For the testing of memory chips, automatic test equipment (ATE) uses pattern program to generate a bit vector for each clock. Pattern programs are not portable across different ATEs, requiring a different program for each ATE, even when they test the same memory chip. Many solutions and in-house tools have been proposed for this portability problem, but they were not completely successful. This paper proposes PaTran, a software translation platform for pattern programs. PaTran employs intermediate representation (IR) based on the pattern itself, generated by simulating the source pattern program. It synthesizes the target program by reconstructing the program statements from the IR. Since the IR size is huge for product-level programs, PaTran employs a concise form of IR, embedded with repetition information. Implementation of PaTran is based on the web and server-client model.

As web pages and web apps increasingly include heavy JavaScript code, JavaScript performance has been a critical issue. Modern JavaScript engines achieve a remarkable performance by employing tiered-execution architecture based on interpreter, baseline just-in-time compiler (JITC), and optimizing JITC. Unfortunately, they suffer from a substantial compilation overhead, which can take more than 50% of the whole running time. A simple idea to reduce the compilation overhead is ahead-of-time compilation (AOTC), which reuses the code generated in the previous run. In fact, existing studies that reuse the bytecode generated by the interpreter or the machine code generated by the baseline JITC have shown tangible performance benefits [12, 31, 41]. However, there has been no study to reuse the machine code generated by the optimizing JITC, which heavily uses profile-based optimizations, thus not easily reusable. We propose a novel AOTC that can reuse the optimized machine code for high-performance JavaScript engines. Unlike previous AOTCs, we need to resolve a few challenging issues related to reusing profile-based optimized code and relocating dynamic addresses. Our AOTC improves the performance of a commercial JavaScript engine by 6.36 times (max) and 1.99 times (average) for Octane benchmarks, by reducing the compilation overhead and by running the optimized code from the first invocation of functions. It also improves the loading time of six web apps by 1.28 times, on average.