[Common] Group NVFP4 Quantize Kernels

[Oleg-Goncharov]

Description

Please include a brief summary of the changes, relevant motivation and context.

Fixes # (issue)

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:

• Change A
• Change B

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

[greptile-apps]

Greptile Summary

This PR adds coordinated grouped NVFP4 1D quantization and transpose kernels, integrates them with grouped dispatch and layout infrastructure, and adds extensive native test coverage.

• Adds grouped NVFP4 quantization, transpose, scaling, and TMA/layout support.
• Restores the delayed-scaling, MXFP8, and FP8 block-scaling dispatch branches identified in the previous review.
• Registers grouped NVFP4 transpose tests covering multiple group counts and shape representations.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported grouped scaling-mode dispatch branches are active at the current head.

Important Files Changed

Filename	Overview
transformer_engine/common/cast/dispatch/quantize.cuh	Integrates grouped NVFP4 dispatch and restores the previously disabled format-specific grouped quantization branches.
transformer_engine/common/cast/nvfp4/specialized/group_quantize_transpose_nvfp4_tuned_1D.cuh	Implements the tuned grouped NVFP4 1D quantize-and-transpose launch path with shape and allocation validation.
transformer_engine/common/cast/core/grouped_layout.cuh	Extends shared grouped-layout infrastructure used to map variable-sized groups to coordinated kernel work.
transformer_engine/common/cast/core/grouped_tma.cuh	Extends grouped TMA support required by the new NVFP4 kernel path.
tests/cpp/operator/test_cast_nvfp4_transpose_grouped.cu	Adds numerical coverage for grouped NVFP4 rowwise and columnwise output across group counts and shape representations.
tests/cpp/operator/CMakeLists.txt	Registers the grouped NVFP4 transpose test in the native operator test executable.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  I[Grouped BF16 input] --> D[Grouped quantization dispatch]
  D --> K[Grouped NVFP4 quantize and transpose kernel]
  K --> R[Rowwise packed FP4 data and scales]
  K --> C[Columnwise packed FP4 data and scales]
  R --> G[Grouped GEMM consumers]
  C --> G

Loading

_{Reviews (3): Last reviewed commit: "[pre-commit.ci] auto fixes from pre-comm..." | Re-trigger Greptile}