[Open64-devel] Re: -OPT flags

[Peng Z. <pen...@cs...>]

Also, If I don't use skip_equal=0 and change it to -Wb,-tr37, the 
executable works fine too.

This is so strange: some unrelated options really influence the generated 
code.


On Mon, 17 Feb 2003, Peng Zhao wrote:

> Hi,
> 
> 	I encountered a very strange problem.
> 
> 	I am debugging the compiler.
> 
> 	In GAP benchmark, all other object files are generated with -O0. 
> And I am debugging a buggy behavior of ORC on statemen.c.
> 1.	When I use 
> orcc  -c statemen.c -o statemen.o    -O2 -DSYS_IS_USG 
> -DSYS_HAS_IOCTL_PROTO -DSYS_HAS_TIME_PROTO -DSYS_HAS_SIGNAL_PROTO 
> -DSYS_HAS_ANSI -DSYS_HAS_CALLOC_PROTO -DSPEC_CPU2000_LP64  -lm  -fb_opt 
> fb.mid -Wb,-OPT:outl=1 -noinline  -v -keep -Wb,-tt72:2,-tt72:4,-tt72:0x80  
> -O2     -Wb,-OPT:skip_before=27:skip_after=27
> 
> 
> 	the generated executable fails (no core dump, just its semantics 
> is changed)
> 
> 2.	But when I use 
> orcc  -c statemen.c -o statemen.o    -O2 -DSYS_IS_USG 
> -DSYS_HAS_IOCTL_PROTO -DSYS_HAS_TIME_PROTO -DSYS_HAS_SIGNAL_PROTO 
> -DSYS_HAS_ANSI -DSYS_HAS_CALLOC_PROTO -DSPEC_CPU2000_LP64  -lm  -fb_opt 
> fb.mid -Wb,-OPT:outl=1 -noinline  -v -keep -Wb,-tt72:2,-tt72:4,-tt72:0x80  
> -O2     -Wb,-OPT:skip_before=27:skip_after=27:skip_equal=0
> 
> 	the generated executable works fine.
> 
> 	For your convenience, the only difference between the two commands 
> are only the ":skip_equal=0", which appears in the second command but not 
> in the first one.
> 
> 	Actually, skip_before=27:skip_after=27 has covered skip_equal=0, 
> right? I just noticed this accidently.
> 
> 
> 	And when I use 
> 
> 	orcc  -c statemen.c -o statemen.o    -O2 -DSYS_IS_USG 
> -DSYS_HAS_IOCTL_PROTO -DSYS_HAS_TIME_PROTO -DSYS_HAS_SIGNAL_PROTO 
> -DSYS_HAS_ANSI -DSYS_HAS_CALLOC_PROTO -DSPEC_CPU2000_LP64  -lm  -fb_opt 
> fb.mid -Wb,-OPT:outl=1 -noinline  -v -keep -Wb,-tt72:2,-tt72:4,-tt72:0x80  
> -O2     -Wb,-OPT:skip_before=27,-OPT:skip_after=27 
> -Wb,-PHASE:c=off:w=off:l=off:p=on
> 
> 	and then use whirl2c to generate the N file back to C file. Then 
> compile the W2C file to statemen.o. The linked executable also works very 
> well. I cannot find anything wrong from the W2C file. Seems that I cannot 
> find the wrong with the IR file.
> 
> 
> 
> 	Any idea?
> 						Peng
> 
> 	
> 
> 

-- 
                         	            |    |    |
Peng Zhao   pen...@cs...	   )_)  )_)  )_)
http://www.cs.ualberta.ca/~pengzhao       )___))___))___)
Phone: 1-780-492-3725	    		 )____)____)_____)\\
Fax:   1-780-492-1071	            ________|____|____|____\\\__
Lab:   CSC-251			-------\                   /---------