Skip to content

Eliminate AIV_HUB init task in paged_attention_unroll orchestration #318

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

Closed
chenshengxin2026 wants to merge 1 commit into from
Closed

Eliminate AIV_HUB init task in paged_attention_unroll orchestration #318

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
21 changes: 9 additions & 12 deletions ...rmap_and_ringbuffer/paged_attention_unroll/kernels/orchestration/paged_attention_orch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -170,15 +170,6 @@ __attribute__((visibility("default"))) void aicpu_orchestration_entry(PTO2Runtim
prof_view_count += 2;
CYCLE_COUNT_LAP(prof_tensor_view);

PTOParam params_inplace;
params_inplace.add_output(oi);
params_inplace.add_output(li_update);
params_inplace.add_output(mi_update);
CYCLE_COUNT_LAP(prof_param_setup);
pto2_rt_submit_aiv_task(rt, FUNC_AIV_HUB, params_inplace);
prof_submit_count++;
CYCLE_COUNT_LAP(prof_submit_task);

// Reusable PTOParam objects — reset() before each use avoids
// repeated stack-frame construction in the inner loop.
// params_qk must persist until params_pv.copy_scalars_from().
Expand Down Expand Up @@ -256,9 +247,15 @@ __attribute__((visibility("default"))) void aicpu_orchestration_entry(PTO2Runtim
params_up.add_input(mi);
params_up.add_input(li);
params_up.add_input(oi_new);
params_up.add_inout(mi_update);
params_up.add_inout(li_update);
params_up.add_inout(oi);
if (is_first) {
params_up.add_output(mi_update);
params_up.add_output(li_update);
params_up.add_output(oi);
} else {
params_up.add_inout(mi_update);
params_up.add_inout(li_update);
params_up.add_inout(oi);
}
params_up.add_output(out_view);
params_up.add_scalar(is_first);
params_up.add_scalar(is_last);
Expand Down
Loading