more issues fixed

WeiqunZhang · Jan 24, 2025 · d9253b6 · d9253b6
1 parent 21e7ab3
commit d9253b6
Show file tree

Hide file tree

Showing 4 changed files with 23 additions and 5 deletions.
diff --git a/.github/workflows/windows.yml b/.github/workflows/windows.yml
@@ -161,7 +161,6 @@ jobs:
     - name: Compile
       run: |
         cmake -S . -B build                       `
-              -DCUDAToolkit_ROOT="$Env:CUDA_PATH" `
               -DCMAKE_VERBOSE_MAKEFILE=ON         `
               -DCMAKE_BUILD_TYPE=Release          `
               -DAMReX_GPU_BACKEND=CUDA            `

diff --git a/Src/Base/AMReX_NonLocalBC.H b/Src/Base/AMReX_NonLocalBC.H
@@ -476,6 +476,9 @@ template <typename... Args>
 using LocalCopy_t = decltype(LocalCopy(std::declval<Args>()...));
 
 //! \brief This type trait tests if a given type DP satisfies the DataPacking concept for type FAB.
+#if defined(AMREX_USE_CUDA) && defined(_WIN32)
+template <typename DP, typename FAB> struct IsDataPacking : std::true_type {};
+#else
 template <typename DP, typename FAB>
 struct IsDataPacking :
     Conjunction<
@@ -487,6 +490,7 @@ struct IsDataPacking :
         IsDetected<UnpackRecvBuffers_t, DP&, FabArray<FAB>&, CommData&>
 #endif
     > {};
+#endif
 
 template <class FAB, class DTOS = Identity, class Proj = Identity>
 std::enable_if_t<IsBaseFab<FAB>() && IsCallableR<Dim3, DTOS, Dim3>() && IsFabProjection<Proj, FAB>()>
@@ -602,10 +606,8 @@ UnpackRecvBuffers (const PackComponents& components, FabArray<FAB>& dest, const
 }
 #endif // AMREX_USE_MPI
 
-#ifndef _WIN32
 static_assert(IsDataPacking<PackComponents, FArrayBox>(), // NOLINT(bugprone-throw-keyword-missing)
               "PackComponents is expected to satisfy the concept DataPacking.");
-#endif
 
 ////////////////////////////////////////////////////////////////////////////////////
 //                                    [DataPacking.ApplyDtosAndProjectionOnReciever]
@@ -665,10 +667,8 @@ UnpackRecvBuffers (const ApplyDtosAndProjectionOnReciever<DTOS, FabProj>& packin
 }
 #endif // AMREX_USE_MPI
 
-#ifndef _WIN32
 static_assert(IsDataPacking<ApplyDtosAndProjectionOnReciever<>, FArrayBox>(), // NOLINT(bugprone-throw-keyword-missing)
               "ApplyDtosAndProjectionOnReciever<> is expected to satisfy the DataPacking concept.");
-#endif
 
 ////////////////////////////////////////////////////////////////////////////////////
 //                                                             [ParallelCopy_nowait]

diff --git a/Src/LinearSolvers/MLMG/AMReX_MLABecLaplacian.H b/Src/LinearSolvers/MLMG/AMReX_MLABecLaplacian.H
@@ -672,6 +672,20 @@ MLABecLaplacianT<MF>::averageDownCoeffsSameAmrLevel (int amrlev, Vector<MF>& a,
                              auto const& by = b[mglev][1].array(mfi);,
                              auto const& bz = b[mglev][2].array(mfi));
                 auto const& osm = this->m_overset_mask[amrlev][mglev]->const_array(mfi);
+#if defined(AMREX_USE_CUDA) && defined(_WIN32)
+                amrex::launch(xbx, [=] AMREX_GPU_DEVICE (Box const& t_xbx)
+                {
+                    overset_rescale_bcoef_x(t_xbx, bx, osm, ncomp, osfac);
+                });
+                amrex::launch(ybx, [=] AMREX_GPU_DEVICE (Box const& t_ybx)
+                {
+                    overset_rescale_bcoef_y(t_ybx, by, osm, ncomp, osfac);
+                });
+                amrex::launch(zbx, [=] AMREX_GPU_DEVICE (Box const& t_zbx)
+                {
+                    overset_rescale_bcoef_z(t_zbx, bz, osm, ncomp, osfac);
+                });
+#else
                 AMREX_LAUNCH_HOST_DEVICE_LAMBDA_DIM
                     (xbx, t_xbx,
                      {
@@ -685,6 +699,7 @@ MLABecLaplacianT<MF>::averageDownCoeffsSameAmrLevel (int amrlev, Vector<MF>& a,
                      {
                          overset_rescale_bcoef_z(t_zbx, bz, osm, ncomp, osfac);
                      });
+#endif
             }
         }
     }

diff --git a/Src/Particle/AMReX_ParticleContainerI.H b/Src/Particle/AMReX_ParticleContainerI.H
@@ -1213,7 +1213,11 @@ ParticleContainer_impl<ParticleType, NArrayReal, NArrayInt, Allocator, CellAssig
     const size_t np_total = np + ptile.numNeighborParticles();
 
     if (memEfficientSort) {
+#if defined(AMREX_USE_CUDA) && defined(_WIN32)
+        if           (!ParticleType::is_soa_particle) {
+#else
         if constexpr (!ParticleType::is_soa_particle) {
+#endif
             static_assert(sizeof(ParticleType)%4 == 0 && sizeof(uint32_t) == 4);
             using tmp_t = std::conditional_t<sizeof(ParticleType)%8 == 0,
                                              uint64_t, uint32_t>;