small formula rendering tests


In the context (but limited to) of apache server with tcl cgi scripts running on safe user I've done some small tests to compare signal processing loops in tcl with C, primarily speed-wise. Of course a loop is not the same as computations, and a compilation not the same as a free run-time interpreted program such as tcl can handle, and moreover tcl/tk scripts can very well drive a C compiler to make it's use more comfortable.

I can get pretty fast not-so-big C compiles running on moderately fast machines running Linux, expecially when using a Ram disk for temporary data storage:

  export TMP=/dev/shm
  export TEMP=/dev/shm

In a bash shell.

The first Tcl test program is very short:

 ~/Math [1008] $ cat > test1.tcl
 for {set i 0} {$i < 100000} {incr i} {set y [expr 10*$i] }
 ~/Math [1009] $ time tclsh test1.tcl 
 real    0m1.021s
 user    0m0.962s
 sys     0m0.006s

The equivalent in C, but with a 1e3 times longer loop:

 ~/Math [1053] $ cat test1.c

    int i;
    long int y = 0;
    for (i=0; i<100000000; i++) y+=10*i;

 ~/Math [1054] $ gcc -o test1 test1.c
 ~/Math [1055] $ time ./test1
 real    0m0.505s
 user    0m0.379s
 sys     0m0.001s

Ok, 2000 times faster loop, and this length is the static compiled crossover point as a result of the compilatation taking a part of 1 second.

Now with processing:

 ~/Math [1056] $ cat test2.tcl 
 set y 0;
 for {set i 0} {$i < 100000} {incr i} {set y [expr $y +  sin($i/10000)]}
 puts $y
 ~/Math [1057] $ time tclsh test2.tcl 
 real    0m3.997s
 user    0m3.173s
 sys     0m0.007s

In C:

 ~/Math [1058] $ cat test2.c


    int i;
    double y = 0;
    for (i=0; i<100000000; i++) y+=sin(i/10000);

 ~/Math [1059] $ gcc -O -o test2 test2.c -lm
 ~/Math [1060] $ time ./test2

 real    0m10.094s
 user    0m7.659s
 sys     0m0.020s
 ~/Math [1061] $ time ./test2

 real    0m8.081s
 user    0m8.030s
 sys     0m0.003s
 ~/Math [1062] $ time ./test2

 real    0m7.949s
 user    0m7.895s
 sys     0m0.009s
 ~/Math [1063] $ 

Of course the overhead of the loop starts to count less, and probably tcl used the same C library sin() call, so not 2000 but 500 times faster. (the last two time-s I switched of the TV running on the same machine).

Someone want to compare multicore or VHDL ?

Oh, another good practicum question: why is the result from the C and the tcl code exactly different?!

TV Sept 3 ´08 I add a page about a similar subject List computations in a FPGA, driven by Tcl which is interesting to compare speeds.