[Oleg-Goncharov]

Description

Add a fused grouped MXFP8 requantization path for Blackwell GPUs. The new operation converts compact rowwise MXFP8 input into columnwise E4M3 MXFP8 without materializing an intermediate high-precision tensor in global memory.

This reduces memory traffic compared with separate grouped dequantization and quantization calls. The kernel uses a tiled TMA pipeline and packed data movement/conversion operations, and supports both compact and GEMM-swizzled output scaling factors. The quantization configuration selects either an FP32 intermediate or a faster BF16 intermediate.

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:

• Add the nvte_group_requantize_mxfp8 C API for fused grouped rowwise-to-columnwise MXFP8 requantization.
• Add an SM100+ TMA-based kernel supporting grouped tensors, FP8 input types, E4M3 output, and compact or GEMM-swizzled output scales.
• Add optimized BF16 fast-math and FP32 intermediate-precision paths.
• Optimize packed data access and conversion in the requantization kernel.
• Add parameterized C++ correctness tests comparing the fused operation against separate dequantization and quantization.

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]

Confidence Score: 4/5

[High impact] Replaces grouped MXFP8 requantization kernel with new implementation.

Fix the missing CuTeDSL bridge interfaces before merging.

Findings

1.  CuTeDSL builds fail ▶

Summary

Replaces the earlier fused path with grouped MXFP8 requantization and adds a CuTeDSL backend.

• Grouped MXFP8 requantization converts rowwise input to columnwise E4M3 output in one operation.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Compact rowwise MXFP8] --> B[Grouped requantize API]
  B --> C{CuTeDSL available?}
  C -->|Yes| D[CuTeDSL kernel]
  C -->|No| E[CUDA kernel]
  D --> F[Columnwise E4M3 and scales]
  E --> F
  D --> G[Optional BF16 output]
  E --> G

Loading

_{Reviews (11) · Last reviewed commit: "Merge main into pr_requantize_mxfp8_perf" · Reviewed by Greptile}

[ptrendx]

@Oleg-Goncharov This PR seems to have some unrelated changes (like the nccl ep submodule). Could you fix that?

[Oleg-Goncharov]

/te-ci

[ptrendx]

@Oleg-Goncharov Please sign every commit in this PR. See https://github.com/NVIDIA/TransformerEngine/blob/main/CONTRIBUTING.rst#sign-your-work

[greptile-apps]

CuTeDSL builds fail

MXFP8RequantConfig::get_kernel uses tvm_ffi_bridge::TVMFFIKernel and make_tvm_ffi_kernel, but neither exists in the repository. Builds with NVTE_WITH_CUTEDSL=ON therefore fail while compiling this header, even when the backend is disabled at runtime.

Use std::optional<tvm::ffi::Function> throughout the cache, as the existing bridge expects.