Re: failure due to compiler?

        |> I am looking for pointers or references to descriptions where software
        |> has compiled without error and later failed due to compiler-introduced
        |> errors in the resultant program. In other words, the source was deemed
        |> to be error free but the compiler botched the code generation.


Some old x86 compilers took


      if( b << c )


and generated:


    move bx, b
    move cl, c
    shift bx, cl
    jne ...


Problem: if the shift count c were 0 then the shift was defined to be
a NOP - *NOT* modifying the flags.


Unless the compiler can prove that c != 0, then the code emmitted must
be:


    move bx, b
    move cl, c
    shift bx, cl
    tst bx,bx ; compare to 0
    jne ...


This bug bit me in a simulator compiled with aforementioned compiler.


---


More apocryphally, I remember spending many days looking for bugs in
my NAPLPS kiosk database code, that turned out to be compiler bugs.
Don't have them now, of course.


---


On a more particular situation that is painfully close to home, does
anyone remember that Intel accidentally quoted SPECint95 numbers for
the P6, ah, Pentium Pro processor, that were about 10% too high? Yes,
it was really and truly due to a compiler bug in code generation for
EQNTOTT.
        Does being embarassed, and having to redo and republish benchmark
results, count as a failure?
---
Andy "Krazy" Glew, glew@ichips.intel.com, Intel,
M/S JF3-206, 2111 NE 25th Ave., Hillsboro, OR 97124
Place URGENT in email subject line for mail filter prioritization.
--

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