cuda-cccl 0.3.0__cp312-cp312-manylinux_2_24_aarch64.whl → 0.3.1__cp312-cp312-manylinux_2_24_aarch64.whl

cuda/cccl/headers/include/cub/device/dispatch/dispatch_reduce_deterministic.cuh

@@ -328,11 +328,6 @@ struct DispatchReduceDeterministic
     // Alias the allocation for the privatized per-block reductions
     deterministic_accum_t* d_block_reductions = (deterministic_accum_t*) allocations[0];
 
-    if (num_chunks > 1 && !detail::all_iterators_support_add_assign_operator(::cuda::std::int32_t{}, d_in))
-    {
-      return cudaErrorInvalidValue;
-    }
-
     auto d_chunk_block_reductions = d_block_reductions;
     for (int chunk_index = 0; chunk_index < num_chunks; chunk_index++)
     {
@@ -372,7 +367,7 @@ struct DispatchReduceDeterministic
 
       if (chunk_index + 1 < num_chunks)
       {
-        detail::advance_iterators_inplace_if_supported(d_in, num_current_items);
+        d_in += num_current_items;
         d_chunk_block_reductions += current_grid_size;
       }
 

cuda/cccl/headers/include/cub/device/dispatch/dispatch_reduce_nondeterministic.cuh

@@ -20,7 +20,6 @@
 
 #include <cub/detail/launcher/cuda_runtime.cuh>
 #include <cub/detail/type_traits.cuh> // for cub::detail::invoke_result_t
-#include <cub/device/dispatch/dispatch_advance_iterators.cuh>
 #include <cub/device/dispatch/kernels/reduce.cuh>
 #include <cub/device/dispatch/tuning/tuning_reduce.cuh>
 #include <cub/grid/grid_even_share.cuh>

cuda/cccl/headers/include/cub/device/dispatch/dispatch_segmented_sort.cuh

@@ -40,7 +40,6 @@
 #include <cub/detail/device_double_buffer.cuh>
 #include <cub/detail/temporary_storage.cuh>
 #include <cub/device/device_partition.cuh>
-#include <cub/device/dispatch/dispatch_advance_iterators.cuh>
 #include <cub/device/dispatch/kernels/segmented_sort.cuh>
 #include <cub/device/dispatch/tuning/tuning_segmented_sort.cuh>
 #include <cub/util_debug.cuh>
@@ -764,8 +763,8 @@ private:
       BeginOffsetIteratorT current_begin_offset = d_begin_offsets;
       EndOffsetIteratorT current_end_offset     = d_end_offsets;
 
-      detail::advance_iterators_inplace_if_supported(current_begin_offset, current_seg_offset);
-      detail::advance_iterators_inplace_if_supported(current_end_offset, current_seg_offset);
+      current_begin_offset += current_seg_offset;
+      current_end_offset += current_seg_offset;
 
       auto medium_indices_iterator =
         ::cuda::std::make_reverse_iterator(large_and_medium_segments_indices.get() + current_num_segments);

cuda/cccl/headers/include/cub/device/dispatch/kernels/reduce.cuh

@@ -47,9 +47,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace reduce
+namespace detail::reduce
 {
 
 /**
@@ -580,7 +578,6 @@ CUB_DETAIL_KERNEL_ATTRIBUTES __launch_bounds__(int(
   }
 }
 
-} // namespace reduce
-} // namespace detail
+} // namespace detail::reduce
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/kernels/scan.cuh

@@ -42,9 +42,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace scan
+namespace detail::scan
 {
 
 /******************************************************************************
@@ -186,7 +184,6 @@ __launch_bounds__(int(ChainedPolicyT::ActivePolicy::ScanPolicyT::BLOCK_THREADS))
   AgentScanT(temp_storage, d_in, d_out, scan_op, real_init_value).ConsumeRange(num_items, tile_state, start_tile);
 }
 
-} // namespace scan
-} // namespace detail
+} // namespace detail::scan
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/kernels/segmented_reduce.cuh

@@ -43,9 +43,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace reduce
+namespace detail::reduce
 {
 
 /// Normalize input iterator to segment offset
@@ -318,7 +316,6 @@ __launch_bounds__(int(ChainedPolicyT::ActivePolicy::ReducePolicy::BLOCK_THREADS)
   }
 }
 
-} // namespace reduce
-} // namespace detail
+} // namespace detail::reduce
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_adjacent_difference.cuh

@@ -43,9 +43,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace adjacent_difference
+namespace detail::adjacent_difference
 {
 template <typename InputIteratorT, bool MayAlias>
 struct policy_hub
@@ -64,7 +62,6 @@ struct policy_hub
 
   using MaxPolicy = Policy500;
 };
-} // namespace adjacent_difference
-} // namespace detail
+} // namespace detail::adjacent_difference
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_batch_memcpy.cuh

@@ -43,9 +43,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace batch_memcpy
+namespace detail::batch_memcpy
 {
 /**
  * Parameterizable tuning policy type for AgentBatchMemcpy
@@ -115,7 +113,6 @@ struct policy_hub
 
   using MaxPolicy = Policy700;
 };
-} // namespace batch_memcpy
-} // namespace detail
+} // namespace detail::batch_memcpy
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_for.cuh

@@ -42,9 +42,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace for_each
+namespace detail::for_each
 {
 
 struct policy_hub_t
@@ -57,7 +55,6 @@ struct policy_hub_t
   using MaxPolicy = policy_500_t;
 };
 
-} // namespace for_each
-} // namespace detail
+} // namespace detail::for_each
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_histogram.cuh

@@ -46,9 +46,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace histogram
+namespace detail::histogram
 {
 enum class primitive_sample
 {
@@ -272,7 +270,6 @@ struct policy_hub
 
   using MaxPolicy = Policy1000;
 };
-} // namespace histogram
-} // namespace detail
+} // namespace detail::histogram
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_merge.cuh

@@ -42,9 +42,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace merge
+namespace detail::merge
 {
 template <typename KeyT, typename ValueT>
 struct policy_hub
@@ -73,7 +71,6 @@ struct policy_hub
 
   using max_policy = policy600;
 };
-} // namespace merge
-} // namespace detail
+} // namespace detail::merge
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_merge_sort.cuh

@@ -62,6 +62,14 @@ struct MergeSortPolicyWrapper<StaticPolicyT, ::cuda::std::void_t<decltype(Static
   {}
 
   CUB_DEFINE_SUB_POLICY_GETTER(MergeSort);
+
+#if defined(CUB_ENABLE_POLICY_PTX_JSON)
+  _CCCL_DEVICE static constexpr auto EncodedPolicy()
+  {
+    using namespace ptx_json;
+    return object<key<"MergeSortPolicy">() = MergeSort().EncodedPolicy()>();
+  }
+#endif
 };
 
 template <typename PolicyT>

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_radix_sort.cuh

@@ -46,9 +46,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace radix
+namespace detail::radix
 {
 // sm90 default
 template <size_t KeySize, size_t ValueSize, size_t OffsetSize>
@@ -1062,7 +1060,6 @@ struct policy_hub
   using MaxPolicy = Policy1000;
 };
 
-} // namespace radix
-} // namespace detail
+} // namespace detail::radix
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_reduce_by_key.cuh

@@ -50,9 +50,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace reduce_by_key
+namespace detail::reduce_by_key
 {
 enum class primitive_key
 {
@@ -939,7 +937,6 @@ struct policy_hub
   };
   using MaxPolicy = Policy1000;
 };
-} // namespace reduce_by_key
-} // namespace detail
+} // namespace detail::reduce_by_key
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_run_length_encode.cuh

@@ -52,9 +52,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace rle
+namespace detail::rle
 {
 enum class primitive_key
 {
@@ -670,7 +668,6 @@ struct policy_hub
   using MaxPolicy = Policy1000;
 };
 } // namespace non_trivial_runs
-} // namespace rle
-} // namespace detail
+} // namespace detail::rle
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_scan.cuh

@@ -53,9 +53,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace scan
+namespace detail::scan
 {
 enum class keep_rejects
 {
@@ -615,7 +613,6 @@ struct policy_hub
 
   using MaxPolicy = Policy1000;
 };
-} // namespace scan
-} // namespace detail
+} // namespace detail::scan
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_scan_by_key.cuh

@@ -49,9 +49,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace scan_by_key
+namespace detail::scan_by_key
 {
 enum class primitive_accum
 {
@@ -1007,7 +1005,6 @@ struct policy_hub
 
   using MaxPolicy = Policy1000;
 };
-} // namespace scan_by_key
-} // namespace detail
+} // namespace detail::scan_by_key
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_segmented_sort.cuh

@@ -43,9 +43,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace segmented_sort
+namespace detail::segmented_sort
 {
 
 template <typename PolicyT, typename = void>
@@ -395,7 +393,6 @@ struct policy_hub
 
   using MaxPolicy = Policy860;
 };
-} // namespace segmented_sort
-} // namespace detail
+} // namespace detail::segmented_sort
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_three_way_partition.cuh

@@ -47,9 +47,7 @@
 
 CUB_NAMESPACE_BEGIN
 
-namespace detail
-{
-namespace three_way_partition
+namespace detail::three_way_partition
 {
 
 template <typename PolicyT, typename = void>
@@ -437,7 +435,6 @@ struct policy_hub
 
   using MaxPolicy = Policy1000;
 };
-} // namespace three_way_partition
-} // namespace detail
+} // namespace detail::three_way_partition
 
 CUB_NAMESPACE_END

cuda/cccl/headers/include/cub/device/dispatch/tuning/tuning_unique_by_key.cuh

@@ -788,6 +788,16 @@ struct UniqueByKeyPolicyWrapper<StaticPolicyT,
   {
     return cub::detail::MakePolicyWrapper(typename StaticPolicyT::UniqueByKeyPolicyT());
   }
+
+#if defined(CUB_ENABLE_POLICY_PTX_JSON)
+  _CCCL_DEVICE static constexpr auto EncodedPolicy()
+  {
+    using namespace ptx_json;
+    return object<key<"UniqueByKeyPolicyT">() = UniqueByKey().EncodedPolicy(),
+                  key<"DelayConstructor">() =
+                    StaticPolicyT::UniqueByKeyPolicyT::detail::delay_constructor_t::EncodedConstructor()>();
+  }
+#endif
 };
 
 template <typename PolicyT>

cuda/cccl/headers/include/cub/warp/specializations/warp_reduce_shfl.cuh

@@ -51,6 +51,7 @@
 #include <cuda/__functional/maximum.h>
 #include <cuda/__functional/minimum.h>
 #include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__bit/countr.h>
 #include <cuda/std/__functional/operations.h>
 #include <cuda/std/__type_traits/enable_if.h>
 #include <cuda/std/__type_traits/integral_constant.h>
@@ -701,7 +702,7 @@ struct WarpReduceShfl
   _CCCL_DEVICE _CCCL_FORCEINLINE T SegmentedReduce(T input, FlagT flag, ReductionOp reduction_op)
   {
     // Get the start flags for each thread in the warp.
-    int warp_flags = __ballot_sync(member_mask, flag);
+    unsigned warp_flags = __ballot_sync(member_mask, flag);
 
     // Convert to tail-segmented
     if (HEAD_SEGMENTED)
@@ -722,7 +723,7 @@ struct WarpReduceShfl
     warp_flags |= 1u << (LOGICAL_WARP_THREADS - 1);
 
     // Find the next set flag
-    int last_lane = __clz(__brev(warp_flags));
+    int last_lane = ::cuda::std::countr_zero(warp_flags);
 
     T output = input;
     // Template-iterate reduction steps

cuda/cccl/headers/include/cub/warp/specializations/warp_reduce_smem.cuh

@@ -49,6 +49,7 @@
 #include <cub/util_type.cuh>
 
 #include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__bit/countr.h>
 #include <cuda/std/__type_traits/integral_constant.h>
 
 CUB_NAMESPACE_BEGIN
@@ -215,7 +216,7 @@ struct WarpReduceSmem
   SegmentedReduce(T input, FlagT flag, ReductionOp reduction_op, ::cuda::std::true_type /*has_ballot*/)
   {
     // Get the start flags for each thread in the warp.
-    int warp_flags = __ballot_sync(member_mask, flag);
+    unsigned warp_flags = __ballot_sync(member_mask, flag);
 
     if (!HEAD_SEGMENTED)
     {
@@ -232,7 +233,7 @@ struct WarpReduceSmem
     }
 
     // Find next flag
-    int next_flag = __clz(__brev(warp_flags));
+    int next_flag = ::cuda::std::countr_zero(warp_flags);
 
     // Clip the next segment at the warp boundary if necessary
     if (LOGICAL_WARP_THREADS != 32)

cuda/cccl/headers/include/cub/warp/specializations/warp_scan_shfl.cuh

@@ -50,8 +50,8 @@
 
 #include <cuda/__ptx/instructions/get_sreg.h>
 #include <cuda/std/__algorithm/clamp.h>
-#include <cuda/std/__algorithm/max.h>
 #include <cuda/std/__bit/has_single_bit.h>
+#include <cuda/std/__bit/integral.h>
 #include <cuda/std/__functional/operations.h>
 #include <cuda/std/__type_traits/integral_constant.h>
 #include <cuda/std/__type_traits/is_integral.h>
@@ -630,7 +630,7 @@ struct WarpScanShfl
     ballot = ballot & ::cuda::ptx::get_sreg_lanemask_le();
 
     // Find index of first set bit
-    int segment_first_lane = ::cuda::std::max(0, 31 - __clz(ballot));
+    int segment_first_lane = ::cuda::std::__bit_log2(ballot);
 
     // Iterate scan steps
     _CCCL_PRAGMA_UNROLL_FULL()

cuda/cccl/headers/include/cuda/__algorithm/common.h

@@ -11,7 +11,7 @@
 #ifndef __CUDA___ALGORITHM_COMMON
 #define __CUDA___ALGORITHM_COMMON
 
-#include <cuda/__cccl_config>
+#include <cuda/std/detail/__config>
 
 #if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
 #  pragma GCC system_header

cuda/cccl/headers/include/cuda/__algorithm/copy.h

@@ -11,7 +11,7 @@
 #ifndef __CUDA___ALGORITHM_COPY_H
 #define __CUDA___ALGORITHM_COPY_H
 
-#include <cuda/__cccl_config>
+#include <cuda/std/detail/__config>
 
 #if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
 #  pragma GCC system_header

cuda/cccl/headers/include/cuda/__algorithm/fill.h

@@ -11,7 +11,7 @@
 #ifndef __CUDA___ALGORITHM_FILL
 #define __CUDA___ALGORITHM_FILL
 
-#include <cuda/__cccl_config>
+#include <cuda/std/detail/__config>
 
 #if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
 #  pragma GCC system_header