The janus triad: exploiting parallelism through dynamic binary modification

Abstract

We present a uni ed approach for exploiting thread-level, data-level, and memory-level parallelism through a same-ISA dynamic binary modi er guided by static binary analysis. A static binary analyser rst examines an executable and determines the operations required to extract parallelism at runtime, encoding them as a series of rewrite rules that a dynamic binary modi er uses to perform binary transformation. We demonstrate this framework by exploiting three di erent kinds of parallelism to perform automatic vectorisation, software prefetching, and automatic parallelisation together on legacy application binaries. Software prefetch insertion alone achieves an average speedup of 1.2×, comparing favourably with an automatic compiler pass. Automatic vectorisation brings speedups of 2.7× on the TSVC benchmarks, signi cantly beating a compiler approach for some workloads. Finally, combining prefetching, vectorisation, and parallelisation realises a speedup of 3.8× on a representative application loop.

Cinnamon: A Domain-Specific Language for Binary Profiling and Monitoring

“…Dyninst [13] offers both static and dynamic binary rewriting approaches and has been used to build tools for profiling and performance analysis [46]. Janus [26] is a hybrid static-dynamic binary modification framework based on DynamoRIO that combines the power of both static and dynamic approaches to perform complex analyses and transformations, such as binary parallelization and vectorization.…”

“…At the entry block of each loop, during its execution, the loop is marked as an active or living loop (lines 10-12) and set back to inactive at the exit block of the loop (lines [13][14][15]. Next, at the entry to each basic block, a global counter of executed blocks is incremented and local counter of each active loop is also incremented (lines [23][24][25][26][27][28][29][30][31] to record that this basic block has been executed. At program exit, the coverage of each loop is calculated as a percentage of the total number of basic blocks executed in the program (lines 31-36).…”

“…These can also be used to identify security vulnerabilities [9], [22]- [25] that can be exploited by malicious attackers. Beyond passive profiling and monitoring, other techniques actively transform the binary to optimize performance through the code that is executed [26]- [28] or its layout in memory [29].…”