Schedule Data, Not Code
نویسندگان
چکیده
In this paper we argue that the scheduler, as the intermediary between hardware and software, needs to be fully data-aware. The old paradigm of envisioning tasks as amorphous blobs of ‘work’ to be assigned to processors is incomplete and needs be expanded. Some techniques and projects have emerged that implicitly use this idea, but either focus on a small aspect of data or are targeted to optimizing specific problems. We argue for more general solutions.
منابع مشابه
Evolutionary Algorithm Based Exploration of Software Schedules for Digital Signal Processors
The simultaneous exploration of tradeo s be tween program memory data memory and ex ecution time requirements D for DSP dig ital signal processing algorithms in embed ded computing environments is a demanding application and example par excellence of a multi objective optimization problem In or der to solve this problem two evolutionary algorithms are shown to be successfully appli cable for ex...
متن کاملCode generation by using integer-controlled dataflow graph
Integer-Controlled Dataflow (IDF) and its code generation applications in Ptolemy are presented. In IDF graphs, which specify data processing systems, data token flow is controlled by integer control tokens and states of actors at run-time. The firing order of actors (schedule) is determined at compile-time, however, the actors are conditionally activated at run-time. This static schedule contr...
متن کاملMemory Efficient Software Synthesis with Mixed Coding Style from Dataflow Graph
Due to the limited amount of memory resources in embedded systems, minimizing the memory requirement is an important goal of software synthesis. This paper presents a set of techniques to reduce the code and data sizes for software synthesis from graphical DSP programs based on the synchronous dataflow (SDF) model. Observing that the required code and buffer sizes of the generated code depend o...
متن کاملQuasi-Static Scheduling of Reconfigurable Dataflow Graphs for DSP Systems
Dataflow programming has proven to be popular for representing applications in rapid prototyping tools for digital signal processing (DSP); however, existing dataflow design tools are limited in their ability to effectively handle dynamic application behavior. In this paper, we develop efficient quasi-static scheduling techniques for a broad class of dynamically-reconfigurable dataflow specific...
متن کاملCode Optimization of Polynomial Approximation Functions on Clustered Instruction-level Parallelism Processors
In this paper, we propose a general code optimization method for implementing polynomial approximation functions on clustered instruction-level parallelism (ILP) processors. In the proposed method, we first introduce the parallel algorithm with minimized data dependency. We then schedule and map the data dependency graph (DDG) constructed based on the parallel algorithm to appropriate clusters ...
متن کاملA Code Motion Framework for Global Instruction Scheduling
A framework for global instruction scheduling is developed that progressively improves a given instruction schedule by eliminating delay slots in the schedule one at a time through the application of global code motions. The elimination of a delay slot is carried out in two steps: a goal oriented search which identiies a global code motion or a cascade of code motions that eliminate the delay w...
متن کامل