* basic support for multicore CPU via a .evalOMP() which

   internaly uses OpenMP if enabled at compile time.
 * added a bench/ folder with a couple benchmarks and benchmark tools.

1 files changed, 46 insertions, 0 deletions

diff --git a/bench/basicbenchmark.cpp b/bench/basicbenchmark.cpp
new file mode 100644
index 000000000..c44ed4514
--- /dev/null
+++ b/bench/basicbenchmark.cpp
@@ -0,0 +1,46 @@
+
+#include "BenchUtil.h"
+#include "basicbenchmark.h"
+
+int main(int argc, char *argv[])
+{
+  // disbale floating point exceptions
+  // this leads to more stable bench results
+  // (this is done by default by ICC)
+  #ifndef __INTEL_COMPILER
+  {
+    int aux;
+    asm(
+    "stmxcsr   %[aux]           \n\t"
+    "orl       $32832, %[aux]   \n\t"
+    "ldmxcsr   %[aux]           \n\t"
+    : : [aux] "m" (aux));
+  }
+  #endif
+
+  // this is the list of matrix type and size we want to bench:
+  // ((suffix) (matrix size) (number of iterations))
+  #define MODES ((3d)(3)(4000000)) ((4d)(4)(1000000)) ((Xd)(4)(1000000)) ((Xd)(20)(10000))
+//   #define MODES ((Xd)(20)(10000))
+
+  #define _GENERATE_HEADER(R,ARG,EL) << BOOST_PP_STRINGIZE(BOOST_PP_SEQ_HEAD(EL)) << "-" \
+    << BOOST_PP_STRINGIZE(BOOST_PP_SEQ_ELEM(1,EL)) << "x" \
+    << BOOST_PP_STRINGIZE(BOOST_PP_SEQ_ELEM(1,EL)) << "   /   "
+
+  std::cout BOOST_PP_SEQ_FOR_EACH(_GENERATE_HEADER, ~, MODES ) << endl;
+
+  const int tries = 10;
+
+  #define _RUN_BENCH(R,ARG,EL) \
+    std::cout << ARG( \
+      BOOST_PP_CAT(Matrix, BOOST_PP_SEQ_HEAD(EL)) (\
+         BOOST_PP_SEQ_ELEM(1,EL),BOOST_PP_SEQ_ELEM(1,EL)), BOOST_PP_SEQ_ELEM(2,EL), tries) \
+    << "   ";
+
+  BOOST_PP_SEQ_FOR_EACH(_RUN_BENCH, benchBasic<LazyEval>, MODES );
+  std::cout << endl;
+  BOOST_PP_SEQ_FOR_EACH(_RUN_BENCH, benchBasic<EarlyEval>, MODES );
+  std::cout << endl;
+
+  return 0;
+}