Command line arguments for dpotri, more cleanups. Enable profiling if…

… the user has enabled it via MCA.
TESSEorg · Apr 4, 2022 · 1d53756 · 1d53756
1 parent bb85c67
commit 1d53756
Show file tree

Hide file tree

Showing 5 changed files with 40 additions and 21 deletions.
diff --git a/examples/potrf/potrf.h b/examples/potrf/potrf.h
@@ -9,9 +9,9 @@
 namespace potrf {
 
 /* FLOP macros taken from DPLASMA */
-#define FMULS_POTRF(__n) ((double)(__n) * (((1. / 6.) * (double)(__n) + 0.5) * (double)(__n) + (1. / 3.)))
-#define FADDS_POTRF(__n) ((double)(__n) * (((1. / 6.) * (double)(__n)      ) * (double)(__n) - (1. / 6.)))
-#define FLOPS_DPOTRF(__n) (     FMULS_POTRF((__n)) +       FADDS_POTRF((__n)) )
+double FMULS_POTRF(double __n) { return (__n * (((1. / 6.) * __n + 0.5) * __n + (1. / 3.))); }
+double FADDS_POTRF(double __n) { return (__n * (((1. / 6.) * __n      ) * __n - (1. / 6.))); }
+double FLOPS_DPOTRF(double __n) { return FMULS_POTRF(__n) + FADDS_POTRF(__n); }
 
 template <typename T>
 auto make_potrf(MatrixT<T>& A,

diff --git a/examples/potrf/potri.h b/examples/potrf/potri.h
@@ -11,10 +11,10 @@
 
 namespace potri {
 
-/* FLOP macros taken from DPLASMA */
-#define FMULS_POTRI(__n) ((double)(__n) * (((1. / 6.) * (double)(__n) + 0.5) * (double)(__n) + (1. / 3.)))
-#define FADDS_POTRI(__n) ((double)(__n) * (((1. / 6.) * (double)(__n)      ) * (double)(__n) - (1. / 6.)))
-#define FLOPS_DPOTRI(__n) (     FMULS_POTRI((__n)) +       FADDS_POTRI((__n)) )
+/* FLOP macros taken from DPLASMA */ 
+double FMULS_POTRI(double __n) { return ( __n * ((2. / 3.) + __n * ((1. / 3.) * __n + 1. )) ); }
+double FADDS_POTRI(double __n) { return ( __n * ((1. / 6.) + __n * ((1. / 3.) * __n - 0.5)) ); }
+double FLOPS_DPOTRI(double __n) { return FMULS_POTRI(__n) + FADDS_POTRI(__n); }
 
 auto make_potri_ttg(MatrixT<double> &A, ttg::Edge<Key2, MatrixTile<double>>&input, ttg::Edge<Key2, MatrixTile<double>>&output ) {
   ttg::Edge<Key2, MatrixTile<double>> trtri_to_lauum("trtri_to_lauum");

diff --git a/examples/potrf/testing_dpotrf.cc b/examples/potrf/testing_dpotrf.cc
@@ -129,7 +129,7 @@ int main(int argc, char **argv)
     auto elapsed = (std::chrono::duration_cast<std::chrono::microseconds>(end - beg).count());
     end = std::chrono::high_resolution_clock::now();
     std::cout << "TTG Execution Time (milliseconds) : "
-              << elapsed / 1E3 << " : Flops " << (FLOPS_DPOTRF(N)) << " " << (FLOPS_DPOTRF(N)/1e9)/(elapsed/1e6) << " GF/s" << std::endl;
+              << elapsed / 1E3 << " : Flops " << (potrf::FLOPS_DPOTRF(N)) << " " << (potrf::FLOPS_DPOTRF(N)/1e9)/(elapsed/1e6) << " GF/s" << std::endl;
   }
 
   world.impl().stop_tracing_dag_of_tasks();

diff --git a/examples/potrf/testing_dpotri.cc b/examples/potrf/testing_dpotri.cc
@@ -20,6 +20,21 @@
 #include <dplasma.h>
 #endif
 
+char* getCmdOption(char ** begin, char ** end, const std::string & option)
+{
+    char ** itr = std::find(begin, end, option);
+    if (itr != end && ++itr != end)
+    {
+        return *itr;
+    }
+    return nullptr;
+}
+
+bool cmdOptionExists(char** begin, char** end, const std::string& option)
+{
+    return std::find(begin, end, option) != end;
+}
+
 int main(int argc, char **argv)
 {
 
@@ -30,25 +45,22 @@ int main(int argc, char **argv)
   int check = 0;
   int nthreads = -1;
   const char* prof_filename = nullptr;
+  char *opt = nullptr;
 
-  if (argc > 1) {
-    N = M = atoi(argv[1]);
-  }
-
-  if (argc > 2) {
-    NB = atoi(argv[2]);
+  if( (opt = getCmdOption(argv+1, argv+argc, "-N")) != nullptr ) {
+    N = M = atoi(opt);
   }
 
-  if (argc > 3) {
-    check = atoi(argv[3]);
+  if( (opt = getCmdOption(argv+1, argv+argc, "-t")) != nullptr ) {
+    NB = atoi(opt);
   }
 
-  if (argc > 4) {
-    nthreads = atoi(argv[4]);
+  if( (opt = getCmdOption(argv+1, argv+argc, "-c")) != nullptr ) {
+    nthreads = atoi(opt);
   }
 
-  if (argc > 5) {
-    prof_filename = argv[5];
+  if( (opt = getCmdOption(argv+1, argv+argc, "-dag")) != nullptr ) {
+    prof_filename = opt;
   }
 
   ttg::initialize(argc, argv, nthreads);
@@ -133,7 +145,7 @@ int main(int argc, char **argv)
     auto elapsed = (std::chrono::duration_cast<std::chrono::microseconds>(end - beg).count());
     end = std::chrono::high_resolution_clock::now();
     std::cout << "TTG Execution Time (milliseconds) : "
-              << elapsed / 1E3 << " : Flops " << (FLOPS_DPOTRF(N)) << " " << (FLOPS_DPOTRF(N)/1e9)/(elapsed/1e6) << " GF/s" << std::endl;
+              << elapsed / 1E3 << " : Flops " << (potri::FLOPS_DPOTRI(N)) << " " << (potri::FLOPS_DPOTRI(N)/1e9)/(elapsed/1e6) << " GF/s" << std::endl;
   }
 
   world.impl().stop_tracing_dag_of_tasks();

diff --git a/ttg/ttg/parsec/ttg.h b/ttg/ttg/parsec/ttg.h
@@ -185,6 +185,13 @@ namespace ttg_parsec {
     {
       ttg::detail::register_world(*this);
       ctx = parsec_init(ncores, argc, argv);
+
+#if defined(PARSEC_PROF_TRACE)
+      if(parsec_profile_enabled) {
+        ttg::profile_on();
+      }
+#endif
+
       es = ctx->virtual_processes[0]->execution_streams[0];
 
       parsec_ce.tag_register(_PARSEC_TTG_TAG, &detail::static_unpack_msg, this, PARSEC_TTG_MAX_AM_SIZE);