Fix CMake and CUDA compiler warnings (#3859)

Description of changes:
- fix several issues in the CMake logic
- fix function visibility issues in CUDA code (fixes #3797)
- document the CPU and GPU implementations of `specfunc`

CMakeLists.txt

@@ -396,7 +396,7 @@ target_compile_options(
             -Wno-unused-variable
             -Wno-unused-parameter
             -Wno-missing-braces
-            $<$<CXX_COMPILER_ID:GNU>:-Wno-clobbered;-Wno-cast-function-type>
+            $<$<CXX_COMPILER_ID:GNU>:-Wno-clobbered>
             $<$<CXX_COMPILER_ID:Intel>:-wd592>)
 
 # disable more warnings from -Wextra
@@ -423,6 +423,10 @@ if(NOT CMAKE_CXX_COMPILER_ID STREQUAL "GNU" OR CMAKE_CXX_COMPILER_VERSION
   # older GCC versions don't support -Wno-pedantic which we need in src/python
   target_compile_options(cxx_interface INTERFACE -pedantic)
 endif()
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION
+                                            VERSION_GREATER_EQUAL "8.1.0")
+  target_compile_options(cxx_interface INTERFACE -Wno-cast-function-type)
+endif()
 if(CMAKE_CXX_COMPILER_ID STREQUAL "Intel" AND CMAKE_CXX_COMPILER_VERSION
                                               VERSION_LESS "16.0")
   # workaround for compiler crash related to decltype() and variadic template

doc/doxygen/bibliography.bib

@@ -412,6 +412,15 @@ @article{martys99a
   doi = {10.1103/PhysRevE.59.3733},
 }
 
+@book{moshier89a,
+  title={{Methods and Programs for Mathematical Functions}},
+  author={Moshier, Stephen Lloyd Baluk},
+  isbn={9780470216095},
+  series={Ellis Horwood Series in Mathematics and Its Applications},
+  year={1989},
+  publisher={Ellis Horwood},
+}
+
 @Article{moss96a,
 author={Moss, G. P.},
 journal={Pure and Applied Chemistry},

doc/sphinx/installation.rst

@@ -612,7 +612,7 @@ each variant having different activated features, and for as many
 platforms as you want.
 
 Once you've run ``ccmake``, you can list the configured variables with
-``cmake -LAH -N | less`` (uses a pager) or with ``ccmake ..`` and pressing
+``cmake -LAH -N .. | less`` (uses a pager) or with ``ccmake ..`` and pressing
 key ``t`` to toggle the advanced mode on (uses the curses interface).
 
 **Example:**

maintainer/CI/dox_warnings.sh

@@ -19,7 +19,7 @@
 
 # This script generates a summary of all Doxygen warnings that require fixing.
 
-DOXYGEN=$(cmake -LA -N | grep "DOXYGEN_EXECUTABLE:FILEPATH" | cut -d'=' -f2-)
+DOXYGEN=$(cmake -LA -N . | grep "DOXYGEN_EXECUTABLE:FILEPATH" | cut -d'=' -f2-)
 if ! hash "${DOXYGEN}" 2>/dev/null; then
     echo "No doxygen found."
     exit 2

src/core/actor/mmm-common_cuda.hpp

@@ -17,7 +17,7 @@
  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 #include "cuda_utils.hpp"
-#include "electrostatics_magnetostatics/mmm-common.hpp"
+#include "electrostatics_magnetostatics/mmm-modpsi.hpp"
 #include "specfunc_cuda.hpp"
 
 // order hardcoded. mmm1d_recalcTables() typically does order less than 30.

src/core/actor/specfunc_cuda.hpp

@@ -16,12 +16,40 @@
  * You should have received a copy of the GNU General Public License
  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
+
+/* Original gsl header
+ * specfunc/bessel_K0.cpp
+ *
+ * Copyright (C) 1996, 1997, 1998, 1999, 2000 Gerard Jungman
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or (at
+ * your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+/* Original Author: G. Jungman */
+
+/** \file
+ *  This file contains implementations for the modified Bessel functions of
+ *  first and second kind. The implementations are based on the GSL code (see
+ *  the original GSL header above) and are duplicated from \ref specfunc.cpp.
+ */
 #include "config.hpp"
 
 const mmm1dgpu_real M_LN2f = M_LN2;
 
-// Adapted from specfunc.c and polynom.h
-
+/** @name Chebyshev expansions based on SLATEC bk0(), bk0e() */
+/*@{*/
 __constant__ static mmm1dgpu_real bk0_data[11] = {
     -.5 - 0.03532739323390276872, 0.3442898999246284869,
     0.03597993651536150163,       0.00126461541144692592,
@@ -47,14 +75,20 @@ __constant__ static mmm1dgpu_real ak02_data[14] = {
     -0.00000000000020743,      0.00000000000001925,  -0.00000000000000192,
     0.00000000000000020,       -0.00000000000000002};
 __constant__ static int ak02_size = 13;
+/*@}*/
 
+/** @name Chebyshev expansions based on SLATEC besi0() */
+/*@{*/
 __constant__ static mmm1dgpu_real bi0_data[12] = {
     5.5 - .07660547252839144951, 1.92733795399380827000, .22826445869203013390,
     .01304891466707290428,       .00043442709008164874,  .00000942265768600193,
     .00000014340062895106,       .00000000161384906966,  .00000000001396650044,
     .00000000000009579451,       .00000000000000053339,  .00000000000000000245};
 __constant__ static int bi0_size = 12;
+/*@}*/
 
+/** @name Chebyshev expansions based on SLATEC besk1(), besk1e() */
+/*@{*/
 __constant__ static mmm1dgpu_real bk1_data[11] = {
     1.5 + 0.0253002273389477705, -0.3531559607765448760, -0.1226111808226571480,
     -0.0069757238596398643,      -0.0001730288957513052, -0.0000024334061415659,
@@ -78,13 +112,17 @@ __constant__ static mmm1dgpu_real ak12_data[14] = {
     0.00000000000023720,       -0.00000000000002176, 0.00000000000000215,
     -0.00000000000000022,      0.00000000000000002};
 __constant__ static int ak12_size = 14;
+/*@}*/
 
+/** @name Chebyshev expansions based on SLATEC besi1(), besi1e() */
+/*@{*/
 __constant__ static mmm1dgpu_real bi1_data[11] = {
     1.75 - 0.001971713261099859, 0.407348876675464810, 0.034838994299959456,
     0.001545394556300123,        0.000041888521098377, 0.000000764902676483,
     0.000000010042493924,        0.000000000099322077, 0.000000000000766380,
     0.000000000000004741,        0.000000000000000024};
 __constant__ static int bi1_size = 11;
+/*@}*/
 
 __device__ mmm1dgpu_real evaluateAsChebychevSeriesAt(mmm1dgpu_real const *c,
                                                      int n, mmm1dgpu_real x) {

src/core/cuda_init.cpp

@@ -61,14 +61,15 @@ std::vector<EspressoGpuDevice> cuda_gather_gpus() {
 
   /* Truncate to 63 chars to fit struct. */
   if (strlen(proc_name) > 63)
-    proc_name[63] = 0;
+    proc_name[63] = '\0';
 
   for (int i = 0; i < n_gpus; ++i) {
     /* Check if device has at least minimum compute capability */
     if (cuda_check_gpu(i) == ES_OK) {
       EspressoGpuDevice device;
       if (cuda_get_device_props(i, device) == ES_OK) {
         strncpy(device.proc_name, proc_name, 64);
+        device.proc_name[63] = '\0';
         device.node = this_node;
         devices.push_back(device);
       }

src/core/electrostatics_magnetostatics/CMakeLists.txt

@@ -6,7 +6,7 @@ target_sources(
           ${CMAKE_CURRENT_SOURCE_DIR}/magnetic_non_p3m_methods.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/mdlc_correction.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/mmm1d.cpp
-          ${CMAKE_CURRENT_SOURCE_DIR}/mmm-common.cpp
+          ${CMAKE_CURRENT_SOURCE_DIR}/mmm-modpsi.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/p3m-common.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/p3m_send_mesh.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/p3m.cpp

src/core/electrostatics_magnetostatics/mmm-common.hpp

@@ -30,20 +30,12 @@
 #ifndef MMM_COMMON_H
 #define MMM_COMMON_H
 
+#include "mmm-modpsi.hpp"
+
 #include "specfunc.hpp"
 
 #include <vector>
 
-/** \name Math constants */
-/*@{*/
-#define C_2PI (2 * M_PI)
-#define C_GAMMA (0.57721566490153286060651209008)
-/*@}*/
-
-/** Table of the Taylor expansions of the modified polygamma functions */
-extern std::vector<std::vector<double>> modPsi;
-extern int n_modPsi;
-
 /** Modified polygamma for even order 2*n, n >= 0 */
 inline double mod_psi_even(int n, double x) {
   return evaluateAsTaylorSeriesAt(modPsi[2 * n], x * x);
@@ -54,7 +46,4 @@ inline double mod_psi_odd(int n, double x) {
   return x * evaluateAsTaylorSeriesAt(modPsi[2 * n + 1], x * x);
 }
 
-/** Create the both the even and odd polygamma functions up to order 2*n */
-void create_mod_psi_up_to(int n);
-
 #endif

src/core/electrostatics_magnetostatics/mmm-common.cpp → src/core/electrostatics_magnetostatics/mmm-modpsi.cpp

@@ -22,6 +22,7 @@
 #include "config.hpp"
 
 #include "mmm-common.hpp"
+#include "mmm-modpsi.hpp"
 
 #include <cmath>
 

src/core/electrostatics_magnetostatics/mmm-modpsi.hpp

@@ -0,0 +1,45 @@
+/*
+ * Copyright (C) 2010-2019 The ESPResSo project
+ * Copyright (C) 2002,2003,2004,2005,2006,2007,2008,2009,2010
+ *   Max-Planck-Institute for Polymer Research, Theory Group
+ *
+ * This file is part of ESPResSo.
+ *
+ * ESPResSo is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * ESPResSo is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+/** \file
+ *  Common parts of the MMM family of methods for the electrostatic
+ *  interaction, MMM1D, MMM2D and ELC. This file contains the code for the
+ *  CPU and GPU implementations.
+ */
+
+#ifndef MMM_BASE_H
+#define MMM_BASE_H
+
+#include <vector>
+
+/** \name Math constants */
+/*@{*/
+#define C_2PI (2 * M_PI)
+#define C_GAMMA (0.57721566490153286060651209008)
+/*@}*/
+
+/** Table of the Taylor expansions of the modified polygamma functions */
+extern std::vector<std::vector<double>> modPsi;
+extern int n_modPsi;
+
+/** Create both the even and odd polygamma functions up to order 2*n */
+void create_mod_psi_up_to(int n);
+
+#endif

src/core/electrostatics_magnetostatics/specfunc.cpp

@@ -266,8 +266,8 @@ static double ai12_cs[22] = {
     -0.00000000000000003};
 /*@}*/
 
-/** Coefficients for Maclaurin summation in hzeta().
- *  B_{2j}/(2j)!
+/** Coefficients for Maclaurin summation in hzeta(). Evaluated as inverse
+ *  numbers, i.e. @f$ \displaystyle\frac{B_{2j}}{(2j)!} @f$.
  */
 static double const hzeta_c[15] = {
     1.00000000000000000000000000000,     0.083333333333333333333333333333,
@@ -291,8 +291,8 @@ double hzeta(double s, double q) {
     double p3 = pow(q / (2.0 + q), s);
     return p1 * (1.0 + p2 + p3);
   }
-  /* Euler-Maclaurin summation formula
-   * [Moshier, p. 400, with several typo corrections]
+  /** Euler-Maclaurin summation formula from @cite moshier89a p. 400, with
+   *  several typo corrections.
    */
   auto const pmax = pow(kmax + q, -s);
   auto scp = s;

src/core/electrostatics_magnetostatics/specfunc.hpp

@@ -25,12 +25,12 @@
  *  implementations are based on the GSL code (see \ref specfunc.cpp for the
  *  original GSL header).
  *
- *  The Hurwitz zeta function is evaluated using the Euler-MacLaurin summation
+ *  The Hurwitz zeta function is evaluated using the Euler-Maclaurin summation
  *  formula, the Bessel functions are evaluated using several different
  *  Chebychev expansions. Both achieve a precision of nearly machine precision,
  *  which is no problem for the Hurwitz zeta function, which is only used when
  *  determining the coefficients for the modified polygamma functions (see \ref
- *  mmm-common.hpp "mmm-common.h"). However, the Bessel functions are actually
+ *  mmm-common.hpp). However, the Bessel functions are actually
  *  used in the near formula of MMM2D, which is therefore slightly slower than
  *  necessary. On the other hand, the number of terms in the Bessel sum is
  *  quite small normally, so that a less precise version will probably not
@@ -84,7 +84,8 @@ double LPK1(double x);
  */
 void LPK01(double x, double *K0, double *K1);
 
-/** evaluate the polynomial interpreted as a Taylor series via the Horner scheme
+/** Evaluate the polynomial interpreted as a Taylor series via the
+ *  Horner scheme.
  */
 inline double evaluateAsTaylorSeriesAt(Utils::Span<const double> series,
                                        double x) {
@@ -97,8 +98,8 @@ inline double evaluateAsTaylorSeriesAt(Utils::Span<const double> series,
   return r;
 }
 
-/** evaluate the polynomial interpreted as a Chebychev series. Requires a series
- *  with at least three coefficients, i.e. no linear approximations!
+/** Evaluate the polynomial interpreted as a Chebychev series. Requires a
+ *  series with at least three coefficients, i.e. no linear approximations!
  */
 inline double evaluateAsChebychevSeriesAt(Utils::Span<const double> series,
                                           double x) {

src/core/grid_based_algorithms/lbgpu.hpp

@@ -45,16 +45,6 @@ typedef double lbForceFloat;
 typedef float lbForceFloat;
 #endif
 
-class LB_particle_allocation_state {
-  bool m_invalid = true;
-
-public:
-  bool operator()() { return m_invalid; }
-  void invalidate() { m_invalid = true; }
-
-  void validate() { m_invalid = false; }
-};
-
 /**-------------------------------------------------------------------------*/
 /** Parameters for the lattice Boltzmann system for GPU. */
 struct LB_parameters_gpu {
@@ -160,7 +150,6 @@ typedef struct {
 /** Switch indicating momentum exchange between particles and fluid */
 extern LB_parameters_gpu lbpar_gpu;
 extern std::vector<LB_rho_v_pi_gpu> host_values;
-extern LB_particle_allocation_state lb_reinit_particles_gpu;
 #ifdef ELECTROKINETICS
 extern LB_node_force_density_gpu node_f;
 extern bool ek_initialized;