Remove macros

CMakeLists.txt

@@ -26,6 +26,9 @@ endif()
 # Set compiler flags
 set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -Wall -Wextra -pedantic -Wno-unused -Wno-psabi -Wfloat-conversion")
 set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -O2 -mtune=generic -Wno-psabi")
+if (CMAKE_COMPILER_IS_GNUCXX)
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} --param max-inline-insns-single=1500")
+endif()
 
 # Set coverage compiler flags - must come before any targets are defined
 if(GMGPOLAR_ENABLE_COVERAGE)

include/common/global_definitions.h

@@ -83,34 +83,44 @@ enum class BetaCoeff
 /* Mumps - Constant Definitions */
 /* ---------------------------- */
 #ifdef GMGPOLAR_USE_MUMPS
-    /* Mumps macro s.t. indices match documentation */
-    #define ICNTL(I) icntl[(I) - 1]
-    #define CNTL(I) cntl[(I) - 1]
-    #define INFOG(I) infog[(I) - 1]
-
-    #define USE_COMM_WORLD -987654
-    #define PAR_NOT_PARALLEL 0
-    #define PAR_PARALLEL 1
-
-    #define JOB_INIT -1
-    #define JOB_END -2
-    #define JOB_REMOVE_SAVED_DATA -3
-    #define JOB_FREE_INTERNAL_DATA -4
-    #define JOB_SUPPRESS_OOC_FILES -200
-
-    #define JOB_ANALYSIS_PHASE 1
-    #define JOB_FACTORIZATION_PHASE 2
-    #define JOB_COMPUTE_SOLUTION 3
-    #define JOB_ANALYSIS_AND_FACTORIZATION 4
-    #define JOB_FACTORIZATION_AND_SOLUTION 5
-    #define JOB_ANALYSIS_FACTORIZATION_SOLUTION 6
-    #define JOB_SAVE_INTERNAL_DATA 7
-    #define JOB_RESTORE_INTERNAL_DATA 8
-    #define JOB_DISTRIBUTE_RHS 9
-
-    #define SYM_UNSYMMETRIC 0
-    #define SYM_POSITIVE_DEFINITE 1
-    #define SYM_GENERAL_SYMMETRIC 2
+    #include "dmumps_c.h"
+/* Mumps inline functions s.t. indices match documentation */
+inline int& ICNTL(DMUMPS_STRUC_C& mumps_solver, int I)
+{
+    return mumps_solver.icntl[(I)-1];
+}
+inline double& CNTL(DMUMPS_STRUC_C& mumps_solver, int I)
+{
+    return mumps_solver.cntl[(I)-1];
+}
+inline int& INFOG(DMUMPS_STRUC_C& mumps_solver, int I)
+{
+    return mumps_solver.infog[(I)-1];
+}
+
+constexpr int USE_COMM_WORLD   = -987654;
+constexpr int PAR_NOT_PARALLEL = 0;
+constexpr int PAR_PARALLEL     = 1;
+
+constexpr int JOB_INIT               = -1;
+constexpr int JOB_END                = -2;
+constexpr int JOB_REMOVE_SAVED_DATA  = -3;
+constexpr int JOB_FREE_INTERNAL_DATA = -4;
+constexpr int JOB_SUPPRESS_OOC_FILES = -200;
+
+constexpr int JOB_ANALYSIS_PHASE                  = 1;
+constexpr int JOB_FACTORIZATION_PHASE             = 2;
+constexpr int JOB_COMPUTE_SOLUTION                = 3;
+constexpr int JOB_ANALYSIS_AND_FACTORIZATION      = 4;
+constexpr int JOB_FACTORIZATION_AND_SOLUTION      = 5;
+constexpr int JOB_ANALYSIS_FACTORIZATION_SOLUTION = 6;
+constexpr int JOB_SAVE_INTERNAL_DATA              = 7;
+constexpr int JOB_RESTORE_INTERNAL_DATA           = 8;
+constexpr int JOB_DISTRIBUTE_RHS                  = 9;
+
+constexpr int SYM_UNSYMMETRIC       = 0;
+constexpr int SYM_POSITIVE_DEFINITE = 1;
+constexpr int SYM_GENERAL_SYMMETRIC = 2;
 #endif
 
 // --------------------------------------- //
@@ -148,4 +158,4 @@ enum class BetaCoeff
     #define LIKWID_START(marker)
     #define LIKWID_STOP(marker)
     #define LIKWID_CLOSE()
-#endif
+#endif

src/DirectSolver/DirectSolver-COO-MUMPS-Give/initializeMumps.cpp

@@ -23,68 +23,68 @@ void DirectSolver_COO_MUMPS_Give::initializeMumpsSolver(DMUMPS_STRUC_C& mumps_so
     mumps_solver.comm_fortran = USE_COMM_WORLD;
     dmumps_c(&mumps_solver);
 
-    mumps_solver.ICNTL(1) = 0; // Output stream for error messages.
-    mumps_solver.ICNTL(2) = 0; // Output stream for diagnostic printing and statistics local to each MPI process.
-    mumps_solver.ICNTL(3) = 0; // Output stream for global information, collected on the host
-    mumps_solver.ICNTL(4) = 0; // Level of printing for error, warning, and diagnostic messages.
-    mumps_solver.ICNTL(5) = 0; // Controls the matrix input format
-    mumps_solver.ICNTL(6) = 7; // Permutes the matrix to a zero-free diagonal and/or scale the matrix
-    mumps_solver.ICNTL(7) =
+    ICNTL(mumps_solver, 1) = 0; // Output stream for error messages.
+    ICNTL(mumps_solver, 2) = 0; // Output stream for diagnostic printing and statistics local to each MPI process.
+    ICNTL(mumps_solver, 3) = 0; // Output stream for global information, collected on the host
+    ICNTL(mumps_solver, 4) = 0; // Level of printing for error, warning, and diagnostic messages.
+    ICNTL(mumps_solver, 5) = 0; // Controls the matrix input format
+    ICNTL(mumps_solver, 6) = 7; // Permutes the matrix to a zero-free diagonal and/or scale the matrix
+    ICNTL(mumps_solver, 7) =
         5; // Computes a symmetric permutation (ordering) to determine the pivot order to be used for the factorization in case of sequential analysis
-    mumps_solver.ICNTL(8)  = 77; // Describes the scaling strategy
-    mumps_solver.ICNTL(9)  = 1; // Computes the solution using A or A^T
-    mumps_solver.ICNTL(10) = 0; // Applies the iterative refinement to the computed solution
-    mumps_solver.ICNTL(11) = 0; // Computes statistics related to an error analysis of the linear system solved
-    mumps_solver.ICNTL(12) = 0; // Defines an ordering strategy for symmetric matrices and is used
-    mumps_solver.ICNTL(13) = 0; // Controls the parallelism of the root node
-    mumps_solver.ICNTL(14) = // Controls the percentage increase in the estimated working space
+    ICNTL(mumps_solver, 8)  = 77; // Describes the scaling strategy
+    ICNTL(mumps_solver, 9)  = 1; // Computes the solution using A or A^T
+    ICNTL(mumps_solver, 10) = 0; // Applies the iterative refinement to the computed solution
+    ICNTL(mumps_solver, 11) = 0; // Computes statistics related to an error analysis of the linear system solved
+    ICNTL(mumps_solver, 12) = 0; // Defines an ordering strategy for symmetric matrices and is used
+    ICNTL(mumps_solver, 13) = 0; // Controls the parallelism of the root node
+    ICNTL(mumps_solver, 14) = // Controls the percentage increase in the estimated working space
         (solver_matrix.is_symmetric() ? 5 : 20);
-    mumps_solver.ICNTL(15) = 0; // Exploits compression of the input matrix resulting from a block format
-    mumps_solver.ICNTL(16) = 0; // Controls the setting of the number of OpenMP threads
+    ICNTL(mumps_solver, 15) = 0; // Exploits compression of the input matrix resulting from a block format
+    ICNTL(mumps_solver, 16) = 0; // Controls the setting of the number of OpenMP threads
     // ICNTL(17) Doesn't exist
-    mumps_solver.ICNTL(18) = 0; // Defines the strategy for the distributed input matrix
-    mumps_solver.ICNTL(19) = 0; // Computes the Schur complement matrix
-    mumps_solver.ICNTL(20) = 0; // Determines the format (dense, sparse, or distributed) of the right-hand sides
-    mumps_solver.ICNTL(21) = 0; // Determines the distribution (centralized or distributed) of the solution vectors.
-    mumps_solver.ICNTL(22) = 0; // Controls the in-core/out-of-core (OOC) factorization and solve.
-    mumps_solver.ICNTL(23) = 0; // Corresponds to the maximum size of the working memory in MegaBytes that MUMPS can
+    ICNTL(mumps_solver, 18) = 0; // Defines the strategy for the distributed input matrix
+    ICNTL(mumps_solver, 19) = 0; // Computes the Schur complement matrix
+    ICNTL(mumps_solver, 20) = 0; // Determines the format (dense, sparse, or distributed) of the right-hand sides
+    ICNTL(mumps_solver, 21) = 0; // Determines the distribution (centralized or distributed) of the solution vectors.
+    ICNTL(mumps_solver, 22) = 0; // Controls the in-core/out-of-core (OOC) factorization and solve.
+    ICNTL(mumps_solver, 23) = 0; // Corresponds to the maximum size of the working memory in MegaBytes that MUMPS can
     //                             allocate per working process
-    mumps_solver.ICNTL(24) = 0; // Controls the detection of “null pivot rows”.
-    mumps_solver.ICNTL(25) =
+    ICNTL(mumps_solver, 24) = 0; // Controls the detection of “null pivot rows”.
+    ICNTL(mumps_solver, 25) =
         0; // Allows the computation of a solution of a deficient matrix and also of a null space basis
-    mumps_solver.ICNTL(26) = 0; // Drives the solution phase if a Schur complement matrix has been computed
-    mumps_solver.ICNTL(27) = -32; // Controls the blocking size for multiple right-hand sides.
-    mumps_solver.ICNTL(28) = 0; // Determines whether a sequential or parallel computation of the ordering is performed
-    mumps_solver.ICNTL(29) =
+    ICNTL(mumps_solver, 26) = 0; // Drives the solution phase if a Schur complement matrix has been computed
+    ICNTL(mumps_solver, 27) = -32; // Controls the blocking size for multiple right-hand sides.
+    ICNTL(mumps_solver, 28) = 0; // Determines whether a sequential or parallel computation of the ordering is performed
+    ICNTL(mumps_solver, 29) =
         0; // Defines the parallel ordering tool (when ICNTL(28)=1) to be used to compute the fill-in reducing permutation.
-    mumps_solver.ICNTL(30) = 0; // Computes a user-specified set of entries in the inverse A^−1 of the original matrix
-    mumps_solver.ICNTL(31) = 0; // Indicates which factors may be discarded during the factorization.
-    mumps_solver.ICNTL(32) = 0; // Performs the forward elimination of the right-hand sides during the factorization
-    mumps_solver.ICNTL(33) = 0; // Computes the determinant of the input matrix.
-    mumps_solver.ICNTL(34) = 0; // Controls the conservation of the OOC files during JOB= –3
-    mumps_solver.ICNTL(35) = 0; // Controls the activation of the BLR feature
-    mumps_solver.ICNTL(36) = 0; // Controls the choice of BLR factorization variant
-    mumps_solver.ICNTL(37) = 0; // Controls the BLR compression of the contribution blocks
-    mumps_solver.ICNTL(38) = 600; // Estimates compression rate of LU factors
-    mumps_solver.ICNTL(39) = 500; // Estimates compression rate of contribution blocks
+    ICNTL(mumps_solver, 30) = 0; // Computes a user-specified set of entries in the inverse A^−1 of the original matrix
+    ICNTL(mumps_solver, 31) = 0; // Indicates which factors may be discarded during the factorization.
+    ICNTL(mumps_solver, 32) = 0; // Performs the forward elimination of the right-hand sides during the factorization
+    ICNTL(mumps_solver, 33) = 0; // Computes the determinant of the input matrix.
+    ICNTL(mumps_solver, 34) = 0; // Controls the conservation of the OOC files during JOB= –3
+    ICNTL(mumps_solver, 35) = 0; // Controls the activation of the BLR feature
+    ICNTL(mumps_solver, 36) = 0; // Controls the choice of BLR factorization variant
+    ICNTL(mumps_solver, 37) = 0; // Controls the BLR compression of the contribution blocks
+    ICNTL(mumps_solver, 38) = 600; // Estimates compression rate of LU factors
+    ICNTL(mumps_solver, 39) = 500; // Estimates compression rate of contribution blocks
     // ICNTL(40-47) Don't exist
-    mumps_solver.ICNTL(48) = 0; // Multithreading with tree parallelism
-    mumps_solver.ICNTL(49) = 0; // Compact workarray id%S at the end of factorization phase
+    ICNTL(mumps_solver, 48) = 0; // Multithreading with tree parallelism
+    ICNTL(mumps_solver, 49) = 0; // Compact workarray id%S at the end of factorization phase
     // ICNTL(50-55) Don't exist
-    mumps_solver.ICNTL(56) =
+    ICNTL(mumps_solver, 56) =
         0; // Detects pseudo-singularities during factorization and factorizes the root node with a rankrevealing method
     // ICNTL(57) Doesn't exist
-    mumps_solver.ICNTL(58) = 2; // Defines options for symbolic factorization
+    ICNTL(mumps_solver, 58) = 2; // Defines options for symbolic factorization
     // ICNTL(59-60) Don't exist
 
-    mumps_solver.CNTL(1) = -1.0; // Relative threshold for numerical pivoting
-    mumps_solver.CNTL(2) = -1.0; // Stopping criterion for iterative refinement
-    mumps_solver.CNTL(3) = 0.0; // Determine null pivot rows
-    mumps_solver.CNTL(4) = -1.0; // Determines the threshold for static pivoting
-    mumps_solver.CNTL(5) =
+    CNTL(mumps_solver, 1) = -1.0; // Relative threshold for numerical pivoting
+    CNTL(mumps_solver, 2) = -1.0; // Stopping criterion for iterative refinement
+    CNTL(mumps_solver, 3) = 0.0; // Determine null pivot rows
+    CNTL(mumps_solver, 4) = -1.0; // Determines the threshold for static pivoting
+    CNTL(mumps_solver, 5) =
         0.0; // Defines the fixation for null pivots and is effective only when null pivot row detection is active
     // CNTL(6) Doesn't exist
-    mumps_solver.CNTL(7) = 0.0; // Defines the precision of the dropping parameter used during BLR compression
+    CNTL(mumps_solver, 7) = 0.0; // Defines the precision of the dropping parameter used during BLR compression
     // CNTL(8-15) Don't exist
 
     mumps_solver.job = JOB_ANALYSIS_AND_FACTORIZATION;
@@ -96,7 +96,7 @@ void DirectSolver_COO_MUMPS_Give::initializeMumpsSolver(DMUMPS_STRUC_C& mumps_so
     mumps_solver.a   = solver_matrix.values_data();
     dmumps_c(&mumps_solver);
 
-    if (mumps_solver.sym == SYM_POSITIVE_DEFINITE && mumps_solver.INFOG(12) != 0) {
+    if (mumps_solver.sym == SYM_POSITIVE_DEFINITE && INFOG(mumps_solver, 12) != 0) {
         std::cout
             << "Warning: DirectSolver matrix is not positive definite: Negative pivots in the factorization phase."
             << std::endl;