From 1cb99250ecd4da445cfc8afe325664030c49c4f0 Mon Sep 17 00:00:00 2001
From: Weiqun Zhang <WeiqunZhang@lbl.gov>
Date: Tue, 1 Dec 2020 10:12:38 -0800
Subject: [PATCH] Update CHANGES (#1584)

---
 CHANGES | 74 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 74 insertions(+)

diff --git a/CHANGES b/CHANGES
index 4b23a0ba44b..5c57036a255 100644
--- a/CHANGES
+++ b/CHANGES
@@ -1,3 +1,77 @@
+# 20.12
+
+  -- EB2::PolyIF (#1567) The existing PolynomialIF contains a dynamic size
+     vector.  This makes it hard to support it on GPU.  This new PolyIF
+     class uses fixed size array and has GPU support.
+
+  -- Tools/Plotfile works for for single precision now. (#1577)
+
+  -- Arena: Add accessor to query Arena attributes through info object
+     (#1575)
+
+  -- Allow MacProjection reuse (#1574)
+
+  -- Remove SUNDIALS Fortran interface support (#1559)
+
+  -- Allow MLEBABecLap to solve for multiple components (#1557)
+
+  -- DPCPP: MPI (#1556) Allow MPI for dpcpp build in gnu make assuming
+     mpiicpx is used.
+
+  -- Non-local boundary conditions (#1544)
+
+  -- If TheZeroVector() is passed in to SortParticlesByBin, do nothing
+     instead of crashing. (#1564)
+
+  -- Scan size type (#1521) Make the size type a template parameter to
+     support the scan of more than INT_MAX elements.
+
+  -- DPCPP: Fortran (#1554) Switch from gfortran to ifx that comes with
+     oneapi in GNU make.
+
+  -- CMake: generate_buildinfo target (#1545) This commit generates a static
+     library `buildInfo::<target>` for the passed user target. This target
+     is then automatically linked.  Also, the generated `.cpp` file is now
+     scoped properly, avoiding collisions.
+
+  -- nvcc --Werror cross-execution-space-call (#1540) This is turned on by
+     default in GNU make, and can be turned on in CMake.
+
+  -- fcompare: Add option for an absolute tolerance check (#1537)
+
+  -- Remove call to cudaDeviceSetCacheConfig (#1536) No longer set CUDA
+     device cache configuration to prefer L1 cache.
+
+  -- Assert no nested MFIters (#1530) Add assertion to catch nested MFIters
+     (e.g., MultiFab functions are called inside MFIter).
+
+  -- Free MPI_Datatype defined by AMReX (#1527) This fixes a one-time memory
+     leak in MPI_Datatypes defined by AMReX as reported in #1525.
+
+  -- make clean & make cleanconfig (#1520) `make clean` now does what `make
+     realclean` does.  `make realclean` is kept.  `make cleanconfig` is
+     introduced to do what `make clean` used to do.  For most users, `make
+     clean` is the one that should be used.
+
+  -- hypre: Adjust singular matrices before hypre solves (#1519)
+
+  -- CMake: overhaul options (#1490)
+
+  -- New Reduce::Sum function taking lambda function (#1508)
+
+  -- GNU Make: AMREX_CUDA_ARCH (#1507) Option to use AMREX_CUDA_ARCH in
+     addition to CUDA_ARCH to set CUDA arch.  If AMREX_CUDA is set, it will
+     be used over CUDA_ARCH.
+
+  -- Add option for device sync at beginning and end of TinyProfiler region
+     (#1505) Now when setting
+     `tiny_profiler.device_synchronize_on_region_end = 1` in the inputs
+     file, we will synchronize before calling nvtxRangePop() and
+     nvtxRangePush(), which means that TINY_PROFILE regions will include the
+     full kernel time, rather than just the kernel launch time.
+
+  -- CMake; bump up hypre minimi required version (#1503)
+
 # 20.11
 
   -- The default hypre interface in MLMG is now IJ matrix for both EB and