[Bug rtl-optimization/21182] gcc can use registers but uses stack instead

[Bug rtl-optimization/21182] gcc can use registers but uses stack instead

[vda.linux at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #6 from Denis Vlasenko <vda.linux at googlemail dot com> 2013-01-18 00:48:23 UTC ---
Created attachment 29200
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29200
Updated testcase, build heper, and results of testing with different gcc
versions

Tarball contains:

serpent.c:
the original testcase, only with "#ifdef NAIL_REGS" instead of "#if 0" which
allows test compiles w/o editing it. Basically, "gcc -DNAIL_REGS serpent.c"
will try to force gcc to use only registers instead of stack.

gencode.sh:
builds serpent.c with -O2 and -O3, with and without -DNAIL_REGS. The object
file names contain gcc version and used options. Then they are objdump'ed and
output saved. Tweakable with setting $PREFIX and/or $CC.
No -fomit-frame-pointer used: the testcase can be compiled so that stack is not
used even without that option.

Disassembly:
serpent-O2-3.4.3.asm
serpent-O2-4.2.1.asm
serpent-O2-4.6.3.asm
serpent-O2-DNAIL_REGS-3.4.3.asm
serpent-O2-DNAIL_REGS-4.2.1.asm
serpent-O2-DNAIL_REGS-4.6.3.asm
serpent-O3-3.4.3.asm
serpent-O3-4.2.1.asm
serpent-O3-4.6.3.asm
serpent-O3-DNAIL_REGS-3.4.3.asm
serpent-O3-DNAIL_REGS-4.2.1.asm
serpent-O3-DNAIL_REGS-4.6.3.asm

Object files:
   text    data     bss     dec     hex filename
   3260       0       0    3260     cbc serpent-O2-DNAIL_REGS-3.4.3.o
   3260       0       0    3260     cbc serpent-O3-DNAIL_REGS-3.4.3.o
   3292       0       0    3292     cdc serpent-O3-3.4.3.o
   3536       0       0    3536     dd0 serpent-O2-4.6.3.o
   3536       0       0    3536     dd0 serpent-O3-4.6.3.o
   3845       0       0    3845     f05 serpent-O2-DNAIL_REGS-4.6.3.o
   3845       0       0    3845     f05 serpent-O3-DNAIL_REGS-4.6.3.o
   3877       0       0    3877     f25 serpent-O2-4.2.1.o
   3877       0       0    3877     f25 serpent-O3-4.2.1.o
   4302       0       0    4302    10ce serpent-O2-3.4.3.o
   4641       0       0    4641    1221 serpent-O2-DNAIL_REGS-4.2.1.o
   4641       0       0    4641    1221 serpent-O3-DNAIL_REGS-4.2.1.o

Take a look inside serpent-O2-DNAIL_REGS-3.4.3.asm file.
This is what I want to get without asm hacks: the smallest code, uses no stack.

gcc-3.4.3 -O3 comes close: it does spill a few words to stack (search for
(%ebp)), but is generally good code (close to ideal?).

All other attempts fare worse:

gcc-3.4.3 -O2: code is significantly worse than -O3.
gcc-4.2.1 -O2/-O3: code is better than gcc-3.4.3 -O2, worse than gcc-4.6.3
gcc-4.6.3 -O2/-O3: six instances of spills to stack . Code is still not as good
as gcc-3.4.3 -O3. (-DNAIL_REGS only confuses it more, unlike 3.4.3).

Stack usage summary:

$ grep 'sub.*,%esp' *.asm | grep -v DNAIL_REGS
serpent-O2-3.4.3.asm:   6:  81 ec 00 01 00 00       sub    $0x100,%esp
serpent-O2-4.2.1.asm:   6:  83 ec 78                sub    $0x78,%esp
serpent-O2-4.6.3.asm:   4:  83 ec 04                sub    $0x4,%esp
serpent-O3-4.2.1.asm:   6:  83 ec 78                sub    $0x78,%esp
serpent-O3-4.6.3.asm:   4:  83 ec 04                sub    $0x4,%esp

(serpent-O3-3.4.3.asm is not listed, but it allocates and uses one word on
stack by push insn).


Modules with best (= minimal) stack usage:

$ grep -F -e '(%esp)' -e '(%ebp)' serpent-O2-DNAIL_REGS-3.4.3.asm
   6:   8b 75 08                mov    0x8(%ebp),%esi
   9:   8b 7d 10                mov    0x10(%ebp),%edi
 ca9:   8b 75 0c                mov    0xc(%ebp),%esi

$ grep -F -e '(%esp)' -e '(%ebp)' serpent-O3-3.4.3.asm
   7:   8b 7d 08                mov    0x8(%ebp),%edi
   a:   8b 4d 10                mov    0x10(%ebp),%ecx
 18c:   89 7d f0                mov    %edi,-0x10(%ebp)
 1dd:   8b 45 f0                mov    -0x10(%ebp),%eax
 23b:   8b 75 f0                mov    -0x10(%ebp),%esi
 299:   8b 7d f0                mov    -0x10(%ebp),%edi
 432:   8b 55 f0                mov    -0x10(%ebp),%edx
 4a0:   8b 4d f0                mov    -0x10(%ebp),%ecx
 50e:   8b 7d f0                mov    -0x10(%ebp),%edi
 84f:   8b 45 f0                mov    -0x10(%ebp),%eax
 8b9:   8b 75 f0                mov    -0x10(%ebp),%esi
 923:   8b 7d f0                mov    -0x10(%ebp),%edi
 cb6:   8b 55 0c                mov    0xc(%ebp),%edx

$ grep -F -e '(%esp)' -e '(%ebp)' serpent-O3-4.6.3.asm
   7:   8b 4c 24 20             mov    0x20(%esp),%ecx
   b:   8b 44 24 18             mov    0x18(%esp),%eax
 22e:   89 0c 24                mov    %ecx,(%esp)
 239:   23 3c 24                and    (%esp),%edi
 588:   89 0c 24                mov    %ecx,(%esp)
 58f:   23 3c 24                and    (%esp),%edi
 8f4:   89 0c 24                mov    %ecx,(%esp)
 8fd:   23 3c 24                and    (%esp),%edi
 c60:   89 0c 24                mov    %ecx,(%esp)
 c6b:   23 3c 24                and    (%esp),%edi
 d37:   89 14 24                mov    %edx,(%esp)
 d5a:   8b 44 24 1c             mov    0x1c(%esp),%eax
 d5e:   33 14 24                xor    (%esp),%edx


Conclusion:
gcc-4.6.3 -O3 was close to ideal.
gcc-4.2.1 is worse.
gcc-4.6.3 got better a bit, still not as good as gcc-4.6.3 -O3.

[Bug rtl-optimization/21182] gcc can use registers but uses stack instead

[vda.linux at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Denis Vlasenko <vda.linux at googlemail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vda.linux at googlemail dot
                   |                            |com

--- Comment #7 from Denis Vlasenko <vda.linux at googlemail dot com> 2013-01-18 00:51:01 UTC ---
"gcc-4.6.3 got better a bit, still not as good as gcc-4.6.3 -O3."

I meant:

gcc-4.6.3 got better a bit, still not as good as gcc-3.4.3 -O3 used to be.

[Bug rtl-optimization/21182] gcc can use registers but uses stack instead

[vda.linux at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #8 from Denis Vlasenko <vda.linux at googlemail dot com> 2013-01-18 00:55:37 UTC ---
Grrr, another mistake. Correcting again:

Conclusion:
gcc-3.4.3 -O3 was close to ideal.
^^^^^^^^^
gcc-4.2.1 is worse.
gcc-4.6.3 got better a bit, still not as good as gcc-3.4.3 -O3 used to be.
                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^

[Bug rtl-optimization/21182] gcc can use registers but uses stack instead

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> 2013-01-18 00:57:00 UTC ---
It would be interesting to try the trunk which has a newer register allocator
than even 4.6.x/4.7.x.

[Bug rtl-optimization/21182] [4.6/4.7/4.8 Regression] gcc can use registers but uses stack instead

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vmakarov at gcc dot gnu.org
   Target Milestone|---                         |4.6.4
            Summary|gcc can use registers but   |[4.6/4.7/4.8 Regression]
                   |uses stack instead          |gcc can use registers but
                   |                            |uses stack instead

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> 2013-01-18 10:38:54 UTC ---
4.4.7 and 4.5.4 generate the same code (no stack use) for -D/-UNAIL_REGS.
With 4.6.3, the -DNAIL_REGS code regresses very much (IRA ...), the
-UNAIL_REGS code is nearly perfect but less good than 4.4/4.5 (if you
only consider grep esp serpent.s | wc -l).  Same behavior with 4.7.2.

Trunk got somewhat worse with -UNAIL_REGS but better with -DNAIL_REGS (at -O2):

     -UNAIL_REGS  -DNAIL_REGS
4.5.4     3             3
4.6.3    15           101
4.7.2    15            93
4.8.0    23            70

The most important thing to fix is the -UNAIL_REGS case of course.

A regression for that from 4.4/4.5.

[Bug rtl-optimization/21182] [4.6/4.7/4.8 Regression] gcc can use registers but uses stack instead

[vda.linux at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #11 from Denis Vlasenko <vda.linux at googlemail dot com> 2013-01-20 14:39:42 UTC ---
(In reply to comment #10)
> 4.4.7 and 4.5.4 generate the same code (no stack use) for -D/-UNAIL_REGS.
> With 4.6.3, the -DNAIL_REGS code regresses very much (IRA ...), the
> -UNAIL_REGS code is nearly perfect but less good than 4.4/4.5 (if you
> only consider grep esp serpent.s | wc -l).  Same behavior with 4.7.2.
> 
> Trunk got somewhat worse with -UNAIL_REGS but better with -DNAIL_REGS (at -O2):
> 
>      -UNAIL_REGS  -DNAIL_REGS
> 4.5.4     3             3
> 4.6.3    15           101

This matches what I see with 4.6.3 - 15 insns with %esp (and no %ebp):

$ grep '%esp' serpent-4.6.3-O2.asm
   4:   83 ec 04                sub    $0x4,%esp
   7:   8b 4c 24 20             mov    0x20(%esp),%ecx
   b:   8b 44 24 18             mov    0x18(%esp),%eax
 22e:   89 0c 24                mov    %ecx,(%esp)
 239:   23 3c 24                and    (%esp),%edi
 588:   89 0c 24                mov    %ecx,(%esp)
 58f:   23 3c 24                and    (%esp),%edi
 8f4:   89 0c 24                mov    %ecx,(%esp)
 8fd:   23 3c 24                and    (%esp),%edi
 c60:   89 0c 24                mov    %ecx,(%esp)
 c6b:   23 3c 24                and    (%esp),%edi
 d37:   89 14 24                mov    %edx,(%esp)
 d5a:   8b 44 24 1c             mov    0x1c(%esp),%eax
 d5e:   33 14 24                xor    (%esp),%edx
 d70:   83 c4 04                add    $0x4,%esp

> The most important thing to fix is the -UNAIL_REGS case of course.

Sure. NAIL_REGS is only a hack meant to demonstrate that regs *can* be
allocated optimally.

[Bug rtl-optimization/21182] [4.6/4.7/4.8 Regression] gcc can use registers but uses stack instead

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #12 from Steven Bosscher <steven at gcc dot gnu.org> 2013-03-13 20:37:23 UTC ---
Curious to hear whether "-fschedule-insns -fsched-pressure" helps.

At least from the %esp and %ebp counts, it looks hopeful:

$ ./cc1 -quiet -m32 -O2 t.c -DNAIL_REGS -o t.s.NAIL
$ ./cc1 -quiet -m32 -O2 t.c -UNAIL_REGS -o t.s
$ ./cc1 -quiet -m32 -O2 t.c -UNAIL_REGS -o t.s.sched_pres \
  -fschedule-insns -fsched-pressure
$ egrep -c '%ebp|%esp' t.s*
t.s:366
t.s.NAIL:305
t.s.sched_pres:277
$ grep ident t.s
.ident  "GCC: (GNU) 4.8.0 20130313 (experimental) [trunk revision 196638]"

It is unfortunate that nobody has put in the resources yet to make options
like "-fschedule-insns -fsched-pressure" the default for x86.

[Bug rtl-optimization/21182] [4.7/4.8/4.9 Regression] gcc can use registers but uses stack instead

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.6.4                       |4.7.4

--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-04-12 15:17:13 UTC ---
GCC 4.6.4 has been released and the branch has been closed.

[Bug rtl-optimization/21182] [4.7/4.8/4.9/4.10 Regression] gcc can use registers but uses stack instead

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.7.4                       |4.8.4

--- Comment #14 from Richard Biener <rguenth at gcc dot gnu.org> ---
The 4.7 branch is being closed, moving target milestone to 4.8.4.

[Bug rtl-optimization/21182] [4.8/4.9/5 Regression] gcc can use registers but uses stack instead

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.4                       |4.8.5

--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.8.4 has been released.

[Bug rtl-optimization/21182] [4.8/4.9/5/6 Regression] gcc can use registers but uses stack instead

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.5                       |4.9.3

--- Comment #16 from Richard Biener <rguenth at gcc dot gnu.org> ---
The gcc-4_8-branch is being closed, re-targeting regressions to 4.9.3.

[Bug rtl-optimization/21182] [4.9/5/6 Regression] gcc can use registers but uses stack instead

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.3 has been released.

[Bug rtl-optimization/21182] [4.9/5/6 Regression] gcc can use registers but uses stack instead

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.3                       |4.9.4

[Bug rtl-optimization/21182] [8/9/10/11 Regression] gcc can use registers but uses stack instead

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

--- Comment #31 from Richard Biener <rguenth at gcc dot gnu.org> ---
-fno-tree-ter improves things quite a bit.  With -DNAILED_REGS gimple doesn't
do much because we treat registers as memory here.

For trunk

-O2               has 52 spills
-O2 -fno-tree-ter has 35 spills
-O2 -fno-tree-ter -fschedule-insns has 74 spills
-O2 -fno-tree-ter -fschedule-insns -fsched-pressure has 18 spills
-O2 -fschedule-insns -fsched-pressure has 17 spills

to me this really hints at out-of-SSA producing a very bad initial
schedule, by TER but also likely due to folding & friends doing
random stmt placing (it's all a single BB).  I think we'd benefit
quite a bit with killing TER (doing all interesting bits pre-RTL
or via SSA RTL forwprop) and ordering SSA def expansion for optimal
register pressure.

[Bug rtl-optimization/21182] [8/9/10/11/12 Regression] gcc can use registers but uses stack instead

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.0                        |11.2

--- Comment #32 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 11.1 has been released, retargeting bugs to GCC 11.2.

[Bug rtl-optimization/21182] [9/10/11/12 Regression] gcc can use registers but uses stack instead

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.2                        |11.3

--- Comment #33 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 11.2 is being released, retargeting bugs to GCC 11.3

[Bug rtl-optimization/21182] [9/10/11/12 Regression] gcc can use registers but uses stack instead

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.3                        |11.4

--- Comment #34 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 11.3 is being released, retargeting bugs to GCC 11.4.

[Bug rtl-optimization/21182] [10/11/12/13/14 Regression] gcc can use registers but uses stack instead

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.4                        |11.5

--- Comment #35 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 11.4 is being released, retargeting bugs to GCC 11.5.

[Bug rtl-optimization/21182] [11/12/13/14 Regression] gcc can use registers but uses stack instead

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=21182

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2006-01-15 20:37:58         |2023-7-15

--- Comment #36 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
turning off TER still helps.