https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65266

            Bug ID: 65266
           Summary: [SH] Use rotcl for bit reversals
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: olegendo at gcc dot gnu.org
            Target: sh*-*-*

The following examples are small bit reversals:

unsigned int test1 (unsigned int a)
{
  return ((a & 1) << 1) | ((a >> 1) & 1);
}

unsigned int test2 (unsigned int a)
{
  return ((a & 1) << 2) | (((a >> 1) & 1) << 1) | ((a >> 2) & 1);
}

unsigned int test3 (unsigned int a)
{
  return ((a & 1) << 3) | (((a >> 1) & 1) << 2) | (((a >> 2) & 1) << 1)
         | (((a >> 3) & 1) << 0);
}

where test3 currently compiles to:
        mov     r4,r0
        mov     #1,r2
        tst     #8,r0
        and     r4,r2
        movt    r1
        mov     r2,r0
        shll2   r0
        tst     r1,r1
        mov     r0,r1
        mov     r4,r0
        and     #2,r0
        rotcl   r1
        add     r0,r0
        mov     r0,r2
        mov     r4,r0
        or      r1,r2
        tst     #4,r0
        mov     #-1,r1
        mov     r2,r0
        negc    r1,r1
        add     r1,r1
        rts
        or      r1,r0

where a minimal sequence would be:
        shlr    r4
        movt    r0
        shlr    r4
        rotcl   r0
        shlr    r4
        rotcl   r0
        shlr    r4
        rotcl   r0


For example, combine tries the following pattern:

Failed to match this instruction:
(set (reg:SI 186 [ D.2006 ])
    (ior:SI (ior:SI (and:SI (ashift:SI (reg/v:SI 176 [ a ])
                                       (const_int 1 [0x1]))
                            (const_int 4 [0x4]))
                    (zero_extract:SI (reg/v:SI 176 [ a ])
                                     (const_int 1 [0x1])
                                     (const_int 3 [0x3])))
             (and:SI (ashift:SI (reg/v:SI 176 [ a ])
                                (const_int 3 [0x3]))
                     (const_int 8 [0x8]))))

A predicate that accepts IOR chains of single bit selections (via zero_extract
or shift-and) could be used to capture the sequence during combine and then
smash it in split1.

Possibly related: PR 65265, PR 63321.

Reply via email to