Optimization flags not working (GCC 4.2.4)

Optimization flags not working (GCC 4.2.4)

[slnc]

I have a file called "main.c":

-- main.c ------------------------------------------
int main(void) {
	int i, j;

	for (i = 0; i < 10; i ++)
		j = 5;

	return 0;
}
----------------------------------------------------

When I compile it with "gcc -O1 -c -o main.o main.c" and I do an 
"objdump -d main.o" I get:

------------------------------------------------------
main.o:     file format elf64-x86-64

Disassembly of section .text:

0000000000000000 <main>:
    0:	31 c0                	xor    %eax,%eax
    2:	c3                   	retq
------------------------------------------------------


But when I try to use just only one optimization flag by compiling with 
"gcc -ftree-dce -c -o main.o main.c" the optimization flag seems to do 
nothing at all:

------------------------------------------------------
main.o:     file format elf64-x86-64

Disassembly of section .text:

0000000000000000 <main>:
    0:	55                   	push   %rbp
    1:	48 89 e5             	mov    %rsp,%rbp
    4:	c7 45 fc 00 00 00 00 	movl   $0x0,-0x4(%rbp)
    b:	eb 0b                	jmp    18 <main+0x18>
    d:	c7 45 f8 05 00 00 00 	movl   $0x5,-0x8(%rbp)
   14:	83 45 fc 01          	addl   $0x1,-0x4(%rbp)
   18:	83 7d fc 09          	cmpl   $0x9,-0x4(%rbp)
   1c:	7e ef                	jle    d <main+0xd>
   1e:	b8 00 00 00 00       	mov    $0x0,%eax
   23:	c9                   	leaveq
   24:	c3                   	retq
------------------------------------------------------


And even if I specify all the optimization flags that are enabled by -O1 
I get the same previous objdump. What am I doing wrong?

Thank you,

-- 
Juan Alonso

RE: Optimization flags not working (GCC 4.2.4)

[John (Eljay) Love-Jensen]

Hi Juan,

If you are not optimizing (-O0), the optimization -f flags have no effect.

HTH,
--Eljay

Re: Optimization flags not working (GCC 4.2.4)

[slnc]

Hi John,

Thank you but -O0 makes no difference.

gcc -c -g -O0 -fdefer-pop -fdelayed-branch -fguess-branch-probability 
-fcprop-registers -fif-conversion -fif-conversion2 -ftree-ccp -ftree-dce 
-ftree-dominator-opts -ftree-dse -ftree-ter -ftree-lrs -ftree-sra 
-ftree-copyrename -ftree-fre -ftree-ch -funit-at-a-time 
-fmerge-constants main.c

(those are all gcc 4.2.4 optimization flags for level -O1 according to 
the docs) generates the same unoptimized code. Changing -O0 with -O1 
generates the optimized version.

John (Eljay) Love-Jensen wrote:
> Hi Juan,
> 
> If you are not optimizing (-O0), the optimization -f flags have no effect.
> 
> HTH,
> --Eljay
> 


-- 
Juan Alonso

Re: Optimization flags not working (GCC 4.2.4)

[David Daney]

slnc wrote:
> Hi John,
> 
> Thank you but -O0 makes no difference.
> 

That's what he just said.  If you use -O0 (the implied default) none of 
the optimizations are done.  Use -O1 (or -O2 or -O3) if you want 
optimizations done.

David Daney

> gcc -c -g -O0 -fdefer-pop -fdelayed-branch -fguess-branch-probability 
> -fcprop-registers -fif-conversion -fif-conversion2 -ftree-ccp -ftree-dce 
> -ftree-dominator-opts -ftree-dse -ftree-ter -ftree-lrs -ftree-sra 
> -ftree-copyrename -ftree-fre -ftree-ch -funit-at-a-time 
> -fmerge-constants main.c
> 
> (those are all gcc 4.2.4 optimization flags for level -O1 according to 
> the docs) generates the same unoptimized code. Changing -O0 with -O1 
> generates the optimized version.
> 
> John (Eljay) Love-Jensen wrote:
>> Hi Juan,
>>
>> If you are not optimizing (-O0), the optimization -f flags have no 
>> effect.
>>
>> HTH,
>> --Eljay
>>
> 
> 

Re: Optimization flags not working (GCC 4.2.4)

[slnc]

David Daney wrote:
> slnc wrote:
>> Hi John,
>>
>> Thank you but -O0 makes no difference.
>>
> 
> That's what he just said.  If you use -O0 (the implied default) none of 
> the optimizations are done.  Use -O1 (or -O2 or -O3) if you want 
> optimizations done.

Oh, I misunderstood him, sorry. It also doesn't work if I omit 
-O<anything>. So to sum it up, there is no way to compile a program in 
gcc enabling only one specific optimization flag. The only option is to 
use the "pre-packaged" sets of flags (-O1, -O2 or -O3) right?

I have tried disabling all -O1 flags just to check if I could select 
only one by disabling all the others but:

gcc -c -O1 -fno-defer-pop -fno-delayed-branch 
-fno-guess-branch-probability -fno-cprop-registers -fno-if-conversion 
-fno-if-conversion2 -fno-tree-ccp -fno-tree-dce -fno-tree-dominator-opts 
-fno-tree-dse -fno-tree-ter -fno-tree-lrs -fno-tree-sra 
-fno-tree-copyrename -fno-tree-fre -fno-tree-ch -fno-unit-at-a-time 
-fno-merge-constants main.c
objdump -d dce.o

generates an optimized version. So why are the flags there if they 
cannot be selected individually? (I feel like I am missing something 
completely obvious..)

Thank you,
-- 
slnc

Re: Optimization flags not working (GCC 4.2.4)

[Robert William Fuller]

slnc wrote:
> David Daney wrote:
>> slnc wrote:
>>> Hi John,
>>>
>>> Thank you but -O0 makes no difference.
>>>
>>
>> That's what he just said.  If you use -O0 (the implied default) none
>> of the optimizations are done.  Use -O1 (or -O2 or -O3) if you want
>> optimizations done.
> 
> Oh, I misunderstood him, sorry. It also doesn't work if I omit
> -O<anything>. So to sum it up, there is no way to compile a program in
> gcc enabling only one specific optimization flag. The only option is to
> use the "pre-packaged" sets of flags (-O1, -O2 or -O3) right?
> 
> I have tried disabling all -O1 flags just to check if I could select
> only one by disabling all the others but:
> 
> gcc -c -O1 -fno-defer-pop -fno-delayed-branch
> -fno-guess-branch-probability -fno-cprop-registers -fno-if-conversion
> -fno-if-conversion2 -fno-tree-ccp -fno-tree-dce -fno-tree-dominator-opts
> -fno-tree-dse -fno-tree-ter -fno-tree-lrs -fno-tree-sra
> -fno-tree-copyrename -fno-tree-fre -fno-tree-ch -fno-unit-at-a-time
> -fno-merge-constants main.c
> objdump -d dce.o
> 
> generates an optimized version. So why are the flags there if they
> cannot be selected individually? (I feel like I am missing something
> completely obvious..)
> 
> Thank you,

The conceptual model is that the flags allow you to tweak the
optimizations selected by various optimization regimes (aka optimization
levels.) First, you specify an optimization regime such as -O1, -O2, or
-O3. Then, if you do not like a specific optimization, such as
-fomit-frame-pointer, which often makes it more difficult to debug
programs, then you can remove it with -fno-omit-frame-pointer.
Conversely, if you want a more aggressive optimization, without
increasing the optimization regime to a more aggressive regime, you can
add that specific optimization. So, if I wanted an optimization from
-O2, but did not want to increase my optimization regime to -O2 from
-O1, I can specify just that optimization. However, you cannot tweak the
optimizations selected by an optimization regime if you do not tell the
compiler to optimize!

Re: Optimization flags not working (GCC 4.2.4)

[Ian Lance Taylor]

slnc <s@slnc.me> writes:

> I have tried disabling all -O1 flags just to check if I could select
> only one by disabling all the others but:
>
> gcc -c -O1 -fno-defer-pop -fno-delayed-branch
> -fno-guess-branch-probability -fno-cprop-registers -fno-if-conversion
> -fno-if-conversion2 -fno-tree-ccp -fno-tree-dce
> -fno-tree-dominator-opts -fno-tree-dse -fno-tree-ter -fno-tree-lrs
> -fno-tree-sra -fno-tree-copyrename -fno-tree-fre -fno-tree-ch
> -fno-unit-at-a-time -fno-merge-constants main.c
> objdump -d dce.o
>
> generates an optimized version. So why are the flags there if they
> cannot be selected individually? (I feel like I am missing something
> completely obvious..)

One view is that they can be selected individually.  However, there are
many optimizations which are not controlled by any flag.  And if you
don't turn on optimization, then gcc won't build the datastructures
needed for most of the optimization passes, so they won't be run even if
you explicitly ask for them.

Another view is that optimizations are not independent.  There is basic
work that has to be done for every optimization, like build the CFG, go
into SSA form, build the dataflow information, etc.  The main difference
between -O0 and -O1 is whether or not this basic work is done.

Ian