Bug#456034: fenix: compile with -O1

[Javier Serrano Polo]

retitle 456034 fenix: fix TYPE_FLOAT functions
thanks

El dj 13 de 12 del 2007 a les 19:17 +0100, en/na Miriam Ruiz va
escriure:
> I guess the solution would not be to remove the optimization level, but to fix
> the bugs that might be causing that problem.

So you like the hard way. Good.

> Which functions are not working
> with -O2 and on which architectures?

The ones performing bad type conversions. I'm attaching a simple test
case for cos(). Output should be:

1.000000
0.707107
0.000000
-0.707107
-1.000000

I'm attaching the solution too. I'll leave to you the remaining
functions and the code arrangement.

program cos_test;
begin
        say(format(cos(0), 6));
        say(format(cos(45000), 6));
        say(format(cos(90000), 6));
        say(format(cos(135000), 6));
        say(format(cos(180000), 6));
end


--- fenix-0.92a.dfsg1.orig/fxi/src/i_func.c	2007-12-14 02:22:21.000000000 +0100
+++ fenix-0.92a.dfsg1/fxi/src/i_func.c	2007-12-14 03:20:43.000000000 +0100
@@ -282,7 +282,12 @@
 {
     float param = *(float *)&params[0] ;
     float result = (float)cos((double)(param*M_PI/180000.0)) ;
-    return *(int *)&result ;
+    union {
+        int i ;
+        float f ;
+    } conv ;
+    conv.f = result ;
+    return conv.i ;
 }
 static int fxi_sin (INSTANCE * my, int * params)
 {