[Bug rtl-optimization/91865] New: Combine misses opportunity to remove (sign_extend (zero_extend)) before searching for insn patterns

[jozefl.gcc at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91865

            Bug ID: 91865
           Summary: Combine misses opportunity to remove (sign_extend
                    (zero_extend)) before searching for insn patterns
           Product: gcc
           Version: 9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jozefl.gcc at gmail dot com
  Target Milestone: ---

Created attachment 46913
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46913&action=edit
msp430-movqipsi.diff

The following program generates poor code for msp430-elf in the large memory
model:

const int table[2] = {1, 2};

int
foo (char i)
{
  return table[i];
}

> msp430-elf-gcc -S tester.i -mlarge -Os
The RTL generated by expand uses two insns to convert "i" to a register's
natural mode; there is a sign extension which would be unnecessary if the first
instruction had a PSImode register as the lvalue:

(insn 2 4 3 2 (set (reg/v:HI 25 [ i ])
        (zero_extend:HI (reg:QI 12 R12 [ i ])))
     (nil))
.....
(insn 7 6 8 2 (set (reg:PSI 28)
        (subreg:PSI (sign_extend:SI (reg/v:HI 25 [ i ])) 0))
     (nil))

All we really need is:

(insn (set (reg:PSI 28 [ i ])
        (zero_extend:PSI (reg:QI 12 R12 [ i ])))
     (nil))

Combine tries to combine the insns but doesn't recognize that 
the following transformation could be applied:
(sign_extend:PSI (zero_extend:HI)) -> (zero_extend:PSI)

Trying 2 -> 7:
    2: r25:HI=zero_extend(R12:QI)
      REG_DEAD R12:QI
    7: r28:PSI=sign_extend(r25:HI)#0
      REG_DEAD r25:HI
Failed to match this instruction:
(set (reg:PSI 28 [ i ])
    (sign_extend:PSI (zero_extend:HI (reg:QI 12 R12 [ i ]))))
Failed to match this instruction:
(set (reg:PSI 28 [ i ])
    (sign_extend:PSI (and:HI (reg:HI 12 R12)
            (const_int 255 [0xff]))))


A separate issue is that the movqipsi pattern is undefined, a patch containing
the pattern is attached and would match if the above transformation were
applied by combine.