ffpa-attn 0.1.4

FFPA: Yet another Faster Flash Prefill Attention for large headdim, 1.8x~3x faster than SDPA EA.

🤖FFPA: Yet another Faster Flash Prefill Attention
with O(1)⚡️GPU SRAM complexity for large headdim🐑

📈L20 ~1.9x↑🎉 | 📈A30 ~1.8x↑🎉 | 📈3080 ~2.9x↑🎉 | 📈4090 ~2.1x↑🎉

FFPA(Split-D): Yet another Faster Flash Prefill Attention with Split-D strategy, achieve O(1) SRAM complexity and O(d/4) register complexity for large headdim (> 256), 1.8x~3x 🎉 faster than SDPA. Currently, FFPA supports self-attention, cross-attention, grouped/multi-query attention, causal attention with large headdim (D=320~1024). While the standard FlashAttention-2 only support headdim <= 256.

Self Attention	Cross/Decode Attention	GQA/MQA Attention	Causal Attention	Headdim
✔️(Nq = Nkv)	✔️(Nq != Nkv)	✔️(Nh_q % Nh_kv == 0)	✔️(causal mask)	32~1024

[!NOTE] FFPA has been tested on Ampere, Ada, Hopper, and Blackwell architectures (e.g., A30, L20, 4090, H200, 5090), achieves 1.8×~3×↑🎉 forward (CUDA) and 1.5×~2.5×↑🎉 backward (Triton w/ autotune) speedup over SDPA for headdim > 256.

📖 Quick Start

First, install the prebuilt package from PyPI or build ffpa-attn from source:

# Required: PyTorch>=2.11.0, CUDA>=13.0, Ubuntu>=22.04
pip3 install -U ffpa-attn # (support: sm_{80,90,...,120})
# Or, build ffpa-attn from source, just follow the cmds:
git clone https://github.com/xlite-dev/ffpa-attn.git
# Then, build the wheel package and install it with pip
cd ffpa-attn && MAX_JOBS=32 python3 setup.py bdist_wheel
# Optional: build ffpa-attn with ccache for faster rebuilds
apt install ccache && bash tools/build_fast.sh bdist_wheel
# Optional: for editable whl, use `pip install -e .` instead.
pip3 install dist/ffpa_attn-*.whl # pip uninstall ffpa-attn -y

Then, try to accelerate your attention computations with just ♥️one line♥️ of code ~

>>> import torch.nn.functional as F
>>> from ffpa_attn import ffpa_attn_func
>>> # Monkey-patch SDPA to point to FFPA attention. Every thing that
>>> # FFPA does not support will automatically fallback to SDPA. For
>>> # example, if the user calls SDPA with headdim <= 256, attn_mask
>>> # not None, and dropout_p > 0.0, it will fallback to the SDPA.
>>> F.scaled_dot_product_attention = ffpa_attn_func

For more advanced features, please refer to our online docs at 📘ffpa-attn.io.

📖 Split-D

We have extended FlashAttention for large headdim (D > 256) by implementing Fine-grained Tiling at the MMA level (GEMM style) for the Q@K^T and P@V matmul (namely, Split-D). This approach results in a constant SRAM usage of Br * 16 or Bc * 16 (Br = Bc) for Q, K, and V, leading to an overall SRAM complexity of O(Br * 16) ≈ O(1) and a register complexity of O(d/4). Consequently, this method allows us to extend headdim > 256 and achieve faster performance compared to SDPA with or without MMA Accumulation F32 (1.8x~3x 🎉 faster than SDPA EA).

We have named this new attention tiling technique FFPA: Faster Flash Prefill Attention. FFPA does not introduce any additional VRAM requirement, so the HBM memory complexity remains the same as FlashAttention.

By leveraging this approach, we can achieve better performance than SDPA EA for very large headdim (D > 256, FA-2 not supported). Approximate SRAM and register complexity analysis for FFPA is as follows: (d=headdim, C,Br,Bc=Constant, Br=Bc, let O(C)≈O(1)) 👇

📚Complexity Analysis	📚FFPA Attention (Split-D)	📚FlashAttention-2
SRAM	O(2xBrx16)≈O(1)	≈O(3xBrxd), d↑
Register	≈O(d/4), d↑	≈O(d/2), d↑
HBM	≈FA2≈O(Nd), O	≈O(Nd), O
Extra HBM	≈FA2≈O(N), m,l	≈O(N), m,l

🎉 Benchmark

Runnable examples are provided under examples. The performance benchmark for the NVIDIA RTX 4090 with large headdim (D=320~1024) is shown below, where FFPA achieves up to 2.1x 🎉 faster than SDPA. For more comprehensive benchmarks, please refer to our benchmark.

🤔 Why not TMA?

FFPA ships an experimental SM90 TMA path (enable_tma=True) that replaces the K/V cp.async global-to-shared transfer with cp.async.bulk.tensor.2d. After tuning (K SWIZZLE_128B, 64-col TMA box) it reaches parity with the cp.async baseline, but does not beat it.

FFPA's Split-D dataflow is a TMA anti-pattern. TMA wins when single thread instruction can amortise its dispatch cost over a large box, but split-D gives it narrow Bc x kMmaAtomK slices. It would require a major redesign (super-tiled K/V on TMA + warp-specialized WGMMA), rather than a drop-in K/V replacement.

©️License

Apache License 2.0

🎉Contribute

How to contribute? Wecome to star⭐️ this repo to support me👆🏻 ~

©️Citations

@misc{ffpa-attn@2025,
  title={FFPA: Yet another Faster Flash Prefill Attention for large headdim.},
  url={https://github.com/xlite-dev/ffpa-attn.git},
  note={Open-source software available at https://github.com/xlite-dev/ffpa-attn.git},
  author={DefTruth},
  year={2025}
}

📖 References

• flash-attention
• LeetCUDA
• flashinfer