Skip to content

[NPUW] Refactor LazyTensor's Constant hashing #32113

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 2 commits into
base: master
Choose a base branch
from
+6 −7
Open

[NPUW] Refactor LazyTensor's Constant hashing #32113

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
37ef280
Refactor Constant hashing
smirnov-alexey Sep 17, 2025
0b56b72
Merge branch 'master' into as/npuw_lt_const_hash
smirnov-alexey Sep 17, 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
12 changes: 6 additions & 6 deletions src/plugins/intel_npu/src/plugin/npuw/lazy_tensor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,18 +26,18 @@ namespace op {
Const::Const(const std::shared_ptr<ov::op::v0::Constant>& n) : m_node(n) {
m_cached_type = m_node->get_element_type();
m_cached_shape = m_node->get_shape();
m_cached_ptr = m_node->get_data_ptr();
m_byte_size = m_node->get_byte_size();

auto rt_info = m_node->get_rt_info();
auto weightless_cache_attr = rt_info.find(ov::WeightlessCacheAttribute::get_type_info_static());
if (weightless_cache_attr != rt_info.end()) {
m_offset = weightless_cache_attr->second.as<ov::WeightlessCacheAttribute>().bin_offset;
}
NPUW_ASSERT(weightless_cache_attr != rt_info.end() &&
"Constant node doesn't have WeightlessCacheAttribute! Likely a new node introduced by some pass.");
Comment on lines +33 to +34
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And why it is an issue? This is a perfectly valid situation.

m_offset = weightless_cache_attr->second.as<ov::WeightlessCacheAttribute>().bin_offset;
}

std::size_t Const::hash() const {
std::size_t seed = std::hash<const void*>()(m_cached_ptr) + 0x9e3779b9;
std::size_t seed = std::hash<std::size_t>()(m_byte_size) + 0x9e3779b9;
seed ^= std::hash<std::size_t>()(m_offset) + 0x9e3779b9;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

do we need this magic 0x9e3779b9 even in case of offset?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Keeping this cargo cult for consistency. Will rework some day

seed ^= m_cached_type.hash() + 0x9e3779b9;
for (const auto& dim : m_cached_shape) {
seed ^= std::hash<std::size_t>()(dim) + 0x9e3779b9;
Expand All @@ -47,7 +47,7 @@ std::size_t Const::hash() const {

bool Const::operator==(const Const& other) const {
return (m_cached_type == other.m_cached_type && m_cached_shape == other.m_cached_shape &&
m_cached_ptr == other.m_cached_ptr);
m_byte_size == other.m_byte_size && m_offset == other.m_offset);
}

ov::Tensor Const::eval() const {
Expand Down
1 change: 0 additions & 1 deletion src/plugins/intel_npu/src/plugin/npuw/lazy_tensor.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,7 +103,6 @@ class Const {
std::shared_ptr<ov::op::v0::Constant> m_node = nullptr;
ov::element::Type m_cached_type;
ov::Shape m_cached_shape;
const void* m_cached_ptr = nullptr;
std::size_t m_offset = 0;
std::size_t m_byte_size = 0;
ov::Tensor m_read_from_bin;
Expand Down
Loading