gcc-2.95.3.test4 miscompiles calc-2.11.4t2 on PowerPC

[Martin Buck]

gcc 2.95.3.test4 seems to miscompile calc-2.11.4t2 (an arbitrary precision
calculator with a really useful testsuite) on PowerPC. calc's testsuite
("make chk") fails, if calc's source file zfunc.c is compiled with -O2.
Going back to -O1 or turning off optimization completely fixes the problem.
I tracked down the problem to be in function zsqrt(), variable d. If you
make d volatile, the errors go away:

*** zfunc.c.orig        Fri Feb 23 18:18:39 2001
--- zfunc.c     Fri Feb 23 18:39:33 2001
***************
*** 1481,1487 ****
  {
        HALF *a, *A, *b, *a0, u;
        int i, j, j1, j2, k, k1, m, m0, m1, n, n0, o;
!       FULL d, e, f, g, h, s, t, x, topbit;
        int remsign;
        BOOL up, onebit;
        ZVALUE sqrt;
--- 1481,1488 ----
  {
        HALF *a, *A, *b, *a0, u;
        int i, j, j1, j2, k, k1, m, m0, m1, n, n0, o;
!       volatile FULL d;
!       FULL e, f, g, h, s, t, x, topbit;
        int remsign;
        BOOL up, onebit;
        ZVALUE sqrt;


Sorry, but I could not reduce this to a simpler test case. zsqrt() is a bit
hard to grok for me.

I've seen this problem only on PowerPC, but it also happens with
gcc-2.95.2.

BTW, calc is at http://www.isthe.com/chongo/tech/comp/calc/

Thanks,
Martin
-- 
Martin Buck               E-mail: m@rtin-buck.de  or  martin.buck@bigfoot.com
Phone/Fax: +41-56-4963841   Snail-mail: Steigstrasse 13, CH-5452 Oberrohrdorf
GPG key available                           WWW: http://home.pages.de/~mbuck/