Freescale Discovery Labs (FDL) - Engineer (Contract Position) (11784685) Freescale Semiconductor Inc.
Primary Location: US-TX-Austin-Oak Hill (TX30)
Job Field: Research and Development
Education Level: Bachelor's Degree
Job Type: Experienced
Relocation Available?: No
The quark highly-efficient embedded architecture research project in Freescale Discovery Labs is investigating future architectures tuned to systems with thousands of cores with minimal area and power footprint in both the cores themselves and in the footprint of their software. This is not a simple re-encoding exercise for an existing architecture, but an attempt to define useful new semantics to substantially improve pathlength, to support low-latency, low-energy intercore communications in usermode, how to get the effect of caches without constructing tag arrays, and much more. We hope eventually to have a coherent and comprehensible programming paradigm for such systems, with an architecture which gives us unfair advantages over others in the basic dimensions of area and power. Initially we will create and exercise interesting ?microbenchmarks?, written in assembler, to arrive at a good architecture candidate. The assembler and simulator will be generated by ADL. We will then take ?real? software written in a high level language and exercise and adapt the architecture with these workloads. We need to be able to retarget our compiler very quickly to track the evolving architecture; we therefore expect to use LLVM as the base compiler technology and hope to adapt ADL to aid in automatic generation of the necessary data for the LLVM target definition.
identify useful microbenchmarks suitable for evaluating the architecture under analysis for functionality and performance
write and verify and exercise and measure the microbenchmarks
identify suitable ?real? code suitable for exercising the architecture under analysis, modify it as needed for execution in our simulation (ADL-generated interpreter) environment, and verify and exercise and measure the benchmarks
transform appropriate ?real? benchmarks into highly-concurrent forms, to demonstrate extreme multicore utility
work with the other team members to construct prototype minimal ?RTOS? runtime infrastructure to support the multicore, multithread execution of the transformed real code
contribute to the architecture, using experience in general and the observations from the workload analysis
write up experiences and knowledge gained so that others can leverage learning as well as deliverables.
This is a consulting assignment which will be assigned to an existing project that has been accepted to Discovery Labs. As such it is a temporary assignment with a variable duration based upon the project's success. Most projects will be 1-1.5 years in duration. Contractors will be payrolled through CoreTech.
general assembler, C, and C++ software experience
general knowledge of network software and other embedded hard realtime application domains
experience with highly-concurrent design, language (csp, occam, go, ...) or software a big plus
good oral and written communication skills
desire and ability to work effectively in a small team
||Austin, TX |