openvino

31a0cfbc - [GPU] Fix implicit onednn concat for 64 byte misalignment regression. (#31762)

[GPU] Fix implicit onednn concat for 64 byte misalignment regression. (#31762) ### Description of the issue(symptom, root-cause, how it was resolved) - Fix PR for https://github.com/openvinotoolkit/openvino/pull/31452 regression. - Solves a problem when using a mix of ocl and onednn impl in concat's input. #### The code and line that caused this issue (if it is not changed directly) - In case of concat that uses impl type mixed together, byte offset calculation is incorrect. Earlier always calculate byte offset only all onednn impl inputs. #### Reproduction step and snapshot (if applicable. Do not attach for customer model) - ./benchmark_app -d GPU.1 -m ww21_weekly_23.0.0-10926-b4452d56304-API2.0/densenet-121/tf/tf_meta/FP16/ INT8/1/dldt/optimized/densenet-121.xml -nstreams 1 -nireq 2 -niter 1 --hint none -infer_precision f32 #### Problematic graph - No graph related. #### Checklist - [O] Is it a proper fix? (not a workaround) - [O] Did you include test case for this fix, if necessary? - [O] Did you review existing test that can be extended to cover this scenario? Which test did you review? concatenation_gpu_test.cpp ### Tickets: - *170847* Signed-off-by: hyunback <hyunback.kim@intel.com>