Posted on February 09, 2006 at 07:06Hello, All.
I discovered a little bit wrong behaviour of Cosmic compiler V4.5c.1 - 20 Jun 2005.
Let's examine following functions. They differ no so much.
1)
void my_func(unsigned char i)
{
unsigned char j;
j = i;
...
}
2)
void my_func(unsigned char i)
{
unsigned char j;
_asm(''ld a,0'');
j = i;
...
}
Value of j should be in both cases the same. But in practice it is not true. Why?
Let's take a look in asm code.
1)
;void my_func(unsigned char i)
;{
switch .text
_my_func:
push a
OFST: set 257
;j = i;
ld x,s
ld (OFST+0,x),a
...
;}
2)
;void my_func(unsigned char i)
;{
switch .text
_my_func:
push a
OFST: set 257
;_asm(''ld a,0'');
ld a,0
;j = i;
ld x,s
ld (OFST+0,x),a
...
;}
The function parameter i is stored in register a and on stack.
In the 1st case as far as register a keeps value of i compiler uses it to fill variable j.
But in the 2nd case register a are „corrupted“ by _asm function. Compiler doesn't analyse this code and skips it. That's why variable j gets a wrong value, in this case allways „0“.
Ok, we can prevent such sutiation by loading register a with value of parameter i ( _asm(''pop a\n push a'') or something else).
I would like to have more nice solution.
Is it a bug or a feature?
Do any compiler keys or parameters exist to make it analyse inlined asm code?
Tnx, Luter
