aix4.3 bootstrap fails with snapshot 20000612
David Edelsohn
dje@watson.ibm.com
Wed Jun 14 12:48:00 GMT 2000
Definitely something screwy with the TOC address of the base of
the jump table. It looks like the new code looking for duplicate TOC
entries is doing something very bad.
read_skip_spaces() tries to load the base of the TOC table from
local symbol LC..313:
lwz 11,LC..313(2)
Just before the function, LC..313 is equated with LC..276:
.set LC..313,LC..276
LC..276 is the label for the TOC entry of label L..33:
LC..276:
.tc L..33[TC],L..33
However, L..33 is the local label for the jump table used by function
copy_rtx()!
read_skip_spaces() and a number of other functions are trying to
collapse the TOC entries, but they refer to an jump table located earlier
in the file which is wrong and the jump computation goes off the rails
jumping to some very large address like 0x70002f5c.
I do not understand Geoff's new TOC code with hash tables well
enough to debug this. I do note that when compiling some of the support
functions in rs6000.c with warnings, GCC complains about prototype
mismatches and one argument does not have the correct number of pointer
indirections.
David
More information about the Gcc-bugs
mailing list