CLA floating point ternary expressions returning negation may generate invalid results for --opt_level=2 or higher when if-conversion optimization is triggered.

Below example code generates incorrect results when compiled at --opt_level=2 or higher
and if-conversion optimization is triggered:
    res[0] = (p > n) ?  p : -n;   
    res[0] = (p > n) ? -p :  n;  
Below if/then cases instead generate correct code due to not if-converting:
   if (p>n) res[0] =  p; else res[0] = -n; 
   if (p>n) res[0] = -p; else res[0] =  n; 

No errors with below assembly without if-conversion:

;*** 10    -----------------------    p = a;
;*** 18    -----------------------    C$1 = b;
;*** 18    -----------------------    if ( p > C$1 ) goto g3;
        MMOV32    MR0,@a 
        MMOV32    MR1,@b  
        MCMPF32   MR0,MR1 
        MNOP 
        MNOP 
        MNOP 
        MBCNDD    $C$L1,GT 
        MNOP 
        MNOP 
        MNOP 

Bug occurs with below assembly with if-conversion conditional execution:

;*** 10    -----------------------    p = a;
;*** 12    -----------------------    C$2 = b;
;*** 12    -----------------------    (p > C$2) ? (S$1 = p) : (S$1 = -C$2);
;*** 12    -----------------------    res[0] = S$1;
;***      -----------------------    return;
        MMOV32    MR0,@a   
        MMOV32    MR1,@b   
        MCMPF32   MR0,MR1  
        MNEGF32   MR0,MR1,LEQ ; conditional execution from if-conversion 
        MMOV32    @res,MR0   
        MNOP    
        MNOP