new api changes for unit tests for supported configurations

[rajprince-intel]

Description

Migrate GEMM tests from legacy Intel Xe-specific API to new APIs.

Changes:

• Grouped GEMM: gemm_universal_mainloopintelxexmx16group.cpp
• Epilogue Fusion GEMM: gemm_universal_lincomb_per_rowbias_eltact.cpp

Type

• Bug
• Feature
• Performance
• Refactor

Testing

• Tests pass
• Xe12
• Xe20

Performance

Metric	Before	After
Performance	No change (API migration only)	No change

References

Related to generic API modernization effort.

Checklist

• Copyright headers preserved (Intel 2025)
• Co-pilot Review (self-reviewed migration)
• Deprecated APIs not used (migrated away from legacy APIs)

---

Quick Reference

Component	Grouped GEMM	Epilogue Fusion GEMM
File	gemm_universal_mainloopintelxexmx16group.cpp	gemm_universal_lincomb_per_rowbias_eltact.cpp
GmemTiledCopy	XE_2D_U16x32x32_* → void	XE_2D_U16x32x32_* → void
MMA Atom	XE_8x16x16_* → XE_DPAS_TT<8, float, bfloat16_t>	XE_8x16x16_* → XE_DPAS_TT<8, float, bfloat16_t>
GEMM Dispatch	MainloopIntelXeXMX16Group → MainloopXeL1StagedGroup	MainloopIntelXeXMX16 → MainloopXeL1Staged
Epilogue Dispatch	IntelXeXMX16Group (unchanged) [based on example09]	IntelXeXMX16 → IntelXeGeneric
Epilogue Tile	N/A [based on example09]	Add void as 3rd param
Epilogue Atoms	Explicit (unchanged) [based on example09]	Explicit → void

Code Examples

Grouped GEMM Changes

- using GmemTiledCopyA = XE_2D_U16x32x32_LD_N;
+ using GmemTiledCopyA = void;

- MMA_Atom<XE_8x16x16_F32BF16BF16F32_TT>
+ MMA_Atom<XE_DPAS_TT<8, float, cute::bfloat16_t>>

- using GEMMDispatchPolicy = cutlass::gemm::MainloopIntelXeXMX16Group<PipelineStages>;
+ using GEMMDispatchPolicy = cutlass::gemm::MainloopXeL1StagedGroup<PipelineStages>;

Epilogue Fusion GEMM Changes

- using GEMMDispatchPolicy = cutlass::gemm::MainloopIntelXeXMX16<PipelineStages>;
+ using GEMMDispatchPolicy = cutlass::gemm::MainloopXeL1Staged<PipelineStages>;

- using EpilogueDispatchPolicy = cutlass::epilogue::IntelXeXMX16;
+ using EpilogueDispatchPolicy = cutlass::epilogue::IntelXeGeneric;

  using CollectiveEpilogue = cutlass::epilogue::collective::CollectiveEpilogue<
      EpilogueDispatchPolicy, TileShape,
+     void,                    // NEW: Epilogue tile (automatic)
      ElementAccumulator, ..., FusionCallBacks,
-     XE_2D_U32x8x16_LD_N, void, void, XE_2D_U32x8x16_ST_N, void, void>;
+     void, void>;             // Automatic load/store atoms