Merge pull request dealii#17474 from masterleinad/remove_cudawrappers

Remove CUDAWrappers, related documentation and macros
mwwt · Aug 16, 2024 · 1223e67 · 1223e67
2 parents 71cca2f + f549fc6
commit 1223e67
Show file tree

Hide file tree

Showing 130 changed files with 60 additions and 10,389 deletions.
diff --git a/.github/workflows/linux.yml b/.github/workflows/linux.yml
@@ -369,7 +369,6 @@ jobs:
               -D CMAKE_CXX_COMPILER=${GITHUB_WORKSPACE}/kokkos/bin/nvcc_wrapper \
               -D DEAL_II_CXX_FLAGS='-Werror -Wno-non-template-friend' \
               -D DEAL_II_EARLY_DEPRECATIONS=ON \
-              -D DEAL_II_WITH_CUDA="ON" \
               -D DEAL_II_WITH_KOKKOS="ON" \
               -D KOKKOS_DIR=${GITHUB_WORKSPACE}/../kokkos-install \
               -D DEAL_II_WITH_MPI="ON" \
@@ -386,10 +385,8 @@ jobs:
     - name: build CUDA tests
       run: |
         cd build
-        make -j2 setup_tests_cuda setup_tests_matrix_free_kokkos
-        cd tests/cuda
-        make -j2 compile_test_executables
-        cd ../matrix_free_kokkos
+        make -j2 setup_tests_matrix_free_kokkos
+        cd tests/matrix_free_kokkos
         make -j2 compile_test_executables
 
   #############################
@@ -463,7 +460,6 @@ jobs:
               -D DEAL_II_FORCE_BUNDLED_BOOST=ON \
               -D DEAL_II_CXX_FLAGS="-std=c++17" \
               -D DEAL_II_EARLY_DEPRECATIONS=ON \
-              -D DEAL_II_WITH_CUDA="ON" \
               -D DEAL_II_WITH_KOKKOS="ON" \
               -D KOKKOS_DIR=${GITHUB_WORKSPACE}/../kokkos-install \
               -D DEAL_II_WITH_MPI="ON" \
@@ -484,8 +480,6 @@ jobs:
     - name: build CUDA tests
       run: |
         cd build
-        make -j2 setup_tests_cuda setup_tests_matrix_free_kokkos
-        cd tests/cuda
-        make -j2 compile_test_executables
-        cd ../matrix_free_kokkos
+        make -j2 setup_tests_matrix_free_kokkos
+        cd tests/matrix_free_kokkos
         make -j2 compile_test_executables
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -94,7 +94,6 @@ verbose_include(${CMAKE_SOURCE_DIR}/cmake/setup_cached_variables.cmake)
 # Now, set the project and set up the rest:
 #
 project(deal.II CXX C)
-enable_language_optional(Fortran)
 
 verbose_include(${CMAKE_SOURCE_DIR}/cmake/setup_deal_ii.cmake)
 

diff --git a/cmake/checks/check_01_cpu_features.cmake b/cmake/checks/check_01_cpu_features.cmake
@@ -327,16 +327,6 @@ if(DEAL_II_HAVE_ARM_NEON)
   set(DEAL_II_VECTORIZATION_WIDTH_IN_BITS 128)
 endif()
 
-#
-# We need to disable SIMD vectorization for CUDA device code.
-# Otherwise, nvcc compilers from version 9 on will emit an error message like:
-# "[...] contains a vector, which is not supported in device code"
-#
-
-if(DEAL_II_WITH_CUDA)
-  set(DEAL_II_VECTORIZATION_WIDTH_IN_BITS 0)
-endif()
-
 #
 # If we have OpenMP SIMD support (i.e. DEAL_II_HAVE_OPENMP_SIMD is true)
 # populate DEAL_II_OPENMP_SIMD_PRAGMA.

diff --git a/cmake/configure/configure_10_mpi.cmake b/cmake/configure/configure_10_mpi.cmake
@@ -54,12 +54,7 @@ macro(feature_mpi_configure_external)
   # TODO: We might consider refactoring this option into an automatic check
   # (in Modules/FindMPI.cmake) at some point. For the time being this is an
   # advanced configuration option.
-  #
-  if(DEAL_II_MPI_WITH_CUDA_SUPPORT)
-    option(DEAL_II_MPI_WITH_DEVICE_SUPPORT "Enable MPI Device support" ON)
-  else()
-    option(DEAL_II_MPI_WITH_DEVICE_SUPPORT "Enable MPI Device support" OFF)
-  endif()
+  option(DEAL_II_MPI_WITH_DEVICE_SUPPORT "Enable MPI Device support" OFF)
   mark_as_advanced(DEAL_II_MPI_WITH_DEVICE_SUPPORT)
 endmacro()
 

diff --git a/cmake/configure/configure_40_cuda.cmake b/cmake/configure/configure_40_cuda.cmake
diff --git a/contrib/docker/Dockerfile b/contrib/docker/Dockerfile
@@ -28,7 +28,6 @@ RUN cd /usr/src && \
     -DDEAL_II_WITH_BOOST=ON \
     -DDEAL_II_WITH_CGAL=ON \
     -DDEAL_II_WITH_COMPLEX_VALUES=ON \
-    -DDEAL_II_WITH_CUDA=OFF \
     -DDEAL_II_WITH_GINKGO=OFF \
     -DDEAL_II_WITH_GMSH=ON \
     -DDEAL_II_WITH_GSL=ON \

diff --git a/doc/doxygen/headers/cuda.h b/doc/doxygen/headers/cuda.h
diff --git a/doc/doxygen/headers/glossary.h b/doc/doxygen/headers/glossary.h
@@ -777,7 +777,7 @@
  * <dt class="glossary">@anchor GlossDevice <b>Device</b></dt>
  *
  * <dd> We commonly refer to GPUs as "devices" in deal.II. The context is
- * always related to Kokkos or CUDA that motivated using this term.
+ * always related to Kokkos that motivated using this term.
  * Occasionally, we also call data corresponding to MemorySpace::Default "device data"
  * (even though it is allocated in CPU memory if Kokkos was configured without
  * a GPU backend) to distinguish between MemorySpace::Default and MemorySpace::Host.

diff --git a/doc/doxygen/headers/main.h b/doc/doxygen/headers/main.h
@@ -74,7 +74,6 @@
     rank="same";
     petsc       [label="PETSc",    URL="\ref PETScWrappers"];
     trilinos    [label="Trilinos", URL="\ref TrilinosWrappers"];
-    cuda        [label="CUDA",     URL="\ref CUDAWrappers"];
   }
   umfpack     [label="UMFPACK"];
 

diff --git a/doc/doxygen/headers/matrixfree.h b/doc/doxygen/headers/matrixfree.h
@@ -42,7 +42,7 @@ digraph G
             color="gray", fontcolor="gray", fillcolor="white", style="filled"];
   fevalues [label="FEEvaluation", fillcolor="deepskyblue"];
   mf       [label="MatrixFree loops", fillcolor="deepskyblue"];
-  cuda     [label="CUDA",     URL="\ref CUDAWrappers", fontname="FreeSans",fontsize=12,
+  portable     [label="Portable",     URL="\ref Portable", fontname="FreeSans",fontsize=12,
             height=0.2,width=0.4,
             color="gray", fontcolor="gray", fillcolor="white", style="filled"];
   tbb      [label="TBB", fontname="FreeSans",fontsize=12,
@@ -51,8 +51,8 @@ digraph G
 {rank=same
   simd -> fevalues        [dir="none", color="transparent"];
   fevalues -> mf          [dir="none", color="transparent"];
-  mf -> cuda              [dir="none", color="transparent"];
-  cuda -> tbb             [dir="none", color="transparent"];
+  mf -> portable              [dir="none", color="transparent"];
+  portable -> tbb             [dir="none", color="transparent"];
 }
   subgraph sol {
     rank="same";

diff --git a/doc/doxygen/options.dox.in b/doc/doxygen/options.dox.in
@@ -204,7 +204,6 @@ PREDEFINED             = DOXYGEN=1 \
                          DEAL_II_CGAL_VERSION_GTE=1 \
                          DEAL_II_WITH_TASKFLOW=1 \
                          DEAL_II_WITH_COMPLEX_VALUES=1 \
-                         DEAL_II_WITH_CUDA=1 \
                          DEAL_II_WITH_GINKGO=1 \
                          DEAL_II_WITH_GMSH=1 \
                          DEAL_II_GMSH_WITH_API=1 \

diff --git a/doc/external-libs/cuda.html b/doc/external-libs/cuda.html
@@ -15,20 +15,6 @@ <h1>Installing deal.II with CUDA</h1>
     <p>
       To compile and run CUDA code, you need to use an external Kokkos
       installation that has been configured with CUDA support.
-
-      To enable deal.II's own CUDA backend, you will need your GPU to have compute
-      capability 6.0 or higher. Independently from the GPU itself, you also need a
-      version of CUDA between 10.2 and 11.8.
-    </p>
-
-    <p>
-      To configure deal.II's own CUDA backend use the following option:
-      <pre>
-
-        -DDEAL_II_WITH_CUDA=ON
-      </pre>
-      CUDA versions prior to 11.0 don't support C++17 or higher. You will have
-      to make sure that C++17 is disabled when using earlier versions.
     </p>
 
     <p>

diff --git a/doc/news/changes/incompatibilities/20240811Arndt b/doc/news/changes/incompatibilities/20240811Arndt
@@ -0,0 +1,4 @@
+Removed: All support for CUDAWrappers and CUDA-related macros have been removed.
+GPU support is provided through Kokkos.
+<br>
+(Daniel Arndt, 2024/08/11)
diff --git a/doc/users/cmake_dealii.html b/doc/users/cmake_dealii.html
@@ -457,7 +457,6 @@ <h3>Feature configuration</h3>
 DEAL_II_WITH_BOOST
 DEAL_II_WITH_CGAL
 DEAL_II_WITH_COMPLEX_VALUES
-DEAL_II_WITH_CUDA
 DEAL_II_WITH_GINKGO
 DEAL_II_WITH_GMSH
 DEAL_II_WITH_GSL

diff --git a/doc/users/cmake_user.html b/doc/users/cmake_user.html
@@ -844,7 +844,6 @@ <h2>  <code>deal.IIConfig.cmake</code> </h2>
 DEAL_II_WITH_ARPACK
 DEAL_II_WITH_ASSIMP
 DEAL_II_WITH_COMPLEX_VALUES
-DEAL_II_WITH_CUDA
 DEAL_II_WITH_GINKGO
 DEAL_II_WITH_GMSH
 DEAL_II_WITH_GSL

diff --git a/doc/users/config.sample b/doc/users/config.sample
@@ -118,7 +118,6 @@
 # set(DEAL_II_WITH_BOOST "ON" CACHE BOOL "")
 # set(DEAL_II_WITH_CGAL "ON" CACHE BOOL "")
 # set(DEAL_II_WITH_COMPLEX_VALUES "ON" CACHE BOOL "")
-# set(DEAL_II_WITH_CUDA "ON" CACHE BOOL "")
 # set(DEAL_II_WITH_GINKGO "ON" CACHE BOOL "")
 # set(DEAL_II_WITH_GMSH "ON" CACHE BOOL "")
 # set(DEAL_II_WITH_GSL "ON" CACHE BOOL "")

diff --git a/examples/step-64/doc/intro.dox b/examples/step-64/doc/intro.dox
@@ -67,8 +67,8 @@ The data movement in deal.II is done using LinearAlgebra::ReadWriteVector. These
 vectors can be seen as buffers on the host that are used to either store data
 received from the device or to send data to the device. There are two types of vectors
 that can be used on the device:
-- LinearAlgebra::CUDAWrappers::Vector, which is similar to the more common
-Vector<Number>, and
+- LinearAlgebra::TpetraWrappers::Vector<Number, MemorySpace>, that has a host
+and device copy of the data, and
 - LinearAlgebra::distributed::Vector<Number,
 MemorySpace::Default>, which is a regular
 LinearAlgebra::distributed::Vector where we have specified which memory

diff --git a/examples/step-64/step-64.cc b/examples/step-64/step-64.cc
@@ -363,8 +363,7 @@ namespace Step64
     // Since all the operations in the `solve()` function are executed on the
     // graphics card, it is necessary for the vectors used to store their values
     // on the GPU as well. LinearAlgebra::distributed::Vector can be told which
-    // memory space to use. There is also LinearAlgebra::CUDAWrappers::Vector
-    // that always uses GPU memory storage but doesn't work with MPI. It might
+    // memory space to use. It might
     // be worth noticing that the communication between different MPI processes
     // can be improved if the MPI implementation is GPU-aware and the configure
     // flag `DEAL_II_MPI_WITH_DEVICE_SUPPORT` is enabled. (The value of this

diff --git a/include/deal.II/base/config.h.in b/include/deal.II/base/config.h.in
@@ -41,7 +41,6 @@
 #cmakedefine DEAL_II_FEATURE_BOOST_BUNDLED_CONFIGURED
 #cmakedefine DEAL_II_WITH_CGAL
 #cmakedefine DEAL_II_WITH_COMPLEX_VALUES
-#cmakedefine DEAL_II_WITH_CUDA
 #cmakedefine DEAL_II_WITH_GINKGO
 #cmakedefine DEAL_II_WITH_GMSH
 #cmakedefine DEAL_II_WITH_GSL
@@ -491,9 +490,6 @@
 #endif
 
 #cmakedefine DEAL_II_MPI_WITH_DEVICE_SUPPORT
-#ifdef DEAL_II_MPI_WITH_DEVICE_SUPPORT
-#cmakedefine DEAL_II_MPI_WITH_CUDA_SUPPORT
-#endif
 
 /***********************************************************************
  * Two macro names that we put at the top and bottom of all deal.II files