Skip to content

Replace deprecated nd_item::barrier() with sycl::group_barrier() in accumulators and gemm kernels #3006

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
antonwolfy merged 3 commits into from
Jul 23, 2026
Merged

Replace deprecated nd_item::barrier() with sycl::group_barrier() in accumulators and gemm kernels #3006

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e48b5cf
Get rid of deprecated nd_item::barrier() member
antonwolfy Jul 22, 2026
65638b7
Update changelog
antonwolfy Jul 22, 2026
4c98f24
Merge branch 'master' into address-deprecation-warning
antonwolfy Jul 23, 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
1 change: 1 addition & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,7 @@ This release is compatible with NumPy 2.5.
* Replaced references to the `dpnp.amax`/`dpnp.amin` aliases with the canonical `dpnp.max`/`dpnp.min` in docstrings and code internally [#2990](https://github.com/IntelPython/dpnp/pull/2990)
* Aligned the signature of `dpnp.tensor.expand_dims` with the Python array API by making `axis` a required argument [#2988](https://github.com/IntelPython/dpnp/pull/2988)
* Removed dead code branches guarded by outdated oneMKL and DPC++ compiler version checks [#2999](https://github.com/IntelPython/dpnp/pull/2999)
* Replaced the deprecated `nd_item::barrier()` member calls in the `accumulators` and `gemm` kernels with the SYCL 2020 `sycl::group_barrier()` free function [#3006](https://github.com/IntelPython/dpnp/pull/3006)

### Deprecated

Expand Down
21 changes: 13 additions & 8 deletions dpnp/tensor/libtensor/include/kernels/accumulators.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -302,11 +302,12 @@ sycl::event inclusive_scan_base_step_blocked(
local_iscan.back(), identity, scan_op);
// ensure all finished reading from SLM, to avoid race condition
// with subsequent writes into SLM
it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(),
sycl::memory_scope::work_group);
}

slm_iscan_tmp[(lid + 1) % wg_size] = wg_iscan_val;
it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);
const outputT modifier = (lid == 0) ? identity : slm_iscan_tmp[lid];

#pragma unroll
Expand Down Expand Up @@ -430,7 +431,8 @@ sycl::event inclusive_scan_base_step_striped(
slm_iscan_tmp[local_offset0 + i] = local_iscan[i];
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(),
sycl::memory_scope::work_group);
}

{
Expand All @@ -452,7 +454,8 @@ sycl::event inclusive_scan_base_step_striped(
slm_iscan_tmp[block_offset + disp_exchanged];
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(),
sycl::memory_scope::work_group);
}
}

Expand All @@ -476,19 +479,20 @@ sycl::event inclusive_scan_base_step_striped(
identity, scan_op);
// ensure all finished reading from SLM, to avoid race condition
// with subsequent writes into SLM
it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(),
sycl::memory_scope::work_group);
}

slm_iscan_tmp[(lid + 1) % wg_size] = wg_iscan_val;
it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);
const outputT modifier = (lid == 0) ? identity : slm_iscan_tmp[lid];

#pragma unroll
for (nwiT m_wi = 0; m_wi < n_wi; ++m_wi) {
local_iscan[m_wi] = scan_op(local_iscan[m_wi], modifier);
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

// convert back to blocked layout
{
Expand All @@ -499,7 +503,8 @@ sycl::event inclusive_scan_base_step_striped(
slm_iscan_tmp[local_offset0 + m_wi] = local_iscan[m_wi];
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(),
sycl::memory_scope::work_group);
}
}

Expand Down
14 changes: 7 additions & 7 deletions dpnp/tensor/libtensor/include/kernels/linalg_functions/gemm.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -653,7 +653,7 @@ class GemmBatchFunctorThreadK
}
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

std::size_t t_shift = block_s * delta_k * n_wi;
std::size_t global_s_offset = i * k + t_shift;
Expand All @@ -672,7 +672,7 @@ class GemmBatchFunctorThreadK
std::size_t workspace_i_shift = local_i * delta_k;
workspace[workspace_i_shift + local_s] = private_sum;

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

if (local_s == 0 && i < n) {
accV_t local_sum(workspace[workspace_i_shift]);
Expand Down Expand Up @@ -1063,7 +1063,7 @@ class GemmBatchFunctorThreadNM_vecm
}
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

const std::uint32_t lo_lhs_st_k = (wg_delta_n * wi_delta_n);
const std::uint32_t lo_rhs_rk_k = (wg_delta_m * wi_delta_m_vecs);
Expand Down Expand Up @@ -1095,7 +1095,7 @@ class GemmBatchFunctorThreadNM_vecm
}
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);
}

if constexpr (m_vec_size == 1) {
Expand Down Expand Up @@ -1942,7 +1942,7 @@ class GemmBatchNoAtomicFunctorThreadNM
}
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

i += local_i * wi_delta_n;
j += local_j * wi_delta_m;
Expand Down Expand Up @@ -2115,7 +2115,7 @@ class GemmBatchNoAtomicFunctorThreadK
}
}

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

std::size_t t_shift = block_s * delta_k * n_wi;
std::size_t global_s_offset = i * k + t_shift;
Expand All @@ -2134,7 +2134,7 @@ class GemmBatchNoAtomicFunctorThreadK
std::size_t workspace_i_shift = local_i * delta_k;
workspace[workspace_i_shift + local_s] = private_sum;

it.barrier(sycl::access::fence_space::local_space);
sycl::group_barrier(it.get_group(), sycl::memory_scope::work_group);

if (local_s == 0 && i < n) {
accV_t local_sum(workspace[workspace_i_shift]);
Expand Down
Loading