Skip to content

[GPU] Fix relevant to zero padding for eltwise #32016

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

Draft
wants to merge 2 commits into
base: master
Choose a base branch
from
+17 −1
Draft

[GPU] Fix relevant to zero padding for eltwise #32016

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2035f91
[GPU] Fix add relevent to zero padding for eltwise
davidsnam-intel Sep 9, 2025
68b258b
Not to skip reset for eltwise
davidsnam-intel Sep 9, 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
2 changes: 1 addition & 1 deletion 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 @@ -2033,7 +2033,7 @@ void primitive_inst::prepare_primitive() {
skip_reset = skip_concat;
}
// Need to reset crop's output to zeros, if followed by onednn concatenation that requires zero-padding for blocked format memory
if (get_node().is_type<crop>() && get_node().can_share_buffer() && _impl_params->get_output_layout(0).format.is_blocked() &&
if ((get_node().is_type<crop>() || get_node().is_type<eltwise>()) && get_node().can_share_buffer() && _impl_params->get_output_layout(0).format.is_blocked() &&
get_node().get_users().size() == 1 && get_node().get_users().front()->is_type<concatenation>() &&
get_node().get_users().front()->get_selected_impl()->is_onednn()) {
skip_reset = false;
Expand Down
10 changes: 10 additions & 0 deletions src/plugins/intel_gpu/src/kernel_selector/cl_kernels/eltwise_blocked_opt.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,16 @@ KERNEL(eltwise_blocked_opt)(INPUTS_DECLS
if ((f_block*VEC_SIZE) >= OUTPUT_FEATURE_NUM || b > OUTPUT_BATCH_NUM) {
return;
}
// if (b > OUTPUT_BATCH_NUM || (f_block*VEC_SIZE + VEC_SIZE) > (OUT_F_BLOCK * FEATURE_BLOCK_SIZE)) {
// return;
// }

// // zero-padding the blocked format padded memory area since it might be used as input of onednn concatenation
// if (NEED_ZERO_PADDING && (f_block*VEC_SIZE) >= OUTPUT_FEATURE_NUM) {
// OUTPUT_TYPE_BLOCK out = (MAKE_VECTOR_TYPE(OUTPUT_TYPE, VEC_SIZE))(0);
// VSTORE_N(out, global_id, output);
// return;
// }

MAKE_VECTOR_TYPE(ACCUMULATOR_TYPE, VEC_SIZE) res;

Expand Down
6 changes: 6 additions & 0 deletions src/plugins/intel_gpu/src/kernel_selector/kernels/eltwise/eltwise_kernel_blocked_opt.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -303,6 +303,12 @@ JitConstants EltwiseKernel_blocked_opt::GetJitConstants(const eltwise_params& pa
jit.AddConstant(MakeJitConstant("OUTPUT_SIZE_XY", params.outputs[0].X().v * params.outputs[0].Y().v));
// To calculate batch, define outer block size of feature axis (divided by the inner feature-block size)
jit.AddConstant(MakeJitConstant("OUT_F_BLOCK", CeilDiv(params.outputs[0].Feature().v, inner_feature_blk_size)));
// jit.AddConstant(MakeJitConstant("FEATURE_BLOCK_SIZE", inner_feature_blk_size));
// if (params.operations[0].mode == EltwiseMode::ASSIGN && inner_feature_blk_size > 1) {
// jit.AddConstant(MakeJitConstant("NEED_ZERO_PADDING", true));
// } else {
// jit.AddConstant(MakeJitConstant("NEED_ZERO_PADDING", false));
// }

bool use_vload = false;
jit.Merge(MakeInputDeclsJitConstants(params, use_vload));
Expand Down
Loading