[Bug target/49497] New: Incorrect lea peephole

[Bug target/49497] New: Incorrect lea peephole

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49497

           Summary: Incorrect lea peephole
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: ubizjak@gmail.com
            Target: x86


i386.md has

;; Convert imul by three, five and nine into lea
(define_peephole2
  [(parallel
    [(set (match_operand:SWI48 0 "register_operand" "")
          (mult:SWI48 (match_operand:SWI48 1 "register_operand" "")
                      (match_operand:SWI48 2 "const_int_operand" ""))) 
     (clobber (reg:CC FLAGS_REG))])]
  "INTVAL (operands[2]) == 3
   || INTVAL (operands[2]) == 5
   || INTVAL (operands[2]) == 9"
  [(set (match_dup 0)
        (plus:SWI48 (mult:SWI48 (match_dup 1) (match_dup 2))
                    (match_dup 1)))] 
  "operands[2] = GEN_INT (INTVAL (operands[2]) - 1);")

(define_peephole2
  [(parallel
    [(set (match_operand:SWI48 0 "register_operand" "")
          (mult:SWI48 (match_operand:SWI48 1 "nonimmediate_operand" "")
                      (match_operand:SWI48 2 "const_int_operand" ""))) 
     (clobber (reg:CC FLAGS_REG))])]
  "optimize_insn_for_speed_p ()
   && (INTVAL (operands[2]) == 3
       || INTVAL (operands[2]) == 5
       || INTVAL (operands[2]) == 9)"
  [(set (match_dup 0) (match_dup 1))
   (set (match_dup 0)
        (plus:SWI48 (mult:SWI48 (match_dup 0) (match_dup 2))
                    (match_dup 0)))]
  "operands[2] = GEN_INT (INTVAL (operands[2]) - 1);")


But lea split has

&& (!TARGET_PARTIAL_REG_STALL || optimize_function_for_size_p (cfun))

When TARGET_PARTIAL_REG_STALL is true, get got

[hjl@gnu-33 1036]$ cat x.i
 char *progName;
 int *tt;
 unsigned char *ll8;
 extern void foo (const char *);
 void cleanUpAndFail ( int ec ) {
    __builtin_exit ( ec );
 }
 void uncompressOutOfMemory ( int draw, int blockSize ) {
    foo ( "\tand failing that, find a machine with more memory.\n");
    cleanUpAndFail(1);
 }
 void setDecompressStructureSizes ( int newSize100k ) {
       int n = 100000 * newSize100k;
       ll8 = __builtin_malloc ( n * sizeof(unsigned char) );
       if (ll8 == ((void *)0) || tt == ((void *)0)) {
          int totalDraw             = n * sizeof(unsigned char) + n *
sizeof(unsigned int);
          uncompressOutOfMemory ( totalDraw, n );
       }
 }
[hjl@gnu-33 1036]$ make x.s
/export/build/gnu/gcc/build-x86_64-linux/gcc/xgcc
-B/export/build/gnu/gcc/build-x86_64-linux/gcc/ -O2  -S x.i
x.i: In function ‘setDecompressStructureSizes’:
x.i:19:2: error: unrecognizable insn:
(insn 43 19 21 4 (set (reg:SI 5 di [69])
        (plus:SI (mult:SI (reg/v:SI 3 bx [orig:59 n ] [59])
                (const_int 4 [0x4]))
            (reg/v:SI 3 bx [orig:59 n ] [59]))) x.i:16 -1
     (nil))
x.i:19:2: internal compiler error: in extract_insn, at recog.c:2113
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
make: *** [x.s] Error 1
[hjl@gnu-33 1036]$

[Bug target/49497] Incorrect lea peephole

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49497

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2011-06-21 23:20:18 UTC ---
Created attachment 24578
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24578
A patch

There is no partial register stall for SImode.

[Bug target/49497] Incorrect lea peephole

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49497

--- Comment #2 from Uros Bizjak <ubizjak at gmail dot com> 2011-06-22 06:14:27 UTC ---
(In reply to comment #1)
> Created attachment 24578 [details]
> A patch
> 
> There is no partial register stall for SImode.

Looks good to me, maybe we should introduce const359_operand predicate (similar
to const248_operand) to move operand checks from insn constraint.

[Bug target/49497] Incorrect lea peephole

[hjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49497

--- Comment #3 from hjl at gcc dot gnu.org <hjl at gcc dot gnu.org> 2011-06-22 15:29:48 UTC ---
Author: hjl
Date: Wed Jun 22 15:29:43 2011
New Revision: 175295

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=175295
Log:
Check TARGET_PARTIAL_REG_STALL in imul to lea peepholes.

2011-06-22  H.J. Lu  <hongjiu.lu@intel.com>

    PR target/49497
    * config/i386/i386.md (*lea_general_2): Always allow SImode.
    (*lea_general_2_zext): Likewise.
    (imul to lea peepholes): Use const359_operand and check
    TARGET_PARTIAL_REG_STALL.

    * config/i386/predicates.md (const359_operand): New.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.md
    trunk/gcc/config/i386/predicates.md

[Bug target/49497] Incorrect lea peephole

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49497

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.7.0

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> 2011-06-23 13:12:01 UTC ---
Fixed.