Merge pull request #1507 from ndellingwood/release-candidate-3.7.00

Release candidate 3.7.00
kokkos · Sep 22, 2022 · 04821ac · 04821ac
2 parents 42ab7a2 + f32debb
commit 04821ac
Show file tree

Hide file tree

Showing 318 changed files with 25,341 additions and 14,885 deletions.
diff --git a/.github/workflows/osx.yml b/.github/workflows/osx.yml
@@ -2,12 +2,19 @@ name: github-OSX
 
 on:
   pull_request:
-    branches:
-      - master
-      - develop
+    types: [ opened, labeled, unlabeled, reopened, synchronize ]
 
 jobs:
+  check-pr-labels:
+    runs-on: [ubuntu-latest]
+    steps:
+      - uses: docker://agilepathway/pull-request-label-checker:latest
+        with:
+          none_of: 'AT: WIP'
+          repo_token: ${{ secrets.GITHUB_TOKEN }}
   osxci:
+    needs: check-pr-labels
+    # TODO: allow re-run via retest label if: ${{ github.event.label.name == 'AT: RETEST' }}
     name: osx-ci
     runs-on: [macos-latest]
 
@@ -16,12 +23,16 @@ jobs:
         include:
           - backend: "SERIAL"
             cmake_build_type: "RelWithDebInfo"
+            debug_bounds_check: "ON"
           - backend: "THREADS"
             cmake_build_type: "RelWithDebInfo"
+            debug_bounds_check: "ON"
           - backend: "SERIAL"
             cmake_build_type: "Debug"
+            debug_bounds_check: "OFF"
           - backend: "SERIAL"
             cmake_build_type: "Release"
+            debug_bounds_check: "ON"
 
     steps:
       - name: checkout_kokkos_kernels
@@ -46,6 +57,7 @@ jobs:
           -DCMAKE_CXX_FLAGS="-Werror" \
           -DCMAKE_CXX_STANDARD=14 \
           -DKokkos_ENABLE_COMPILER_WARNINGS=ON \
+          -DKokkos_ENABLE_DEBUG_BOUNDS_CHECK:BOOL=${{ matrix.debug_bounds_check }} \
           -DKokkos_ENABLE_DEPRECATED_CODE_3=OFF \
           -DCMAKE_BUILD_TYPE=${{ matrix.cmake_build_type }} \
           -DCMAKE_INSTALL_PREFIX=$PWD/../install \
@@ -73,6 +85,8 @@ jobs:
           -DKokkosKernels_INST_FLOAT=ON \
           -DKokkosKernels_INST_LAYOUTLEFT:BOOL=ON \
           -DKokkosKernels_INST_LAYOUTRIGHT:BOOL=ON \
+          -DKokkosKernels_INST_OFFSET_INT=ON \
+          -DKokkosKernels_INST_OFFSET_SIZE_T=ON \
           -DKokkosKernels_ENABLE_TPL_CUSPARSE=OFF \
           -DKokkosKernels_ENABLE_TPL_CUBLAS=OFF \
           ..
@@ -83,4 +97,4 @@ jobs:
 
       - name: test
         working-directory: kokkos-kernels/build
-        run: ctest -j2 --output-on-failure
+        run: ctest -j2 --output-on-failure --timeout 3600
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,119 @@
 # Change Log
 
+## [3.7.00](https://github.com/kokkos/kokkos-kernels/tree/3.7.00) (2022-08-18)
+[Full Changelog](https://github.com/kokkos/kokkos-kernels/compare/3.6.01...3.7.00)
+
+### Features:
+
+#### Final Bsr algorithms implemented for multigrid:
+- Sparse: bsr transpose algorithm [\#1477](https://github.com/kokkos/kokkos-kernels/pull/1477)
+- BSR block SpGEMM implementation [\#1099](https://github.com/kokkos/kokkos-kernels/pull/1099)
+
+#### Adding batched dense linear and non-linear system solvers:
+- Add batched GESV [\#1384](https://github.com/kokkos/kokkos-kernels/pull/1384)
+- Newton solver: serial on device implementation of Newton's method [\#1479](https://github.com/kokkos/kokkos-kernels/pull/1479)
+
+#### Add sparse matrix conversion:
+- Add csc2csr [\#1342](https://github.com/kokkos/kokkos-kernels/pull/1342)
+- csc2csr: update Kokkos_Numeric.hpp header inclusion [\#1449](https://github.com/kokkos/kokkos-kernels/pull/1449)
+- sparse: Remove csc2csr copy [\#1375](https://github.com/kokkos/kokkos-kernels/pull/1375)
+
+#### New documentation in readthedocs
+- Added https://kokkos-kernels.readthedocs.io [\#1451](https://github.com/kokkos/kokkos-kernels/pull/1451)
+- Restructure docs [\#1368](https://github.com/kokkos/kokkos-kernels/pull/1368)
+
+#### Fix issues with TPLs for mutlivector SPMV
+- Add cuSparse TPL files for CrsMatrix-multivector product [\#1427](https://github.com/kokkos/kokkos-kernels/pull/1427)
+
+### Deprecations:
+- Add template params to forwarding calls in deprecated KokkosKernels::… [\#1441](https://github.com/kokkos/kokkos-kernels/pull/1441)
+
+### Implemented enhancements:
+
+####
+- SPILUK: Move host allocations to symbolic [\#1480](https://github.com/kokkos/kokkos-kernels/pull/1480)
+- trsv: remove assumptions about entry order within rows [\#1463](https://github.com/kokkos/kokkos-kernels/pull/1463)
+
+#### Hierarchical BLAS algorithms, added and moved from batched:
+- Blas serial axpy and nrm2 [\#1460](https://github.com/kokkos/kokkos-kernels/pull/1460)
+- Move Set/Scale unit test to KokkosBlas [\#1455](https://github.com/kokkos/kokkos-kernels/pull/1455)
+- Move {Serial,Team,TeamVector} Set to KokkosBlas [\#1454](https://github.com/kokkos/kokkos-kernels/pull/1454)
+- Move {Serial,Team,TeamVector}Scale to KokkosBlas [\#1448](https://github.com/kokkos/kokkos-kernels/pull/1448)
+
+#### Code base organization and clean-ups:
+- Common Utils: removing dependency on Sparse Utils in Common Utils [\#1436](https://github.com/kokkos/kokkos-kernels/pull/1436)
+- Common cleanup [\#1431](https://github.com/kokkos/kokkos-kernels/pull/1431)
+- Clean-up src: re-organizing the src directory [\#1398](https://github.com/kokkos/kokkos-kernels/pull/1398)
+- Sparse utils namespace [\#1439](https://github.com/kokkos/kokkos-kernels/pull/1439)
+
+#### perf tests updates, fixes and clean-ups:
+- dot perf test: adding support for HIP and SYCL backend [\#1453](https://github.com/kokkos/kokkos-kernels/pull/1453)
+- Add verbosity parameter to GMRES example. Turn off for testing. [\#1385](https://github.com/kokkos/kokkos-kernels/pull/1385)
+- KokkosSparse_spiluk.cpp perf test: add int-int guards to cusparse codes [\#1369](https://github.com/kokkos/kokkos-kernels/pull/1369)
+- perf_test/blas: Check ARMPL build version [\#1352](https://github.com/kokkos/kokkos-kernels/pull/1352)
+- Clean-up batched block tridiag perf test [\#1343](https://github.com/kokkos/kokkos-kernels/pull/1343)
+- Reduce lots of macro duplication in sparse unit tests [\#1340](https://github.com/kokkos/kokkos-kernels/pull/1340)
+
+#### Infrastructure changes: ETI and testing upgrades, minor fixes
+- sycl: re-enabling test now that dpcpp has made progress [\#1473](https://github.com/kokkos/kokkos-kernels/pull/1473)
+- Only instantiate Kokkos's default Cuda mem space [\#1361](https://github.com/kokkos/kokkos-kernels/pull/1361)
+- Sparse and CI updates [\#1411](https://github.com/kokkos/kokkos-kernels/pull/1411)
+- Newer sparse tests were not following the new testing pattern [\#1356](https://github.com/kokkos/kokkos-kernels/pull/1356)
+- Add ETI for D1 coloring [\#1401](https://github.com/kokkos/kokkos-kernels/pull/1401)
+- Add ETI to SpAdd (symbolic and numeric) [\#1399](https://github.com/kokkos/kokkos-kernels/pull/1399)
+- Reformat example/fenl files changed in 1382 [\#1464](https://github.com/kokkos/kokkos-kernels/pull/1464)
+- Change Controls::getParameter error message from stdout to stderr [\#1416](https://github.com/kokkos/kokkos-kernels/pull/1416)
+
+#### Kokkos alignment: update our implementations to use newer Kokkos features
+- Arith traits integral nan [\#1438](https://github.com/kokkos/kokkos-kernels/pull/1438)
+- Kokkos_ArithTraits: re-implementation using Kokkos Core [\#1406](https://github.com/kokkos/kokkos-kernels/pull/1406)
+- Value-initialize result of MaxLoc reduction to avoid maybe uninitialized warning [\#1383](https://github.com/kokkos/kokkos-kernels/pull/1383)
+- Remove volatile qualifiers in reducer join(), init(), and operator+= methods [\#1382](https://github.com/kokkos/kokkos-kernels/pull/1382)
+
+#### BLAS and batched algorithms updates
+- Update Batched GMRES [\#1392](https://github.com/kokkos/kokkos-kernels/pull/1392)
+- GEMV: accumulate in float for scalar = bhalf_t [\#1360](https://github.com/kokkos/kokkos-kernels/pull/1360)
+- Restore BLAS-1 MV paths for 1 column [\#1354](https://github.com/kokkos/kokkos-kernels/pull/1354)
+
+#### Sparse and Graph updates
+- Minor updates to cluster Gauss-Seidel [\#1372](https://github.com/kokkos/kokkos-kernels/pull/1372)
+- Add unit test for BsrMatrix and BlockCrsMatrix spmv [\#1338](https://github.com/kokkos/kokkos-kernels/pull/1338)
+- Refactor SPGEMM MKL Impl [\#1244](https://github.com/kokkos/kokkos-kernels/pull/1244)
+- D1 coloring: remove unused but set variable [\#1403](https://github.com/kokkos/kokkos-kernels/pull/1403)
+
+#### half precision paper
+- Minor changes for half precision paper [\#1429](https://github.com/kokkos/kokkos-kernels/pull/1429)
+- Add benchmarks for us-rse escience 2022 half precision paper [\#1422](https://github.com/kokkos/kokkos-kernels/pull/1422)
+
+
+### Bug Fixes:
+- TPLs: adding CUBLAS in the list of dependencies [\#1482](https://github.com/kokkos/kokkos-kernels/pull/1482)
+- Fix MKL build errors [\#1478](https://github.com/kokkos/kokkos-kernels/pull/1478)
+- Fixup drop layout template param in rank-0 views [\#1476](https://github.com/kokkos/kokkos-kernels/pull/1476)
+- BLAS: fixing test that access results before synching [\#1472](https://github.com/kokkos/kokkos-kernels/pull/1472)
+- Fix D1 color ETI with both CudaSpace and UVM [\#1471](https://github.com/kokkos/kokkos-kernels/pull/1471)
+- Fix arithtraits warning [\#1468](https://github.com/kokkos/kokkos-kernels/pull/1468)
+- Fix build when double not instantiated [\#1467](https://github.com/kokkos/kokkos-kernels/pull/1467)
+- Fix -Werror [\#1466](https://github.com/kokkos/kokkos-kernels/pull/1466)
+- Fix GitHub CI failing on broken develop [\#1461](https://github.com/kokkos/kokkos-kernels/pull/1461)
+- HIP: fix warning from ExecSpaceUtils and GEMV [\#1459](https://github.com/kokkos/kokkos-kernels/pull/1459)
+- Removes a duplicate cuda_data_type_from when KOKKOS_HALF_T_IS_FLOAT [\#1456](https://github.com/kokkos/kokkos-kernels/pull/1456)
+- Fix incorrect function call in KokkosBatched::TeamGEMV unit test [\#1444](https://github.com/kokkos/kokkos-kernels/pull/1444)
+- Fix SYCL nightly test [\#1419](https://github.com/kokkos/kokkos-kernels/pull/1419)
+- Fix issues with cuSparse TPL availability for BsrMatrix SpMV [\#1418](https://github.com/kokkos/kokkos-kernels/pull/1418)
+- SpMV: fixing issues with unit-tests tolerance [\#1412](https://github.com/kokkos/kokkos-kernels/pull/1412)
+- Address 1409 [\#1410](https://github.com/kokkos/kokkos-kernels/pull/1410)
+- Fix colliding include guards (copy-paste mistake) [\#1408](https://github.com/kokkos/kokkos-kernels/pull/1408)
+- src/sparse: Fix & check for fence post errors [\#1405](https://github.com/kokkos/kokkos-kernels/pull/1405)
+- Bspgemm fixes [\#1396](https://github.com/kokkos/kokkos-kernels/pull/1396)
+- Fix unused parameter warnings in GEMM test. [\#1381](https://github.com/kokkos/kokkos-kernels/pull/1381)
+- Fixes code deprecation warnings. [\#1379](https://github.com/kokkos/kokkos-kernels/pull/1379)
+- Fix sign-compare warning in SPMV perf test [\#1371](https://github.com/kokkos/kokkos-kernels/pull/1371)
+- Minor MKL fixes [\#1365](https://github.com/kokkos/kokkos-kernels/pull/1365)
+- perf_test/batched: Temporarily disable tests [\#1359](https://github.com/kokkos/kokkos-kernels/pull/1359)
+- Fix nightly builds following promotion of the math functions in Kokkos [\#1339](https://github.com/kokkos/kokkos-kernels/pull/1339)
+
+
 ## [3.6.01](https://github.com/kokkos/kokkos-kernels/tree/3.6.01) (2022-05-23)
 [Full Changelog](https://github.com/kokkos/kokkos-kernels/compare/3.6.00...3.6.01)
 

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -24,8 +24,8 @@ IF(NOT KOKKOSKERNELS_HAS_TRILINOS)
     PROJECT(KokkosKernels CXX)
   ENDIF()
   SET(KokkosKernels_VERSION_MAJOR 3)
-  SET(KokkosKernels_VERSION_MINOR 6)
-  SET(KokkosKernels_VERSION_PATCH 01)
+  SET(KokkosKernels_VERSION_MINOR 7)
+  SET(KokkosKernels_VERSION_PATCH 00)
   SET(KokkosKernels_VERSION "${KokkosKernels_VERSION_MAJOR}.${KokkosKernels_VERSION_MINOR}.${KokkosKernels_VERSION_PATCH}")
   MATH(EXPR KOKKOSKERNELS_VERSION "${KokkosKernels_VERSION_MAJOR} * 10000 + ${KokkosKernels_VERSION_MINOR} * 100 + ${KokkosKernels_VERSION_PATCH}")
 ENDIF()
@@ -35,7 +35,7 @@ CMAKE_POLICY(SET CMP0074 NEW)
 
 INCLUDE(GNUInstallDirs)
 IF (KOKKOSKERNELS_HAS_TRILINOS)
- SET(TRILINOS_INCDIR ${CMAKE_INSTALL_PREFIX}/${${PROJECT_NAME}_INSTALL_INCLUDE_DIR})
+ SET(TRILINOS_INCDIR ${${PROJECT_NAME}_INSTALL_INCLUDE_DIR})
  SET(KOKKOSKERNELS_HEADER_INSTALL_DIR ${TRILINOS_INCDIR})
  SET(KOKKOS_ENABLE_CUDA_UVM ${Kokkos_ENABLE_CUDA_UVM})
 ELSEIF(KOKKOSKERNELS_HAS_PARENT)

diff --git a/README.md b/README.md
@@ -1,3 +1,5 @@
+[![Generic badge](https://readthedocs.org/projects/pip/badge/?version=latest&style=flat)](https://kokkos-kernels.readthedocs.io/en/latest/)
+
 ![KokkosKernels](https://avatars2.githubusercontent.com/u/10199860?s=200&v=4)
 
 # Kokkos Kernels

diff --git a/cm_generate_makefile.bash b/cm_generate_makefile.bash
@@ -230,7 +230,7 @@ display_help_text() {
       echo "--with-openmptarget:                          Enable OpenMPTarget backend."
       echo "--with-sycl:                                  Enable Sycl backend."
       echo "--with-openmp:                                Enable OpenMP backend."
-      echo "--with-pthread:                               Enable Pthreads backend."
+      echo "--with-threads:                               Enable Threads backend."
       echo "--with-serial:                                Enable Serial backend."
       echo "--with-devices:                               Explicitly add a set of backends."
       echo ""
@@ -274,6 +274,8 @@ display_help_text() {
       echo "                 Pascal61        = NVIDIA Pascal generation CC 6.1"
       echo "                 Volta70         = NVIDIA Volta generation CC 7.0"
       echo "                 Volta72         = NVIDIA Volta generation CC 7.2"
+      echo "                 Ampere80        = NVIDIA Ampere generation CC 8.0"
+      echo "                 Ampere86        = NVIDIA Ampere generation CC 8.6"
       echo ""
       echo "--compiler=/Path/To/Compiler  Set the compiler."
       echo ""
@@ -335,6 +337,7 @@ display_help_text() {
       echo "--kokkos-make-j=[NUM]:        Set -j parallel level for kokkos install"
       echo "                                Default: j == 4"
       echo "--enable-tests: build Kokkos Kernels unit and performance tests"
+      echo "--deprecated-code             Enable deprecated code (disabled by default)"
       echo "--enable-perfsuite: build Kokkos Kernels performance tests with
 RAJAPerf Suite"
 
@@ -360,6 +363,8 @@ KERNELS_DEFAULT_ETI_OPTION=""
 WITH_CUDA_BACKEND=OFF
 WITH_HIP_BACKEND=OFF
 
+KOKKOS_DEPRECATED_CODE=OFF
+
 while [[ $# > 0 ]]
 do
   key="$1"
@@ -415,8 +420,8 @@ do
     --with-sycl)
       update_kokkos_devices Sycl
       ;;
-    --with-pthread)
-      update_kokkos_devices Pthread
+    --with-threads)
+      update_kokkos_devices Threads
       ;;
     --with-serial)
       update_kokkos_devices Serial
@@ -522,6 +527,9 @@ do
     --disable-examples)
       KOKKOSKERNELS_DO_EXAMPLES=OFF
       ;;
+    --deprecated-code)
+      KOKKOS_DEPRECATED_CODE=ON
+      ;;
     --compiler*)
       COMPILER="${key#*=}"
       CNUM=$(command -v ${COMPILER} 2>&1 >/dev/null | grep "no ${COMPILER}" | wc -l)
@@ -738,9 +746,9 @@ cd ${KOKKOS_INSTALL_PATH}
 
 # Configure kokkos
 echo ""
-echo cmake $COMPILER_CMD  -DCMAKE_CXX_FLAGS="${KOKKOS_CXXFLAGS}" -DCMAKE_EXE_LINKER_FLAGS="${KOKKOS_LDFLAGS}" -DCMAKE_INSTALL_PREFIX=${KOKKOS_INSTALL_PATH} ${KOKKOS_DEVICE_CMD} ${KOKKOS_ARCH_CMD} -DKokkos_ENABLE_TESTS=${KOKKOS_DO_TESTS} -DKokkos_ENABLE_EXAMPLES=${KOKKOS_DO_EXAMPLES} ${KOKKOS_OPTION_CMD} ${KOKKOS_CUDA_OPTION_CMD} ${KOKKOS_HIP_OPTION_CMD} -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_CXX_EXTENSIONS=OFF ${STANDARD_CMD} ${KOKKOS_BUILDTYPE_CMD} -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBRARIES} ${KOKKOS_BC_CMD} ${KOKKOS_HWLOC_CMD} ${KOKKOS_HWLOC_PATH_CMD} ${KOKKOS_MEMKIND_CMD} ${KOKKOS_MEMKIND_PATH_CMD} -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF ${KOKKOS_PATH}
+echo cmake $COMPILER_CMD  -DCMAKE_CXX_FLAGS="${KOKKOS_CXXFLAGS}" -DCMAKE_EXE_LINKER_FLAGS="${KOKKOS_LDFLAGS}" -DCMAKE_INSTALL_PREFIX=${KOKKOS_INSTALL_PATH} ${KOKKOS_DEVICE_CMD} ${KOKKOS_ARCH_CMD} -DKokkos_ENABLE_TESTS=${KOKKOS_DO_TESTS} -DKokkos_ENABLE_EXAMPLES=${KOKKOS_DO_EXAMPLES} ${KOKKOS_OPTION_CMD} ${KOKKOS_CUDA_OPTION_CMD} ${KOKKOS_HIP_OPTION_CMD} -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_CXX_EXTENSIONS=OFF ${STANDARD_CMD} ${KOKKOS_BUILDTYPE_CMD} -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBRARIES} ${KOKKOS_BC_CMD} ${KOKKOS_HWLOC_CMD} ${KOKKOS_HWLOC_PATH_CMD} ${KOKKOS_MEMKIND_CMD} ${KOKKOS_MEMKIND_PATH_CMD} -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF -DKokkos_ENABLE_DEPRECATED_CODE_3=${KOKKOS_DEPRECATED_CODE} ${KOKKOS_PATH}
 echo ""
-cmake $COMPILER_CMD  -DCMAKE_CXX_FLAGS="${KOKKOS_CXXFLAGS//\"}" -DCMAKE_EXE_LINKER_FLAGS="${KOKKOS_LDFLAGS//\"}" -DCMAKE_INSTALL_PREFIX=${KOKKOS_INSTALL_PATH} ${KOKKOS_DEVICE_CMD} ${KOKKOS_ARCH_CMD} -DKokkos_ENABLE_TESTS=${KOKKOS_DO_TESTS} -DKokkos_ENABLE_EXAMPLES=${KOKKOS_DO_EXAMPLES} ${KOKKOS_OPTION_CMD} ${KOKKOS_CUDA_OPTION_CMD} ${KOKKOS_HIP_OPTION_CMD} -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_CXX_EXTENSIONS=OFF ${STANDARD_CMD} ${KOKKOS_BUILDTYPE_CMD} -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBRARIES}${KOKKOS_BC_CMD} ${KOKKOS_HWLOC_CMD} ${KOKKOS_HWLOC_PATH_CMD} ${KOKKOS_MEMKIND_CMD} ${KOKKOS_MEMKIND_PATH_CMD} -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF ${KOKKOS_PATH}
+cmake $COMPILER_CMD  -DCMAKE_CXX_FLAGS="${KOKKOS_CXXFLAGS//\"}" -DCMAKE_EXE_LINKER_FLAGS="${KOKKOS_LDFLAGS//\"}" -DCMAKE_INSTALL_PREFIX=${KOKKOS_INSTALL_PATH} ${KOKKOS_DEVICE_CMD} ${KOKKOS_ARCH_CMD} -DKokkos_ENABLE_TESTS=${KOKKOS_DO_TESTS} -DKokkos_ENABLE_EXAMPLES=${KOKKOS_DO_EXAMPLES} ${KOKKOS_OPTION_CMD} ${KOKKOS_CUDA_OPTION_CMD} ${KOKKOS_HIP_OPTION_CMD} -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_CXX_EXTENSIONS=OFF ${STANDARD_CMD} ${KOKKOS_BUILDTYPE_CMD} -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBRARIES}${KOKKOS_BC_CMD} ${KOKKOS_HWLOC_CMD} ${KOKKOS_HWLOC_PATH_CMD} ${KOKKOS_MEMKIND_CMD} ${KOKKOS_MEMKIND_PATH_CMD} -DKokkos_ENABLE_DEPRECATION_WARNINGS=OFF -DKokkos_ENABLE_DEPRECATED_CODE_3=${KOKKOS_DEPRECATED_CODE} ${KOKKOS_PATH}
 
 # Install kokkos library
 make install -j $KOKKOS_MAKEINSTALL_J

diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
@@ -1,7 +1,12 @@
 TRIBITS_PACKAGE_DEFINE_DEPENDENCIES(
         LIB_REQUIRED_PACKAGES KokkosCore KokkosContainers KokkosAlgorithms
-        LIB_OPTIONAL_TPLS quadmath MKL BLAS LAPACK CUSPARSE MAGMA METIS SuperLU Cholmod LAPACKE CBLAS ARMPL ROCBLAS ROCSPARSE
+        LIB_OPTIONAL_TPLS quadmath MKL BLAS LAPACK CUSPARSE MAGMA METIS SuperLU Cholmod LAPACKE CBLAS ARMPL ROCBLAS ROCSPARSE CUBLAS
         TEST_OPTIONAL_TPLS yaml-cpp
 )
 # NOTE: If you update names in LIB_OPTIONAL_TPLS above, make sure to map those names in
-# the macro 'KOKKOSKERNELS_ADD_TPL_OPTION' that resides in cmake/kokkoskernels_tpls.cmake.
+# the macro 'KOKKOSKERNELS_ADD_TPL_OPTION' that resides in cmake/kokkoskernels_tpls.cmake.
+
+if (TPL_ENABLE_CUDA)
+  tribits_tpl_tentatively_enable(CUBLAS)
+endif()
+
diff --git a/cmake/KokkosKernels_config.h.in b/cmake/KokkosKernels_config.h.in
@@ -70,6 +70,8 @@
 #cmakedefine KOKKOSKERNELS_INST_FLOAT
 /* Whether to build kernels for scalar type Kokkos::Experimental::half_t */
 #cmakedefine KOKKOSKERNELS_INST_HALF
+/* Whether to build kernels for scalar type Kokkos::Experimental::bhalf_t */
+#cmakedefine KOKKOSKERNELS_INST_BHALF
 /* Whether to build kernels for scalar type complex<double> */
 #cmakedefine KOKKOSKERNELS_INST_COMPLEX_DOUBLE
 /* Whether to build kernels for scalar type complex<float> */

diff --git a/cmake/Modules/FindTPLMKL.cmake b/cmake/Modules/FindTPLMKL.cmake
@@ -41,6 +41,10 @@ ELSE()
       LIBRARY_PATHS
         ${MKL_ROOT}/lib/intel64
         ${ENV_LIBDIRS}
+      HEADER
+        mkl.h
+      HEADER_PATHS
+        ${MKL_ROOT}/include
     )
   ENDIF()
 ENDIF()