add make support for magma

CMakeLists.txt

@@ -215,6 +215,7 @@ if (MFEM_USE_HIP)
   endif()
   find_package(HIP REQUIRED)
   find_package(HIPSPARSE REQUIRED)
+  find_package(HIPBLAS REQUIRED)
 endif()
 
 # OpenMP
@@ -482,6 +483,11 @@ if (MFEM_USE_ENZYME)
    find_package(ENZYME REQUIRED)
 endif()
 
+# MAGMA
+if (MFEM_USE_MAGMA)
+   find_package(MAGMA REQUIRED)
+endif()
+
 # MFEM_TIMER_TYPE
 if (NOT DEFINED MFEM_TIMER_TYPE)
   if (APPLE)
@@ -522,7 +528,7 @@ set(MFEM_TPLS OPENMP HYPRE LAPACK BLAS SuperLUDist STRUMPACK METIS SuiteSparse
     SUNDIALS PETSC SLEPC MUMPS AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB
     NETCDF MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE
     ADIOS2 CUBLAS CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG
-    MPI_CXX HIP HIPSPARSE MOONOLITH BLITZ ALGOIM ENZYME)
+    MPI_CXX HIP HIPSPARSE HIPBLAS MOONOLITH BLITZ ALGOIM ENZYME MAGMA)
 
 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")

config/cmake/MFEMConfig.cmake.in

@@ -61,6 +61,7 @@ set(MFEM_USE_ALGOIM @MFEM_USE_ALGOIM@)
 set(MFEM_USE_BENCHMARK @MFEM_USE_BENCHMARK@)
 set(MFEM_USE_PARELAG @MFEM_USE_PARELAG@)
 set(MFEM_USE_ENZYME @MFEM_USE_ENZYME@)
+set(MFEM_USE_MAGMA @MFEM_USE_MAGMA@)
 
 set(MFEM_CXX_COMPILER "@CMAKE_CXX_COMPILER@")
 set(MFEM_CXX_FLAGS "@CMAKE_CXX_FLAGS@")

config/cmake/config.hpp.in

@@ -193,4 +193,8 @@
 // Enable Enzyme for AD
 #cmakedefine MFEM_USE_ENZYME
 
+// Enable MAGMA linear algebra package
+#cmakedefine MFEM_USE_MAGMA
+
+
 #endif // MFEM_CONFIG_HEADER

config/cmake/modules/FindMAGMA.cmake

@@ -0,0 +1,19 @@
+# Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
+# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+# LICENSE and NOTICE for details. LLNL-CODE-806117.
+#
+# This file is part of the MFEM library. For more information and source code
+# availability visit https://mfem.org.
+#
+# MFEM is free software; you can redistribute it and/or modify it under the
+# terms of the BSD-3 license. We welcome feedback and contributions, see file
+# CONTRIBUTING.md for details.
+
+# Defines the following variables:
+#   - MAGMA_FOUND
+#   - MAGMA_LIBRARIES
+#   - MAGMA_INCLUDE_DIRS
+
+include(MfemCmakeUtilities)
+mfem_find_package(MAGMA MAGMA MAGMA_DIR "include" "magma_v2.h" "lib" "magma;magma_sparse"
+  "Paths to headers required by MAGMA." "Libraries required by MAGMA.")

config/cmake/modules/MfemCmakeUtilities.cmake

@@ -873,7 +873,7 @@ function(mfem_export_mk_files)
       MFEM_USE_HIP MFEM_USE_RAJA MFEM_USE_OCCA MFEM_USE_CEED MFEM_USE_CALIPER
       MFEM_USE_UMPIRE MFEM_USE_SIMD MFEM_USE_ADIOS2 MFEM_USE_MKL_CPARDISO
       MFEM_USE_ADFORWARD MFEM_USE_CODIPACK MFEM_USE_BENCHMARK MFEM_USE_PARELAG
-      MFEM_USE_MOONOLITH MFEM_USE_ALGOIM MFEM_USE_ENZYME)
+      MFEM_USE_MOONOLITH MFEM_USE_ALGOIM MFEM_USE_ENZYME MFEM_USE_MAGMA)
   foreach(var ${CONFIG_MK_BOOL_VARS})
     if (${var})
       set(${var} YES)

config/config.hpp.in

@@ -198,4 +198,8 @@
 // Enable the Enzyme LLVM plugin
 // #define MFEM_USE_ENZYME
 
+// Enable the MAGMA package
+// #define MFEM_USE_MAGMA
+
+
 #endif // MFEM_CONFIG_HEADER

config/config.mk.in

@@ -63,6 +63,7 @@ MFEM_USE_CODIPACK      = @MFEM_USE_CODIPACK@
 MFEM_USE_BENCHMARK     = @MFEM_USE_BENCHMARK@
 MFEM_USE_PARELAG       = @MFEM_USE_PARELAG@
 MFEM_USE_ENZYME        = @MFEM_USE_ENZYME@
+MFEM_USE_MAGMA         = @MFEM_USE_MAGMA@
 
 # Compiler, compile options, and link options
 MFEM_CXX       = @MFEM_CXX@

config/defaults.cmake

@@ -65,6 +65,7 @@ option(MFEM_USE_CODIPACK "Enable automatic differentiation (AD) using CoDiPack"
 option(MFEM_USE_BENCHMARK "Enable Google Benchmark" OFF)
 option(MFEM_USE_PARELAG "Enable ParELAG" OFF)
 option(MFEM_USE_ENZYME "Enable Enzyme" OFF)
+option(MFEM_USE_MAGMA "Enable MAGMA" OFF)
 
 # Optional overrides for autodetected MPIEXEC and MPIEXEC_NUMPROC_FLAG
 # set(MFEM_MPIEXEC "mpirun" CACHE STRING "Command for running MPI tests")
@@ -256,6 +257,8 @@ set(CODIPACK_INCLUDE_DIRS "${MFEM_DIR}/../CoDiPack/include" CACHE STRING
     "Path to CoDiPack headers.")
 set(CODIPACK_LIBRARIES "")
 
+set(MAGMA_DIR "${MFEM_DIR}/../magma" CACHE PATH "Path to MAGMA")
+
 # Some useful variables:
 set(CMAKE_SKIP_PREPROCESSED_SOURCE_RULES ON) # Skip *.i rules
 set(CMAKE_SKIP_ASSEMBLY_SOURCE_RULES  ON)    # Skip *.s rules

config/defaults.mk

@@ -166,6 +166,7 @@ MFEM_USE_CODIPACK      = NO
 MFEM_USE_BENCHMARK     = NO
 MFEM_USE_PARELAG       = NO
 MFEM_USE_ENZYME        = NO
+MFEM_USE_MAGMA         = NO
 
 # MPI library compile and link flags
 # These settings are used only when building MFEM with MPI + HIP
@@ -459,11 +460,11 @@ GSLIB_LIB = -L$(GSLIB_DIR)/lib -lgs
 
 # CUDA library configuration
 CUDA_OPT =
-CUDA_LIB = -lcusparse
+CUDA_LIB = -lcusparse -lcublas
 
 # HIP library configuration
 HIP_OPT =
-HIP_LIB = -L$(HIP_DIR)/lib $(XLINKER)-rpath,$(HIP_DIR)/lib -lhipsparse
+HIP_LIB = -L$(HIP_DIR)/lib $(XLINKER)-rpath,$(HIP_DIR)/lib -lhipsparse -lhipblas
 
 # OCCA library configuration
 OCCA_DIR = @MFEM_DIR@/../occa
@@ -560,6 +561,10 @@ ENZYME_VERSION ?= 14
 ENZYME_OPT = -fno-experimental-new-pass-manager -Xclang -load -Xclang $(ENZYME_DIR)/ClangEnzyme-$(ENZYME_VERSION).so
 ENZYME_LIB = ""
 
+MAGMA_DIR ?= @MFEM_DIR@/../magma
+MAGMA_OPT = -I${MAGMA_DIR}/include/
+MAGMA_LIB = -L${MAGMA_DIR}/lib/ -lmagma -lmagma_sparse
+
 # If YES, enable some informational messages
 VERBOSE = NO
 

linalg/CMakeLists.txt

@@ -27,6 +27,7 @@ list(APPEND SRCS
   sparsemat.cpp
   sparsesmoothers.cpp
   vector.cpp
+  libBatchSolver.cpp
   )
 
 list(APPEND HDRS
@@ -57,6 +58,7 @@ list(APPEND HDRS
   tensor.hpp
   dual.hpp
   vector.hpp
+  libBatchSolver.hpp
   )
 
 if (MFEM_USE_MPI)

linalg/libBatchSolver.cpp

@@ -0,0 +1,115 @@
+// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#include "libBatchSolver.hpp"
+
+namespace mfem
+{
+
+void LibBatchMult::Mult(const Vector &b, Vector &x)
+{
+
+#if defined(MFEM_USE_CUDA) || defined(MFEM_USE_HIP)
+   if (Device::Allows(Backend::DEVICE_MASK))
+   {
+#if defined(MFEM_USE_MAGMA)
+      const double alpha = 1.0, beta = 0.0;
+#else
+      const double alpha_t = 1.0, beta_t = 0.0;
+      const double* alpha = &alpha_t;
+      const double* beta = &beta_t;
+#endif
+
+#if defined(MFEM_USE_HIP) || (defined(MFEM_USE_CUDA) && CUDA_VERSION >= 11070) || defined(MFEM_USE_MAGMA)
+#if defined(MFEM_USE_MAGMA)
+      magmablas_dgemv_batched_strided(MagmaNoTrans,
+#else
+      MFEM_SUB_cu_or_hip(blasStatus_t)
+      status = MFEM_SUB_cu_or_hip(blasDgemvStridedBatched)(MFEM_SUB_Cuda_or_Hip(
+                                                              BLAS::Handle)(),
+                                                           MFEM_SUB_CU_or_HIP(BLAS_OP_N),
+#endif
+                                      mat_size,
+                                      mat_size,
+                                      alpha,
+                                      MatrixBatch.Read(),
+                                      mat_size,
+                                      mat_size * mat_size,
+                                      b.Read(),
+                                      1,
+                                      mat_size,
+                                      beta,
+                                      x.Write(),
+                                      1,
+                                      mat_size,
+                                      num_mats
+#if defined(MFEM_USE_MAGMA)
+                                      , queue
+#endif
+                                     );
+#else
+      MFEM_SUB_cu_or_hip(blasStatus_t)
+      status = MFEM_SUB_cu_or_hip(blasDgemmStridedBatched)(MFEM_SUB_Cuda_or_Hip(
+                                                              BLAS::Handle)(),
+                                                           MFEM_SUB_CU_or_HIP(BLAS_OP_N),
+                                                           MFEM_SUB_CU_or_HIP(BLAS_OP_N),
+                                                           mat_size,
+                                                           1,
+                                                           mat_size,
+                                                           alpha,
+                                                           MatrixBatch.Read(),
+                                                           mat_size,
+                                                           mat_size * mat_size,
+                                                           b.Read(),
+                                                           mat_size,
+                                                           mat_size,
+                                                           beta,
+                                                           x.Write(),
+                                                           mat_size,
+                                                           mat_size,
+                                                           num_mats);
+
+
+#endif
+
+#if !defined(MFEM_USE_MAGMA)
+      MFEM_VERIFY(status == MFEM_SUB_CU_or_HIP(BLAS_STATUS_SUCCESS),
+                  "blasDgemvStridedBatched");
+#endif
+   }
+   else
+#endif
+   {
+      const int rows = MatrixBatch.SizeI();
+      const int cols = MatrixBatch.SizeJ();
+      const int N = MatrixBatch.SizeK();
+
+      for (int e=0; e<N; ++e)
+      {
+
+         //matvec
+         for (int c=0; c<cols; ++c)
+         {
+            double dot=0.0;
+            for (int r=0; r<rows; ++r)
+            {
+               int idx = r  +  rows * e;
+               dot += MatrixBatch(c, r, e) * b(idx);
+            }
+            int idx = c + cols * e;
+            x(idx) = dot;
+         }
+      }
+   }
+}
+
+
+} //mfem namespace

linalg/libBatchSolver.hpp

@@ -0,0 +1,115 @@
+// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
+// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
+// LICENSE and NOTICE for details. LLNL-CODE-806117.
+//
+// This file is part of the MFEM library. For more information and source code
+// availability visit https://mfem.org.
+//
+// MFEM is free software; you can redistribute it and/or modify it under the
+// terms of the BSD-3 license. We welcome feedback and contributions, see file
+// CONTRIBUTING.md for details.
+
+#ifndef MFEM_LIB_BATCH_SOLVER
+#define MFEM_LIB_BATCH_SOLVER
+
+#include "../config/config.hpp"
+#include "../general/globals.hpp"
+#include "matrix.hpp"
+#include "densemat.hpp"
+
+#if defined(MFEM_USE_CUDA)
+// These macros prefixes the argument with `cuda` or `hip`, e.g. `cudaStream_t` or `hipStream_t`
+#define MFEM_SUB_cuda_or_hip(stub) cuda##stub
+#define MFEM_SUB_cu_or_hip(stub) cu##stub
+#define MFEM_SUB_Cuda_or_Hip(stub) Cuda##stub
+#define MFEM_SUB_CUDA_or_HIP(stub) CUDA##stub
+#define MFEM_SUB_CU_or_HIP(stub) CU##stub
+
+#include <cuda.h>
+#include <cublas.h>
+#include <cusolverDn.h>
+#include <cuda_runtime.h>
+#endif  // MFEM_USE_CUDA
+
+#if defined(MFEM_USE_HIP)
+// These macros prefixes the argument with `cuda` or `hip`, e.g. `cudaStream_t` or `hipStream_t`
+#define MFEM_SUB_cu_or_hip(stub) hip##stub
+#define MFEM_SUB_cuda_or_hip(stub) hip##stub
+#define MFEM_SUB_Cuda_or_Hip(stub) Hip##stub
+#define MFEM_SUB_CUDA_or_HIP(stub) HIP##stub
+#define MFEM_SUB_CU_or_HIP(stub) HIP##stub
+#include <hipblas/hipblas.h>
+#include <hip/hip_runtime.h>
+#include <hip/hip_runtime_api.h>
+#endif  // MFEM_USE_HIP
+
+#if defined(MFEM_USE_MAGMA)
+#include <magma_v2.h>
+#endif
+
+namespace mfem
+{
+
+#if defined(MFEM_USE_CUDA) || defined(MFEM_USE_HIP)
+class MFEM_SUB_Cuda_or_Hip(BLAS)
+{
+protected:
+   MFEM_SUB_cu_or_hip(blasHandle_t) handle = nullptr;
+
+   MFEM_SUB_Cuda_or_Hip(BLAS)()
+   {
+      MFEM_SUB_cu_or_hip(blasStatus_t) status = MFEM_SUB_cu_or_hip(blasCreate)(
+                                                   &handle);
+      MFEM_VERIFY(status == MFEM_SUB_CU_or_HIP(BLAS_STATUS_SUCCESS),
+                  "Cannot initialize BLAS.");
+   }
+
+   ~MFEM_SUB_Cuda_or_Hip(BLAS)() { MFEM_SUB_cu_or_hip(blasDestroy)(handle); }
+
+   static MFEM_SUB_Cuda_or_Hip(BLAS) & Instance()
+   {
+      static MFEM_SUB_Cuda_or_Hip(BLAS) instance;
+      return instance;
+   }
+
+public:
+   static MFEM_SUB_cu_or_hip(blasHandle_t) Handle() { return Instance().handle; }
+};
+#endif
+
+/**
+   Use GPU library calls to perform action of block diagonal matrices
+**/
+class LibBatchMult
+{
+
+private:
+   DenseTensor &MatrixBatch;
+
+   int mat_size, num_mats;
+#if defined(MFEM_USE_MAGMA)
+   magma_int_t device = 0;
+   magma_queue_t queue ;
+#endif
+public:
+   LibBatchMult() = delete;
+
+   LibBatchMult(DenseTensor &MatrixBatch_) :
+      MatrixBatch(MatrixBatch_),
+      mat_size(MatrixBatch.SizeI()),
+      num_mats(MatrixBatch.SizeK())
+   {
+#if defined(MFEM_USE_MAGMA)
+      magma_setdevice(device);
+      magma_queue_create(device, &queue);
+#endif
+   };
+
+   //Action of block diagonal matrix
+   void Mult(const Vector &b, Vector &x);
+};
+
+
+} // namespace mfem
+
+#endif

linalg/linalg.hpp

@@ -18,6 +18,7 @@
 #include "operator.hpp"
 #include "matrix.hpp"
 #include "sparsemat.hpp"
+#include "libBatchSolver.hpp"
 #include "complex_operator.hpp"
 #include "complex_densemat.hpp"
 #include "blockvector.hpp"

makefile

@@ -278,7 +278,7 @@ endif
 MFEM_REQ_LIB_DEPS = ENZYME SUPERLU MUMPS METIS FMS CONDUIT SIDRE LAPACK SUNDIALS\
  SUITESPARSE STRUMPACK GINKGO GNUTLS NETCDF PETSC SLEPC MPFR PUMI HIOP\
  GSLIB OCCA CEED RAJA UMPIRE MKL_CPARDISO AMGX CALIPER PARELAG BENCHMARK\
- MOONOLITH ALGOIM
+ MOONOLITH ALGOIM MAGMA
 
 
 PETSC_ERROR_MSG = $(if $(PETSC_FOUND),,. PETSC config not found: $(PETSC_VARS))
@@ -347,7 +347,7 @@ MFEM_DEFINES = MFEM_VERSION MFEM_VERSION_STRING MFEM_GIT_STRING MFEM_USE_MPI\
  MFEM_USE_SIMD MFEM_USE_ADIOS2 MFEM_USE_MKL_CPARDISO MFEM_USE_AMGX\
  MFEM_USE_MUMPS MFEM_USE_ADFORWARD MFEM_USE_CODIPACK MFEM_USE_CALIPER\
  MFEM_USE_BENCHMARK MFEM_USE_PARELAG MFEM_USE_ALGOIM MFEM_USE_ENZYME\
- MFEM_SOURCE_DIR MFEM_INSTALL_DIR MFEM_SHARED_BUILD
+ MFEM_USE_MAGMA MFEM_SOURCE_DIR MFEM_INSTALL_DIR MFEM_SHARED_BUILD
 
 # List of makefile variables that will be written to config.mk:
 MFEM_CONFIG_VARS = MFEM_CXX MFEM_HOST_CXX MFEM_CPPFLAGS MFEM_CXXFLAGS\
@@ -710,6 +710,7 @@ status info:
 	$(info MFEM_USE_BENCHMARK     = $(MFEM_USE_BENCHMARK))
 	$(info MFEM_USE_PARELAG       = $(MFEM_USE_PARELAG))
 	$(info MFEM_USE_ENZYME        = $(MFEM_USE_ENZYME))
+	$(info MFEM_USE_MAGMA         = $(MFEM_USE_MAGMA))
 	$(info MFEM_CXX               = $(value MFEM_CXX))
 	$(info MFEM_HOST_CXX          = $(value MFEM_HOST_CXX))
 	$(info MFEM_CPPFLAGS          = $(value MFEM_CPPFLAGS))