Skip to content
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

[GPU] update shape for fused prims #25363

Merged
merged 2 commits into from
Jul 4, 2024
+34 −11
Merged

[GPU] update shape for fused prims #25363

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a9f76d8
[GPU] update shape for fused prims
ahnyoung-paul Jul 3, 2024
4e8f81e
Support to check multiple fused inputs
ahnyoung-paul Jul 4, 2024
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
40 changes: 29 additions & 11 deletions src/plugins/intel_gpu/src/graph/primitive_inst.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -586,19 +586,37 @@ event::ptr primitive_inst::realloc_if_needed() {
user_insts.size(), " and ", user_insts_origin.size());
}
for (auto user : user_insts) {
auto is_fused_prim_of_user = [&](primitive_id id) -> bool {
for (auto& p : user->get_node().get_fused_primitives()) {
if (p.has_outer_dep()) {
if (user->get_node().get_dependency(p.outer_dep_start_idx).id() == id) {
Copy link
Contributor

Choose a reason for hiding this comment

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

What if there is multiple fused inputs?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Fixed it.

return true;
}
}
}
return false;
};
// Since fake alignment is applicable for input tensor as well, make sure we allocate enough memory
// to prevent reading beyond the allocated memory bounds
if (user->get_node().is_type<fully_connected>() && user->is_dynamic() && user->_deps[0].first == this) {
GPU_DEBUG_TRACE_DETAIL << "Check fc user " << user->id() << "'s fake alignment-ed input size" << std::endl;
user->update_shape();
user->update_shape_done_by_other = true;

auto fc_impl_params = *user->_impl_params;
auto fc_input_layout = user->get_node().type()->get_fake_aligned_params(fc_impl_params).input_layouts[0];
if (fc_input_layout.bytes_count() > updated_layout.bytes_count()) {
GPU_DEBUG_TRACE_DETAIL << id() << ": increase output layout allocation size from " << actual_layout.to_short_string() << " -> "
<< fc_input_layout.to_short_string() << " to meet the input buffer alignment requirements for FC\n";
updated_layout = fc_input_layout;
if (user->get_node().is_type<fully_connected>() && user->is_dynamic()) {
if (user->_deps[0].first == this) {
GPU_DEBUG_TRACE_DETAIL << "Check fc user " << user->id() << "'s fake alignment-ed input size" << std::endl;
user->update_shape();
user->update_shape_done_by_other = true;

auto fc_impl_params = *user->_impl_params;
auto fc_input_layout = user->get_node().type()->get_fake_aligned_params(fc_impl_params).input_layouts[0];
if (fc_input_layout.bytes_count() > updated_layout.bytes_count()) {
GPU_DEBUG_TRACE_DETAIL << id() << ": increase output layout allocation size from " << actual_layout.to_short_string() << " -> "
<< fc_input_layout.to_short_string() << " to meet the input buffer alignment requirements for FC\n";
updated_layout = fc_input_layout;
}
} else if (is_fused_prim_of_user(id()) && user->update_shape_done_by_other) {
// Since the output layout of fused prim in user is determined after user's update_shape
// Rerun update_shape w/ new output layout of fused prim
user->update_shape_done_by_other = false;
user->update_shape();
user->update_shape_done_by_other = true;
}
}
}
Expand Down
Loading