www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Recursive template problem

reply Lars Kyllingstad <public kyllingen.NOSPAMnet> writes:
Hello,

I want to make a template that evaluates to the summed absolute values 
of the elements of an array. In other words, I want something like this:

     real[3] foo;
     real sum = absSum!(foo);

to expand (at compile time) to

     real[3] foo;
     real sum = abs(foo[0]) + abs(foo[1]) + abs(foo[2]);

So, I wrote the following templates:

     private template absSum(real[N] V, ulong N) {
         real absSum = absSum!(V, N-1) + abs(V[N-1]);
     }
     private template absSum(real[N] V, ulong M, ulong N) {
         real absSum = absSum!(M-1, V) + abs(V[M-1]);
     }
     private template absSum(real[N] V, ulong M : 1, ulong N) {
         real absSum = abs(V[0]);
     }

but on compilation I get (for each template!) the error messages

   variable N forward referenced
   Error: Integer constant expression expected instead of N
   Error: Integer constant expression expected instead of N
   Error: Integer constant expression expected instead of N
   variable N forward referenced
   Error: Integer constant expression expected instead of N
   Error: Integer constant expression expected instead of cast(ulong)N
   Error: Integer constant expression expected instead of cast(ulong)N
   Error: arithmetic/string type expected for value-parameter, not real[N]

I haven't a clue why I am getting these messages. Any tips?

Thanks,

-Lars
Jul 29 2008
next sibling parent BCS <ao pathlink.com> writes:
Reply to Lars,

 Hello,
 
 I want to make a template that evaluates to the summed absolute values
 of the elements of an array. In other words, I want something like
 this:
 
 real[3] foo;
 real sum = absSum!(foo);
 to expand (at compile time) to
 
 real[3] foo;
 real sum = abs(foo[0]) + abs(foo[1]) + abs(foo[2]);
 So, I wrote the following templates:
 
 private template absSum(real[N] V, ulong N) {
 real absSum = absSum!(V, N-1) + abs(V[N-1]);
 }
right off that should be "const real ..."
 private template absSum(real[N] V, ulong M, ulong N) {
 real absSum = absSum!(M-1, V) + abs(V[M-1]);
 }
 private template absSum(real[N] V, ulong M : 1, ulong N) {
 real absSum = abs(V[0]);
 }
 but on compilation I get (for each template!) the error messages
 
 variable N forward referenced
 Error: Integer constant expression expected instead of N
 Error: Integer constant expression expected instead of N
 Error: Integer constant expression expected instead of N
 variable N forward referenced
 Error: Integer constant expression expected instead of N
 Error: Integer constant expression expected instead of cast(ulong)N
 Error: Integer constant expression expected instead of cast(ulong)N
 Error: arithmetic/string type expected for value-parameter, not
 real[N]
I'm not shure arrays of real are allowd for template args 2 options: CTFE: real AbsSum(real[] args){ real ret=0; foreach(real r;args) if(r > 0) ret += r; else if(r < 0) ret -= r; //else 0 or nan: do nothing return ret; } tuples: template AbsSum(T...) { static if(T.length == 0) const real AbsSum = 0; else { static if(T[0] > 0 const real AbsSum = AbsSum(T[1..$]) + T[0]; else static if(T[0] < 0 const real AbsSum = AbsSum(T[1..$]) - T[0]; else const real AbsSum = AbsSum(T[1..$]); } }
Jul 29 2008
prev sibling parent reply maelp <mael.primet gmail.com> writes:
I think what he looks for is not the actual result since his array of reals
isn't const (not known at compile time), and he simply wants the actual *code*
to be generated at compile time, so none of your solution will work. I've had
similar problems of recursive template construction, so since there are several
posts, maybe we'll have some answers.

My problem was to construct a struct having k (known at compile time) elements
named (for instance)
v1 ... vk,

ie. something like

MyStruct!(5) 

=>

struct
{
  int v1, v2, v3, v4, v5 ;
}

someone could help me with this? Or is it possible to do this?

 
Jul 30 2008
next sibling parent Max Samukha <samukha voliacable.com.removethis> writes:
On Wed, 30 Jul 2008 04:39:42 -0400, maelp <mael.primet gmail.com>
wrote:

I think what he looks for is not the actual result since his array of reals
isn't const (not known at compile time), and he simply wants the actual *code*
to be generated at compile time, so none of your solution will work. I've had
similar problems of recursive template construction, so since there are several
posts, maybe we'll have some answers.

My problem was to construct a struct having k (known at compile time) elements
named (for instance)
v1 ... vk,

ie. something like

MyStruct!(5) 

=>

struct
{
  int v1, v2, v3, v4, v5 ;
}

someone could help me with this? Or is it possible to do this?

 
One of a few ways to do that: import std.stdio; import std.metastrings; template Fields(uint n) { static if (n) { mixin ("int v" ~ ToString!(n) ~ ";"); mixin Fields!(n - 1); } } struct S(uint n) { mixin Fields!(n); } void main() { S!(5) s; s.v1 = 1; s.v2 = 2; s.v5 = 3; }
Jul 30 2008
prev sibling parent reply Lars Kyllingstad <public kyllingen.NOSPAMnet> writes:
maelp wrote:
 I think what he looks for is not the actual result since his array of reals
isn't const (not known at compile time), and he simply wants the actual *code*
to be generated at compile time, so none of your solution will work. I've had
similar problems of recursive template construction, so since there are several
posts, maybe we'll have some answers.
That's right. I am using D for numerical computations, so fast-executing code is essential. Therefore, I want to use static arrays whenever possible, and instead of having to write foreach (real r; v) sum += r; or similar, i would like expressions such as sum = v[0] + v[1] + ... to be created at compile time for arrays of any given length. I was kinda hoping templates could provide a solution... -Lars
Jul 30 2008
parent BCS <ao pathlink.com> writes:
Reply to Lars,

 maelp wrote:
 
 I think what he looks for is not the actual result since his array of
 reals isn't const (not known at compile time), and he simply wants
 the actual *code* to be generated at compile time, so none of your
 solution will work. I've had similar problems of recursive template
 construction, so since there are several posts, maybe we'll have some
 answers.
 
That's right. I am using D for numerical computations, so fast-executing code is essential. Therefore, I want to use static arrays whenever possible, and instead of having to write foreach (real r; v) sum += r; or similar, i would like expressions such as sum = v[0] + v[1] + ... to be created at compile time for arrays of any given length. I was kinda hoping templates could provide a solution... -Lars
First take a look at the code here, it might give you some ideas: http://www.dsource.org/projects/scrapple/browser/trunk/bignum/bignum.d somewhere in there is a hunk of code that generate a tuple like T!(1, 2, 4, 5, ...) using a foreach over that will let you get compile time enumeration. Another point to be made is that if you are reading more than just a few numbers you may actually be ahead to use the foreach as its code will be much shorter (lower memory usage = better cache usage). As a first guess, if the unrolled code is more than a memory page long, loop.
Jul 30 2008