Comparisons with zero
lazarus / 64-FPC development

----------------------------------------------------------------------
lagprogramming · 2015-11-21 10:19

variable_in_register=0 and variable_in_register<>0 comparisons are
optimized to use "test" instead of "cmp" based on the zero flag. This
means we have something like:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i if
variable_in_register<>0 then variable_in_register:=0;testq  
%rax,%raxje      .Lj6movq    $0,%rax

When comparing for sign, fpc produces the following code:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i if
variable_in_register<0 then variable_in_register:=0;cmpq    $0,%raxjnl
    .Lj6movq    $0,%raxWhen dealing with "<0" and ">=0", I'd like
"test" to be used instead of "cmp", too.

I've looked at compiler/i386/popt386.pas and I've seen:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i           
  A_CMP:                begin                  if
(taicpu(p).oper[0]^.typ = top_const) and                   
 (taicpu(p).oper[0]^.val = 0) and                   
 (taicpu(p).oper[1]^.typ = top_reg) then                   {change
"cmp $0, %reg" to "test %reg, %reg"}                    begin        
            taicpu(p).opcode := A_TEST;                     
taicpu(p).loadreg(0,taicpu(p).oper[1]^.reg);                     
continue;                    end;                end; So, I've said to
myself that the "cmp" should have already been changed with "test".
Paying further attention I've noticed that "jnl" would not work with
"test" because "jnl" involves the overflow flag.
Now, I'm thinking that maybe if "jnl" would be replaced with "jns" at
variable_in_register<0 then the peephole optimizer might do the job.
If I'm on the right track, where do I have to look further?

   P.S. It should be obvious to you that I'm not experienced with
compiler internals and I couldn't find documentation regarding how the
compiler optimizes the code.