[Common] row-scaled nvfp4 path: add single-launch group fused amax

[cael-ling]

Description

Adds a single-kernel row/col amax pass for grouped (MoE) row-scaled NVFP4. It replaces the per-expert loop of compute_rowwise_amax + compute_columnwise_amax with one launch over a packed (sum_M, K) input, exposed as nvte_group_nvfp4_compute_amax. The kernel is the grouped generalization of #3454 - the single-tensor compute_fused_amax_kernel (same tiling / SMEM pipeline / acquire-barrier reduction). A single input TMA map covers all experts; per-expert amax vectors are selected via a prefix sum of row counts, so a 128-row chunk always lands in one expert.

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

New gtest tests/cpp/operator/test_row_scaled_nvfp4_grouped_amax.cu:

• NVFP4GroupedRowScaledAmaxTestSuite.MatchesPerExpertOracle — grouped amax vs a CPU per-row/per-col max(|x|) reference over the same BF16 bytes, plus cross-launch determinism, across 6 expert/shape configs.
• DISABLED_LoopVsGrouped — microbench comparing the same kernel called per-expert (loop)
  vs once (grouped).

cmake -Btests/cpp/build tests/cpp && cmake --build tests/cpp/build -j
./tests/cpp/build/operator/test_operator --gtest_filter='*NVFP4GroupedRowScaledAmax*'
./tests/cpp/build/operator/test_operator --gtest_also_run_disabled_tests --gtest_filter='*GroupedRowScaledAmaxBench*'

Performance

Same kernel called per-expert (loop) vs a single grouped launch, GB200, K=4096. Speedup is loop / grouped.

experts	M=128 loop→grouped (ms)	speedup	M=512 loop→grouped (ms)	speedup
2	0.0179 → 0.0117	1.52x	0.0180 → 0.0124	1.45x
4	0.0288 → 0.0120	2.39x	0.0271 → 0.0138	1.97x
8	0.0513 → 0.0124	4.15x	0.0514 → 0.0165	3.11x
16	0.0950 → 0.0140	6.81x	0.0960 → 0.0238	4.04x
32	0.1852 → 0.0173	10.70x	0.2009 → 0.0370	5.43x
64	0.3631 → 0.0232	15.65x	0.3960 → 0.0607	6.53x

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

The PR adds a public grouped NVFP4 amax operation that computes per-expert rowwise and columnwise maxima from a packed BF16 input using one CUDA launch.

• Adds the grouped fused amax kernel, argument preparation, output initialization, and input-shape validation.
• Supports independently optional rowwise and columnwise amax buffers for each expert.
• Adds numerical, determinism, heterogeneous-buffer, and disabled benchmark coverage to the native C++ operator suite.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the trailing-column truncation is prevented by explicit alignment validation, and nullable per-expert amax buffers are guarded before every zeroing or atomic-write access.

Important Files Changed

Filename	Overview
transformer_engine/common/cast/nvfp4/group_quantize_transpose_nvfp4.cuh	Implements the grouped fused kernel and now validates tile alignment while safely skipping null per-expert amax buffers.
transformer_engine/common/cast/cast_grouped.cu	Adds the ABI implementation that converts tensor handles and dispatches the grouped amax operation.
transformer_engine/common/include/transformer_engine/cast.h	Declares and documents the packed-input and per-expert-output contract for the new C API.
tests/cpp/operator/test_row_scaled_nvfp4_grouped_amax.cu	Adds reference comparisons, determinism checks, heterogeneous columnwise-buffer coverage, and an optional performance benchmark.
tests/cpp/operator/CMakeLists.txt	Registers the grouped NVFP4 amax test source in the native operator test executable.

_{Reviews (2): Last reviewed commit: "[Common] Enforce K alignment and per-exp..." | Re-trigger Greptile}