Difference when calling guile script from C vs interpreter

[Jean Rene Dawin]

Hi,

when following guile script: 
______ gp.guile _________

(use-modules (statprof))
(use-modules (system vm program))

;Scalar product
(define (sp a b)
   (+ 
      (* (car a)(car b)) 
      (* (cdr a)(cdr b))
   )
)
(define size 444100)
(define (run)
   (let ((accu 0.001) 
         (r (make-f32vector size 0.0))
        )
         (let lp ((j 0) )
            (when (< j size)
               (set! accu (+ accu 0.001))
               (f32vector-set! r j (sp (cons accu (* 2 accu)) (cons (* 3 accu) 
(* 5 accu))))
               (lp (+ j 1) )
            )
         )
         (display (f32vector-ref r (- size 1)))(newline)
   )
)
(statprof run)
(display (program? run))(newline)

______________________________

is run from the interpreter, the output shows the following:

scheme@(guile-user)> (load "gp.guile")
;;; note: source file /home/gp.guile
;;;       newer than compiled 
/home/.cache/guile/ccache/2.2-LE-8-3.A/home/gp.guile.go
;;; compiling /home/gp.guile
;;; compiled /home/.cache/guile/ccache/2.2-LE-8-3.A/home/gp.guile.go
2563934.0
%     cumulative   self             
time   seconds     seconds  procedure
 50.00      0.33      0.16  /home/gp.guile:12:0:run
 28.57      0.09      0.09  %after-gc-thunk
 21.43      0.14      0.07  /home/gp.guile:5:0:sp
  0.00      0.09      0.00  anon #x7fe262d41380
---
Sample count: 14
Total time: 0.327908556 seconds (0.121140347 seconds in GC)
#t



When the same script is loaded from following C program:
_____ gp.c _______
#include <libguile.h>

int main(int argc, char **argv)
{
   scm_init_guile();
   scm_c_primitive_load("/home/gp.guile");
   return 0;
}
___________________

the result looks like this:

$ gcc -o gp gp.c $(guile-config compile) $(guile-config link)
$ ./gp 
2563934.0
%     cumulative   self             
time   seconds     seconds  procedure
 15.60      0.30      0.30  ice-9/eval.scm:226:7
 12.84      0.59      0.25  ice-9/eval.scm:159:9
 12.84      0.25      0.25  ice-9/eval.scm:182:7
  6.42      3.49      0.12  ice-9/eval.scm:618:6
  6.42      0.55      0.12  ice-9/eval.scm:159:9
  5.50      0.25      0.11  ice-9/eval.scm:625:6
  5.50      0.11      0.11  ice-9/eval.scm:123:11
  3.67      0.07      0.07  ice-9/eval.scm:282:4
  3.67      0.07      0.07  ice-9/eval.scm:124:11
  3.67      0.07      0.07  srfi/srfi-4.scm:85:20:f32vector-set!
  2.75      0.53      0.05  ice-9/eval.scm:159:9
  2.75      0.20      0.05  ice-9/eval.scm:191:12
  2.75      0.12      0.05  ice-9/eval.scm:263:9
  2.75      0.07      0.05  ice-9/eval.scm:155:9
  1.83      1.10      0.04  ice-9/eval.scm:163:9
  1.83      0.46      0.04  ice-9/eval.scm:159:9
  1.83      0.04      0.04  ice-9/eval.scm:336:13
  1.83      0.04      0.04  ice-9/eval.scm:222:7
  1.83      0.04      0.04  ice-9/eval.scm:590:16
  1.83      0.04      0.04  ice-9/eval.scm:333:13
  0.92      0.02      0.02  ice-9/eval.scm:273:7
  0.92      0.02      0.02  ice-9/eval.scm:224:11
  0.00      0.05      0.00  ice-9/eval.scm:155:9
---
Sample count: 109
Total time: 1.939099187 seconds (0.570765622 seconds in GC)
#t

Is this difference expected?
It is 6 times slower when called from C versus from the interpreter.
And the statprof output of the C version shows calls in ice-9/eval.scm
but from the interpreter it doesn't. As I'm new to guile, any
hints would be helpful.

This is GNU Guile 2.2.6 on x86_64 GNU/Linux.

Regards,
Jean Rene Dawin