[PATCH] Remove gen_binary from combine.c (take 3)

[Joseph S. Myers]

On Mon, 21 Mar 2005, Paolo Bonzini wrote:

> > > This patch fixed bug 19454 (regression for bf64-1.c on ia64-hpux) on
> > > mainline and when applied to 4.0 branch fixes the regression there as well
> > > with no other changes to testsuite results, so I believe it should be
> > > considered for 4.0 branch.
> > 
> > 
> > Agreed, but I'd like to see 4.0 test results on at least two other
> > platforms, as this is rather heavy hammer.
> 
> And I would like to see diffs in .combine output on the failing test
> case, for the same reason.  (The patch has been in my tree for months,
> but it is indeed a big hammer!)

I tested on x86_64-unknown-linux-gnu, and found the patch isn't yet 
suitable for 4.0 branch because it causes regressions:

FAIL: gcc.c-torture/unsorted/pass.c,  -O3 -fomit-frame-pointer  
FAIL: gcc.c-torture/unsorted/pass.c,  -O3 -g  

The same tests fail on mainline on x86_64-unknown-linux-gnu, but didn't a 
few days ago, so I suspect this patch of causing them there too.

/scratch/joseph/gcc-mainline/gcc/testsuite/gcc.c-torture/unsorted/pass.c: In function 'bar':
/scratch/joseph/gcc-mainline/gcc/testsuite/gcc.c-torture/unsorted/pass.c:13: error: unrecognizable insn:
(insn 30 10 15 0 (set (reg:SI 0 ax [61])
        (subreg:SI (plus:DI (plus:DI (reg:DI 0 ax [orig:58 w ] [58])
                    (subreg:DI (reg:SI 7 sp) 0))
                (const_int -4 [0xfffffffffffffffc])) 0)) -1 (nil)
    (nil))
/scratch/joseph/gcc-mainline/gcc/testsuite/gcc.c-torture/unsorted/pass.c:13: internal compiler error: in extract_insn, at recog.c:2042

(The patch might still be suitable for 4.0 branch after this issue is 
fixed.  Or a smaller local fix for 4.0 branch for the original problem 
would be fine.)

-- 
Joseph S. Myers               http://www.srcf.ucam.org/~jsm28/gcc/
    jsm@polyomino.org.uk (personal mail)
    joseph@codesourcery.com (CodeSourcery mail)
    jsm28@gcc.gnu.org (Bugzilla assignments and CCs)