[Bug middle-end/49491] New: Superflouos move because of unnecessary spill for 2-operand insn.

["gjl at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49491

           Summary: Superflouos move because of unnecessary spill for
                    2-operand insn.
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Keywords: missed-optimization, ra
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: gjl@gcc.gnu.org
            Target: avr


This is an optimization flaw I observe in current 4.7 trunk (r175201) and also
in upcoming 4.6.1.

This is the C source:

unsigned int shift1 (unsigned char x, unsigned char y)
{
    return (x|y) >> 3;
}

The machine (AVR) just has 2-operand instructions.  It occurs for AVR but
presumably is not target-specific and so I chose "middle-end" as component
because reload generates insn 27.

Function shift1 gets x in r24, y in r22 and has to return the result in r24
(LSB) and r25 (MSB).

The code produced is:

shift1:
    or r22,r24     ;  7    iorqi3/1    [length = 1]
    mov r24,r22     ;  27    *movqi/1    [length = 1]
    lsr r24     ;  24    *lshrqi3/5    [length = 3]
    lsr r24
    lsr r24
    ldi r25,lo8(0)     ;  25    *movqi/1    [length = 1]
    ret     ;  30    return    [length = 1]

Instead of a simple

    or r24,r22

there is  

    or r22,r24
    mov r24,r22

There is nothing I can see in avr BE that forces this move; iorqi3 insn is
commutative and no insn involved needs an (early) clobber.

Without the shift, code is as expected:

unsigned int or1 (unsigned char x, unsigned char y)
{
    return x|y;
}

or1:
    or r24,r22     ;  23    iorqi3/1    [length = 1]
    ldi r25,lo8(0)     ;  24    *movqi/1    [length = 1]
    ret     ;  28    return    [length = 1]

===================

Command line:

avr-gcc shr.c -Os -dp -S -v

===================

Using built-in specs.
COLLECT_GCC=avr-gcc
COLLECT_LTO_WRAPPER=/local/gnu/install/gcc-4.7/libexec/gcc/avr/4.7.0/lto-wrapper
Target: avr
Configured with: ../../gcc.gnu.org/trunk/configure --target=avr
--prefix=/local/gnu/install/gcc-4.7 --disable-nls --disable-shared
--enable-languages=c,c++ --with-dwarf2
Thread model: single
gcc version 4.7.0 20110620 (experimental) (GCC)
COLLECT_GCC_OPTIONS='-Os' '-dp' '-S' '-v'
 /local/gnu/install/gcc-4.7/libexec/gcc/avr/4.7.0/cc1 -quiet -v shr.c -quiet
-dumpbase shr.c -dp -auxbase shr -Os -version -o shr.s
GNU C (GCC) version 4.7.0 20110620 (experimental) (avr)
        compiled by GNU C version 4.3.2 [gcc-4_3-branch revision 141291], GMP
version 5.0.1, MPFR version 3.0.0-p8, MPC version 0.8.2
GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096