Re: optimization/8872: Constant propagation and conditional statements

Re: optimization/8872: Constant propagation and conditional statements

[reichelt]

Synopsis: Constant propagation and conditional statements

State-Changed-From-To: open->closed
State-Changed-By: reichelt
State-Changed-When: Sun Dec  8 13:40:21 2002
State-Changed-Why:
    Fixed in mainline.
    
    With gcc 3.3-20021118 I get the following assembler output,
    when optimizing with -O2:
    
    foo:
    	pushl	%ebp
    	movl	$1, %eax
    	movl	%esp, %ebp
    	popl	%ebp
    	ret
    
    That's what you expected, right?
    
    Since the bad optimization in gcc 3.2.1 is no regression,
    the fix will most likely not be backported for gcc 3.2.2.
    
    But thanks for spotting this problem.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=8872

optimization/8872: Constant propagation and conditional statements

[s.chauveau]

>Number:         8872
>Category:       optimization
>Synopsis:       Constant propagation and conditional statements
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Dec 08 02:56:01 PST 2002
>Closed-Date:
>Last-Modified:
>Originator:     s.chauveau@chello.nl
>Release:        gcc 3.2.1
>Organization:
>Environment:
Debian ix86
>Description:
Gcc seems to have problem propagating constants 
when they are defined within a condition. Only a 
few levels of constants are propagated. 
Consider the following code: 

int foo(void)
{
  int a = 0 ; 
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  if (a==0) a=1 ;  else if (a==1) a=0 ;
  return a ;
}

A good optimiser should be able to reduce this code
to a single constant (0 or 1). Gcc manages to reduce 
the first 4 conditional statements. The rest is not
simplified and so generates a very poor assembly code 
(with tests and jumps).

>How-To-Repeat:

>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: