Optimisation from assembler output

This is probably a dumb questions but.... Is it sensible to optimise
simplistic assembler output from a compiler rather than build a
complex optimiser into the compiler? If so does one exist for SPARC?


The question is prompted by recent examination of the assembler
output from the g++ compiler. I was disappointed as I could, without
much effort, greatly improve the code due to poor register tracking,
missed common subexpression elimination and worst of all pointless
introduction of temporaries (constructors). If these could be
eliminated I think there would be around half? the instructions.


Any help much appreciated


Kevin
[Depends on what kinds of optimizations you want to do. Peephole opts are
always done on something pretty close to assembler, other kinds of opts need a
higher level representation. For G++ I'd think it'd be better to work at
a higher level than assembler so the work you do is applicable to more than
one platform. -John]
--

Return to the comp.compilers page.
Search the comp.compilers archives again.