| 摘要 |
Compilers are tools that generate efficient mappings from programs to machines. A Java "Just-In-Time" runs as part of an application, and as such, it must be fast and efficient in its use of memory. To achieve good performance and further optimize code generation, the present invention introduces a method for eliminating common subexpressions from Java bytecodes. The method of the present invention first loads a code stream containing sequences of computer code into computer memory. The expression value for a first expression of a first code sequence is computed and stored in a memory location. A tag is assigned to the memory location holding this expression value for tracking which expression sequences' values are held in memory locations. As code compilation continues, the code selector looks ahead in the code stream to see if any upcoming expression sequences already have expression values stored in a memory location. The code selector compares the expression of a second code sequence with the code sequences annotated by the tags of expression values currently stored in memory. If the second code sequence matches a sequence already associated with a memory location, then the value of the matched sequence is pushed from the memory location onto a stack, and the computations of the expression of the second code sequence is skipped. If the second expression does not match any of the expressions represented by the tags, the expression value of the second expression is calculated and stored in a memory location. This memory location is then annotated with its own expression tag for future comparisons with upcoming expressions in the code stream.
|
