Compilers: Principles and Practice by Pearson

Code Generation and Machine-dependent Optimization 275

movl a, %eax

addl b, %eax

imull c, %eax

movl %eax, %edx

idiv d

movl %edx, t

9.1.5 Order of Evaluation Sequence

In a long numerical expression, we may find that some particular execution order results in a more

efficient target code, at least in terms of register usage. Most of the modern CPUs are pipelined, i.e.

execute several instructions in parallel. This opens up a possibility of considering the evaluation

order in depth, on the one hand, and the correct computation must be achieved, on the other hand;

reordering of instructions will result in faster code. We postpone the discussion of this ...