www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - array advantage?

reply Lucas Goss <lgoss007 gmail.com> writes:
Is there any advantage or benefit to writing a class as:

class SomeClass
{
    float number[3];
    //...
}

Instead of:

class SomeClass
{
    float number1;
    float number2;
    float number3;
    //...
}

I see a lot of C++ classes that are like the first. Could it be memory 
alignment reasons? Or maybe for speed?
Jan 06 2006
next sibling parent reply Fredrik Olsson <peylow gmail.com> writes:
Lucas Goss skrev:

 Is there any advantage or benefit to writing a class as:
 
 class SomeClass
 {
    float number[3];
    //...
 }
 
 Instead of:
 
 class SomeClass
 {
    float number1;
    float number2;
    float number3;
    //...
 }
 
 I see a lot of C++ classes that are like the first. Could it be memory 
 alignment reasons? Or maybe for speed?


Now lets see.
1. The array version is shorter to write (Imagine 1000 numbers).
2. An array logically groups related values.
3. You can do for and foreach loops over the array.
4. No speed loss with arrays.
5. Should the need arise the static array is quite easy to make dynamic.


Actually, I have a harder time seeing why I would want it the second 
way? Well except cases when the naming and grouping is set in stone and 
becomes more logical, like:
class MyPoint {
   float x, y, z;
   // ...
}

// Fredrik Olsson
Jan 07 2006
next sibling parent reply Lars Ivar Igesund <larsivar igesund.net> writes:
Fredrik Olsson wrote:


 Lucas Goss skrev:

 Is there any advantage or benefit to writing a class as:
 
 class SomeClass
 {
    float number[3];
    //...
 }
 
 Instead of:
 
 class SomeClass
 {
    float number1;
    float number2;
    float number3;
    //...
 }
 
 I see a lot of C++ classes that are like the first. Could it be memory
 alignment reasons? Or maybe for speed?


 Now lets see.
 1. The array version is shorter to write (Imagine 1000 numbers).
 2. An array logically groups related values.
 3. You can do for and foreach loops over the array.
 4. No speed loss with arrays.
 5. Should the need arise the static array is quite easy to make dynamic.
 
 
 Actually, I have a harder time seeing why I would want it the second
 way? Well except cases when the naming and grouping is set in stone and
 becomes more logical, like:
 class MyPoint {
    float x, y, z;
    // ...
 }
 
 // Fredrik Olsson


And even then you use arrays, just having

class MyPoint {

   float _point[3];

   float x() { return _point[0]; }
   void x(float newx) { _point[0] = newx; }
   etc;
}

Lars Ivar Igesund
Jan 07 2006
parent pragma <pragma_member pathlink.com> writes:
In article <dpo5ud$3m3$1 digitaldaemon.com>, Lars Ivar Igesund says...

Fredrik Olsson wrote:


 Lucas Goss skrev:

 Is there any advantage or benefit to writing a class as:
 
 class SomeClass
 {
    float number[3];
    //...
 }
 
 Instead of:
 
 class SomeClass
 {
    float number1;
    float number2;
    float number3;
    //...
 }
 
 I see a lot of C++ classes that are like the first. Could it be memory
 alignment reasons? Or maybe for speed?


 Now lets see.
 1. The array version is shorter to write (Imagine 1000 numbers).
 2. An array logically groups related values.
 3. You can do for and foreach loops over the array.
 4. No speed loss with arrays.
 5. Should the need arise the static array is quite easy to make dynamic.
 
 
 Actually, I have a harder time seeing why I would want it the second
 way? Well except cases when the naming and grouping is set in stone and
 becomes more logical, like:
 class MyPoint {
    float x, y, z;
    // ...
 }
 
 // Fredrik Olsson


And even then you use arrays, just having

class MyPoint {

   float _point[3];

   float x() { return _point[0]; }
   void x(float newx) { _point[0] = newx; }
   etc;
}

Lars Ivar Igesund


Or just exploit D's "array properties feature" and remove the class completely:


   alias float[3] MyPoint;
   float x(MyPoint _point) { return _point[0]; }
   float x(MyPoint _point,float newx) { _point[0] = newx; return newx; }
   /* ... */
   MyPoint p;
   p.x = 5.5;


- EricAnderton at yahoo
Jan 07 2006
prev sibling next sibling parent Tom S <h3r3tic remove.mat.uni.torun.pl> writes:
Fredrik Olsson wrote:

 Actually, I have a harder time seeing why I would want it the second 
 way? Well except cases when the naming and grouping is set in stone and 
 becomes more logical, like:
 class MyPoint {
   float x, y, z;
   // ...
 }


In that case we can have the best of both worlds:

struct vec3
{
     union
     {
         struct {
             float x, y, z;
         }
         struct {
             float r, g, b;
         }
         float[3] cell;
     }

     // ...
}

Sometimes it's useful to perform operations on just one component of 
R^3. If the 'cell' index can be calculated earlier and reused, code 
duplication can be avoided


-- 
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCS/M d-pu s+: a-->----- C+++$>++++ UL P+ L+ E--- W++ N++ o? K? w++ !O 
!M V? PS- PE- Y PGP t 5 X? R tv-- b DI- D+ G e>+++ h>++ !r !y
------END GEEK CODE BLOCK------

Tomasz Stachowiak  /+ a.k.a. h3r3tic +/
Jan 07 2006
prev sibling parent reply Hasan Aljudy <hasan.aljudy gmail.com> writes:
Fredrik Olsson wrote:

 Lucas Goss skrev:
 

 Is there any advantage or benefit to writing a class as:

 class SomeClass
 {
    float number[3];
    //...
 }

 Instead of:

 class SomeClass
 {
    float number1;
    float number2;
    float number3;
    //...
 }

 I see a lot of C++ classes that are like the first. Could it be memory 
 alignment reasons? Or maybe for speed?


 
 Now lets see.
 1. The array version is shorter to write (Imagine 1000 numbers).
 2. An array logically groups related values.
 3. You can do for and foreach loops over the array.
 4. No speed loss with arrays.
 5. Should the need arise the static array is quite easy to make dynamic.
 
 
 Actually, I have a harder time seeing why I would want it the second 
 way? Well except cases when the naming and grouping is set in stone and 
 becomes more logical, like:
 class MyPoint {
   float x, y, z;
   // ...
 }
 
 // Fredrik Olsson


I think he's referring to using arrays to group non-related things, 
because, come to think of it, if the things were logically related, you 
would use a dynamic array.
Jan 07 2006
next sibling parent Fredrik Olsson <peylow gmail.com> writes:
Hasan Aljudy skrev:

 Fredrik Olsson wrote:
 

 Lucas Goss skrev:


 Is there any advantage or benefit to writing a class as:






<snip>

 I see a lot of C++ classes that are like the first. Could it be 
 memory alignment reasons? Or maybe for speed?






<snip>

 
 I think he's referring to using arrays to group non-related things, 
 because, come to think of it, if the things were logically related, you 
 would use a dynamic array.


For grouping things not logically related I see no reason at all to use 
arrays, a compiler not s good at optimization might even do worse with 
an array. But more importantly, it is against KISS, keep it simple stupid.

An array is an ordered list of a single type of data, using it for 
anything else is inviting trouble if you ask me. If you put temperature 
and stock changes in the same array, you will have to rely on external 
mechanism, or even worse; coding practices, for knowing what is what. 
Sure you can say element 0 IS the temp, and element 1 IS the stocks. A 
good recipe for writing solid and maintainable code is to make it hard 
to write the code wrong. "Hmm... was the temperature with offset 0 or 
1?", is a probably question to rise, and a good one. "Hmm... was the 
temperature in temp or stockRate?" Os not quite as likely, and the 
person asking it should be fired on the spot.

Now if I was only better at coding as I preach :).

// Fredrik Olsson
Jan 07 2006
prev sibling parent reply Lucas Goss <lgoss007 gmail.com> writes:
Hasan Aljudy wrote:

 Fredrik Olsson wrote:
 

 Actually, I have a harder time seeing why I would want it the second 
 way? Well except cases when the naming and grouping is set in stone 
 and becomes more logical, like:
 class MyPoint {
   float x, y, z;
   // ...
 }

 // Fredrik Olsson


 
 
 I think he's referring to using arrays to group non-related things, 
 because, come to think of it, if the things were logically related, you 
 would use a dynamic array.


Well I was sort of looking for a general reason since I have other 
classes that are structured differently, but I'm actually working on a 
vector class at the moment. So yeah I could do an array like:


 Lars Ivar Igesund wrote:
 class MyPoint {

    float _point[3];

    float x() { return _point[0]; }
    void x(float newx) { _point[0] = newx; }
    etc;
 }


Of course keeping the "_point" private, and just allowing the x, y, and 
z properties. But why do that if the properties do nothing but set the 
value, they might as well be public:

class MyPoint {
public:
    float x, y, z;
    ...
}

Unless there is some benefit to the array? Especially if it's speed. I 
know in C++ classes (and C structs) there is no guarantee to how the 
members are aligned. In D?

Now there may be other benefits to using the array that might come in 
handy that I don't see till later... I was just trying to plan ahead :)
Jan 07 2006
parent reply BCS <BCS_member pathlink.com> writes:
Lucas Goss wrote:

 Well I was sort of looking for a general reason since I have other 
 classes that are structured differently, but I'm actually working on a 
 vector class at the moment. So yeah I could do an array like:



If you are looking to do 3D math I have a fairly compete library.

http://www.cs.uidaho.edu/~temp0092/point_math.d

This is an older version (the newer one is on a differnt system), I will post 
the newer one in a few hours.
Jan 07 2006
parent reply Lucas Goss <lgoss007 gmail.com> writes:
BCS wrote:

 
 If you are looking to do 3D math I have a fairly compete library.
 
 http://www.cs.uidaho.edu/~temp0092/point_math.d
 
 This is an older version (the newer one is on a differnt system), I will 
 post the newer one in a few hours.


Thanks! (sorry I was a little slow to respond).

My vector class actually looked pretty similar (which means I must be 
doing something right :) ), though I'm using templates so that you can 
use float, double, or real. I'd post mine but I no longer have a web 
site (I'll have to remedy that soon).
Jan 13 2006
parent reply BCS <BCS_member pathlink.com> writes:
Lucas Goss wrote:

 BCS wrote:
 

 If you are looking to do 3D math I have a fairly compete library.

 http://www.cs.uidaho.edu/~temp0092/point_math.d

 This is an older version (the newer one is on a differnt system), I 
 will post the newer one in a few hours.


 
 
 Thanks! (sorry I was a little slow to respond).
 
 My vector class actually looked pretty similar (which means I must be 
 doing something right :) ), though I'm using templates so that you can 
 use float, double, or real. I'd post mine but I no longer have a web 
 site (I'll have to remedy that soon).


Making it into a template ought not to be to bad as the actual type is done 
using a alias. (if you don't see that in the version you have, I put up a more 
recent version a while back).

I have some more code that can go along with it (matrix transformations etc.), 
If i can find it I'll put it up over the weekend.
Jan 13 2006
parent reply Lucas Goss <lgoss007 gmail.com> writes:
BCS wrote:

 
 Making it into a template ought not to be to bad as the actual type is 
 done using a alias. (if you don't see that in the version you have, I 
 put up a more recent version a while back).
 


Yeah, that's how I'm doing it. I have:
alias TVector3!(float) Vector3f;
alias TVector3!(double) Vector3d;
alias TVector3!(real) Vector3;


 I have some more code that can go along with it (matrix transformations 
 etc.), If i can find it I'll put it up over the weekend.


I'd like to see it if you can find it. I'm doing my matrix stuff now and 
I have part of it done, but I'm not positive about what all methods to 
include (I have all the operators done). Currently I have: adjoint, 
determinant, (get/set)Row, (get/set)Column, inverse, quadraticForm, 
transpose, and orthonormalize. Right now I'm getting ready to add 
eigenDecomposition and after that some euler conversions.
Jan 14 2006
parent reply BCS <BCS_member pathlink.com> writes:
In article <dqb9jm$2t6$1 digitaldaemon.com>, Lucas Goss says...

BCS wrote:

 I have some more code that can go along with it (matrix transformations 
 etc.), If i can find it I'll put it up over the weekend.


I'd like to see it if you can find it. I'm doing my matrix stuff now and 
I have part of it done, but I'm not positive about what all methods to 
include (I have all the operators done). Currently I have: adjoint, 
determinant, (get/set)Row, (get/set)Column, inverse, quadraticForm, 
transpose, and orthonormalize. Right now I'm getting ready to add 
eigenDecomposition and after that some euler conversions.


I was sticking to just the geomectric stuff like transformations (translate ,
rotate, skew), inverse and combining. The sticky one has been inverse (To many
special cases)
Jan 14 2006
parent reply Lucas Goss <lgoss007 gmail.com> writes:
 In article <dqb9jm$2t6$1 digitaldaemon.com>, Lucas Goss says...
 
 I was sticking to just the geomectric stuff like transformations (translate ,
 rotate, skew), inverse and combining. The sticky one has been inverse (To many
 special cases)
 


Ah, I'm modeling mine off of David Eberly's books. For the inverse I 
just use cofactors (courtesy of David Eberly :) ).
Jan 14 2006
parent reply BCS <BCS_member pathlink.com> writes:
In article <dqcc8u$vqr$1 digitaldaemon.com>, Lucas Goss says...

Ah, I'm modeling mine off of David Eberly's books. For the inverse I 
just use cofactors (courtesy of David Eberly :) ).


I I'm using row ops (not sure what the method is called)

I've got most of it working and tested, I'll get it posted sometime tomorrow.
BTW, is the point3d struct useful enough to merit inclusion in phobos and/or
aries?
Jan 16 2006
parent reply Tom S <h3r3tic remove.mat.uni.torun.pl> writes:
BCS wrote:

 In article <dqcc8u$vqr$1 digitaldaemon.com>, Lucas Goss says...
 

Ah, I'm modeling mine off of David Eberly's books. For the inverse I 
just use cofactors (courtesy of David Eberly :) ).


 
 
 I I'm using row ops (not sure what the method is called)


I believe it's Gaussian elimination :)

BTW, I've also got a 3d math module: http://158.75.59.9/~h3/tmp/maths.d


-- 
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCS/M d-pu s+: a-->----- C+++$>++++ UL P+ L+ E--- W++ N++ o? K? w++ !O 
!M V? PS- PE- Y PGP t 5 X? R tv-- b DI- D+ G e>+++ h>++ !r !y
------END GEEK CODE BLOCK------

Tomasz Stachowiak  /+ a.k.a. h3r3tic +/
Jan 17 2006
next sibling parent BCS <BCS_member pathlink.com> writes:
Tom S wrote:

 BCS wrote:
 

 In article <dqcc8u$vqr$1 digitaldaemon.com>, Lucas Goss says...


 Ah, I'm modeling mine off of David Eberly's books. For the inverse I 
 just use cofactors (courtesy of David Eberly :) ).




 I I'm using row ops (not sure what the method is called)


 
 
 I believe it's Gaussian elimination :)
 
 BTW, I've also got a 3d math module: http://158.75.59.9/~h3/tmp/maths.d
 
 


It's up at http://www.cs.uidaho.edu/~temp0092/point_lib.d

Consider it bata (a.k.a. Probably has a few bugs). The unittest doesn't cover 
all cases and omits a few functions all together. Also I have a few function I 
might add later
Jan 17 2006
prev sibling parent reply Lucas Goss <lgoss007 gmail.com> writes:
Tom S wrote:

 BCS wrote:

 In article <dqcc8u$vqr$1 digitaldaemon.com>, Lucas Goss says...


 Ah, I'm modeling mine off of David Eberly's books. For the inverse I 
 just use cofactors (courtesy of David Eberly :) ).



 I I'm using row ops (not sure what the method is called)


 
 I believe it's Gaussian elimination :)
 
 BTW, I've also got a 3d math module: http://158.75.59.9/~h3/tmp/maths.d
 
 


I'm  curious, what do you have in:
	import core.common;
	import core.serializer;

I see this is probably pulled from core.serializer:
	mixin Serializer.RawStruct;

Would you be open to showing those as well?
Jan 30 2006
parent reply Tom S <h3r3tic remove.mat.uni.torun.pl> writes:
Lucas Goss wrote:

 I'm  curious, what do you have in:
     import core.common;
     import core.serializer;
 
 I see this is probably pulled from core.serializer:
     mixin Serializer.RawStruct;
 
 Would you be open to showing those as well?


Here's the rest of my engine: 
http://158.75.59.9/~h3/code/heresy/backup/src-2006-01-28_0.rar

It doesn't currently compile as I'm in the middle of a refactoring process.



-- 
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCS/M d-pu s+: a-->----- C+++$>++++ UL P+ L+ E--- W++ N++ o? K? w++ !O 
!M V? PS- PE- Y PGP t 5 X? R tv-- b DI- D+ G e>+++ h>++ !r !y
------END GEEK CODE BLOCK------

Tomasz Stachowiak  /+ a.k.a. h3r3tic +/
Jan 30 2006
parent reply BCS <BCS_member pathlink.com> writes:
I just put an update of my 3d vector library up. Bug fixes better docs etc.

http://www.cs.uidaho.edu/~temp0092/point_lib.d
http://www.cs.uidaho.edu/~temp0092/point_lib.html
http://www.cs.uidaho.edu/~temp0092/point_math.d
http://www.cs.uidaho.edu/~temp0092/point_math.html
Jan 30 2006
parent Lucas Goss <lgoss007 gmail.com> writes:
BCS wrote:

 I just put an update of my 3d vector library up. Bug fixes better docs etc.
 
 http://www.cs.uidaho.edu/~temp0092/point_lib.d
 http://www.cs.uidaho.edu/~temp0092/point_lib.html
 http://www.cs.uidaho.edu/~temp0092/point_math.d
 http://www.cs.uidaho.edu/~temp0092/point_math.html


Thanks!
Jan 31 2006
prev sibling parent reply nick <nick.atamas gmail.com> writes:
Lucas Goss wrote:

 Is there any advantage or benefit to writing a class as:
 
 class SomeClass
 {
    float number[3];
    //...
 }
 
 Instead of:
 
 class SomeClass
 {
    float number1;
    float number2;
    float number3;
    //...
 }
 
 I see a lot of C++ classes that are like the first. Could it be memory 
 alignment reasons? Or maybe for speed?


Referring to array elements requires a few extra operations. For 
starters, you have to LOAD the base_address. Then, depending on your 
architecture you may have to do some math (like a multiply) to figure 
out the offset address and finally perform another LOAD to get the 
actual element into memory. This is definitely worth it if you are going 
to be processing a large number of elements because the overhead 
associated with an array is ameliorated. (only only load the base 
address once and most machines tend to have address generation units).
On the other hand, if you are going to be performing several extra 
operations just to increment a couple integers then the performance may 
suffer.

With that said, I don't know if anybody did a benchmark of a vec3 
implemented with an array vs. a vec3 implemented with 3 variables.

Also, unless this class is on the absolute critical path, the 
performance gain will be less than insignificant.
Jan 29 2006
parent Oskar Linde <oskar.lindeREM OVEgmail.com> writes:
nick wrote:

 With that said, I don't know if anybody did a benchmark of a vec3 
 implemented with an array vs. a vec3 implemented with 3 variables.
 
 Also, unless this class is on the absolute critical path, the 
 performance gain will be less than insignificant.


To get the best of both worlds, the following should work:

struct vec3 {
	union {
		float number[3];
		struct {
			float x,y,z;
		}
	}
}
Jan 29 2006