Skip to content

Fix env-based DeviceReduce kernel-allowlist tests on MSVC + CUDA 13.3 #9726

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
edenfunf wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix env-based DeviceReduce kernel-allowlist tests on MSVC + CUDA 13.3 #9726

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7a2efcc
Fix CUB reduce env kernel-allowlist tests on MSVC + CUDA 13.3
edenfunf Jul 7, 2026
eb895af
Address review: dedup allowlist construction, add missing include
edenfunf Jul 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
149 changes: 91 additions & 58 deletions cub/cub/device/dispatch/dispatch_reduce_deterministic.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,8 +24,11 @@
#include <cub/device/dispatch/tuning/tuning_reduce.cuh>
#include <cub/util_debug.cuh>
#include <cub/util_device.cuh>
#include <cub/util_macro.cuh>
#include <cub/util_temporary_storage.cuh>

#include <thrust/type_traits/unwrap_contiguous_iterator.h>

#include <cuda/__cmath/ceil_div.h>
#include <cuda/__type_traits/is_floating_point.h>
#include <cuda/std/__algorithm/min.h>
Expand All @@ -35,6 +38,7 @@
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/decay.h>
#include <cuda/std/__type_traits/enable_if.h>
#include <cuda/std/__type_traits/is_empty.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

Expand Down Expand Up @@ -84,13 +88,72 @@ struct deterministic_sum_t
template <typename PolicySelector,
typename InputIteratorT,
typename OutputIteratorT,
typename OffsetT,
typename ReductionOpT,
typename InitValueT,
typename DeterministicAccumT,
typename TransformOpT>
struct DeterministicDeviceReduceKernelSource
{
// PolicySelector must be stateless, so we can pass the type to the kernel
static_assert(::cuda::std::is_empty_v<PolicySelector>);

CUB_DEFINE_KERNEL_GETTER(
SingleTileKernel,
reduce::DeterministicDeviceReduceSingleTileKernel<
PolicySelector,
InputIteratorT,
OutputIteratorT,
ReductionOpT,
InitValueT,
DeterministicAccumT,
TransformOpT>)

CUB_DEFINE_KERNEL_GETTER(
ReductionKernel,
reduce::
DeterministicDeviceReduceKernel<PolicySelector, InputIteratorT, ReductionOpT, DeterministicAccumT, TransformOpT>)

// The second single-tile pass reduces the per-block partials and always uses an identity transform
CUB_DEFINE_KERNEL_GETTER(
SingleTileSecondKernel,
reduce::DeterministicDeviceReduceSingleTileKernel<
PolicySelector,
DeterministicAccumT*,
OutputIteratorT,
ReductionOpT,
InitValueT,
DeterministicAccumT>)
};

//! Kernel source instantiation matching the default used by `rfa::dispatch` below. Tests use this alias to
//! reference the exact kernel instantiations the dispatch launches.
template <typename InputIteratorT,
typename OutputIteratorT,
typename OffsetT,
typename InitValueT,
typename TransformOpT = ::cuda::std::identity,
typename AccumT = accum_t<InitValueT, InputIteratorT, TransformOpT>,
typename PolicySelector =
policy_selector_from_types<AccumT, OffsetT, deterministic_sum_t<AccumT>, __determinism_t::__gpu_to_gpu>>
using default_kernel_source_t = DeterministicDeviceReduceKernelSource<
PolicySelector,
THRUST_NS_QUALIFIER::try_unwrap_contiguous_iterator_t<InputIteratorT>,
OutputIteratorT,
deterministic_sum_t<AccumT>,
InitValueT,
typename deterministic_sum_t<AccumT>::DeterministicAcc,
TransformOpT>;

template <typename SingleTileKernelT,
typename InputIteratorT,
typename OutputIteratorT,
typename OffsetT,
typename ReductionOpT,
typename InitValueT,
typename TransformOpT,
typename KernelLauncherFactory>
CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invoke_single_tile(
SingleTileKernelT single_tile_kernel,
void* d_temp_storage,
size_t& temp_storage_bytes,
InputIteratorT d_in,
Expand Down Expand Up @@ -122,20 +185,7 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invok
// Invoke single_reduce_sweep_kernel
if (const auto error = CubDebug(
launcher_factory(1, active_policy.single_tile.threads_per_block, 0, stream)
.doit(detail::reduce::DeterministicDeviceReduceSingleTileKernel<
PolicySelector,
InputIteratorT,
OutputIteratorT,
ReductionOpT,
InitValueT,
DeterministicAccumT,
TransformOpT>,
d_in,
d_out,
static_cast<int>(num_items),
reduction_op,
init,
transform_op)))
.doit(single_tile_kernel, d_in, d_out, static_cast<int>(num_items), reduction_op, init, transform_op)))
{
return error;
}
Expand All @@ -150,16 +200,19 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invok
return CubDebug(detail::DebugSyncStream(stream));
}

template <typename PolicySelector,
template <typename DeterministicAccumT,
typename ReduceKernelT,
typename SingleTileKernelT,
typename InputIteratorT,
typename OutputIteratorT,
typename OffsetT,
typename ReductionOpT,
typename InitValueT,
typename DeterministicAccumT,
typename TransformOpT,
typename KernelLauncherFactory>
CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invoke_passes(
ReduceKernelT reduce_kernel,
SingleTileKernelT single_tile_second_kernel,
void* d_temp_storage,
size_t& temp_storage_bytes,
InputIteratorT d_in,
Expand All @@ -179,10 +232,7 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invok
}

KernelConfig reduce_config;
if (const auto error = CubDebug(reduce_config.__init(
detail::reduce::
DeterministicDeviceReduceKernel<PolicySelector, InputIteratorT, ReductionOpT, DeterministicAccumT, TransformOpT>,
active_policy.multi_tile)))
if (const auto error = CubDebug(reduce_config.__init(reduce_kernel, active_policy.multi_tile)))
{
return error;
}
Expand Down Expand Up @@ -249,11 +299,7 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invok

if (const auto error = CubDebug(
launcher_factory(current_grid_size, active_policy.multi_tile.threads_per_block, 0, stream)
.doit(detail::reduce::DeterministicDeviceReduceKernel<PolicySelector,
InputIteratorT,
ReductionOpT,
DeterministicAccumT,
TransformOpT>,
.doit(reduce_kernel,
d_in,
d_chunk_block_reductions,
num_current_items,
Expand Down Expand Up @@ -295,13 +341,7 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invok
// Invoke DeterministicDeviceReduceSingleTileKernel
if (const auto error = CubDebug(
launcher_factory(1, active_policy.single_tile.threads_per_block, 0, stream)
.doit(detail::reduce::DeterministicDeviceReduceSingleTileKernel<
PolicySelector,
DeterministicAccumT*,
OutputIteratorT,
ReductionOpT,
InitValueT,
DeterministicAccumT>,
.doit(single_tile_second_kernel,
d_block_reductions,
d_out,
reduce_grid_size,
Expand All @@ -322,15 +362,18 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE cudaError_t invok
return CubDebug(detail::DebugSyncStream(stream));
}

template <typename InputIteratorT,
typename OutputIteratorT,
typename OffsetT,
typename InitValueT,
typename TransformOpT = ::cuda::std::identity,
typename AccumT = accum_t<InitValueT, InputIteratorT, TransformOpT>,
typename PolicySelector =
policy_selector_from_types<AccumT, OffsetT, deterministic_sum_t<AccumT>, __determinism_t::__gpu_to_gpu>,
typename KernelLauncherFactory = CUB_DETAIL_DEFAULT_KERNEL_LAUNCHER_FACTORY>
template <
typename InputIteratorT,
typename OutputIteratorT,
typename OffsetT,
typename InitValueT,
typename TransformOpT = ::cuda::std::identity,
typename AccumT = accum_t<InitValueT, InputIteratorT, TransformOpT>,
typename PolicySelector =
policy_selector_from_types<AccumT, OffsetT, deterministic_sum_t<AccumT>, __determinism_t::__gpu_to_gpu>,
typename KernelSource =
default_kernel_source_t<InputIteratorT, OutputIteratorT, OffsetT, InitValueT, TransformOpT, AccumT, PolicySelector>,
typename KernelLauncherFactory = CUB_DETAIL_DEFAULT_KERNEL_LAUNCHER_FACTORY>
CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
void* d_temp_storage,
size_t& temp_storage_bytes,
Expand All @@ -341,6 +384,7 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
cudaStream_t stream = {},
TransformOpT transform_op = {},
PolicySelector policy_selector = {},
KernelSource kernel_source = {},
KernelLauncherFactory launcher_factory = {})
{
// Get CC
Expand Down Expand Up @@ -373,14 +417,8 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(

if (num_items <= tile_items)
{
return invoke_single_tile<PolicySelector,
input_unwrapped_it_t,
OutputIteratorT,
OffsetT,
deterministic_add_t,
InitValueT,
typename deterministic_add_t::DeterministicAcc,
TransformOpT>(
return invoke_single_tile(
kernel_source.SingleTileKernel(),
d_temp_storage,
temp_storage_bytes,
d_in_unwrapped,
Expand All @@ -394,14 +432,9 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
launcher_factory);
}

return invoke_passes<PolicySelector,
input_unwrapped_it_t,
OutputIteratorT,
OffsetT,
deterministic_add_t,
InitValueT,
typename deterministic_add_t::DeterministicAcc,
TransformOpT>(
return invoke_passes<typename deterministic_add_t::DeterministicAcc>(
kernel_source.ReductionKernel(),
kernel_source.SingleTileSecondKernel(),
d_temp_storage,
temp_storage_bytes,
d_in_unwrapped,
Expand Down
Loading
Loading