Skip to content

Encourage vectorization in Hll8Array::mergeHll #553

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

Merged
proost merged 2 commits into from
Oct 30, 2026
Merged

Encourage vectorization in Hll8Array::mergeHll #553

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
57cdeb6
Refactor `Hll8Array::mergeHll` to encourage autovectorization
hadrian-reppas Oct 27, 2026
872533c
review feedback
hadrian-reppas Oct 29, 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
43 changes: 21 additions & 22 deletions hll/include/Hll8Array-internal.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,30 +120,29 @@ void Hll8Array<A>::mergeHll(const HllArray<A>& src) {
// we can optimize further when the k values are equal
if (this->getLgConfigK() == src.getLgConfigK()) {
if (src.getTgtHllType() == target_hll_type::HLL_8) {
uint32_t i = 0;
for (const auto value: src.getHllArray()) {
this->hllByteArr_[i] = std::max(this->hllByteArr_[i], value);
++i;
// Hoist the data pointer out of the loop to encourage vectorization
const uint8_t* src_data = src.getHllArray().data();
uint8_t* data = this->hllByteArr_.data();
Comment thread
hadrian-reppas marked this conversation as resolved.
const size_t n = src.getHllArray().size();
for (size_t i = 0; i < n; ++i) {
data[i] = std::max(data[i], src_data[i]);
}
} else if (src.getTgtHllType() == target_hll_type::HLL_6) {
const uint32_t src_k = 1 << src.getLgConfigK();
uint32_t i = 0;
const uint8_t* ptr = src.getHllArray().data();
while (i < src_k) {
uint8_t value = *ptr & 0x3f;
this->hllByteArr_[i] = std::max(this->hllByteArr_[i], value);
++i;
value = *ptr++ >> 6;
value |= (*ptr & 0x0f) << 2;
this->hllByteArr_[i] = std::max(this->hllByteArr_[i], value);
++i;
value = *ptr++ >> 4;
value |= (*ptr & 3) << 4;
this->hllByteArr_[i] = std::max(this->hllByteArr_[i], value);
++i;
value = *ptr++ >> 2;
this->hllByteArr_[i] = std::max(this->hllByteArr_[i], value);
++i;
const uint8_t* src_data = src.getHllArray().data();
uint8_t* data = this->hllByteArr_.data();
const size_t src_k = 1 << src.getLgConfigK();
const size_t num_groups = src_k / 4;
for (size_t g = 0; g < num_groups; ++g) {
const size_t i = 4 * g;
const size_t j = 3 * g;
// Read into locals so that stores don't force reloads
const uint8_t b0 = src_data[j];
const uint8_t b1 = src_data[j + 1];
const uint8_t b2 = src_data[j + 2];
data[i] = std::max<uint8_t>(data[i], b0 & 0x3f);
data[i + 1] = std::max<uint8_t>(data[i + 1], (b0 >> 6) | ((b1 & 0x0f) << 2));
data[i + 2] = std::max<uint8_t>(data[i + 2], (b1 >> 4) | ((b2 & 3) << 4));
data[i + 3] = std::max<uint8_t>(data[i + 3], b2 >> 2);
}
} else { // HLL_4
const auto& src4 = static_cast<const Hll4Array<A>&>(src);
Expand Down
Loading