NVIDIA/cccl v3.3.0

on GitHub

Full Changelog: v3.3.0...v3.3.0

What's Changed

📚 Libcudacxx

• [libcudacxx] Fix a typo in the documentation by @caugonnet in #7330
• Add a test for <nv/target> to validate old dialect support. by @wmaxey in #7241

🔄 Other Changes

• Implement cudax::cufile by @davebayer in #6122
• Update linear_congruential_generator with constexpr, tests and a fast discard by @RAMitchell in #6402
• Replace _CCCL_HAS_CUDA_COMPILER() with _CCCL_CUDA_COMPILATION() by @davebayer in #6399
• Remove unnecessary casts in complex multiplication/division by @davebayer in #6670
• Add benchmark batch script by @bernhardmgruber in #6661
• Improvements and testing for inspect_changes CI functionality. by @alliepiper in #6535
• Improve clarity of CCCL assert macro documentation by @jrhemstad in #6675
• Fix oversubscription issue with lit precompile, label hack by @alliepiper in #6554
• Make missing sccache nonfatal. by @alliepiper in #6582
• Address pending comments for make_tma_descriptor by @fbusato in #6662
• Add nvhpc 25.9. by @alliepiper in #6003
• Test building for all arches. by @alliepiper in #6113
• Add nvbench_helper tests to CI. by @alliepiper in #6679
• Add more targets to pytorch build. by @alliepiper in #6685
• Add host std lib version detection by @davebayer in #6678
• Improve CUB benchmark docs by @bernhardmgruber in #6640
• Use if consteval in libcu++ by @davebayer in #6424
• Update docs for _CCCL_IF_CONSTEVAL by @davebayer in #6692
• Fixes issue with select close to int_max by @elstehle in #6641
• Update libcudacxx C++ dialect handling. by @alliepiper in #6693
• Simplifies env usage in DeviceTopK tests by @elstehle in #6680
• Switch to S3 preprocessor cache by @alliepiper in #6561
• fix omp scan bug by @charan-003 in #6560
• Refactor out variant from transform tunings by @bernhardmgruber in #6669
• [libcu++] Waive hierarchy constexpr testing on GCC8 by @pciolkosz in #6707
• Use wrapper with void* argument types for iterator advance/dereference signature by @shwina in #6634
• Restore libcudacxx dialect presets. by @alliepiper in #6705
• Refactor error handling in radix sort dispatch by @bernhardmgruber in #6681
• Remove special dialect handling from cudax build system. by @alliepiper in #6702
• Segmented scan followup by @oleksandr-pavlyk in #6706
• Fix electing leader from any group in cuda::memcpy_async by @bernhardmgruber in #6710
• Avoid scaling twice in ReduceNondeterministicPolicy by @bernhardmgruber in #6711
• Remove special handling of C++ dialect in CUB's build system by @alliepiper in #6713
• [libcu++] Use resource test fixture members through this by @pciolkosz in #6717
• Improves top-k examples to illustrate stream usage by @elstehle in #6723
• Tweak sol.py a bit by @bernhardmgruber in #6721
• Implement PCG64 as extension by @RAMitchell in #6292
• Use PDL in cub::DeviceScan by @bernhardmgruber in #6639
• Fix header in libcudacxx test by @alliepiper in #6726
• Remove dead code. by @alliepiper in #6725
• Add deps on thrust/cub to libcudacxx. by @alliepiper in #6694
• Remove special handling for dialect in Thrust's build system. by @alliepiper in #6722
• [libcu++] Automatically bump up the release threshold of default mempools by @pciolkosz in #6718
• Backport cuda::std::reference_wrapper C++20 features by @davebayer in #6709
• Relax error tolerance for deterministic_device_reduce (RFA) test by @srinivasyadav18 in #6720
• [DOC] Add temp_storage_bytes usage guide by @Aminsed in #6208
• Improve charconv test compile times by @davebayer in #6687
• Move source location builtins directly to <cuda/std/source_location> by @davebayer in #6738
• Small improvements for cuda::ipow by @davebayer in #6736
• Add support for clang's alignment builtins by @davebayer in #6741
• Disable test that is failing in multiple configurations by @miscco in #6745
• Implement std::normal_distribution by @RAMitchell in #6585
• Update cuda::std::span concepts by @davebayer in #6744
• Improve bit builtins support by @davebayer in #6737
• Implement ranges::drop_view by @miscco in #5049
• Improve fp decompose by @davebayer in #6749
• Enable caching of advance/dereference methods for Zipiterator and PermutationIterator by @shwina in #6753
• implement indeterminate_domain from P3826R2 by @ericniebler in #6628
• Fix cuda::std::reference_wrapper noexcept test with gcc-8 by @davebayer in #6757
• cuda.compute: In TransformIterator, use type annotations (if available) to determine the output type of user-provided op by @shwina in #6760
• cuda.compute: Fixes and improvements to function caching by @shwina in #6758
• Fix __throw_cuda_error availability with nvrtc by @davebayer in #6759
• Implement ranges::find_if and ranges::find_if_not by @miscco in #6752
• Fix radix_sort tuning namespace by @bernhardmgruber in #6755
• [libcu++] Add sm_62 arch traits by @pciolkosz in #6772
• fix(readme): Update broken Godbolt example link by @miyanyan in #6773
• Implement CUDA backend for parallel cuda::std::for_each by @miscco in #5610
• Ensure that we properly warn about device lambdas that need to query the return type by @miscco in #6765
• Add missing test for thrust::reduce_into by @Pansysk75 in #6572
• cuda.compute: Add select algorithm based on three_way_partition by @shwina in #6766
• Add queries for CUB ptx version as arch_id by @bernhardmgruber in #6776
• Add operator<< to some CUB enums by @bernhardmgruber in #6774
• cuda.compute: Fix caching of functions that call other functions by @shwina in #6770
• Implement std::exponential_distribution by @RAMitchell in #6750
• Fix issue with libcudacxx header tests. by @alliepiper in #6785
• Add a type and operation enum to CUB by @bernhardmgruber in #6780
• Use conventional order of _CCCL_API friend consistently by @miscco in #6781
• Implement std::binomial_distribution by @RAMitchell in #6747
• Fixes i32 overflow for benchmark data generation of more than INT_MAX number of items by @elstehle in #6809
• Temporarily add upper bound to numba-cuda dependency by @shwina in #6815
• Make cuda capabilities part of cccl config by @davebayer in #6806
• Update std::uniform_real_distribution by @RAMitchell in #6798
• [cub] Implement cub::MaxPotentialDynamicSmemBytes by @davebayer in #6818
• libcudacxx: streamline simple trait aliases by @Aminsed in #6740
• Fix a typo in compute.rst by @shwina in #6826
• Improve our WarpReduce implementation by @miscco in #6814
• Implement cuda::sincos by @davebayer in #6742
• Replace inline ptx with intrinsics by @davebayer in #6810
• [cudax->libcu++] Move buffer type from cudax to libcu++ by @pciolkosz in #6627
• Improve CMake package handling, add MSVC compat flags to libcudacxx's public interface. by @alliepiper in #6791
• Fix arch related cuda::device:: APIs for nvhpc in CUDA mode by @davebayer in #6829
• Implement the new tuning API for DeviceReduce by @bernhardmgruber in #6544
• Replace internal assert with _CCCL_ASSERT in libcu++ by @davebayer in #6825
• Implement std::gamma_distribution by @RAMitchell in #6786
• Run scan benchmark for 2^32 elements by @bernhardmgruber in #6834
• Remove upper bound on numba-cuda by @shwina in #6835
• Use lit for cuda::arch_id and cuda::compute_capability tests by @davebayer in #6775
• Extends DeviceScan tests in preparation for the warpspeed scan implementation by @elstehle in #6836
• smoke test for all_of algorithm by @viralbhadeshiya in #6828
• [CUB][device] Add a env-based overload of the device segmented reductions primitives by @rbourgeois33 in #6674
• Avoid use of cccl namespace macros in cub by @davebayer in #6844
• Beautify vector mismatch reporting by @bernhardmgruber in #6837
• Implement std::lognormal_distribution by @RAMitchell in #6789
• Implement std::weibull_distribution by @RAMitchell in #6797
• [PTX] Add cp.async.bulk.dst.src.mbarrier::complete_tx::bytes.ignore_oob by @bernhardmgruber in #6854
• Complex asinh accuracy refinement by @s-oboyle in #6428
• Allow numpy struct types as initial value for Zipiterator inputs by @shwina in #6861
• In test_device_segmented_scan_api change type from int to unsigned by @oleksandr-pavlyk in #6868
• Add missing doc strings to support old CMake. by @alliepiper in #6869
• avoid error adding pointer to reference in any_resource by @ericniebler in #6875
• smoke test for adjacent_difference by @viralbhadeshiya in #6872
• Bump minimum CMake to 3.18, add CI testing of public packages with it. by @alliepiper in #6871
• [PTX] Regenerate by @bernhardmgruber in #6859
• Implement std::poisson distribution by @RAMitchell in #6748
• [libcu++] Add memory_pool header and correct legacy resources namespace by @pciolkosz in #6852
• [cuda.compute]: Fix issue with get_dtype() not working anymore for pytorch arrays by @NaderAlAwar in #6882
• [cuda.compute]: Add fast path to extract PyTorch array pointer by @NaderAlAwar in #6884
• accommodate new behavior of clang's __builtin_structured_binding_size by @ericniebler in #6888
• [libcu++] Don't require accessibility property on type erased wrappers by @pciolkosz in #6851
• Move launch API from cudax to libcu++ by @pciolkosz in #6667
• [libcu++] Fix minor version compatibility in 13.X by @pciolkosz in #6895
• [libcu++] Leak static CUDA resources and add missing release on memory pool by @pciolkosz in #6892
• Add limited RTX PRO 6000 coverage. by @alliepiper in #6841
• Add quotes and error checking to devcontainer init. by @alliepiper in #6886
• Update std::uniform_int_distribution by @RAMitchell in #6799
• portability macro for checking whether an expression satisfies a concept in a _CCCL_REQUIRES_EXPR clause by @ericniebler in #6890
• Add ptxas local memory usage warnings to cub builds by @davebayer in #6838
• Pcg64 uint128 fallback implementation for MSVC by @RAMitchell in #6746
• [cuda.compute] Add dependency on nvidia-nvvm by @shwina in #6909
• Add missing OpKind docs entries by @ktaletsk in #6910
• Fix overflow issue in histogram even benchmark when the number of bins exceeds what LevelT can represent by @NaderAlAwar in #6908
• [libcu++] Add as_ref() to memory pool types by @pciolkosz in #6900
• Fix exhaustive policy chain pruning test by @bernhardmgruber in #6903
• Extend transform benchmarks to 2^32 elements by @bernhardmgruber in #6920
• Implement std::cauchy_distribution by @RAMitchell in #6787
• Implement std::extreme_value_distribution by @RAMitchell in #6788
• Implement std::fisher_f_distribution by @RAMitchell in #6857
• change concepts portability macros to avoid use of macro EXPAND by @ericniebler in #6366
• Implement cuda::__all_arch_ids and cuda::__is_specific_arch by @davebayer in #6916
• [libcu++] Rename device_transform back to launch_transform by @pciolkosz in #6927
• Add unsupported compiler flag to .clangd by @ericniebler in #6911
• Add an option to use CCCL from CTK for C2H by @bernhardmgruber in #6848
• Avoid waring about missing braces for subobject by @miscco in #6929
• Add missing nvrtc nv target archs by @davebayer in #6880
• Make sure we actually use overflow builtins by @davebayer in #6904
• Implement std::chi_squared_distribution by @RAMitchell in #6856
• [libcu++] Static assert that resource is copyable in buffer constructors by @pciolkosz in #6928
• Use vectorized transform kernel for sizeof(T) < 4 workloads of arity >1 on Hopper by @bernhardmgruber in #6921
• port the trampoline_scheduler from stdexec to cudax::execution by @ericniebler in #6894
• Properly specialize cub functions for __nv_bfloat16 by @miscco in #6931
• [CUB] Fix mask types in block_radix_rank.cuh by @Aminsed in #6189
• [CUB]: Use the new tuning API for nondeterministic reduce by @NaderAlAwar in #6932
• clean up some allocator and memory utilities by @ericniebler in #6939
• Unify operator handling in cuda.compute by @shwina in #6938
• Use integer promotion for warp_reduce by @miscco in #6819
• implement task_scheduler from C++26 ([exec.task.scheduler]) by @ericniebler in #5975
• Implement std::negative_binomial_distribution by @RAMitchell in #6879
• Implement std::student_t_distribution by @RAMitchell in #6858
• Remove [[nodiscard]] from barrier's .arrive(...) method by @davebayer in #6947
• Implement std::geometric_distribution by @RAMitchell in #6924
• [cuda.compute] Refactor code for creating void* wrappers by @shwina in #6941
• Expose not guaranteed determinism to reduce in cuda.compute by @NaderAlAwar in #6926
• Make __cccl_is_floating_point_v consistent with __cccl_is_integer_v by @davebayer in #6952
• Disable __builtin_structured_binding_size with nvcc by @davebayer in #6961
• Provide thrust::find_if benchmark by @gonidelis in #6956
• Remove all usage of old experimental MR macro by @pciolkosz in #6962
• Fix cuda::std::abs for floating points by @davebayer in #6958
• Expose <cuda/std/charconv> by @davebayer in #6672
• Don't use __builtin_bswap128 during constant evaluation by @davebayer in #6967
• Avoid using _CCCL_UNREACHABLE() unless it's necessary by @davebayer in #6948
• Cleanups for random module by @RAMitchell in #6951
• Remove need for hardcoded LevelT for histogram in c.parallel and cuda.compute by @NaderAlAwar in #6915
• Use ublkcp/memcpy_async in transform when dtype size is not a power of two by @NaderAlAwar in #6972
• Add internal cuda::__is_device_or_managed_memory by @fbusato in #6918
• Improve and apply _CCCL_THROW by @fbusato in #6684
• Define _CCCL_ASSERT_IMPL_HOST correctly for clang on Windows by @asmelko in #6971
• Allow if consteval in device code with nvcc 13.1 by @davebayer in #6902
• c.parallel: reuse CUB agent policies for histogram by @NaderAlAwar in #6974
• [libcu++] Dynamically load CUDA library instead of using the runtime by @pciolkosz in #6899
• [libcu++] Uncomment some tests and fix launch include after launch was moved to libcu++ by @pciolkosz in #6966
• Re-enable using std::meow builtins by @davebayer in #6978
• use cooperative_groups in execution::bulk to synchronize across thread blocks by @ericniebler in #6992
• Provide Shared Memory mdspan and accessor by @fbusato in #6703
• Add non-throwing overloads to is_pointer_accessible by @fbusato in #6988
• [cuda.compute]: fix alignment not being set properly for gpu_struct types by @NaderAlAwar in #6995
• Workaround for a potential bug in the driver related to TMA descriptor by @fbusato in #6985
• do not introduce a pack in a structured binding with nvcc by @ericniebler in #6994
• Extract environment boilerplate code from within the device interfaces to a separate header by @gonidelis in #6622
• Remove _CCCL_HAS_CUDA_COMPILER() by @davebayer in #6984
• [libcu++] Fix memory pool and buffer test issues on Windows by @pciolkosz in #6993
• fix off-by-one error in the implementation of cuda::std::__tuple by @ericniebler in #6996
• Small fixes around warpspeed scan by @bernhardmgruber in #6998
• Remove <version> include by @davebayer in #7001
• Upgrade GitHub Actions to latest versions by @salmanmkc in #6991
• cuda.coop: Use cuda.core.experimental.Linker instead of internal numba-cuda _Linker by @shwina in #7011
• Make c2h vector utils constexpr by @davebayer in #7009
• Improves comments on decoupled look-back code example by @elstehle in #7015
• Extract reduce_op_sync into a free function by @bernhardmgruber in #7004
• Remove experimental namespace from cuda.core import by @NaderAlAwar in #7022
• reexpress completion signature transform alias to make clangd happy by @ericniebler in #7026
• Qualify call to __launch_impl in launch.h to avoid ambiguity errors by @ericniebler in #7024
• Rework hierarchy levels by @davebayer in #6957
• [CUB]: use vectorized kernel for triad and add benchmark for dtypes of size 2 by @NaderAlAwar in #7019
• [libcu++] Fix synchronous resource adapter property passing by @pciolkosz in #6976
• [libcu++] Remove _view from the shared memory getter name by @pciolkosz in #6997
• [thrust] Ignore CUDA free errors in thrust memory resource by @pciolkosz in #7002
• [libcu++] Correctly handle extended lambda in cuda::launch by @pciolkosz in #6987
• Use <stdexcept> header unconditionally by @fbusato in #7028
• Error out when nvrtcc cannot parse cuda_thread_count by @bernhardmgruber in #7035
• Allow all public headers to be included with host compilers only by @davebayer in #7012
• [cuda.compute]: Fixes and updates to benchmarks by @shwina in #6999
• Support operations with side-effects (state) in cuda.compute by @shwina in #7008
• Fix cuda::memcpy async edge cases and add more tests by @bernhardmgruber in #6608
• Explicitly set CCCL_TOPLEVEL_PROJECT to OFF when needed by @KyleFromNVIDIA in #7016
• [libcu++] Add explicit alignment specification in buffer by @pciolkosz in #7005
• Use the sccache-dist build cluster for RAPIDS CI jobs by @trxcllnt in #7014
• tidy up the primitive variant type used by cudax::execution by @ericniebler in #7029
• Fix docs by @gevtushenko in #7052
• Disable LDL/STL checks, for failures seen with NVRTC 13.1 by @shwina in #7054
• Enhance DLPack compatibility by @fbusato in #7045
• Support lambdas as operators in cuda.compute by @shwina in #7058
• [libcu++] Make kernel_config member private and allow it in hierarchy queries by @pciolkosz in #7034
• [libcu++] Remove mentions of cuda/event header from docs by @pciolkosz in #7066
• [BUG] use references for mdspan internal methods by @fbusato in #7059
• Avoid invalid compiler warning with VS2026 by @miscco in #7077
• Avoid compiler issue with MSVC _CCCL_UNREACHABLE by @miscco in #7080
• cuda.compute: Allow multiple uses of the same function in single compilation by @shwina in #7072
• Refactor c2h generator to ensure teardown before main exits by @bernhardmgruber in #7067
• Remove cumlprims_mg from RAPIDS workflows/devcontainers by @bdice in #7082
• [DOCS] Clarifies DeviceTopK docs that inputs and output ranges may not overlap by @elstehle in #7078
• Enhance RDC detection and add _CCCL_HAS_DEVICE_RUNTIME() macro by @davebayer in #7049
• Expand warning suppression for braces around subobject by @miscco in #7087
• [STF] Document how to enable assertions by @caugonnet in #7084
• Simplify cuda::host_launch API by @davebayer in #6689
• Improvements to cuda.compute documentation by @shwina in #7061
• [libcu++] Add tests for some buffer members and alignment passing by @pciolkosz in #7055
• [libcu++] Fix driver api test after curand changes by @pciolkosz in #7095
• Add DeviceTransform to device wide CUB docs by @bernhardmgruber in #7101
• Fix incorrect if else logic in fmax by @miscco in #7107
• Add -device-type128 flags only once by @davebayer in #7100
• [libcu++] Check if managed pools are accessible in is_pointer_accessible test by @pciolkosz in #7096
• Fix calculation of necessary bits in feistel projection by @miscco in #7098
• Fix deferred annotations handling in gpu_struct by @shwina in #7121
• Use cudaMemcpyDefault for trivial copies by @bdice in #7006
• Disable NVHPC builds for pull request CI by @miscco in #7135
• Generator for prologue/epilogue by @davebayer in #7099
• Refactor mdspan cuda::std::__detectably_invalid by @fbusato in #6733
• Fix nvrtcc minimum arch for __float128 support by @davebayer in #7119
• Disable cudax with msvc in CI for now by @pciolkosz in #7139
• Simplify namespace definitions by @davebayer in #7104
• Move DLPack include to separate file by @davebayer in #7108
• Replace and deprecate compute_capability::major() and compute_capability::minor() by @davebayer in #7118
• Disable reference_wrapper test for VS2026 by @miscco in #7088
• Clean up hierarchy by @davebayer in #7023
• Implement new tuning API arch dispatching by @bernhardmgruber in #7093
• Improve std:: builtin handling with nvrtc by @davebayer in #7131
• libcu++: silence msvc+nvcc12.9 warning plaguing c.parallel. by @griwes in #7144
• Implement cub::DeviceFind::FindIf by @gonidelis in #2405
• Fix/modernize thrust examples by @Flawxd in #7094
• Modularize chrono by @miscco in #6671
• Those are unused internal traits by @miscco in #7148
• Implement ranges::reverse_view by @miscco in #6751
• Fixes for shuffle_iterator by @RAMitchell in #7130
• [STF] Use execution places without STF contexts by @caugonnet in #7149
• Revert nested namespace change to <nv/target> by @wmaxey in #7151
• Replace internal uses of thrust::tuple with cuda::std::tuple by @miscco in #6629
• Add Android-specific assert handling in __cccl/assert.h by @fbusato in #7156
• Fix make_tma_descriptor() unit test by @fbusato in #7152
• Rename new tuning API policies and fix MSVC warning by @bernhardmgruber in #7103
• Align local vector storage arrays in vec transform by @bernhardmgruber in #7162
• Try and avoid GCC-15 warning about expected ) by @miscco in #7166
• Fix build issues with documentation by @miscco in #7122
• [DOCS] Improves docs for DeviceTopK, clarifying that inputs and output ranges must not overlap by @elstehle in #7086
• Test passing a custom policy to DispatchRadixSort by @bernhardmgruber in #7170
• Avoid benign overflow in __calloc_device by @miscco in #7176
• Fixes for thrust::shuffle by @RAMitchell in #7172
• cub, c.parallel: {lower,upper}_bound by @griwes in #7007
• Initial nvrtcc implementation by @davebayer in #7051
• Implement comparison operators for thrust::reference and thrust::pointer by @miscco in #7190
• Agent Updates by @alliepiper in #7194
• Add missing CUB_RUNTIME_FUNCTION annotations. by @alliepiper in #7195
• Define methods for test ranges by @miscco in #7220
• Fix noexcept specification of extreme_value_distribution by @miscco in #7219
• Drop thrust::detail::is_commutative by @miscco in #7218
• Try and work around NVHPC issue with is_metafunction_defined by @miscco in #7217
• Reenable MSVC cudax CI by @miscco in #7221
• Add support for [[lifetimebound]] by @fbusato in #7155
• [cccl.c] Use function try blocks by @davebayer in #7236
• Do not try to run catch2 tests with nvrtc by @miscco in #7242
• Add .branch_notes. by @alliepiper in #7238
• Make cuda.compute importable in a CPU-only environment by @shwina in #7171
• Drop libcudacxx ABI Evolution clause by @bernhardmgruber in #7247
• [CI] MSVC sccache auth -> file by @alliepiper in #7257
• Implement the new tuning API for DeviceTransform by @bernhardmgruber in #6914
• Refactor some bits of DeviceRadixSort by @bernhardmgruber in #7193
• Drop leftover code after tuning API migration by @bernhardmgruber in #7264
• Fix extracting CUDA stream in cub::DeviceTransform by @bernhardmgruber in #7239
• Run build/test commands with 5h30m timeout on CI. by @alliepiper in #7213
• Change the order of conditions in cuda::barrier by @davebayer in #7259
• Don't run CPU-only import test if the wheel artifact doesn't exist by @shwina in #7270
• Test passing more stream types to cub::DeviceTransform by @bernhardmgruber in #7278
• Add versionadded directives to all public API functions by @cliffburdick in #7215
• Refactor cub::DeviceRadixSort by @bernhardmgruber in #7282
• [FEA]: Add DevEx/Infra ticket templates by @alliepiper in #7261
• Fix __query_or CPO by @miscco in #7266
• Test passing a custom policy to DispatchUniqueByKey by @bernhardmgruber in #7296
• Test passing a custom policy to DispatchSelectIf by @bernhardmgruber in #7294
• Test passing a custom policy to DispatchHistogram by @bernhardmgruber in #7288
• Fix is_address_from for cluster_shared for pre-sm_90 by @davebayer in #7245
• Test passing a custom policy to DispatchThreeWayPartitionIf by @bernhardmgruber in #7295
• cuda.compute: Consolidate caching logic across all algorithms by @shwina in #7281
• Test passing a custom policy to DispatchAdjacentDifference, DispatchMergeSort, DispatchScan, DispatchBatchMemcpy by @bernhardmgruber in #7289
• Test passing a custom policy to DispatchSegmentedSort by @bernhardmgruber in #7307
• Test passing a custom policy to DispatchSegmentedRadixSort by @bernhardmgruber in #7308
• Test passing a custom policy to DispatchSegmentedReduce by @bernhardmgruber in #7311
• Add accessor methods to shared_resource by @bdice in #7315
• cub, c.parallel: change {lower,upper}_bound to return indices. by @griwes in #7320
• [nvrtcc] Add __NVRTCC_USE_NVRTC__ macro to nvrtcc by @davebayer in #7293
• Remove fmtlib from CCCL by @davebayer in #7300
• Fix clang warning about missing braces again by @miscco in #7302
• Test passing a custom policy to DispatchReduceByKey by @bernhardmgruber in #7310
• Fix missing newline for spdx/pragma once by @msarahan in #7306
• mdspan to DLPack by @fbusato in #7027
• Test passing a custom policy to DeviceRleDispatch by @bernhardmgruber in #7314
• Skip checking build prereqs if installing by @wmaxey in #7316
• Implement the new tuning API for DeviceRadixSort by @bernhardmgruber in #6767
• Add cuda.compute APIs for upper_bound and lower_bound by @shwina in #7250
• cuda.compute: Fix deferred annotations handling in signature_from_annotations by @shwina in #7321
• Test passing a custom policy to DispatchScanByKey by @bernhardmgruber in #7309
• Fixup missed feedback on #7311 by @bernhardmgruber in #7323
• Enhance internal vector type utilities by @fbusato in #7327
• Fix MSVC version detection by @miscco in #7305
• Implement parallel cuda::std::reduce by @miscco in #6777
• Remove _CCCL_HAS_INCLUDE by @davebayer in #7304
• Fix narrow conversion in __is_valid_address_range when compling on 32-bit systems by @davebayer in #7333
• [nvrtcc] Add nvrtcc dependency by @davebayer in #7287
• Fix double destroy in vector by @miscco in #7331
• Optimize cuda::add_overflow for unsigned types by @davebayer in #7340
• [STF] Make data place more extensible by @caugonnet in #7252
• Formatters for cuda::arch_id and cuda::compute_capability by @davebayer in #7335
• Drop most parts of thrust::allocator_traits by @miscco in #7286
• [CI] Re-enable RAPIDS builds in PRs by @alliepiper in #7255
• Use arch dispatch workaround on GCC 8-9 as well by @bernhardmgruber in #7349
• Drop partial specialization in friend functions of fast_mod_div by @miscco in #7348
• Add blackduck-sca.yml by @alliepiper in #7360
• [cccl.c]: Disable SASS check for merge_sort pairs by @NaderAlAwar in #7357
• Assert before terminating on throw in device code by @davebayer in #7358
• DLPack to mdspan by @fbusato in #7047
• Test VerifyCodegen only with latest CTK by @davebayer in #7351
• Ensure device_find_if works with non-default-constructible types by @miscco in #7337
• Fix _CCCL_THROW in dlpack_to_mdspan by @fbusato in #7363
• Expose CUDA vector type traits by @fbusato in #7364
• Add some ramblings about symbol visibility by @miscco in #6114
• Fix handling of boolean types in cuda.compute by @shwina in #7389
• Decouple numba-cuda from type system and other internals by @shwina in #7342
• Fix DeviceTransform docs by @bernhardmgruber in #7392
• [libcu++] Add runtime check if memory pools are supported by @pciolkosz in #7339
• Pass -std flag from CLI to cmake in c.parallel build scripts by @NaderAlAwar in #7394
• Migrate cuco HLL by @srinivasyadav18 in #6666
• Add CTK 13.1 CI jobs, devcontainers. by @alliepiper in #6887
• make the abi of __basic_any compatible between c++17 and c++20 by @ericniebler in #7401
• Enforce supported long double format by @davebayer in #7345
• Implement passing stream and memory resource to execution policies by @miscco in #7299
• Reenable overflow builtins with nvc++ 26.1+ by @davebayer in #7414
• Implement the new tuning API for DeviceSegmentedReduce by @bernhardmgruber in #7334
• part deux: make the abi of __basic_any compatible between c++17 and c++20 by @ericniebler in #7405
• Move host stdlib wrappers to <cuda/std/__host_stdlib> directory by @davebayer in #7411
• Improves error handling in CUB algorithms using thrust::triple_chevron calls by @elstehle in #7415
• [DOCS] Clarifies that thrust::reduce_by_key actually selects the last item of a range of consecutively equal keys by @elstehle in #7408
• fix determinism rejection logic for scan by @srinivasyadav18 in #7133
• Upgrade GitHub Actions for Node 24 compatibility by @salmanmkc in #6990
• [libcu++] Fix handle-type mask checks in memory pool tests. by @pciolkosz in #7428
• Avoid deallocate throwing by @gonidelis in #7233
• Drop accidental [[nodiscard]] on constructor by @miscco in #7413
• Use SPDX license headers in thrust/thrust/system/detail/generic by @bernhardmgruber in #7422
• Use SPDX license headers in thrust/thrust/detail by @bernhardmgruber in #7423
• Update Catch2 to 3.12 by @bernhardmgruber in #6067
• Add DeviceTransform benchmarks from pytorch by @bernhardmgruber in #7391
• Refactor cub::ThreadLoad by @bernhardmgruber in #7419
• cuda.compute: Don't attempt to set host_advance by @shwina in #7425
• Use SPDX license headers in thrust/system/cpp by @bernhardmgruber in #7430
• Use SPDX license headers in thrust/random by @bernhardmgruber in #7435
• Use SPDX license headers in thrust/system/cuda by @bernhardmgruber in #7432
• Use SPDX license headers in thrust/iterator by @bernhardmgruber in #7433
• Use SPDX license headers some thrust files 2/2 by @bernhardmgruber in #7437
• Use SPDX license headers in thrust/system/detail/sequential by @bernhardmgruber in #7431
• Use SPDX license headers some thrust files 1/2 by @bernhardmgruber in #7436
• Optimize cuda::sub_overflow by @davebayer in #7344
• [c.parallel]: migrate transform to use jit templates instead of string based implementations by @NaderAlAwar in #7399
• Fix build against libc++ by @miscco in #7448
• cuda.compute: Fix struct comparison (ordering matters) by @shwina in #7451
• Disable batch benchmarks for DeviceTransform by @bernhardmgruber in #7450
• Initial version of DeviceSegmentedTopk for fixed-size segments by @elstehle in #6980
• Document random module by @RAMitchell in #7412
• Check for __cpp_xxx value, not definition by @davebayer in #4811
• Fix missing c2h symbol when compiling with clang-cuda by @davebayer in #7454
• Optimize cuda::add_overflow for signed types by @davebayer in #7343
• Implement parallel cuda::std::transform by @miscco in #7395
• [cuda.coop]: add device-side coop.warp.sum benchmark with pynvbench by @NaderAlAwar in #6846
• Replace typedef with using by @davebayer in #7271
• Refactor cub::ThreadStore by @bernhardmgruber in #7418
• Fix non default constructible input types test for cub::FindIf by @gonidelis in #7447
• Add versionadded annotations to CUB public APIs by @cliffburdick in #7406
• Fix random doc formatting by @RAMitchell in #7492
• Include ninja, ctest, and sccache logs in build artifacts by @trxcllnt in #7487
• Suppress MSVC-specific warnings on linux. by @alliepiper in #7462
• Drop usage of pickle by @miscco in #7491
• Add cub::DeviceTransform N->M API entrypoint by @bernhardmgruber in #7473
• Improve reduce implementation by @miscco in #7493
• Minor libcu++ lit config improvements by @trxcllnt in #7486
• Implement parallel cuda::std::replace by @miscco in #7407
• cuda.compute: improve caching performance by not relying on isinstance() checks for protocols by @shwina in #7501
• Retry failed image pulls 10 times by @trxcllnt in #7488
• Fix DeviceReduce env test for rfa by @bernhardmgruber in #7481
• Pass PARALLEL_LEVEL to cmake --build in ci/build_stdpar.sh by @trxcllnt in #7483
• cuda.compute: Use native CCCL.c support for stateful ops by @shwina in #7500
• [STF] Add exec_place_guard RAII helper for scoped exec_place activation by @caugonnet in #7434
• Implement parallel cuda::std::replace_copy by @miscco in #7410
• Provide Operator Properties by @fbusato in #7240
• Implement parallel cuda::std::generate by @miscco in #7416
• [STF] Fix equality operators for places by @caugonnet in #7494
• Implement parallel cuda::std::count by @miscco in #7382
• Clarify Docker install step for WSL in .devcontainer README by @acosmicflamingo in #7516
• Restrict to numba-cuda less than 0.27 by @shwina in #7529
• Fix caching of functions referencing numpy ufuncs (or any "dotted" functions) by @shwina in #7535
• Add docs for cccl-runtime 3.2 additions and cccl-runtime landing page by @pciolkosz in #7489
• Tweak random docs by @RAMitchell in #7534
• Use modern __syncthreads_or primitive by @gonidelis in #7509
• Fix identity_element tests by @miscco in #7526
• [STF] Use relaxed capture mode by @caugonnet in #7566
• Remove recursion from __internal_is_address_from by @dkolsen-pgi in #7561
• Fix operator identity and absorbing element for char by @bernhardmgruber in #7568
• Rewrite iterators to not depend on Numba by @shwina in #7441
• [Backport branch/3.3.x] Fix ranges_overlap for nvc++ -cuda by @github-actions[bot] in #7599
• [Backport branch/3.3.x] Fix cuda::device::current_arch_id by @github-actions[bot] in #7602
• [Backport branch/3.3.x] Fix cuda::barrier missing accounting of results in try_wait by @github-actions[bot] in #7635
• [Backport branch/3.3.x] Check for _GLIBCXX_USE_CXX11_ABI only when compiling with libstdc++ by @github-actions[bot] in #7631

New Contributors

• @miyanyan made their first contribution in #6773
• @Pansysk75 made their first contribution in #6572
• @rbourgeois33 made their first contribution in #6674
• @ktaletsk made their first contribution in #6910
• @asmelko made their first contribution in #6971
• @salmanmkc made their first contribution in #6991
• @KyleFromNVIDIA made their first contribution in #7016
• @Flawxd made their first contribution in #7094
• @msarahan made their first contribution in #7306
• @acosmicflamingo made their first contribution in #7516

Full Changelog: v3.3.0.dev...v3.3.0