Abstract
The VLIW architecture can be exploited to greatly enhance instruction level parallelism, thus it can provide computation power and energy efficiency advantages, which satisfies the requirements of future sensor-based systems. However, as VLIW codes are mainly compiled statically, the performance of a VLIW processor is dominated by the behavior of its compiler. In this paper, we present an advanced compiler designed for a VLIW DSP named Magnolia, which will be used in sensor-based systems. This compiler is based on the Open64 compiler. We have implemented several advanced optimization techniques in the compiler, and fulfilled the O3 level optimization. Benchmarks from the DSPstone test suite are used to verify the compiler. Results show that the code generated by our compiler can make the performance of Magnolia match that of the current state-of-the-art DSP processors.
Highlights
Nowadays, sensor-based systems are becoming more and more widely used in many domains due to their possibilities in collecting various data from the environment, such as the temperature, humidity, luminosity and many other parameters, and measuring and otherwise processing it for different purposes
As development of the Magnolia processor is still in progress, we only evaluated the performance of running the compiled code on a simulator designed for Magnolia, which is based on the gem5 [9] simulator
These results indicated that with all the optimization techniques working, the performance gain can be around 4.39 times on average against to the code generated without any optimization
Summary
Sensor-based systems are becoming more and more widely used in many domains due to their possibilities in collecting various data from the environment, such as the temperature, humidity, luminosity and many other parameters, and measuring and otherwise processing it for different purposes. The huge number of emerging applications has imposed strong requirements, such as real-time processing, low-power consumption, reduced size, high-precision algorithms, efficient and secure communications, and many others, on the sensor-based systems, increasing demands for architecture improvement and optimization. Due to their balanced combination of flexibility and hardware performance, Digital Signal Processors (DSPs) have been more and more adopted in sensor-based systems. Out-of-order execution architecture executes instructions in an order different from the one they appear in the program, it can make use of clock cycles that would otherwise be wasted by a certain type of costly delay [1] These techniques all come at a cost: increased hardware complexity.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
