Experience in fine-tuning a compiler generated from an attribute grammar

Compilers are one of the tools that are used daily by software developers. Generating efficient compilers is thus important for compiler generators such as attribute grammar (AG) systems. This paper presents our experience at fine-tuning a compiler generated from an AG. The techniques adopted for fine-tuning include memory management, hash symbol tables, and function inline. These techniques have been widely used in the development of hand-coded compilers, but are rarely applied in AG generators. The speedup obtained in the attribute evaluation is 2.80, and the overall speedup is 1.73. These results indicate that the compilers generated from AGs can be improved considerably with the above techniques. They also indicate a new direction in optimization techniques for generating compilers from AGs.