Font Size: a A A

Research On Key Technology Of Multi - Core Processor Mapping

Posted on:2015-02-19Degree:MasterType:Thesis
Country:ChinaCandidate:S H FanFull Text:PDF
GTID:2208330464959683Subject:Integrated circuit engineering
Abstract/Summary:PDF Full Text Request
The fact that multi-core processor provides a more powerful performance than single-core meets the people’s endless desire for performance. But how to fully exploit multi-core processor’s efficient processing capabilities to enable parallel implementation of a variable of applications has been a hot research subject.For those traditional programmer, the lack of the parallel program development and not deep enough understanding of the underlying hardware and parallel algorithms, these factors make it a very difficult problem to program in parallel which in turn limits the application of multi-core processors.For this situation, our group has developed a tool direct to multi-core systems which can parallelize the serial program. But the tool’s input must be single function program, based on this, this thesis extends the function of the tool so that it can support the partition in function level. In a program, function is a basic form and the function callings constitute the main framework. Therefore supporting the function calling is significant for making the tool more universal and more practical.To achieve the above goal, first, the dependency relationship of each function will be analyzed to get the extended data flow graph (EDFG). Second, according to the calling relationship between functions, the EDFGs of the calling function and the called function will be integrated so that we will get the whole EDFG of the program. Last, the whole EDFG will be divided to several threads by an efficient fusion algorithm to fulfill the automatic parallel partition. The tool developed according to the above solution supports the conversion from the serial program with the function calling to the parallel one in a static way and inserts the communication between the threads for the specific hardware platform so that fulfill the whole function from the serial source program to mapping to hardware. The test results show that the execution efficiency has improved obviously after the conversion for the source program from the serial to the parallel by the tool.
Keywords/Search Tags:parallelization, function calling, task partition, multi-core mapping
PDF Full Text Request
Related items