[clef] timing examples

benchmark/clef_speed.cpp

@@ -0,0 +1,179 @@
+
+#include <triqs/clef.hpp>
+#include <triqs/gfs.hpp>
+#include <triqs/utility/timer.hpp>
+
+using namespace triqs;   
+using namespace triqs::gfs;
+
+// Index placeholders
+clef::placeholder<0> i;
+clef::placeholder<1> j;
+clef::placeholder<2> k;
+clef::placeholder<3> l;
+
+// Frequency placeholders
+clef::placeholder<4> w;
+clef::placeholder<5> n1;
+clef::placeholder<6> n2;
+clef::placeholder<7> n3;
+
+typedef gf<cartesian_product<imfreq, imfreq, imfreq>, tensor_valued<4>> g2_iw_t;
+
+double test_clef_clef(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  auto G2_iw_ph = g2_iw_t{
+    {{beta, Boson, n}, {beta, Fermion, m}, {beta, Fermion, m}}, {1, 1, 1, 1}};
+  timer.start();
+
+  G2_iw_ph(w,n1,n2)(i,j,k,l) << 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+double test_loop_clef(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+  timer.start();
+
+  for(auto const& w : mesh_b)
+  for(auto const& n1 : mesh_f)
+  for(auto const& n2 : mesh_f)
+    G2_iw_ph[w,n1,n2](i,j,k,l) << 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+double test_loop_loop(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+  timer.start();
+
+  int size_0 = G2_iw_ph.target_shape()[0];
+  int size_1 = G2_iw_ph.target_shape()[1];
+  int size_2 = G2_iw_ph.target_shape()[2];
+  int size_3 = G2_iw_ph.target_shape()[3];
+
+  for(auto const& w : mesh_b)
+  for(auto const& n1 : mesh_f)
+  for(auto const& n2 : mesh_f)
+  for(int i : range(size_0))
+  for(int j : range(size_1))
+  for(int k : range(size_2))
+  for(int l : range(size_3))
+    G2_iw_ph[w,n1,n2](i,j,k,l) = 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+double test_loop_loop_bracket(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+  timer.start();
+
+  int size_0 = G2_iw_ph.target_shape()[0];
+  int size_1 = G2_iw_ph.target_shape()[1];
+  int size_2 = G2_iw_ph.target_shape()[2];
+  int size_3 = G2_iw_ph.target_shape()[3];
+
+  for(auto const& w : mesh_b)
+  for(auto const& n1 : mesh_f)
+  for(auto const& n2 : mesh_f)
+  for(int i : range(size_0))
+  for(int j : range(size_1))
+  for(int k : range(size_2))
+  for(int l : range(size_3))
+    G2_iw_ph[{w,n1,n2}](i,j,k,l) = 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+double test_loop_loop_constexpr_target(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+  timer.start();
+
+  auto size_0 = 1;
+  auto size_1 = 1;
+  auto size_2 = 1;  
+  auto size_3 = 1;
+
+  for(auto const& w : mesh_b)
+  for(auto const& n1 : mesh_f)
+  for(auto const& n2 : mesh_f)
+  for(int i : range(size_0))
+  for(int j : range(size_1))
+  for(int k : range(size_2))
+  for(int l : range(size_3))
+    G2_iw_ph[{w,n1,n2}](i,j,k,l) = 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+double test_loop_fixed(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+  timer.start();
+
+  for(auto const& w : mesh_b)
+  for(auto const& n1 : mesh_f)
+  for(auto const& n2 : mesh_f)
+    G2_iw_ph[w,n1,n2](0,0,0,0) = 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+double test_array(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+  timer.start();
+
+  G2_iw_ph.data() *= 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+/*
+
+//#include <Eigen/Core>
+//#include <Eigen/Dense>
+//#include <unsupported/Eigen/CXX11/Tensor>
+
+double test_eigen_tensor(int n=20, int m=20, double beta=1.0) {
+  utility::timer timer;
+  gf_mesh<imfreq> mesh_b{beta, Boson, n};
+  gf_mesh<imfreq> mesh_f{beta, Fermion, m};
+  auto G2_iw_ph = g2_iw_t{{mesh_b, mesh_f, mesh_f}, {1, 1, 1, 1}};
+
+  // create eigen tensor view of the data ?!?!
+  template<int I> using Tensor = Eigen::Tensor<std::complex<double>, I, Eigen::RowMajor>;
+  Eigen::TensorMap<Tensor<7>> data(G2_iw_ph.data(), n, m, m, 1, 1, 1, 1);
+
+  timer.start();
+
+  data = 0.0;
+
+  timer.stop();  
+  return double(timer);
+}
+
+*/

benchmark/clef_speed.py

@@ -0,0 +1,23 @@
+import numpy as np
+from cpp2py.compiler import compile
+
+filename = 'clef_speed.cpp'
+
+code = ''.join(open(filename, 'r').readlines())
+
+cxxflags = '-O3 -march=native -Ofast -mavx -mfma -mavx2 -g -ggdb -Wno-invalid-partial-specialization '
+
+includes = ' -I /opt/local/include/ -I /opt/local/include/openmpi-clang40 '
+
+M = compile(code, modules = "pytriqs", cxxflags=cxxflags + includes)
+
+print '--> Running clef timing tests'
+tests = np.array([
+    getattr(M, method) for method in M.__dict__.keys() if 'test_' in method])
+times = np.array([test(n=80, m=80) for test in tests])
+
+sidx = np.argsort(times)[::-1]
+times, tests = times[sidx], tests[sidx]
+
+for time, test in zip(times, tests):
+    print '%40s -- %10.2f ms' % (test.__name__, 1000.0 * time)