www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - CTFE Bug?!

reply Mehrdad <wfunction hotmail.com> writes:
I tried converting some C code I found on the internet to D.

When I run the code below:

     import std.math;
     import std.stdio;

     static double[2][] fft(in double[2][] timeDomain)
     {
         static void fft_rec(sizediff_t n, sizediff_t offset, sizediff_t 
delta, in double[2][] timeDomain, double[2][] freqDomain, double[2][] 
scratch)
         {
             if (n > 2) /* Perform recursive step. */
             {
                 const N2 = n / 2; /* half the number of points in FFT */
                 /* Calculate two (n/2)-point DFT's. */
                 fft_rec(N2, offset, 2 * delta, timeDomain, scratch, 
freqDomain);
                 fft_rec(N2, offset + delta, 2 * delta, timeDomain, 
scratch, freqDomain);

                 /* Combine the two (n/2)-point DFT's into one n-point 
DFT. */
                 foreach (k; 0 .. N2)
                 {
                     const double TWO_PI = 
6.2831853071795864769252867665590057683943L;
                     const
                         k00 = offset + k * delta,
                         k01 = k00 + N2 * delta,
                         k10 = offset + 2 * k * delta,
                         k11 = k10 + delta,
                         cs = cos(k * TWO_PI / n),
                         sn = sin(k * TWO_PI / n),
                         tmp0 = cs * scratch[k11][0] + sn * scratch[k11][1],
                         tmp1 = cs * scratch[k11][1] - sn * scratch[k11][0];
                     freqDomain[k01][0] = scratch[k10][0] - tmp0;
                     freqDomain[k01][1] = scratch[k10][1] - tmp1;
                     freqDomain[k00][0] = scratch[k10][0] + tmp0;
                     freqDomain[k00][1] = scratch[k10][1] + tmp1;
                 }
             }
             else  /* Perform 2-point DFT. */
             {
                 const k00 = offset, k01 = k00 + delta;
                 freqDomain[k01][0] = timeDomain[k00][0] - 
timeDomain[k01][0];
                 freqDomain[k01][1] = timeDomain[k00][1] - 
timeDomain[k01][1];
                 freqDomain[k00][0] = timeDomain[k00][0] + 
timeDomain[k01][0];
                 freqDomain[k00][1] = timeDomain[k00][1] + 
timeDomain[k01][1];
             }
         }

         auto freqDomain = new double[2][timeDomain.length];
         fft_rec(timeDomain.length, 0, 1, timeDomain, freqDomain, new 
double[2][2 * timeDomain.length]);
         return freqDomain;
     }
     void main()
     {
         const result = fft([[0, 1], [1, -1]]);
         pragma(msg, result);
         writeln(result);
     }


I get the following results:
Compile time: [[1,0],[1,0]]
Run time: [[1, 0], [-1, 2]]

Why is this?
Oct 29 2011
parent reply "Vladimir Panteleev" <vladimir thecybershadow.net> writes:
On Sun, 30 Oct 2011 03:39:57 +0300, Mehrdad <wfunction hotmail.com> wrote:

 I tried converting some C code I found on the internet to D.
Reduced version: import std.stdio; static double[1][] f() { auto array = new double[1][2]; array[0][0] = 1; return array; } void main() { const result = f(); pragma(msg, result); // [[1],[1]] writeln(result); // [[1], [nan]] } I don't know what's going on here. -- Best regards, Vladimir mailto:vladimir thecybershadow.net
Oct 30 2011
next sibling parent bearophile <bearophileHUGS lycos.com> writes:
Vladimir Panteleev:

 Reduced version:
Nice reduction, it looks fit for Bugzilla. Bye, bearophile
Oct 30 2011
prev sibling parent Don <nospam nospam.com> writes:
On 30.10.2011 22:42, Vladimir Panteleev wrote:
 On Sun, 30 Oct 2011 03:39:57 +0300, Mehrdad <wfunction hotmail.com> wrote:

 I tried converting some C code I found on the internet to D.
Reduced version: import std.stdio; static double[1][] f() { auto array = new double[1][2]; array[0][0] = 1; return array; } void main() { const result = f(); pragma(msg, result); // [[1],[1]] writeln(result); // [[1], [nan]] } I don't know what's going on here.
Bug 6885. I'm working on a fix.
Nov 03 2011