Develop stream 2024-09-12 (#462)

* Fixed overflow bug for large sizes in thrust::shuffle

* Added definitions of execution space macros

* Add missing overloads for thrust::pow

* Refactors thrust::unique_by_key to use cub::DeviceSelect::UniqueByKey

* Fix a typo in thrust-config.cmake

* Check that thrust::pair is trivially copyable

* Remove double ignore in discard_iterator.h docs

* Replace deprecated _VSTD macro with std

* Update mode example to use thrust::unique_count

* Ensure that thrust fancy iterators are trivially_copy_constructible when possible

* Use checked allocators in CUB catch2 tests

* Refactors thrust::copy_if to use cub::DeviceSelect

* Refactor thrust::[stable_]partition[_copy] to use cub::DevicePartition

* Fix include of <thrust/random.h> with NVC++

* Cleanup diagnostic handling

* Rework config.h

* Bump version to 2.4.0

* Fix issues with ambiguous calls to addressof in thrust::optional

* Try harder to unwrap nested thrust::tuple_of_iterator_references, CUDA backend

* Added missing element from thrust's tuple implementation

* Ensure that we can run reduce_by_key with const inputs

* Leave definitions of __host__ and __device__

This prevents CCCL/thrust's build breakage because of v2.4.0 changes

* Patched up CI because of CCCL2.4.0 tests' build failure

* Updated tests and examples for __host__ __device__ use

* Updated CHANGELOG

* Added operator to transform_reduce benchmark

* Added mem allocator in benchmarks

* Changes for review

* ci: set up sccache

* Added helper functions for choosing between different custom reporter

* Added json and csv custom reporter for benchmarks

* Changes for review

* Added hipstdpar tests

* Relocated our ParallelSTL additions

* Fixed several naming issues

* Added missing unimplemented algorithms

* Split hipstdpar_lib.hpp

* Added relevant information to README and CHANGELOG regarding HIPSTDPAR

* Clarified upstream LLVM offload support

* Emit error when HIPSTDPAR macros are not defined

* Move forwarding calls to rocPRIM to thrust's stubs

* Fix path to hipstdpar impl headers

* Prevent building hipstdpar tests when no compatible libstdc++ is present

* Disable TBB tests build

---------

Co-authored-by: Beatriz Navidad Vilches <beatriz@streamhpc.com>
Co-authored-by: Robin Voetter <robin@streamhpc.com>

.clang-format

@@ -1,11 +1,11 @@
 # Style file for MLSE Libraries based on the modified rocBLAS style
 
 # Common settings
-BasedOnStyle:  WebKit
-TabWidth:        4
-IndentWidth:     4
+BasedOnStyle:  LLVM
+TabWidth:        2
+IndentWidth:     2
 UseTab:          Never
-ColumnLimit: 100
+ColumnLimit: 120
 
 # Other languages JavaScript, Proto
 
@@ -20,14 +20,14 @@ Language: Cpp
 # void formatted_code_again;
 
 DisableFormat:   false
-Standard: Cpp11
-
-AccessModifierOffset: -4
+Standard: c++14
+AccessModifierOffset: -2
 AlignAfterOpenBracket: true
 AlignConsecutiveAssignments: true
 AlignConsecutiveDeclarations: true
 AlignEscapedNewlinesLeft: true
 AlignOperands:   true
+AllowAllArgumentsOnNextLine: true
 AlignTrailingComments: false
 AllowAllParametersOfDeclarationOnNextLine: true
 AllowShortBlocksOnASingleLine: false
@@ -39,13 +39,26 @@ AlwaysBreakAfterDefinitionReturnType: false
 AlwaysBreakAfterReturnType: None
 AlwaysBreakBeforeMultilineStrings: false
 AlwaysBreakTemplateDeclarations: true
+AttributeMacros: [
+                  'THRUST_DEVICE',
+                  'THRUST_FORCEINLINE',
+                  'THRUST_HOST_DEVICE',
+                  'THRUST_HOST',
+                  '_CCCL_DEVICE',
+                  '_CCCL_FORCEINLINE',
+                  '_CCCL_HOST_DEVICE',
+                  '_CCCL_HOST',
+                  'THRUST_RUNTIME_FUNCTION',
+                  'THRUST_DETAIL_KERNEL_ATTRIBUTES',
+                 ]
 BinPackArguments: false
 BinPackParameters: false
 
 # Configure each individual brace in BraceWrapping
 BreakBeforeBraces: Custom
 # Control of individual brace wrapping cases
 BraceWrapping: {
+    AfterCaseLabel: 'false'
     AfterClass: 'true'
     AfterControlStatement: 'true'
     AfterEnum : 'true'
@@ -56,52 +69,69 @@ BraceWrapping: {
     BeforeCatch : 'true'
     BeforeElse : 'true'
     IndentBraces : 'false'
-#    AfterExternBlock : 'true'
+    SplitEmptyFunction: 'false'
+    SplitEmptyRecord: 'false'
 }
 
-#BreakAfterJavaFieldAnnotations: true
-#BreakBeforeInheritanceComma: false
-#BreakBeforeBinaryOperators: None
-#BreakBeforeTernaryOperators: true
-#BreakConstructorInitializersBeforeComma: true
-#BreakStringLiterals: true
+BreakBeforeConceptDeclarations: true
+BreakBeforeBinaryOperators: NonAssignment
+BreakBeforeTernaryOperators: true
+BreakConstructorInitializers: BeforeComma
+BreakInheritanceList: BeforeComma
+EmptyLineAfterAccessModifier: Never
+EmptyLineBeforeAccessModifier: Always
+
+InsertBraces: true
+InsertNewlineAtEOF: true
+InsertTrailingCommas: Wrapped
+IndentRequires: true
+IndentPPDirectives: AfterHash
+PackConstructorInitializers: Never
+PenaltyBreakAssignment: 30
+PenaltyBreakTemplateDeclaration: 0
+PenaltyIndentedWhitespace: 2
+RemoveSemicolon: false
+SpaceAfterLogicalNot: false
+SpaceAfterTemplateKeyword: true
+SpaceBeforeCtorInitializerColon: true
+SpaceBeforeInheritanceColon: true
+SpaceBeforeRangeBasedForLoopColon: true
+
 
 CommentPragmas:  '^ IWYU pragma:'
-#CompactNamespaces: false
+CompactNamespaces: false
 ConstructorInitializerAllOnOneLineOrOnePerLine: false
 ConstructorInitializerIndentWidth: 4
-ContinuationIndentWidth: 4
+ContinuationIndentWidth: 2
 Cpp11BracedListStyle: true
-#SpaceBeforeCpp11BracedList: false
-DerivePointerAlignment: false
+SpaceBeforeCpp11BracedList: false
 ExperimentalAutoDetectBinPacking: false
 ForEachMacros:   [ foreach, Q_FOREACH, BOOST_FOREACH ]
-IndentCaseLabels: false
-#FixNamespaceComments: true
+IndentCaseLabels: true
+FixNamespaceComments: true
 IndentWrappedFunctionNames: false
-KeepEmptyLinesAtTheStartOfBlocks: true
+KeepEmptyLinesAtTheStartOfBlocks: false
 MacroBlockBegin: ''
 MacroBlockEnd:   ''
 #JavaScriptQuotes: Double
 MaxEmptyLinesToKeep: 1
-NamespaceIndentation: Inner
+NamespaceIndentation: None
 ObjCBlockIndentWidth: 4
 #ObjCSpaceAfterProperty: true
 #ObjCSpaceBeforeProtocolList: true
-PenaltyBreakBeforeFirstCallParameter: 19
-PenaltyBreakComment: 300
-PenaltyBreakFirstLessLess: 120
-PenaltyBreakString: 1000
-
-PenaltyExcessCharacter: 1000000
-PenaltyReturnTypeOnItsOwnLine: 60
+PenaltyBreakBeforeFirstCallParameter: 50
+PenaltyBreakComment: 0
+PenaltyBreakFirstLessLess: 0
+PenaltyBreakString: 70
+PenaltyExcessCharacter: 100
+PenaltyReturnTypeOnItsOwnLine: 90
 PointerAlignment: Left
-SpaceAfterCStyleCast: false
+SpaceAfterCStyleCast: true
 SpaceBeforeAssignmentOperators: true
-SpaceBeforeParens: Never
+SpaceBeforeParens: ControlStatements
 SpaceInEmptyParentheses: false
 SpacesBeforeTrailingComments: 1
-SpacesInAngles:  false
+SpacesInAngles: Never
 SpacesInContainerLiterals: true
 SpacesInCStyleCastParentheses: false
 SpacesInParentheses: false
@@ -110,11 +140,25 @@ SpacesInSquareBrackets: false
 #SpaceBeforeInheritanceColon: true
 
 #SortUsingDeclarations: true
-SortIncludes: true
+SortIncludes: CaseInsensitive
 
-# Comments are for developers, they should arrange them
-ReflowComments: false
+ReflowComments: true
 
 #IncludeBlocks: Preserve
 #IndentPPDirectives: AfterHash
+
+StatementMacros: [
+  'THRUST_EXEC_CHECK_DISABLE',
+  'THRUST_NAMESPACE_BEGIN',
+  'THRUST_NAMESPACE_END',
+  'THRUST_EXEC_CHECK_DISABLE',
+  'CUB_NAMESPACE_BEGIN',
+  'CUB_NAMESPACE_END',
+  'THRUST_NAMESPACE_BEGIN',
+  'THRUST_NAMESPACE_END',
+  '_LIBCUDACXX_BEGIN_NAMESPACE_STD',
+  '_LIBCUDACXX_END_NAMESPACE_STD',
+]
+TabWidth: 2
+UseTab: Never
 ---

.gitlab-ci.yml

@@ -12,6 +12,7 @@ include:
       - /deps-rocm.yaml
       - /deps-windows.yaml
       - /deps-nvcc.yaml
+      - /deps-compiler-acceleration.yaml
       - /gpus-rocm.yaml
       - /gpus-nvcc.yaml
       - /rules.yaml
@@ -46,17 +47,21 @@ copyright-date:
   extends:
     - .deps:rocm
     - .deps:cmake-latest
+    - .deps:compiler-acceleration
   before_script:
     - !reference [".deps:rocm", before_script]
     - !reference [".deps:cmake-latest", before_script]
+    - !reference [".deps:compiler-acceleration", before_script]
 
 .cmake-minimum:
   extends:
     - .deps:rocm
     - .deps:cmake-minimum
+    - .deps:compiler-acceleration
   before_script:
     - !reference [".deps:rocm", before_script]
     - !reference [".deps:cmake-minimum", before_script]
+    - !reference [".deps:compiler-acceleration", before_script]
 
 .install-rocprim:
   script:
@@ -69,8 +74,11 @@ copyright-date:
       -D CMAKE_CXX_COMPILER=hipcc
       -D CMAKE_BUILD_TYPE=Release
       -D BUILD_TEST=OFF
+      -D BUILD_HIPSTDPAR_TEST=OFF
       -D BUILD_EXAMPLE=OFF
       -D ROCM_DEP_ROCMCORE=OFF
+      -D CMAKE_C_COMPILER_LAUNCHER=phc_sccache_c
+      -D CMAKE_CXX_COMPILER_LAUNCHER=phc_sccache_cxx
       -S $ROCPRIM_DIR
       -B $ROCPRIM_DIR/build
     - cd $ROCPRIM_DIR/build
@@ -91,7 +99,7 @@ copyright-date:
     - !reference [.install-rocprim, script]
     - | # Setup env vars for testing
         rng_seed_count=0; prng_seeds="0";
-        if [[ $CI_COMMIT_BRANCH == "develop_stream" ]]; then 
+        if [[ $CI_COMMIT_BRANCH == "develop_stream" ]]; then
             rng_seed_count=3
             prng_seeds="0, 1000"
         fi
@@ -111,6 +119,9 @@ copyright-date:
       -D AMDGPU_TEST_TARGETS=$GPU_TARGETS
       -D RNG_SEED_COUNT=$rng_seed_count
       -D PRNG_SEEDS=$prng_seeds
+      -D CMAKE_C_COMPILER_LAUNCHER=phc_sccache_c
+      -D CMAKE_CXX_COMPILER_LAUNCHER=phc_sccache_cxx
+      -D CMAKE_CUDA_COMPILER_LAUNCHER=phc_sccache_cuda
       -S $CI_PROJECT_DIR
       -B $CI_PROJECT_DIR/build
     - cmake --build $CI_PROJECT_DIR/build
@@ -198,10 +209,10 @@ build:windows:
       -D CMAKE_INSTALL_PREFIX:PATH="$ROCPRIM_DIR/build/install" *>&1
     - \& cmake --build "$ROCPRIM_DIR/build" --target install *>&1
     # Configure and build rocThrust
-    - \& cmake 
-      -S "$CI_PROJECT_DIR" 
-      -B "$CI_PROJECT_DIR/build" 
-      -G Ninja 
+    - \& cmake
+      -S "$CI_PROJECT_DIR"
+      -B "$CI_PROJECT_DIR/build"
+      -G Ninja
       -D CMAKE_BUILD_TYPE=Release
       -D GPU_TARGETS=$GPU_TARGET
       -D BUILD_TEST=ON
@@ -327,10 +338,12 @@ test:rocm-windows-install:
     - .deps:nvcc
     - .gpus:nvcc-gpus
     - .deps:cmake-latest
+    - .deps:compiler-acceleration
     - .rules:manual
   before_script:
     - !reference [".deps:nvcc", before_script]
     - !reference [".deps:cmake-latest", before_script]
+    - !reference [".deps:compiler-acceleration", before_script]
 
 build:cuda-and-omp:
   stage: build
@@ -340,7 +353,7 @@ build:cuda-and-omp:
   tags:
     - build
   variables:
-    CCCL_GIT_BRANCH: v2.3.2
+    CCCL_GIT_BRANCH: v2.4.0
     CCCL_DIR: ${CI_PROJECT_DIR}/cccl
   needs: []
   script:
@@ -349,16 +362,21 @@ build:cuda-and-omp:
     - rm -R $CCCL_DIR/thrust/thrust
     - cp -r $CI_PROJECT_DIR/thrust $CCCL_DIR/thrust
     # Build tests and examples from CCCL Thrust
+    # CCCL 2.4.0 breaks compilation of tests. Compile examples only until we
+    # match v2.5.0.
     - cmake
       -G Ninja
       -D CMAKE_BUILD_TYPE=Release
       -D CMAKE_CUDA_ARCHITECTURES="$GPU_TARGETS"
-      -D THRUST_ENABLE_TESTING=ON
+      -D THRUST_ENABLE_TESTING=OFF
       -D THRUST_ENABLE_EXAMPLES=ON
       -D THRUST_ENABLE_BENCHMARKS=OFF
       -D THRUST_ENABLE_MULTICONFIG=ON
       -D THRUST_MULTICONFIG_ENABLE_SYSTEM_OMP=ON
       -D THRUST_MULTICONFIG_ENABLE_SYSTEM_CUDA=ON
+      -D CMAKE_C_COMPILER_LAUNCHER=phc_sccache_c
+      -D CMAKE_CXX_COMPILER_LAUNCHER=phc_sccache_cxx
+      -D CMAKE_CUDA_COMPILER_LAUNCHER=phc_sccache_cuda
       -B $CI_PROJECT_DIR/build
       -S $CCCL_DIR/thrust
     - cmake --build $CI_PROJECT_DIR/build

CHANGELOG.md

@@ -3,16 +3,18 @@
 Documentation for rocThrust available at
 [https://rocm.docs.amd.com/projects/rocThrust/en/latest/](https://rocm.docs.amd.com/projects/rocThrust/en/latest/).
 
-## (Unreleased) rocThrust 3.2.0 for ROCm 6.4
+## (Unreleased) rocThrust 3.3.0 for ROCm 6.4
 
 ### Added
 * Added extended tests to `rtest.py`. These tests are extra tests that did not fit the criteria of smoke and regression tests. These tests will take much longer to run relative to smoke and regression tests. Use `python rtest.py [--emulation|-e|--test|-t]=extended` to run these tests.
 * Added regression tests to `rtest.py`. These tests recreate scenarios that have caused hardware problems in past emulation environments. Use `python rtest.py [--emulation|-e|--test|-t]=regression` to run these tests.
 * Added smoke test options, which runs a subset of the unit tests and ensures that less than 2gb of VRAM will be used. Use `python rtest.py [--emulation|-e|--test|-t]=smoke` to run these tests.
 * Added `--emulation` option for `rtest.py`
+* Merged changes from upstream CCCL/thrust 2.4.0
 
 ### Changed
 * `--test|-t` is no longer a required flag for `rtest.py`. Instead, the user can use either `--emulation|-e` or `--test|-t`, but not both.
+* Split the contents of HIPSTDPAR's forwarding header into several implementation headers.
 
 ## (Unreleased) rocThrust 3.2.0 for ROCm 6.3
 
@@ -38,6 +40,7 @@ Documentation for rocThrust available at
 
 * Merged changes from upstream CCCL/thrust 2.2.0
   * Updated the contents of `system/hip` and `test` with the upstream changes to `system/cuda` and `testing`
+* Added HIPSTDPAR library as part of rocThrust.
 
 ### Changes
 

CMakeLists.txt

@@ -80,6 +80,7 @@ endif()
 # Disable -Werror
 option(DISABLE_WERROR "Disable building with Werror" ON)
 option(BUILD_TEST "Build tests" OFF)
+option(BUILD_HIPSTDPAR_TEST "Build hipstdpar tests" OFF)
 option(BUILD_EXAMPLES "Build examples" OFF)
 option(BUILD_BENCHMARKS "Build benchmarks" OFF)
 option(DOWNLOAD_ROCPRIM "Download rocPRIM and do not search for rocPRIM package" OFF)
@@ -143,14 +144,14 @@ if(BUILD_TEST OR BUILD_BENCHMARKS)
 endif()
 
 # Tests
-if(BUILD_TEST)
+if(BUILD_TEST OR BUILD_HIPSTDPAR_TEST)
   rocm_package_setup_client_component(tests)
   if (ENABLE_UPSTREAM_TESTS)
     enable_testing()
   endif()
   # We still want the testing to be compiled to catch some errors
   #TODO: Get testing folder working with HIP on Windows
-  if (NOT WIN32)
+  if (NOT WIN32 AND BUILD_TEST)
     add_subdirectory(testing)
   endif()
   enable_testing()