assembler routines and stack frame generation
lazarus / 67-Windows

----------------------------------------------------------------------
Marladu · 2015-01-21 04:02

Well, I had typed up this as a report but remembered being asked to
write in forums before filing reports so here's what I was going to
report.
---------------
assembler routines won't respect {$OPTIMIZATION STACKFRAME} and
{$STACKFRAMES OFF} but respects nostackframe function modifier
---------------

 In the following code with {$OPTIMIZATION STACKFRAME} and
{$STACKFRAMES OFF}, if I leave commented out the "nostackframe;"
function modifer the following code is generated:

push   %ebp
mov    %esp,%ebp
lea    -0x4(%esp),%esp
add    %edx,%eax
leave 
ret   

 However, if I uncomment the "nostackframe;" function modifer the
following code is generated:

add    %edx,%eax
ret   

 From the documentation of {$STACKFRAMES OFF} it says it won't turn
off stackframes for assembler routines. Alright then so that's why it
doesn't work in this case. But the documentation of {$OPTIMIZATION
STACKFRAME} does not mention what rules it follows.

 In the end if {$OPTIMIZATION STACKFRAME} is completely identical to
{$STACKFRAMES OFF} then it should be mentioned in the documentation,
else if it's supposed to apply the "nostackframe;" to every function
then it's not doing it to assembler routines.

--------------------------

program Project1;

{$APPTYPE GUI}
{$STACKFRAMES OFF}
{$INLINE ON}
{$MODE FPC}
{$ASMMODE INTEL}
{$MMX-}
{$FPUTYPE SSE2}
{$OPTIMIZATION LEVEL3}
{$OPTIMIZATION REGVAR}
{$OPTIMIZATION STACKFRAME}
{$OPTIMIZATION PEEPHOLE}
{$OPTIMIZATION ASMCSE}
{$OPTIMIZATION LOOPUNROLL}
{$OPTIMIZATION TAILREC}
{$OPTIMIZATION CSE}
{$OPTIMIZATION DFA}
{$OPTIMIZATION USEEBP}
{$OPTIMIZATION ORDERFIELDS}
{$OPTIMIZATION FASTMATH}
{$OPTIMIZATION REMOVEEMPTYPROCS}
{$OPTIMIZATION CONSTPROP}

function test_it(a,b:longint):longint;assembler;//nostackframe;
asm
 add eax,edx
end;

procedure test;
begin
 if test_it(10,20)<>30 then
  halt(1);
end;

begin
 test;
end.

----------------------------

 So, the question is, should the "nostackframe" function modifier be
required after specifying {$OPTIMIZATION STACKFRAME} and {$STACKFRAMES
OFF}?

----------------------------------------------------------------------
engkin · 2015-01-21 14:22

Quote from: Marladu on January 21, 2015, 05:02:37 amFrom the
documentation of {$STACKFRAMES OFF} it says it won't turn off
stackframes for assembler routines.
Where did you read that?
According to this, your test_it function should have a stackframe
because it has parameters.

----------------------------------------------------------------------
Marladu · 2015-01-21 15:15

Hi and thanks for your reply.

 In the very same page you linked it says: "If the procedure is not an
assembler procedure, it must not have a asm …end; block.", i figured
that meant that compiler directive doesn't apply to assembler
routines.

 Interestingly enough, changing mode to {$MODE DELPHI} without
changing anything else in the code in the first post (so without
nostackframe function modifier) then no stack frame is generated.
Also, by digging up the compiler source code I found an undocumented
compiler directive {$OPTIMIZATION FORCENOSTACKFRAME} that did remove
that stackframe. I'm guessing it adds the nostackframe function
modifier to every function so should be used carefully.

----------------------------------------------------------------------
engkin · 2015-01-21 15:56

Quote from: Marladu on January 21, 2015, 04:15:36 pmIn the very same
page you linked it says: "If the procedure is not an assembler
procedure, it must not have a asm …end; block.", i figured that meant
that compiler directive doesn't apply to assembler routines.
My understanding is that it should *not* have an asm block *inside* a
pascal procedure, it should be a pure pascal procedure or a pure
assembly procedure. So mixing assembly and pascal is going to cause a
stackframe, as in:
Code: [Select]procedure SomeProc;
begin
..
//The following asm block is going to cause a stackframe:
  asm
    //some assembly
    ..
  end;
..
end;

----------------------------------------------------------------------
Marladu · 2015-01-21 17:09

Ah ok thanks for the clarification, that wasn't what I had understood.

 As for {$OPTIMIZATION FORCENOSTACKFRAME} I made a mistake, it doesn't
seem to do anything at this time, it's probably a work in progress. No
idea why the stack frame in the code presented disappears when using
{$mode delphi} but it does, just like it does in {$mode fpc} with
"nostackframe;" on function implementation.