Support communication/gemm overlap for [Wgrad->Dgrad] execution order in the bwd pass.

[fanshiqing]

Description

Support communication/gemm overlap for [Wgrad->Dgrad] execution order in the bwd pass.

Taking MLP modules as an example:

New added comm/gemm patterns include:

• ⑥ AG->fc2_wgrad / proj_wgrad
• ⑦ AG->fc1_wgrad / qkv_wgrad

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

• transformer_engine/common/comm_gemm_overlap/comm_gemm_overlap.{cpp, h}: support new patterns ⑥ and ⑦; aggregate 2X AG and GEMM chunks to lower then overhead
• transformer_engine/pytorch/module/base.py: verify and clean the user specified comm/gemm overlap knobs to avoid conflicts.

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[zhongbozhu]

/te-ci pytorch L1