|
From: Julian S. <js...@ac...> - 2006-04-04 13:31:21
|
Well, I can't reproduce this, on a R H 7.3 box, but that's with the default 2.96 gcc, and IIRC you're using gcc 3.0.X ? Could you send me the executable? J On Tuesday 04 April 2006 09:41, Tom Hughes wrote: > In message <200...@ac...> > > Julian Seward <js...@ac...> wrote: > > On Tuesday 04 April 2006 03:32, Tom Hughes wrote: > >> Nightly build on alvis ( i686, Red Hat 7.3 ) > >> > >> memcheck/tests/sh-mem-random (stdout) > >> memcheck/tests/sh-mem-random (stderr) > > > > That's not good. Tom, can you send the diffs? > > The output is: > > -------- testing non-auxmap range -------- > initialising > post-initialisation check > test passed, sum = 38338686 (127.79562 per byte) > doing copies > final check > > > FAILURE: final check, byte 82 -- is 0xf6, should be 0xfe > > I think the problem is in the eight byte floating point copies as > suppressing those makes it pass. The disassembly for the floating > point copy on that machine is: > > 8048b3a: 8b 45 d4 mov 0xffffffd4(%ebp),%eax > 8048b3d: 8b 55 08 mov 0x8(%ebp),%edx > 8048b40: 01 c2 add %eax,%edx > 8048b42: 8b 45 d8 mov 0xffffffd8(%ebp),%eax > 8048b45: 03 45 08 add 0x8(%ebp),%eax > 8048b48: dd 00 fldl (%eax) > 8048b4a: dd 1a fstpl (%edx) > > Tom |