 |
HAL-LIRMM - Open Archive |  |
Deterministic OpenMP and the LBP Parallelizing Manycore Processor
Abstract : This paper presents Deterministic OpenMP, a new runtime for OpenMP programs , and the Little Big Processor (LBP) manycore processor design. LBP isolates the application it runs, using its multiple cores to distribute a non in-terruptible single parallelized computation. When run on LBP, a Deterministic OpenMP code produces cycle by cycle deterministic computations. LBP and Deterministic OpenMP are particularly suited to safely accelerate real time embedded applications through their parallel execution. The paper reports experimental results to measure the LBP performance on an FPGA based simulator and compares LBP with a Xeon Phi 2. The instruction retired and Instruction Per Cycle (IPC) measured in the experiment indicate that LBP performance does not suffer from the elimination of all the high performance features incompatible with an embedded processor (predictors, caches). LBP performance comes from its hardware parallelizing capability, its capacity to capture distant Instruction Level Parallelism (ILP) and from our proposed Deterministic OpenMP runtime favoring local memory accesses.
Cited literature [19 references]
https://hal-lirmm.ccsd.cnrs.fr/lirmm-02767830
Contributor : David Parello <>
Submitted on : Thursday, June 4, 2020 - 10:09:25 AM
Last modification on : Monday, June 8, 2020 - 4:00:25 AM
Long-term archiving on: : Friday, December 4, 2020 - 5:08:22 PM
Contributor : David Parello <>
Submitted on : Thursday, June 4, 2020 - 10:09:25 AM
Last modification on : Monday, June 8, 2020 - 4:00:25 AM
Long-term archiving on: : Friday, December 4, 2020 - 5:08:22 PM
File
goossens-jpdc-si.pdf
Files produced by the author(s)