Merge branch 'development'

AMReX-Codes · Mar 2, 2019 · 91323f2 · 91323f2
2 parents 749721f + 6e1acca
commit 91323f2
Show file tree

Hide file tree

Showing 7 changed files with 44 additions and 21 deletions.
diff --git a/Src/Base/AMReX_BLassert.H b/Src/Base/AMReX_BLassert.H
@@ -48,4 +48,11 @@
 #define AMREX_ALWAYS_ASSERT_WITH_MESSAGE(EX,MSG) (EX)?((void)0):amrex::Assert( # EX , __FILE__, __LINE__ , # MSG)
 #define AMREX_ALWAYS_ASSERT(EX) (EX)?((void)0):amrex::Assert( # EX , __FILE__, __LINE__)
 
+
+#if !defined(AMREX_USE_GPU) || !defined(AMREX_DEBUG) && !defined(AMREX_USE_ASSERTION)
+#define AMREX_GPU_ASSERT(EX) ((void)0)
+#else
+#define AMREX_GPU_ASSERT(EX) (EX)?((void)0):amrex::Assert( # EX , __FILE__, __LINE__)
+#endif
+
 #endif /*BL_BL_ASSERT_H*/
diff --git a/Src/Base/AMReX_CArena.H b/Src/Base/AMReX_CArena.H
@@ -5,6 +5,8 @@
 #include <set>
 #include <vector>
 #include <mutex>
+#include <unordered_set>
+#include <functional>
 
 #include <AMReX_Arena.H>
 
@@ -89,6 +91,12 @@ protected:
             return m_owner == rhs.m_owner;
         }
 
+        struct hash {
+            std::size_t operator() (const Node& n) const noexcept {
+                return std::hash<void*>{}(n.m_block);
+            }
+        };
+
     private:
         //! The block of memory we reference.
         void* m_block;
@@ -117,7 +125,8 @@ protected:
     * \brief The list of busy blocks.
     * A block is either on the freelist or on the blocklist, but not on both.
     */
-    NL m_busylist;
+//    NL m_busylist;
+    std::unordered_set<Node, Node::hash> m_busylist;
     //! The minimal size of hunks to request via ::operator new().
     std::size_t m_hunk;
     //! The amount of heap space currently allocated.

diff --git a/Src/Base/AMReX_CArena.cpp b/Src/Base/AMReX_CArena.cpp
@@ -116,7 +116,7 @@ CArena::free (void* vp)
     //
     // `vp' had better be in the busy list.
     //
-    NL::iterator busy_it = m_busylist.find(Node(vp,0,0));
+    auto busy_it = m_busylist.find(Node(vp,0,0));
 
     BL_ASSERT(!(busy_it == m_busylist.end()));
     BL_ASSERT(m_freelist.find(*busy_it) == m_freelist.end());

diff --git a/Src/Base/AMReX_FabArray.H b/Src/Base/AMReX_FabArray.H
@@ -782,7 +782,9 @@ FabArray<FAB>::fabDevicePtr (const MFIter& mfi)
 {
     BL_ASSERT(mfi.LocalIndex() < indexArray.size());
     BL_ASSERT(DistributionMap() == mfi.DistributionMap());
-    return m_fabs_v[mfi.LocalIndex()];
+    int li = mfi.LocalIndex();
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
+    return m_fabs_v[li];
 }
 
 template <class FAB>
@@ -791,7 +793,9 @@ FabArray<FAB>::fabDevicePtr (const MFIter& mfi) const
 {
     BL_ASSERT(mfi.LocalIndex() < indexArray.size());
     BL_ASSERT(DistributionMap() == mfi.DistributionMap());
-    return m_fabs_v[mfi.LocalIndex()];
+    int li = mfi.LocalIndex();
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
+    return m_fabs_v[li];
 }
 
 template <class FAB>
@@ -800,6 +804,7 @@ FabArray<FAB>::fabDevicePtr (int K)
 {
     int li = localindex(K);
     BL_ASSERT(li >=0 && li < indexArray.size());
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
     return m_fabs_v[li];
 }
 
@@ -818,10 +823,12 @@ FabArray<FAB>::fabHostPtr (const MFIter& mfi)
 {
     BL_ASSERT(mfi.LocalIndex() < indexArray.size());
     BL_ASSERT(DistributionMap() == mfi.DistributionMap());
+    int li = mfi.LocalIndex();
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
 #if AMREX_USE_GPU
-    return m_host_fabs_v[mfi.LocalIndex()];
+    return m_host_fabs_v[li];
 #else
-    return m_fabs_v[mfi.LocalIndex()];
+    return m_fabs_v[li];
 #endif
 }
 
@@ -831,10 +838,12 @@ FabArray<FAB>::fabHostPtr (const MFIter& mfi) const
 {
     BL_ASSERT(mfi.LocalIndex() < indexArray.size());
     BL_ASSERT(DistributionMap() == mfi.DistributionMap());
+    int li = mfi.LocalIndex();
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
 #if AMREX_USE_GPU
-    return m_host_fabs_v[mfi.LocalIndex()];
+    return m_host_fabs_v[li];
 #else
-    return m_fabs_v[mfi.LocalIndex()];
+    return m_fabs_v[li];
 #endif
 }
 
@@ -844,6 +853,7 @@ FabArray<FAB>::fabHostPtr (int K)
 {
     int li = localindex(K);
     BL_ASSERT(li >=0 && li < indexArray.size());
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
 #if AMREX_USE_GPU
     return m_host_fabs_v[li];
 #else
@@ -857,6 +867,7 @@ FabArray<FAB>::fabHostPtr (int K) const
 {
     int li = localindex(K);
     BL_ASSERT(li >=0 && li < indexArray.size());
+    AMREX_GPU_ASSERT(m_fabs_v[li]->dataPtr() == m_host_fabs_v[li]->dataPtr());
 #if AMREX_USE_GPU
     return m_host_fabs_v[li];
 #else

diff --git a/Src/Base/AMReX_FabArrayCommI.H b/Src/Base/AMReX_FabArrayCommI.H
@@ -301,8 +301,6 @@ FabArray<FAB>::FBEP_nowait (int scomp, int ncomp, const IntVect& nghost,
 
             if (Gpu::inLaunchRegion())
             {
-                LayoutData<Array<Vector<FabCopyTag<FAB> >,AMREX_SPACEDIM> >
-                    dtags(loc_copy_tags.boxArray(), loc_copy_tags.DistributionMap());
                 for (MFIter mfi(*this); mfi.isValid(); ++mfi)
                 {
                     FAB* dfab = this->fabPtr(mfi);
@@ -318,7 +316,7 @@ FabArray<FAB>::FBEP_nowait (int scomp, int ncomp, const IntVect& nghost,
                             gbx_lo.grow(jdim,nghost[jdim]);
                         }
                         const int ncells_to_hi = vbx.length(idim)+nghost[idim];
-                        const Box gbx_hi = gbx_lo + IntVect{ncells_to_hi};
+                        const Box gbx_hi = amrex::shift(gbx_lo, idim, ncells_to_hi);
 
                         Vector<FabCopyTag<FAB> > tags_lo, tags_hi;
                         for (const auto& lc_tag : lc_tags) {
@@ -1503,8 +1501,6 @@ FillBoundary (Vector<FabArray<FAB>*> const& mf, const Periodicity& period)
 
                     if (Gpu::inLaunchRegion())
                     {
-                        LayoutData<Array<Vector<FabCopyTag<FAB> >,AMREX_SPACEDIM> >
-                            dtags(loc_copy_tags.boxArray(), loc_copy_tags.DistributionMap());
                         for (MFIter mfi(fa); mfi.isValid(); ++mfi)
                         {
                             FAB* dfab = fa.fabPtr(mfi);
@@ -1520,7 +1516,7 @@ FillBoundary (Vector<FabArray<FAB>*> const& mf, const Periodicity& period)
                                     gbx_lo.grow(jdim,ng[jdim]);
                                 }
                                 const int ncells_to_hi = vbx.length(idim)+ng[idim];
-                                const Box gbx_hi = gbx_lo + IntVect{ncells_to_hi};
+                                const Box gbx_hi = amrex::shift(gbx_lo, idim, ncells_to_hi);
 
                                 Vector<FabCopyTag<FAB> > vtags_lo, vtags_hi;
                                 for (const auto& lc_tag : lc_tags) {

diff --git a/Src/LinearSolvers/MLMG/AMReX_MLEBABecLap.cpp b/Src/LinearSolvers/MLMG/AMReX_MLEBABecLap.cpp
@@ -603,7 +603,7 @@ MLEBABecLap::FFlux (int amrlev, const MFIter& mfi, const Array<FArrayBox*,AMREX_
                     const FArrayBox& sol, Location loc, const int face_only) const
 {
     // todo: gpu
-    Gpu::LaunchSafeGuard(false);
+    Gpu::LaunchSafeGuard lg(false);
     BL_PROFILE("MLEBABecLap::FFlux()");
     const int at_centroid = (Location::FaceCentroid == loc) ? 1 : 0;
     const int mglev = 0; 
@@ -665,7 +665,7 @@ MLEBABecLap::compGrad (int amrlev, const Array<MultiFab*,AMREX_SPACEDIM>& grad,
                        MultiFab& sol, Location loc) const
 {
     // todo: gpu
-    Gpu::LaunchSafeGuard(false);
+    Gpu::LaunchSafeGuard lg(false);
     BL_PROFILE("MLEBABecLap::compGrad()");
 
     const int at_centroid = (Location::FaceCentroid == loc) ? 1 : 0;
@@ -753,7 +753,7 @@ void
 MLEBABecLap::normalize (int amrlev, int mglev, MultiFab& mf) const
 {
     // todo: gpu
-    Gpu::LaunchSafeGuard(false);
+    Gpu::LaunchSafeGuard lg(false);
     const MultiFab& acoef = m_a_coeffs[amrlev][mglev];
     AMREX_D_TERM(const MultiFab& bxcoef = m_b_coeffs[amrlev][mglev][0];,
                  const MultiFab& bycoef = m_b_coeffs[amrlev][mglev][1];,
@@ -840,7 +840,7 @@ void
 MLEBABecLap::interpolation (int amrlev, int fmglev, MultiFab& fine, const MultiFab& crse) const
 {
     // todo: gpu
-    Gpu::LaunchSafeGuard(false);
+    Gpu::LaunchSafeGuard lg(false);
     auto factory = dynamic_cast<EBFArrayBoxFactory const*>(m_factory[amrlev][fmglev].get());
     const FabArray<EBCellFlagFab>* flags = (factory) ? &(factory->getMultiEBCellFlagFab()) : nullptr;
 
@@ -891,7 +891,7 @@ MLEBABecLap::applyBC (int amrlev, int mglev, MultiFab& in, BCMode bc_mode, State
                       const MLMGBndry* bndry, bool skip_fillboundary) const
 {
     // todo: gpu
-    Gpu::LaunchSafeGuard(false);
+    Gpu::LaunchSafeGuard lg(false);
     BL_PROFILE("MLEBABecLap::applyBC()");
 
     // No coarsened boundary values, cannot apply inhomog at mglev>0.

diff --git a/Src/LinearSolvers/MLMG/AMReX_MLMG.cpp b/Src/LinearSolvers/MLMG/AMReX_MLMG.cpp
@@ -561,7 +561,7 @@ MLMG::interpCorrection (int alev)
 
     if (linop.isCellCentered())
     {
-        Gpu::LaunchSafeGuard(!isEB); // turn off gpu for eb for now TODO
+        Gpu::LaunchSafeGuard lg(!isEB); // turn off gpu for eb for now TODO
         MFItInfo mfi_info;
         if (Gpu::notInLaunchRegion()) mfi_info.EnableTiling().SetDynamic(true);
 #ifdef _OPENMP
@@ -692,7 +692,7 @@ MLMG::interpCorrection (int alev, int mglev)
 
     if (linop.isCellCentered())
     {
-        Gpu::LaunchSafeGuard(!isEB); // turn off gpu for eb for now TODO
+        Gpu::LaunchSafeGuard lg(!isEB); // turn off gpu for eb for now TODO
         MFItInfo mfi_info;
         if (Gpu::notInLaunchRegion()) mfi_info.EnableTiling().SetDynamic(true);
 #ifdef _OPENMP