Skip to content

Use static_for for fixed-size loops #771

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
wants to merge 3 commits into
base: dev
Choose a base branch
from
Open

Use static_for for fixed-size loops #771

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fabc460
Use static_for for fixed-size loops
PointKernel Oct 23, 2025
634c671
Use ftors instead of lambda to fix 12.0 warnings
PointKernel Oct 24, 2025
17ec7ec
CMake cleanups
PointKernel Oct 24, 2025
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
73 changes: 40 additions & 33 deletions include/cuco/detail/bloom_filter/bloom_filter_impl.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include <cuda/std/tuple>
#include <cuda/std/type_traits>
#include <cuda/stream_ref>
#include <cuda/utility>
#include <thrust/iterator/constant_iterator.h>

#include <cooperative_groups.h>
Expand Down Expand Up @@ -142,15 +143,14 @@ class bloom_filter_impl {
template <class HashValue, class BlockIndex>
__device__ void add_impl(HashValue const& hash_value, BlockIndex block_index)
{
#pragma unroll words_per_block
for (uint32_t i = 0; i < words_per_block; ++i) {
auto const word = policy_.word_pattern(hash_value, i);
cuda::static_for<words_per_block>([&](auto i) {
auto const word = policy_.word_pattern(hash_value, i());
if (word != 0) {
auto atom_word = cuda::atomic_ref<word_type, thread_scope>{
*(words_ + (block_index * words_per_block + i))};
*(words_ + (block_index * words_per_block + i()))};
atom_word.fetch_or(word, cuda::memory_order_relaxed);
}
}
});
}

template <class CG, class ProbeKey>
Expand Down Expand Up @@ -205,9 +205,11 @@ class bloom_filter_impl {
block_index = policy_.block_index(hash_value, num_blocks_);
}

for (uint32_t j = 0; (j < num_threads) and (i + j < num_keys); ++j) {
this->add_impl(group, group.shfl(hash_value, j), group.shfl(block_index, j));
}
cuda::static_for<num_threads>([&](auto j) {
if ((j() < num_threads) and (i + j() < num_keys)) {
this->add_impl(group, group.shfl(hash_value, j()), group.shfl(block_index, j()));
}
});
}
} else { // subdivide given CG into multiple optimal CGs
typename policy_type::hash_result_type hash_value;
Expand All @@ -225,10 +227,13 @@ class bloom_filter_impl {
block_index = policy_.block_index(hash_value, num_blocks_);
}

for (uint32_t j = 0; (j < worker_num_threads) and (i + worker_offset + j < num_keys); ++j) {
this->add_impl(
worker_group, worker_group.shfl(hash_value, j), worker_group.shfl(block_index, j));
}
cuda::static_for<worker_num_threads>([&](auto j) {
if ((j() < worker_num_threads) and (i + worker_offset + j() < num_keys)) {
this->add_impl(worker_group,
worker_group.shfl(hash_value, j()),
worker_group.shfl(block_index, j()));
}
});
}
}
}
Expand All @@ -245,12 +250,13 @@ class bloom_filter_impl {
*(words_ + (block_index * words_per_block + rank))};
atom_word.fetch_or(policy_.word_pattern(hash_value, rank), cuda::memory_order_relaxed);
} else {
#pragma unroll
for (auto i = rank; i < words_per_block; i += num_threads) {
auto atom_word = cuda::atomic_ref<word_type, thread_scope>{
*(words_ + (block_index * words_per_block + i))};
atom_word.fetch_or(policy_.word_pattern(hash_value, i), cuda::memory_order_relaxed);
}
cuda::static_for<words_per_block>([&](auto i) {
if (i() >= rank && (i() - rank) % num_threads == 0) {
auto atom_word = cuda::atomic_ref<word_type, thread_scope>{
*(words_ + (block_index * words_per_block + i()))};
atom_word.fetch_or(policy_.word_pattern(hash_value, i()), cuda::memory_order_relaxed);
}
});
}
}

Expand Down Expand Up @@ -330,11 +336,12 @@ class bloom_filter_impl {
auto const stored_pattern = this->vec_load_words<words_per_block>(
policy_.block_index(hash_value, num_blocks_) * words_per_block);

#pragma unroll words_per_block
for (uint32_t i = 0; i < words_per_block; ++i) {
auto const expected_pattern = policy_.word_pattern(hash_value, i);
if ((stored_pattern[i] & expected_pattern) != expected_pattern) { return false; }
}
bool result = true;
cuda::static_for<words_per_block>([&](auto i) {
auto const expected_pattern = policy_.word_pattern(hash_value, i());
if ((stored_pattern[i()] & expected_pattern) != expected_pattern) { result = false; }
});
if (!result) { return false; }

return true;
}
Expand All @@ -354,17 +361,17 @@ class bloom_filter_impl {
auto const hash_value = policy_.hash(key);
bool success = true;

#pragma unroll
for (uint32_t i = rank; i < optimal_num_threads; i += num_threads) {
auto const thread_offset = i * words_per_thread;
auto const stored_pattern = this->vec_load_words<words_per_thread>(
policy_.block_index(hash_value, num_blocks_) * words_per_block + thread_offset);
#pragma unroll words_per_thread
for (uint32_t j = 0; j < words_per_thread; ++j) {
auto const expected_pattern = policy_.word_pattern(hash_value, thread_offset + j);
if ((stored_pattern[j] & expected_pattern) != expected_pattern) { success = false; }
cuda::static_for<optimal_num_threads>([&](auto i) {
if (i() >= rank && (i() - rank) % num_threads == 0) {
auto const thread_offset = i() * words_per_thread;
auto const stored_pattern = this->vec_load_words<words_per_thread>(
policy_.block_index(hash_value, num_blocks_) * words_per_block + thread_offset);
cuda::static_for<words_per_thread>([&](auto j) {
auto const expected_pattern = policy_.word_pattern(hash_value, thread_offset + j());
if ((stored_pattern[j()] & expected_pattern) != expected_pattern) { success = false; }
});
}
}
});

return group.all(success);
}
Expand Down
13 changes: 7 additions & 6 deletions include/cuco/detail/hyperloglog/kernels.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@

#include <cuda/std/array>
#include <cuda/std/span>
#include <cuda/utility>

#include <cooperative_groups.h>

Expand Down Expand Up @@ -71,16 +72,16 @@ CUCO_KERNEL void add_shmem_vectorized(typename RefType::value_type const* first,
#if defined(CUCO_HAS_CG_INVOKE_ONE)
cooperative_groups::invoke_one(grid, [&]() {
auto const remainder = n % VectorSize;
for (int i = 0; i < remainder; ++i) {
local_ref.add(*(first + n - i - 1));
}
cuda::static_for<VectorSize>([&](auto i) {
if (i() < remainder) { local_ref.add(*(first + n - i() - 1)); }
});
});
#else
if (grid.thread_rank() == 0) {
auto const remainder = n % VectorSize;
for (int i = 0; i < remainder; ++i) {
local_ref.add(*(first + n - i - 1));
}
cuda::static_for<VectorSize>([&](auto i) {
if (i() < remainder) { local_ref.add(*(first + n - i() - 1)); }
});
}
#endif
block.sync();
Expand Down
Loading
Loading