[Bug target/17622] New: Non-optimal code sequence for floating point "x=0; x=a*b;"

[bangerth at dealii dot org]

This would fix part of PR 17169: 
 
Take this code: 
-------------------- 
double a,b;  
  
double foobar () {  
  double s = 0;  
  s += a * b; 
  return s;  
}  
-------------------- 
The compiler should realize that the two statements can be reduced 
to "s=a*b". However, the compiler does so only when given -ffast-math. 
 
Here's the assembler when compiled with -O3: 
.LC1: 
	.long	0 
	.long	0 
_Z6foobarv: 
	fldl	b 
	fmull	a 
	pushl	%ebp 
	movl	%esp, %ebp 
	popl	%ebp 
	faddl	.LC1 
	ret 
 
And here with -O3 -ffast-math: 
_Z6foobarv: 
	fldl	b 
	fmull	a 
	pushl	%ebp 
	movl	%esp, %ebp 
	popl	%ebp 
	ret 
 
I would argue that the addition of a zero can't change anything 
if we have already determined overflow flags etc during the 
multiplication immediately before. 
 
(A point I'm not quite sure about, because I don't know the least 
about cycle counts etc: wouldn't it be faster to use the  
sequence "fldz ; faddp st,st(1)" rather than "faddl .LC1"? gcc2.95 
used to create that sequence.) 
 
W.

-- 
           Summary: Non-optimal code sequence for floating point "x=0;
                    x=a*b;"
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: bangerth at dealii dot org
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17622