Update CHANGES for 21.10 (#2372)

CHANGES

@@ -1,3 +1,62 @@
+# 21.10
+
+  -- Update for oneapi 2021.4.0 (#2368)
+
+  -- Port CellConservativeQuartic to C++ with GPU support (#2362)
+
+  -- Port CellConservativeProtected::protect() to C++ with GPU support (#2347)
+
+  -- CMake CUDA Debug: -G by Default (#2355)
+
+  -- OverlapSync nowait and finish. (#2346)
+
+  -- Add another overload for PC::Checkpoint (#2353)
+
+  -- Call setLevelBC from getFluxes in case we have set up the projector
+     class but not yet called project (#2345)
+
+  -- Function to atomically set particle ids (#2344)
+
+  -- Remove macro: CRSEGRNDOMP and AMREX_CRSEGRNDOMP (#2339)
+
+  -- Don't rely on mananaged memory in PC::InitOnePerCell (#2337)
+
+  -- TinyProfile parser that saves data in a machine-readable format (#2335)
+
+  -- Fix return type of MLEBNodeFDLaplacian::setEBDirichlet (#2329)
+
+  -- ParReduce (#2327)
+
+  -- Remove some versions of ParallelFor(MF) (#2326)
+
+  -- Update MLMG solver to allow variable number of ghost nodes for variable refinement ratios (#2319)
+
+  -- CMake: Always CUDA Perf-Neutral Debug (#2321)
+
+  -- Add version of SumBoundary that takes src_nghost (#2322)
+
+  -- Remove the Old Fusing Approach (#2308)
+
+  -- Add const to Array's sum and product functions (#2316)
+
+  -- Fix signed/unsigned comparison in diagnostic neighbor list function. (#2312)
+
+  -- HDF5 I/O and Compression  (#2220)
+
+  -- Use ParallelFor(Tag) in Linear Solver Preparation (#2297)
+
+  -- TagBoxArray: Remove omp parallel inside GPU launch region (#2309)
+
+  -- Update DiffMultiFab tool (#2307)
+
+  -- Kernel fusing in linear solver's applyBC (#2295)
+
+  -- Adding new parallel for into normalize method of node laplacian. (#2299)
+
+  -- Gpu::Atomic::If (#2301)
+
+  -- add adaptor and bridge for amrMesh+Particles (#2285, #2364)
+
 # 21.09
 
   -- Migrate Fortran kernles in the Advection_AmrLevel test to C++. (#2296,