regression updates and SIMD size 1 for reverse AD

RichRoos · Sep 18, 2022 · 4357145 · 4357145
1 parent c0a8d7c
commit 4357145
Show file tree

Hide file tree

Showing 4 changed files with 17 additions and 4 deletions.
diff --git a/Common/include/parallelization/vectorization.hpp b/Common/include/parallelization/vectorization.hpp
@@ -58,7 +58,17 @@ template<class T>
 constexpr size_t preferredLen() { return PREFERRED_SIZE / sizeof(T); }
 
 template<>
-constexpr size_t preferredLen<su2double>() { return PREFERRED_SIZE / sizeof(passivedouble); }
+constexpr size_t preferredLen<su2double>() {
+#ifdef CODI_REVERSE_TYPE
+  /*--- Use a SIMD size of 1 for reverse AD, larger sizes increase
+   * the pre-accumulation time with no performance benefit. ---*/
+  return 1;
+#else
+  /*--- For forward AD there is a performance benefit. This covers
+   * forward AD and primal mode (su2double == passivedouble). ---*/
+  return PREFERRED_SIZE / sizeof(passivedouble);
+#endif
+}
 
 /*!
  * \class Array

diff --git a/SU2_CFD/include/numerics_simd/CNumericsSIMD.cpp b/SU2_CFD/include/numerics_simd/CNumericsSIMD.cpp
@@ -130,9 +130,12 @@ CNumericsSIMD* createNumerics(const CConfig& config, int iMesh, const CVariable*
  * numerical methods.
  */
 CNumericsSIMD* CNumericsSIMD::CreateNumerics(const CConfig& config, int nDim, int iMesh, const CVariable* turbVars) {
+#ifndef CODI_REVERSE_TYPE
   if ((Double::Size < 4) && (SU2_MPI::GetRank() == MASTER_NODE)) {
-    cout << "WARNING: SU2 was not compiled for an AVX-capable architecture." << endl;
+    cout << "WARNING: SU2 was not compiled for an AVX-capable architecture. Performance could be better,\n"
+            "         see https://su2code.github.io/docs_v7/Build-SU2-Linux-MacOS/#compiler-optimizations" << endl;
   }
+#endif
   if (nDim == 2) return createNumerics<2>(config, iMesh, turbVars);
   if (nDim == 3) return createNumerics<3>(config, iMesh, turbVars);
 

diff --git a/TestCases/incomp_navierstokes/streamwise_periodic/chtPinArray_2d/of_grad_findiff.csv.ref b/TestCases/incomp_navierstokes/streamwise_periodic/chtPinArray_2d/of_grad_findiff.csv.ref
@@ -1,2 +1,2 @@
 "VARIABLE"      , "AVG_DENSITY[0]", "AVG_ENTHALPY[0]", "AVG_NORMALVEL[0]", "DRAG[0]"       , "EFFICIENCY[0]" , "FORCE_X[0]"    , "FORCE_Y[0]"    , "FORCE_Z[0]"    , "LIFT[0]"       , "MOMENT_X[0]"   , "MOMENT_Y[0]"   , "MOMENT_Z[0]"   , "SIDEFORCE[0]"  , "SURFACE_MACH[0]", "SURFACE_MASSFLOW[0]", "SURFACE_MOM_DISTORTION[0]", "SURFACE_PRESSURE_DROP[0]", "SURFACE_SECONDARY[0]", "SURFACE_SECOND_OVER_UNIFORM[0]", "SURFACE_STATIC_PRESSURE[0]", "SURFACE_STATIC_TEMPERATURE[0]", "SURFACE_TOTAL_PRESSURE[0]", "SURFACE_TOTAL_TEMPERATURE[0]", "SURFACE_UNIFORMITY[0]", "AVG_TEMPERATURE[1]", "MAXIMUM_HEATFLUX[1]", "TOTAL_HEATFLUX[1]", "FINDIFF_STEP"  
-0               , 0.0             , -199999.9862164259, -2.2204999999731917e-08, 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , -0.04999999997368221, -2.2200000003768e-08, -2.0599999983605954      , 0.0                     , 2.1199999991616814  , 3.6999999980524834            , 330.00000030369847        , -30.00000106112566           , 314.000000400938         , -30.00000106112566          , -1.3999999826097564  , -139.99999737279722, 0.0                , -509.99999530176865, 1e-08           
+0               , 0.0             , -99999.96982514858, -2.2204000000409886e-08, 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , 0.0             , -0.04000000053405728, 0.0                , -2.060000001136153       , 0.0                     , 2.12000000054946    , 3.700000000828041             , 330.00000030369847        , -30.00000106112566           , 314.99999977313564       , -30.00000106112566          , -1.400000004814217   , -140.0000030571391, 0.0                , -509.99999530176865, 1e-08           
diff --git a/TestCases/serial_regression.py b/TestCases/serial_regression.py
@@ -336,7 +336,7 @@ def main():
     turb_wallfunction_flatplate_sa.cfg_dir   = "wallfunctions/flatplate/compressible_SA"
     turb_wallfunction_flatplate_sa.cfg_file  = "turb_SA_flatplate.cfg"
     turb_wallfunction_flatplate_sa.test_iter = 10
-    turb_wallfunction_flatplate_sa.test_vals = [-4.684321, -2.249602, -2.229407, 0.779062, -5.306220, 10, -1.984965, 0.069232, 0.002496] #last 9 columns
+    turb_wallfunction_flatplate_sa.test_vals = [-4.436048, -2.044706, -2.114644, 0.979771, -5.393729, 10, -1.589465, 0.069744, 0.002686] #last 9 columns
     turb_wallfunction_flatplate_sa.su2_exec  = "SU2_CFD"
     turb_wallfunction_flatplate_sa.timeout   = 1600
     turb_wallfunction_flatplate_sa.tol       = 0.00001