Bridging the Compiler-Simulator Gap: Faster and Easier Hardware/Software Optimization

要让计算机运行得更快，需要优化利用硬件的软件。探索新的硬件想法需要硬件模拟器，并优化与想象的硬件相匹配的软件。构建它们并使它们匹配既耗时又容易出错。这个令人信服的项目是自动构建模拟器和匹配的软件工具，以便更多的人可以更快地为现代计算机和编程语言探索更多的想法。CoGenT，代表联合生成(模拟和编译)工具，带有机器描述。根据指令格式、执行和时序信息，我们在所描述的硬件上为Java程序生成了预测模拟器。Java系统包括优化编译器和非优化编译器，我们将这两个编译器都重定为目标硬件。我们通过使用与硬件目标相同的编译器抽象机的描述形式来实现这一点。我们通过搜索目标指令来匹配每个可能的抽象代码片段来生成代码生成规则。我们还关注编译器后端的自动构建，以及我们构建的模拟器的效率。虽然某些硬件创新显然需要全新的技术来有效地利用它们，但我们自动化了现有优化编译器中涉及的那些方面。我们的贡献在于自动化程度的提高和技术的集成化，以及模拟器性能的改善。虽然主要设置是模拟，但结果在编译器重定目标、动态二进制翻译等方面很有用。

Making computers go fast requires optimizing software that exploits thehardware. Exploring new hardware ideas requires hardware simulators, ANDoptimizing software matching the imagined hardware. It is time-consumingand error-prone to build these and get them to match. The CoGenT project isautomating the building of simulators and matching software tools, so thatmore people can explore more ideas, more quickly, for modern computers andprogramming languages.CoGenT, which stands for Co-Generation of (simulation and compilation)Tools, with machine descriptions. From instruction format, execution, andtiming information, we produce predictive simulators for Java programs onthe described hardware. The Java system includes optimizing andnon-optimizing compilers, and we retarget both compilers to the targethardware. We do this by using the same description forms for the compiler'sabstract machine as for the hardware target. We generate code generationrules by searching for target instructions to match each possible abstractcode fragment. We are also concerned with automated construction of therest of the compiler back end, and with efficiency of the simulators we build.While certain hardware innovations clearly demand entirely new techniquesto exploit them effectively, we automate those aspects addressed in theexisting optimizing compiler. Our contributions lie in the increased degreeof automation and integration of the techniques, and in improved simulatorperformance. While the primary setting is simulation, the results areuseful in compiler retargeting, dynamic binary translation, and beyond.