Merge pull request #1530 from LLNL/v2023.06.1-RC

V2023.06.1 rc

.gitlab-ci.yml

@@ -59,7 +59,7 @@ stages:
     include:
       - local: '.gitlab/custom-jobs-and-variables.yml'
       - project: 'radiuss/radiuss-shared-ci'
-        ref: v2023.03.1
+        ref: v2023.06.0
         file: '${CI_MACHINE}-build-and-test.yml'
       - local: '.gitlab/${CI_MACHINE}-build-and-test-extra.yml'
     strategy: depend

.gitlab/corona-build-and-test-extra.yml

@@ -21,8 +21,8 @@
 # ${PROJECT_<MACHINE>_DEPS} in the extra jobs. There is no reason not to fully
 # describe the spec here.
 
-rocmcc_5_4_1_hip_desul_atomics:
+rocmcc_5_5_0_hip_desul_atomics:
   variables:
-    SPEC: " ~shared +rocm ~openmp +tests +desul amdgpu_target=gfx906 %rocmcc@5.4.1 ^hip@5.4.1 ^blt@develop"
+    SPEC: " ~shared +rocm ~openmp +tests +desul amdgpu_target=gfx906 %rocmcc@5.5.0 ^hip@5.5.0 ^blt@develop"
   extends: .build_and_test_on_corona
 

.gitlab/lassen-build-and-test-extra.yml

@@ -36,6 +36,11 @@ xl_2022_08_19_gcc_8_3_1_cuda_11_7_0:
 # ${PROJECT_<MACHINE>_DEPS} in the extra jobs. There is no reason not to fully
 # describe the spec here.
 
+gcc_8_3_1_omptask:
+  variables:
+    SPEC: " ~shared +openmp +omptask +tests %gcc@8.3.1"
+  extends: .build_and_test_on_lassen
+
 gcc_8_3_1_cuda_11_5_0_ats_disabled:
   extends: .build_and_test_on_lassen
   variables:

.gitlab/ruby-build-and-test-extra.yml

@@ -12,7 +12,22 @@
 # We keep ${PROJECT_<MACHINE>_VARIANTS} and ${PROJECT_<MACHINE>_DEPS} So that
 # the comparison with the original job is easier.
 
-# No overridden jobs so far.
+clang_14_0_6:
+  variables:
+    SPEC: " ~shared +openmp +omptask +tests %clang@14.0.6"
+  extends: .build_and_test_on_ruby
+
+gcc_10_3_1:
+  variables:
+    SPEC: " ~shared +openmp +omptask +tests %gcc@10.3.1"
+    RUBY_BUILD_AND_TEST_JOB_ALLOC: "--time=60 --nodes=1"
+  extends: .build_and_test_on_ruby
+
+intel_19_1_2_gcc_8_5_0:
+  variables:
+    SPEC: " ~shared +openmp +omptask +tests %intel@19.1.2.gcc.8.5.0"
+    RUBY_BUILD_AND_TEST_JOB_ALLOC: "--time=90 --nodes=1"
+  extends: .build_and_test_on_ruby
 
 ############
 # Extra jobs

.gitlab/tioga-build-and-test-extra.yml

@@ -28,5 +28,5 @@ rocmcc_5_4_3_hip_desul_atomics:
 
 rocmcc_5_4_3_hip_openmp:
   variables:
-    SPEC: "~shared +rocm +openmp +tests amdgpu_target=gfx90a %rocmcc@5.4.3 ^hip@5.4.3 ^blt@develop"
+    SPEC: "~shared +rocm +openmp +omptask +tests amdgpu_target=gfx90a %rocmcc@5.4.3 ^hip@5.4.3 ^blt@develop"
   extends: .build_and_test_on_tioga

CMakeLists.txt

@@ -14,9 +14,9 @@ endif()
 include(CMakeDependentOption)
 
 # Set version number
-set(RAJA_VERSION_MAJOR 2022)
-set(RAJA_VERSION_MINOR 10)
-set(RAJA_VERSION_PATCHLEVEL 5)
+set(RAJA_VERSION_MAJOR 2023)
+set(RAJA_VERSION_MINOR 06)
+set(RAJA_VERSION_PATCHLEVEL 1)
 
 if (RAJA_LOADED AND (NOT RAJA_LOADED STREQUAL "${RAJA_VERSION_MAJOR}.${RAJA_VERSION_MINOR}.${RAJA_VERSION_PATCHLEVEL}"))
   message(FATAL_ERROR "You are mixing RAJA versions. Loaded is ${RAJA_LOADED}, expected ${RAJA_VERSION_MAJOR}.${RAJA_VERSION_MINOR}.${RAJA_VERSION_PATCHLEVEL}")

Dockerfile

@@ -11,31 +11,35 @@ COPY . /home/raja/workspace
 WORKDIR /home/raja/workspace/build
 RUN cmake -DCMAKE_CXX_COMPILER=g++ -DRAJA_ENABLE_WARNINGS=On -DRAJA_ENABLE_TBB=On -DRAJA_DEPRECATED_TESTS=On -DENABLE_OPENMP=On .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/gcc-ubuntu-20.04:gcc-8.1.0 AS gcc8.1.0
 ENV GTEST_COLOR=1
 COPY . /home/raja/workspace
 WORKDIR /home/raja/workspace/build
 RUN cmake -DCMAKE_CXX_COMPILER=g++ -DRAJA_ENABLE_WARNINGS=On -DRAJA_ENABLE_WARNINGS_AS_ERRORS=On -DENABLE_COVERAGE=On -DRAJA_ENABLE_TBB=On -DENABLE_OPENMP=On .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/gcc-ubuntu-20.04:gcc-9.4.0 AS gcc9.4.0
 ENV GTEST_COLOR=1
 COPY . /home/raja/workspace
 WORKDIR /home/raja/workspace/build
 RUN cmake -DCMAKE_CXX_COMPILER=g++ -DRAJA_ENABLE_WARNINGS=On -DRAJA_ENABLE_TBB=On -DRAJA_ENABLE_RUNTIME_PLUGINS=On -DENABLE_OPENMP=On .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/gcc-ubuntu-20.04:gcc-11.2.0 AS gcc11.2.0
 ENV GTEST_COLOR=1
 COPY . /home/raja/workspace
 WORKDIR /home/raja/workspace/build
 RUN cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_CXX_COMPILER=g++ -DRAJA_ENABLE_WARNINGS=On -DRAJA_ENABLE_TBB=On -DRAJA_ENABLE_BOUNDS_CHECK=ON -DENABLE_OPENMP=On .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/clang-ubuntu-20.04:llvm-11.0.0 AS clang11.0.0
 ENV GTEST_COLOR=1
@@ -44,7 +48,8 @@ WORKDIR /home/raja/workspace/build
 RUN . /opt/spack/share/spack/setup-env.sh && export LD_LIBRARY_PATH=/opt/view/lib:$LD_LIBRARY_PATH && \
     cmake -DCMAKE_CXX_COMPILER=clang++ -DRAJA_ENABLE_TBB=On -DENABLE_OPENMP=On .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/clang-ubuntu-20.04:llvm-11.0.0 AS clang11.0.0-debug
 ENV GTEST_COLOR=1
@@ -53,7 +58,8 @@ WORKDIR /home/raja/workspace/build
 RUN . /opt/spack/share/spack/setup-env.sh && export LD_LIBRARY_PATH=/opt/view/lib:$LD_LIBRARY_PATH && \
     cmake -DCMAKE_CXX_COMPILER=clang++ -DENABLE_OPENMP=On -DCMAKE_BUILD_TYPE=Debug .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/clang-ubuntu-22.04:llvm-13.0.0 AS clang13.0.0
 ENV GTEST_COLOR=1
@@ -62,40 +68,45 @@ WORKDIR /home/raja/workspace/build
 RUN . /opt/spack/share/spack/setup-env.sh && export LD_LIBRARY_PATH=/opt/view/lib:$LD_LIBRARY_PATH && \
     cmake -DCMAKE_CXX_COMPILER=clang++ -DENABLE_OPENMP=On -DCMAKE_BUILD_TYPE=Release .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure
+    ctest -T test --output-on-failure && \
+    cd .. && rm -rf build
 
-FROM ghcr.io/rse-ops/cuda:cuda-10.1.243-ubuntu-18.04 AS nvcc10.1.243
-ENV GTEST_COLOR=1
-COPY . /home/raja/workspace
-WORKDIR /home/raja/workspace/build
-RUN . /opt/spack/share/spack/setup-env.sh && spack load cuda && \
-    cmake -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=On -DCMAKE_CUDA_STANDARD=14 -DCMAKE_CUDA_ARCHITECTURES=70 -DENABLE_OPENMP=On .. && \
-    make -j 4
+##FROM ghcr.io/rse-ops/cuda:cuda-10.1.243-ubuntu-18.04 AS nvcc10.1.243
+##ENV GTEST_COLOR=1
+##COPY . /home/raja/workspace
+##WORKDIR /home/raja/workspace/build
+##RUN . /opt/spack/share/spack/setup-env.sh && spack load cuda && \
+##    cmake -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=On -DCMAKE_CUDA_STANDARD=14 -DCMAKE_CUDA_ARCHITECTURES=70 -DENABLE_OPENMP=On .. && \
+##    make -j 4 && \
+##    cd .. && rm -rf build
 
-FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.1.1 AS nvcc11.1.1
-ENV GTEST_COLOR=1
-COPY . /home/raja/workspace
-WORKDIR /home/raja/workspace/build
-RUN . /opt/spack/share/spack/setup-env.sh && spack load cuda && \
-    cmake -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=On -DCMAKE_CUDA_STANDARD=14 -DCMAKE_CUDA_ARCHITECTURES=70 -DENABLE_OPENMP=On .. && \
-    make -j 4
+##FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.1.1 AS nvcc11.1.1
+##ENV GTEST_COLOR=1
+##COPY . /home/raja/workspace
+##WORKDIR /home/raja/workspace/build
+##RUN . /opt/spack/share/spack/setup-env.sh && spack load cuda && \
+##    cmake -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=On -DCMAKE_CUDA_STANDARD=14 -DCMAKE_CUDA_ARCHITECTURES=70 -DENABLE_OPENMP=On .. && \
+##    make -j 4 && \
+##    cd .. && rm -rf build
 
-FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.1.1 AS nvcc11.1.-debug
-ENV GTEST_COLOR=1
-COPY . /home/raja/workspace
-WORKDIR /home/raja/workspace/build
-RUN . /opt/spack/share/spack/setup-env.sh && spack load cuda && \
-    cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=On -DCMAKE_CUDA_STANDARD=14 -DCMAKE_CUDA_ARCHITECTURES=70 -DENABLE_OPENMP=On .. && \
-    make -j 4
+##FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.1.1 AS nvcc11.1.-debug
+##ENV GTEST_COLOR=1
+##COPY . /home/raja/workspace
+##WORKDIR /home/raja/workspace/build
+##RUN . /opt/spack/share/spack/setup-env.sh && spack load cuda && \
+##    cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_COMPILER=g++ -DENABLE_CUDA=On -DCMAKE_CUDA_STANDARD=14 -DCMAKE_CUDA_ARCHITECTURES=70 -DENABLE_OPENMP=On .. && \
+##    make -j 4 && \
+##    cd .. && rm -rf build
 
-FROM ghcr.io/rse-ops/hip-ubuntu-20.04:hip-5.1.3 AS hip5.1.3
-ENV GTEST_COLOR=1
-ENV HCC_AMDGPU_TARGET=gfx900
-COPY . /home/raja/workspace
-WORKDIR /home/raja/workspace/build
-RUN . /opt/spack/share/spack/setup-env.sh && spack load hip llvm-amdgpu && \
-    cmake -DCMAKE_CXX_COMPILER=clang++ -DHIP_PATH=/opt -DENABLE_HIP=On -DENABLE_CUDA=Off -DRAJA_ENABLE_WARNINGS_AS_ERRORS=Off .. && \
-    make -j 6
+##FROM ghcr.io/rse-ops/hip-ubuntu-20.04:hip-5.1.3 AS hip5.1.3
+##ENV GTEST_COLOR=1
+##ENV HCC_AMDGPU_TARGET=gfx900
+##COPY . /home/raja/workspace
+##WORKDIR /home/raja/workspace/build
+##RUN . /opt/spack/share/spack/setup-env.sh && spack load hip llvm-amdgpu && \
+##    cmake -DCMAKE_CXX_COMPILER=clang++ -DHIP_PATH=/opt -DENABLE_HIP=On -DENABLE_CUDA=Off -DRAJA_ENABLE_WARNINGS_AS_ERRORS=Off .. && \
+##    make -j 6 && \
+##    cd .. && rm -rf build
 
 FROM ghcr.io/rse-ops/intel-ubuntu-22.04:intel-2022.1.0 AS sycl
 ENV GTEST_COLOR=1
@@ -104,4 +115,5 @@ WORKDIR /home/raja/workspace/build
 RUN /bin/bash -c "source /opt/view/setvars.sh && \
     cmake -DCMAKE_CXX_COMPILER=dpcpp -DRAJA_ENABLE_SYCL=On -DENABLE_OPENMP=Off -DENABLE_ALL_WARNINGS=Off -DBLT_CXX_STD=c++17 .. && \
     make -j 6 &&\
-    ctest -T test --output-on-failure"
+    ctest -T test --output-on-failure" && \
+    cd .. && rm -rf build

RELEASE_NOTES.md

@@ -6,7 +6,7 @@
 [comment]: # (# SPDX-License-Identifier: BSD-3-Clause)
 [comment]: # (#################################################################)
 
-Version vxx.yy.zz -- Release date 20yy-mm-dd
+Version YYYY.MM.PP -- Release date 20yy-mm-dd
 ============================================
 
 This release contains ...
@@ -20,6 +20,35 @@ Notable changes include:
   * Bug fixes/improvements:
 
 
+Version 2023.06.1 -- Release date 2023-08-16
+============================================
+
+This release contains various smaller RAJA improvements. 
+
+Notable changes include:
+
+  * New features / API changes:
+     * Add compile time block size optimization for new reduction interface.
+     * Changed default stream usage for Workgroup constructs to use the 
+       stream associated with the default (camp) resource. Previously, we were 
+       using stream zero. Specifically, this change affects where we memset 
+       memory in the zeroed device memory pool and where we get device function
+       pointers for WorkGroup.  
+
+  * Build changes/improvements:
+     * RAJA_ENABLE_OPENMP_TASK CMake option added to enable/disable algorithm 
+       options based on OpenMP task construct. Currently, this only applies
+       to RAJA's OpenMP sort implementation. The default is 'Off'. The option
+       allows users to choose a task implementation if they wish.
+     * Resolve several compiler warnings.
+
+  * Bug fixes/improvements:
+     * Fix compilation of GPU occupancy calculator and use common types for 
+       HIP and CUDA backends in the occupancy calculator, kernel policies, 
+       and kernel launch helper routines.
+     * Fix direct cudaMalloc/hipMalloc calls and memory leaks.
+
+
 Version 2023.06.0 -- Release date 2023-07-06
 ============================================
 

azure-pipelines.yml

@@ -48,14 +48,14 @@ jobs:
         docker_target: clang11.0.0-debug
       clang13.0.0:
         docker_target: clang13.0.0
-      nvcc10.1.243:
-        docker_target: nvcc10.1.243
-      nvcc11.1.1:
-        docker_target: nvcc11.1.1
+##      nvcc10.1.243:
+##        docker_target: nvcc10.1.243
+##      nvcc11.1.1:
+##        docker_target: nvcc11.1.1
 ##      nvcc11.1.1-debug:
 ##        docker_target: nvcc11.1.1-debug
-      hip5.1.3:
-        docker_target: hip5.1.3
+##      hip5.1.3:
+##        docker_target: hip5.1.3
       sycl:
         docker_target: sycl
   pool:

cmake/SetupRajaOptions.cmake

@@ -17,6 +17,8 @@ option(RAJA_ENABLE_SYCL "Build SYCL support" Off)
 
 option(RAJA_ENABLE_VECTORIZATION "Build experimental vectorization support" On)
 
+option(RAJA_ENABLE_OPENMP_TASK "Build OpenMP task variants of certain algorithms" Off)
+
 option(RAJA_ENABLE_REPRODUCERS "Build issue reproducers" Off)
 
 option(RAJA_ENABLE_EXERCISES "Build exercises " On)

docs/sphinx/user_guide/feature/policies.rst

@@ -279,12 +279,37 @@ policies have the prefix ``hip_``.
  CUDA/HIP Execution Policies               Works with    Brief description
  ========================================= ============= =======================================
  cuda/hip_exec<BLOCK_SIZE>                 forall,       Execute loop iterations
-                                           scan,         in a GPU kernel launched
-                                           sort          with given thread-block
-                                                         size. Note that the
+                                           scan,         directly mapped to global threads
+                                           sort          in a GPU kernel launched
+                                                         with given thread-block
+                                                         size and unbounded grid size.
+                                                         Note that the thread-block
+                                                         size must be provided,
+                                                         there is no default.
+ cuda/hip_exec_grid<BLOCK_SIZE, GRID_SIZE> forall,       Execute loop iterations
+                                                         mapped to global threads via
+                                                         grid striding with multiple
+                                                         iterations per global thread
+                                                         in a GPU kernel launched
+                                                         with given thread-block
+                                                         size and grid size.
+                                                         Note that the thread-block
+                                                         size and grid size must be
+                                                         provided, there is no default.
+ cuda/hip_exec_occ_calc<BLOCK_SIZE>        forall        Execute loop iterations
+                                                         mapped to global threads via
+                                                         grid striding with multiple
+                                                         iterations per global thread
+                                                         in a GPU kernel launched
+                                                         with given thread-block
+                                                         size and grid size bounded
+                                                         by the maximum occupancy of
+                                                         the kernel. Note that the
                                                          thread-block size must
-                                                         be provided, there is
-                                                         no default.
+                                                         be provided, there is no
+                                                         default. Note this can improve
+                                                         reducer performance in kernels
+                                                         with large iteration counts.
  cuda/hip_launch_t                         launch        Launches a device kernel,
                                                          any code expressed within
                                                          the lambda is executed

host-configs/lc-builds/toss3/icpc_X_gcc8headers.cmake

@@ -7,7 +7,7 @@
 
 set(RAJA_COMPILER "RAJA_COMPILER_ICC" CACHE STRING "")
 
-set(COMMON_FLAGS "-gxx-name=/usr/tce/packages/gcc/gcc-8.1.0/bin/g++")
+set(COMMON_FLAGS "-gxx-name=/usr/tce/packages/gcc/gcc-8.3.1/bin/g++")
 
 set(CMAKE_CXX_FLAGS_RELEASE "${COMMON_FLAGS} -O3 -march=native -ansi-alias -diag-disable cpu-dispatch" CACHE STRING "")
 set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${COMMON_FLAGS} -O3 -g -march=native -ansi-alias -diag-disable cpu-dispatch" CACHE STRING "")

host-configs/lc-builds/toss4/icpc-classic_X.cmake

@@ -0,0 +1,18 @@
+###############################################################################
+# Copyright (c) 2016-23, Lawrence Livermore National Security, LLC
+# and RAJA project contributors. See the RAJA/LICENSE file for details.
+#
+# SPDX-License-Identifier: (BSD-3-Clause)
+###############################################################################
+
+set(RAJA_COMPILER "RAJA_COMPILER_ICC" CACHE STRING "")
+
+set(COMMON_FLAGS "-gxx-name=/usr/tce/packages/gcc/gcc-10.3.1/bin/g++")
+
+set(CMAKE_CXX_FLAGS_RELEASE "${COMMON_FLAGS} -O3 -march=native -ansi-alias -diag-disable cpu-dispatch" CACHE STRING "")
+set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${COMMON_FLAGS} -O3 -g -march=native -ansi-alias -diag-disable cpu-dispatch" CACHE STRING "")
+set(CMAKE_CXX_FLAGS_DEBUG "${COMMON_FLAGS} -O0 -g" CACHE STRING "")
+
+set(RAJA_DATA_ALIGN 64 CACHE STRING "")
+
+set(RAJA_HOST_CONFIG_LOADED On CACHE BOOL "")

include/RAJA/config.hpp.in

@@ -176,6 +176,8 @@ static_assert(RAJA_HAS_SOME_CXX14,
 #cmakedefine RAJA_ENABLE_CLANG_CUDA
 #cmakedefine RAJA_ENABLE_HIP
 #cmakedefine RAJA_ENABLE_SYCL
+
+#cmakedefine RAJA_ENABLE_OMP_TASK
 #cmakedefine RAJA_ENABLE_VECTORIZATION
 
 #cmakedefine RAJA_ENABLE_NV_TOOLS_EXT
@@ -254,12 +256,14 @@ namespace RAJA {
 #if defined(RAJA_ENABLE_OPENMP) && !defined(__HIP_DEVICE_COMPILE__)
 #if defined(_OPENMP)
 #if (_OPENMP >= 200805)
-#define RAJA_ENABLE_OPENMP_TASK
+#if defined(RAJA_ENABLE_OPENMP_TASK)
+#define RAJA_ENABLE_OPENMP_TASK_INTERNAL
 #endif
+#endif // _OPENMP >= 200805
 #else
 #error RAJA configured with RAJA_ENABLE_OPENMP, but _OPENMP is not defined in this code section
-#endif // _OPENMP
-#endif // RAJA_ENABLE_OPENMP && __HIP_DEVICE_COMPILE__
+#endif // else
+#endif // RAJA_ENABLE_OPENMP && !__HIP_DEVICE_COMPILE__
 
 #if defined(RAJA_ENABLE_CUDA) && defined(__CUDACC__)
 #define RAJA_CUDA_ACTIVE