MASQ: Accelerating Masked Diffusion via Stage-Wise Multi-Precision Quantization
This work addresses the computational inefficiency of masked diffusion for region-specific image synthesis, enabling faster and more energy-efficient deployment on hardware.
MASQ introduces a hardware-software co-designed accelerator for masked diffusion that reduces computational redundancy by processing only the masked region. It achieves up to 16.06x speedup and 4.18x energy-efficiency gain over A100 while preserving quality.
Masked diffusion enables region-specific image synthesis but suffers from computational redundancy, since the entire image is processed each timestep even though only the masked region requires generation. To address this, we introduce MASQ, a hardware-software co-designed accelerator for masked diffusion. Our approach performs stage-wise MXINT8/4/2 precision assignment that dynamically reflects spatial and semantic importance, complemented by timestep-aware scheduling and optimized non-matrix operations. MASQ features a block-wise multi-precision compute engine and mask management unit, efficiently handling our approach. It achieves up to 16.06x and 5.39x speedup and 4.18x and 4.93x energy-efficiency gain over A100 and Orin NX, respectively, while preserving quality.