Fixes for navigation index table in single precision mode.

VecGeom/base/Transformation3D.h

@@ -426,8 +426,8 @@ public:
   size_t DeviceSizeOf() const { return DevicePtr<cuda::Transformation3D>::SizeOf(); }
   DevicePtr<cuda::Transformation3D> CopyToGpu() const;
   DevicePtr<cuda::Transformation3D> CopyToGpu(DevicePtr<cuda::Transformation3D> const gpu_ptr) const;
-  static void CopyManyToGpu(const std::vector<Transformation3D const *>& trafos,
-                            const std::vector<DevicePtr<cuda::Transformation3D>>& gpu_ptrs);
+  static void CopyManyToGpu(const std::vector<Transformation3D const *> &trafos,
+                            const std::vector<DevicePtr<cuda::Transformation3D>> &gpu_ptrs);
 #endif
 
 #ifdef VECGEOM_ROOT
@@ -611,7 +611,7 @@ void Transformation3D::DoRotation_new(Vector3D<InputType> const &master, Vector3
   }
 
   // General case
-  local = Vector3D<double>(); // reset to zero -- any better way to do this???
+  local = Vector3D<InputType>(); // reset to zero -- any better way to do this???
   if (code & 0x001) {
     local[0] += master[0] * fRotation[0];
   }

source/NavIndexTable.cpp

@@ -21,7 +21,7 @@ NavIndex_t BuildNavIndexVisitor::apply(NavStatePath *state, int level, NavIndex_
     return NavIndexTable::Instance()->ValidateState(state);
   }
   // Size in bytes of the current node data
-  size_t current_size = (3 + nd + ((nd + 1) & 1)) * sizeof(unsigned int) + int(cacheTrans) * 12 * sizeof(double);
+  size_t current_size = (3 + nd + ((nd + 1) & 1)) * sizeof(unsigned int) + int(cacheTrans) * 12 * sizeof(Precision);
   if (fDoCount) {
     fTableSize += current_size;
     return 0;
@@ -66,14 +66,15 @@ NavIndex_t BuildNavIndexVisitor::apply(NavStatePath *state, int level, NavIndex_
   *content_hasm = 0x04 + 0x02 * (unsigned short)mat.HasTranslation() + (unsigned short)mat.HasRotation();
 
   // Write the transformation elements
-  auto content_mat = (double *)(&fNavInd[fCurrent]);
+  auto content_mat = (Precision *)(&fNavInd[fCurrent]);
   for (auto i = 0; i < 3; ++i)
     content_mat[i] = mat.Translation(i);
   for (auto i = 0; i < 9; ++i)
     content_mat[i + 3] = mat.Rotation(i);
 
   // Set new value for fCurrent
-  fCurrent += 24;
+  fCurrent += 12 * sizeof(Precision) / sizeof(NavIndex_t);
+  assert((fCurrent - new_mother) * sizeof(NavIndex_t) == current_size);
   return new_mother;
 }
 

test/benchmark/TestNavIndex.cpp

@@ -115,12 +115,13 @@ int visitAllPlacedVolumesPassNavIndex(VPlacedVolume const *currentvolume, Visito
     visitor->apply(state, nav_ind);
     auto ierr = visitor->GetError();
     if (ierr) {
-      printf("=== EEE === TestNavIndex: %s\n", errcodes[ierr]);
+      printf("=== EEE === TestNavIndex: %s\n", errcodes[ierr - 1]);
       return ierr;
     }
     for (auto daughter : currentvolume->GetDaughters()) {
       auto nav_ind_d = NavStateIndex::PushImpl(nav_ind, daughter);
-      visitAllPlacedVolumesPassNavIndex(daughter, visitor, state, nav_ind_d);
+      ierr           = visitAllPlacedVolumesPassNavIndex(daughter, visitor, state, nav_ind_d);
+      if (ierr) return ierr;
     }
     state->Pop();
   }

test/benchmark/TestNavIndex.cu

@@ -29,8 +29,8 @@ namespace visitorcuda {
 
 class GlobalToLocalVisitor {
 private:
-  int fError = 0;   ///< error code
-  int fNiter = 0;   ///< number of iterations
+  int fError = 0; ///< error code
+  int fNiter = 0; ///< number of iterations
 public:
   VECCORE_ATT_HOST_DEVICE
   GlobalToLocalVisitor() {}
@@ -115,21 +115,21 @@ int visitAllPlacedVolumesPassNavIndex(VPlacedVolume const *currentvolume, Visito
                             "level mismatch",
                             "navigation index inconsistency for Push/Pop",
                             "number of daughters mismatch",
-                            "transformation matrix mismatch"
-                           };
-  constexpr int maxiter = 100000; // limit the maximum number of iterations (slow on 1 GPU thread)
+                            "transformation matrix mismatch"};
+  constexpr int maxiter  = 100000; // limit the maximum number of iterations (slow on 1 GPU thread)
   if (currentvolume != NULL) {
     state->Push(currentvolume);
     visitor->apply(state, nav_ind);
     auto ierr = visitor->GetError();
     if (ierr) {
-      printf("=== EEE === TestNavIndex: %s\n", errcodes[ierr]);
+      printf("=== EEE === TestNavIndex: %s\n", errcodes[ierr - 1]);
       return ierr;
     }
     if (visitor->GetNiter() > maxiter) return 0;
     for (auto daughter : currentvolume->GetDaughters()) {
       auto nav_ind_d = NavStateIndex::PushImpl(nav_ind, daughter);
-      visitAllPlacedVolumesPassNavIndex(daughter, visitor, state, nav_ind_d);
+      ierr           = visitAllPlacedVolumesPassNavIndex(daughter, visitor, state, nav_ind_d);
+      if (ierr > 0) return ierr;
       if (visitor->GetNiter() > maxiter) return 0;
     }
     state->Pop();
@@ -139,11 +139,11 @@ int visitAllPlacedVolumesPassNavIndex(VPlacedVolume const *currentvolume, Visito
 
 } // namespace visitorcuda
 
-__global__
-void TestNavIndexGPUKernel(vecgeom::cuda::VPlacedVolume const* const gpu_world, vecgeom::cuda::NavStatePath * const state, int *ierr)
+__global__ void TestNavIndexGPUKernel(vecgeom::cuda::VPlacedVolume const *const gpu_world,
+                                      vecgeom::cuda::NavStatePath *const state, int *ierr)
 {
   using namespace visitorcuda;
-  
+
   state->Clear();
   GlobalToLocalVisitor visitor;
 
@@ -152,14 +152,14 @@ void TestNavIndexGPUKernel(vecgeom::cuda::VPlacedVolume const* const gpu_world,
   *ierr = visitAllPlacedVolumesPassNavIndex(gpu_world, &visitor, state, nav_ind_top);
 }
 
-int TestNavIndexGPU(vecgeom::cxx::VPlacedVolume const* const world, int maxdepth)
+int TestNavIndexGPU(vecgeom::cxx::VPlacedVolume const *const world, int maxdepth)
 {
   // Load and synchronize the geometry on the GPU
   size_t statesize = NavigationState::SizeOfInstance(maxdepth);
 
   vecgeom::cxx::CudaManager::Instance().LoadGeometry(world);
   vecgeom::cxx::CudaManager::Instance().Synchronize();
-  
+
   auto gpu_world = vecgeom::cxx::CudaManager::Instance().world_gpu();
   assert(gpu_world && "GPU world volume is a null pointer");
 
@@ -182,9 +182,7 @@ int TestNavIndexGPU(vecgeom::cxx::VPlacedVolume const* const world, int maxdepth
 
   checkCudaErrors(cudaFree(input_buffer));
   auto tvalidate = timer.Stop();
-  if (!ierr)
-    std::cout << "=== Info navigation table validation on GPU took: " << tvalidate << " sec.\n";
+  if (!ierr) std::cout << "=== Info navigation table validation on GPU took: " << tvalidate << " sec.\n";
 
   return ierr;
 }
-