public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug tree-optimization/65412] New: missing control flow optimisation
@ 2015-03-12 21:33 skvadrik at gmail dot com
  2015-03-13  9:22 ` [Bug tree-optimization/65412] " rguenth at gcc dot gnu.org
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: skvadrik at gmail dot com @ 2015-03-12 21:33 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65412

            Bug ID: 65412
           Summary: missing control flow optimisation
           Product: gcc
           Version: 4.9.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: skvadrik at gmail dot com

Created attachment 35023
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35023&action=edit
files 1.c (source), 1.s (gcc assembly), 2.s (clang assembly)

Consider the following code:

void f(int x)
{
   if (x == 0) f0();
   else if (x == 1) f1();
   else if (x == 2) f2();
   else if (x == 3) f3();
   else if (x == 4) f4();
   else if (x == 5) f5();
}

gcc-4.9.2 doesn't optimize ifs to decision tree, jump table or something (see
1.s in attach). clang does (2.s in attach, jump table).

Maybe you could improve that?


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [Bug tree-optimization/65412] missing control flow optimisation
  2015-03-12 21:33 [Bug tree-optimization/65412] New: missing control flow optimisation skvadrik at gmail dot com
@ 2015-03-13  9:22 ` rguenth at gcc dot gnu.org
  2021-06-04 21:38 ` [Bug tree-optimization/65412] sequence of ifs not turned into a switch statement pinskia at gcc dot gnu.org
  2021-06-04 21:41 ` pinskia at gcc dot gnu.org
  2 siblings, 0 replies; 4+ messages in thread
From: rguenth at gcc dot gnu.org @ 2015-03-13  9:22 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65412

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think there exist patches for if-to-switch conversion.


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [Bug tree-optimization/65412] sequence of ifs not turned into a switch statement
  2015-03-12 21:33 [Bug tree-optimization/65412] New: missing control flow optimisation skvadrik at gmail dot com
  2015-03-13  9:22 ` [Bug tree-optimization/65412] " rguenth at gcc dot gnu.org
@ 2021-06-04 21:38 ` pinskia at gcc dot gnu.org
  2021-06-04 21:41 ` pinskia at gcc dot gnu.org
  2 siblings, 0 replies; 4+ messages in thread
From: pinskia at gcc dot gnu.org @ 2021-06-04 21:38 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65412

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |10.0
   Last reconfirmed|2015-12-28 00:00:00         |2021-6-4

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
>From *.iftoswitch on the trunk:
;;  Canonical GIMPLE case clusters: 0 1 2 3 4 5


So I looked and this is fixed but not for the low if count.
Take:
#define B(y) void f##y(void);
#define B10(y)  B(y##0)  B(y##1)  B(y##2) \
                B(y##3)  B(y##4)  B(y##5) \
                B(y##6)  B(y##7)  B(y##8) \
                B(y##9)
#define B100(y)  B10(y##0)  B10(y##1)  B10(y##2) \
                 B10(y##3)  B10(y##4)  B10(y##5) \
                 B10(y##6)  B10(y##7)  B10(y##8) \
                 B10(y##9)

B10(1)
B100(1)

#define A(y) else if (x == y) f##y();
#define A10(y)  A(y##0) \
                A(y##1) \
                A(y##2) \
                A(y##3) \
                A(y##4) \
                A(y##5) \
                A(y##6) \
                A(y##7) \
                A(y##8) \
                A(y##9)

#define A100(y) A10(y##0)  \
                A10(y##1) \
                A10(y##2) \
                A10(y##3) \
                A10(y##4) \
                A10(y##5) \
                A10(y##6) \
                A10(y##7) \
                A10(y##8) \
                A10(y##9)

void f10(int x)
{
  if (0) ;
  A10(1)
}
void f100(int x)
{
  if (0) ;
  A100(1)
}

--- CUT ----
f100 is converted while f10 is not.  This is a heuristics at work to see if it
is better as a jump table vs if statements.

https://godbolt.org/z/ofWGb5aKn for reference.

Clang decides (I don't know if they lower switches back to if statements) to do
the switch for all cases.

So we can close this as fixed.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* [Bug tree-optimization/65412] sequence of ifs not turned into a switch statement
  2015-03-12 21:33 [Bug tree-optimization/65412] New: missing control flow optimisation skvadrik at gmail dot com
  2015-03-13  9:22 ` [Bug tree-optimization/65412] " rguenth at gcc dot gnu.org
  2021-06-04 21:38 ` [Bug tree-optimization/65412] sequence of ifs not turned into a switch statement pinskia at gcc dot gnu.org
@ 2021-06-04 21:41 ` pinskia at gcc dot gnu.org
  2 siblings, 0 replies; 4+ messages in thread
From: pinskia at gcc dot gnu.org @ 2021-06-04 21:41 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65412

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
clang decides at 4 to change it to a switch statement while GCC takes a few
extra ifs to change it.  Again this is heuristics at work.


#define B(y) void f##y(void);
#define B10(y)  B(y##0)  B(y##1)  B(y##2) \
                B(y##3)  B(y##4)  B(y##5) \
                B(y##6)  B(y##7)  B(y##8) \
                B(y##9)

B10(1)

#define A(y) else if (x == y) f##y();

void f10(int x)
{
  if (0) ;
  A(11)
  A(12)
  A(13)
  A(14)
}

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2021-06-04 21:41 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-03-12 21:33 [Bug tree-optimization/65412] New: missing control flow optimisation skvadrik at gmail dot com
2015-03-13  9:22 ` [Bug tree-optimization/65412] " rguenth at gcc dot gnu.org
2021-06-04 21:38 ` [Bug tree-optimization/65412] sequence of ifs not turned into a switch statement pinskia at gcc dot gnu.org
2021-06-04 21:41 ` pinskia at gcc dot gnu.org

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).