Problems with -mregparm=2 and unsigned long long

Lars Hamren hamren@sdu.se
Fri Oct 29 15:48:00 GMT 2004


This program

	int main()
	{
	    unsigned long      high  = 0x0013ABC;
            unsigned long      low   = 0xB388B94A;
	    unsigned long long time1 = (((unsigned long long)high << 32L) + low);
	    unsigned long long time  = time1 / 450;
	    return time == 0;
	}


will return different values depending on the -mregparm compiler
option. With a value of zero or one it will compute the correct
result, but with a value of two or three the value of `time' will be
zero and the return code will be one:

	$ for i in 0 1 2 3 ; do gcc34 rdtsc2.c -o rdtsc2 -mregparm=$i; ./rdtsc2 ; echo $? ; done;
	0
	0
	1
	1

This may be because the compiler generates calls to __udivdi3
uses parameter passing in register, but linking is done with libraries
that use stack parameter passing.

Extracting __udivdi3 from glibc and including it in the source code
will solve the problem.

This problem is present in both gcc 3.3.3 and gcc 3.4 on Fedora
Core 2. I have not tried later versions of gcc.

	$ gcc -v
	Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/3.3.3/specs
	Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --enable-shared --enable-threads=posix --disable-checking --disable-libunwind-exceptions --with-system-zlib --enable-__cxa_atexit --host=i386-redhat-linux
	Thread model: posix
	gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)

This problem surfaced when I tried using `unsigned long long' in a Linux
kernel driver. The kernel is compiled with a high value for -mregparm.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: rdtsc2.i
Type: application/octet-stream
Size: 295 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-bugs/attachments/20041029/60868941/attachment.obj>
-------------- next part --------------


Regards
/Lars
-------------------------------------------------------------------
Lars Hamr??n              Tel...: +46(46)189090
Svensk Datorutveckling   e-post: hamren@sdu.se
Vadm??llan 211            WWW...: www.sdu.se
S-225 94 Lund
Sweden


More information about the Gcc-bugs mailing list