PR #35335 Extend moe_3gemm to all oneDNN aware GPUs

Extend moe_3gemm to all oneDNN aware GPUs #35335

peterchen-intel wants to merge 22 commits into openvinotoolkit:master from peterchen-intel:oom/fixing

intel_gpu: fix Qwen3 MoE GEMM3_SWIGLU on MTL-class (non-systolic) iGPU

a7fa364a

intel_gpu: scope MOE3GemmFusedCompressed queue fix to in_order only

2bcc00a2

intel_gpu: revert skip_transfer_on_igpu extension to MTL

d442d393

intel_gpu: revert moe_gather.hpp rank-2 fix (not needed in final path)

4a0a5044

peterchen-intel requested a review 50 days ago

github-actions added category: GPU

peterchen-intel requested a review from

riverlijunjie 50 days ago

peterchen-intel assigned

e-ddykim 50 days ago

peterchen-intel added under_perf_check

intel_gpu: fix oneDNN engine init for MoE on non-systolic GPU

e1a7c831

peterchen-intel requested a review from

copilot-pull-request-reviewer 49 days ago

copilot-pull-request-reviewer commented on 2026-04-15

rkazants commented on 2026-04-16

Merge branch 'master' into oom/fixing

8a030fd1

peterchen-intel added do_not_merge

intel_gpu: fix unnecessary tmp_out buffer per-layer in paged_attention

1498dd1f

peterchen-intel force pushed to 1498dd1f 43 days ago

Merge branch 'master' into oom/fixing

20045578

peterchen-intel removed under_perf_check

peterchen-intel removed do_not_merge

Merge branch 'master' into oom/fixing

c9eb5c68

peterchen-intel added under_perf_check

peterchen-intel changed the title ~~Extend moe_3gemm to all Ultra series iGPU~~ Extend moe_3gemm to all GPUs and reduce internal buffer holding in paged_attention_opt 42 days ago

riverlijunjie commented on 2026-04-24

Merge branch 'master' into oom/fixing

046c575a

riverlijunjie approved these changes on 2026-05-11

Merge branch 'master' into oom/fixing

3ebb016f

Roll back the mistaken change

990c0bc7

Merge branch 'master' into oom/fixing

39ba7fca

peterchen-intel commented on 2026-05-21

iintel_gpu: enable compressed MoE fusion chain on non-systolic devices

2ee0c089

peterchen-intel force pushed to 2ee0c089 12 days ago

Merge branch 'master' into oom/fixing

12847654

peterchen-intel changed the title ~~Extend moe_3gemm to all GPUs and reduce internal buffer holding in paged_attention_opt~~ Extend moe_3gemm to all Intel GPUs 12 days ago

peterchen-intel requested a review from

copilot-pull-request-reviewer 12 days ago

peterchen-intel requested a review from

riverlijunjie 12 days ago

peterchen-intel requested a review from

rkazants 12 days ago

copilot-pull-request-reviewer commented on 2026-05-22

Support oneDNN known gpu_arch only

f8d0cc30

peterchen-intel requested a review from

copilot-pull-request-reviewer 1 day ago

copilot-pull-request-reviewer commented on 2026-06-02

device_info.arch >= cldnn::gpu_arch::xe_lp

3875448b

peterchen-intel changed the title ~~Extend moe_3gemm to all Intel GPUs~~ Extend moe_3gemm to all oneDNN known GPUs 1 day ago

peterchen-intel changed the title ~~Extend moe_3gemm to all oneDNN known GPUs~~ Extend moe_3gemm to all oneDNN aware GPUs 1 day ago

peterchen-intel requested a review from

copilot-pull-request-reviewer 1 day ago

Merge branch 'master' into oom/fixing

ae72a35e

copilot-pull-request-reviewer commented on 2026-06-02

peterchen-intel commented on 2026-06-02

For ENABLE_ONEDNN_FOR_GPU only

fa26fd1f

peterchen-intel commented on 2026-06-02

Remove duplicate

4ce2b9a5

Remove duplicate

dfd234cc

peterchen-intel requested a review from

copilot-pull-request-reviewer 1 day ago

copilot-pull-request-reviewer commented on 2026-06-02

Define disable_moe_opt only if ENABLE_ONEDNN_FOR_GPU

8edab885

Reviewers

riverlijunjie

rkazants

copilot-pull-request-reviewer

Assignees

e-ddykim

Labels

category: GPU under_perf_check

Milestone

No milestone

openvino Extend moe_3gemm to all oneDNN aware GPUs #35335 Open

Extend moe_3gemm to all oneDNN aware GPUs #35335

openvino
Extend moe_3gemm to all oneDNN aware GPUs
#35335

Open