In this paper we show a new execution paradigm based on Decoupled Software Pipelining in the context of Scheduled Dataflow (SDF) architecture.  We call the new architecture MT-SDF. We introduce mini-threads to execute loops as a software pipeline. We permit the mini-threads to share registers. We present a qualitative and quantitative comparison of the mini-threads with the original SDF architecture, and out-of-order superscalar architecture. We use several benchmark