division of integers by floats

division of integers by floats

[Richard Stallman]

It seems like this is the best change to make.  Does it give correct
results?


*** data.c	27 Apr 2004 04:36:00 -0400	1.235
--- data.c	05 May 2004 14:09:22 -0400	
***************
*** 2520,2525 ****
--- 2520,2533 ----
      default:
        break;
      }
+ 
+   /* For division, if any arg is a float, do the computation
+      in floating point.  */
+   if (code == Adiv)
+     for (argnum = 0; argnum < nargs; argnum++)
+       if (FLOATP (args[argnum]))
+ 	return float_arith_driver ((double) accum, 0, code,
+ 				   nargs, args);
  
    for (argnum = 0; argnum < nargs; argnum++)
      {

Re: division of integers by floats

[Peter Whaite]

Richard Stallman <rms@gnu.org> wrote:
> It seems like this is the best change to make.  Does it give correct
> results?

It does with limited testing, but having it in the arith_driver like
that means the argument list gets unnecessarily scanned for floats when
there are 2 arguments.

Here's a patch to put the change in the DEFUN.  Note that there is no
effective change for the two arg form, so its still efficient and there
wont be any "surprises" for the dominant usage.

--- data.c      6 May 2004 00:10:15 -0000       1.237
+++ data.c      6 May 2004 20:24:25 -0000
@@ -2698,7 +2698,13 @@
      int nargs;
      Lisp_Object *args;
 {
-  return arith_driver (Adiv, nargs, args);
+  int argnum;
+  if (nargs == 2)
+    return arith_driver (Adiv, nargs, args);
+  for (argnum = 0; argnum < nargs; argnum++)
+    if (FLOATP (args[argnum]))
+      return float_arith_driver (0, 0, Adiv, nargs, args);
+  return arith_driver (Adiv, nargs, args);
 }
 
 DEFUN ("%", Frem, Srem, 2, 2, 0,


-- 
Peter Whaite (http://whaite.ca)

Re: division of integers by floats

[Richard Stallman]

    Here's a patch to put the change in the DEFUN.  Note that there is no
    effective change for the two arg form, so its still efficient and there
    wont be any "surprises" for the dominant usage.

Your change is ok too.
(Could someone install it, as a "tiny change"?)

--- data.c      6 May 2004 00:10:15 -0000       1.237
+++ data.c      6 May 2004 20:24:25 -0000
@@ -2698,7 +2698,13 @@
      int nargs;
      Lisp_Object *args;
 {
-  return arith_driver (Adiv, nargs, args);
+  int argnum;
+  if (nargs == 2)
+    return arith_driver (Adiv, nargs, args);
+  for (argnum = 0; argnum < nargs; argnum++)
+    if (FLOATP (args[argnum]))
+      return float_arith_driver (0, 0, Adiv, nargs, args);
+  return arith_driver (Adiv, nargs, args);
 }
 
 DEFUN ("%", Frem, Srem, 2, 2, 0,

Re: division of integers by floats

[Piet van Oostrum]

>>>>> Richard Stallman <rms@gnu.org> (RS) wrote:

RS>     Here's a patch to put the change in the DEFUN.  Note that there is no
RS>     effective change for the two arg form, so its still efficient and there
RS>     wont be any "surprises" for the dominant usage.

RS> Your change is ok too.
RS> (Could someone install it, as a "tiny change"?)

RS> --- data.c      6 May 2004 00:10:15 -0000       1.237
RS> +++ data.c      6 May 2004 20:24:25 -0000
RS> @@ -2698,7 +2698,13 @@
RS>       int nargs;
RS>       Lisp_Object *args;
RS>  {
RS> -  return arith_driver (Adiv, nargs, args);
RS> +  int argnum;
RS> +  if (nargs == 2)
RS> +    return arith_driver (Adiv, nargs, args);
RS> +  for (argnum = 0; argnum < nargs; argnum++)
RS> +    if (FLOATP (args[argnum]))
RS> +      return float_arith_driver (0, 0, Adiv, nargs, args);
RS> +  return arith_driver (Adiv, nargs, args);
RS>  }

This can even be combined as:

+  int argnum;
+  for (argnum = 2; argnum < nargs; argnum++)
+    if (FLOATP (args[argnum]))
+      return float_arith_driver (0, 0, Adiv, nargs, args);
+  return arith_driver (Adiv, nargs, args);

-- 
Piet van Oostrum <piet@cs.uu.nl>
URL: http://www.cs.uu.nl/~piet [PGP]
Private email: P.van.Oostrum@hccnet.nl

Re: division of integers by floats

[Juanma Barranquero]

On 08 May 2004 12:28:12 +0200, Piet van Oostrum <piet@cs.uu.nl> wrote:

> +  int argnum;
> +  for (argnum = 2; argnum < nargs; argnum++)
> +    if (FLOATP (args[argnum]))
> +      return float_arith_driver (0, 0, Adiv, nargs, args);
> +  return arith_driver (Adiv, nargs, args);

Installed.

                                                           /L/e/k/t/u

Re: division of integers by floats

[Peter Whaite]

Piet van Oostrum <piet@cs.uu.nl> wrote:
> This can even be combined as:
> 
> +  int argnum;
> +  for (argnum = 2; argnum < nargs; argnum++)
> +    if (FLOATP (args[argnum]))
> +      return float_arith_driver (0, 0, Adiv, nargs, args);
> +  return arith_driver (Adiv, nargs, args);

Thats a nice optimization.

-- 
Peter Whaite (http://whaite.ca)