Mailing-List: contact dev-help@stdcxx.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@stdcxx.apache.org
Received-SPF: pass (nike.apache.org: local policy)
Content-class: urn:content-classes:message
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
Subject: RE: svn commit: r663377 - in /stdcxx/branches/4.2.x: include/valarray
 src/valarray.cpp tests/numerics/26.class.gslice.cpp
 tests/regress/26.valarray.sub.stdcxx-955.cpp
Date: Fri, 6 Jun 2008 12:40:23 -0600
Message-ID: 
 <CFFDD219128FD94FB4F92B99F52D0A499AA886@exchmail01.Blue.Roguewave.Com>
Thread-Topic: svn commit: r663377 - in /stdcxx/branches/4.2.x:
 include/valarray src/valarray.cpp tests/numerics/26.class.gslice.cpp
 tests/regress/26.valarray.sub.stdcxx-955.cpp
Thread-Index: AcjH6yxyYq28ay2KRdOi4MEcSZMWugADdcDw
References: <20080604214836.C2E492388A06@eris.apache.org>
 <48495931.4070409@roguewave.com>
From: "Travis Vitek" <Travis.Vitek@roguewave.com>
To: <dev@stdcxx.apache.org>

=20

Martin Sebor wrote:
>
>vitek@apache.org wrote:
>> Author: vitek
>> Date: Wed Jun  4 14:48:36 2008
>> New Revision: 663377
>>=20
>[...]
>> Modified: stdcxx/branches/4.2.x/src/valarray.cpp
>> URL:=20
>http://svn.apache.org/viewvc/stdcxx/branches/4.2.x/src/valarray
>.cpp?rev=3D663377&r1=3D663376&r2=3D663377&view=3Ddiff
>>=20
>=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- stdcxx/branches/4.2.x/src/valarray.cpp (original)
>> +++ stdcxx/branches/4.2.x/src/valarray.cpp Wed Jun  4 14:48:36 2008
>> @@ -41,8 +41,12 @@
>>  {
>>      _RWSTD_SIZE_T __n =3D _C_length.size ();
>> =20
>> -    while (__n && _C_r_length [__n - 1] =3D=3D _C_length [__n - 1] - =
1)
>> -        --__n;
>> +    for (/**/; __n; --__n)
>> +    {
>
>The brace should be on the line above :)
>

Blarg!

>> +        if (   _C_length [__n - 1]
>> +            && _C_r_length [__n - 1] !=3D _C_length [__n - 1] - 1)
>
>Also, I wonder if it might help generate more optimal code to write
>the loop like so:
>
>     while (__n) {
>         --__n;
>
>         if (_C_length [n] && _C_r_length [n] !=3D _C_length [n] - 1)
>             break;
>     }
>

It is possible it would help generate better code, but this code changes
the behavior of the loop. We now decrement __n before checking the
condition; when we break __n will be off-by-one. Obviously we can deal
with that by incrementing __n before the break.

I'll take a quick look at some disassembly and decide if it is worth
making the change or not.

>The duplicate check for (0 =3D=3D n) below could probably be hoisted
>into the loop for even more optimal code, something like this:
>
>     for ( ; ; ) {
>         if (0 =3D=3D n) {
>             _C_reset =3D true;
>             break;
>         }
>
>         --n;
>
>         if (_C_length [n] && _C_r_length [n] !=3D _C_length [n] - 1)
>             break;
>     }
>
>> +            break;
>> +    }
>> =20
>>      if (0 =3D=3D __n) {
>>          _C_reset    =3D true;
>>=20