Mailing-List: contact dev-help@commons.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "Commons Developers List" <dev@commons.apache.org>
Received-SPF: pass (nike.apache.org: domain of sebbaz@gmail.com designates
 209.85.220.171 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <20110907094657.GQ2590@dusk.harfang.homelinux.org>
References: <20110906210659.51A182388A67@eris.apache.org>
	<20110906230459.GZ19625@dusk.harfang.homelinux.org>
	<CAOGo0VYfbRxg3r9GA1XCzVj3D4a5JCK_6eLqq9Aemp=x6B=t2A@mail.gmail.com>
	<20110907094657.GQ2590@dusk.harfang.homelinux.org>
Date: Wed, 7 Sep 2011 11:22:24 +0100
Message-ID: 
 <CAOGo0VbSho-5_9qugiDwHfbGD0AHLHO7oaED5KnjzrcgucXzpg@mail.gmail.com>
Subject: Re: svn commit: r1165846 [2/2] -
 /commons/proper/math/trunk/src/main/java/org/apache/commons/math/util/FastMath.java
From: sebb <sebbaz@gmail.com>
To: Commons Developers List <dev@commons.apache.org>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

On 7 September 2011 10:46, Gilles Sadowski <gilles@harfang.homelinux.org> w=
rote:
> On Wed, Sep 07, 2011 at 12:42:06AM +0100, sebb wrote:
>> On 7 September 2011 00:04, Gilles Sadowski <gilles@harfang.homelinux.org=
> wrote:
>> > Hello.
>> >
>> >> Modified: commons/proper/math/trunk/src/main/java/org/apache/commons/=
math/util/FastMath.java
>> >> URL: http://svn.apache.org/viewvc/commons/proper/math/trunk/src/main/=
java/org/apache/commons/math/util/FastMath.java?rev=3D1165846&r1=3D1165845&=
r2=3D1165846&view=3Ddiff
>> >> =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
>> >> --- commons/proper/math/trunk/src/main/java/org/apache/commons/math/u=
til/FastMath.java (original)
>> >> +++ commons/proper/math/trunk/src/main/java/org/apache/commons/math/u=
til/FastMath.java Tue Sep =A06 21:06:58 2011
>> >> @@ -74,31 +74,6147 @@ public class FastMath {
>> >> =A0 =A0 =A0/** Napier's constant e, base of the natural logarithm. */
>> >> =A0 =A0 =A0public static final double E =3D 2850325.0 / 1048576.0 + 8=
.254840070411028747e-8;
>> >>
>> >> + =A0 =A0private static final int EXP_INT_TABLE_MAX_INDEX =3D 750;
>> >> + =A0 =A0private static final int EXP_INT_TABLE_LEN =3D EXP_INT_TABLE=
_MAX_INDEX * 2;
>> >> +
>> >> =A0 =A0 =A0/** Exponential evaluated at integer values,
>> >> - =A0 =A0 * exp(x) =3D =A0expIntTableA[x + 750] + expIntTableB[x+750]=
.
>> >> + =A0 =A0 * exp(x) =3D =A0expIntTableA[x + EXP_INT_TABLE_MAX_INDEX] +=
 expIntTableB[x+EXP_INT_TABLE_MAX_INDEX].
>> >> =A0 =A0 =A0 */
>> >> - =A0 =A0private static final double EXP_INT_TABLE_A[] =3D new double=
[1500];
>> >> + =A0 =A0private static final double EXP_INT_TABLE_A[] =3D
>> >> + =A0 =A0{
>> >> + =A0 =A0 =A0 =A0+0.0d,
>> >> + =A0 =A0 =A0 =A0Double.NaN,
>> >
>> > [More than 6000 lines stripped.]
>> >
>> > Wouldn't it be advantageous to store those tabulated data in separate
>> > Java files? E.g.
>> > ---
>> > class ExpIntTables {
>> > =A0 =A0static final double[] A =3D {
>> > =A0 =A0 =A0// Very long table.
>> > =A0 =A0};
>> > =A0 =A0static final double[] B =3D {
>> > =A0 =A0 =A0// ...
>> > =A0 =A0};
>> > ---
>> >
>> > And in "FastMath.java":
>> > ---
>> > public class FastMath {
>> > =A0 =A0private static final double[] EXP_INT_TABLE_A =3D ExpIntTables.=
A;
>> > =A0 =A0private static final double[] EXP_INT_TABLE_B =3D ExpIntTables.=
B;
>> > }
>> > ---
>>
>> That would be possible, but would require the tables to be
>> non-private, increasing the theoretical risk of accidental changes.
>
> I know, but having "package" access level reduces the risk to our own
> mistakes, similar to any bug we can introduce when writing code.

Except that there are now many more files that may need to be debugged.

> One can also assumes that the tables =A0will need much less changes (if
> at all) than actual code in "FastMath.java"; so having them in separate
> files could reduce the risk of messing with them... And of course, there =
is
> the practical advantage of not having to load/scroll 6000+ lines in orde =
to
> have a look at the code.

We could move the tables and disabled setup code to the end of the source f=
ile.

Some of the setup code could be moved into a separate class - e.g. the
slowExp, slowSin and SlowCos methods use the FACT table, which is not
needed elsewhere.

> By the way, how much faster is now the first use of a method from
> "FastMath"?

Not checked. Perhaps the OP of the JIRA will be able to help there.

That probably needs to be the next step.

>
> Regards,
> Gilles
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org