Re: bug in -O -finline-functions Solaris 2.6 egcs1.1b g++

[andtheworldandtheworld <aghosh at chem dot columbia dot edu>]

	I am forwarding this from earlier as it got rejected by the
mail deamon because of the file size. The actual input file/test
case can be found via anonymous ftp at

elroy.chem.columbia.edu 

in the directory /pub/aghosh/sgb2.tar.gz

> The version of egcs I'm using is 
> 
> egcs version
> 
> /egcs-2.91.57/specs
> gcc version egcs-2.91.57 19980901 (egcs-1.1 release)
> 
> --------------------------------------------
> How installed :: standard install
> 
> ./configure
> make bootstrap
> make install
> 
> --------------------------------------------
> 
> 	I believe I have found a bug in a mixture of -O and
> -finline-functions w/ the egcs 1.1b g++ compiler. This did not
> show up w/ egcs1.02 
> 
> 	I've attached a Makefile and the source code and some
> parameter files for the code in tar.gz form.
> 
> 	The program will untar itself in sgb2/. The makefile 
> make all will create the executable in that directory. 
> 
> --- Running the program-------------------------------
> 
> To run the program set the environmental variable Sgb_Dir to
> the directory that directory
> 
> i.e 
> 
> export Sgb_Dir=~/sgb2
> 
> and to run it 
> 
> ./sgbmain
> 
> ----
> 
> 	THe problem occurs in line 2537 of the code. (I merged
> the entire code into one for ease for sending).
> 
> inside_mike_beachy=0;
> 
> for(j=0;0 == inside_mike_beachy;j++)
>     {                           //  loop over neighbors
>     printf("THis is a line\n");
>     printf("This is another line\n");
>     printf("Ho hum\n");
>     printf("THe inside on the outside is %d\n",inside_mike_beachy);
> 
> 
> What happens is that inside_mike_beachy gets assigned to the value
> 1 during the comparision statement of the for loop. The disassembly
> of the code yields the following ::
> 
> 
> Using -g and ddd data display debugger.
> 
> Dump of assembler code from 0x8055b00 to 0x8055c00:
>     0x8055b00 <check_set_point+12>:	movl   $0x0,0xffffff90(%ebp)
>     0x8055b07 <check_set_point+19>:	movl $0x40f86a00,0xffffff94(%ebp)
>     0x8055b0e <check_set_point+26>:	movl   $0x0,0xffffff88(%ebp)
>     0x8055b15 <check_set_point+33>:	movl $0x3ff00000,0xffffff8c(%ebp)
>     0x8055b1c <check_set_point+40>:	movl   $0x0,0xffffff98(%ebp)
>     0x8055b23 <check_set_point+47>:	xorl   %edi,%edi
>     0x8055b25 <check_set_point+49>:	movl   $0x0,0xffffff80(%ebp)
>     0x8055b2c <check_set_point+56>:	testl  %edi,%edi
>     0x8055b2e <check_set_point+58>:	jne    0x8055e9c <check_set_point+936>
>     0x8055b34 <check_set_point+64>:	leal   0xffffff98(%ebp),%eax
>     0x8055b37 <check_set_point+67>:	movl   %eax,0xffffff7c(%ebp)
>     0x8055b3d <check_set_point+73>:	leal   0xffffff90(%ebp),%ecx
>     0x8055b40 <check_set_point+76>:	movl   %ecx,0xffffff78(%ebp)
>     0x8055b46 <check_set_point+82>:	leal   0xffffff88(%ebp),%eax
>     0x8055b49 <check_set_point+85>:	movl   %eax,0xffffff74(%ebp)
>     0x8055b4f <check_set_point+91>:	leal   0xffffff84(%ebp),%ecx
>     0x8055b52 <check_set_point+94>:	movl   %ecx,0xffffff70(%ebp)
>     0x8055b58 <check_set_point+100>:	movl   $0x1,%edi
>     0x8055b5d <check_set_point+105>:	nop    
>     0x8055b5e <check_set_point+106>:	nop    
> 
> 
> This is the dump of that particular routine sua_t::check_set_point up to  
> and including the for statement. The last instruction at 0x8055b5d before
> the nops is the one that is  assigning the value of inside_mike_beachy to
> be 1.
> 
> 	I hope this is enough information but let me know if you would
> like anything else w/ regard to this problem ..Or if it is
> patchable/fixable :)
> 
> 
> 	-me
> 
> 
> 
> 
> 
>