[Bug optimization/10982] New: [arm] poor optimisation of "if (var & const)"

[Bug optimization/10982] New: [arm] poor optimisation of "if (var & const)"

[pb@gcc.gnu.org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982

           Summary: [arm] poor optimisation of "if (var & const)"
           Product: gcc
           Version: unknown
            Status: NEW
          Severity: enhancement
          Priority: P2
         Component: optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: pb@gcc.gnu.org
                CC: gcc-bugs@gcc.gnu.org
GCC target triplet: arm-linux

Compiling this code:

f(int a)
{
  if (a & 0xfff)
    return 1;
  return 0;
}

with -O2 yields:

        mov     r0, r0, asl #20
        mov     r0, r0, lsr #20
        cmp     r0, #0
        movne   r0, #1
        moveq   r0, #0

There's no need to fully calculate the value of the AND, since all that we care
about is whether it's zero or not.  So, I think the first three instructions
could be replaced by a single one:

        movs     r0, r0, asl #20



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug optimization/10982] [arm] poor optimisation of "if (var & const)"

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4                         |---

[Bug optimization/10982] [arm] poor optimisation of "if (var & const)"

[pb at gcc dot gnu dot org]

------- Additional Comments From pb at gcc dot gnu dot org  2004-05-09 17:11 -------
For that particular testcase, adding a pattern like the one below seems to allow
combine to do the right thing.  However, changing the "return 1" to "return 2"
prevents this from matching.  Something a bit more sophisticated might be needed.

--

(define_insn_and_split "*zero_extract_compare0_shifted"
  [(set (match_operand:SI 0 "s_register_operand" "=r")
	(ne:SI (zero_extract:SI (match_operand:SI 1 "s_register_operand" "r")
		        	   (match_operand:SI 2 "const_int_operand" "n")
			      	   (const_int 0))
	            	  (const_int 0)))
   (clobber (reg:CC CC_REGNUM))]
  "TARGET_ARM"
  "#"
  ""
  [(parallel [(set (reg:CC_NOOV CC_REGNUM)
		   (compare:CC_NOOV (ashift:SI (match_dup 1)
			                       (match_dup 2))
			            (const_int 0)))
	      (set (match_dup 0)
		   (ashift:SI (match_dup 1) (match_dup 2)))])
   (set (match_dup 0)
        (if_then_else (eq:SI (reg:CC_NOOV CC_REGNUM) (const_int 0))
	              (match_dup 0)
		      (const_int 1)))]
  ""
)

;; Match the second half of *zero_extract_compare0_shifted
(define_insn "*movsi_conditional"
  [(set (match_operand:SI 0 "s_register_operand" "=r")
        (if_then_else (eq:SI (reg:CC_NOOV CC_REGNUM) (const_int 0))
	              (match_dup 0)
		      (match_operand:SI 1 "const_int_operand" "n")))]
  "TARGET_ARM"
  "movne\\t%0, %1"
  [(set_attr "conds" "use")]
)


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|unknown                     |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982

[Bug target/10982] [arm] poor optimisation of "if (var & const)"

[rearnsha at gcc dot gnu dot org]

------- Additional Comments From rearnsha at gcc dot gnu dot org  2004-05-16 22:19 -------
The proposed patch isn't quite correct.  There are already patterns that can be
used as the target of the split, so there's no need to add new ones, and the
split needs to 

1) Convert operand2
2) Not apply to Thumb state


Patch forthcoming.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
          Component|rtl-optimization            |target
   Last reconfirmed|2004-01-01 07:59:12         |2004-05-16 22:19:36
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982

[Bug target/10982] [arm] poor optimisation of "if (var & const)"

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-05-16 22:22 -------
Subject: Bug 10982

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rearnsha@gcc.gnu.org	2004-05-16 22:22:49

Modified files:
	gcc            : ChangeLog 
	gcc/config/arm : arm.md 

Log message:
	PR target/10982
	* arm.md (ne_zeroextractsi): Convert to insn-and-split.
	(ne_zeroextractsi_shifted): New pattern.
	(ite_ne_zeroextractsi): New pattern.
	(ite_ne_zeroextractsi_shifted): New pattern.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.3669&r2=2.3670
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/arm/arm.md.diff?cvsroot=gcc&r1=1.167&r2=1.168



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982

[Bug target/10982] [arm] poor optimisation of "if (var & const)"

[rearnsha at gcc dot gnu dot org]

------- Additional Comments From rearnsha at gcc dot gnu dot org  2004-05-16 22:23 -------
Fixed with applied patch

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982

[Bug target/10982] [arm] poor optimisation of "if (var & const)"

[rearnsha at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10982