[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gcc-bugs
Subject:    [Bug target/54089] [SH] Refactor shift patterns
From:       "olegendo at gcc dot gnu.org" <gcc-bugzilla () gcc ! gnu ! org>
Date:       2012-09-30 18:45:56
Message-ID: bug-54089-4-akKCAhyksn () http ! gcc ! gnu ! org/bugzilla/
[Download RAW message or body]


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54089

--- Comment #21 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-09-30 18:45:56 UTC ---
I've noticed that there seems to be a problem with register allocation related
to shift insns.  For example in the CSiBE set, I've seen sequences such as

        mov.w   .L342,r2
        mov     #0,r7
        .align 2
.L340:
        mov     r7,r1
        add     r2,r1
        shar    r1
        mov     r1,r3      <<<
        shll2   r3         <<<
        mov     r3,r0      <<<
        mov.l   @(r0,r5),r3
        cmp/gt  r4,r3
        bf      0f

quite often.  Not sure why this happens.  Maybe because 'gen_shifty_op' (which
emits the shift sequence insns) is called after reload and not during the first
split pass after combine...
[prev in list] [next in list] [prev in thread] [next in thread]