support of sequential and MPI enabled version

exanauts · Apr 19, 2018 · cefa8b7 · cefa8b7
1 parent cd2c26d
commit cefa8b7
Show file tree

Hide file tree

Showing 4 changed files with 68 additions and 25 deletions.
diff --git a/include/ad.hpp b/include/ad.hpp
@@ -31,7 +31,6 @@ original variable.
 #include "perf.hpp"
 #include "alg.hpp"
 #include "user.hpp"
-#include <mpi.h>
 #include <stdlib.h>
 #include "parallel.hpp"
 #ifdef EIGEN
@@ -873,11 +872,11 @@ void propagateAD(pVector<double>& m0, pMatrix<double>& cv0, System& sys,
       // exit(1);
       // paduprop_gather(H);
       // H.cutoff(0.7);
-      std::cout << "H nz: " << H.nz() << std::endl;
+      // std::cout << "H nz: " << H.nz() << std::endl;
       H_tmp=H;
-      std::cout << "Cutrate: " << cutrate << std::endl;
+      // std::cout << "Cutrate: " << cutrate << std::endl;
       // H_tmp.cutoff(cutrate);
-      std::cout << "H_tmp nz: " << H_tmp.nz() << std::endl;
+      // std::cout << "H_tmp nz: " << H_tmp.nz() << std::endl;
       drivers.t3s_t2s_t1s_driver(m0, J, H_tmp, T, start, end);
       drivers.t1s_driver(m0, J);
       // T.zeros();
@@ -951,10 +950,10 @@ void propagateAD(pVector<double>& m0, pMatrix<double>& cv0, System& sys,
 
   if (degree > 2) {
     if(paduprop_getrank() == 0) {
-      std::cout << "Propagating covariance 3rd order part1" << std::endl;
-      std::cout << "NZ: " << cv0.nz() << std::endl;
-      std::cout << "Entries: " << dim*dim << std::endl;
-      std::cout << "Percent NZ: " << (cv0.nz()*100)/(dim*dim) << std::endl;
+      // std::cout << "Propagating covariance 3rd order part1" << std::endl;
+      // std::cout << "NZ: " << cv0.nz() << std::endl;
+      // std::cout << "Entries: " << dim*dim << std::endl;
+      // std::cout << "Percent NZ: " << (cv0.nz()*100)/(dim*dim) << std::endl;
     }
 
     double aux, kurt;
@@ -967,7 +966,7 @@ void propagateAD(pVector<double>& m0, pMatrix<double>& cv0, System& sys,
 
     for (size_t pm = 0; pm < dim; ++pm) { 
       if(paduprop_getrank() == 0) {
-        std::cout << "pm: " << pm << std::endl;
+        // std::cout << "pm: " << pm << std::endl;
       }
       for (size_t pn = 0; pn < dim; ++pn) { 
         for (size_t i = 0; i < dim; ++i) { 

diff --git a/include/parallel.hpp b/include/parallel.hpp
@@ -1,9 +1,11 @@
 #ifndef ADUPROP_PARALLEL_HPP_
 #define ADUPROP_PARALLEL_HPP_
-
+#ifdef MPI_VERSION
 #include <mpi.h>
+#endif
 #include <fstream>
 #include "alg.hpp"
+#include <cstring>
 using namespace alg;
 
 int paduprop_init();

diff --git a/include/perf.hpp b/include/perf.hpp
@@ -4,11 +4,17 @@
 
 #include <map>
 #include <iostream>
+#ifdef MPI_VERSION
 #include <mpi.h>
+#endif
 
 
 // E.g. __rdtsc() or MPI_Wtime()
+#ifdef MPI_VERSION
 #define ADUPROP_TIMER MPI_Wtime()
+#else
+#define ADUPROP_TIMER __rdtsc()
+#endif
 // double for MPI_Wtime and uint64_t for __rdtsc()
 #define ADUPROP_TIMER_TYPE double
 

diff --git a/src/parallel.cpp b/src/parallel.cpp
@@ -2,7 +2,11 @@
 
 std::ofstream paduprop_sink("/dev/null");
 int paduprop_init() {
+#ifdef MPI_VERSION
   int ierr = MPI_Init(NULL, NULL);
+#else
+  int ierr = 0;
+#endif
 
   if (paduprop_getrank() != 0) {
     // Mute standard output
@@ -13,12 +17,15 @@ int paduprop_init() {
   return ierr;
 }
 int paduprop_destroy() {
+#ifdef MPI_VERSION
   return MPI_Finalize();
+#else
+  return 0;
+#endif
 }
 size_t paduprop_getstart(size_t dim) {
-  int rank; int commsize;
-  MPI_Comm_rank(MPI_COMM_WORLD, &rank);
-  MPI_Comm_size(MPI_COMM_WORLD, &commsize);
+  int commsize = paduprop_getcommsize();
+  int rank = paduprop_getrank();
   size_t chunk = dim/commsize;
   size_t remain = dim%commsize;
   size_t addleft = 0;
@@ -30,8 +37,7 @@ size_t paduprop_getstart(size_t dim) {
 }
 
 size_t paduprop_getstart(size_t dim, int rank) {
-  int commsize;
-  MPI_Comm_size(MPI_COMM_WORLD, &commsize);
+  int commsize = paduprop_getcommsize();
   size_t chunk = dim/commsize;
   size_t remain = dim%commsize;
   size_t addleft = 0;
@@ -43,9 +49,8 @@ size_t paduprop_getstart(size_t dim, int rank) {
 }
 
 size_t paduprop_getend(size_t dim) {
-  int rank; int commsize;
-  MPI_Comm_rank(MPI_COMM_WORLD, &rank);
-  MPI_Comm_size(MPI_COMM_WORLD, &commsize);
+  int commsize = paduprop_getcommsize();
+  int rank = paduprop_getrank();
   size_t chunk = dim/commsize;
   size_t remain = dim%commsize;
   size_t addright = 0;
@@ -57,8 +62,7 @@ size_t paduprop_getend(size_t dim) {
 }
 
 size_t paduprop_getend(size_t dim, int rank) {
-  int commsize;
-  MPI_Comm_size(MPI_COMM_WORLD, &commsize);
+  int commsize = paduprop_getcommsize();
   size_t chunk = dim/commsize;
   size_t remain = dim%commsize;
   size_t addright = 0;
@@ -71,29 +75,46 @@ size_t paduprop_getend(size_t dim, int rank) {
 
 int paduprop_getrank() {
   int rank;
+#ifdef MPI_VERSION
   MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+#else
+  rank = 0;
+#endif
   return rank;
 }
 int paduprop_getcommsize() {
   int size;
+#ifdef MPI_VERSION
   MPI_Comm_size(MPI_COMM_WORLD, &size);
+#else
+  size = 1;
+#endif
   return size;
 }
 int paduprop_sum(pMatrix<double> &mat) {
   size_t nrows = mat.nrows();
   size_t ncols = mat.ncols();
   size_t size = nrows*ncols;
   double *ptr = mat.get_datap();
-  return MPI_Allreduce(MPI_IN_PLACE, ptr, size, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
+  int ierr = 0;
+#ifdef MPI_VERSION
+  ierr = MPI_Allreduce(MPI_IN_PLACE, ptr, size, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
+#endif
+  return ierr;
 }
 
 int paduprop_sum(pVector<double> &vec) {
   size_t size = vec.dim();
   double *ptr = vec.get_datap();
-  return MPI_Allreduce(MPI_IN_PLACE, ptr, size, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
+  int ierr = 0;
+#ifdef MPI_VERSION
+  ierr = MPI_Allreduce(MPI_IN_PLACE, ptr, size, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
+#endif
+  return ierr;
 }
 
 int paduprop_gather(pTensor3<double> &ten) {
+  int ierr = 0;
   size_t dim = ten.get_d1();
   size_t start = paduprop_getstart(dim);
   size_t end = paduprop_getend(dim);
@@ -108,16 +129,22 @@ int paduprop_gather(pTensor3<double> &ten) {
     recvcounts[i] = (paduprop_getend(dim,i) - paduprop_getstart(dim,i))*dim*dim;
     count += recvcounts[i];
   }
-  int ierr = MPI_Allgatherv(sendbuf, size, MPI_DOUBLE,
+#ifdef MPI_VERSION
+  ierr = MPI_Allgatherv(sendbuf, size, MPI_DOUBLE,
                   &ten[0][0][0], recvcounts, displs, MPI_DOUBLE,
                   MPI_COMM_WORLD);
+#else
+  std::memcpy (sendbuf, &ten[0][0][0], recvcounts[0] * sizeof(double));
+#endif
+
   delete [] recvcounts;
   delete [] displs;
   delete [] sendbuf;
   return ierr;
 }
 
 int paduprop_gather(pMatrix<double> &mat) {
+  int ierr = 0;
   size_t dim = mat.nrows();
   size_t start = paduprop_getstart(dim);
   size_t end = paduprop_getend(dim);
@@ -132,16 +159,21 @@ int paduprop_gather(pMatrix<double> &mat) {
     recvcounts[i] = (paduprop_getend(dim,i) - paduprop_getstart(dim,i))*dim;
     count += recvcounts[i];
   }
-  int ierr = MPI_Allgatherv(sendbuf, size, MPI_DOUBLE,
+#ifdef MPI_VERSION
+  ierr = MPI_Allgatherv(sendbuf, size, MPI_DOUBLE,
                   &mat[0][0], recvcounts, displs, MPI_DOUBLE,
                   MPI_COMM_WORLD);
+#else
+  std::memcpy (sendbuf, &mat[0][0], recvcounts[0] * sizeof(double));
+#endif
   delete [] recvcounts;
   delete [] displs;
   delete [] sendbuf;
   return ierr;
 }
 
 int paduprop_gather(pVector<double> &vec) {
+  int ierr = 0;
   size_t dim = vec.dim();
   size_t start = paduprop_getstart(dim);
   size_t end = paduprop_getend(dim);
@@ -156,9 +188,13 @@ int paduprop_gather(pVector<double> &vec) {
     recvcounts[i] = (paduprop_getend(dim,i) - paduprop_getstart(dim,i));
     count += recvcounts[i];
   }
-  int ierr = MPI_Allgatherv(sendbuf, size, MPI_DOUBLE,
+  #ifdef MPI_VERSION
+  ierr = MPI_Allgatherv(sendbuf, size, MPI_DOUBLE,
                   &vec[0], recvcounts, displs, MPI_DOUBLE,
                   MPI_COMM_WORLD);
+  #else
+  std::memcpy (sendbuf, &vec[0], recvcounts[0] * sizeof(double));
+  #endif
   delete [] recvcounts;
   delete [] displs;
   delete [] sendbuf;