30

I recently received a bug report about a program which fails to compile with the -O3 switch (see https://github.com/cschwan/sage-on-gentoo/issues/66). In particular, the problem is that compilation hangs at a certain point. The issue is solved by compiling with -O2 (I am well aware of the fact that programs compiled with -O3 may be broken, but I did not know that -O3 may hangup the compiler). If you want to reproduce the issue run

wget http://perso.ens-lyon.fr/xavier.pujol/fplll/libfplll-3.0.12.tar.gz
tar -xf libfplll-3.0.12.tar.gz
cd libfplll-3.0.12
./configure CXXFLAGS="-O3"
make

I wondered why -O3 hangs up the compiler and so I tried to track down the issue. First, I tried to find out the the difference between -O2 between -O3. Gcc's man page states that -O3 enables the switches of -O2 and the following ones (lets call them x):

-finline-functions -funswitch-loops -fpredictive-commoning -fgcse-after-reload
-ftree-vectorize -fipa-cp-clone

I verified that by comparing the output of gcc when invoked with -Q -O2 --help=optimizers and -Q -O3 --help=optimizers. I then planned to selectively remove the switches in order to find the one which causes the problem. However, compilation works fine with -O2 and the additional switches above, so I conclude

-O3 != -O2 x

Now my question: Does anybody know if there is a further difference between -O2 and -O3 (undocumented?), has anyone experienced a similar behavior ? Is this maybe a compiler bug ?

Ciro Santilli OurBigBook.com
  • 347,512
  • 102
  • 1,199
  • 985
cschwan
  • 3,283
  • 3
  • 22
  • 32
  • I think, you should add to the problem -O3 run such options "-ftree-dump-all-all -frtl-dump-all-all" and to find last stage which have a hang in it. Also, chech the stack of gcc with starting it under `gdb --args` and stopping with Ctrl-C. – osgx Jun 23 '11 at 13:45
  • For google users without C++/make experience who are trying to make libfplll like me, without specifying any flag it "hanged" about 30 minutes on `enum/libfplll_la-enumerate_base.lo` and then continued. – Albert Hendriks May 23 '20 at 13:06

2 Answers2

23

Man pages can be outdated, but you can find actual lists for O2 and O3.

To get full list (almost, check "update") of -f optimization options actually used, I suggest you use the -fverbose-asm -save-temps (or -fverbose-asm -S) - there is a full list at a top of asm file (*.s).

For gcc-4.6.0 I got x (the difference between O2 and O3) to be:

 -fgcse-after-reload
 -finline-functions
 -fipa-cp-clone
 -fpredictive-commoning
 -ftree-loop-distribute-patterns
 -ftree-vectorize
 -funswitch-loops

Another source of information for your question is the sources of GCC (file gcc/opts.c and possibly gcc/common.opt) as gcc-4.6.0:

/* -O3 optimizations.  */
{ OPT_LEVELS_3_PLUS, OPT_ftree_loop_distribute_patterns, NULL, 1 },
{ OPT_LEVELS_3_PLUS, OPT_fpredictive_commoning, NULL, 1 },
/* Inlining of functions reducing size is a good idea with -Os
   regardless of them being declared inline.  */
{ OPT_LEVELS_3_PLUS_AND_SIZE, OPT_finline_functions, NULL, 1 },
{ OPT_LEVELS_3_PLUS, OPT_funswitch_loops, NULL, 1 },
{ OPT_LEVELS_3_PLUS, OPT_fgcse_after_reload, NULL, 1 },
{ OPT_LEVELS_3_PLUS, OPT_ftree_vectorize, NULL, 1 },
{ OPT_LEVELS_3_PLUS, OPT_fipa_cp_clone, NULL, 1 },

I have also checked, does gcc check -On in other files (cscope symbol search for x_optimize).

The only additional usage of n from option -On is saving it value into macro __OPTIMIZE__. So some headers can behave differently for value of this macro equal 2 or 3.

UPDATE: There are questions about it in GCC WIKI:

  • "Is -O1 (-O2,-O3 or -Os) equivalent to individual -foptimization options?"

No. First, individual optimization options (-f*) do not enable optimization, an option -Os or -Ox with x > 0 is required. Second, the -Ox flags enable many optimizations that are not controlled by any individual -f* option. There are no plans to add individual options for controlling all these optimizations.

  • "What specific flags are enabled by -O1 (-O2, -O3 or -Os)?"

Varies by platform and GCC version. You can get GCC to tell you what flags it enables by doing this: 

touch empty.c
gcc -O1 -S -fverbose-asm empty.c
cat empty.s
osgx
  • 90,338
  • 53
  • 357
  • 513
  • 5
    Your answer gave me the right hint - in gcc/opts.c I found additional parameters which are set with -O3: `max-aliased-vops` and `avg-aliased-vops`. The latter is the source of my problem. When set to `3` (which is default for -O3), the compiler shows the same behavior. Thanks! – cschwan Jun 24 '11 at 08:16
  • For the sake of completeness: `max-partial-antic-length` is also effected by -O3. This is also correctly documented in gcc's man page, though it is not mentioned in the description of -O3. – cschwan Jun 24 '11 at 08:26
  • cschwan, I happy to hear that you find the problem, but I think, that *vops and *antic* are added in gentoo version of gcc or smth like, because there are no such options in "vanilla" gcc-4.6 – osgx Jun 24 '11 at 11:20
  • 1
    No, these options are present in vanilla gcc 4.4.6, see [GCC 4.4.6 Optimization Options](http://gcc.gnu.org/onlinedocs/gcc-4.4.6/gcc/Optimize-Options.html#Optimize-Options). In 4.5 `max-aliased-vops` and `avg-alias-vops` were apparently removed (at least they are no longer documented in [GCC 4.5.3 Optiomization Options](http://gcc.gnu.org/onlinedocs/gcc-4.5.3/gcc/Optimize-Options.html#Optimize-Options)), 4.6 removes `max-partial-antic-length`. – cschwan Jun 24 '11 at 16:55
3

If your compiler hangs, then yes - I would regard that as a compiler bug. Compilers have bugs too.

(even if the compiler used to compile your compiler have bugs, a bug could be introduced in the new compiler - gcc takes some steps to avoid that by its staging bootstrapping though.)

It could be other things as well, e.g. the optimization done just takes much, much more time to perform, or the increased optimization level causes more memory to be used, and your system starts trashing.

nos
  • 223,662
  • 58
  • 417
  • 506