Re: Effectiveness of compilers today

>Could someone tell me more about this. Are there any tests between common
>compilers and assembler programmers. Who is the best?


>[This has indeed been discussed to death in the past. My impression is that
>for a small chunk of code, a human can almost always meet or beat a compiler,
>but for large chunks people aren't as persistent as compilers at optimizing


Assembly programmers will never have job security! Take a look at the GNU
superoptimizer ---


GSO is a function sequence generator that uses an exhaustive
generate-and-test approach to find the shortest instruction sequence for a
given function. You have to tell the superoptimizer which function and
which CPU you want to get code for.


Sample GSO results:


    SPARC code for i5 = ((signed) i0 > (signed) i1):
                    subcc %i1,%i0,%i2
                    srl %i1,31,%i3
                    sra %i0,31,%i4
                    addx %i4,%i3,%i5


    Faster 80x86 code than the standard cmpl/seta/and for
    ecx = ((unsigned) eax > (unsigned) edx):
                    cmpl %eax,%edx
                    sbbl %ecx,%ecx
                    negl %ecx


    MC88000 code for 'find first set bit', i.e., r6 = ffs(r2):
                    subu.co r3,r0,r2
                    and r4,r3,r2
                    addu.ci r5,r4,r4
                    ff1 r6,r5


    POWER (RS/6000) code for r6 = ((unsigned) r3 < (unsigned) r4) + r5
                    sf r6,r4,r3
                    sfe r7,r5,r6
                    sf r8,r7,r6


Aide from the CPUs in the examples above, GSO generates code for
MC680x0, Amd290xx, Pyramid(SP,AP,XP), and Alpha.


joshp@cmu.edu
--

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