CCCL 2.8.0

What's Changed

• Adds benchmarks for DeviceSelect::Unique by @elstehle in #2359
• CUB - Enable DPX Reduction by @fbusato in #2286
• [CUDAX] add a small c++17 implementation of std::execution (aka P2300) by @ericniebler in #2301
• Add thurst::transform_inclusive_scan with init value by @gonidelis in #2326
• Widen histogram agent constructor to more types by @bernhardmgruber in #2380
• Use a constant for the amount of static SMEM by @bernhardmgruber in #2374
• Add cub::DeviceTransform by @bernhardmgruber in #2086
• Update toolkit to CTK 12.6 by @miscco in #2348
• implement make_integer_sequence in terms of intrinsics whenever possible by @ericniebler in #2384
• Implement cuda::mr::cuda_async_memory_resource by @miscco in #1637
• Drop implementation of thrust::pair and thrust::tuple by @miscco in #2395
• Pull out _LIBCUDACXX_UNREACHABLE into its own file by @miscco in #2399
• Share common compiler flags in new CCCL-level targets. by @alliepiper in #2386
• conditionally include <crt/host_defines.h> from __cccl/execution_space.h header by @ericniebler in #2406
• add some simple utilities for manipulating lists of types by @ericniebler in #2370
• Drop thrusts diagnostic suppression warnings by @miscco in #2392
• [PoC]: Implement cuda::experimental::uninitialized_async_buffer by @miscco in #1854
• Fix thrust package to work with newer FindOpenMP.cmake. by @alliepiper in #2421
• Introduce cccl_configure_target cmake function. by @alliepiper in #2388
• Fix sccache errors in RAPIDS builds by @trxcllnt in #2417
• Replace CUDA C++ Core Libraries with CUDA Core Compute Libraries (only in README.md). by @rwgk in #2424
• Minor cleanup with cuda/atomic by @miscco in #2418
• uninitialized_buffer::get_resource returns a ref to an any_resource that can be copied by @ericniebler in #2431
• Refactor cuda::ceil_div to take two different types by @miscco in #2376
• Reduce PR testing matrix. by @alliepiper in #2436
• Implement cudax::shared_resource by @miscco in #2398
• Increase the libcu++ timeout by @miscco in #2435
• Move c/include/cccl/.h files to c/include/cccl/c/.h by @rwgk in #2428
• Make any_resource emplacable by @miscco in #2425
• Fix issues with __host__ and __device__ definitions by @miscco in #2413
• Make bit_cast play nice with extended floating point types by @miscco in #2434
• Do not include our own string.h file by @miscco in #2444
• Move nightly time by @bdice in #2437
• Remove a ton of lines in thrust tests by @gonidelis in #2356
• [CUDAX] Add placeholder green context type and logical device that can hold both a green ctx and a device by @pciolkosz in #2446
• Fix typo in CCCLBuildCompilerTargets.cmake by @alliepiper in #2453
• Drop superflous compile definition from thrust tests by @miscco in #2450
• Consolidate packages and install rules by @alliepiper in #2456
• Prune CUB's ChainedPolicy by CUDA_ARCH_LIST by @bernhardmgruber in #2154
• fixes merge conflict for policy pruning by @elstehle in #2466
• Add CCCL_ENABLE_WERROR flag. by @alliepiper in #2463
• Add CUB tests for segmented sort/radix sort with 64-bit num. items and segments by @fbusato in #2254
• Propagate compiler flags down to libcu++ LIT tests by @Artem-B in #2420
• Drop remaining uses of _LIBCUDACXX_COMPILER_* by @miscco in #2467
• Avoid C++17 extension in c++11 tests by @miscco in #2469
• Add span to example and templated block size by @Kh4ster in #2470
• Drop Objective C++ support by @miscco in #2468
• removes superfluous template keyword in call to Dereference by @andrewcorrigan in #2482
• Improve build times in several heavyweight libcudacxx tests. by @wmaxey in #2478
• Drop __availability header by @miscco in #2484
• Replace a few more instances of CUDA C++ Core Libraries with CUDA Core Compute Libraries`. by @rwgk in #2447
• Fix common_type specialization for extended floating point types by @miscco in #2483
• Implement some CUDA API calls for async_memory_pool by @miscco in #2455
• Move cudax example project to CCCL project examples. by @alliepiper in #2462
• Disable system header for narrowing conversion check by @miscco in #2465
• Require resources to always provide at least one execution space property by @miscco in #2489
• Rework builtin handling by @miscco in #2461
• Disable execution checks for std::equal by @miscco in #2491
• replace _CCCL_ALWAYS_INLINE with _CCCL_FORCEINLINE by @ericniebler in #2439
• Drop 2 relative includes that snuck in by @miscco in #2492
• re-express the cudax::__tupl::__apply member to make nvc++ happy by @ericniebler in #2493
• Drop badly named _One_of concept by @miscco in #2490
• Unify assert handling in cccl by @miscco in #2382
• Reduce scope of Thrust linkage in cudax. by @alliepiper in #2496
• Centralize CPM logic. by @alliepiper in #2495
• Fix typo in presets. by @alliepiper in #2497
• Refactor away per-project TOPLEVEL flags. by @alliepiper in #2498
• [FEA]: Validate cuda.parallel type matching in build and execution by @rwgk in #2429
• avoid gcc optimizer bug by not force inlining part of thrust::transform by @ericniebler in #2509
• Cleanup and modularize <cuda/std/barrier> by @miscco in #2443
• Consolidate header testing infra. by @alliepiper in #2460
• Add ForEachN from CUB to cccl/c. by @wmaxey in #2378
• Adds support for large number of items in DeviceSelect and DevicePartition by @elstehle in #2400
• Adds support for large number of items to DeviceScan::*ByKey family of algorithms by @elstehle in #2477
• Integrate c/parallel with CCCL build system and CI. by @alliepiper in #2514
• Create a command list utility for nvrtc/jitlink steps. by @wmaxey in #2511
• Fix the example project which the documentation refers too by @caugonnet in #2531
• Enable tests/headertests for c/parallel in all-dev presets. by @alliepiper in #2566
• Rename cudax test targets to match CCCL conventions. by @alliepiper in #2568
• Update project list in issue template by @alliepiper in #2532
• Disable compiler extensions on CCCL targets. by @alliepiper in #2559
• Fixes cub::DeviceMemcpy::Batched to be able to copy from const source pointers by @elstehle in #2573
• Fix documentation error in ci/build_common.sh for -arch by @caugonnet in #2574
• gcc-14 gained the ability to mangle noexcept expressions by @ericniebler in #2565
• Miscellaneous simple fixes by @rwgk in #2575
• Avoid including yvals.h when the compiler is not MSVC. by @wmaxey in #2545
• Fix popc.h when architecture is not x86 on MSVC. by @wmaxey in #2524
• test for exceptions support on msvc with the _CPPUNWIND macro by @ericniebler in #2576
• fix the forwarding of the receiver in the just_from algorithm by @ericniebler in #2569
• Block type pack indexing on NVCC by @wmaxey in #2563
• Cleanup the semaphore headers by @miscco in #2441
• Add _CCCL_GRID_CONSTANT macro by @fbusato in #2530
• Add _CCCL_RESTRICT macro by @fbusato in #2529
• Try to use the same redefinition of __assert_fail as pytorch has by @miscco in #2577
• Fix miscellaneous bugs in cub/iterator documentation. by @rwgk in #2580
• Expose parts of <cuda/std/memory> by @fbusato in #2502
• add a config macro for testing support for inline variables by @ericniebler in #2581
• add dialect macros _CCCL_NO_RTTI and _CCCL_NO_TYPEID by @ericniebler in #2578
• fix misspelling in the _CCCL_NO_VARIABLE_TEMPLATES macro by @ericniebler in #2584
• Add atomic_ref support for 8 and 16b types. by @wmaxey in #2255
• add _LIBCUDACXX_REQUIRES_EXPR to the concepts emulation macros by @ericniebler in #2564
• Ensure CuPy arrays can be used with cuda.parallel too by @leofang in #2335
• assert that cuda::std::declval is noexcept by @ericniebler in #2588
• Revert accidental force push to main. by @wmaxey in #2596
• add __is_callable_v variable template when possible by @ericniebler in #2598
• Cleanup threading support by @miscco in #2507
• CCCL_TOPLEVEL_PROJECT always needs to be defined by @robertmaynard in #2597
• Strip prefix paths from cudax documentation by @caugonnet in #2603
• examples/cudax/CMakeLists.txt should not be executable by @caugonnet in #2594
• [CUDAX] Peer access control on async_memory_pool and async_memory_resource by @pciolkosz in #2587
• Introduce _CCCL_PRAGMA to CCCL by @davebayer in #2610
• Only enable CUDA language when needed. by @alliepiper in #2612
• Modularize latch by @miscco in #2508
• Unify kernel dispatch paths for device reduce between CUB and c.parallel. by @griwes in #2591
• Integrate CUDASTF -> CudaX by @caugonnet in #2572
• [STF] The cmake example for stf was not updated when moving to main branch by @caugonnet in #2618
• Rework head_flags so that we do not rely on the tuple being unevaluated by @miscco in #2619
• [CUDAX] size_bytes in buffer types by @pciolkosz in #2621
• fix portability bug in libcu++'s implementation of char_traits by @ericniebler in #2623
• [cccl/c] Unify some build boilerplate by @wmaxey in #2625
• devcontainer: replace VAULT_HOST with AWS_ROLE_ARN by @jjacobelli in #2604
• Add checks to unique_id by @andralex in #2622
• Add cuda::get_device_address by @miscco in #2611
• Do not pass integral constants to ptx by @miscco in #2229
• Add nvhpc devcontainer to CI by @miscco in #1488
• Use a default initialization for CUDA graph mem alloc nodes by @caugonnet in #2632
• [CUDAX] Add get_name to device_ref by @pciolkosz in #2631
• Add 12.5 devcontainer needed for nvhpc by @miscco in #2634
• a substitute for std::type_info when the compiler doesn't support RTTI by @ericniebler in #2582
• Check for missing inline on functions in public headers. by @alliepiper in #2570
• fix linker errors about multiply defined symbols in STF by @ericniebler in #2641
• Add installation presets and update README with install steps by @alliepiper in #2643
• Fix annotated_ptr test failures. by @wmaxey in #2607
• Issue a deprecation warning when compiling with ICC by @bernhardmgruber in #2076
• Include all python libs in inspect_changes. by @alliepiper in #2648
• Add reusable workflow for updating version in branch with a PR by @wmaxey in #2589
• define _CCCL_NO_RTTI in device code; RTTI isn't available there by @ericniebler in #2639
• Migrate C2H library to top-level library by @alliepiper in #2629
• [CUDAX] Add can_peer_access_to API to device_ref and check both ways access in get_peers by @pciolkosz in #2642
• Use _CCCL_ASSERT for stf by @miscco in #2645
• un-templatize CUDASTF's callback_completion_kernel per @robertmaynard by @ericniebler in #2656
• Implement C++20 <source_location> by @miscco in #2628
• Disable [[no_unique_address]] for clang and mdspan by @miscco in #2646
• [STF] Adapt timing_with_fences test to be more reliable by @caugonnet in #2658
• Add prefetching kernel as new fallback for cub::DeviceTransform by @bernhardmgruber in #2396
• Drop cub::DeviceTransform fallback to cub::DeviceFor by @bernhardmgruber in #2660
• Ignore more files when detecting CI changes. by @alliepiper in #2654
• Add thrust::universal_host_pinned_vector by @bernhardmgruber in #2653
• add new type-list algorithms copy_if, remove_if, find_if, and unique by @ericniebler in #2644
• abide by CCCL config macro naming conventions for _CCCL_PRETTY_FUNCTION and _CCCL_NO_BUILTIN_STRLEN by @ericniebler in #2640
• [STF] Fix how we define multi-dimensional shapes in the documentation by @caugonnet in #2662
• Automate creating a CCCL release from RC tags. by @wmaxey in #2657
• Enable span to work with contiguous std containers in C++17 by @miscco in #2613
• [Version] Update main to v2.8.0 by @github-actions in #2670
• promote the cudax __async/config.cuh to be the config for all of cudax by @ericniebler in #2638
• avoid using nvcc's __type_pack_element before 12.2 by @ericniebler in #2673
• Update ninja_summary.py to support ninja log v6. by @alliepiper in #2663
• Rename new CUB headers to follow conventions. by @alliepiper in #2675
• consistent use of _CUDAX function attributes in the cudax __async/ directory by @ericniebler in #2676
• [CUDAX] Add forwarding reference to functor accepting launch by @pciolkosz in #2677
• [CUDAX] Add initial bits of copy_bytes and fill_bytes by @pciolkosz in #2608
• suppress msvc warning "qualifier applied to function type" in is_function by @ericniebler in #2683
• Disable ublkcp CUB transform kernel for NVHPC by @bernhardmgruber in #2664
• Deprecate thrust::cuda_cub::identity by @bernhardmgruber in #2688
• Remove an unused variable by @bernhardmgruber in #2690
• Setup cudax examples. by @alliepiper in #2697
• portability fixes for _CCCL_BUILTIN_PRETTY_FUNCTION and _CCCL_TYPEID by @ericniebler in #2695
• address portability issues found while using the typelist/typeset utities by @ericniebler in #2694
• Make tests technically correct by initializing the barrier by @miscco in #2701
• Fix invalid memory reads in test_device_batch_copy. by @alliepiper in #2698
• revert config macros _CCCL_CUDACC_BELOW_XX_X to their original semantics by @ericniebler in #2700
• This cleanes up our function objects a bit by @miscco in #2702
• Drop handling of 32bit Windows by @bernhardmgruber in #2689
• Guard inclusion of cuda_runtime_api by using a cuda compiler by @miscco in #2704
• Fix race condition in block_reduce_raking. by @alliepiper in #2699
• Honor CCCL_ENABLE_WERROR for CUDA targets. by @alliepiper in #2705
• Fix nvbench helper compilation for clang-18 by @bernhardmgruber in #2707
• Default ctor of device_ptr and normal_iterator by @bernhardmgruber in #2708
• Add cuda::minimum and cuda::maximum by @Jacobfaib in #2681
• Various fixes to cub::DeviceTransform by @bernhardmgruber in #2709
• Make thrust::transform use cub::DeviceTransform by @bernhardmgruber in #2389
• Ensure that we only use the inline variable trait when it is actually available by @miscco in #2712
• [CUDAX] Rename memory resource and memory pool from async to device by @pciolkosz in #2710
• triple_chevron fix by @fbusato in #2720
• Improve uninitialized_{async_}buffer API by @miscco in #2713
• Fix merge conflict from renaming of async_memory_resource by @miscco in #2728
• [STF] Improve DOT graph outputs by @caugonnet in #2703
• Implement _CCCL_SUPPRESS_DEPRECATED_[PUSH|POP] for ICC and NVHPC by @bernhardmgruber in #2730
• Clean up CUB thread operators by @bernhardmgruber in #2716
• Deprecate/replace more of Thrust functional by @bernhardmgruber in #2105
• Alias cuda::std::identity to __identity by @bernhardmgruber in #2733
• Do not read uninitialized memory for OOB elements. by @alliepiper in #2739
• Add option to conditionally build CUDASTF by @miscco in #2731
• fix cuda::std::bit_width() return type by @fbusato in #2745
• [STF] Option to disable kernel generation in CUDASTF by @caugonnet in #2723
• fix static_extent() return type by @fbusato in #2751
• make the empty parens after level constructors optional by @ericniebler in #2750
• cudax: rename ustdex's __query member function to query by @ericniebler in #2757
• Implement execution policies by @miscco in #2715
• Document some transform iterator corner cases by @bernhardmgruber in #2740
• Shorten the git commit message in the ci scripts by @miscco in #2760
• Separate CUDA and C++ code in C2H by @bernhardmgruber in #2734
• Make get_stream work with queries by @miscco in #2761
• Allow thrust::identity to forward value category by @bernhardmgruber in #2732
• Proclaim Thrust/CUB/libcu++ functor address stability by @bernhardmgruber in #2719
• give declval an implementation that compiles 2x faster by @ericniebler in #2758
• [CUDAX] Add modernized simpleP2P sample by @pciolkosz in #2696
• s/get_delegatee_scheduler/get_delegation_scheduler/ by @ericniebler in #2766
• remove duplicated __apply_cv type trait by @ericniebler in #2754
• merge metaprogramming libs from libcudac++ and µstdex by @ericniebler in #2767
• Doc fix scan by @karthikeyann in #2769
• Remove obsolete ways to set iterator category in CUB by @bernhardmgruber in #2759
• Run thrust::transform benchmarks with more elements by @bernhardmgruber in #2764
• Increase libcu++ timeout by @miscco in #2774
• [STF] Rename the redux access mode into relaxed by @caugonnet in #2776
• Enable type trait aliases in all standard modes by @miscco in #2763
• Optimize, Cleanup, and Expose CUB Thread-Level Reduction by @fbusato in #2756
• Disable execution checks for tuple by @miscco in #2780
• Avoid benchmarking first-time setup in Thrust algorithms by @bernhardmgruber in #2782
• Improve listing benchmarks and text by @bernhardmgruber in #2778
• Fix thrust partition docs typo by @gonidelis in #2791
• Drop unused sanitizer hook by @miscco in #2793
• use _CCCL_HAS_FEATURE instead of plain __has_feature everywhere by @davebayer in #2794
• Avoid make_zip_iterator(make_tuple(...)) by @bernhardmgruber in #2796
• implement _CCCL_HAS_INCLUDE by @davebayer in #2786
• add __cpp_lib_mdspan feature-test macro by @fbusato in #2787
• Remove redundant cmake from example. by @alliepiper in #2804
• change __as_type_list so it doesn't cause the instantiation of its argument by @ericniebler in #2803
• [CUDAX] Enable passing hierarchy levels directly into make_config by @pciolkosz in #2755
• Fix cudacc/cluster detection macro in launch path of libcudacxx tests by @wmaxey in #2811
• [STF] Replace CUDASTF_CODE_GENERATION by !CUDASTF_DISABLE_CODE_GENERATION by @caugonnet in #2797
• Reduce P0 benchmark variations for merge_sort_pairs by @bernhardmgruber in #2798
• Replace macros by lambdas in cub::DeviceTransform by @bernhardmgruber in #2817
• Add nvrtc_sm_top_level::add_link_list() and use in c/parallel/src/reduce.cu by @rwgk in #2781
• give completion_signatures a fast lookup cache by @ericniebler in #2812
• implement new compiler checks for NVHPC by @davebayer in #2816
• Unify [CCCL|CUB|THRUST]_ENABLE_BENCHMARKS by @bernhardmgruber in #2827
• Remove traces of metal from CCCL by @bernhardmgruber in #2828
• Move our CUDACC version checks towards the new version check design by @miscco in #2826
• Extend CUB benchmarking documentation by @bernhardmgruber in #2831
• Remove all warm-up runs from Thrust benchmarks by @bernhardmgruber in #2838
• Utility scripts for benchmark database by @gevtushenko in #2847
• [CUDAX] Add missing sm_61 traits by @pciolkosz in #2848
• Move _CCCL_COMPILER_ICC to the new macro by @miscco in #2849
• Fix wrong include in Thrust benchmark by @bernhardmgruber in #2854
• Add missing include by @bernhardmgruber in #2855
• Move _CCCL_COMPILER_GCC to the new macro by @davebayer in #2850
• Add benchmarking and tuning presets by @bernhardmgruber in #2856
• Fix race condition in block-RLD test harness. by @alliepiper in #2706
• Add MatX build to CCCL CI by @alliepiper in #2682
• Fix DeviceSegmentedSort NVTX range name by @davidwendt in #2857
• Make discovery mechanism for cuda/_include directory compatible with pip install --editable by @rwgk in #2846
• add missing DOXYGEN_* predefined macros when building the cudax docs by @ericniebler in #2858
• correct the names of shared_resource's async allocate/deallocate members by @ericniebler in #2880
• [Docs/PTX] Add device tensor map init example by @ahendriksen in #1983
• Fix rst typos in benchmarking.html by @gonidelis in #2868
• Include use of NVHPC in CUB/Thrust magic namespace by @bernhardmgruber in #2771
• backport to_underlying by @davebayer in #2853
• move _CCCL_COMPILER_CLANG to the new macro by @davebayer in #2859
• Automate release branch creation by @wmaxey in #2685
• Add thrust_create_target DISPATCH option. by @alliepiper in #2844
• for_each_in_extent by @fbusato in #2518
• Fix old gcc version check by @davebayer in #2904
• Move implementation of _LIBCUDACXX_TEMPLATE to CCCL by @miscco in #2832
• Try to work around issue with NVHPC in conjunction with older CTK versions by @miscco in #2889
• Refactor nvbench helper less_t by @bernhardmgruber in #2905
• add "interface" to _CCCL_PUSH_MACROS by @ericniebler in #2919
• Replace inconsistent Doxygen macros with _CCCL_DOXYGEN_INVOKED by @ericniebler in #2921
• implement C++26 std::span::at by @davebayer in #2924
• move msvc compiler macros to new version by @davebayer in #2885
• Reorganize PTX tests to match generator by @bernhardmgruber in #2930
• Reorganize PTX docs to match generator by @bernhardmgruber in #2929
• Improve build instructions for libcu++ by @miscco in #2881
• Reorganize PTX headers to match generator by @bernhardmgruber in #2925
• implement C++26 std::span's constructor from std::initializer_list by @davebayer in #2923
• Add tuple protocol to cuda::std::complex from C++26 by @davebayer in #2882
• Add missing qualifier for cuda namespace by @bernhardmgruber in #2940
• Try to fix a clang warning: by @bernhardmgruber in #2941
• minor consistency improvements in concepts macros by @ericniebler in #2928
• Drop some of the mdspan fold implementation by @miscco in #2949
• [STF] Implement CUDASTF_DOT_TIMING for the ctx.cuda_kernel construct by @caugonnet in #2950
• Avoid potential null dereference in annotated_ptr by @miscco in #2951
• make compiler version comparison utility generic by @davebayer in #2952
• Add SM100 descriptor to target by @miscco in #2954
• Regenerate cuda::ptx headers/docs and run format by @bernhardmgruber in #2937
• Regenerate cuda::ptx test by @bernhardmgruber in #2953
• Do not include extended floating point headers if they are not needed by @miscco in #2956
• [CUDAX] Add copy_bytes and fill_bytes overloads for mdspan by @pciolkosz in #2932
• add a _CCCL_NO_CONCEPTS config macro by @ericniebler in #2945
• remove definition of macro (_LIBCUDACXX_NO_RTTI) that is no longer used by @ericniebler in #2957
• Avoid symbol clashes with libc++ by @miscco in #2955
• Add more CUB transform benchmarks by @bernhardmgruber in #2906
• Start reworking our math functions by @miscco in #2749
• Drop memory resources in libcu++ by @miscco in #2860
• std::dims by @fbusato in #2961
• Fix merge conflict from moving resources up a namespace by @miscco in #2965
• [CUDAX] Add a way to combine thread hierarchies by @pciolkosz in #2746
• Require approval to run CI on draft PRs by @bdice in #2969
• fix thread-reduce performance regression by @fbusato in #2944
• add a __type_switch utility and use it the ptx generator by @ericniebler in #2946
• replace use of old _CONCEPT_FRAGMENT macro in cudax by @ericniebler in #2973
• remove vestigal uses of the old DOXYGEN_SHOULD_SKIP_THIS macro by @ericniebler in #2978
• Fix proclaim_copyable_arguments for lambdas by @bernhardmgruber in #2833
• Forward declare half types in cuda::ptx by @ahendriksen in #2981
• Fix tuning benchmark for cub::DeviceTransform by @bernhardmgruber in #2970
• fix old gcc version check by @davebayer in #2989
• Fix a typo in thrust/binary_search.h (#2980) by @hzhangxyz in #2992
• Enable assertions for CCCL users in CMake Debug builds by @bernhardmgruber in #2986
• Fix CMake warning for FindPythonInterp by @bernhardmgruber in #2982
• Further clarify host compiler support by @bernhardmgruber in #2991
• Drop _CCCL_ELSE_IF_CONSTEXPR by @bernhardmgruber in #2966
• implement C++26 std::ignore by @davebayer in #2922
• make the upper limit on TMP loop unrolling configurable by @ericniebler in #2971
• Update docs with recent features by @davebayer in #2994
• Restore thrust single config options. by @alliepiper in #2977
• Document tuning DB comparison scripts by @bernhardmgruber in #2968
• Build CUB and Thrust tests with assertions by @bernhardmgruber in #2987
• Issue a deprecation warning when compiling with Visual Studio 2017 by @bernhardmgruber in #2990
• Guard forward declarations of extended FP types by @bernhardmgruber in #2998
• [STF] Create dot sections to possibly collapse nodes when displaying large DOT graphs by @caugonnet in #2988
• Remove redundant pre c++11 checks by @davebayer in #2999
• Avoid checking unsigned values for negativity by @bernhardmgruber in #2997
• Rename thrust example version.cu to print_version.cu by @j3soon in #3002
• don't bother sync-ing a stream with itself by @ericniebler in #3007
• Backport is_scoped_enum by @davebayer in #3003
• Put monostate in <utility> by @davebayer in #3000
• backport std integer comparison functions to C++11 by @davebayer in #2805
• backport forward_like by @davebayer in #2995
• Document how to profile benchmarks by @bernhardmgruber in #3015
• Update Thrust examples ReadMe by @bernhardmgruber in #3004
• Deprecate public CUB/Thrust deprecation macros by @bernhardmgruber in #3010
• Fix libcudacxx example by @j3soon in #3013
• Refactor BlockLoad test by @bernhardmgruber in #3005
• Fix NVBench profile flags in docs by @bernhardmgruber in #3016
• Update RAPIDS to 25.02. by @bdice in #2967
• Tweak tuning database plot and comparison scripts by @bernhardmgruber in #2883
• Allow passing debug flags to NVRTC in libcudacxx tests by @wmaxey in #3020
• Add missing template parameter to BlockRadixRank example. by @esoha-nvidia in #2736
• Fix value range overflows in tests by @Artem-B in #3022
• Avoid relative includesthat have slipped in by @miscco in #3042
• Fix word count example in Thrust by @caugonnet in #3014
• revise <cuda/std/version> by @davebayer in #3043
• Replace thrust::swap by cuda::std::swap by @bernhardmgruber in #2985
• add a converting constructor to cudax::stream_ref from cuda::stream_ref by @ericniebler in #3052
• [CUDAX] Remove launch overloads taking dimensions and make everything except make_hierarchy return kernel_config by @pciolkosz in #2979
• move sender support library to __async/sender/ by @ericniebler in #3056
• [cuda.cooperative] Add block.load and block.store. by @brycelelbach in #2693
• Backport unreachable by @davebayer in #3018
• Define the destructor of kernel_arg by @miscco in #3060
• Add missing __syncthreads() to test by @miscco in #3061
• Add assertions in the mdspan accessors that we are not out of bounds by @miscco in #3055
• Do not use cudaGetErrorString on GPU. by @Artem-B in #3059
• Reduce number of per-PR CI jobs. by @alliepiper in #2931
• Rework CUDA compiler checks by @davebayer in #3057
• implement C++23 invoke_r by @davebayer in #3041
• Consider NV_TARGET_SM_INTEGER_LIST for ChainedPolicy pruning by @bernhardmgruber in #2772
• Add environment to encapsulate information needed for cudax::vector by @miscco in #2775
• We should not call cudaGetErrorString on device by @miscco in #3062
• Introduce cuda.cooperative overloads not requiring temporary storage by @gevtushenko in #2528
• basic_any: a utility for defining type-erasing wrappers in terms of an interface description by @ericniebler in #2633
• Fix Thrust/CUB tests by adding empty base opt-ins to iterator classes by @wmaxey in #3066
• Don't use exact comparison for FP values. by @Artem-B in #2742
• Use consistent spelling for aliasing select benchmarks by @bernhardmgruber in #3073
• Improve handling of language level features by @miscco in #3069
• Only tune streaming DeviceSelect versions for 64-bit offsets by @bernhardmgruber in #3072
• Disable nvrtc workaround by @miscco in #1116
• fix assorted problems in cudax memory resource equality fns by @ericniebler in #3079
• Support fancy iterators in cuda.parallel by @rwgk in #2788
• fix feature test for operator<=> by @ericniebler in #3075
• Mark test as potentially passing by @miscco in #3078
• Avoid padding warning with MSVC by @miscco in #3077
• Improve CUB tuning documentation by @bernhardmgruber in #3058
• Optimise tuning compile-time by @bernhardmgruber in #3074
• Use consistent spelling for CounterT in histogram benchmarks by @bernhardmgruber in #3089
• [Improvement] Don't require specifying output type when constructing TransformIterator (cuda.parallel) by @shwina in #3083
• simplify the definition of the basic_any class template by @ericniebler in #3085
• Use only signed offset types in CUB benchmarks by @bernhardmgruber in #3087
• Improve readability of DispatchSelectIf parameterization by @bernhardmgruber in #3092
• [cudax] Simplify implementation of device attributes by @davebayer in #3084
• suppress -Werror=empty-body in char_traits implementation by @ericniebler in #3098
• help older clang and gcc to disambiguate basic_any<__ireference<I>> and basic_any<I&> bases by @ericniebler in #3102
• [PERF] cuda.parallel: Cache intermediate results to improve performance of cudax.reduce_into by @shwina in #3001
• [Improvement] cuda.parallel: Don't require value_type when constructing iterators by @shwina in #3105
• Fix zip and permutation iterator EBO on MSVC by @wmaxey in #3106
• Avoid signed unsigned warnings in annotated_ptr test by @miscco in #3076
• Changes DispatchScan[ByKey] documentation to advise using unsigned offset types by @elstehle in #3111
• [STF] reduce access mode by @caugonnet in #2830
• add support for comparing type-erased wrappers to non-type-erased objects by @ericniebler in #3100
• backport byte by @davebayer in #3091
• Add bound checks for each dimension of mdspan by @fbusato in #3065
• Move some CUB tunings to dedicated headers by @bernhardmgruber in #3096
• [CUDAX] Add combine API to kernel_config and allow adding default configuration to kernel functors by @pciolkosz in #3082
• Extend tuning guide by @bernhardmgruber in #3117
• Densen sm90 policy by @gonidelis in #3121
• Fix a typo in the documentation of cub::DeviceReduce::Reduce by @caugonnet in #3123
• Cleanup select if tuning by @bernhardmgruber in #3120
• Modularize <cuda/std/cstddef> by @davebayer in #3119
• Use programmatic dependent launch in CUB merge sort by @bernhardmgruber in #3114
• Refactor selecting default tuning for select_if by @bernhardmgruber in #3124
• Refactor SM90 radix_sort tuning by @bernhardmgruber in #3125
• [STF] Improved sparse CG example and rename scalar to scalar_view by @caugonnet in #3112
• [CUDAX] Fix the other copy of vector_add after migration to use configs in launch by @pciolkosz in #3129
• Refactor cub histogram tuning by @bernhardmgruber in #3128
• Refactor RLE tuning by @bernhardmgruber in #3127
• Make PDL available with CTK 12.0 by @bernhardmgruber in #3136
• Refactor reduce_by_key tuning by @bernhardmgruber in #3137
• Refactor scan tunings by @bernhardmgruber in #3138
• Fix analyze.py bug by @gonidelis in #3067
• Refactor scan_by_key tuning by @bernhardmgruber in #3139
• Refactor three_way_parition tuning by @bernhardmgruber in #3140
• Clarify passing ValueT to scan_by_key tuning by @bernhardmgruber in #3143
• Move remaining CUB policy hubs to tuning headers by @bernhardmgruber in #3141
• [Internal Cleanup] pre-commit ruff (excluding docs/tools, libcudacxx/test) by @rwgk in #3110
• Add Python codeowners by @jrhemstad in #3150
• make basic_any compile for device by stubbing out the virtual tables by @ericniebler in #3109
• Refactoring unique by key by @gonidelis in #3145
• Add missing header in bench scan exclusive base header by @gonidelis in #3157
• Use synchronize_optional for device-to-device copy in thrust::copy() by @davidwendt in #3149
• [Internal Cleanup] pre-commit ruff libcudacxx/tests by @rwgk in #3152
• Clarify unknown tuning axis are ignored by @bernhardmgruber in #3156
• address portability issue in basic_any with older nvcc versions by @ericniebler in #3160
• Add limited H100 testing for CUB by @jrhemstad in #3151
• Unify policy hub handling and update documentation by @bernhardmgruber in #3142
• make the _CCCL_REQUIRES_EXPR macro more robust by @ericniebler in #3164
• [Refactor] cuda.parallel: Simplify TransformIterator implementation and refactor iterators to derive from a common base by @shwina in #3118
• the streams created by cudax::stream should not synchronize with the null stream by @ericniebler in #3167
• [STF] Implement CUDASTF_DOT_TIMING for the host_launch construct by @caugonnet in #3170
• Add support for sm_101 and sm_101a to NV_TARGET by @bernhardmgruber in #3166
• implement C++23 byteswap by @davebayer in #3093
• Unifies large problem test helper infrastructure by @elstehle in #3171
• Deprectate C++11 and C++14 for libcu++ by @miscco in #3173
• Implement abs and div from cstdlib by @davebayer in #3153
• Fix missing radix sort policies by @bernhardmgruber in #3174
• Introduces new DeviceReduce::Arg{Min,Max} interface with two output iterators by @elstehle in #3148
• Extend tuning documentation by @bernhardmgruber in #3179
• Add codespell pre-commit hook, fix typos in CCCL by @bdice in #3168
• Fix parameter space for TUNE_LOAD in scan benchmark by @bernhardmgruber in #3176
• Fix various old compiler version checks by @davebayer in #3178
• Implement ADL-proof std::projected from C++26 by @davebayer in #3175
• Fix pre-commit config for codespell and remaining typos by @shwina in #3182
• Massive cleanup of our config by @miscco in #3155
• Fix UB in atomics with automatic storage by @wmaxey in #2586
• Refactor the source code layout for cuda.parallel by @shwina in #3177
• new type-erased memory resources by @ericniebler in #2824
• rename _LIBCUDACXX_DECLSPEC_EMPTY_BASES to _CCCL_DECLSPEC_EMPTY_BASES by @ericniebler in #3186
• Document address stability of thrust::transform by @bernhardmgruber in #3181
• turn off cuda version check for clangd by @ericniebler in #3194
• [STF] jacobi example based on parallel_for by @caugonnet in #3187
• Fixes pre-CTK 11.5 diag suppression issues by @elstehle in #3189
• Prefer c2h::type_name over c2h::demangle by @bernhardmgruber in #3195
• Fix memcpy_async* tests by @ahendriksen in #3197
• Add type annotations and mypy checks for cuda.parallel by @shwina in #3180
• Fix rendering of cuda.parallel docs by @shwina in #3192
• Enable PDL for DeviceMergeSortBlockSortKernel by @bernhardmgruber in #3199
• Adds support for large num_items to DeviceReduce::{ArgMin,ArgMax} by @elstehle in #2647
• Fixes for Python 3.7 docs environment by @shwina in #3206
• Adds support for large number of items to DeviceTransform by @elstehle in #3172
• cp_async_bulk: Fix test by @ahendriksen in #3198
• cudax fixes for msvc 14.41 by @ericniebler in #3200
• avoid instantiating class templates in is_same implementation when possible by @ericniebler in #3203
• Fix: make launchers a CUB detail; make kernel source functions hidden. by @griwes in #3209
• help the ranges concepts recognize standard contiguous iterators in c++14/17 by @ericniebler in #3202
• unify macros and cmake options that control the suppression of deprecation warnings by @ericniebler in #3220
• Fx thread-reduce performance regression by @fbusato in #3225
• cuda.parallel: In-memory caching of cuda.parallel build objects by @shwina in #3216
• clean up the cuda::std::span implementation with minimal c++14 range support by @ericniebler in #3211
• use generalized concepts portability macros to simplify the range concept by @ericniebler in #3217
• Use Ruff to sort imports by @shwina in #3230
• Fix scan / sm90 perf regression by @gevtushenko in #3236
• [STF] Logical token by @caugonnet in #3196
• Fix ReduceByKey tuning by @gevtushenko in #3240
• Fix RLE tuning by @gevtushenko in #3239
• cuda.parallel: Forbid non-contiguous arrays as inputs (or outputs) by @shwina in #3233
• Backport to 2.8: Make CUB NVRTC commandline arguments come from a cmake template (#3292) by @bernhardmgruber in #3322
• Backport to 2.8: Deprecate GridBarrier and GridBarrierLifetime (#3258) by @bernhardmgruber in #3288
• Backport to 2.8: Deprecate cub::Swap (#3333) by @bernhardmgruber in #3350
• Backport to 2.8: Deprecate Thrust's cpp_compatibility.h macros (#3299) by @bernhardmgruber in #3321
• Backport to 2.8: Deprecate cub::IterateThreadStore (#3337) by @bernhardmgruber in #3351
• Backport to 2.8: Deprecate thrust::null_type (#3367) by @bernhardmgruber in #3373
• Backport to 2.8: Review/Deprecate CUB util.ptx for CCCL 2.x (#3342) by @bernhardmgruber in #3389
• Backport to 2.8: Deprecate thrust::optional (#3307) by @bernhardmgruber in #3393
• Backport to 2.8: Deprecate thrust::numeric_limits (#3366) by @bernhardmgruber in #3392
• Backport to 2.8: Redefine and deprecate thrust::remove_cvref (#3394) by @bernhardmgruber in #3420
• Backport to 2.8: Replace and deprecate thrust::cuda_cub::terminate (#3421) by @bernhardmgruber in #3425
• [BACKPORT]: Deprecate cub::{min, max} and replace internal uses with those from libcu++ (#3419) by @miscco in #3447
• Backport to 2.8: Deprecate thrust::async (#3324) by @bernhardmgruber in #3388
• [BACKPORT]: Moves agents to detail::<algorithm_name> namespace by @elstehle in #3454
• Backport to 2.8: Deprecate a few CUB macros (#3456) by @bernhardmgruber in #3463
• [BACKPORT]: Fix assert definition for NVHPC due to constexpr issues (#3418) by @miscco in #3448
• Backport to 2.8: Deprecate cub::DeviceSpmv (#3320) by @bernhardmgruber in #3374
• Backport to 2.8: some FP8 support by @bernhardmgruber in #3479
• Backport to 2.8: Deprecate block/warp algo specializations (#3455) by @bernhardmgruber in #3481
• Backport to 2.8: Refactor limits and climits (#3221) by @bernhardmgruber in #3488
• Backport to 2.8: Fix typo in limits (#3491) by @bernhardmgruber in #3498
• Backport to 2.8: Update upload-pages-artifact to v3 (#3423) by @bernhardmgruber in #3513
• Backport to 2.8: Implement cuda::std::numeric_limits for __half and __nv_bfloat16 (#3361) by @bernhardmgruber in #3490
• Backport PRs #3201, #3523, #3547, #3580 to the 2.8.x branch. by @rwgk in #3536
• [Backport 2.8] work around msvc bug exposed by __type_index in type_list.h (#3487) by @wmaxey in #3537
• [Backport] #3572 to the 2.8.x branch. by @miscco in #3605
• Backport to 2.8: Specialize cuda::std::numeric_limits for FP8 types (#3478) by @bernhardmgruber in #3492
• Backport to 2.8: Deprecate thrust universal iterator categories (#3461) by @bernhardmgruber in #3471
• Backport to 2.8: Deprecate and replace thrust::cuda_cub iterators (#3422) by @bernhardmgruber in #3510
• Backport to 2.8: Deprecate thrust macros from type_deduction.h (#3501) by @bernhardmgruber in #3511
• Backport to 2.8: Deprecate macros from cuda/detail/core/util.h (#3504) by @bernhardmgruber in #3520
• [BACKPORT]:: Try to always include the definition of barrier_native_handle when needed (#3556) by @miscco in #3569
• Backport to 2.8: Deprecates tuning policy hubs by @elstehle in #3531
• [Backport 2.8] Add extended data type macro identification by @fbusato in #3586
• Backport to 2.8: Deprecate thrust logical meta functions (#3538) by @bernhardmgruber in #3567
• Backport to 2.8: Refactor (#3561) by @bernhardmgruber in #3566
• Backport to 2.8: Tune cub::DeviceTransform for Blackwell (#3543) by @bernhardmgruber in #3565
• Backport to 2.8: Deprecate and replace CUB_IS_INT128_ENABLED (#3427) by @bernhardmgruber in #3629
• Backport to 2.8: Deprecate CUB iterators existing in Thrust (#3304) by @bernhardmgruber in #3534
• Backport to 2.8: Deprecate thrust event, future and more (#3457) by @bernhardmgruber in #3512
• Backport to 2.8: PTX support for Blackwell by @bernhardmgruber in #3624
• Backport to 2.8: Support FP16 traits on CTK 12.0 (#3535) by @bernhardmgruber in #3625
• [Backport 2.8] Deprecate AgentSegmentFixupPolicy by @fbusato in #3638
• Backport to 2.8: PTX: fix cp.async.bulk.tensor and mbarrier.arrive (#3628) by @bernhardmgruber in #3630
• Backport to 2.8: Suppress execution checks for vocabulary types (#3578) by @miscco in #3599
• [BACKPORT]: Try and get rapids green (#3503) by @miscco in #3598
• Backport to 2.8: Internalize triple_chevron (#3648) by @bernhardmgruber in #3650
• [BACKPORT]: Ensure that headers in <cuda/*> can be build with a C++ only compiler (#3472) by @miscco in #3651
• Backport to 2.8: __builtin_isfinite is only available above nvrtc 12.2 by @leofang in #3653
• [Backport 2.8.x] Backport #3575 deprecating old ABIs in libcudacxx by @wmaxey in #3660
• [Backport 2.8.x] Backport [nv/target] Add sm_120 macros. (#3550) by @wmaxey in #3661
• Backport to 2.8: Add b200 policies for device.select.if,flagged,unique (#3545) by @bernhardmgruber in #3667
• Backport to 2.8: Add b200 tunings for radix_sort.pairs (#3626) by @bernhardmgruber in #3668
• [Backport branch/2.8.x] PTX: mbarrier.{test,try}_wait: Fix return value by @github-actions in #3672
• Backport to 2.8: Add b200 tunings for radix_sort.keys (#3611) by @bernhardmgruber in #3655
• [Backport branch/2.8.x] Fix issues with nvrtc compilation by @github-actions in #3674
• [Backport branch/2.8.x] Add b200 policies for cub.select.unique_by_key by @github-actions in #3673
• [Backport branch/2.8.x] Deprecate cub::FpLimits in favor of cuda::std::numeric_limits by @github-actions in #3658
• Backport to 2.8: Deprecate cub::AliasTemporaries (#3679) and cub::PolicyWrapper (#3681) by @bernhardmgruber in #3690
• [Backport branch/2.8.x] Internalize cub::KernelConfig by @github-actions in #3688
• Backport to 2.8: Fix transform_iterator (#3652) and Deprecate thrust::identity (#3649) by @bernhardmgruber in #3693
• Backport to 2.8: Add b200 policies for cub.device.run_length_encode.encode,non_trivialruns (#3546) by @bernhardmgruber in #3704
• [BACKPORT] Remove cugraph-ops from RAPIDS 25.04 builds. (#3675) by @miscco in #3696
• Backport to 2.8: Make thrust iterators work with NVRTC (#3676) and replace CUB iterators by Thrust ones (#3480) by @bernhardmgruber in #3697
• [Backport branch/2.8.x] Deprecate cub::RegBoundScaling and cub::MemBoundScaling by @github-actions in #3706
• [backport 2.8] Deprecate and replace Int2Type by @fbusato in #3705
• [Backport branch/2.8.x] Add b200 policies for partition.three_way by @github-actions in #3710
• Backport to 2.8: Deprecate cub::Trait::CATEGORY|PRIMITIVE|NULL_TYPE (#3689) by @bernhardmgruber in #3703
• [Backport branch/2.8.x] Add b200 tunings for scan.exclusive.by_key by @github-actions in #3719
• Backport to 2.8: B200 reduce.by_key tunings by @bernhardmgruber in #3726
• Backport to 2.8: B200 tunings for histogram by @bernhardmgruber in #3728
• Backport to 2.8: B200 reduce tunings by @bernhardmgruber in #3735
• Backport to 2.8: Add b200 policies for cub.device.partition.flagged,if (#3617) by @bernhardmgruber in #3736
• Backport to 2.8: Add b200 tunings for scan.exclusive.sum (#3559) by @bernhardmgruber in #3738
• [Backport branch/2.8.x] fix ::cuda::discard_memory by @github-actions in #3737
• [Backport branch/2.8.x] Fix cub trait deprecations by @github-actions in #3744
• [Backport branch/2.8.x] [Automation] Add release workflow for tagging and testing new RCs by @github-actions in #3754
• Suppress deprecatings on logical meta functions by @bernhardmgruber in #3795
• Revert back to cub::Traits::CATEGORY|PRIMITIVE by @bernhardmgruber in #3866
• [2.8.x] Disable [[no_unique_address]] for MSVC (#3757) by @miscco in #3869
• [Backport branch/2.8.x] do not try to use clang-19's support for c++26 pack indexing by @github-actions in #3903

New Contributors

• @Artem-B made their first contribution in #2420
• @Kh4ster made their first contribution in #2470
• @andrewcorrigan made their first contribution in #2482
• @jjacobelli made their first contribution in #2604
• @andralex made their first contribution in #2622
• @Jacobfaib made their first contribution in #2681
• @karthikeyann made their first contribution in #2769
• @davidwendt made their first contribution in #2857
• @hzhangxyz made their first contribution in #2992
• @j3soon made their first contribution in #3002
• @esoha-nvidia made their first contribution in #2736

Full Changelog: v2.7.0...v2.8.0