what is the problem with the profile optimization?

["Chen Ken" <chenken at mprc dot pku dot edu dot cn>]

When I used the gcc to compile the compress92.c with -fprofile-arcs -O
option, and ran the compress, then recompiled the compress92.c with
-fbranch-probabilities -O option, it reported

${HOME}/usr/bin/gcc -fprofile-arcs -O -o compress compress.c
./compress tarfile
${HOME}/usr/bin/gcc -fbranch-probabilities -O -o compress1 compress.c
compress.c: In function `main':
compress.c:438: warning: assignment from incompatible pointer type
compress.c:438: warning: comparison of distinct pointer types lacks a cast
compress.c:439: warning: passing arg 2 of `signal' from incompatible pointer
type
compress.c:440: warning: passing arg 2 of `signal' from incompatible pointer
type
compress.c:717: corrupted profile info: prob for 15-16 thought to be -10000
compress.c:717: corrupted profile info: prob for 15-41 thought to be 20000
compress.c:717: corrupted profile info: prob for 62--2 thought to be 20000
compress.c:717: corrupted profile info: prob for 62-63 thought to be -10000
compress.c:717: corrupted profile info: prob for 127--2 thought to be 40000
compress.c:717: corrupted profile info: prob for 127-128 thought to
be -30000
${HOME}/usr/include/sys/stat.h: At top level:
${HOME}/usr/include/sys/stat.h:367: .da file contents not exhausted

And, I found different version of gcc will not give the same error. The gcc
I
used above was built by myself(gcc-3.2.1). And the gcc in Redhat Linux8.0
will
not report the error when add -O option. However, my own gcc will not report
error when use the -O2 option instead, and the gcc in Redhat Linux8.0 will
report. I don't know how this error was produced, and how can I check and
debug
with the .da file?


yours sincerely,
Chen Ken
Microprocessor Research & Development Center, Peking University
Tel: +86(010)62765828  ext. 872
chenken at mprc dot pku dot edu dot cn