float *lsq_qtmat_c, float *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
-
-template void recon_lsq_cell_q<double>(
- const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const double *lsq_rmat_rdiag_c, const double *lsq_rmat_utri_c,
- const double *lsq_moments, const double *lsq_qtmat_c, double *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_Q);
template <typename T>
void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
@@ -529,21 +483,7 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
Kokkos::fence();
}
-template void recon_lsq_cell_q_svd<float>(
- const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const float *lsq_pseudoinv, const float *lsq_moments, float *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
-
-template void recon_lsq_cell_q_svd<double>(
- const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const double *lsq_pseudoinv, const double *lsq_moments, double *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_Q_SVD);
template <typename T>
void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
@@ -785,23 +725,7 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::fence();
}
-template void recon_lsq_cell_c<float>(
- const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const float *lsq_rmat_rdiag_c, const float *lsq_rmat_utri_c,
- const float *lsq_moments, const float *lsq_qtmat_c, float *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
-
-template void recon_lsq_cell_c<double>(
- const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const double *lsq_rmat_rdiag_c, const double *lsq_rmat_utri_c,
- const double *lsq_moments, const double *lsq_qtmat_c, double *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_C);
template <typename T>
void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
@@ -988,21 +912,7 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
Kokkos::fence();
}
-template void recon_lsq_cell_c_svd<float>(
- const float *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const float *lsq_pseudoinv, const float *lsq_moments, float *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
-
-template void recon_lsq_cell_c_svd<double>(
- const double *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const double *lsq_pseudoinv, const double *lsq_moments, double *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, int patch_id, int lsq_high_set_dim_c,
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_C_SVD);
template <typename T>
void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
@@ -1046,19 +956,7 @@ void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
}
}
-template void div3d<float>(const float *vec_e, const int *cell_edge_idx,
- const int *cell_edge_blk, const float *geofac_div,
- float *div_vec_c, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, bool lacc, int nlev,
- int nblks_c, int nblks_e);
-
-template void div3d<double>(const double *vec_e, const int *cell_edge_idx,
- const int *cell_edge_blk, const double *geofac_div,
- double *div_vec_c, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, bool lacc, int nlev,
- int nblks_c, int nblks_e);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_DIV3D);
template <typename T>
void div3d_2field(const T *vec_e, const int *cell_edge_idx,
@@ -1115,19 +1013,7 @@ void div3d_2field(const T *vec_e, const int *cell_edge_idx,
}
}
-template void div3d_2field<float>(
- const float *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
- const float *geofac_div, float *div_vec_c, const float *in2, float *out2,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, int slev,
- int elev, int nproma, bool lacc, int nlev, int nblks_c, int nblks_e);
-
-template void
-div3d_2field<double>(const double *vec_e, const int *cell_edge_idx,
- const int *cell_edge_blk, const double *geofac_div,
- double *div_vec_c, const double *in2, double *out2,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma, bool lacc,
- int nlev, int nblks_c, int nblks_e);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_DIV3D_2FIELD);
template <typename T>
void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
@@ -1180,19 +1066,7 @@ void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
}
}
-template void div4d<float>(const int *cell_edge_idx, const int *cell_edge_blk,
- const float *geofac_div, const float *f4din,
- float *f4dout, int dim4d, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- const int *slev, const int *elev, int nproma,
- bool lacc, int nlev, int nblks_c, int nblks_e);
-
-template void div4d<double>(const int *cell_edge_idx, const int *cell_edge_blk,
- const double *geofac_div, const double *f4din,
- double *f4dout, int dim4d, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- const int *slev, const int *elev, int nproma,
- bool lacc, int nlev, int nblks_c, int nblks_e);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_DIV4D);
template <typename T>
void div_avg(const T *vec_e, const int *cell_neighbor_idx,
@@ -1373,28 +1247,7 @@ void div_avg(const T *vec_e, const int *cell_neighbor_idx,
}
}
-template void div_avg<float>(const float *vec_e, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk,
- const int *cell_edge_idx, const int *cell_edge_blk,
- const float *geofac_div, const float *avg_coeff,
- float *div_vec_c, const float *opt_in2,
- float *opt_out2, const int *i_startblk_in,
- const int *i_endblk_in, const int *i_startidx_in,
- const int *i_endidx_in, int slev, int elev,
- int nproma, int patch_id, bool l_limited_area,
- bool l2fields, bool lacc, int nlev, int nblks_c,
- int nblks_e);
-
-template void
-div_avg<double>(const double *vec_e, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const int *cell_edge_idx,
- const int *cell_edge_blk, const double *geofac_div,
- const double *avg_coeff, double *div_vec_c,
- const double *opt_in2, double *opt_out2,
- const int *i_startblk_in, const int *i_endblk_in,
- const int *i_startidx_in, const int *i_endidx_in, int slev,
- int elev, int nproma, int patch_id, bool l_limited_area,
- bool l2fields, bool lacc, int nlev, int nblks_c, int nblks_e);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_DIV_AVG);
template <typename T>
void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
@@ -1447,17 +1300,7 @@ void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
}
}
-template void rot_vertex_atmos<float>(
- const float *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
- const float *geofac_rot, float *rot_vec, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool lacc, int nlev, int nblks_e, int nblks_v);
-
-template void rot_vertex_atmos<double>(
- const double *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
- const double *geofac_rot, double *rot_vec, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool lacc, int nlev, int nblks_e, int nblks_v);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_ROT_VERTEX_ATMOS);
template <typename T>
void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
@@ -1513,14 +1356,4 @@ void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
Kokkos::fence();
}
-template void rot_vertex_ri<float>(
- const float *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
- const float *geofac_rot, float *rot_vec, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool lacc, bool acc_async, int nlev, int nblks_e, int nblks_v);
-
-template void rot_vertex_ri<double>(
- const double *vec_e, const int *vert_edge_idx, const int *vert_edge_blk,
- const double *geofac_rot, double *rot_vec, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool lacc, bool acc_async, int nlev, int nblks_e, int nblks_v);
+ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_ROT_VERTEX_RI);
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
index db60b29..a0cc8cf 100644
--- a/src/horizontal/lib_divrot.hpp
+++ b/src/horizontal/lib_divrot.hpp
@@ -12,106 +12,122 @@
#pragma once
#include <Kokkos_Core.hpp>
+#include <types.hpp>
-template <typename T>
-void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const T *lsq_qtmat_c,
- const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, T *p_coeff, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- int slev, int elev, int nproma, bool l_consv, bool lacc,
- bool acc_async, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+#define ICONMATH_DECLARE_RECON_LSQ_CELL_L(_type) \
+ void recon_lsq_cell_l( \
+ const _type *p_cc, const int *cell_neighbor_idx, \
+ const int *cell_neighbor_blk, const _type *lsq_qtmat_c, \
+ const _type *lsq_rmat_rdiag_c, const _type *lsq_rmat_utri_c, \
+ const _type *lsq_moments, _type *p_coeff, int i_startblk, int i_endblk, \
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma, \
+ bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev, \
+ int lsq_dim_unk, int lsq_dim_c)
-template <typename T>
-void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const T *lsq_pseudoinv,
- const T *lsq_moments, T &p_coeff, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- int slev, int elev, int nproma, bool l_consv,
- bool lacc, bool acc_async, int nblks_c, int nlev,
- int lsq_dim_unk, int lsq_dim_c);
+#define ICONMATH_DECLARE_RECON_LSQ_CELL_L_SVD(_type) \
+ void recon_lsq_cell_l_svd( \
+ const _type *p_cc, const int *cell_neighbor_idx, \
+ const int *cell_neighbor_blk, const _type *lsq_pseudoinv, \
+ const _type *lsq_moments, _type *p_coeff, int i_startblk, int i_endblk, \
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma, \
+ bool l_consv, bool lacc, bool acc_async, int nblks_c, int nlev, \
+ int lsq_dim_unk, int lsq_dim_c)
-template <typename T>
-void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma,
- int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+#define ICONMATH_DECLARE_RECON_LSQ_CELL_Q(_type) \
+ void recon_lsq_cell_q( \
+ const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
+ const _type *lsq_rmat_rdiag_c, const _type *lsq_rmat_utri_c, \
+ const _type *lsq_moments, const _type *lsq_qtmat_c, _type *p_coeff, \
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
+ int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
+ int lsq_dim_c)
-template <typename T>
-void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
- const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T &p_coeff, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- int slev, int elev, int nproma, int patch_id,
- int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+#define ICONMATH_DECLARE_RECON_LSQ_CELL_Q_SVD(_type) \
+ void recon_lsq_cell_q_svd( \
+ const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
+ const _type *lsq_pseudoinv, const _type *lsq_moments, _type *p_coeff, \
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
+ int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
+ int lsq_dim_c)
-template <typename T>
-void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T &p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma,
- int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+#define ICONMATH_DECLARE_RECON_LSQ_CELL_C(_type) \
+ void recon_lsq_cell_c( \
+ const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
+ const _type *lsq_rmat_rdiag_c, const _type *lsq_rmat_utri_c, \
+ const _type *lsq_moments, const _type *lsq_qtmat_c, _type *p_coeff, \
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
+ int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
+ int lsq_dim_c)
-template <typename T>
-void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
- const int *lsq_blk_c, const T *lsq_pseudoinv,
- const T *lsq_moments, T &p_coeff, int i_startblk,
- int i_endblk, int i_startidx_in, int i_endidx_in,
- int slev, int elev, int nproma, int patch_id,
- int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c);
+#define ICONMATH_DECLARE_RECON_LSQ_CELL_C_SVD(_type) \
+ void recon_lsq_cell_c_svd( \
+ const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
+ const _type *lsq_pseudoinv, const _type *lsq_moments, _type *p_coeff, \
+ int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
+ int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
+ bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
+ int lsq_dim_c)
-template <typename T>
-void div3d(const T *vec_e, const int *cell_edge_idx, const int *cell_edge_blk,
- const T *geofac_div, T &div_vec_c, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma,
- bool lacc, int nlev, int nblks_c, int nblks_e);
+#define ICONMATH_DECLARE_DIV3D(_type) \
+ void div3d(const _type *vec_e, const int *cell_edge_idx, \
+ const int *cell_edge_blk, const _type *geofac_div, \
+ _type *div_vec_c, int i_startblk, int i_endblk, \
+ int i_startidx_in, int i_endidx_in, int slev, int elev, \
+ int nproma, bool lacc, int nlev, int nblks_c, int nblks_e)
-template <typename T>
-void div3d_2field(const T *vec_e, const int *cell_edge_idx,
- const int *cell_edge_blk, const T *geofac_div, T &div_vec_c,
- const T *in2, T &out2, int i_startblk, int i_endblk,
- int i_startidx_in, int i_endidx_in, int slev, int elev,
- int nproma, bool lacc, int nlev, int nblks_c, int nblks_e);
+#define ICONMATH_DECLARE_DIV3D_2FIELD(_type) \
+ void div3d_2field(const _type *vec_e, const int *cell_edge_idx, \
+ const int *cell_edge_blk, const _type *geofac_div, \
+ _type *div_vec_c, const _type *in2, _type *out2, \
+ int i_startblk, int i_endblk, int i_startidx_in, \
+ int i_endidx_in, int slev, int elev, int nproma, \
+ bool lacc, int nlev, int nblks_c, int nblks_e)
-template <typename T>
-void div4d(const int *cell_edge_idx, const int *cell_edge_blk,
- const T *geofac_div, const T *f4din, T &f4dout, int dim4d,
- int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in,
- const int *slev, const int *elev, int nproma, bool lacc, int nlev,
- int nblks_c, int nblks_e);
+#define ICONMATH_DECLARE_DIV4D(_type) \
+ void div4d(const int *cell_edge_idx, const int *cell_edge_blk, \
+ const _type *geofac_div, const _type *f4din, _type *f4dout, \
+ int dim4d, int i_startblk, int i_endblk, int i_startidx_in, \
+ int i_endidx_in, const int *slev, const int *elev, int nproma, \
+ bool lacc, int nlev, int nblks_c, int nblks_e)
-template <typename T>
-void div_avg(const T *vec_e, const int *cell_neighbor_idx,
- const int *cell_neighbor_blk, const int *cell_edge_idx,
- const int *cell_edge_blk, const T *geofac_div, const T *avg_coeff,
- T &div_vec_c, const T *opt_in2, T &opt_out2,
- const int *i_startblk_in, const int *i_endblk_in,
- const int *i_startidx_in, const int *i_endidx_in, int slev,
- int elev, int nproma, int patch_id, bool l_limited_area,
- bool l2fields, bool lacc, int nlev, int nblks_c, int nblks_e);
+#define ICONMATH_DECLARE_DIV_AVG(_type) \
+ void div_avg(const _type *vec_e, const int *cell_neighbor_idx, \
+ const int *cell_neighbor_blk, const int *cell_edge_idx, \
+ const int *cell_edge_blk, const _type *geofac_div, \
+ const _type *avg_coeff, _type *div_vec_c, const _type *opt_in2, \
+ _type *opt_out2, const int *i_startblk_in, \
+ const int *i_endblk_in, const int *i_startidx_in, \
+ const int *i_endidx_in, int slev, int elev, int nproma, \
+ int patch_id, bool l_limited_area, bool l2fields, bool lacc, \
+ int nlev, int nblks_c, int nblks_e)
-template <typename T>
-void rot_vertex_atmos(const T *vec_e, const int *vert_edge_idx,
- const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma,
- bool lacc, int nlev, int nblks_e, int nblks_v);
+#define ICONMATH_DECLARE_ROT_VERTEX_ATMOS(_type) \
+ void rot_vertex_atmos( \
+ const _type *vec_e, const int *vert_edge_idx, const int *vert_edge_blk, \
+ const _type *geofac_rot, _type *rot_vec, int i_startblk, int i_endblk, \
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma, \
+ bool lacc, int nlev, int nblks_e, int nblks_v)
-template <typename T>
-void rot_vertex_ri(const T *vec_e, const int *vert_edge_idx,
- const int *vert_edge_blk, const T *geofac_rot, T &rot_vec,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma, bool lacc,
- bool acc_async, int nlev, int nblks_e, int nblks_v);
+#define ICONMATH_DECLARE_ROT_VERTEX_RI(_type) \
+ void rot_vertex_ri( \
+ const _type *vec_e, const int *vert_edge_idx, const int *vert_edge_blk, \
+ const _type *geofac_rot, _type *rot_vec, int i_startblk, int i_endblk, \
+ int i_startidx_in, int i_endidx_in, int slev, int elev, int nproma, \
+ bool lacc, bool acc_async, int nlev, int nblks_e, int nblks_v)
+
+// Declare as templates
+template <typename T> ICONMATH_DECLARE_RECON_LSQ_CELL_L(T);
+template <typename T> ICONMATH_DECLARE_RECON_LSQ_CELL_L_SVD(T);
+template <typename T> ICONMATH_DECLARE_RECON_LSQ_CELL_Q(T);
+template <typename T> ICONMATH_DECLARE_RECON_LSQ_CELL_Q_SVD(T);
+template <typename T> ICONMATH_DECLARE_RECON_LSQ_CELL_C(T);
+template <typename T> ICONMATH_DECLARE_RECON_LSQ_CELL_C_SVD(T);
+template <typename T> ICONMATH_DECLARE_DIV3D(T);
+template <typename T> ICONMATH_DECLARE_DIV3D_2FIELD(T);
+template <typename T> ICONMATH_DECLARE_DIV4D(T);
+template <typename T> ICONMATH_DECLARE_DIV_AVG(T);
+template <typename T> ICONMATH_DECLARE_ROT_VERTEX_ATMOS(T);
+template <typename T> ICONMATH_DECLARE_ROT_VERTEX_RI(T);
diff --git a/src/types.hpp b/src/types.hpp
new file mode 100644
index 0000000..7192e18
--- /dev/null
+++ b/src/types.hpp
@@ -0,0 +1,16 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#pragma once
+
+#define ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(_macro) \
+ template _macro(float); \
+ template _macro(double)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index a99556f..29693ff 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -9,10 +9,11 @@
// SPDX-License-Identifier: BSD-3-Clause
// ---------------------------------------------------------------
+#include <vector>
+
#include <Kokkos_Core.hpp>
#include <gtest/gtest.h>
#include <horizontal/lib_divrot.hpp>
-#include <vector>
// Template helpers for combining multiple dimension array sizes.
// The base function of dimension combine. Should not be used.
--
GitLab
From bc09632d8d907efa7937e23c1ce22fa3eea0c7be Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Sat, 8 Mar 2025 20:53:35 +0100
Subject: [PATCH 15/35] Add comments and adapt Doxygen style
---
test/c/test_horizontal_divrot.cpp | 75 ++++++++++++++++++++++---------
1 file changed, 54 insertions(+), 21 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 29693ff..3a46b9e 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -15,27 +15,43 @@
#include <gtest/gtest.h>
#include <horizontal/lib_divrot.hpp>
-// Template helpers for combining multiple dimension array sizes.
-// The base function of dimension combine. Should not be used.
+// Template function for computing array size.
+// For example, we get the array size of a 4-dimensional array A(2, 3, 4, 5) by
+// dim_combine(2, 3, 4, 5).
+// Which will automatically instantiate
+// dim_combine<int, int, int, int>(2, 3, 4, 5).
+// The function then call dim_combine recursively
+// dim_combine<int, int, int, int>(2, 3, 4, 5) {
+// return static_cast<size_t>(2) * dim_combine<int, int, int>(3, 4, 5);
+// }
+// dim_combine<int, int, int>(3, 4, 5) {
+// return static_cast<size_t>(3) * dim_combine<int, int>(4, 5);
+// }
+// dim_combine<int, int>(4, 5) {
+// return static_cast<size_t>(4) * dim_combine<int>(5);
+// }
+// Where the last dim_combine is specialized as
+// dim_combine<int>(5) {
+// return static_cast<size_t>(5);
+// }
+// Which gives
+// dim_combine<int, int, int, int>(2, 3, 4, 5) =
+// static_cast<size_t>(2) * static_cast<size_t>(3) *
+// static_cast<size_t>(4) * static_cast<size_t>(5)
+/// Template helpers for combining multiple dimension array sizes.
+/// The base function of dimension combine. Should not be used.
template <typename... Ts> size_t dim_combine(Ts... dims) { return 0; }
-// Template specialization of only one dimension, returns the dimension itself.
+/// Template specialization of only one dimension, returns the dimension itself.
template <typename T> size_t dim_combine(T dim) {
return static_cast<size_t>(dim);
}
-// Template specialization of picking out the first dimension. The combined
-// dimension is the first dimension times the combined dimension of the rest.
+/// Template specialization of picking out the first dimension. The combined
+/// dimension is the first dimension times the combined dimension of the rest.
template <typename T, typename... Ts> size_t dim_combine(T dim, Ts... dims) {
return static_cast<size_t>(dim) * dim_combine(dims...);
}
-// Enum class for the reconstruction method
-enum class ReconstructionMethod {
- linear,
- quadratic,
- cubic,
-};
-
-// Template function for LayoutLeft ID access in compile time
+// Template function for LayoutLeft ID access in compile time.
// For example, a multi-dimensional array A of dimensions <2, 3, 4, 5> gets its
// corresponding vector id (LayoutLeft) by
// at<2, 3, 4, 5>(id1, id2, id3, id4).
@@ -56,29 +72,38 @@ enum class ReconstructionMethod {
// Which gives
// at<2, 3, 4, 5>(id1, id2, id3, id4) = id1 + id2 * 2 +
// id3 * 2 * 3 + id4 * 2 * 3 * 4
-// Helper type converting integer numbers to int
+/// Helper type converting integer numbers to int
template <class T, auto> using always_t = T;
-// Base function of at_impl. Should not be used.
+/// Base function of at_impl. Should not be used.
template <int... Dims> int at_impl(always_t<int, Dims>... ids) { return 0; }
-// Template specialization of the last ID
+/// Template specialization of the last ID
template <int LastDim> int at_impl(int prefix, int id) { return id * prefix; }
-// Template specialization of at_impl, accumulate the return value using the
-// first id and pass the prefix to the next recursive at_impl function.
+/// Template specialization of at_impl, accumulate the return value using the
+/// first id and pass the prefix to the next recursive at_impl function.
template <int FirstDim, int... Dims>
int at_impl(int prefix, int id, always_t<int, Dims>... ids) {
return id * prefix + at_impl<Dims...>(prefix * FirstDim, ids...);
}
-// at<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
-// LayoutLeft. Use this function instead of at_impl.
+/// at<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
+/// LayoutLeft. Use this function instead of at_impl.
template <int FirstDim, int... Dims>
int at(int id, always_t<int, Dims>... ids) {
return id + at_impl<Dims...>(FirstDim, ids...);
}
+/// Enum class for the reconstruction method
+enum class ReconstructionMethod {
+ linear,
+ quadratic,
+ cubic,
+};
+
+/// Base test class for the horizontal divrot tests. Templated for the ValueType
+/// and ReconMethod for the reconstruction method.
template <typename ValueType, int ReconMethod>
class HorizontalDivrotTest : public ::testing::Test {
protected:
- // [lsq_dim_c, lsq_dim_unk]
+ // lsq_dim_c and lsq_dim_unk are instantiated in compile time.
static constexpr std::tuple<int, int>
init_lsq_dim(ReconstructionMethod method) {
switch (method) {
@@ -133,16 +158,22 @@ protected:
}
};
+/// Test class for the horizontal tests. The reconstruction method is specified
+/// to linear.
template <typename ValueType>
class HorizontalDivrotLinearTest
: public HorizontalDivrotTest<
ValueType, static_cast<int>(ReconstructionMethod::linear)> {};
+/// Test class for the horizontal tests. The reconstruction method is specified
+/// to quadratic.
template <typename ValueType>
class HorizontalDivrotQuadraticTest
: public HorizontalDivrotTest<
ValueType, static_cast<int>(ReconstructionMethod::quadratic)> {};
+/// Test class for the horizontal tests. The reconstruction method is specified
+/// to cubic.
template <typename ValueType>
class HorizontalDivrotCubicTest
: public HorizontalDivrotTest<ValueType, static_cast<int>(
@@ -194,6 +225,7 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
this->lsq_moments[at<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
}
+ // Test function
recon_lsq_cell_l<TypeParam>(
this->p_cc.data(), this->cell_neighbor_idx.data(),
this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
@@ -203,6 +235,7 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+ // Check result
EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(0, 0, 0, 0))],
0.34, 1e-6);
EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(1, 0, 0, 0))],
--
GitLab
From c911a6bb5ab813d551e3c0da908a50a9feb8ea75 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Sat, 8 Mar 2025 22:37:00 +0100
Subject: [PATCH 16/35] Add first random test
---
test/c/test_horizontal_divrot.cpp | 158 ++++++++++++++++++++++++++----
1 file changed, 138 insertions(+), 20 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 3a46b9e..a57f31a 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -9,11 +9,14 @@
// SPDX-License-Identifier: BSD-3-Clause
// ---------------------------------------------------------------
+#include <iostream>
+#include <random>
#include <vector>
#include <Kokkos_Core.hpp>
#include <gtest/gtest.h>
#include <horizontal/lib_divrot.hpp>
+#include <support/mo_lib_loopindices.hpp>
// Template function for computing array size.
// For example, we get the array size of a 4-dimensional array A(2, 3, 4, 5) by
@@ -180,6 +183,7 @@ class HorizontalDivrotCubicTest
ReconstructionMethod::cubic)> {
};
+/// ValueTypes which the divrot tests should run with
typedef ::testing::Types<float, double> ValueTypes;
TYPED_TEST_SUITE(HorizontalDivrotLinearTest, ValueTypes);
@@ -196,33 +200,38 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_c, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
// Initialization
for (int i = 0; i < nproma; ++i) {
- this->p_cc[at<nproma, nlev, nblks_c>(i, 0, 0)] = (i + 1);
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
- this->cell_neighbor_idx[at<nproma, nblks_c, 3>(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_idx[at<nproma, nblks_c, 3>(i, 0, 1)] = i;
- this->cell_neighbor_idx[at<nproma, nblks_c, 3>(i, 0, 2)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
for (int j = 0; j < 3; ++j) {
- this->cell_neighbor_blk[at<nproma, nblks_c, 3>(i, 0, j)] = 0;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
}
for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_qtmat_c[at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 0, j,
- 0)] = 1.0;
- this->lsq_qtmat_c[at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>(i, 1, j,
- 0)] = 0.5;
- this->p_coeff[at<lsq_dim_c, nproma, nlev, nblks_c>(j, i, 0, 0)] = 0.0;
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
}
- this->lsq_rmat_rdiag_c[at<nproma, lsq_dim_unk, nblks_c>(i, 0, 0)] = 2.0;
- this->lsq_rmat_rdiag_c[at<nproma, lsq_dim_unk, nblks_c>(i, 1, 0)] = 2.0;
- this->lsq_rmat_utri_c
- [at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>(
- i, 0, 0)] = 0.1;
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
+ this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
- this->lsq_moments[at<nproma, nblks_c, lsq_dim_unk>(i, 0, 0)] = 0.2;
- this->lsq_moments[at<nproma, nblks_c, lsq_dim_unk>(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
}
// Test function
@@ -244,7 +253,118 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
1.0, 1e-6);
}
-typedef ::testing::Types<float, double> ValueTypes;
+TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
+ constexpr int nproma = TestFixture::nproma;
+ constexpr int nlev = TestFixture::nlev;
+ constexpr int nblks_c = TestFixture::nblks_c;
+ constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
+ constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_c, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < 3; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = real_distrib(gen);
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = real_distrib(gen);
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = real_distrib(gen);
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = real_distrib(gen);
+ this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = real_distrib(gen);
+
+ this->lsq_moments[moments_at(i, 0, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, 1)] = real_distrib(gen);
+ }
+
+ // Test function
+ recon_lsq_cell_l<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(3);
+ std::vector<TypeParam> z_qt_times_d(2);
+ std::vector<TypeParam> p_result(lsq_dim_c * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ z_d[0] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ z_d[1] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ z_d[2] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ z_qt_times_d[0] = this->lsq_qtmat_c[qtmat_at(jc, 0, 0, jb)] * z_d[0] +
+ this->lsq_qtmat_c[qtmat_at(jc, 0, 1, jb)] * z_d[1] +
+ this->lsq_qtmat_c[qtmat_at(jc, 0, 2, jb)] * z_d[2];
+ z_qt_times_d[1] = this->lsq_qtmat_c[qtmat_at(jc, 1, 0, jb)] * z_d[0] +
+ this->lsq_qtmat_c[qtmat_at(jc, 1, 1, jb)] * z_d[1] +
+ this->lsq_qtmat_c[qtmat_at(jc, 1, 2, jb)] * z_d[2];
+ p_result[at<lsq_dim_c, nproma>(2, jc)] =
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 1, jb)] * z_qt_times_d[1];
+ p_result[at<lsq_dim_c, nproma>(1, jc)] =
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 0, jb)] *
+ (z_qt_times_d[0] - this->lsq_rmat_utri_c[rmat_utri_at(jc, 0, jb)] *
+ p_result[at<lsq_dim_c, nproma>(2, jc)]);
+ p_result[at<lsq_dim_c, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ }
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ p_result[at<lsq_dim_c, nproma>(0, jc)] =
+ p_result[at<lsq_dim_c, nproma>(0, jc)] -
+ p_result[at<lsq_dim_c, nproma>(1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 0)] -
+ p_result[at<lsq_dim_c, nproma>(2, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 1)];
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_c, nproma>(i, jc))], 1e-6);
+ }
+ }
+}
TYPED_TEST_SUITE(HorizontalDivrotQuadraticTest, ValueTypes);
@@ -253,8 +373,6 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestLsqDimensions) {
EXPECT_EQ(TestFixture::lsq_dim_unk, 5);
}
-typedef ::testing::Types<float, double> ValueTypes;
-
TYPED_TEST_SUITE(HorizontalDivrotCubicTest, ValueTypes);
TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
--
GitLab
From c7a777cc52e367f23f1786712a90c3dea421c2f7 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Sun, 9 Mar 2025 11:06:02 +0100
Subject: [PATCH 17/35] Fix bug and add tests
---
src/horizontal/lib_divrot.cpp | 88 ++++-----
src/horizontal/lib_divrot.hpp | 23 +--
test/c/test_horizontal_divrot.cpp | 314 ++++++++++++++++++++++++------
3 files changed, 306 insertions(+), 119 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index be6d9da..164be38 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -36,14 +36,14 @@ void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- Kokkos::View<T *> z_d("z_d", 3);
- Kokkos::View<T *> z_qt_times_d("z_qt_times_d", 2);
+ Kokkos::View<T *> z_d("z_d", lsq_dim_c);
+ Kokkos::View<T *> z_qt_times_d("z_qt_times_d", lsq_dim_unk);
- UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
- UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, lsq_dim_c);
+ UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
- UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_unk + 1, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -124,13 +124,13 @@ void recon_lsq_cell_l_svd(const T *p_cc, const int *cell_neighbor_idx,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- Kokkos::View<T *> z_b("z_b", 3);
+ Kokkos::View<T *> z_b("z_b", lsq_dim_c);
- UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, 3);
- UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, 3);
+ UnmanagedConstInt3D iidx(cell_neighbor_idx, nproma, nblks_c, lsq_dim_c);
+ UnmanagedConstInt3D iblk(cell_neighbor_blk, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
- UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_unk + 1, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -183,13 +183,12 @@ ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_L_SVD);
template <typename T>
void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma,
- int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c) {
+ const T *lsq_qtmat_c, const T *lsq_rmat_rdiag_c,
+ const T *lsq_rmat_utri_c, const T *lsq_moments,
+ T *p_coeff, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev,
+ int nproma, int patch_id, bool l_limited_area, bool lacc,
+ int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
@@ -202,14 +201,14 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- Kokkos::View<T ***> z_d("z_d", lsq_high_set_dim_c, nproma, elev);
- Kokkos::View<T *> z_qt_times_d("z_qt_times_d", 5);
+ Kokkos::View<T ***> z_d("z_d", lsq_dim_c, nproma, nlev);
+ Kokkos::View<T *> z_qt_times_d("z_qt_times_d", lsq_dim_unk);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
- UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_unk + 1, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -219,9 +218,10 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
- if (patch_id > 1 || l_limited_area) {
+ if (patch_id > 0 || l_limited_area) {
Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
- {0, i_startidx_in, slev, i_startblk}, {6, i_endidx_in, elev, i_endblk});
+ {0, i_startidx_in, slev, i_startblk},
+ {lsq_dim_unk + 1, i_endidx_in, elev, i_endblk});
Kokkos::parallel_for(
"recon_lsq_cell_q_init", initPolicy,
KOKKOS_LAMBDA(const int ji, const int jc, const int jk, const int jb) {
@@ -351,9 +351,8 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
const T *lsq_moments, T *p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
- int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c) {
+ bool l_limited_area, bool lacc, int nblks_c, int nlev,
+ int lsq_dim_unk, int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
@@ -366,21 +365,22 @@ void recon_lsq_cell_q_svd(const T *p_cc, const int *lsq_idx_c,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- Kokkos::View<T ***> z_b("z_b", lsq_high_set_dim_c, nproma, elev);
+ Kokkos::View<T ***> z_b("z_b", lsq_dim_c, nproma, elev);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
- UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_unk + 1, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
- if (patch_id > 1 || l_limited_area) {
+ if (patch_id > 0 || l_limited_area) {
Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
- {0, i_startidx_in, slev, i_startblk}, {6, i_endidx_in, elev, i_endblk});
+ {0, i_startidx_in, slev, i_startblk},
+ {lsq_dim_unk + 1, i_endidx_in, elev, i_endblk});
Kokkos::parallel_for(
"recon_lsq_cell_q_svd_init", initPolicy,
KOKKOS_LAMBDA(const int ji, const int jc, const int jk, const int jb) {
@@ -487,13 +487,12 @@ ICONMATH_INSTANTIATE_FOR_EACH_VALUE_TYPE(ICONMATH_DECLARE_RECON_LSQ_CELL_Q_SVD);
template <typename T>
void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
- const T *lsq_rmat_rdiag_c, const T *lsq_rmat_utri_c,
- const T *lsq_moments, const T *lsq_qtmat_c, T *p_coeff,
- int i_startblk, int i_endblk, int i_startidx_in,
- int i_endidx_in, int slev, int elev, int nproma,
- int patch_id, int lsq_high_set_dim_c, bool l_limited_area,
- bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
- int lsq_dim_c) {
+ const T *lsq_qtmat_c, const T *lsq_rmat_rdiag_c,
+ const T *lsq_rmat_utri_c, const T *lsq_moments,
+ T *p_coeff, int i_startblk, int i_endblk,
+ int i_startidx_in, int i_endidx_in, int slev, int elev,
+ int nproma, int patch_id, bool l_limited_area, bool lacc,
+ int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
typedef Kokkos::View<const T ***, Kokkos::LayoutLeft, Kokkos::MemoryUnmanaged>
UnmanagedConstT3D;
@@ -506,14 +505,14 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::MemoryUnmanaged>
UnmanagedConstInt3D;
- Kokkos::View<T ***> z_d("z_d", lsq_high_set_dim_c, nproma, elev);
+ Kokkos::View<T ***> z_d("z_d", lsq_dim_c, nproma, elev);
Kokkos::View<T *> z_qt_times_d("z_qt_times_d", 9);
UnmanagedConstInt3D iidx(lsq_idx_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
- UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_unk + 1, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_qtmat_c_view(lsq_qtmat_c, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
@@ -523,9 +522,10 @@ void recon_lsq_cell_c(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
- if (patch_id > 1 || l_limited_area) {
+ if (patch_id > 0 || l_limited_area) {
Kokkos::MDRangePolicy<Kokkos::Rank<4>> initPolicy(
- {0, i_startidx_in, slev, i_startblk}, {9, i_endidx_in, elev, i_endblk});
+ {0, i_startidx_in, slev, i_startblk},
+ {lsq_dim_unk + 1, i_endidx_in, elev, i_endblk});
Kokkos::parallel_for(
"recon_lsq_cell_c_init", initPolicy,
KOKKOS_LAMBDA(const int ji, const int jc, const int jk, const int jb) {
@@ -754,20 +754,20 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
UnmanagedConstInt3D iblk(lsq_blk_c, nproma, nblks_c, lsq_dim_c);
UnmanagedConstT3D p_cc_view(p_cc, nproma, nlev, nblks_c);
- UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_c, nproma, nlev, nblks_c);
+ UnmanagedT4D p_coeff_view(p_coeff, lsq_dim_unk + 1, nproma, nlev, nblks_c);
UnmanagedConstT4D lsq_pseudoinv_view(lsq_pseudoinv, nproma, lsq_dim_unk,
lsq_dim_c, nblks_c);
UnmanagedConstT3D lsq_moments_view(lsq_moments, nproma, nblks_c, lsq_dim_unk);
- if (patch_id > 1 || l_limited_area) {
+ if (patch_id > 0 || l_limited_area) {
for (int jb = i_startblk; jb < i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(i_startidx_in, i_endidx_in, nproma, jb, i_startblk,
i_endblk, i_startidx, i_endidx);
- Kokkos::MDRangePolicy<Kokkos::Rank<3>> initPolicy({slev, i_startidx, 0},
- {elev, i_endidx, 9});
+ Kokkos::MDRangePolicy<Kokkos::Rank<3>> initPolicy(
+ {slev, i_startidx, 0}, {elev, i_endidx, lsq_dim_unk + 1});
Kokkos::parallel_for(
"recon_lsq_cell_c_svd_init", initPolicy,
KOKKOS_LAMBDA(const int jk, const int jc, const int ji) {
@@ -1156,7 +1156,7 @@ void div_avg(const T *vec_e, const int *cell_neighbor_idx,
}
}
- if (patch_id > 1 || l_limited_area) {
+ if (patch_id > 0 || l_limited_area) {
i_startblk = i_startblk_in[1];
i_endblk = i_endblk_in[1];
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
index a0cc8cf..dae8282 100644
--- a/src/horizontal/lib_divrot.hpp
+++ b/src/horizontal/lib_divrot.hpp
@@ -36,31 +36,28 @@
#define ICONMATH_DECLARE_RECON_LSQ_CELL_Q(_type) \
void recon_lsq_cell_q( \
const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
- const _type *lsq_rmat_rdiag_c, const _type *lsq_rmat_utri_c, \
- const _type *lsq_moments, const _type *lsq_qtmat_c, _type *p_coeff, \
+ const _type *lsq_qtmat_c, const _type *lsq_rmat_rdiag_c, \
+ const _type *lsq_rmat_utri_c, const _type *lsq_moments, _type *p_coeff, \
int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
- int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
- int lsq_dim_c)
+ int slev, int elev, int nproma, int patch_id, bool l_limited_area, \
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c)
#define ICONMATH_DECLARE_RECON_LSQ_CELL_Q_SVD(_type) \
void recon_lsq_cell_q_svd( \
const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
const _type *lsq_pseudoinv, const _type *lsq_moments, _type *p_coeff, \
int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
- int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
- int lsq_dim_c)
+ int slev, int elev, int nproma, int patch_id, bool l_limited_area, \
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c)
#define ICONMATH_DECLARE_RECON_LSQ_CELL_C(_type) \
void recon_lsq_cell_c( \
const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
- const _type *lsq_rmat_rdiag_c, const _type *lsq_rmat_utri_c, \
- const _type *lsq_moments, const _type *lsq_qtmat_c, _type *p_coeff, \
+ const _type *lsq_qtmat_c, const _type *lsq_rmat_rdiag_c, \
+ const _type *lsq_rmat_utri_c, const _type *lsq_moments, _type *p_coeff, \
int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
- int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
- bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
- int lsq_dim_c)
+ int slev, int elev, int nproma, int patch_id, bool l_limited_area, \
+ bool lacc, int nblks_c, int nlev, int lsq_dim_unk, int lsq_dim_c)
#define ICONMATH_DECLARE_RECON_LSQ_CELL_C_SVD(_type) \
void recon_lsq_cell_c_svd( \
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index a57f31a..37110db 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -134,10 +134,12 @@ protected:
int i_startidx_in = 0;
int i_endidx_in = nproma; // Full range: 0 .. nproma-1
int slev = 0;
- int elev = nlev; // Full vertical range (0 .. nlev-1)
- bool lacc = false; // Not using ACC-specific behavior.
- bool acc_async = false; // No asynchronous execution.
- bool l_consv = true; // No conservative correction
+ int elev = nlev; // Full vertical range (0 .. nlev-1)
+ int patch_id = 0;
+ bool lacc = false; // Not using ACC-specific behavior.
+ bool acc_async = false; // No asynchronous execution.
+ bool l_consv = true; // With conservative correction.
+ bool l_limited_area = true; // Limited area setup
std::vector<ValueType> p_cc;
std::vector<int> cell_neighbor_idx;
@@ -150,14 +152,14 @@ protected:
HorizontalDivrotTest() {
p_cc.resize(dim_combine(nproma, nlev, nblks_c));
- cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, 3));
- cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, 3));
+ cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
+ cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
lsq_qtmat_c.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
lsq_rmat_rdiag_c.resize(dim_combine(nproma, lsq_dim_unk, nblks_c));
lsq_rmat_utri_c.resize(dim_combine(
nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c));
lsq_moments.resize(dim_combine(nproma, nblks_c, lsq_dim_unk));
- p_coeff.resize(dim_combine(lsq_dim_c, nproma, nlev, nblks_c));
+ p_coeff.resize(dim_combine(lsq_dim_unk + 1, nproma, nlev, nblks_c));
}
};
@@ -201,9 +203,9 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_c, nproma, nlev, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
const auto &rmat_utri_at =
at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
@@ -216,13 +218,12 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
- for (int j = 0; j < 3; ++j) {
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
}
@@ -245,12 +246,15 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
// Check result
- EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- 0.34, 1e-6);
- EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.8, 1e-6);
- EXPECT_NEAR(this->p_coeff[(at<lsq_dim_c, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.34, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 1.0, 1e-6);
}
TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
@@ -261,9 +265,9 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_c, nproma, nlev, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
const auto &rmat_utri_at =
at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
@@ -278,14 +282,13 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
for (int i = 0; i < nproma; ++i) {
this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
- for (int j = 0; j < 3; ++j) {
+ for (int j = 0; j < lsq_dim_c; ++j) {
this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = real_distrib(gen);
this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = real_distrib(gen);
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
}
@@ -308,60 +311,50 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
// Compute reference result
- std::vector<TypeParam> z_d(3);
- std::vector<TypeParam> z_qt_times_d(2);
- std::vector<TypeParam> p_result(lsq_dim_c * nproma);
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
this->i_startblk, this->i_endblk, i_startidx, i_endidx);
for (int jk = this->slev; jk < this->elev; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
- z_d[0] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- z_d[1] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- z_d[2] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- z_qt_times_d[0] = this->lsq_qtmat_c[qtmat_at(jc, 0, 0, jb)] * z_d[0] +
- this->lsq_qtmat_c[qtmat_at(jc, 0, 1, jb)] * z_d[1] +
- this->lsq_qtmat_c[qtmat_at(jc, 0, 2, jb)] * z_d[2];
- z_qt_times_d[1] = this->lsq_qtmat_c[qtmat_at(jc, 1, 0, jb)] * z_d[0] +
- this->lsq_qtmat_c[qtmat_at(jc, 1, 1, jb)] * z_d[1] +
- this->lsq_qtmat_c[qtmat_at(jc, 1, 2, jb)] * z_d[2];
- p_result[at<lsq_dim_c, nproma>(2, jc)] =
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ z_qt_times_d[0] = 0.0;
+ z_qt_times_d[1] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_qt_times_d[0] += this->lsq_qtmat_c[qtmat_at(jc, 0, i, jb)] * z_d[i];
+ z_qt_times_d[1] += this->lsq_qtmat_c[qtmat_at(jc, 1, i, jb)] * z_d[i];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] =
this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 1, jb)] * z_qt_times_d[1];
- p_result[at<lsq_dim_c, nproma>(1, jc)] =
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] =
this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 0, jb)] *
- (z_qt_times_d[0] - this->lsq_rmat_utri_c[rmat_utri_at(jc, 0, jb)] *
- p_result[at<lsq_dim_c, nproma>(2, jc)]);
- p_result[at<lsq_dim_c, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- }
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- p_result[at<lsq_dim_c, nproma>(0, jc)] =
- p_result[at<lsq_dim_c, nproma>(0, jc)] -
- p_result[at<lsq_dim_c, nproma>(1, jc)] *
+ (z_qt_times_d[0] -
+ this->lsq_rmat_utri_c[rmat_utri_at(jc, 0, jb)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)]);
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
this->lsq_moments[moments_at(jc, jb, 0)] -
- p_result[at<lsq_dim_c, nproma>(2, jc)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
this->lsq_moments[moments_at(jc, jb, 1)];
}
}
}
// Check result
- for (int i = 0; i < lsq_dim_c; ++i) {
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
for (int jc = 0; jc < nproma; ++jc) {
EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
- p_result[(at<lsq_dim_c, nproma>(i, jc))], 1e-6);
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
}
}
}
@@ -373,6 +366,203 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestLsqDimensions) {
EXPECT_EQ(TestFixture::lsq_dim_unk, 5);
}
+TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
+ constexpr int nproma = TestFixture::nproma;
+ constexpr int nlev = TestFixture::nlev;
+ constexpr int nblks_c = TestFixture::nblks_c;
+ constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
+ constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
+ this->lsq_qtmat_c[qtmat_at(i, 2, j, 0)] = 0.2;
+ this->lsq_qtmat_c[qtmat_at(i, 3, j, 0)] = 0.7;
+ this->lsq_qtmat_c[qtmat_at(i, 4, j, 0)] = 1.3;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = 2.0;
+ }
+
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = 1.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ }
+
+ // Test function
+ recon_lsq_cell_q<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.24, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 3.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ -2.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 2.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ -3.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 2.6, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticRandom) {
+ constexpr int nproma = TestFixture::nproma;
+ constexpr int nlev = TestFixture::nlev;
+ constexpr int nblks_c = TestFixture::nblks_c;
+ constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
+ constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_qtmat_c[qtmat_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_q<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ z_qt_times_d[j] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_qt_times_d[j] +=
+ this->lsq_qtmat_c[qtmat_at(jc, j, i, jb)] * z_d[i];
+ }
+ }
+ int utri_id = 0;
+ for (int j = lsq_dim_unk; j > 0; --j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = z_qt_times_d[j - 1];
+ for (int k = j + 1; k <= lsq_dim_unk; ++k) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] -=
+ this->lsq_rmat_utri_c[rmat_utri_at(jc, utri_id++, jb)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(k, jc)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] *=
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, j - 1, jb)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
+
TYPED_TEST_SUITE(HorizontalDivrotCubicTest, ValueTypes);
TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
--
GitLab
From 601153ea69581a5dff8942ecc7e1f1c3475e0f86 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Sun, 9 Mar 2025 11:14:13 +0100
Subject: [PATCH 18/35] Add tests
---
test/c/test_horizontal_divrot.cpp | 217 ++++++++++++++++++++++++++++++
1 file changed, 217 insertions(+)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 37110db..f61fc17 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -569,3 +569,220 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
EXPECT_EQ(TestFixture::lsq_dim_c, 9);
EXPECT_EQ(TestFixture::lsq_dim_unk, 9);
}
+
+TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
+ constexpr int nproma = TestFixture::nproma;
+ constexpr int nlev = TestFixture::nlev;
+ constexpr int nblks_c = TestFixture::nblks_c;
+ constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
+ constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.9;
+ this->lsq_qtmat_c[qtmat_at(i, 2, j, 0)] = 0.8;
+ this->lsq_qtmat_c[qtmat_at(i, 3, j, 0)] = 0.7;
+ this->lsq_qtmat_c[qtmat_at(i, 4, j, 0)] = 0.6;
+ this->lsq_qtmat_c[qtmat_at(i, 5, j, 0)] = 0.5;
+ this->lsq_qtmat_c[qtmat_at(i, 6, j, 0)] = 0.4;
+ this->lsq_qtmat_c[qtmat_at(i, 7, j, 0)] = 0.3;
+ this->lsq_qtmat_c[qtmat_at(i, 8, j, 0)] = 0.2;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = 2.0;
+ }
+
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = 1.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
+ this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
+ this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
+ this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
+ }
+
+ // Test function
+ recon_lsq_cell_c<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.28, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
+ 0.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicRandom) {
+ constexpr int nproma = TestFixture::nproma;
+ constexpr int nlev = TestFixture::nlev;
+ constexpr int nblks_c = TestFixture::nblks_c;
+ constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
+ constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_qtmat_c[qtmat_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_c<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ z_qt_times_d[j] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_qt_times_d[j] +=
+ this->lsq_qtmat_c[qtmat_at(jc, j, i, jb)] * z_d[i];
+ }
+ }
+ int utri_id = 0;
+ for (int j = lsq_dim_unk; j > 0; --j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = z_qt_times_d[j - 1];
+ for (int k = j + 1; k <= lsq_dim_unk; ++k) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] -=
+ this->lsq_rmat_utri_c[rmat_utri_at(jc, utri_id++, jb)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(k, jc)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] *=
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, j - 1, jb)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
--
GitLab
From 1f6c95b342ea4c6e42388ccd112049e51502dff7 Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Wed, 12 Mar 2025 15:54:21 +0100
Subject: [PATCH 19/35] Inner product using lambda functions
---
src/horizontal/CMakeLists.txt | 4 +-
src/horizontal/lib_divrot.cpp | 71 +++++++++++++----------------------
2 files changed, 29 insertions(+), 46 deletions(-)
diff --git a/src/horizontal/CMakeLists.txt b/src/horizontal/CMakeLists.txt
index d403cb2..af52206 100644
--- a/src/horizontal/CMakeLists.txt
+++ b/src/horizontal/CMakeLists.txt
@@ -22,7 +22,9 @@ set(Fortran_MODULE_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/mod")
set_target_properties(
iconmath-horizontal
PROPERTIES Fortran_MODULE_DIRECTORY "${Fortran_MODULE_DIRECTORY}"
- EXPORT_NAME ${PROJECT_NAME}::horizontal)
+ EXPORT_NAME ${PROJECT_NAME}::horizontal
+ CXX_STANDARD 20
+ CXX_STANDARD_REQUIRED ON)
if(IM_ENABLE_LOOP_EXCHANGE)
target_compile_definitions(iconmath-horizontal PRIVATE __LOOP_EXCHANGE)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 164be38..49e72c3 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -15,6 +15,18 @@
#include <horizontal/lib_divrot.hpp>
#include <support/mo_lib_loopindices.hpp>
+#define DECLARE_LAMBDA_INNER_PRODUCT(_func_name, _output, _id, _lambda) \
+ auto inner_product = [=, &_output](int _id, auto &&...ts) { \
+ return [=, &_output] { \
+ _output(_id) = 0.0; \
+ int dummy[sizeof...(ts)]{(_lambda, 0)...}; \
+ }; \
+ }; \
+ auto _func_name = [=]<int... Is>(int _id, \
+ std::integer_sequence<int, Is...>) { \
+ return inner_product(_id, Is...)(); \
+ };
+
template <typename T>
void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const T *lsq_qtmat_c,
@@ -261,51 +273,20 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::parallel_for(
"recon_lsq_cell_q_step2", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
- z_qt_times_d(0) = lsq_qtmat_c_view(jc, 0, 0, jb) * z_d(0, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 1, jb) * z_d(1, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 2, jb) * z_d(2, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 3, jb) * z_d(3, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 4, jb) * z_d(4, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 5, jb) * z_d(5, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 6, jb) * z_d(6, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 0, 8, jb) * z_d(8, jc, jk);
- z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 0, jb) * z_d(0, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 4, jb) * z_d(4, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 5, jb) * z_d(5, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 6, jb) * z_d(6, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 1, 8, jb) * z_d(8, jc, jk);
- z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 0, jb) * z_d(0, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 4, jb) * z_d(4, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 5, jb) * z_d(5, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 6, jb) * z_d(6, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 2, 8, jb) * z_d(8, jc, jk);
- z_qt_times_d(3) = lsq_qtmat_c_view(jc, 3, 0, jb) * z_d(0, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 1, jb) * z_d(1, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 2, jb) * z_d(2, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 3, jb) * z_d(3, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 4, jb) * z_d(4, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 5, jb) * z_d(5, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 6, jb) * z_d(6, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 3, 8, jb) * z_d(8, jc, jk);
- z_qt_times_d(4) = lsq_qtmat_c_view(jc, 4, 0, jb) * z_d(0, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 1, jb) * z_d(1, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 2, jb) * z_d(2, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 3, jb) * z_d(3, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 4, jb) * z_d(4, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 5, jb) * z_d(5, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 6, jb) * z_d(6, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 7, jb) * z_d(7, jc, jk) +
- lsq_qtmat_c_view(jc, 4, 8, jb) * z_d(8, jc, jk);
+ auto lambda_add = [=, &z_qt_times_d](auto lsq_qtmat_c_view, auto z_d,
+ int jb, int jk, int jc, int unk,
+ int i) {
+ z_qt_times_d(unk) +=
+ lsq_qtmat_c_view(jc, unk, i, jb) * z_d(i, jc, jk);
+ };
+ DECLARE_LAMBDA_INNER_PRODUCT(
+ dot_product, z_qt_times_d, unk,
+ lambda_add(lsq_qtmat_c_view, z_d, jb, jk, jc, unk, ts));
+ dot_product(0, std::make_integer_sequence<int, 9>());
+ dot_product(1, std::make_integer_sequence<int, 9>());
+ dot_product(2, std::make_integer_sequence<int, 9>());
+ dot_product(3, std::make_integer_sequence<int, 9>());
+ dot_product(4, std::make_integer_sequence<int, 9>());
p_coeff_view(5, jc, jk, jb) = ptr_rrdiag(jc, 4, jb) * z_qt_times_d(4);
p_coeff_view(4, jc, jk, jb) =
--
GitLab
From 23e92b7162abea31d868c4aa9717cd31e4a1b94d Mon Sep 17 00:00:00 2001
From: Yen-Chen Chen <yen-chen.chen@tum.de>
Date: Wed, 12 Mar 2025 15:55:43 +0100
Subject: [PATCH 20/35] Revert "Inner product using lambda functions"
This reverts commit 1f6c95b342ea4c6e42388ccd112049e51502dff7.
---
src/horizontal/CMakeLists.txt | 4 +-
src/horizontal/lib_divrot.cpp | 71 ++++++++++++++++++++++-------------
2 files changed, 46 insertions(+), 29 deletions(-)
diff --git a/src/horizontal/CMakeLists.txt b/src/horizontal/CMakeLists.txt
index af52206..d403cb2 100644
--- a/src/horizontal/CMakeLists.txt
+++ b/src/horizontal/CMakeLists.txt
@@ -22,9 +22,7 @@ set(Fortran_MODULE_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/mod")
set_target_properties(
iconmath-horizontal
PROPERTIES Fortran_MODULE_DIRECTORY "${Fortran_MODULE_DIRECTORY}"
- EXPORT_NAME ${PROJECT_NAME}::horizontal
- CXX_STANDARD 20
- CXX_STANDARD_REQUIRED ON)
+ EXPORT_NAME ${PROJECT_NAME}::horizontal)
if(IM_ENABLE_LOOP_EXCHANGE)
target_compile_definitions(iconmath-horizontal PRIVATE __LOOP_EXCHANGE)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 49e72c3..164be38 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -15,18 +15,6 @@
#include <horizontal/lib_divrot.hpp>
#include <support/mo_lib_loopindices.hpp>
-#define DECLARE_LAMBDA_INNER_PRODUCT(_func_name, _output, _id, _lambda) \
- auto inner_product = [=, &_output](int _id, auto &&...ts) { \
- return [=, &_output] { \
- _output(_id) = 0.0; \
- int dummy[sizeof...(ts)]{(_lambda, 0)...}; \
- }; \
- }; \
- auto _func_name = [=]<int... Is>(int _id, \
- std::integer_sequence<int, Is...>) { \
- return inner_product(_id, Is...)(); \
- };
-
template <typename T>
void recon_lsq_cell_l(const T *p_cc, const int *cell_neighbor_idx,
const int *cell_neighbor_blk, const T *lsq_qtmat_c,
@@ -273,20 +261,51 @@ void recon_lsq_cell_q(const T *p_cc, const int *lsq_idx_c, const int *lsq_blk_c,
Kokkos::parallel_for(
"recon_lsq_cell_q_step2", innerPolicy,
KOKKOS_LAMBDA(const int jk, const int jc) {
- auto lambda_add = [=, &z_qt_times_d](auto lsq_qtmat_c_view, auto z_d,
- int jb, int jk, int jc, int unk,
- int i) {
- z_qt_times_d(unk) +=
- lsq_qtmat_c_view(jc, unk, i, jb) * z_d(i, jc, jk);
- };
- DECLARE_LAMBDA_INNER_PRODUCT(
- dot_product, z_qt_times_d, unk,
- lambda_add(lsq_qtmat_c_view, z_d, jb, jk, jc, unk, ts));
- dot_product(0, std::make_integer_sequence<int, 9>());
- dot_product(1, std::make_integer_sequence<int, 9>());
- dot_product(2, std::make_integer_sequence<int, 9>());
- dot_product(3, std::make_integer_sequence<int, 9>());
- dot_product(4, std::make_integer_sequence<int, 9>());
+ z_qt_times_d(0) = lsq_qtmat_c_view(jc, 0, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 0, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(1) = lsq_qtmat_c_view(jc, 1, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 1, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(2) = lsq_qtmat_c_view(jc, 2, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 2, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(3) = lsq_qtmat_c_view(jc, 3, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 3, 8, jb) * z_d(8, jc, jk);
+ z_qt_times_d(4) = lsq_qtmat_c_view(jc, 4, 0, jb) * z_d(0, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 1, jb) * z_d(1, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 2, jb) * z_d(2, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 3, jb) * z_d(3, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 4, jb) * z_d(4, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 5, jb) * z_d(5, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 6, jb) * z_d(6, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 7, jb) * z_d(7, jc, jk) +
+ lsq_qtmat_c_view(jc, 4, 8, jb) * z_d(8, jc, jk);
p_coeff_view(5, jc, jk, jb) = ptr_rrdiag(jc, 4, jb) * z_qt_times_d(4);
p_coeff_view(4, jc, jk, jb) =
--
GitLab
From acf957e43ac783f5a154816f2ce2cab17567e75b Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Fri, 14 Mar 2025 21:22:35 +0100
Subject: [PATCH 21/35] replaced TestFixture:: with this->
---
test/c/test_horizontal_divrot.cpp | 60 +++++++++++++++----------------
1 file changed, 30 insertions(+), 30 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index f61fc17..5e7bb25 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -196,11 +196,11 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestLsqDimensions) {
}
TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
- constexpr int nproma = TestFixture::nproma;
- constexpr int nlev = TestFixture::nlev;
- constexpr int nblks_c = TestFixture::nblks_c;
- constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
- constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
@@ -258,11 +258,11 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
}
TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
- constexpr int nproma = TestFixture::nproma;
- constexpr int nlev = TestFixture::nlev;
- constexpr int nblks_c = TestFixture::nblks_c;
- constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
- constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
@@ -367,11 +367,11 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestLsqDimensions) {
}
TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
- constexpr int nproma = TestFixture::nproma;
- constexpr int nlev = TestFixture::nlev;
- constexpr int nblks_c = TestFixture::nblks_c;
- constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
- constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
@@ -452,11 +452,11 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
}
TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticRandom) {
- constexpr int nproma = TestFixture::nproma;
- constexpr int nlev = TestFixture::nlev;
- constexpr int nblks_c = TestFixture::nblks_c;
- constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
- constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
@@ -571,11 +571,11 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
}
TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
- constexpr int nproma = TestFixture::nproma;
- constexpr int nlev = TestFixture::nlev;
- constexpr int nblks_c = TestFixture::nblks_c;
- constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
- constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
@@ -676,11 +676,11 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
}
TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicRandom) {
- constexpr int nproma = TestFixture::nproma;
- constexpr int nlev = TestFixture::nlev;
- constexpr int nblks_c = TestFixture::nblks_c;
- constexpr int lsq_dim_c = TestFixture::lsq_dim_c;
- constexpr int lsq_dim_unk = TestFixture::lsq_dim_unk;
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
const auto &p_cc_at = at<nproma, nlev, nblks_c>;
const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
--
GitLab
From fd8760ae7f161e74731d3edab176dfe7fe069c9e Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Fri, 14 Mar 2025 21:33:27 +0100
Subject: [PATCH 22/35] removed an unused argument to one of the function in
mo_lib_divrot
---
src/horizontal/lib_divrot.cpp | 2 +-
src/horizontal/lib_divrot.hpp | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/lib_divrot.cpp
index 164be38..a24981d 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/lib_divrot.cpp
@@ -733,7 +733,7 @@ void recon_lsq_cell_c_svd(const T *p_cc, const int *lsq_idx_c,
const T *lsq_moments, T *p_coeff, int i_startblk,
int i_endblk, int i_startidx_in, int i_endidx_in,
int slev, int elev, int nproma, int patch_id,
- int lsq_high_set_dim_c, bool l_limited_area,
+ bool l_limited_area,
bool lacc, int nblks_c, int nlev, int lsq_dim_unk,
int lsq_dim_c) {
// Wrap raw pointers in unmanaged Kokkos Views.
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/lib_divrot.hpp
index dae8282..b8e9743 100644
--- a/src/horizontal/lib_divrot.hpp
+++ b/src/horizontal/lib_divrot.hpp
@@ -64,7 +64,7 @@
const _type *p_cc, const int *lsq_idx_c, const int *lsq_blk_c, \
const _type *lsq_pseudoinv, const _type *lsq_moments, _type *p_coeff, \
int i_startblk, int i_endblk, int i_startidx_in, int i_endidx_in, \
- int slev, int elev, int nproma, int patch_id, int lsq_high_set_dim_c, \
+ int slev, int elev, int nproma, int patch_id, \
bool l_limited_area, bool lacc, int nblks_c, int nlev, int lsq_dim_unk, \
int lsq_dim_c)
--
GitLab
From 5d7f4c34c27afaec018e05f5c603c0cc3b530765 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Fri, 14 Mar 2025 21:34:07 +0100
Subject: [PATCH 23/35] added unit tests for all the svd functions
---
test/c/test_horizontal_divrot.cpp | 508 ++++++++++++++++++++++++++++++
1 file changed, 508 insertions(+)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 5e7bb25..11f98aa 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -148,6 +148,7 @@ protected:
std::vector<ValueType> lsq_rmat_rdiag_c;
std::vector<ValueType> lsq_rmat_utri_c;
std::vector<ValueType> lsq_moments;
+ std::vector<ValueType> lsq_pseudoinv;
std::vector<ValueType> p_coeff;
HorizontalDivrotTest() {
@@ -159,6 +160,7 @@ protected:
lsq_rmat_utri_c.resize(dim_combine(
nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c));
lsq_moments.resize(dim_combine(nproma, nblks_c, lsq_dim_unk));
+ lsq_pseudoinv.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
p_coeff.resize(dim_combine(lsq_dim_unk + 1, nproma, nlev, nblks_c));
}
};
@@ -257,6 +259,64 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
1.0, 1e-6);
}
+TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
+ // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
+ // this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ }
+
+ // Test function
+ recon_lsq_cell_l_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.65, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.5, 1e-6);
+}
+
TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -359,6 +419,95 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
}
}
+TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVDRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = real_distrib(gen);
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = real_distrib(gen);
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, 1)] = real_distrib(gen);
+ }
+
+ // Test function
+ recon_lsq_cell_l_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] =
+ this->lsq_pseudoinv[pseudoinv_at(jc, 1, 0, jb)] * z_d[0] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 1, 1, jb)] * z_d[1] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 1, 2, jb)] * z_d[2];
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] =
+ this->lsq_pseudoinv[pseudoinv_at(jc, 0, 0, jb)] * z_d[0] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 0, 1, jb)] * z_d[1] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 0, 2, jb)] * z_d[2];
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 0)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 1)];
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
+
TYPED_TEST_SUITE(HorizontalDivrotQuadraticTest, ValueTypes);
TYPED_TEST(HorizontalDivrotQuadraticTest, TestLsqDimensions) {
@@ -451,6 +600,79 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
2.6, 1e-6);
}
+TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
+ this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.2;
+ this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
+ this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 1.3;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ }
+
+ // Test function
+ recon_lsq_cell_q_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ -0.56, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.5, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ 0.7, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 1.3, 1e-6);
+}
+
TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -563,6 +785,104 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticRandom) {
}
}
+TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVDRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization is done only for iblk = 0 and ilev = 0
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_pseudoinv[pseudoinv_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_q_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ // for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ {int jb = 0;
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ // for (int jk = this->slev; jk < this->elev; ++jk) {
+ {int jk = 0;
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 1; j < lsq_dim_unk + 1; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
+ this->lsq_pseudoinv[pseudoinv_at(jc, j-1, i, jb)] * z_d[i];
+ }
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+
+ // Check result
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(j, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(j, jc))], 1e-5)
+ << "For loop result fails for j = " << j << ", jc = " << jc;
+ }
+ }
+}
+
TYPED_TEST_SUITE(HorizontalDivrotCubicTest, ValueTypes);
TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
@@ -675,6 +995,99 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
0.4, 1e-6);
}
+TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.9;
+ this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.8;
+ this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
+ this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 0.6;
+ this->lsq_pseudoinv[pseudoinv_at(i, 5, j, 0)] = 0.5;
+ this->lsq_pseudoinv[pseudoinv_at(i, 6, j, 0)] = 0.4;
+ this->lsq_pseudoinv[pseudoinv_at(i, 7, j, 0)] = 0.3;
+ this->lsq_pseudoinv[pseudoinv_at(i, 8, j, 0)] = 0.2;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
+ this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
+ this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
+ this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
+ }
+
+ // Test function
+ recon_lsq_cell_c_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ -1.64, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.9, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ 0.7, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 0.6, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
+ 0.5, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
+ 0.3, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
+ 0.2, 1e-6);
+}
+
TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -786,3 +1199,98 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicRandom) {
}
}
}
+
+TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicSVDRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_pseudoinv[pseudoinv_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_c_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+
+ // for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ {int jb = 0;
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ // for (int jk = this->slev; jk < this->elev; ++jk) {
+ {int jk = 0;
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 1; j < lsq_dim_unk + 1; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
+ this->lsq_pseudoinv[pseudoinv_at(jc, j-1, i, jb)] * z_d[i];
+ }
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
--
GitLab
From 0dca3eff9d4920e4af90095384771838e5637298 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Fri, 14 Mar 2025 21:34:53 +0100
Subject: [PATCH 24/35] removed all TestLsqDimensions
---
test/c/test_horizontal_divrot.cpp | 15 ---------------
1 file changed, 15 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 11f98aa..6d0b3c6 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -192,11 +192,6 @@ typedef ::testing::Types<float, double> ValueTypes;
TYPED_TEST_SUITE(HorizontalDivrotLinearTest, ValueTypes);
-TYPED_TEST(HorizontalDivrotLinearTest, TestLsqDimensions) {
- EXPECT_EQ(TestFixture::lsq_dim_c, 3);
- EXPECT_EQ(TestFixture::lsq_dim_unk, 2);
-}
-
TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -510,11 +505,6 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVDRandom) {
TYPED_TEST_SUITE(HorizontalDivrotQuadraticTest, ValueTypes);
-TYPED_TEST(HorizontalDivrotQuadraticTest, TestLsqDimensions) {
- EXPECT_EQ(TestFixture::lsq_dim_c, 9);
- EXPECT_EQ(TestFixture::lsq_dim_unk, 5);
-}
-
TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -885,11 +875,6 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVDRandom) {
TYPED_TEST_SUITE(HorizontalDivrotCubicTest, ValueTypes);
-TYPED_TEST(HorizontalDivrotCubicTest, TestLsqDimensions) {
- EXPECT_EQ(TestFixture::lsq_dim_c, 9);
- EXPECT_EQ(TestFixture::lsq_dim_unk, 9);
-}
-
TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
--
GitLab
From a43cebbf77f72c2b8ec4a03f14c8c06953a568a1 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Sun, 16 Mar 2025 08:35:04 +0100
Subject: [PATCH 25/35] renamed the unit test to remove redundant parts
---
test/c/test_horizontal_divrot.cpp | 46 +++++++++++++++----------------
1 file changed, 23 insertions(+), 23 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 6d0b3c6..4ef6b7d 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -104,7 +104,7 @@ enum class ReconstructionMethod {
/// Base test class for the horizontal divrot tests. Templated for the ValueType
/// and ReconMethod for the reconstruction method.
template <typename ValueType, int ReconMethod>
-class HorizontalDivrotTest : public ::testing::Test {
+class HorizontalReconTest : public ::testing::Test {
protected:
// lsq_dim_c and lsq_dim_unk are instantiated in compile time.
static constexpr std::tuple<int, int>
@@ -151,7 +151,7 @@ protected:
std::vector<ValueType> lsq_pseudoinv;
std::vector<ValueType> p_coeff;
- HorizontalDivrotTest() {
+ HorizontalReconTest() {
p_cc.resize(dim_combine(nproma, nlev, nblks_c));
cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
@@ -168,31 +168,31 @@ protected:
/// Test class for the horizontal tests. The reconstruction method is specified
/// to linear.
template <typename ValueType>
-class HorizontalDivrotLinearTest
- : public HorizontalDivrotTest<
+class HorizontalReconLinearTest
+ : public HorizontalReconTest<
ValueType, static_cast<int>(ReconstructionMethod::linear)> {};
/// Test class for the horizontal tests. The reconstruction method is specified
/// to quadratic.
template <typename ValueType>
-class HorizontalDivrotQuadraticTest
- : public HorizontalDivrotTest<
+class HorizontalReconQuadraticTest
+ : public HorizontalReconTest<
ValueType, static_cast<int>(ReconstructionMethod::quadratic)> {};
/// Test class for the horizontal tests. The reconstruction method is specified
/// to cubic.
template <typename ValueType>
-class HorizontalDivrotCubicTest
- : public HorizontalDivrotTest<ValueType, static_cast<int>(
+class HorizontalReconCubicTest
+ : public HorizontalReconTest<ValueType, static_cast<int>(
ReconstructionMethod::cubic)> {
};
/// ValueTypes which the divrot tests should run with
typedef ::testing::Types<float, double> ValueTypes;
-TYPED_TEST_SUITE(HorizontalDivrotLinearTest, ValueTypes);
+TYPED_TEST_SUITE(HorizontalReconLinearTest, ValueTypes);
-TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCell) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -254,7 +254,7 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinear) {
1.0, 1e-6);
}
-TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVD) {
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVD) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -312,7 +312,7 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVD) {
0.5, 1e-6);
}
-TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -414,7 +414,7 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearRandom) {
}
}
-TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVDRandom) {
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVDRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -503,9 +503,9 @@ TYPED_TEST(HorizontalDivrotLinearTest, TestReconLsqCellLinearSVDRandom) {
}
}
-TYPED_TEST_SUITE(HorizontalDivrotQuadraticTest, ValueTypes);
+TYPED_TEST_SUITE(HorizontalReconQuadraticTest, ValueTypes);
-TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCell) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -590,7 +590,7 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadratic) {
2.6, 1e-6);
}
-TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVD) {
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVD) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -663,7 +663,7 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVD) {
1.3, 1e-6);
}
-TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticRandom) {
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -775,7 +775,7 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticRandom) {
}
}
-TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVDRandom) {
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -873,9 +873,9 @@ TYPED_TEST(HorizontalDivrotQuadraticTest, TestReconLsqCellQuadraticSVDRandom) {
}
}
-TYPED_TEST_SUITE(HorizontalDivrotCubicTest, ValueTypes);
+TYPED_TEST_SUITE(HorizontalReconCubicTest, ValueTypes);
-TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCell) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -980,7 +980,7 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubic) {
0.4, 1e-6);
}
-TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicSVD) {
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVD) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1073,7 +1073,7 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicSVD) {
0.2, 1e-6);
}
-TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicRandom) {
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1185,7 +1185,7 @@ TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicRandom) {
}
}
-TYPED_TEST(HorizontalDivrotCubicTest, TestReconLsqCellCubicSVDRandom) {
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
--
GitLab
From 24f64c4cfaee1a6a112b04b88bcb3fac010b28cc Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Sun, 16 Mar 2025 15:29:23 +0100
Subject: [PATCH 26/35] reordered the unit tests
---
test/c/test_horizontal_divrot.cpp | 448 +++++++++++++++---------------
1 file changed, 224 insertions(+), 224 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 4ef6b7d..60f0641 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -254,64 +254,6 @@ TYPED_TEST(HorizontalReconLinearTest, TestLsqCell) {
1.0, 1e-6);
}
-TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVD) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
- // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
- // this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- }
-
- // Test function
- recon_lsq_cell_l_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
- this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- 0.65, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.0, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 0.5, 1e-6);
-}
-
TYPED_TEST(HorizontalReconLinearTest, TestLsqCellRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -414,6 +356,64 @@ TYPED_TEST(HorizontalReconLinearTest, TestLsqCellRandom) {
}
}
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
+ // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
+ // this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ }
+
+ // Test function
+ recon_lsq_cell_l_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.65, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.5, 1e-6);
+}
+
TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVDRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -590,79 +590,6 @@ TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCell) {
2.6, 1e-6);
}
-TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVD) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
- for (int j = 1; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
- this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.2;
- this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
- this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 1.3;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
- this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
- this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
- }
-
- // Test function
- recon_lsq_cell_q_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- -0.56, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.0, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 0.5, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
- 0.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
- 0.7, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
- 1.3, 1e-6);
-}
-
TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -775,6 +702,79 @@ TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellRandom) {
}
}
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
+ this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.2;
+ this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
+ this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 1.3;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ }
+
+ // Test function
+ recon_lsq_cell_q_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ -0.56, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.5, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ 0.7, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 1.3, 1e-6);
+}
+
TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -980,99 +980,6 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCell) {
0.4, 1e-6);
}
-TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVD) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
- for (int j = 1; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.9;
- this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.8;
- this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
- this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 0.6;
- this->lsq_pseudoinv[pseudoinv_at(i, 5, j, 0)] = 0.5;
- this->lsq_pseudoinv[pseudoinv_at(i, 6, j, 0)] = 0.4;
- this->lsq_pseudoinv[pseudoinv_at(i, 7, j, 0)] = 0.3;
- this->lsq_pseudoinv[pseudoinv_at(i, 8, j, 0)] = 0.2;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
- this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
- this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
- this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
- this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
- this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
- this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
- }
-
- // Test function
- recon_lsq_cell_c_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- -1.64, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.0, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 0.9, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
- 0.8, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
- 0.7, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
- 0.6, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
- 0.5, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
- 0.4, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
- 0.3, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
- 0.2, 1e-6);
-}
-
TYPED_TEST(HorizontalReconCubicTest, TestLsqCellRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
@@ -1185,6 +1092,99 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCellRandom) {
}
}
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.9;
+ this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.8;
+ this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
+ this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 0.6;
+ this->lsq_pseudoinv[pseudoinv_at(i, 5, j, 0)] = 0.5;
+ this->lsq_pseudoinv[pseudoinv_at(i, 6, j, 0)] = 0.4;
+ this->lsq_pseudoinv[pseudoinv_at(i, 7, j, 0)] = 0.3;
+ this->lsq_pseudoinv[pseudoinv_at(i, 8, j, 0)] = 0.2;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
+ this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
+ this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
+ this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
+ }
+
+ // Test function
+ recon_lsq_cell_c_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ -1.64, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.9, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ 0.7, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 0.6, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
+ 0.5, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
+ 0.3, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
+ 0.2, 1e-6);
+}
+
TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
--
GitLab
From 3bf9a38b5a4cb67d5b1be7acbda20300d82bbd03 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Sun, 16 Mar 2025 22:18:39 +0100
Subject: [PATCH 27/35] added the unit test for div3d
---
test/c/test_horizontal_divrot.cpp | 183 ++++++++++++++++++++++++++++++
1 file changed, 183 insertions(+)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 60f0641..078e753 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -1279,3 +1279,186 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
}
}
}
+
+template <typename ValueType>
+class HorizontalDivTest : public ::testing::Test {
+protected:
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 2; // number of vertical levels
+ static constexpr int nblks_c = 1; // number of cell blocks
+ static constexpr int nblks_e = 1; // number of edge blocks
+ static constexpr int dim4d = 2; // 4th dimension size
+
+ int i_startblk = 0;
+ int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
+ int i_startidx_in = 0;
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ std::vector<int> slev;
+ std::vector<int> elev;
+ bool lacc = false; // Not using ACC-specific behavior.
+
+ std::vector<ValueType> vec_e;
+ std::vector<int> cell_edge_idx;
+ std::vector<int> cell_edge_blk;
+ std::vector<ValueType> geofac_div;
+ std::vector<ValueType> div_vec_c;
+ std::vector<ValueType> f4din;
+ std::vector<ValueType> f4dout;
+
+ HorizontalDivTest() {
+ slev.resize(dim4d, 0);
+ elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
+
+ vec_e.resize(dim_combine(nproma, nlev, nblks_e));
+ cell_edge_idx.resize(dim_combine(nproma, nblks_c, 3));
+ cell_edge_blk.resize(dim_combine(nproma, nblks_c, 3));
+ geofac_div.resize(dim_combine(nproma, 3, nblks_c));
+ div_vec_c.resize(dim_combine(nproma, nlev, nblks_c));
+ f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
+ f4dout.resize(dim_combine(nproma, nlev, nblks_c, dim4d));
+ }
+};
+
+template <typename ValueType>
+class HorizontalDiv3DTest
+ : public HorizontalDivTest<ValueType> {};
+TYPED_TEST_SUITE(HorizontalDiv3DTest, ValueTypes);
+
+TYPED_TEST(HorizontalDiv3DTest, TestSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ }
+
+ // Set edge indices to point to specific cells (including self)
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Initialize div_vec_c to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the div3d function
+ div3d<TypeParam>(
+ this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
+
+}
+
+TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize div_vec_c to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div3d function
+ div3d<TypeParam>(
+ this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "Results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
--
GitLab
From 03cbb3f2c755756a90dfc6013238eb5d35ec0b38 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Sun, 16 Mar 2025 22:22:20 +0100
Subject: [PATCH 28/35] added rest of the unit tests for div3d, div4d and
divavg
removed some redundant things
---
test/c/test_horizontal_divrot.cpp | 850 ++++++++++++++++++++++++++++++
1 file changed, 850 insertions(+)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 078e753..2ad95d2 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -1305,6 +1305,13 @@ protected:
std::vector<ValueType> f4din;
std::vector<ValueType> f4dout;
+ // Followings are needed in HorizontalDivAvgTest
+ std::vector<int> cell_neighbor_idx;
+ std::vector<int> cell_neighbor_blk;
+ std::vector<ValueType> avg_coeff;
+ std::vector<ValueType> opt_in2;
+ std::vector<ValueType> opt_out2;
+
HorizontalDivTest() {
slev.resize(dim4d, 0);
elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
@@ -1316,12 +1323,30 @@ protected:
div_vec_c.resize(dim_combine(nproma, nlev, nblks_c));
f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
f4dout.resize(dim_combine(nproma, nlev, nblks_c, dim4d));
+ cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, 3));
+ cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, 3));
+ avg_coeff.resize(dim_combine(nproma, 4, nblks_c));
+ opt_in2.resize(dim_combine(nproma, nlev, nblks_e));
+ opt_out2.resize(dim_combine(nproma, nlev, nblks_c));
}
};
template <typename ValueType>
class HorizontalDiv3DTest
: public HorizontalDivTest<ValueType> {};
+
+template <typename ValueType>
+class HorizontalDiv3D2FTest
+ : public HorizontalDivTest<ValueType> {};
+
+template <typename ValueType>
+class HorizontalDiv4DTest
+ : public HorizontalDivTest<ValueType> {};
+
+template <typename ValueType>
+class HorizontalDivAvgTest
+ : public HorizontalDivTest<ValueType> {};
+
TYPED_TEST_SUITE(HorizontalDiv3DTest, ValueTypes);
TYPED_TEST(HorizontalDiv3DTest, TestSpecific) {
@@ -1462,3 +1487,828 @@ TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
}
}
}
+
+TYPED_TEST_SUITE(HorizontalDiv3D2FTest, ValueTypes);
+
+TYPED_TEST(HorizontalDiv3D2FTest, TestSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+ const auto &f4d_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ this->f4din[f4d_at(i, k, 0, 0)] = (i + 1) * (k + 2); // Different pattern for second field
+ }
+
+ // Set edge indices to point to specific cells (including self)
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Initialize div_vec_c and f4dout to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ this->f4dout[f4dout_at(i, k, 0, 0)] = 0.0;
+ }
+ }
+
+ // Call the div3d_2field function
+ div3d_2field<TypeParam>(
+ this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->f4din.data(), this->f4dout.data(),
+ this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Check first field (same as in div3d test)
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
+
+ // Check second field (expected values calculated manually)
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 0)], 5.1, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 0)], 6.3, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 0)], 4.4, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 6.6, 1e-6);
+}
+
+TYPED_TEST(HorizontalDiv3D2FTest, TestRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+ const auto &f4d_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ this->f4din[f4d_at(i, k, 0, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize div_vec_c and f4dout to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->f4dout[f4dout_at(i, k, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div3d_2field function
+ div3d_2field<TypeParam>(
+ this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->f4din.data(), this->f4dout.data(),
+ this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
+ std::vector<TypeParam> ref_f4dout(nproma * nlev * nblks_c * dim4d, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ // Calculate reference value for first field
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+
+ // Calculate reference value for second field
+ ref_f4dout[f4dout_at(jc, jk, jb, 0)] =
+ this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)], 0)] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)], 0)] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)], 0)] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Verify results for first field
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "First field results differ at i=" << i << ", k=" << k;
+ }
+ }
+
+ // Verify results for second field
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->f4dout[f4dout_at(i, k, 0, 0)],
+ ref_f4dout[f4dout_at(i, k, 0, 0)], 1e-5)
+ << "Second field results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalDiv4DTest, ValueTypes);
+
+TYPED_TEST(HorizontalDiv4DTest, TestSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &f4din_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = (i + j) % nproma;
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->geofac_div[geofac_div_at(i, j, 0)] = 0.1 * (j + 1);
+ }
+
+ for (int k = 0; k < nlev; ++k) {
+ for (int d = 0; d < dim4d; ++d) {
+ this->f4din[f4din_at(i, k, 0, d)] = 1.0 + i + k + d;
+ this->f4dout[f4dout_at(i, k, 0, d)] = 0.0;
+ }
+ }
+ }
+
+ // Test function
+ div4d<TypeParam>(
+ this->cell_edge_idx.data(), this->cell_edge_blk.data(),
+ this->geofac_div.data(), this->f4din.data(), this->f4dout.data(),
+ this->dim4d, this->i_startblk, this->i_endblk, this->i_startidx_in,
+ this->i_endidx_in, this->slev.data(), this->elev.data(), this->nproma,
+ this->lacc, this->nlev, this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 1.4, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 1.1, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 0)], 1.1, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 0)], 2.0, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 1)], 2.0, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 1)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 1)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 1)], 2.6, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 1)], 2.3, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 1)], 2.3, 1e-6);
+
+}
+
+TYPED_TEST(HorizontalDiv4DTest, TestDiv4dRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &f4din_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
+
+ // Initialize with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ for (int k = 0; k < nlev; ++k) {
+ for (int d = 0; d < dim4d; ++d) {
+ this->f4din[f4din_at(i, k, 0, d)] = real_distrib(gen);
+ this->f4dout[f4dout_at(i, k, 0, d)] = 0.0;
+ }
+ }
+ }
+
+ // Test function
+ div4d<TypeParam>(
+ this->cell_edge_idx.data(), this->cell_edge_blk.data(),
+ this->geofac_div.data(), this->f4din.data(), this->f4dout.data(),
+ this->dim4d, this->i_startblk, this->i_endblk, this->i_startidx_in,
+ this->i_endidx_in, this->slev.data(), this->elev.data(), this->nproma,
+ this->lacc, this->nlev, this->nblks_c, this->nblks_e);
+
+ // Compute reference result and check
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int ji = 0; ji < dim4d; ++ji) {
+ for (int jk = this->slev[ji]; jk < this->elev[ji]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ TypeParam expected = 0.0;
+ for (int je = 0; je < 3; ++je) {
+ expected += this->f4din[f4din_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, je)],
+ jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, je)],
+ ji)] * this->geofac_div[geofac_div_at(jc, je, jb)];
+ }
+
+ EXPECT_NEAR(this->f4dout[f4dout_at(jc, jk, jb, ji)], expected, 1e-5)
+ << "Random test fails at jc=" << jc << ", jk=" << jk
+ << ", jb=" << jb << ", ji=" << ji;
+ }
+ }
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalDivAvgTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivAvgTest, TestSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for additional parameters
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = true;
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Initialize the vectors with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ this->opt_in2[vec_e_at(i, k, 0)] = (i + 1) * (k + 1) * 0.5; // Half of vec_e
+ }
+
+ // Set edge indices to point to specific cells
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // Set neighbor indices similarly
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges and neighbors are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Average coefficients
+ this->avg_coeff[avg_coeff_at(i, 0, 0)] = 0.4; // Self
+ this->avg_coeff[avg_coeff_at(i, 1, 0)] = 0.2; // First neighbor
+ this->avg_coeff[avg_coeff_at(i, 2, 0)] = 0.2; // Second neighbor
+ this->avg_coeff[avg_coeff_at(i, 3, 0)] = 0.2; // Third neighbor
+
+ // Initialize div_vec_c and opt_out2 to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ this->opt_out2[div_vec_c_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
+ this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
+ i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
+ this->elev[0], this->nproma, patch_id, l_limited_area,
+ l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.16, 1e-6);
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 0, 0)], 0.94, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 1, 0)], 1.88, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 0, 0)], 1.02, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 2.04, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 1.04, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 2.08, 1e-6);
+
+}
+
+TYPED_TEST(HorizontalDivAvgTest, TestRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = true;
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialize with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ this->opt_in2[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random average coefficients
+ for (int j = 0; j < 4; ++j) {
+ this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random initial values for div_vec_c and opt_out2
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
+ this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
+ i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
+ this->elev[0], this->nproma, patch_id, l_limited_area,
+ l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values manually
+ std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> aux_c2(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_opt_out2(dim_combine(nproma, nlev, nblks_c));
+
+ // Step 1: Calculate aux_c and aux_c2
+ for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ aux_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+
+ aux_c2[div_vec_c_at(jc, jk, jb)] =
+ this->opt_in2[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->opt_in2[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->opt_in2[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Step 2: Assign aux_c to div_vec_c and aux_c2 to opt_out2 for patch_id > 0
+ for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] = aux_c[div_vec_c_at(jc, jk, jb)];
+ ref_opt_out2[div_vec_c_at(jc, jk, jb)] = aux_c2[div_vec_c_at(jc, jk, jb)];
+ }
+ }
+ }
+
+ // Step 3: Perform averaging for the rest of the blocks
+ for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)] * this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+
+ ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
+ aux_c2[div_vec_c_at(jc, jk, jb)] * this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c2[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c2[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c2[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "div_vec_c results differ at i=" << i << ", k=" << k;
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(i, k, 0)],
+ ref_opt_out2[div_vec_c_at(i, k, 0)], 1e-5)
+ << "opt_out2 results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = false;
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Initialize the vectors with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ this->opt_in2[vec_e_at(i, k, 0)] = (i + 1) * (k + 1) * 0.5; // Half of vec_e
+ }
+
+ // Set edge indices to point to specific cells
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // Set neighbor indices similarly
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges and neighbors are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Average coefficients
+ this->avg_coeff[avg_coeff_at(i, 0, 0)] = 0.4; // Self
+ this->avg_coeff[avg_coeff_at(i, 1, 0)] = 0.2; // First neighbor
+ this->avg_coeff[avg_coeff_at(i, 2, 0)] = 0.2; // Second neighbor
+ this->avg_coeff[avg_coeff_at(i, 3, 0)] = 0.2; // Third neighbor
+
+ // Initialize div_vec_c and opt_out2 to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ this->opt_out2[div_vec_c_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
+ this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
+ i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
+ this->elev[0], this->nproma, patch_id, l_limited_area,
+ l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.16, 1e-6);
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 0, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 1, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 0, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 0.0, 1e-6);
+
+}
+
+TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = false; // Set to false for this test
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialize with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ this->opt_in2[vec_e_at(i, k, 0)] = real_distrib(gen); // Not used but initialize anyway
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random average coefficients
+ for (int j = 0; j < 4; ++j) {
+ this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random initial values for div_vec_c and opt_out2
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen); // Not used but initialize anyway
+ }
+ }
+
+ // Call the div_avg function with l2fields=false
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
+ this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
+ i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
+ this->elev[0], this->nproma, patch_id, l_limited_area,
+ l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values manually
+ std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
+
+ // Step 1: Calculate aux_c (but not aux_c2 since l2fields=false)
+ for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ aux_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Step 2: Assign aux_c to div_vec_c for patch_id > 0 (opt_out2 not updated since l2fields=false)
+ for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] = aux_c[div_vec_c_at(jc, jk, jb)];
+ }
+ }
+ }
+
+ // Step 3: Perform averaging for the rest of the blocks (only for div_vec_c, not opt_out2)
+ for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)] * this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+ }
+ }
+ }
+
+ // Verify results - only check div_vec_c since l2fields=false means opt_out2 isn't updated
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "div_vec_c results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
--
GitLab
From 5575b916c2cad0a1f9de33edafd57e10d456e325 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Mon, 17 Mar 2025 11:13:48 +0100
Subject: [PATCH 29/35] added unit tests for rest of the functions
---
test/c/test_horizontal_divrot.cpp | 352 ++++++++++++++++++++++++++++++
1 file changed, 352 insertions(+)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 2ad95d2..8f95e6e 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -2312,3 +2312,355 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
}
}
}
+
+template <typename ValueType>
+class HorizontalRotVertexTest : public ::testing::Test {
+protected:
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 2; // number of vertical levels
+ static constexpr int nblks_e = 1; // number of edge blocks
+ static constexpr int nblks_v = 1; // number of vertex blocks
+ static constexpr int dim4d = 2; // 4th dimension size
+
+ int i_startblk = 0;
+ int i_endblk = nblks_v; // Test blocks [0 .. nblks_v-1]
+ int i_startidx_in = 0;
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ std::vector<int> slev;
+ std::vector<int> elev;
+ bool lacc = false; // Not using ACC-specific behavior.
+ bool acc_async = false; // Not using ACC-specific behavior.
+
+ std::vector<ValueType> vec_e;
+ std::vector<int> vert_edge_idx;
+ std::vector<int> vert_edge_blk;
+ std::vector<ValueType> geofac_rot;
+ std::vector<ValueType> rot_vec;
+ std::vector<ValueType> f4din;
+ std::vector<ValueType> f4dout;
+
+ HorizontalRotVertexTest () {
+ slev.resize(dim4d, 0);
+ elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
+
+ vec_e.resize(dim_combine(nproma, nlev, nblks_e));
+ vert_edge_idx.resize(dim_combine(nproma, nblks_v, 6));
+ vert_edge_blk.resize(dim_combine(nproma, nblks_v, 6));
+ geofac_rot.resize(dim_combine(nproma, 6, nblks_v));
+ rot_vec.resize(dim_combine(nproma, nlev, nblks_v));
+ f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
+ f4dout.resize(dim_combine(nproma, nlev, nblks_v, dim4d));
+ }
+};
+
+template <typename ValueType>
+class HorizontalRotVertexAtmosTest
+ : public HorizontalRotVertexTest<ValueType> {};
+
+template <typename ValueType>
+class HorizontalRotVertexRITest
+ : public HorizontalRotVertexTest<ValueType> {};
+
+TYPED_TEST_SUITE(HorizontalRotVertexAtmosTest, ValueTypes);
+
+TYPED_TEST(HorizontalRotVertexAtmosTest, TestSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ }
+
+ // Set edge indices to point to specific edges
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = (i + j) % nproma;
+ // All edges are in the same block for this test
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors for rotation
+ this->geofac_rot[geofac_rot_at(i, 0, 0)] = 0.3;
+ this->geofac_rot[geofac_rot_at(i, 1, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 2, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 3, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 4, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 5, 0)] = 0.1;
+
+ // Initialize rot_vec to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the rot_vertex_atmos function
+ rot_vertex_atmos<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(),
+ this->rot_vec.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_e, this->nblks_v);
+
+ // Expected values based on the initialization pattern
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 1, 0)], 4.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalRotVertexAtmosTest, TestRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 6; ++j) {
+ this->geofac_rot[geofac_rot_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize rot_vec to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the rot_vertex_atmos function
+ rot_vertex_atmos<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(),
+ this->rot_vec.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_e, this->nblks_v);
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_rot_vec(nproma * nlev * nblks_v, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_v_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jv = i_startidx; jv < i_endidx; ++jv) {
+ ref_rot_vec[rot_vec_at(jv, jk, jb)] =
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
+ this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
+ this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
+ this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
+ this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
+ this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
+ this->geofac_rot[geofac_rot_at(jv, 5, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(i, k, 0)],
+ ref_rot_vec[rot_vec_at(i, k, 0)], 1e-5)
+ << "Results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalRotVertexRITest, ValueTypes);
+
+TYPED_TEST(HorizontalRotVertexRITest, TestSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ }
+
+ // Set edge indices to point to specific edges
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = (i + j) % nproma;
+ // All edges are in the same block for this test
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors for rotation
+ this->geofac_rot[geofac_rot_at(i, 0, 0)] = 0.3;
+ this->geofac_rot[geofac_rot_at(i, 1, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 2, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 3, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 4, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 5, 0)] = 0.1;
+
+ // Initialize rot_vec to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the rot_vertex_ri function
+ rot_vertex_ri<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(),
+ this->rot_vec.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->acc_async,
+ this->nlev, this->nblks_e, this->nblks_v);
+
+ // Expected values based on the initialization pattern
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 1, 0)], 4.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalRotVertexRITest, TestRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 6; ++j) {
+ this->geofac_rot[geofac_rot_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize rot_vec to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the rot_vertex_ri function
+ rot_vertex_ri<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(),
+ this->rot_vec.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->acc_async,
+ this->nlev, this->nblks_e, this->nblks_v);
+
+ // Ensure computation is complete for both modes
+ Kokkos::fence();
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_rot_vec(nproma * nlev * nblks_v, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_v_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jv = i_startidx; jv < i_endidx; ++jv) {
+ ref_rot_vec[rot_vec_at(jv, jk, jb)] =
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
+ this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
+ this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
+ this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
+ this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
+ this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
+ this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
+ this->geofac_rot[geofac_rot_at(jv, 5, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(i, k, 0)],
+ ref_rot_vec[rot_vec_at(i, k, 0)], 1e-5)
+ << "Results differ at i=" << i << ", k=" << k << ")";
+ }
+ }
+
+}
--
GitLab
From a8245d1276a6efbc480231da0c171662fb9b485c Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Mon, 17 Mar 2025 11:16:18 +0100
Subject: [PATCH 30/35] applied clang-format to the test file
---
test/c/test_horizontal_divrot.cpp | 706 ++++++++++++++++--------------
1 file changed, 374 insertions(+), 332 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 8f95e6e..a97c1da 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -169,8 +169,9 @@ protected:
/// to linear.
template <typename ValueType>
class HorizontalReconLinearTest
- : public HorizontalReconTest<
- ValueType, static_cast<int>(ReconstructionMethod::linear)> {};
+ : public HorizontalReconTest<ValueType, static_cast<int>(
+ ReconstructionMethod::linear)> {
+};
/// Test class for the horizontal tests. The reconstruction method is specified
/// to quadratic.
@@ -184,7 +185,7 @@ class HorizontalReconQuadraticTest
template <typename ValueType>
class HorizontalReconCubicTest
: public HorizontalReconTest<ValueType, static_cast<int>(
- ReconstructionMethod::cubic)> {
+ ReconstructionMethod::cubic)> {
};
/// ValueTypes which the divrot tests should run with
@@ -483,12 +484,12 @@ TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVDRandom) {
this->lsq_pseudoinv[pseudoinv_at(jc, 0, 1, jb)] * z_d[1] +
this->lsq_pseudoinv[pseudoinv_at(jc, 0, 2, jb)] * z_d[2];
p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)] -
- p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
- this->lsq_moments[moments_at(jc, jb, 0)] -
- p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
- this->lsq_moments[moments_at(jc, jb, 1)];
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 0)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 1)];
}
}
}
@@ -832,12 +833,14 @@ TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
// for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- {int jb = 0;
+ {
+ int jb = 0;
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
this->i_startblk, this->i_endblk, i_startidx, i_endidx);
// for (int jk = this->slev; jk < this->elev; ++jk) {
- {int jk = 0;
+ {
+ int jk = 0;
for (int jc = i_startidx; jc < i_endidx; ++jc) {
for (int i = 0; i < lsq_dim_c; ++i) {
z_d[i] = this->p_cc[p_cc_at(
@@ -849,7 +852,7 @@ TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
for (int i = 0; i < lsq_dim_c; ++i) {
p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
- this->lsq_pseudoinv[pseudoinv_at(jc, j-1, i, jb)] * z_d[i];
+ this->lsq_pseudoinv[pseudoinv_at(jc, j - 1, i, jb)] * z_d[i];
}
}
p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
@@ -1240,12 +1243,14 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
// for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- {int jb = 0;
+ {
+ int jb = 0;
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
this->i_startblk, this->i_endblk, i_startidx, i_endidx);
// for (int jk = this->slev; jk < this->elev; ++jk) {
- {int jk = 0;
+ {
+ int jk = 0;
for (int jc = i_startidx; jc < i_endidx; ++jc) {
for (int i = 0; i < lsq_dim_c; ++i) {
z_d[i] = this->p_cc[p_cc_at(
@@ -1257,7 +1262,7 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
for (int i = 0; i < lsq_dim_c; ++i) {
p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
- this->lsq_pseudoinv[pseudoinv_at(jc, j-1, i, jb)] * z_d[i];
+ this->lsq_pseudoinv[pseudoinv_at(jc, j - 1, i, jb)] * z_d[i];
}
}
p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
@@ -1280,14 +1285,13 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
}
}
-template <typename ValueType>
-class HorizontalDivTest : public ::testing::Test {
+template <typename ValueType> class HorizontalDivTest : public ::testing::Test {
protected:
static constexpr int nproma = 3; // inner loop length
static constexpr int nlev = 2; // number of vertical levels
static constexpr int nblks_c = 1; // number of cell blocks
static constexpr int nblks_e = 1; // number of edge blocks
- static constexpr int dim4d = 2; // 4th dimension size
+ static constexpr int dim4d = 2; // 4th dimension size
int i_startblk = 0;
int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
@@ -1332,20 +1336,16 @@ protected:
};
template <typename ValueType>
-class HorizontalDiv3DTest
- : public HorizontalDivTest<ValueType> {};
+class HorizontalDiv3DTest : public HorizontalDivTest<ValueType> {};
template <typename ValueType>
-class HorizontalDiv3D2FTest
- : public HorizontalDivTest<ValueType> {};
+class HorizontalDiv3D2FTest : public HorizontalDivTest<ValueType> {};
template <typename ValueType>
-class HorizontalDiv4DTest
- : public HorizontalDivTest<ValueType> {};
+class HorizontalDiv4DTest : public HorizontalDivTest<ValueType> {};
template <typename ValueType>
-class HorizontalDivAvgTest
- : public HorizontalDivTest<ValueType> {};
+class HorizontalDivAvgTest : public HorizontalDivTest<ValueType> {};
TYPED_TEST_SUITE(HorizontalDiv3DTest, ValueTypes);
@@ -1388,13 +1388,12 @@ TYPED_TEST(HorizontalDiv3DTest, TestSpecific) {
}
// Call the div3d function
- div3d<TypeParam>(
- this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
+ div3d<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
@@ -1402,7 +1401,6 @@ TYPED_TEST(HorizontalDiv3DTest, TestSpecific) {
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
-
}
TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
@@ -1431,7 +1429,8 @@ TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
// Set random edge indices
for (int j = 0; j < 3; ++j) {
this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
}
// Random geometric factors
@@ -1446,13 +1445,12 @@ TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
}
// Call the div3d function
- div3d<TypeParam>(
- this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
+ div3d<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
// Calculate reference values separately and verify results
std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
@@ -1460,19 +1458,22 @@ TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
this->geofac_div[geofac_div_at(jc, 2, jb)];
}
}
@@ -1508,7 +1509,8 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestSpecific) {
for (int i = 0; i < nproma; ++i) {
for (int k = 0; k < nlev; ++k) {
this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- this->f4din[f4d_at(i, k, 0, 0)] = (i + 1) * (k + 2); // Different pattern for second field
+ this->f4din[f4d_at(i, k, 0, 0)] =
+ (i + 1) * (k + 2); // Different pattern for second field
}
// Set edge indices to point to specific cells (including self)
@@ -1534,14 +1536,13 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestSpecific) {
}
// Call the div3d_2field function
- div3d_2field<TypeParam>(
- this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->f4din.data(), this->f4dout.data(),
- this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
+ div3d_2field<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->f4din.data(),
+ this->f4dout.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
// Check first field (same as in div3d test)
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
@@ -1590,7 +1591,8 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestRandom) {
// Set random edge indices
for (int j = 0; j < 3; ++j) {
this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
}
// Random geometric factors
@@ -1606,14 +1608,13 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestRandom) {
}
// Call the div3d_2field function
- div3d_2field<TypeParam>(
- this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->f4din.data(), this->f4dout.data(),
- this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
+ div3d_2field<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->f4din.data(),
+ this->f4dout.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
// Calculate reference values separately and verify results
std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
@@ -1622,32 +1623,38 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestRandom) {
for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
// Calculate reference value for first field
ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
this->geofac_div[geofac_div_at(jc, 2, jb)];
// Calculate reference value for second field
ref_f4dout[f4dout_at(jc, jk, jb, 0)] =
this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)], 0)] *
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)],
+ 0)] *
this->geofac_div[geofac_div_at(jc, 0, jb)] +
this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)], 0)] *
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)],
+ 0)] *
this->geofac_div[geofac_div_at(jc, 1, jb)] +
this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)], 0)] *
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)],
+ 0)] *
this->geofac_div[geofac_div_at(jc, 2, jb)];
}
}
@@ -1703,12 +1710,12 @@ TYPED_TEST(HorizontalDiv4DTest, TestSpecific) {
}
// Test function
- div4d<TypeParam>(
- this->cell_edge_idx.data(), this->cell_edge_blk.data(),
- this->geofac_div.data(), this->f4din.data(), this->f4dout.data(),
- this->dim4d, this->i_startblk, this->i_endblk, this->i_startidx_in,
- this->i_endidx_in, this->slev.data(), this->elev.data(), this->nproma,
- this->lacc, this->nlev, this->nblks_c, this->nblks_e);
+ div4d<TypeParam>(this->cell_edge_idx.data(), this->cell_edge_blk.data(),
+ this->geofac_div.data(), this->f4din.data(),
+ this->f4dout.data(), this->dim4d, this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev.data(), this->elev.data(), this->nproma,
+ this->lacc, this->nlev, this->nblks_c, this->nblks_e);
EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 1.4, 1e-6);
EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 1.1, 1e-6);
@@ -1722,7 +1729,6 @@ TYPED_TEST(HorizontalDiv4DTest, TestSpecific) {
EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 1)], 2.6, 1e-6);
EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 1)], 2.3, 1e-6);
EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 1)], 2.3, 1e-6);
-
}
TYPED_TEST(HorizontalDiv4DTest, TestDiv4dRandom) {
@@ -1759,29 +1765,29 @@ TYPED_TEST(HorizontalDiv4DTest, TestDiv4dRandom) {
}
// Test function
- div4d<TypeParam>(
- this->cell_edge_idx.data(), this->cell_edge_blk.data(),
- this->geofac_div.data(), this->f4din.data(), this->f4dout.data(),
- this->dim4d, this->i_startblk, this->i_endblk, this->i_startidx_in,
- this->i_endidx_in, this->slev.data(), this->elev.data(), this->nproma,
- this->lacc, this->nlev, this->nblks_c, this->nblks_e);
+ div4d<TypeParam>(this->cell_edge_idx.data(), this->cell_edge_blk.data(),
+ this->geofac_div.data(), this->f4din.data(),
+ this->f4dout.data(), this->dim4d, this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev.data(), this->elev.data(), this->nproma,
+ this->lacc, this->nlev, this->nblks_c, this->nblks_e);
// Compute reference result and check
for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
for (int ji = 0; ji < dim4d; ++ji) {
for (int jk = this->slev[ji]; jk < this->elev[ji]; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
TypeParam expected = 0.0;
for (int je = 0; je < 3; ++je) {
- expected += this->f4din[f4din_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, je)],
- jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, je)],
- ji)] * this->geofac_div[geofac_div_at(jc, je, jb)];
+ expected +=
+ this->f4din[f4din_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, je)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, je)], ji)] *
+ this->geofac_div[geofac_div_at(jc, je, jb)];
}
EXPECT_NEAR(this->f4dout[f4dout_at(jc, jk, jb, ji)], expected, 1e-5)
@@ -1826,7 +1832,8 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecific) {
for (int i = 0; i < nproma; ++i) {
for (int k = 0; k < nlev; ++k) {
this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- this->opt_in2[vec_e_at(i, k, 0)] = (i + 1) * (k + 1) * 0.5; // Half of vec_e
+ this->opt_in2[vec_e_at(i, k, 0)] =
+ (i + 1) * (k + 1) * 0.5; // Half of vec_e
}
// Set edge indices to point to specific cells
@@ -1867,12 +1874,11 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecific) {
div_avg<TypeParam>(
this->vec_e.data(), this->cell_neighbor_idx.data(),
this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
- this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
- i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
- this->elev[0], this->nproma, patch_id, l_limited_area,
- l2fields, this->lacc, this->nlev,
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
this->nblks_c, this->nblks_e);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
@@ -1888,7 +1894,6 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecific) {
EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 2.04, 1e-6);
EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 1.04, 1e-6);
EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 2.08, 1e-6);
-
}
TYPED_TEST(HorizontalDivAvgTest, TestRandom) {
@@ -1932,141 +1937,158 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandom) {
// Set random edge indices
for (int j = 0; j < 3; ++j) {
this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 3; ++j) {
- this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Random average coefficients
- for (int j = 0; j < 4; ++j) {
- this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Random initial values for div_vec_c and opt_out2
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- }
- }
-
- // Call the div_avg function
- div_avg<TypeParam>(
- this->vec_e.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
- this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
- i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
- this->elev[0], this->nproma, patch_id, l_limited_area,
- l2fields, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- // Calculate reference values manually
- std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> aux_c2(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> ref_opt_out2(dim_combine(nproma, nlev, nblks_c));
-
- // Step 1: Calculate aux_c and aux_c2
- for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
- i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- aux_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
-
- aux_c2[div_vec_c_at(jc, jk, jb)] =
- this->opt_in2[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->opt_in2[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->opt_in2[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
- }
- }
- }
-
- // Step 2: Assign aux_c to div_vec_c and aux_c2 to opt_out2 for patch_id > 0
- for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
- i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] = aux_c[div_vec_c_at(jc, jk, jb)];
- ref_opt_out2[div_vec_c_at(jc, jk, jb)] = aux_c2[div_vec_c_at(jc, jk, jb)];
- }
- }
- }
-
- // Step 3: Perform averaging for the rest of the blocks
- for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
- i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- aux_c[div_vec_c_at(jc, jk, jb)] * this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
- aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
- this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
- aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
- this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
- aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
- this->avg_coeff[avg_coeff_at(jc, 3, jb)];
-
- ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
- aux_c2[div_vec_c_at(jc, jk, jb)] * this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
- aux_c2[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
- this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
- aux_c2[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
- this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
- aux_c2[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
- this->avg_coeff[avg_coeff_at(jc, 3, jb)];
- }
- }
- }
-
- // Verify results
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
- ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
- << "div_vec_c results differ at i=" << i << ", k=" << k;
-
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(i, k, 0)],
- ref_opt_out2[div_vec_c_at(i, k, 0)], 1e-5)
- << "opt_out2 results differ at i=" << i << ", k=" << k;
- }
- }
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random average coefficients
+ for (int j = 0; j < 4; ++j) {
+ this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random initial values for div_vec_c and opt_out2
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values manually
+ std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> aux_c2(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_opt_out2(dim_combine(nproma, nlev, nblks_c));
+
+ // Step 1: Calculate aux_c and aux_c2
+ for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ aux_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+
+ aux_c2[div_vec_c_at(jc, jk, jb)] =
+ this->opt_in2[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->opt_in2[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->opt_in2[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Step 2: Assign aux_c to div_vec_c and aux_c2 to opt_out2 for patch_id > 0
+ for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)];
+ ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
+ aux_c2[div_vec_c_at(jc, jk, jb)];
+ }
+ }
+ }
+
+ // Step 3: Perform averaging for the rest of the blocks
+ for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)] *
+ this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+
+ ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
+ aux_c2[div_vec_c_at(jc, jk, jb)] *
+ this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c2[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c2[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c2[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "div_vec_c results differ at i=" << i << ", k=" << k;
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(i, k, 0)],
+ ref_opt_out2[div_vec_c_at(i, k, 0)], 1e-5)
+ << "opt_out2 results differ at i=" << i << ", k=" << k;
+ }
+ }
}
TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
@@ -2099,7 +2121,8 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
for (int i = 0; i < nproma; ++i) {
for (int k = 0; k < nlev; ++k) {
this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- this->opt_in2[vec_e_at(i, k, 0)] = (i + 1) * (k + 1) * 0.5; // Half of vec_e
+ this->opt_in2[vec_e_at(i, k, 0)] =
+ (i + 1) * (k + 1) * 0.5; // Half of vec_e
}
// Set edge indices to point to specific cells
@@ -2140,15 +2163,13 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
div_avg<TypeParam>(
this->vec_e.data(), this->cell_neighbor_idx.data(),
this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
- this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
- i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
- this->elev[0], this->nproma, patch_id, l_limited_area,
- l2fields, this->lacc, this->nlev,
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
this->nblks_c, this->nblks_e);
-
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
@@ -2162,7 +2183,6 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 0.0, 1e-6);
EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 0.0, 1e-6);
EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 0.0, 1e-6);
-
}
TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
@@ -2185,7 +2205,7 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
// Parameters for the test
int patch_id = 1;
bool l_limited_area = true;
- bool l2fields = false; // Set to false for this test
+ bool l2fields = false; // Set to false for this test
const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
@@ -2200,16 +2220,19 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
for (int i = 0; i < nproma; ++i) {
for (int k = 0; k < nlev; ++k) {
this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- this->opt_in2[vec_e_at(i, k, 0)] = real_distrib(gen); // Not used but initialize anyway
+ this->opt_in2[vec_e_at(i, k, 0)] =
+ real_distrib(gen); // Not used but initialize anyway
}
// Set random edge indices
for (int j = 0; j < 3; ++j) {
this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
}
// Random geometric factors
@@ -2225,7 +2248,8 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
// Random initial values for div_vec_c and opt_out2
for (int k = 0; k < nlev; ++k) {
this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen); // Not used but initialize anyway
+ this->opt_out2[div_vec_c_at(i, k, 0)] =
+ real_distrib(gen); // Not used but initialize anyway
}
}
@@ -2233,12 +2257,11 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
div_avg<TypeParam>(
this->vec_e.data(), this->cell_neighbor_idx.data(),
this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(), this->avg_coeff.data(),
- this->div_vec_c.data(), this->opt_in2.data(), this->opt_out2.data(),
- i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0],
- this->elev[0], this->nproma, patch_id, l_limited_area,
- l2fields, this->lacc, this->nlev,
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
this->nblks_c, this->nblks_e);
// Calculate reference values manually
@@ -2254,20 +2277,24 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
aux_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
this->geofac_div[geofac_div_at(jc, 2, jb)];
}
}
}
- // Step 2: Assign aux_c to div_vec_c for patch_id > 0 (opt_out2 not updated since l2fields=false)
+ // Step 2: Assign aux_c to div_vec_c for patch_id > 0 (opt_out2 not updated
+ // since l2fields=false)
for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
@@ -2275,12 +2302,14 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] = aux_c[div_vec_c_at(jc, jk, jb)];
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)];
}
}
}
- // Step 3: Perform averaging for the rest of the blocks (only for div_vec_c, not opt_out2)
+ // Step 3: Perform averaging for the rest of the blocks (only for div_vec_c,
+ // not opt_out2)
for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
@@ -2289,21 +2318,26 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- aux_c[div_vec_c_at(jc, jk, jb)] * this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
- aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ aux_c[div_vec_c_at(jc, jk, jb)] *
+ this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
- aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
- aux_c[div_vec_c_at(this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
this->avg_coeff[avg_coeff_at(jc, 3, jb)];
}
}
}
- // Verify results - only check div_vec_c since l2fields=false means opt_out2 isn't updated
+ // Verify results - only check div_vec_c since l2fields=false means opt_out2
+ // isn't updated
for (int i = 0; i < nproma; ++i) {
for (int k = 0; k < nlev; ++k) {
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
@@ -2316,20 +2350,20 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
template <typename ValueType>
class HorizontalRotVertexTest : public ::testing::Test {
protected:
- static constexpr int nproma = 3; // inner loop length
- static constexpr int nlev = 2; // number of vertical levels
- static constexpr int nblks_e = 1; // number of edge blocks
- static constexpr int nblks_v = 1; // number of vertex blocks
- static constexpr int dim4d = 2; // 4th dimension size
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 2; // number of vertical levels
+ static constexpr int nblks_e = 1; // number of edge blocks
+ static constexpr int nblks_v = 1; // number of vertex blocks
+ static constexpr int dim4d = 2; // 4th dimension size
int i_startblk = 0;
- int i_endblk = nblks_v; // Test blocks [0 .. nblks_v-1]
+ int i_endblk = nblks_v; // Test blocks [0 .. nblks_v-1]
int i_startidx_in = 0;
- int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
std::vector<int> slev;
std::vector<int> elev;
- bool lacc = false; // Not using ACC-specific behavior.
- bool acc_async = false; // Not using ACC-specific behavior.
+ bool lacc = false; // Not using ACC-specific behavior.
+ bool acc_async = false; // Not using ACC-specific behavior.
std::vector<ValueType> vec_e;
std::vector<int> vert_edge_idx;
@@ -2339,9 +2373,9 @@ protected:
std::vector<ValueType> f4din;
std::vector<ValueType> f4dout;
- HorizontalRotVertexTest () {
+ HorizontalRotVertexTest() {
slev.resize(dim4d, 0);
- elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
+ elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
vec_e.resize(dim_combine(nproma, nlev, nblks_e));
vert_edge_idx.resize(dim_combine(nproma, nblks_v, 6));
@@ -2354,12 +2388,11 @@ protected:
};
template <typename ValueType>
-class HorizontalRotVertexAtmosTest
- : public HorizontalRotVertexTest<ValueType> {};
+class HorizontalRotVertexAtmosTest : public HorizontalRotVertexTest<ValueType> {
+};
template <typename ValueType>
-class HorizontalRotVertexRITest
- : public HorizontalRotVertexTest<ValueType> {};
+class HorizontalRotVertexRITest : public HorizontalRotVertexTest<ValueType> {};
TYPED_TEST_SUITE(HorizontalRotVertexAtmosTest, ValueTypes);
@@ -2404,10 +2437,9 @@ TYPED_TEST(HorizontalRotVertexAtmosTest, TestSpecific) {
// Call the rot_vertex_atmos function
rot_vertex_atmos<TypeParam>(
this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(),
- this->rot_vec.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->nlev,
this->nblks_e, this->nblks_v);
// Expected values based on the initialization pattern
@@ -2445,7 +2477,8 @@ TYPED_TEST(HorizontalRotVertexAtmosTest, TestRandom) {
// Set random edge indices
for (int j = 0; j < 6; ++j) {
this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
- this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
}
// Random geometric factors
@@ -2462,10 +2495,9 @@ TYPED_TEST(HorizontalRotVertexAtmosTest, TestRandom) {
// Call the rot_vertex_atmos function
rot_vertex_atmos<TypeParam>(
this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(),
- this->rot_vec.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->nlev,
this->nblks_e, this->nblks_v);
// Calculate reference values separately and verify results
@@ -2479,23 +2511,29 @@ TYPED_TEST(HorizontalRotVertexAtmosTest, TestRandom) {
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jv = i_startidx; jv < i_endidx; ++jv) {
ref_rot_vec[rot_vec_at(jv, jk, jb)] =
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
this->geofac_rot[geofac_rot_at(jv, 5, jb)];
}
}
@@ -2554,10 +2592,9 @@ TYPED_TEST(HorizontalRotVertexRITest, TestSpecific) {
// Call the rot_vertex_ri function
rot_vertex_ri<TypeParam>(
this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(),
- this->rot_vec.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->acc_async,
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->acc_async,
this->nlev, this->nblks_e, this->nblks_v);
// Expected values based on the initialization pattern
@@ -2595,7 +2632,8 @@ TYPED_TEST(HorizontalRotVertexRITest, TestRandom) {
// Set random edge indices
for (int j = 0; j < 6; ++j) {
this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
- this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0; // Keep in same block for simplicity
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
}
// Random geometric factors
@@ -2612,10 +2650,9 @@ TYPED_TEST(HorizontalRotVertexRITest, TestRandom) {
// Call the rot_vertex_ri function
rot_vertex_ri<TypeParam>(
this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(),
- this->rot_vec.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->acc_async,
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->acc_async,
this->nlev, this->nblks_e, this->nblks_v);
// Ensure computation is complete for both modes
@@ -2632,23 +2669,29 @@ TYPED_TEST(HorizontalRotVertexRITest, TestRandom) {
for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
for (int jv = i_startidx; jv < i_endidx; ++jv) {
ref_rot_vec[rot_vec_at(jv, jk, jb)] =
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
- this->vec_e[vec_e_at(this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
this->geofac_rot[geofac_rot_at(jv, 5, jb)];
}
}
@@ -2662,5 +2705,4 @@ TYPED_TEST(HorizontalRotVertexRITest, TestRandom) {
<< "Results differ at i=" << i << ", k=" << k << ")";
}
}
-
}
--
GitLab
From bad072426c8bf88f4b3c6de95968d65efe9fc8f9 Mon Sep 17 00:00:00 2001
From: Pradipta Samanta <samanta@dkrz.de>
Date: Mon, 17 Mar 2025 11:22:04 +0100
Subject: [PATCH 31/35] reverted back some changes
---
test/c/test_horizontal_divrot.cpp | 16 ++++------------
1 file changed, 4 insertions(+), 12 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index a97c1da..10725a5 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -832,15 +832,11 @@ TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
std::vector<TypeParam> z_d(lsq_dim_c);
std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- // for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- {
- int jb = 0;
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- // for (int jk = this->slev; jk < this->elev; ++jk) {
- {
- int jk = 0;
+ for (int jk = this->slev; jk < this->elev; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
for (int i = 0; i < lsq_dim_c; ++i) {
z_d[i] = this->p_cc[p_cc_at(
@@ -1242,15 +1238,11 @@ TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- // for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- {
- int jb = 0;
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
int i_startidx, i_endidx;
get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- // for (int jk = this->slev; jk < this->elev; ++jk) {
- {
- int jk = 0;
+ for (int jk = this->slev; jk < this->elev; ++jk) {
for (int jc = i_startidx; jc < i_endidx; ++jc) {
for (int i = 0; i < lsq_dim_c; ++i) {
z_d[i] = this->p_cc[p_cc_at(
--
GitLab
From e03d73d9f3cf6ac16edb5a24164820fad5dec760 Mon Sep 17 00:00:00 2001
From: Yen-Chen <yen-chen.chen@tum.de>
Date: Mon, 17 Mar 2025 13:46:00 +0100
Subject: [PATCH 32/35] Remove redundant code
---
test/c/test_horizontal_divrot.cpp | 4 ----
1 file changed, 4 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 10725a5..0a0aba3 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -386,10 +386,6 @@ TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVD) {
this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
}
- // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
- // this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
- // this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
-
this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
}
--
GitLab
From 4107598a8bdfa336f134c0809e314baebddd8ba9 Mon Sep 17 00:00:00 2001
From: Yen-Chen <yen-chen.chen@tum.de>
Date: Mon, 17 Mar 2025 13:48:34 +0100
Subject: [PATCH 33/35] Rename lib_divrot to mo_lib_divrot
---
src/horizontal/CMakeLists.txt | 2 +-
src/horizontal/{lib_divrot.cpp => mo_lib_divrot.cpp} | 2 +-
src/horizontal/{lib_divrot.hpp => mo_lib_divrot.hpp} | 0
test/c/test_horizontal_divrot.cpp | 2 +-
4 files changed, 3 insertions(+), 3 deletions(-)
rename src/horizontal/{lib_divrot.cpp => mo_lib_divrot.cpp} (99%)
rename src/horizontal/{lib_divrot.hpp => mo_lib_divrot.hpp} (100%)
diff --git a/src/horizontal/CMakeLists.txt b/src/horizontal/CMakeLists.txt
index d403cb2..f3b75c0 100644
--- a/src/horizontal/CMakeLists.txt
+++ b/src/horizontal/CMakeLists.txt
@@ -11,7 +11,7 @@
add_library(
iconmath-horizontal
- lib_divrot.cpp
+ mo_lib_divrot.cpp
mo_lib_divrot.F90
mo_lib_laplace.F90
mo_lib_gradients.F90)
diff --git a/src/horizontal/lib_divrot.cpp b/src/horizontal/mo_lib_divrot.cpp
similarity index 99%
rename from src/horizontal/lib_divrot.cpp
rename to src/horizontal/mo_lib_divrot.cpp
index a24981d..d086e8b 100644
--- a/src/horizontal/lib_divrot.cpp
+++ b/src/horizontal/mo_lib_divrot.cpp
@@ -12,7 +12,7 @@
#include <iostream>
#include <vector>
-#include <horizontal/lib_divrot.hpp>
+#include <horizontal/mo_lib_divrot.hpp>
#include <support/mo_lib_loopindices.hpp>
template <typename T>
diff --git a/src/horizontal/lib_divrot.hpp b/src/horizontal/mo_lib_divrot.hpp
similarity index 100%
rename from src/horizontal/lib_divrot.hpp
rename to src/horizontal/mo_lib_divrot.hpp
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 0a0aba3..6a6e458 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -15,7 +15,7 @@
#include <Kokkos_Core.hpp>
#include <gtest/gtest.h>
-#include <horizontal/lib_divrot.hpp>
+#include <horizontal/mo_lib_divrot.hpp>
#include <support/mo_lib_loopindices.hpp>
// Template function for computing array size.
--
GitLab
From 29f23212c9fc1983b96899bec2277cf67e94b1a0 Mon Sep 17 00:00:00 2001
From: Yen-Chen <yen-chen.chen@tum.de>
Date: Mon, 17 Mar 2025 13:53:42 +0100
Subject: [PATCH 34/35] Combine Div tests
---
test/c/test_horizontal_divrot.cpp | 40 ++++++++++---------------------
1 file changed, 12 insertions(+), 28 deletions(-)
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
index 6a6e458..f3cb3c8 100644
--- a/test/c/test_horizontal_divrot.cpp
+++ b/test/c/test_horizontal_divrot.cpp
@@ -1323,21 +1323,9 @@ protected:
}
};
-template <typename ValueType>
-class HorizontalDiv3DTest : public HorizontalDivTest<ValueType> {};
-
-template <typename ValueType>
-class HorizontalDiv3D2FTest : public HorizontalDivTest<ValueType> {};
-
-template <typename ValueType>
-class HorizontalDiv4DTest : public HorizontalDivTest<ValueType> {};
+TYPED_TEST_SUITE(HorizontalDivTest, ValueTypes);
-template <typename ValueType>
-class HorizontalDivAvgTest : public HorizontalDivTest<ValueType> {};
-
-TYPED_TEST_SUITE(HorizontalDiv3DTest, ValueTypes);
-
-TYPED_TEST(HorizontalDiv3DTest, TestSpecific) {
+TYPED_TEST(HorizontalDivTest, TestDiv3DSpecific) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1391,7 +1379,7 @@ TYPED_TEST(HorizontalDiv3DTest, TestSpecific) {
EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
}
-TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
+TYPED_TEST(HorizontalDivTest, TestDiv3DRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1477,9 +1465,7 @@ TYPED_TEST(HorizontalDiv3DTest, TestRandom) {
}
}
-TYPED_TEST_SUITE(HorizontalDiv3D2FTest, ValueTypes);
-
-TYPED_TEST(HorizontalDiv3D2FTest, TestSpecific) {
+TYPED_TEST(HorizontalDivTest, TestDiv3D2FSpecific) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1549,7 +1535,7 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestSpecific) {
EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 6.6, 1e-6);
}
-TYPED_TEST(HorizontalDiv3D2FTest, TestRandom) {
+TYPED_TEST(HorizontalDivTest, TestDiv3D2FRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1667,9 +1653,7 @@ TYPED_TEST(HorizontalDiv3D2FTest, TestRandom) {
}
}
-TYPED_TEST_SUITE(HorizontalDiv4DTest, ValueTypes);
-
-TYPED_TEST(HorizontalDiv4DTest, TestSpecific) {
+TYPED_TEST(HorizontalDivTest, TestDiv4DSpecific) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1719,7 +1703,7 @@ TYPED_TEST(HorizontalDiv4DTest, TestSpecific) {
EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 1)], 2.3, 1e-6);
}
-TYPED_TEST(HorizontalDiv4DTest, TestDiv4dRandom) {
+TYPED_TEST(HorizontalDivTest, TestDiv4DRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1787,9 +1771,9 @@ TYPED_TEST(HorizontalDiv4DTest, TestDiv4dRandom) {
}
}
-TYPED_TEST_SUITE(HorizontalDivAvgTest, ValueTypes);
+TYPED_TEST_SUITE(HorizontalDivTest, ValueTypes);
-TYPED_TEST(HorizontalDivAvgTest, TestSpecific) {
+TYPED_TEST(HorizontalDivTest, TestDivAvgSpecific) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -1884,7 +1868,7 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecific) {
EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 2.08, 1e-6);
}
-TYPED_TEST(HorizontalDivAvgTest, TestRandom) {
+TYPED_TEST(HorizontalDivTest, TestDivAvgRandom) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -2079,7 +2063,7 @@ TYPED_TEST(HorizontalDivAvgTest, TestRandom) {
}
}
-TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
+TYPED_TEST(HorizontalDivTest, TestDivAvgSpecificNoL2fields) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
@@ -2173,7 +2157,7 @@ TYPED_TEST(HorizontalDivAvgTest, TestSpecificNoL2fields) {
EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 0.0, 1e-6);
}
-TYPED_TEST(HorizontalDivAvgTest, TestRandomNoL2fields) {
+TYPED_TEST(HorizontalDivTest, TestDivAvgRandomNoL2fields) {
constexpr int nproma = this->nproma;
constexpr int nlev = this->nlev;
constexpr int nblks_c = this->nblks_c;
--
GitLab
From 4503701926af6bec4590d0b3f41879f8181d9b9d Mon Sep 17 00:00:00 2001
From: Yen-Chen <yen-chen.chen@tum.de>
Date: Mon, 17 Mar 2025 14:07:16 +0100
Subject: [PATCH 35/35] Split tests into three files
---
test/c/CMakeLists.txt | 6 +-
test/c/dim_helper.hpp | 88 +
test/c/test_horizontal_div.cpp | 1070 ++++++++++++
test/c/test_horizontal_divrot.cpp | 2680 -----------------------------
test/c/test_horizontal_recon.cpp | 1199 +++++++++++++
test/c/test_horizontal_rot.cpp | 378 ++++
6 files changed, 2740 insertions(+), 2681 deletions(-)
create mode 100644 test/c/dim_helper.hpp
create mode 100644 test/c/test_horizontal_div.cpp
delete mode 100644 test/c/test_horizontal_divrot.cpp
create mode 100644 test/c/test_horizontal_recon.cpp
create mode 100644 test/c/test_horizontal_rot.cpp
diff --git a/test/c/CMakeLists.txt b/test/c/CMakeLists.txt
index 18c2710..98a21b2 100644
--- a/test/c/CMakeLists.txt
+++ b/test/c/CMakeLists.txt
@@ -27,7 +27,9 @@ endif()
set(SOURCES
main.cpp
- test_horizontal_divrot.cpp
+ test_horizontal_div.cpp
+ test_horizontal_recon.cpp
+ test_horizontal_rot.cpp
test_tdma_solver.cpp
test_interpolation_vector.cpp
test_intp_rbf.cpp
@@ -36,6 +38,8 @@ set(SOURCES
# Create the test executable from your test files, including main.cpp.
add_executable(iconmath_test_c ${SOURCES})
+target_include_directories(iconmath_test_c PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
+
# Link the test executable with GoogleTest and Kokkos.
target_link_libraries(iconmath_test_c
PUBLIC
diff --git a/test/c/dim_helper.hpp b/test/c/dim_helper.hpp
new file mode 100644
index 0000000..165d5d9
--- /dev/null
+++ b/test/c/dim_helper.hpp
@@ -0,0 +1,88 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#pragma once
+
+// Template function for computing array size.
+// For example, we get the array size of a 4-dimensional array A(2, 3, 4, 5) by
+// dim_combine(2, 3, 4, 5).
+// Which will automatically instantiate
+// dim_combine<int, int, int, int>(2, 3, 4, 5).
+// The function then call dim_combine recursively
+// dim_combine<int, int, int, int>(2, 3, 4, 5) {
+// return static_cast<size_t>(2) * dim_combine<int, int, int>(3, 4, 5);
+// }
+// dim_combine<int, int, int>(3, 4, 5) {
+// return static_cast<size_t>(3) * dim_combine<int, int>(4, 5);
+// }
+// dim_combine<int, int>(4, 5) {
+// return static_cast<size_t>(4) * dim_combine<int>(5);
+// }
+// Where the last dim_combine is specialized as
+// dim_combine<int>(5) {
+// return static_cast<size_t>(5);
+// }
+// Which gives
+// dim_combine<int, int, int, int>(2, 3, 4, 5) =
+// static_cast<size_t>(2) * static_cast<size_t>(3) *
+// static_cast<size_t>(4) * static_cast<size_t>(5)
+/// Template helpers for combining multiple dimension array sizes.
+/// The base function of dimension combine. Should not be used.
+template <typename... Ts> size_t dim_combine(Ts... dims) { return 0; }
+/// Template specialization of only one dimension, returns the dimension itself.
+template <typename T> size_t dim_combine(T dim) {
+ return static_cast<size_t>(dim);
+}
+/// Template specialization of picking out the first dimension. The combined
+/// dimension is the first dimension times the combined dimension of the rest.
+template <typename T, typename... Ts> size_t dim_combine(T dim, Ts... dims) {
+ return static_cast<size_t>(dim) * dim_combine(dims...);
+}
+
+// Template function for LayoutLeft ID access in compile time.
+// For example, a multi-dimensional array A of dimensions <2, 3, 4, 5> gets its
+// corresponding vector id (LayoutLeft) by
+// at<2, 3, 4, 5>(id1, id2, id3, id4).
+// The at_impl then adds the id from beginning to the end and pass the id prefix
+// to the next recursive at_impl function. In this example,
+// at<2, 3, 4, 5>(id1, id2, id3, id4) {
+// return id1 + at_impl<3, 4, 5>(2, id2, id3, id4);
+// }
+// at_impl<3, 4, 5>(2, id2, id3, id4) {
+// return id2 * 2 + at_impl<4, 5>(2 * 3, id3, id4);
+// }
+// at_impl<4, 5>(2 * 3, id3, id4) {
+// return id3 * 2 * 3 + at_impl<5>(2 * 3 * 4, id4);
+// }
+// at_impl<5>(2 * 3 * 4, id4) {
+// return id4 * 2 * 3 * 4;
+// }
+// Which gives
+// at<2, 3, 4, 5>(id1, id2, id3, id4) = id1 + id2 * 2 +
+// id3 * 2 * 3 + id4 * 2 * 3 * 4
+/// Helper type converting integer numbers to int
+template <class T, auto> using always_t = T;
+/// Base function of at_impl. Should not be used.
+template <int... Dims> int at_impl(always_t<int, Dims>... ids) { return 0; }
+/// Template specialization of the last ID
+template <int LastDim> int at_impl(int prefix, int id) { return id * prefix; }
+/// Template specialization of at_impl, accumulate the return value using the
+/// first id and pass the prefix to the next recursive at_impl function.
+template <int FirstDim, int... Dims>
+int at_impl(int prefix, int id, always_t<int, Dims>... ids) {
+ return id * prefix + at_impl<Dims...>(prefix * FirstDim, ids...);
+}
+/// at<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
+/// LayoutLeft. Use this function instead of at_impl.
+template <int FirstDim, int... Dims>
+int at(int id, always_t<int, Dims>... ids) {
+ return id + at_impl<Dims...>(FirstDim, ids...);
+}
diff --git a/test/c/test_horizontal_div.cpp b/test/c/test_horizontal_div.cpp
new file mode 100644
index 0000000..596d19e
--- /dev/null
+++ b/test/c/test_horizontal_div.cpp
@@ -0,0 +1,1070 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#include <iostream>
+#include <random>
+#include <vector>
+
+#include <Kokkos_Core.hpp>
+#include <gtest/gtest.h>
+#include <dim_helper.hpp>
+#include <horizontal/mo_lib_divrot.hpp>
+#include <support/mo_lib_loopindices.hpp>
+
+/// Test class for the horizontal divergence tests. Templated for the ValueType
+template <typename ValueType> class HorizontalDivTest : public ::testing::Test {
+protected:
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 2; // number of vertical levels
+ static constexpr int nblks_c = 1; // number of cell blocks
+ static constexpr int nblks_e = 1; // number of edge blocks
+ static constexpr int dim4d = 2; // 4th dimension size
+
+ int i_startblk = 0;
+ int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
+ int i_startidx_in = 0;
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ std::vector<int> slev;
+ std::vector<int> elev;
+ bool lacc = false; // Not using ACC-specific behavior.
+
+ std::vector<ValueType> vec_e;
+ std::vector<int> cell_edge_idx;
+ std::vector<int> cell_edge_blk;
+ std::vector<ValueType> geofac_div;
+ std::vector<ValueType> div_vec_c;
+ std::vector<ValueType> f4din;
+ std::vector<ValueType> f4dout;
+
+ // Followings are needed in HorizontalDivAvgTest
+ std::vector<int> cell_neighbor_idx;
+ std::vector<int> cell_neighbor_blk;
+ std::vector<ValueType> avg_coeff;
+ std::vector<ValueType> opt_in2;
+ std::vector<ValueType> opt_out2;
+
+ HorizontalDivTest() {
+ slev.resize(dim4d, 0);
+ elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
+
+ vec_e.resize(dim_combine(nproma, nlev, nblks_e));
+ cell_edge_idx.resize(dim_combine(nproma, nblks_c, 3));
+ cell_edge_blk.resize(dim_combine(nproma, nblks_c, 3));
+ geofac_div.resize(dim_combine(nproma, 3, nblks_c));
+ div_vec_c.resize(dim_combine(nproma, nlev, nblks_c));
+ f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
+ f4dout.resize(dim_combine(nproma, nlev, nblks_c, dim4d));
+ cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, 3));
+ cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, 3));
+ avg_coeff.resize(dim_combine(nproma, 4, nblks_c));
+ opt_in2.resize(dim_combine(nproma, nlev, nblks_e));
+ opt_out2.resize(dim_combine(nproma, nlev, nblks_c));
+ }
+};
+
+/// ValueTypes which the divrot tests should run with
+typedef ::testing::Types<float, double> ValueTypes;
+
+TYPED_TEST_SUITE(HorizontalDivTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivTest, TestDiv3DSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ }
+
+ // Set edge indices to point to specific cells (including self)
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Initialize div_vec_c to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the div3d function
+ div3d<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivTest, TestDiv3DRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize div_vec_c to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div3d function
+ div3d<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "Results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalDivTest, TestDiv3D2FSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+ const auto &f4d_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ this->f4din[f4d_at(i, k, 0, 0)] =
+ (i + 1) * (k + 2); // Different pattern for second field
+ }
+
+ // Set edge indices to point to specific cells (including self)
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Initialize div_vec_c and f4dout to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ this->f4dout[f4dout_at(i, k, 0, 0)] = 0.0;
+ }
+ }
+
+ // Call the div3d_2field function
+ div3d_2field<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->f4din.data(),
+ this->f4dout.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Check first field (same as in div3d test)
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
+
+ // Check second field (expected values calculated manually)
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 0)], 5.1, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 0)], 6.3, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 0)], 4.4, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 6.6, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivTest, TestDiv3D2FRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+ const auto &f4d_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ this->f4din[f4d_at(i, k, 0, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize div_vec_c and f4dout to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->f4dout[f4dout_at(i, k, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div3d_2field function
+ div3d_2field<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->div_vec_c.data(), this->f4din.data(),
+ this->f4dout.data(), this->i_startblk, this->i_endblk,
+ this->i_startidx_in, this->i_endidx_in, this->slev[0],
+ this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
+ std::vector<TypeParam> ref_f4dout(nproma * nlev * nblks_c * dim4d, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ // Calculate reference value for first field
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+
+ // Calculate reference value for second field
+ ref_f4dout[f4dout_at(jc, jk, jb, 0)] =
+ this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)],
+ 0)] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)],
+ 0)] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)],
+ 0)] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Verify results for first field
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "First field results differ at i=" << i << ", k=" << k;
+ }
+ }
+
+ // Verify results for second field
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->f4dout[f4dout_at(i, k, 0, 0)],
+ ref_f4dout[f4dout_at(i, k, 0, 0)], 1e-5)
+ << "Second field results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalDivTest, TestDiv4DSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &f4din_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = (i + j) % nproma;
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->geofac_div[geofac_div_at(i, j, 0)] = 0.1 * (j + 1);
+ }
+
+ for (int k = 0; k < nlev; ++k) {
+ for (int d = 0; d < dim4d; ++d) {
+ this->f4din[f4din_at(i, k, 0, d)] = 1.0 + i + k + d;
+ this->f4dout[f4dout_at(i, k, 0, d)] = 0.0;
+ }
+ }
+ }
+
+ // Test function
+ div4d<TypeParam>(this->cell_edge_idx.data(), this->cell_edge_blk.data(),
+ this->geofac_div.data(), this->f4din.data(),
+ this->f4dout.data(), this->dim4d, this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev.data(), this->elev.data(), this->nproma,
+ this->lacc, this->nlev, this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 1.4, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 1.1, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 0)], 1.1, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 0)], 2.0, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 1)], 2.0, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 1)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 1)], 1.7, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 1)], 2.6, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 1)], 2.3, 1e-6);
+ EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 1)], 2.3, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivTest, TestDiv4DRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &f4din_at = at<nproma, nlev, nblks_e, dim4d>;
+ const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
+
+ // Initialize with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ for (int k = 0; k < nlev; ++k) {
+ for (int d = 0; d < dim4d; ++d) {
+ this->f4din[f4din_at(i, k, 0, d)] = real_distrib(gen);
+ this->f4dout[f4dout_at(i, k, 0, d)] = 0.0;
+ }
+ }
+ }
+
+ // Test function
+ div4d<TypeParam>(this->cell_edge_idx.data(), this->cell_edge_blk.data(),
+ this->geofac_div.data(), this->f4din.data(),
+ this->f4dout.data(), this->dim4d, this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev.data(), this->elev.data(), this->nproma,
+ this->lacc, this->nlev, this->nblks_c, this->nblks_e);
+
+ // Compute reference result and check
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int ji = 0; ji < dim4d; ++ji) {
+ for (int jk = this->slev[ji]; jk < this->elev[ji]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ TypeParam expected = 0.0;
+ for (int je = 0; je < 3; ++je) {
+ expected +=
+ this->f4din[f4din_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, je)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, je)], ji)] *
+ this->geofac_div[geofac_div_at(jc, je, jb)];
+ }
+
+ EXPECT_NEAR(this->f4dout[f4dout_at(jc, jk, jb, ji)], expected, 1e-5)
+ << "Random test fails at jc=" << jc << ", jk=" << jk
+ << ", jb=" << jb << ", ji=" << ji;
+ }
+ }
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalDivTest, ValueTypes);
+
+TYPED_TEST(HorizontalDivTest, TestDivAvgSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for additional parameters
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = true;
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Initialize the vectors with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ this->opt_in2[vec_e_at(i, k, 0)] =
+ (i + 1) * (k + 1) * 0.5; // Half of vec_e
+ }
+
+ // Set edge indices to point to specific cells
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // Set neighbor indices similarly
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges and neighbors are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Average coefficients
+ this->avg_coeff[avg_coeff_at(i, 0, 0)] = 0.4; // Self
+ this->avg_coeff[avg_coeff_at(i, 1, 0)] = 0.2; // First neighbor
+ this->avg_coeff[avg_coeff_at(i, 2, 0)] = 0.2; // Second neighbor
+ this->avg_coeff[avg_coeff_at(i, 3, 0)] = 0.2; // Third neighbor
+
+ // Initialize div_vec_c and opt_out2 to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ this->opt_out2[div_vec_c_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.16, 1e-6);
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 0, 0)], 0.94, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 1, 0)], 1.88, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 0, 0)], 1.02, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 2.04, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 1.04, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 2.08, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivTest, TestDivAvgRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = true;
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialize with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ this->opt_in2[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random average coefficients
+ for (int j = 0; j < 4; ++j) {
+ this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random initial values for div_vec_c and opt_out2
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values manually
+ std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> aux_c2(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_opt_out2(dim_combine(nproma, nlev, nblks_c));
+
+ // Step 1: Calculate aux_c and aux_c2
+ for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ aux_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+
+ aux_c2[div_vec_c_at(jc, jk, jb)] =
+ this->opt_in2[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->opt_in2[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->opt_in2[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Step 2: Assign aux_c to div_vec_c and aux_c2 to opt_out2 for patch_id > 0
+ for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)];
+ ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
+ aux_c2[div_vec_c_at(jc, jk, jb)];
+ }
+ }
+ }
+
+ // Step 3: Perform averaging for the rest of the blocks
+ for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)] *
+ this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+
+ ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
+ aux_c2[div_vec_c_at(jc, jk, jb)] *
+ this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c2[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c2[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c2[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "div_vec_c results differ at i=" << i << ", k=" << k;
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(i, k, 0)],
+ ref_opt_out2[div_vec_c_at(i, k, 0)], 1e-5)
+ << "opt_out2 results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalDivTest, TestDivAvgSpecificNoL2fields) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int dim4d = this->dim4d;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = false;
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Initialize the vectors with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ this->opt_in2[vec_e_at(i, k, 0)] =
+ (i + 1) * (k + 1) * 0.5; // Half of vec_e
+ }
+
+ // Set edge indices to point to specific cells
+ this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
+ this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // Set neighbor indices similarly
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = (i + 2) % nproma;
+
+ // All edges and neighbors are in the same block for this test
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors
+ this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
+ this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
+ this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
+
+ // Average coefficients
+ this->avg_coeff[avg_coeff_at(i, 0, 0)] = 0.4; // Self
+ this->avg_coeff[avg_coeff_at(i, 1, 0)] = 0.2; // First neighbor
+ this->avg_coeff[avg_coeff_at(i, 2, 0)] = 0.2; // Second neighbor
+ this->avg_coeff[avg_coeff_at(i, 3, 0)] = 0.2; // Third neighbor
+
+ // Initialize div_vec_c and opt_out2 to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
+ this->opt_out2[div_vec_c_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the div_avg function
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.08, 1e-6);
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.16, 1e-6);
+
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 0, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 1, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 0, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 0.0, 1e-6);
+ EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 0.0, 1e-6);
+}
+
+TYPED_TEST(HorizontalDivTest, TestDivAvgRandomNoL2fields) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int nblks_e = this->nblks_e;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &cell_edge_at = at<nproma, nblks_c, 3>;
+ const auto &geofac_div_at = at<nproma, 3, nblks_c>;
+ const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
+
+ // Vectors for block and index ranges
+ std::vector<int> i_startblk_in(3, 0);
+ std::vector<int> i_endblk_in(3, nblks_c);
+ std::vector<int> i_startidx_in(3, 0);
+ std::vector<int> i_endidx_in(3, nproma);
+
+ // Parameters for the test
+ int patch_id = 1;
+ bool l_limited_area = true;
+ bool l2fields = false; // Set to false for this test
+
+ const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
+ const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialize with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ this->opt_in2[vec_e_at(i, k, 0)] =
+ real_distrib(gen); // Not used but initialize anyway
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 3; ++j) {
+ this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
+ this->cell_edge_blk[cell_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 3; ++j) {
+ this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random average coefficients
+ for (int j = 0; j < 4; ++j) {
+ this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Random initial values for div_vec_c and opt_out2
+ for (int k = 0; k < nlev; ++k) {
+ this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
+ this->opt_out2[div_vec_c_at(i, k, 0)] =
+ real_distrib(gen); // Not used but initialize anyway
+ }
+ }
+
+ // Call the div_avg function with l2fields=false
+ div_avg<TypeParam>(
+ this->vec_e.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
+ this->cell_edge_blk.data(), this->geofac_div.data(),
+ this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
+ this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
+ i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
+ this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
+ this->nblks_c, this->nblks_e);
+
+ // Calculate reference values manually
+ std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
+ std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
+
+ // Step 1: Calculate aux_c (but not aux_c2 since l2fields=false)
+ for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
+ i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ aux_c[div_vec_c_at(jc, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
+ this->geofac_div[geofac_div_at(jc, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
+ this->geofac_div[geofac_div_at(jc, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
+ this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
+ this->geofac_div[geofac_div_at(jc, 2, jb)];
+ }
+ }
+ }
+
+ // Step 2: Assign aux_c to div_vec_c for patch_id > 0 (opt_out2 not updated
+ // since l2fields=false)
+ for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
+ i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)];
+ }
+ }
+ }
+
+ // Step 3: Perform averaging for the rest of the blocks (only for div_vec_c,
+ // not opt_out2)
+ for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
+ i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
+ aux_c[div_vec_c_at(jc, jk, jb)] *
+ this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
+ this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
+ this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
+ aux_c[div_vec_c_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
+ this->avg_coeff[avg_coeff_at(jc, 3, jb)];
+ }
+ }
+ }
+
+ // Verify results - only check div_vec_c since l2fields=false means opt_out2
+ // isn't updated
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
+ ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
+ << "div_vec_c results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
diff --git a/test/c/test_horizontal_divrot.cpp b/test/c/test_horizontal_divrot.cpp
deleted file mode 100644
index f3cb3c8..0000000
--- a/test/c/test_horizontal_divrot.cpp
+++ /dev/null
@@ -1,2680 +0,0 @@
-// ICON
-//
-// ---------------------------------------------------------------
-// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
-// Contact information: icon-model.org
-//
-// See AUTHORS.TXT for a list of authors
-// See LICENSES/ for license information
-// SPDX-License-Identifier: BSD-3-Clause
-// ---------------------------------------------------------------
-
-#include <iostream>
-#include <random>
-#include <vector>
-
-#include <Kokkos_Core.hpp>
-#include <gtest/gtest.h>
-#include <horizontal/mo_lib_divrot.hpp>
-#include <support/mo_lib_loopindices.hpp>
-
-// Template function for computing array size.
-// For example, we get the array size of a 4-dimensional array A(2, 3, 4, 5) by
-// dim_combine(2, 3, 4, 5).
-// Which will automatically instantiate
-// dim_combine<int, int, int, int>(2, 3, 4, 5).
-// The function then call dim_combine recursively
-// dim_combine<int, int, int, int>(2, 3, 4, 5) {
-// return static_cast<size_t>(2) * dim_combine<int, int, int>(3, 4, 5);
-// }
-// dim_combine<int, int, int>(3, 4, 5) {
-// return static_cast<size_t>(3) * dim_combine<int, int>(4, 5);
-// }
-// dim_combine<int, int>(4, 5) {
-// return static_cast<size_t>(4) * dim_combine<int>(5);
-// }
-// Where the last dim_combine is specialized as
-// dim_combine<int>(5) {
-// return static_cast<size_t>(5);
-// }
-// Which gives
-// dim_combine<int, int, int, int>(2, 3, 4, 5) =
-// static_cast<size_t>(2) * static_cast<size_t>(3) *
-// static_cast<size_t>(4) * static_cast<size_t>(5)
-/// Template helpers for combining multiple dimension array sizes.
-/// The base function of dimension combine. Should not be used.
-template <typename... Ts> size_t dim_combine(Ts... dims) { return 0; }
-/// Template specialization of only one dimension, returns the dimension itself.
-template <typename T> size_t dim_combine(T dim) {
- return static_cast<size_t>(dim);
-}
-/// Template specialization of picking out the first dimension. The combined
-/// dimension is the first dimension times the combined dimension of the rest.
-template <typename T, typename... Ts> size_t dim_combine(T dim, Ts... dims) {
- return static_cast<size_t>(dim) * dim_combine(dims...);
-}
-
-// Template function for LayoutLeft ID access in compile time.
-// For example, a multi-dimensional array A of dimensions <2, 3, 4, 5> gets its
-// corresponding vector id (LayoutLeft) by
-// at<2, 3, 4, 5>(id1, id2, id3, id4).
-// The at_impl then adds the id from beginning to the end and pass the id prefix
-// to the next recursive at_impl function. In this example,
-// at<2, 3, 4, 5>(id1, id2, id3, id4) {
-// return id1 + at_impl<3, 4, 5>(2, id2, id3, id4);
-// }
-// at_impl<3, 4, 5>(2, id2, id3, id4) {
-// return id2 * 2 + at_impl<4, 5>(2 * 3, id3, id4);
-// }
-// at_impl<4, 5>(2 * 3, id3, id4) {
-// return id3 * 2 * 3 + at_impl<5>(2 * 3 * 4, id4);
-// }
-// at_impl<5>(2 * 3 * 4, id4) {
-// return id4 * 2 * 3 * 4;
-// }
-// Which gives
-// at<2, 3, 4, 5>(id1, id2, id3, id4) = id1 + id2 * 2 +
-// id3 * 2 * 3 + id4 * 2 * 3 * 4
-/// Helper type converting integer numbers to int
-template <class T, auto> using always_t = T;
-/// Base function of at_impl. Should not be used.
-template <int... Dims> int at_impl(always_t<int, Dims>... ids) { return 0; }
-/// Template specialization of the last ID
-template <int LastDim> int at_impl(int prefix, int id) { return id * prefix; }
-/// Template specialization of at_impl, accumulate the return value using the
-/// first id and pass the prefix to the next recursive at_impl function.
-template <int FirstDim, int... Dims>
-int at_impl(int prefix, int id, always_t<int, Dims>... ids) {
- return id * prefix + at_impl<Dims...>(prefix * FirstDim, ids...);
-}
-/// at<dim1, dim2, ...>(id1, id2, ...) gets its memory index in vector assuming
-/// LayoutLeft. Use this function instead of at_impl.
-template <int FirstDim, int... Dims>
-int at(int id, always_t<int, Dims>... ids) {
- return id + at_impl<Dims...>(FirstDim, ids...);
-}
-
-/// Enum class for the reconstruction method
-enum class ReconstructionMethod {
- linear,
- quadratic,
- cubic,
-};
-
-/// Base test class for the horizontal divrot tests. Templated for the ValueType
-/// and ReconMethod for the reconstruction method.
-template <typename ValueType, int ReconMethod>
-class HorizontalReconTest : public ::testing::Test {
-protected:
- // lsq_dim_c and lsq_dim_unk are instantiated in compile time.
- static constexpr std::tuple<int, int>
- init_lsq_dim(ReconstructionMethod method) {
- switch (method) {
- case ReconstructionMethod::linear:
- return std::make_tuple(3, 2);
- case ReconstructionMethod::quadratic:
- return std::make_tuple(9, 5);
- case ReconstructionMethod::cubic:
- return std::make_tuple(9, 9);
- }
- }
-
- // Constant dimensions.
- static constexpr int nproma = 3; // inner loop length
- static constexpr int nlev = 1; // number of vertical levels
- static constexpr int nblks_c = 1; // number of cell blocks (for p_e_in)
- static constexpr std::tuple<int, int> lsq_dim =
- init_lsq_dim(static_cast<ReconstructionMethod>(ReconMethod));
- static constexpr int lsq_dim_c = std::get<0>(lsq_dim);
- static constexpr int lsq_dim_unk = std::get<1>(lsq_dim);
-
- // Parameter values.
- int i_startblk = 0;
- int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
- int i_startidx_in = 0;
- int i_endidx_in = nproma; // Full range: 0 .. nproma-1
- int slev = 0;
- int elev = nlev; // Full vertical range (0 .. nlev-1)
- int patch_id = 0;
- bool lacc = false; // Not using ACC-specific behavior.
- bool acc_async = false; // No asynchronous execution.
- bool l_consv = true; // With conservative correction.
- bool l_limited_area = true; // Limited area setup
-
- std::vector<ValueType> p_cc;
- std::vector<int> cell_neighbor_idx;
- std::vector<int> cell_neighbor_blk;
- std::vector<ValueType> lsq_qtmat_c;
- std::vector<ValueType> lsq_rmat_rdiag_c;
- std::vector<ValueType> lsq_rmat_utri_c;
- std::vector<ValueType> lsq_moments;
- std::vector<ValueType> lsq_pseudoinv;
- std::vector<ValueType> p_coeff;
-
- HorizontalReconTest() {
- p_cc.resize(dim_combine(nproma, nlev, nblks_c));
- cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
- cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
- lsq_qtmat_c.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
- lsq_rmat_rdiag_c.resize(dim_combine(nproma, lsq_dim_unk, nblks_c));
- lsq_rmat_utri_c.resize(dim_combine(
- nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c));
- lsq_moments.resize(dim_combine(nproma, nblks_c, lsq_dim_unk));
- lsq_pseudoinv.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
- p_coeff.resize(dim_combine(lsq_dim_unk + 1, nproma, nlev, nblks_c));
- }
-};
-
-/// Test class for the horizontal tests. The reconstruction method is specified
-/// to linear.
-template <typename ValueType>
-class HorizontalReconLinearTest
- : public HorizontalReconTest<ValueType, static_cast<int>(
- ReconstructionMethod::linear)> {
-};
-
-/// Test class for the horizontal tests. The reconstruction method is specified
-/// to quadratic.
-template <typename ValueType>
-class HorizontalReconQuadraticTest
- : public HorizontalReconTest<
- ValueType, static_cast<int>(ReconstructionMethod::quadratic)> {};
-
-/// Test class for the horizontal tests. The reconstruction method is specified
-/// to cubic.
-template <typename ValueType>
-class HorizontalReconCubicTest
- : public HorizontalReconTest<ValueType, static_cast<int>(
- ReconstructionMethod::cubic)> {
-};
-
-/// ValueTypes which the divrot tests should run with
-typedef ::testing::Types<float, double> ValueTypes;
-
-TYPED_TEST_SUITE(HorizontalReconLinearTest, ValueTypes);
-
-TYPED_TEST(HorizontalReconLinearTest, TestLsqCell) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
- this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
- this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- }
-
- // Test function
- recon_lsq_cell_l<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
- this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
- this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- 0.34, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.8, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 1.0, 1e-6);
-}
-
-TYPED_TEST(HorizontalReconLinearTest, TestLsqCellRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = real_distrib(gen);
- this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = real_distrib(gen);
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
- }
-
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = real_distrib(gen);
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = real_distrib(gen);
- this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = real_distrib(gen);
-
- this->lsq_moments[moments_at(i, 0, 0)] = real_distrib(gen);
- this->lsq_moments[moments_at(i, 0, 1)] = real_distrib(gen);
- }
-
- // Test function
- recon_lsq_cell_l<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
- this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
- this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
-
- // Compute reference result
- std::vector<TypeParam> z_d(lsq_dim_c);
- std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
- std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_d[i] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- z_qt_times_d[0] = 0.0;
- z_qt_times_d[1] = 0.0;
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_qt_times_d[0] += this->lsq_qtmat_c[qtmat_at(jc, 0, i, jb)] * z_d[i];
- z_qt_times_d[1] += this->lsq_qtmat_c[qtmat_at(jc, 1, i, jb)] * z_d[i];
- }
- p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] =
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 1, jb)] * z_qt_times_d[1];
- p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] =
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 0, jb)] *
- (z_qt_times_d[0] -
- this->lsq_rmat_utri_c[rmat_utri_at(jc, 0, jb)] *
- p_result[at<lsq_dim_unk + 1, nproma>(2, jc)]);
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)] -
- p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
- this->lsq_moments[moments_at(jc, jb, 0)] -
- p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
- this->lsq_moments[moments_at(jc, jb, 1)];
- }
- }
- }
-
- // Check result
- for (int i = 0; i < lsq_dim_unk + 1; ++i) {
- for (int jc = 0; jc < nproma; ++jc) {
- EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
- p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
- << "For loop result fails for i = " << i << ", jc = " << jc;
- }
- }
-}
-
-TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVD) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- }
-
- // Test function
- recon_lsq_cell_l_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
- this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- 0.65, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.0, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 0.5, 1e-6);
-}
-
-TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVDRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = real_distrib(gen);
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = real_distrib(gen);
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = real_distrib(gen);
- this->lsq_moments[moments_at(i, 0, 1)] = real_distrib(gen);
- }
-
- // Test function
- recon_lsq_cell_l_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
- this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
-
- // Compute reference result
- std::vector<TypeParam> z_d(lsq_dim_c);
- std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_d[i] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] =
- this->lsq_pseudoinv[pseudoinv_at(jc, 1, 0, jb)] * z_d[0] +
- this->lsq_pseudoinv[pseudoinv_at(jc, 1, 1, jb)] * z_d[1] +
- this->lsq_pseudoinv[pseudoinv_at(jc, 1, 2, jb)] * z_d[2];
- p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] =
- this->lsq_pseudoinv[pseudoinv_at(jc, 0, 0, jb)] * z_d[0] +
- this->lsq_pseudoinv[pseudoinv_at(jc, 0, 1, jb)] * z_d[1] +
- this->lsq_pseudoinv[pseudoinv_at(jc, 0, 2, jb)] * z_d[2];
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)] -
- p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
- this->lsq_moments[moments_at(jc, jb, 0)] -
- p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
- this->lsq_moments[moments_at(jc, jb, 1)];
- }
- }
- }
-
- // Check result
- for (int i = 0; i < lsq_dim_unk + 1; ++i) {
- for (int jc = 0; jc < nproma; ++jc) {
- EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
- p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
- << "For loop result fails for i = " << i << ", jc = " << jc;
- }
- }
-}
-
-TYPED_TEST_SUITE(HorizontalReconQuadraticTest, ValueTypes);
-
-TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCell) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
- for (int j = 1; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
- this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
- this->lsq_qtmat_c[qtmat_at(i, 2, j, 0)] = 0.2;
- this->lsq_qtmat_c[qtmat_at(i, 3, j, 0)] = 0.7;
- this->lsq_qtmat_c[qtmat_at(i, 4, j, 0)] = 1.3;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- for (int j = 0; j < lsq_dim_unk; ++j) {
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = 2.0;
- }
-
- for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
- this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = 1.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
- this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
- this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
- }
-
- // Test function
- recon_lsq_cell_q<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
- this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- 0.24, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 3.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- -2.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
- 2.8, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
- -3.8, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
- 2.6, 1e-6);
-}
-
-TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_unk; ++j) {
- for (int k = 0; k < lsq_dim_c; ++k) {
- this->lsq_qtmat_c[qtmat_at(i, j, k, 0)] = real_distrib(gen);
- }
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = real_distrib(gen);
- this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
- }
- for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
- this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = real_distrib(gen);
- }
-
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
- }
- }
-
- // Test function
- recon_lsq_cell_q<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
- this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Compute reference result
- std::vector<TypeParam> z_d(lsq_dim_c);
- std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
- std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_d[i] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- for (int j = 0; j < lsq_dim_unk; ++j) {
- z_qt_times_d[j] = 0.0;
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_qt_times_d[j] +=
- this->lsq_qtmat_c[qtmat_at(jc, j, i, jb)] * z_d[i];
- }
- }
- int utri_id = 0;
- for (int j = lsq_dim_unk; j > 0; --j) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = z_qt_times_d[j - 1];
- for (int k = j + 1; k <= lsq_dim_unk; ++k) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] -=
- this->lsq_rmat_utri_c[rmat_utri_at(jc, utri_id++, jb)] *
- p_result[at<lsq_dim_unk + 1, nproma>(k, jc)];
- }
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] *=
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, j - 1, jb)];
- }
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)];
- for (int j = 0; j < lsq_dim_unk; ++j) {
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
- p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
- this->lsq_moments[moments_at(jc, jb, j)];
- }
- }
- }
- }
-
- // Check result
- for (int i = 0; i < lsq_dim_unk + 1; ++i) {
- for (int jc = 0; jc < nproma; ++jc) {
- EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
- p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
- << "For loop result fails for i = " << i << ", jc = " << jc;
- }
- }
-}
-
-TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVD) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
- for (int j = 1; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
- this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.2;
- this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
- this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 1.3;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
- this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
- this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
- }
-
- // Test function
- recon_lsq_cell_q_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- -0.56, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.0, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 0.5, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
- 0.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
- 0.7, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
- 1.3, 1e-6);
-}
-
-TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
-
- // Initialization is done only for iblk = 0 and ilev = 0
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_unk; ++j) {
- for (int k = 0; k < lsq_dim_c; ++k) {
- this->lsq_pseudoinv[pseudoinv_at(i, j, k, 0)] = real_distrib(gen);
- }
- this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
- }
-
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
- }
- }
-
- // Test function
- recon_lsq_cell_q_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Compute reference result
- std::vector<TypeParam> z_d(lsq_dim_c);
- std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
- std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_d[i] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- for (int j = 1; j < lsq_dim_unk + 1; ++j) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
- for (int i = 0; i < lsq_dim_c; ++i) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
- this->lsq_pseudoinv[pseudoinv_at(jc, j - 1, i, jb)] * z_d[i];
- }
- }
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)];
- for (int j = 0; j < lsq_dim_unk; ++j) {
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
- p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
- this->lsq_moments[moments_at(jc, jb, j)];
- }
- }
- }
- }
-
- // Check result
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- for (int jc = 0; jc < nproma; ++jc) {
- EXPECT_NEAR(this->p_coeff[(p_coeff_at(j, jc, 0, 0))],
- p_result[(at<lsq_dim_unk + 1, nproma>(j, jc))], 1e-5)
- << "For loop result fails for j = " << j << ", jc = " << jc;
- }
- }
-}
-
-TYPED_TEST_SUITE(HorizontalReconCubicTest, ValueTypes);
-
-TYPED_TEST(HorizontalReconCubicTest, TestLsqCell) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
- for (int j = 1; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
- this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.9;
- this->lsq_qtmat_c[qtmat_at(i, 2, j, 0)] = 0.8;
- this->lsq_qtmat_c[qtmat_at(i, 3, j, 0)] = 0.7;
- this->lsq_qtmat_c[qtmat_at(i, 4, j, 0)] = 0.6;
- this->lsq_qtmat_c[qtmat_at(i, 5, j, 0)] = 0.5;
- this->lsq_qtmat_c[qtmat_at(i, 6, j, 0)] = 0.4;
- this->lsq_qtmat_c[qtmat_at(i, 7, j, 0)] = 0.3;
- this->lsq_qtmat_c[qtmat_at(i, 8, j, 0)] = 0.2;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- for (int j = 0; j < lsq_dim_unk; ++j) {
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = 2.0;
- }
-
- for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
- this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = 1.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
- this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
- this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
- this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
- this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
- this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
- this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
- }
-
- // Test function
- recon_lsq_cell_c<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
- this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- 0.28, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 0.4, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- -0.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
- 0.4, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
- -0.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
- 0.4, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
- -0.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
- 0.4, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
- -0.2, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
- 0.4, 1e-6);
-}
-
-TYPED_TEST(HorizontalReconCubicTest, TestLsqCellRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
- const auto &rmat_utri_at =
- at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_unk; ++j) {
- for (int k = 0; k < lsq_dim_c; ++k) {
- this->lsq_qtmat_c[qtmat_at(i, j, k, 0)] = real_distrib(gen);
- }
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = real_distrib(gen);
- this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
- }
- for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
- this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = real_distrib(gen);
- }
-
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
- }
- }
-
- // Test function
- recon_lsq_cell_c<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
- this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Compute reference result
- std::vector<TypeParam> z_d(lsq_dim_c);
- std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
- std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_d[i] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- for (int j = 0; j < lsq_dim_unk; ++j) {
- z_qt_times_d[j] = 0.0;
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_qt_times_d[j] +=
- this->lsq_qtmat_c[qtmat_at(jc, j, i, jb)] * z_d[i];
- }
- }
- int utri_id = 0;
- for (int j = lsq_dim_unk; j > 0; --j) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = z_qt_times_d[j - 1];
- for (int k = j + 1; k <= lsq_dim_unk; ++k) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] -=
- this->lsq_rmat_utri_c[rmat_utri_at(jc, utri_id++, jb)] *
- p_result[at<lsq_dim_unk + 1, nproma>(k, jc)];
- }
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] *=
- this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, j - 1, jb)];
- }
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)];
- for (int j = 0; j < lsq_dim_unk; ++j) {
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
- p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
- this->lsq_moments[moments_at(jc, jb, j)];
- }
- }
- }
- }
-
- // Check result
- for (int i = 0; i < lsq_dim_unk + 1; ++i) {
- for (int jc = 0; jc < nproma; ++jc) {
- EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
- p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
- << "For loop result fails for i = " << i << ", jc = " << jc;
- }
- }
-}
-
-TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVD) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
- for (int j = 1; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
- this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.9;
- this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.8;
- this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
- this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 0.6;
- this->lsq_pseudoinv[pseudoinv_at(i, 5, j, 0)] = 0.5;
- this->lsq_pseudoinv[pseudoinv_at(i, 6, j, 0)] = 0.4;
- this->lsq_pseudoinv[pseudoinv_at(i, 7, j, 0)] = 0.3;
- this->lsq_pseudoinv[pseudoinv_at(i, 8, j, 0)] = 0.2;
- }
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
- }
-
- this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
- this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
- this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
- this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
- this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
- this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
- this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
- this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
- this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
- }
-
- // Test function
- recon_lsq_cell_c_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Check result
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
- -1.64, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
- 1.0, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
- 0.9, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
- 0.8, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
- 0.7, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
- 0.6, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
- 0.5, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
- 0.4, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
- 0.3, 1e-6);
- EXPECT_NEAR(
- this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
- 0.2, 1e-6);
-}
-
-TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int lsq_dim_c = this->lsq_dim_c;
- constexpr int lsq_dim_unk = this->lsq_dim_unk;
-
- const auto &p_cc_at = at<nproma, nlev, nblks_c>;
- const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
- const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
- const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
- const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
-
- for (int j = 0; j < lsq_dim_c; ++j) {
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- for (int j = 0; j < lsq_dim_unk; ++j) {
- for (int k = 0; k < lsq_dim_c; ++k) {
- this->lsq_pseudoinv[pseudoinv_at(i, j, k, 0)] = real_distrib(gen);
- }
- this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
- }
-
- for (int j = 0; j < lsq_dim_unk + 1; ++j) {
- this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
- }
- }
-
- // Test function
- recon_lsq_cell_c_svd<TypeParam>(
- this->p_cc.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
- this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
- this->elev, this->nproma, this->patch_id, this->l_limited_area,
- this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
- this->lsq_dim_c);
-
- // Compute reference result
- std::vector<TypeParam> z_d(lsq_dim_c);
- std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
- std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
-
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
- for (int jk = this->slev; jk < this->elev; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- for (int i = 0; i < lsq_dim_c; ++i) {
- z_d[i] = this->p_cc[p_cc_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
- this->p_cc[p_cc_at(jc, jk, jb)];
- }
- for (int j = 1; j < lsq_dim_unk + 1; ++j) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
- for (int i = 0; i < lsq_dim_c; ++i) {
- p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
- this->lsq_pseudoinv[pseudoinv_at(jc, j - 1, i, jb)] * z_d[i];
- }
- }
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
- this->p_cc[p_cc_at(jc, jk, jb)];
- for (int j = 0; j < lsq_dim_unk; ++j) {
- p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
- p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
- this->lsq_moments[moments_at(jc, jb, j)];
- }
- }
- }
- }
- // Check result
- for (int i = 0; i < lsq_dim_unk + 1; ++i) {
- for (int jc = 0; jc < nproma; ++jc) {
- EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
- p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
- << "For loop result fails for i = " << i << ", jc = " << jc;
- }
- }
-}
-
-template <typename ValueType> class HorizontalDivTest : public ::testing::Test {
-protected:
- static constexpr int nproma = 3; // inner loop length
- static constexpr int nlev = 2; // number of vertical levels
- static constexpr int nblks_c = 1; // number of cell blocks
- static constexpr int nblks_e = 1; // number of edge blocks
- static constexpr int dim4d = 2; // 4th dimension size
-
- int i_startblk = 0;
- int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
- int i_startidx_in = 0;
- int i_endidx_in = nproma; // Full range: 0 .. nproma-1
- std::vector<int> slev;
- std::vector<int> elev;
- bool lacc = false; // Not using ACC-specific behavior.
-
- std::vector<ValueType> vec_e;
- std::vector<int> cell_edge_idx;
- std::vector<int> cell_edge_blk;
- std::vector<ValueType> geofac_div;
- std::vector<ValueType> div_vec_c;
- std::vector<ValueType> f4din;
- std::vector<ValueType> f4dout;
-
- // Followings are needed in HorizontalDivAvgTest
- std::vector<int> cell_neighbor_idx;
- std::vector<int> cell_neighbor_blk;
- std::vector<ValueType> avg_coeff;
- std::vector<ValueType> opt_in2;
- std::vector<ValueType> opt_out2;
-
- HorizontalDivTest() {
- slev.resize(dim4d, 0);
- elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
-
- vec_e.resize(dim_combine(nproma, nlev, nblks_e));
- cell_edge_idx.resize(dim_combine(nproma, nblks_c, 3));
- cell_edge_blk.resize(dim_combine(nproma, nblks_c, 3));
- geofac_div.resize(dim_combine(nproma, 3, nblks_c));
- div_vec_c.resize(dim_combine(nproma, nlev, nblks_c));
- f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
- f4dout.resize(dim_combine(nproma, nlev, nblks_c, dim4d));
- cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, 3));
- cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, 3));
- avg_coeff.resize(dim_combine(nproma, 4, nblks_c));
- opt_in2.resize(dim_combine(nproma, nlev, nblks_e));
- opt_out2.resize(dim_combine(nproma, nlev, nblks_c));
- }
-};
-
-TYPED_TEST_SUITE(HorizontalDivTest, ValueTypes);
-
-TYPED_TEST(HorizontalDivTest, TestDiv3DSpecific) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
-
- // Initialization with specific values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- }
-
- // Set edge indices to point to specific cells (including self)
- this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
- this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
- this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
-
- // All edges are in the same block for this test
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
- }
-
- // Geometric factors
- this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
- this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
- this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
-
- // Initialize div_vec_c to zero
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
- }
- }
-
- // Call the div3d function
- div3d<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.1, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
-}
-
-TYPED_TEST(HorizontalDivTest, TestDiv3DRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
-
- // Set up random number generators
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
-
- // Initialization with random values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- }
-
- // Set random edge indices
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 3; ++j) {
- this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Initialize div_vec_c to random values
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- }
- }
-
- // Call the div3d function
- div3d<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- // Calculate reference values separately and verify results
- std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
-
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
- }
- }
- }
-
- // Verify results
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
- ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
- << "Results differ at i=" << i << ", k=" << k;
- }
- }
-}
-
-TYPED_TEST(HorizontalDivTest, TestDiv3D2FSpecific) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
- constexpr int dim4d = this->dim4d;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
- const auto &f4d_at = at<nproma, nlev, nblks_e, dim4d>;
- const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
-
- // Initialization with specific values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- this->f4din[f4d_at(i, k, 0, 0)] =
- (i + 1) * (k + 2); // Different pattern for second field
- }
-
- // Set edge indices to point to specific cells (including self)
- this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
- this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
- this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
-
- // All edges are in the same block for this test
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
- }
-
- // Geometric factors
- this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
- this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
- this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
-
- // Initialize div_vec_c and f4dout to zero
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
- this->f4dout[f4dout_at(i, k, 0, 0)] = 0.0;
- }
- }
-
- // Call the div3d_2field function
- div3d_2field<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->f4din.data(),
- this->f4dout.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- // Check first field (same as in div3d test)
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.7, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.4, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.1, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.2, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.2, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.4, 1e-6);
-
- // Check second field (expected values calculated manually)
- EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 3.4, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 0)], 5.1, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 4.2, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 0)], 6.3, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 0)], 4.4, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 6.6, 1e-6);
-}
-
-TYPED_TEST(HorizontalDivTest, TestDiv3D2FRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
- constexpr int dim4d = this->dim4d;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
- const auto &f4d_at = at<nproma, nlev, nblks_e, dim4d>;
- const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
-
- // Set up random number generators
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
-
- // Initialization with random values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- this->f4din[f4d_at(i, k, 0, 0)] = real_distrib(gen);
- }
-
- // Set random edge indices
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 3; ++j) {
- this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Initialize div_vec_c and f4dout to random values
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- this->f4dout[f4dout_at(i, k, 0, 0)] = real_distrib(gen);
- }
- }
-
- // Call the div3d_2field function
- div3d_2field<TypeParam>(this->vec_e.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->div_vec_c.data(), this->f4din.data(),
- this->f4dout.data(), this->i_startblk, this->i_endblk,
- this->i_startidx_in, this->i_endidx_in, this->slev[0],
- this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- // Calculate reference values separately and verify results
- std::vector<TypeParam> ref_div_vec_c(nproma * nlev * nblks_c, 0.0);
- std::vector<TypeParam> ref_f4dout(nproma * nlev * nblks_c * dim4d, 0.0);
-
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- // Calculate reference value for first field
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
-
- // Calculate reference value for second field
- ref_f4dout[f4dout_at(jc, jk, jb, 0)] =
- this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)],
- 0)] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)],
- 0)] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->f4din[f4d_at(this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)],
- 0)] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
- }
- }
- }
-
- // Verify results for first field
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
- ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
- << "First field results differ at i=" << i << ", k=" << k;
- }
- }
-
- // Verify results for second field
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->f4dout[f4dout_at(i, k, 0, 0)],
- ref_f4dout[f4dout_at(i, k, 0, 0)], 1e-5)
- << "Second field results differ at i=" << i << ", k=" << k;
- }
- }
-}
-
-TYPED_TEST(HorizontalDivTest, TestDiv4DSpecific) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
- constexpr int dim4d = this->dim4d;
-
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &f4din_at = at<nproma, nlev, nblks_e, dim4d>;
- const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
-
- // Initialization
- for (int i = 0; i < nproma; ++i) {
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_idx[cell_edge_at(i, 0, j)] = (i + j) % nproma;
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
- this->geofac_div[geofac_div_at(i, j, 0)] = 0.1 * (j + 1);
- }
-
- for (int k = 0; k < nlev; ++k) {
- for (int d = 0; d < dim4d; ++d) {
- this->f4din[f4din_at(i, k, 0, d)] = 1.0 + i + k + d;
- this->f4dout[f4dout_at(i, k, 0, d)] = 0.0;
- }
- }
- }
-
- // Test function
- div4d<TypeParam>(this->cell_edge_idx.data(), this->cell_edge_blk.data(),
- this->geofac_div.data(), this->f4din.data(),
- this->f4dout.data(), this->dim4d, this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in,
- this->slev.data(), this->elev.data(), this->nproma,
- this->lacc, this->nlev, this->nblks_c, this->nblks_e);
-
- EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 0)], 1.4, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 0)], 1.1, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 0)], 1.1, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 0)], 2.0, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 0)], 1.7, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 0)], 1.7, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(0, 0, 0, 1)], 2.0, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(1, 0, 0, 1)], 1.7, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(2, 0, 0, 1)], 1.7, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(0, 1, 0, 1)], 2.6, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(1, 1, 0, 1)], 2.3, 1e-6);
- EXPECT_NEAR(this->f4dout[f4dout_at(2, 1, 0, 1)], 2.3, 1e-6);
-}
-
-TYPED_TEST(HorizontalDivTest, TestDiv4DRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
- constexpr int dim4d = this->dim4d;
-
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &f4din_at = at<nproma, nlev, nblks_e, dim4d>;
- const auto &f4dout_at = at<nproma, nlev, nblks_c, dim4d>;
-
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
-
- // Initialize with random values
- for (int i = 0; i < nproma; ++i) {
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
- this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
- }
-
- for (int k = 0; k < nlev; ++k) {
- for (int d = 0; d < dim4d; ++d) {
- this->f4din[f4din_at(i, k, 0, d)] = real_distrib(gen);
- this->f4dout[f4dout_at(i, k, 0, d)] = 0.0;
- }
- }
- }
-
- // Test function
- div4d<TypeParam>(this->cell_edge_idx.data(), this->cell_edge_blk.data(),
- this->geofac_div.data(), this->f4din.data(),
- this->f4dout.data(), this->dim4d, this->i_startblk,
- this->i_endblk, this->i_startidx_in, this->i_endidx_in,
- this->slev.data(), this->elev.data(), this->nproma,
- this->lacc, this->nlev, this->nblks_c, this->nblks_e);
-
- // Compute reference result and check
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
-
- for (int ji = 0; ji < dim4d; ++ji) {
- for (int jk = this->slev[ji]; jk < this->elev[ji]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- TypeParam expected = 0.0;
- for (int je = 0; je < 3; ++je) {
- expected +=
- this->f4din[f4din_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, je)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, je)], ji)] *
- this->geofac_div[geofac_div_at(jc, je, jb)];
- }
-
- EXPECT_NEAR(this->f4dout[f4dout_at(jc, jk, jb, ji)], expected, 1e-5)
- << "Random test fails at jc=" << jc << ", jk=" << jk
- << ", jb=" << jb << ", ji=" << ji;
- }
- }
- }
- }
-}
-
-TYPED_TEST_SUITE(HorizontalDivTest, ValueTypes);
-
-TYPED_TEST(HorizontalDivTest, TestDivAvgSpecific) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
- constexpr int dim4d = this->dim4d;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
-
- // Vectors for additional parameters
- // Vectors for block and index ranges
- std::vector<int> i_startblk_in(3, 0);
- std::vector<int> i_endblk_in(3, nblks_c);
- std::vector<int> i_startidx_in(3, 0);
- std::vector<int> i_endidx_in(3, nproma);
-
- // Parameters for the test
- int patch_id = 1;
- bool l_limited_area = true;
- bool l2fields = true;
-
- const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
- const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
-
- // Initialize the vectors with specific values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- this->opt_in2[vec_e_at(i, k, 0)] =
- (i + 1) * (k + 1) * 0.5; // Half of vec_e
- }
-
- // Set edge indices to point to specific cells
- this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
- this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
- this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
-
- // Set neighbor indices similarly
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = i;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = (i + 1) % nproma;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = (i + 2) % nproma;
-
- // All edges and neighbors are in the same block for this test
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- // Geometric factors
- this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
- this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
- this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
-
- // Average coefficients
- this->avg_coeff[avg_coeff_at(i, 0, 0)] = 0.4; // Self
- this->avg_coeff[avg_coeff_at(i, 1, 0)] = 0.2; // First neighbor
- this->avg_coeff[avg_coeff_at(i, 2, 0)] = 0.2; // Second neighbor
- this->avg_coeff[avg_coeff_at(i, 3, 0)] = 0.2; // Third neighbor
-
- // Initialize div_vec_c and opt_out2 to zero
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
- this->opt_out2[div_vec_c_at(i, k, 0)] = 0.0;
- }
- }
-
- // Call the div_avg function
- div_avg<TypeParam>(
- this->vec_e.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
- this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
- this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.08, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.08, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.16, 1e-6);
-
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 0, 0)], 0.94, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 1, 0)], 1.88, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 0, 0)], 1.02, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 2.04, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 1.04, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 2.08, 1e-6);
-}
-
-TYPED_TEST(HorizontalDivTest, TestDivAvgRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
-
- // Vectors for block and index ranges
- std::vector<int> i_startblk_in(3, 0);
- std::vector<int> i_endblk_in(3, nblks_c);
- std::vector<int> i_startidx_in(3, 0);
- std::vector<int> i_endidx_in(3, nproma);
-
- // Parameters for the test
- int patch_id = 1;
- bool l_limited_area = true;
- bool l2fields = true;
-
- const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
- const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
-
- // Set up random number generators
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
-
- // Initialize with random values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- this->opt_in2[vec_e_at(i, k, 0)] = real_distrib(gen);
- }
-
- // Set random edge indices
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 3; ++j) {
- this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Random average coefficients
- for (int j = 0; j < 4; ++j) {
- this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Random initial values for div_vec_c and opt_out2
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- this->opt_out2[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- }
- }
-
- // Call the div_avg function
- div_avg<TypeParam>(
- this->vec_e.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
- this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
- this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- // Calculate reference values manually
- std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> aux_c2(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> ref_opt_out2(dim_combine(nproma, nlev, nblks_c));
-
- // Step 1: Calculate aux_c and aux_c2
- for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
- i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- aux_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
-
- aux_c2[div_vec_c_at(jc, jk, jb)] =
- this->opt_in2[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->opt_in2[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->opt_in2[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
- }
- }
- }
-
- // Step 2: Assign aux_c to div_vec_c and aux_c2 to opt_out2 for patch_id > 0
- for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
- i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- aux_c[div_vec_c_at(jc, jk, jb)];
- ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
- aux_c2[div_vec_c_at(jc, jk, jb)];
- }
- }
- }
-
- // Step 3: Perform averaging for the rest of the blocks
- for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
- i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- aux_c[div_vec_c_at(jc, jk, jb)] *
- this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
- aux_c[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
- this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
- aux_c[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
- this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
- aux_c[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
- this->avg_coeff[avg_coeff_at(jc, 3, jb)];
-
- ref_opt_out2[div_vec_c_at(jc, jk, jb)] =
- aux_c2[div_vec_c_at(jc, jk, jb)] *
- this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
- aux_c2[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
- this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
- aux_c2[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
- this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
- aux_c2[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
- this->avg_coeff[avg_coeff_at(jc, 3, jb)];
- }
- }
- }
-
- // Verify results
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
- ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
- << "div_vec_c results differ at i=" << i << ", k=" << k;
-
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(i, k, 0)],
- ref_opt_out2[div_vec_c_at(i, k, 0)], 1e-5)
- << "opt_out2 results differ at i=" << i << ", k=" << k;
- }
- }
-}
-
-TYPED_TEST(HorizontalDivTest, TestDivAvgSpecificNoL2fields) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
- constexpr int dim4d = this->dim4d;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
-
- // Vectors for block and index ranges
- std::vector<int> i_startblk_in(3, 0);
- std::vector<int> i_endblk_in(3, nblks_c);
- std::vector<int> i_startidx_in(3, 0);
- std::vector<int> i_endidx_in(3, nproma);
-
- // Parameters for the test
- int patch_id = 1;
- bool l_limited_area = true;
- bool l2fields = false;
-
- const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
- const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
-
- // Initialize the vectors with specific values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- this->opt_in2[vec_e_at(i, k, 0)] =
- (i + 1) * (k + 1) * 0.5; // Half of vec_e
- }
-
- // Set edge indices to point to specific cells
- this->cell_edge_idx[cell_edge_at(i, 0, 0)] = i;
- this->cell_edge_idx[cell_edge_at(i, 0, 1)] = (i + 1) % nproma;
- this->cell_edge_idx[cell_edge_at(i, 0, 2)] = (i + 2) % nproma;
-
- // Set neighbor indices similarly
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = i;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = (i + 1) % nproma;
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = (i + 2) % nproma;
-
- // All edges and neighbors are in the same block for this test
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_blk[cell_edge_at(i, 0, j)] = 0;
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
- }
-
- // Geometric factors
- this->geofac_div[geofac_div_at(i, 0, 0)] = 0.5;
- this->geofac_div[geofac_div_at(i, 1, 0)] = 0.3;
- this->geofac_div[geofac_div_at(i, 2, 0)] = 0.2;
-
- // Average coefficients
- this->avg_coeff[avg_coeff_at(i, 0, 0)] = 0.4; // Self
- this->avg_coeff[avg_coeff_at(i, 1, 0)] = 0.2; // First neighbor
- this->avg_coeff[avg_coeff_at(i, 2, 0)] = 0.2; // Second neighbor
- this->avg_coeff[avg_coeff_at(i, 3, 0)] = 0.2; // Third neighbor
-
- // Initialize div_vec_c and opt_out2 to zero
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = 0.0;
- this->opt_out2[div_vec_c_at(i, k, 0)] = 0.0;
- }
- }
-
- // Call the div_avg function
- div_avg<TypeParam>(
- this->vec_e.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
- this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
- this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 0, 0)], 1.88, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(0, 1, 0)], 3.76, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 0, 0)], 2.04, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(1, 1, 0)], 4.08, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 0, 0)], 2.08, 1e-6);
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(2, 1, 0)], 4.16, 1e-6);
-
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 0, 0)], 0.0, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(0, 1, 0)], 0.0, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 0, 0)], 0.0, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(1, 1, 0)], 0.0, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 0, 0)], 0.0, 1e-6);
- EXPECT_NEAR(this->opt_out2[div_vec_c_at(2, 1, 0)], 0.0, 1e-6);
-}
-
-TYPED_TEST(HorizontalDivTest, TestDivAvgRandomNoL2fields) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_c = this->nblks_c;
- constexpr int nblks_e = this->nblks_e;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &cell_edge_at = at<nproma, nblks_c, 3>;
- const auto &geofac_div_at = at<nproma, 3, nblks_c>;
- const auto &div_vec_c_at = at<nproma, nlev, nblks_c>;
-
- // Vectors for block and index ranges
- std::vector<int> i_startblk_in(3, 0);
- std::vector<int> i_endblk_in(3, nblks_c);
- std::vector<int> i_startidx_in(3, 0);
- std::vector<int> i_endidx_in(3, nproma);
-
- // Parameters for the test
- int patch_id = 1;
- bool l_limited_area = true;
- bool l2fields = false; // Set to false for this test
-
- const auto &cell_neighbor_at = at<nproma, nblks_c, 3>;
- const auto &avg_coeff_at = at<nproma, 4, nblks_c>;
-
- // Set up random number generators
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
-
- // Initialize with random values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- this->opt_in2[vec_e_at(i, k, 0)] =
- real_distrib(gen); // Not used but initialize anyway
- }
-
- // Set random edge indices
- for (int j = 0; j < 3; ++j) {
- this->cell_edge_idx[cell_edge_at(i, 0, j)] = int_distrib(gen);
- this->cell_edge_blk[cell_edge_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
-
- this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
- this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 3; ++j) {
- this->geofac_div[geofac_div_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Random average coefficients
- for (int j = 0; j < 4; ++j) {
- this->avg_coeff[avg_coeff_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Random initial values for div_vec_c and opt_out2
- for (int k = 0; k < nlev; ++k) {
- this->div_vec_c[div_vec_c_at(i, k, 0)] = real_distrib(gen);
- this->opt_out2[div_vec_c_at(i, k, 0)] =
- real_distrib(gen); // Not used but initialize anyway
- }
- }
-
- // Call the div_avg function with l2fields=false
- div_avg<TypeParam>(
- this->vec_e.data(), this->cell_neighbor_idx.data(),
- this->cell_neighbor_blk.data(), this->cell_edge_idx.data(),
- this->cell_edge_blk.data(), this->geofac_div.data(),
- this->avg_coeff.data(), this->div_vec_c.data(), this->opt_in2.data(),
- this->opt_out2.data(), i_startblk_in.data(), i_endblk_in.data(),
- i_startidx_in.data(), i_endidx_in.data(), this->slev[0], this->elev[0],
- this->nproma, patch_id, l_limited_area, l2fields, this->lacc, this->nlev,
- this->nblks_c, this->nblks_e);
-
- // Calculate reference values manually
- std::vector<TypeParam> aux_c(dim_combine(nproma, nlev, nblks_c));
- std::vector<TypeParam> ref_div_vec_c(dim_combine(nproma, nlev, nblks_c));
-
- // Step 1: Calculate aux_c (but not aux_c2 since l2fields=false)
- for (int jb = i_startblk_in[0]; jb < i_endblk_in[0]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[0], i_endidx_in[0], nproma, jb,
- i_startblk_in[0], i_endblk_in[0], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- aux_c[div_vec_c_at(jc, jk, jb)] =
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 0)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 0)])] *
- this->geofac_div[geofac_div_at(jc, 0, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 1)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 1)])] *
- this->geofac_div[geofac_div_at(jc, 1, jb)] +
- this->vec_e[vec_e_at(
- this->cell_edge_idx[cell_edge_at(jc, jb, 2)], jk,
- this->cell_edge_blk[cell_edge_at(jc, jb, 2)])] *
- this->geofac_div[geofac_div_at(jc, 2, jb)];
- }
- }
- }
-
- // Step 2: Assign aux_c to div_vec_c for patch_id > 0 (opt_out2 not updated
- // since l2fields=false)
- for (int jb = i_startblk_in[1]; jb < i_endblk_in[1]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[1], i_endidx_in[1], nproma, jb,
- i_startblk_in[1], i_endblk_in[1], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- aux_c[div_vec_c_at(jc, jk, jb)];
- }
- }
- }
-
- // Step 3: Perform averaging for the rest of the blocks (only for div_vec_c,
- // not opt_out2)
- for (int jb = i_startblk_in[2]; jb < i_endblk_in[2]; ++jb) {
- int i_startidx, i_endidx;
- get_indices_c_lib(i_startidx_in[2], i_endidx_in[2], nproma, jb,
- i_startblk_in[2], i_endblk_in[2], i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jc = i_startidx; jc < i_endidx; ++jc) {
- ref_div_vec_c[div_vec_c_at(jc, jk, jb)] =
- aux_c[div_vec_c_at(jc, jk, jb)] *
- this->avg_coeff[avg_coeff_at(jc, 0, jb)] +
- aux_c[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 0)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 0)])] *
- this->avg_coeff[avg_coeff_at(jc, 1, jb)] +
- aux_c[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 1)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 1)])] *
- this->avg_coeff[avg_coeff_at(jc, 2, jb)] +
- aux_c[div_vec_c_at(
- this->cell_neighbor_idx[cell_neighbor_at(jc, jb, 2)], jk,
- this->cell_neighbor_blk[cell_neighbor_at(jc, jb, 2)])] *
- this->avg_coeff[avg_coeff_at(jc, 3, jb)];
- }
- }
- }
-
- // Verify results - only check div_vec_c since l2fields=false means opt_out2
- // isn't updated
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->div_vec_c[div_vec_c_at(i, k, 0)],
- ref_div_vec_c[div_vec_c_at(i, k, 0)], 1e-5)
- << "div_vec_c results differ at i=" << i << ", k=" << k;
- }
- }
-}
-
-template <typename ValueType>
-class HorizontalRotVertexTest : public ::testing::Test {
-protected:
- static constexpr int nproma = 3; // inner loop length
- static constexpr int nlev = 2; // number of vertical levels
- static constexpr int nblks_e = 1; // number of edge blocks
- static constexpr int nblks_v = 1; // number of vertex blocks
- static constexpr int dim4d = 2; // 4th dimension size
-
- int i_startblk = 0;
- int i_endblk = nblks_v; // Test blocks [0 .. nblks_v-1]
- int i_startidx_in = 0;
- int i_endidx_in = nproma; // Full range: 0 .. nproma-1
- std::vector<int> slev;
- std::vector<int> elev;
- bool lacc = false; // Not using ACC-specific behavior.
- bool acc_async = false; // Not using ACC-specific behavior.
-
- std::vector<ValueType> vec_e;
- std::vector<int> vert_edge_idx;
- std::vector<int> vert_edge_blk;
- std::vector<ValueType> geofac_rot;
- std::vector<ValueType> rot_vec;
- std::vector<ValueType> f4din;
- std::vector<ValueType> f4dout;
-
- HorizontalRotVertexTest() {
- slev.resize(dim4d, 0);
- elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
-
- vec_e.resize(dim_combine(nproma, nlev, nblks_e));
- vert_edge_idx.resize(dim_combine(nproma, nblks_v, 6));
- vert_edge_blk.resize(dim_combine(nproma, nblks_v, 6));
- geofac_rot.resize(dim_combine(nproma, 6, nblks_v));
- rot_vec.resize(dim_combine(nproma, nlev, nblks_v));
- f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
- f4dout.resize(dim_combine(nproma, nlev, nblks_v, dim4d));
- }
-};
-
-template <typename ValueType>
-class HorizontalRotVertexAtmosTest : public HorizontalRotVertexTest<ValueType> {
-};
-
-template <typename ValueType>
-class HorizontalRotVertexRITest : public HorizontalRotVertexTest<ValueType> {};
-
-TYPED_TEST_SUITE(HorizontalRotVertexAtmosTest, ValueTypes);
-
-TYPED_TEST(HorizontalRotVertexAtmosTest, TestSpecific) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_e = this->nblks_e;
- constexpr int nblks_v = this->nblks_v;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &vert_edge_at = at<nproma, nblks_v, 6>;
- const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
- const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
-
- // Initialization with specific values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- }
-
- // Set edge indices to point to specific edges
- for (int j = 0; j < 6; ++j) {
- this->vert_edge_idx[vert_edge_at(i, 0, j)] = (i + j) % nproma;
- // All edges are in the same block for this test
- this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0;
- }
-
- // Geometric factors for rotation
- this->geofac_rot[geofac_rot_at(i, 0, 0)] = 0.3;
- this->geofac_rot[geofac_rot_at(i, 1, 0)] = 0.2;
- this->geofac_rot[geofac_rot_at(i, 2, 0)] = 0.1;
- this->geofac_rot[geofac_rot_at(i, 3, 0)] = 0.2;
- this->geofac_rot[geofac_rot_at(i, 4, 0)] = 0.1;
- this->geofac_rot[geofac_rot_at(i, 5, 0)] = 0.1;
-
- // Initialize rot_vec to zero
- for (int k = 0; k < nlev; ++k) {
- this->rot_vec[rot_vec_at(i, k, 0)] = 0.0;
- }
- }
-
- // Call the rot_vertex_atmos function
- rot_vertex_atmos<TypeParam>(
- this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
- this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
- this->slev[0], this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_e, this->nblks_v);
-
- // Expected values based on the initialization pattern
- EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 0, 0)], 1.7, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 1, 0)], 3.4, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 0, 0)], 2.1, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 1, 0)], 4.2, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 0, 0)], 2.2, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 1, 0)], 4.4, 1e-6);
-}
-
-TYPED_TEST(HorizontalRotVertexAtmosTest, TestRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_e = this->nblks_e;
- constexpr int nblks_v = this->nblks_v;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &vert_edge_at = at<nproma, nblks_v, 6>;
- const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
- const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
-
- // Set up random number generators
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
-
- // Initialization with random values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- }
-
- // Set random edge indices
- for (int j = 0; j < 6; ++j) {
- this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
- this->vert_edge_blk[vert_edge_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 6; ++j) {
- this->geofac_rot[geofac_rot_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Initialize rot_vec to random values
- for (int k = 0; k < nlev; ++k) {
- this->rot_vec[rot_vec_at(i, k, 0)] = real_distrib(gen);
- }
- }
-
- // Call the rot_vertex_atmos function
- rot_vertex_atmos<TypeParam>(
- this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
- this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
- this->slev[0], this->elev[0], this->nproma, this->lacc, this->nlev,
- this->nblks_e, this->nblks_v);
-
- // Calculate reference values separately and verify results
- std::vector<TypeParam> ref_rot_vec(nproma * nlev * nblks_v, 0.0);
-
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_v_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jv = i_startidx; jv < i_endidx; ++jv) {
- ref_rot_vec[rot_vec_at(jv, jk, jb)] =
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
- this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
- this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
- this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
- this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
- this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
- this->geofac_rot[geofac_rot_at(jv, 5, jb)];
- }
- }
- }
-
- // Verify results
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->rot_vec[rot_vec_at(i, k, 0)],
- ref_rot_vec[rot_vec_at(i, k, 0)], 1e-5)
- << "Results differ at i=" << i << ", k=" << k;
- }
- }
-}
-
-TYPED_TEST_SUITE(HorizontalRotVertexRITest, ValueTypes);
-
-TYPED_TEST(HorizontalRotVertexRITest, TestSpecific) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_e = this->nblks_e;
- constexpr int nblks_v = this->nblks_v;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &vert_edge_at = at<nproma, nblks_v, 6>;
- const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
- const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
-
- // Initialization with specific values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
- }
-
- // Set edge indices to point to specific edges
- for (int j = 0; j < 6; ++j) {
- this->vert_edge_idx[vert_edge_at(i, 0, j)] = (i + j) % nproma;
- // All edges are in the same block for this test
- this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0;
- }
-
- // Geometric factors for rotation
- this->geofac_rot[geofac_rot_at(i, 0, 0)] = 0.3;
- this->geofac_rot[geofac_rot_at(i, 1, 0)] = 0.2;
- this->geofac_rot[geofac_rot_at(i, 2, 0)] = 0.1;
- this->geofac_rot[geofac_rot_at(i, 3, 0)] = 0.2;
- this->geofac_rot[geofac_rot_at(i, 4, 0)] = 0.1;
- this->geofac_rot[geofac_rot_at(i, 5, 0)] = 0.1;
-
- // Initialize rot_vec to zero
- for (int k = 0; k < nlev; ++k) {
- this->rot_vec[rot_vec_at(i, k, 0)] = 0.0;
- }
- }
-
- // Call the rot_vertex_ri function
- rot_vertex_ri<TypeParam>(
- this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
- this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
- this->slev[0], this->elev[0], this->nproma, this->lacc, this->acc_async,
- this->nlev, this->nblks_e, this->nblks_v);
-
- // Expected values based on the initialization pattern
- EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 0, 0)], 1.7, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 1, 0)], 3.4, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 0, 0)], 2.1, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 1, 0)], 4.2, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 0, 0)], 2.2, 1e-6);
- EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 1, 0)], 4.4, 1e-6);
-}
-
-TYPED_TEST(HorizontalRotVertexRITest, TestRandom) {
- constexpr int nproma = this->nproma;
- constexpr int nlev = this->nlev;
- constexpr int nblks_e = this->nblks_e;
- constexpr int nblks_v = this->nblks_v;
-
- const auto &vec_e_at = at<nproma, nlev, nblks_e>;
- const auto &vert_edge_at = at<nproma, nblks_v, 6>;
- const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
- const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
-
- // Set up random number generators
- std::random_device rd;
- std::mt19937 gen(rd());
- std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
- std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
-
- // Initialization with random values
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
- }
-
- // Set random edge indices
- for (int j = 0; j < 6; ++j) {
- this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
- this->vert_edge_blk[vert_edge_at(i, 0, j)] =
- 0; // Keep in same block for simplicity
- }
-
- // Random geometric factors
- for (int j = 0; j < 6; ++j) {
- this->geofac_rot[geofac_rot_at(i, j, 0)] = real_distrib(gen);
- }
-
- // Initialize rot_vec to random values
- for (int k = 0; k < nlev; ++k) {
- this->rot_vec[rot_vec_at(i, k, 0)] = real_distrib(gen);
- }
- }
-
- // Call the rot_vertex_ri function
- rot_vertex_ri<TypeParam>(
- this->vec_e.data(), this->vert_edge_idx.data(),
- this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
- this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
- this->slev[0], this->elev[0], this->nproma, this->lacc, this->acc_async,
- this->nlev, this->nblks_e, this->nblks_v);
-
- // Ensure computation is complete for both modes
- Kokkos::fence();
-
- // Calculate reference values separately and verify results
- std::vector<TypeParam> ref_rot_vec(nproma * nlev * nblks_v, 0.0);
-
- for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
- int i_startidx, i_endidx;
- get_indices_v_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
- this->i_startblk, this->i_endblk, i_startidx, i_endidx);
-
- for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
- for (int jv = i_startidx; jv < i_endidx; ++jv) {
- ref_rot_vec[rot_vec_at(jv, jk, jb)] =
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
- this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
- this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
- this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
- this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
- this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
- this->vec_e[vec_e_at(
- this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
- this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
- this->geofac_rot[geofac_rot_at(jv, 5, jb)];
- }
- }
- }
-
- // Verify results
- for (int i = 0; i < nproma; ++i) {
- for (int k = 0; k < nlev; ++k) {
- EXPECT_NEAR(this->rot_vec[rot_vec_at(i, k, 0)],
- ref_rot_vec[rot_vec_at(i, k, 0)], 1e-5)
- << "Results differ at i=" << i << ", k=" << k << ")";
- }
- }
-}
diff --git a/test/c/test_horizontal_recon.cpp b/test/c/test_horizontal_recon.cpp
new file mode 100644
index 0000000..8938a10
--- /dev/null
+++ b/test/c/test_horizontal_recon.cpp
@@ -0,0 +1,1199 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#include <iostream>
+#include <random>
+#include <vector>
+
+#include <Kokkos_Core.hpp>
+#include <gtest/gtest.h>
+#include <dim_helper.hpp>
+#include <horizontal/mo_lib_divrot.hpp>
+#include <support/mo_lib_loopindices.hpp>
+
+/// Enum class for the reconstruction method
+enum class ReconstructionMethod {
+ linear,
+ quadratic,
+ cubic,
+};
+
+/// Base test class for the horizontal reconstruct tests. Templated for the ValueType
+/// and ReconMethod for the reconstruction method.
+template <typename ValueType, int ReconMethod>
+class HorizontalReconTest : public ::testing::Test {
+protected:
+ // lsq_dim_c and lsq_dim_unk are instantiated in compile time.
+ static constexpr std::tuple<int, int>
+ init_lsq_dim(ReconstructionMethod method) {
+ switch (method) {
+ case ReconstructionMethod::linear:
+ return std::make_tuple(3, 2);
+ case ReconstructionMethod::quadratic:
+ return std::make_tuple(9, 5);
+ case ReconstructionMethod::cubic:
+ return std::make_tuple(9, 9);
+ }
+ }
+
+ // Constant dimensions.
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 1; // number of vertical levels
+ static constexpr int nblks_c = 1; // number of cell blocks (for p_e_in)
+ static constexpr std::tuple<int, int> lsq_dim =
+ init_lsq_dim(static_cast<ReconstructionMethod>(ReconMethod));
+ static constexpr int lsq_dim_c = std::get<0>(lsq_dim);
+ static constexpr int lsq_dim_unk = std::get<1>(lsq_dim);
+
+ // Parameter values.
+ int i_startblk = 0;
+ int i_endblk = nblks_c; // Test blocks [0 .. nblks_c-1]
+ int i_startidx_in = 0;
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ int slev = 0;
+ int elev = nlev; // Full vertical range (0 .. nlev-1)
+ int patch_id = 0;
+ bool lacc = false; // Not using ACC-specific behavior.
+ bool acc_async = false; // No asynchronous execution.
+ bool l_consv = true; // With conservative correction.
+ bool l_limited_area = true; // Limited area setup
+
+ std::vector<ValueType> p_cc;
+ std::vector<int> cell_neighbor_idx;
+ std::vector<int> cell_neighbor_blk;
+ std::vector<ValueType> lsq_qtmat_c;
+ std::vector<ValueType> lsq_rmat_rdiag_c;
+ std::vector<ValueType> lsq_rmat_utri_c;
+ std::vector<ValueType> lsq_moments;
+ std::vector<ValueType> lsq_pseudoinv;
+ std::vector<ValueType> p_coeff;
+
+ HorizontalReconTest() {
+ p_cc.resize(dim_combine(nproma, nlev, nblks_c));
+ cell_neighbor_idx.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
+ cell_neighbor_blk.resize(dim_combine(nproma, nblks_c, lsq_dim_c));
+ lsq_qtmat_c.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
+ lsq_rmat_rdiag_c.resize(dim_combine(nproma, lsq_dim_unk, nblks_c));
+ lsq_rmat_utri_c.resize(dim_combine(
+ nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c));
+ lsq_moments.resize(dim_combine(nproma, nblks_c, lsq_dim_unk));
+ lsq_pseudoinv.resize(dim_combine(nproma, lsq_dim_unk, lsq_dim_c, nblks_c));
+ p_coeff.resize(dim_combine(lsq_dim_unk + 1, nproma, nlev, nblks_c));
+ }
+};
+
+/// Test class for the horizontal tests. The reconstruction method is specified
+/// to linear.
+template <typename ValueType>
+class HorizontalReconLinearTest
+ : public HorizontalReconTest<ValueType, static_cast<int>(
+ ReconstructionMethod::linear)> {
+};
+
+/// Test class for the horizontal tests. The reconstruction method is specified
+/// to quadratic.
+template <typename ValueType>
+class HorizontalReconQuadraticTest
+ : public HorizontalReconTest<
+ ValueType, static_cast<int>(ReconstructionMethod::quadratic)> {};
+
+/// Test class for the horizontal tests. The reconstruction method is specified
+/// to cubic.
+template <typename ValueType>
+class HorizontalReconCubicTest
+ : public HorizontalReconTest<ValueType, static_cast<int>(
+ ReconstructionMethod::cubic)> {
+};
+
+/// ValueTypes which the divrot tests should run with
+typedef ::testing::Types<float, double> ValueTypes;
+
+TYPED_TEST_SUITE(HorizontalReconLinearTest, ValueTypes);
+
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCell) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = 2.0;
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = 2.0;
+ this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = 0.1;
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ }
+
+ // Test function
+ recon_lsq_cell_l<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.34, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 1.0, 1e-6);
+}
+
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = real_distrib(gen);
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = real_distrib(gen);
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 0, 0)] = real_distrib(gen);
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, 1, 0)] = real_distrib(gen);
+ this->lsq_rmat_utri_c[rmat_utri_at(i, 0, 0)] = real_distrib(gen);
+
+ this->lsq_moments[moments_at(i, 0, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, 1)] = real_distrib(gen);
+ }
+
+ // Test function
+ recon_lsq_cell_l<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ z_qt_times_d[0] = 0.0;
+ z_qt_times_d[1] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_qt_times_d[0] += this->lsq_qtmat_c[qtmat_at(jc, 0, i, jb)] * z_d[i];
+ z_qt_times_d[1] += this->lsq_qtmat_c[qtmat_at(jc, 1, i, jb)] * z_d[i];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] =
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 1, jb)] * z_qt_times_d[1];
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] =
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, 0, jb)] *
+ (z_qt_times_d[0] -
+ this->lsq_rmat_utri_c[rmat_utri_at(jc, 0, jb)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)]);
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 0)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 1)];
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 1)] = i;
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 2)] = i;
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ }
+
+ // Test function
+ recon_lsq_cell_l_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.65, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.5, 1e-6);
+}
+
+TYPED_TEST(HorizontalReconLinearTest, TestLsqCellSVDRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 3.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = real_distrib(gen);
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = real_distrib(gen);
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, 1)] = real_distrib(gen);
+ }
+
+ // Test function
+ recon_lsq_cell_l_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->l_consv, this->lacc, this->acc_async,
+ this->nblks_c, this->nlev, this->lsq_dim_unk, this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] =
+ this->lsq_pseudoinv[pseudoinv_at(jc, 1, 0, jb)] * z_d[0] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 1, 1, jb)] * z_d[1] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 1, 2, jb)] * z_d[2];
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] =
+ this->lsq_pseudoinv[pseudoinv_at(jc, 0, 0, jb)] * z_d[0] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 0, 1, jb)] * z_d[1] +
+ this->lsq_pseudoinv[pseudoinv_at(jc, 0, 2, jb)] * z_d[2];
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 0)] -
+ p_result[at<lsq_dim_unk + 1, nproma>(2, jc)] *
+ this->lsq_moments[moments_at(jc, jb, 1)];
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalReconQuadraticTest, ValueTypes);
+
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCell) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.5;
+ this->lsq_qtmat_c[qtmat_at(i, 2, j, 0)] = 0.2;
+ this->lsq_qtmat_c[qtmat_at(i, 3, j, 0)] = 0.7;
+ this->lsq_qtmat_c[qtmat_at(i, 4, j, 0)] = 1.3;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = 2.0;
+ }
+
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = 1.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ }
+
+ // Test function
+ recon_lsq_cell_q<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.24, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 3.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ -2.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 2.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ -3.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 2.6, 1e-6);
+}
+
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_qtmat_c[qtmat_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_q<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ z_qt_times_d[j] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_qt_times_d[j] +=
+ this->lsq_qtmat_c[qtmat_at(jc, j, i, jb)] * z_d[i];
+ }
+ }
+ int utri_id = 0;
+ for (int j = lsq_dim_unk; j > 0; --j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = z_qt_times_d[j - 1];
+ for (int k = j + 1; k <= lsq_dim_unk; ++k) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] -=
+ this->lsq_rmat_utri_c[rmat_utri_at(jc, utri_id++, jb)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(k, jc)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] *=
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, j - 1, jb)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.5;
+ this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.2;
+ this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
+ this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 1.3;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ }
+
+ // Test function
+ recon_lsq_cell_q_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ -0.56, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.5, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ 0.7, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 1.3, 1e-6);
+}
+
+TYPED_TEST(HorizontalReconQuadraticTest, TestLsqCellSVDRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization is done only for iblk = 0 and ilev = 0
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_pseudoinv[pseudoinv_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_q_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 1; j < lsq_dim_unk + 1; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
+ this->lsq_pseudoinv[pseudoinv_at(jc, j - 1, i, jb)] * z_d[i];
+ }
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+
+ // Check result
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(j, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(j, jc))], 1e-5)
+ << "For loop result fails for j = " << j << ", jc = " << jc;
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalReconCubicTest, ValueTypes);
+
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCell) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_qtmat_c[qtmat_at(i, 0, j, 0)] = 1.0;
+ this->lsq_qtmat_c[qtmat_at(i, 1, j, 0)] = 0.9;
+ this->lsq_qtmat_c[qtmat_at(i, 2, j, 0)] = 0.8;
+ this->lsq_qtmat_c[qtmat_at(i, 3, j, 0)] = 0.7;
+ this->lsq_qtmat_c[qtmat_at(i, 4, j, 0)] = 0.6;
+ this->lsq_qtmat_c[qtmat_at(i, 5, j, 0)] = 0.5;
+ this->lsq_qtmat_c[qtmat_at(i, 6, j, 0)] = 0.4;
+ this->lsq_qtmat_c[qtmat_at(i, 7, j, 0)] = 0.3;
+ this->lsq_qtmat_c[qtmat_at(i, 8, j, 0)] = 0.2;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = 2.0;
+ }
+
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = 1.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
+ this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
+ this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
+ this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
+ }
+
+ // Test function
+ recon_lsq_cell_c<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ 0.28, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
+ -0.2, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
+ 0.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &qtmat_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &rmat_rdiag_at = at<nproma, lsq_dim_unk, nblks_c>;
+ const auto &rmat_utri_at =
+ at<nproma, (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_qtmat_c[qtmat_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(i, j, 0)] = real_distrib(gen);
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+ for (int j = 0; j < (lsq_dim_unk * lsq_dim_unk - lsq_dim_unk) / 2; ++j) {
+ this->lsq_rmat_utri_c[rmat_utri_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_c<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_qtmat_c.data(),
+ this->lsq_rmat_rdiag_c.data(), this->lsq_rmat_utri_c.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ z_qt_times_d[j] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_qt_times_d[j] +=
+ this->lsq_qtmat_c[qtmat_at(jc, j, i, jb)] * z_d[i];
+ }
+ }
+ int utri_id = 0;
+ for (int j = lsq_dim_unk; j > 0; --j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = z_qt_times_d[j - 1];
+ for (int k = j + 1; k <= lsq_dim_unk; ++k) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] -=
+ this->lsq_rmat_utri_c[rmat_utri_at(jc, utri_id++, jb)] *
+ p_result[at<lsq_dim_unk + 1, nproma>(k, jc)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] *=
+ this->lsq_rmat_rdiag_c[rmat_rdiag_at(jc, j - 1, jb)];
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
+
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVD) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = (i + 1);
+
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, 0)] = (i + 1) % nproma;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, 0)] = 0;
+ for (int j = 1; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = i;
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->lsq_pseudoinv[pseudoinv_at(i, 0, j, 0)] = 1.0;
+ this->lsq_pseudoinv[pseudoinv_at(i, 1, j, 0)] = 0.9;
+ this->lsq_pseudoinv[pseudoinv_at(i, 2, j, 0)] = 0.8;
+ this->lsq_pseudoinv[pseudoinv_at(i, 3, j, 0)] = 0.7;
+ this->lsq_pseudoinv[pseudoinv_at(i, 4, j, 0)] = 0.6;
+ this->lsq_pseudoinv[pseudoinv_at(i, 5, j, 0)] = 0.5;
+ this->lsq_pseudoinv[pseudoinv_at(i, 6, j, 0)] = 0.4;
+ this->lsq_pseudoinv[pseudoinv_at(i, 7, j, 0)] = 0.3;
+ this->lsq_pseudoinv[pseudoinv_at(i, 8, j, 0)] = 0.2;
+ }
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = 0.0;
+ }
+
+ this->lsq_moments[moments_at(i, 0, 0)] = 0.2;
+ this->lsq_moments[moments_at(i, 0, 1)] = 0.3;
+ this->lsq_moments[moments_at(i, 0, 2)] = 0.4;
+ this->lsq_moments[moments_at(i, 0, 3)] = 0.5;
+ this->lsq_moments[moments_at(i, 0, 4)] = 0.6;
+ this->lsq_moments[moments_at(i, 0, 5)] = 0.7;
+ this->lsq_moments[moments_at(i, 0, 6)] = 0.8;
+ this->lsq_moments[moments_at(i, 0, 7)] = 0.9;
+ this->lsq_moments[moments_at(i, 0, 8)] = 1.0;
+ }
+
+ // Test function
+ recon_lsq_cell_c_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Check result
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(0, 0, 0, 0))],
+ -1.64, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(1, 0, 0, 0))],
+ 1.0, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(2, 0, 0, 0))],
+ 0.9, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(3, 0, 0, 0))],
+ 0.8, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(4, 0, 0, 0))],
+ 0.7, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(5, 0, 0, 0))],
+ 0.6, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(6, 0, 0, 0))],
+ 0.5, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(7, 0, 0, 0))],
+ 0.4, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(8, 0, 0, 0))],
+ 0.3, 1e-6);
+ EXPECT_NEAR(
+ this->p_coeff[(at<lsq_dim_unk + 1, nproma, nlev, nblks_c>(9, 0, 0, 0))],
+ 0.2, 1e-6);
+}
+
+TYPED_TEST(HorizontalReconCubicTest, TestLsqCellSVDRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_c = this->nblks_c;
+ constexpr int lsq_dim_c = this->lsq_dim_c;
+ constexpr int lsq_dim_unk = this->lsq_dim_unk;
+
+ const auto &p_cc_at = at<nproma, nlev, nblks_c>;
+ const auto &cell_neighbor_at = at<nproma, nblks_c, lsq_dim_c>;
+ const auto &pseudoinv_at = at<nproma, lsq_dim_unk, lsq_dim_c, nblks_c>;
+ const auto &p_coeff_at = at<lsq_dim_unk + 1, nproma, nlev, nblks_c>;
+ const auto &moments_at = at<nproma, nblks_c, lsq_dim_unk>;
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(0.0, 1.0);
+
+ // Initialization
+ for (int i = 0; i < nproma; ++i) {
+ this->p_cc[p_cc_at(i, 0, 0)] = real_distrib(gen);
+
+ for (int j = 0; j < lsq_dim_c; ++j) {
+ this->cell_neighbor_idx[cell_neighbor_at(i, 0, j)] = int_distrib(gen);
+ this->cell_neighbor_blk[cell_neighbor_at(i, 0, j)] = 0;
+ }
+
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ for (int k = 0; k < lsq_dim_c; ++k) {
+ this->lsq_pseudoinv[pseudoinv_at(i, j, k, 0)] = real_distrib(gen);
+ }
+ this->lsq_moments[moments_at(i, 0, j)] = real_distrib(gen);
+ }
+
+ for (int j = 0; j < lsq_dim_unk + 1; ++j) {
+ this->p_coeff[p_coeff_at(j, i, 0, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Test function
+ recon_lsq_cell_c_svd<TypeParam>(
+ this->p_cc.data(), this->cell_neighbor_idx.data(),
+ this->cell_neighbor_blk.data(), this->lsq_pseudoinv.data(),
+ this->lsq_moments.data(), this->p_coeff.data(), this->i_startblk,
+ this->i_endblk, this->i_startidx_in, this->i_endidx_in, this->slev,
+ this->elev, this->nproma, this->patch_id, this->l_limited_area,
+ this->lacc, this->nblks_c, this->nlev, this->lsq_dim_unk,
+ this->lsq_dim_c);
+
+ // Compute reference result
+ std::vector<TypeParam> z_d(lsq_dim_c);
+ std::vector<TypeParam> z_qt_times_d(lsq_dim_unk);
+ std::vector<TypeParam> p_result((lsq_dim_unk + 1) * nproma);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_c_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+ for (int jk = this->slev; jk < this->elev; ++jk) {
+ for (int jc = i_startidx; jc < i_endidx; ++jc) {
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ z_d[i] = this->p_cc[p_cc_at(
+ this->cell_neighbor_idx[cell_neighbor_at(jc, jb, i)], jk,
+ this->cell_neighbor_blk[cell_neighbor_at(jc, jb, i)])] -
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ }
+ for (int j = 1; j < lsq_dim_unk + 1; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] = 0.0;
+ for (int i = 0; i < lsq_dim_c; ++i) {
+ p_result[at<lsq_dim_unk + 1, nproma>(j, jc)] +=
+ this->lsq_pseudoinv[pseudoinv_at(jc, j - 1, i, jb)] * z_d[i];
+ }
+ }
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] =
+ this->p_cc[p_cc_at(jc, jk, jb)];
+ for (int j = 0; j < lsq_dim_unk; ++j) {
+ p_result[at<lsq_dim_unk + 1, nproma>(0, jc)] -=
+ p_result[at<lsq_dim_unk + 1, nproma>(j + 1, jc)] *
+ this->lsq_moments[moments_at(jc, jb, j)];
+ }
+ }
+ }
+ }
+ // Check result
+ for (int i = 0; i < lsq_dim_unk + 1; ++i) {
+ for (int jc = 0; jc < nproma; ++jc) {
+ EXPECT_NEAR(this->p_coeff[(p_coeff_at(i, jc, 0, 0))],
+ p_result[(at<lsq_dim_unk + 1, nproma>(i, jc))], 1e-5)
+ << "For loop result fails for i = " << i << ", jc = " << jc;
+ }
+ }
+}
diff --git a/test/c/test_horizontal_rot.cpp b/test/c/test_horizontal_rot.cpp
new file mode 100644
index 0000000..68e8024
--- /dev/null
+++ b/test/c/test_horizontal_rot.cpp
@@ -0,0 +1,378 @@
+// ICON
+//
+// ---------------------------------------------------------------
+// Copyright (C) 2004-2025, DWD, MPI-M, DKRZ, KIT, ETH, MeteoSwiss
+// Contact information: icon-model.org
+//
+// See AUTHORS.TXT for a list of authors
+// See LICENSES/ for license information
+// SPDX-License-Identifier: BSD-3-Clause
+// ---------------------------------------------------------------
+
+#include <iostream>
+#include <random>
+#include <vector>
+
+#include <Kokkos_Core.hpp>
+#include <gtest/gtest.h>
+#include <dim_helper.hpp>
+#include <horizontal/mo_lib_divrot.hpp>
+#include <support/mo_lib_loopindices.hpp>
+
+/// Test class for the horizontal rotation tests. Templated for the ValueType.
+template <typename ValueType>
+class HorizontalRotVertexTest : public ::testing::Test {
+protected:
+ static constexpr int nproma = 3; // inner loop length
+ static constexpr int nlev = 2; // number of vertical levels
+ static constexpr int nblks_e = 1; // number of edge blocks
+ static constexpr int nblks_v = 1; // number of vertex blocks
+ static constexpr int dim4d = 2; // 4th dimension size
+
+ int i_startblk = 0;
+ int i_endblk = nblks_v; // Test blocks [0 .. nblks_v-1]
+ int i_startidx_in = 0;
+ int i_endidx_in = nproma; // Full range: 0 .. nproma-1
+ std::vector<int> slev;
+ std::vector<int> elev;
+ bool lacc = false; // Not using ACC-specific behavior.
+ bool acc_async = false; // Not using ACC-specific behavior.
+
+ std::vector<ValueType> vec_e;
+ std::vector<int> vert_edge_idx;
+ std::vector<int> vert_edge_blk;
+ std::vector<ValueType> geofac_rot;
+ std::vector<ValueType> rot_vec;
+ std::vector<ValueType> f4din;
+ std::vector<ValueType> f4dout;
+
+ HorizontalRotVertexTest() {
+ slev.resize(dim4d, 0);
+ elev.resize(dim4d, nlev); // Full vertical range (0 .. nlev-1)
+
+ vec_e.resize(dim_combine(nproma, nlev, nblks_e));
+ vert_edge_idx.resize(dim_combine(nproma, nblks_v, 6));
+ vert_edge_blk.resize(dim_combine(nproma, nblks_v, 6));
+ geofac_rot.resize(dim_combine(nproma, 6, nblks_v));
+ rot_vec.resize(dim_combine(nproma, nlev, nblks_v));
+ f4din.resize(dim_combine(nproma, nlev, nblks_e, dim4d));
+ f4dout.resize(dim_combine(nproma, nlev, nblks_v, dim4d));
+ }
+};
+
+/// ValueTypes which the divrot tests should run with
+typedef ::testing::Types<float, double> ValueTypes;
+
+TYPED_TEST_SUITE(HorizontalRotVertexTest, ValueTypes);
+
+TYPED_TEST(HorizontalRotVertexTest, TestRotVertexAtmosSpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ }
+
+ // Set edge indices to point to specific edges
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = (i + j) % nproma;
+ // All edges are in the same block for this test
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors for rotation
+ this->geofac_rot[geofac_rot_at(i, 0, 0)] = 0.3;
+ this->geofac_rot[geofac_rot_at(i, 1, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 2, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 3, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 4, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 5, 0)] = 0.1;
+
+ // Initialize rot_vec to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the rot_vertex_atmos function
+ rot_vertex_atmos<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_e, this->nblks_v);
+
+ // Expected values based on the initialization pattern
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 1, 0)], 4.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalRotVertexTest, TestRotVertexAtmosRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 6; ++j) {
+ this->geofac_rot[geofac_rot_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize rot_vec to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the rot_vertex_atmos function
+ rot_vertex_atmos<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->nlev,
+ this->nblks_e, this->nblks_v);
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_rot_vec(nproma * nlev * nblks_v, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_v_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jv = i_startidx; jv < i_endidx; ++jv) {
+ ref_rot_vec[rot_vec_at(jv, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
+ this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
+ this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
+ this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
+ this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
+ this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
+ this->geofac_rot[geofac_rot_at(jv, 5, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(i, k, 0)],
+ ref_rot_vec[rot_vec_at(i, k, 0)], 1e-5)
+ << "Results differ at i=" << i << ", k=" << k;
+ }
+ }
+}
+
+TYPED_TEST_SUITE(HorizontalRotVertexTest, ValueTypes);
+
+TYPED_TEST(HorizontalRotVertexTest, TestRotVertexRISpecific) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Initialization with specific values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = (i + 1) * (k + 1); // Simple pattern
+ }
+
+ // Set edge indices to point to specific edges
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = (i + j) % nproma;
+ // All edges are in the same block for this test
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] = 0;
+ }
+
+ // Geometric factors for rotation
+ this->geofac_rot[geofac_rot_at(i, 0, 0)] = 0.3;
+ this->geofac_rot[geofac_rot_at(i, 1, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 2, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 3, 0)] = 0.2;
+ this->geofac_rot[geofac_rot_at(i, 4, 0)] = 0.1;
+ this->geofac_rot[geofac_rot_at(i, 5, 0)] = 0.1;
+
+ // Initialize rot_vec to zero
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = 0.0;
+ }
+ }
+
+ // Call the rot_vertex_ri function
+ rot_vertex_ri<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->acc_async,
+ this->nlev, this->nblks_e, this->nblks_v);
+
+ // Expected values based on the initialization pattern
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 0, 0)], 1.7, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(0, 1, 0)], 3.4, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 0, 0)], 2.1, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(1, 1, 0)], 4.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 0, 0)], 2.2, 1e-6);
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(2, 1, 0)], 4.4, 1e-6);
+}
+
+TYPED_TEST(HorizontalRotVertexTest, TestRotVertexRIRandom) {
+ constexpr int nproma = this->nproma;
+ constexpr int nlev = this->nlev;
+ constexpr int nblks_e = this->nblks_e;
+ constexpr int nblks_v = this->nblks_v;
+
+ const auto &vec_e_at = at<nproma, nlev, nblks_e>;
+ const auto &vert_edge_at = at<nproma, nblks_v, 6>;
+ const auto &geofac_rot_at = at<nproma, 6, nblks_v>;
+ const auto &rot_vec_at = at<nproma, nlev, nblks_v>;
+
+ // Set up random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<int> int_distrib(0, nproma - 1);
+ std::uniform_real_distribution<TypeParam> real_distrib(-10.0, 10.0);
+
+ // Initialization with random values
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ this->vec_e[vec_e_at(i, k, 0)] = real_distrib(gen);
+ }
+
+ // Set random edge indices
+ for (int j = 0; j < 6; ++j) {
+ this->vert_edge_idx[vert_edge_at(i, 0, j)] = int_distrib(gen);
+ this->vert_edge_blk[vert_edge_at(i, 0, j)] =
+ 0; // Keep in same block for simplicity
+ }
+
+ // Random geometric factors
+ for (int j = 0; j < 6; ++j) {
+ this->geofac_rot[geofac_rot_at(i, j, 0)] = real_distrib(gen);
+ }
+
+ // Initialize rot_vec to random values
+ for (int k = 0; k < nlev; ++k) {
+ this->rot_vec[rot_vec_at(i, k, 0)] = real_distrib(gen);
+ }
+ }
+
+ // Call the rot_vertex_ri function
+ rot_vertex_ri<TypeParam>(
+ this->vec_e.data(), this->vert_edge_idx.data(),
+ this->vert_edge_blk.data(), this->geofac_rot.data(), this->rot_vec.data(),
+ this->i_startblk, this->i_endblk, this->i_startidx_in, this->i_endidx_in,
+ this->slev[0], this->elev[0], this->nproma, this->lacc, this->acc_async,
+ this->nlev, this->nblks_e, this->nblks_v);
+
+ // Ensure computation is complete for both modes
+ Kokkos::fence();
+
+ // Calculate reference values separately and verify results
+ std::vector<TypeParam> ref_rot_vec(nproma * nlev * nblks_v, 0.0);
+
+ for (int jb = this->i_startblk; jb < this->i_endblk; ++jb) {
+ int i_startidx, i_endidx;
+ get_indices_v_lib(this->i_startidx_in, this->i_endidx_in, nproma, jb,
+ this->i_startblk, this->i_endblk, i_startidx, i_endidx);
+
+ for (int jk = this->slev[0]; jk < this->elev[0]; ++jk) {
+ for (int jv = i_startidx; jv < i_endidx; ++jv) {
+ ref_rot_vec[rot_vec_at(jv, jk, jb)] =
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 0)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 0)])] *
+ this->geofac_rot[geofac_rot_at(jv, 0, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 1)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 1)])] *
+ this->geofac_rot[geofac_rot_at(jv, 1, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 2)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 2)])] *
+ this->geofac_rot[geofac_rot_at(jv, 2, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 3)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 3)])] *
+ this->geofac_rot[geofac_rot_at(jv, 3, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 4)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 4)])] *
+ this->geofac_rot[geofac_rot_at(jv, 4, jb)] +
+ this->vec_e[vec_e_at(
+ this->vert_edge_idx[vert_edge_at(jv, jb, 5)], jk,
+ this->vert_edge_blk[vert_edge_at(jv, jb, 5)])] *
+ this->geofac_rot[geofac_rot_at(jv, 5, jb)];
+ }
+ }
+ }
+
+ // Verify results
+ for (int i = 0; i < nproma; ++i) {
+ for (int k = 0; k < nlev; ++k) {
+ EXPECT_NEAR(this->rot_vec[rot_vec_at(i, k, 0)],
+ ref_rot_vec[rot_vec_at(i, k, 0)], 1e-5)
+ << "Results differ at i=" << i << ", k=" << k << ")";
+ }
+ }
+}
+
--
GitLab