This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: rs6000 geu/ltu patterns

From: Alan Modra <amodra at bigpond dot net dot au>
To: David Edelsohn <dje at watson dot ibm dot com>
Cc: gcc-patches at gcc dot gnu dot org
Date: Tue, 9 Mar 2004 10:06:12 +1030
Subject: Re: rs6000 geu/ltu patterns
References: <20040308131622.GO18129@bubble.modra.org> <200403082122.i28LMXT31522@makai.watson.ibm.com>

On Mon, Mar 08, 2004 at 04:22:33PM -0500, David Edelsohn wrote:
> 	I guess in the short-run, if GCC does generate those comparisons,
> we need your proposed patch as an optimization.  We really don't want GCC
> to generate the actual comparison, assuming GCC generates something
> stupid if zero is rejected.

For the testcase in PR 14478, rejecting zero in reg_or_neg_short_operand
generates (-m32 -O1):

foo:
        cmpwi 0,3,0
        beq- 0,.L6
.L4:
        addi 4,4,1
        slwi. 3,3,1
        bne+ 0,.L4
.L6:
        li 3,0
        subfc 3,3,4
        li 3,0
        adde 3,3,3
        blr

That looks like correct code to me, but inferior to that generated
with my patch:

foo:
        cmpwi 0,3,0
        beq- 0,.L6
.L4:
        addi 4,4,1
        slwi. 3,3,1
        bne+ 0,.L4
.L6:
        li 3,1
        blr

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

References:
- rs6000 geu/ltu patterns
  - From: Alan Modra
- Re: rs6000 geu/ltu patterns
  - From: David Edelsohn

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]