Static arrays problem

[Denis Koroskin]

On Thu, 18 Dec 2008 00:52:43 +0300, bearophile <bearophileHUGS at lycos.com>  
wrote:

> At the bottom of this post there are two little test programs that may  
> show a problem with static arrays.
>
> D code compiled with DMD 1.038:
> dmd test_d.d
> Compile time: 8.29 s
> Resulting exe and obj:
> 17.756.188 test_d.exe
>      2.390 test_d.map
> 17.734.261 test_d.obj
>
>
> C code compiled with DMC 8.42n:
> dmc test_c.c
> Compile time: 0.04 s
> Resulting exe:
>  51.740 test_c.exe
>
>
> C code compiled with GCC (4.2.1):
> gcc test_c.c -o test_c
> Compile time: 0.16 s
> Resulting exe:
>     16.781 test_c.exe
>
>
> C code compiled with LLVM-GCC (2.5):
> llvm-gcc test_c.c -o test_c
> Compile time: 0.11 s
> Resulting exe:
>     13.462 test_c.exe
>
>
> (The following code is just reduced toy, but the full program runs up to  
> about 13 times faster with static arrays, so I'd like to use them  
> instead of the dynamic ones.)
>
> Using obj2asm on the test_d.obj it's made mostly of:
>
> _D6test_d1aG3G65G65G130f:
> 	db	000h,000h,0c0h,07fh,000h,000h,0c0h,07fh
> 	db	000h,000h,0c0h,07fh,000h,000h,0c0h,07fh
> 	db	000h,000h,0c0h,07fh,000h,000h,0c0h,07fh
> 	db	000h,000h,0c0h,07fh,000h,000h,0c0h,07fh
> ...
>
> -------------------------
>
> // C code:
>
> #include "stdio.h"
>
> #define N 65
> #define M (N+N)
>
> float a[3][N][N][M];
> float b[4][N][N][M];
> float c[N][N][M];
>
> int main() {
>     int i, j, k;
>
>     for (i = 0; i < N; i++)
>         for (j = 0; j < N; j++)
>             for (k = 0; k < M; k++) {
>                 a[0][i][j][k] = 0.0;
>                 a[1][i][j][k] = 0.0;
>                 a[2][i][j][k] = 0.0;
>                 b[0][i][j][k] = 0.0;
>                 b[1][i][j][k] = 0.0;
>                 b[2][i][j][k] = 0.0;
>                 b[3][i][j][k] = 0.0;
>                 c[i][j][k] = 0.0;
>             }
>
>     printf("%f %f %f\n", a[2][10][10][10], a[2][10][10][10],  
> c[10][10][10]);
>     return 0;
> }
>
> -------------------------
>
> // D code, it's the same,
> // I have just changed the first 3 lines.
>
> import std.c.stdio: printf;
>
> const int N = 65;
> const int M = (N+N);
>
> float a[3][N][N][M];
> float b[4][N][N][M];
> float c[N][N][M];
>
> int main() {
>     int i, j, k;
>
>     for (i = 0; i < N; i++)
>         for (j = 0; j < N; j++)
>             for (k = 0; k < M; k++) {
>                 a[0][i][j][k] = 0.0;
>                 a[1][i][j][k] = 0.0;
>                 a[2][i][j][k] = 0.0;
>                 b[0][i][j][k] = 0.0;
>                 b[1][i][j][k] = 0.0;
>                 b[2][i][j][k] = 0.0;
>                 b[3][i][j][k] = 0.0;
>                 c[i][j][k] = 0.0;
>             }
>
>     printf("%f %f %f\n", a[2][10][10][10], a[2][10][10][10],  
> c[10][10][10]);
>     return 0;
> }
>
> Bye,
> bearophile


This is a linker issue. DMD stores a, b,c AND a.init, b.init, c.init  
(doubling data size).
Intelligent linker could strip a.init, b.init and c.init if they are never  
accessed.