Full text
| |
| Author(s): |
Juan Jesús Salamanca Guillén
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Computação |
| Defense date: | 2016-12-21 |
| Examining board members: |
Guido Costa Souza de Araújo;
Alexandro José Baldassin;
Márcio Bastos Castro;
Sandro Rigo;
Emilio de Camargo Francesquini
|
| Advisor: | Guido Costa Souza de Araújo; José Nelson Amaral |
| Abstract | |
Thread-Level Speculation (TLS) is a hardware/software technique that enables the execution of multiple loop iterations in parallel, even in the presence of some loop-carried dependences. TLS requires hardware mechanisms to support conflict detection, speculative storage, in-order commit of transactions, and transaction roll-back. Prior research has investigated approaches to implement TLS, either on dedicated hardware or purely in software, and has attempted to predict the performance of future TLS hardware implementations. Nevertheless, there is no off-the-shelf processor that provides direct support for TLS. Speculative execution is supported, however, in the form of Hardware Transactional Memory (HTM) ¿ available in recent processors such as the Intel Core and the IBM POWER8. HTM implements three key features required by TLS: conflict detection, speculative storage, and transaction roll-back. Before applying TLS to a hot loop, it is necessary to determine if the loop has potential to be amenable. A loop could be amenable if the probability of loop-carried dependences at runtime is low; to measure this probability loop dependence profiling is used. This project presents a novel dynamic loop-carried dependence checker integrated as a new extension to OpenMP, the parallel for check construct, which can be used to help programmers identify the existence of loop-carried dependences in parallel for constructs. This work also presents a detailed analysis of the application of HTM support for loop parallelization with TLS and describes a careful evaluation of the implementation of TLS on the HTM extensions available in such machines. As a result, it provides evidence to support several important claims about the performance of TLS over HTM in the Intel Core and the IBM POWER8 architectures. Experimental results reveal that by implementing TLS on top of HTM, speed-ups of up to 3.8× can be obtained for some loops. Finally, this work describes a novel speculation technique for the optimization, and simultaneous execution, of multiple alternative traces of hot code regions. This technique, called Speculative Trace Optimization (STO), enumerates, optimizes, and speculatively executes traces of hot loops. It requires hardware support that can be provided in a similar fashion as that available in HTM systems. This work discusses the necessary features to support STO, namely multi-versioning, lazy conflict resolution, eager conflict detection, and transaction synchronization. A review of existing HTM architectures ¿ Intel TSX, IBM BG/Q, and IBM POWER8 ¿ shows that none of them has all the features required to implement STO. However, this work demonstrates that STO can be implemented on top of existing HTM architectures through the addition of privatization and wait/resume code (AU) | |
| FAPESP's process: | 15/04285-5 - Speculative code execution for multicore architectures |
| Grantee: | Juan Jesús Salamanca Guillén |
| Support Opportunities: | Scholarships in Brazil - Doctorate (Direct) |