Question about optimization

Dear Andrew and Gregory,
Thanks for the prompt reply. I have just moved the assembly code to Linear
assembly (took me some time ... ). Generally speaking I am using the 6711 but
now I am trying the simulator as I do not have a parallel port on my laptop
(only USB ...). Per your request, I am posting the C code, the assembly code,
and the linear assembly code. Just to make it easier to understand the input is:
y[]={0.220649,
;	0.439838,
;	0.656116,
;	0.868058,
;	1.074272,
;	1.273410,
;	1.464181,
;	1.645360,
;	}
and the output should be:
x[]={0.220649,
;0.0
;	0.439838,
;	0.0
;	0.656116,
;	0.0
;	0.868058,
;	0.0
;	1.074272,
;	0.0
;	1.273410,
;	0.0
;	1.464181,
;	0.0
;	1.645360,
;	};
Linear assembly:
.global _load_array;	declare the function as global so that it ;	can ber
eferenced from C
_load_array .cproc x_addr, y_addr	;
.reg	memOffset
.reg	FFTPnts
.reg	jCnt
.reg	zeroValue	.reg	x_value
.reg	y_value
ZERO memOffset	;offset register, used for pointer aritmetic ZERO jCnt	;used for
generating the values of memOffset	ZERO zeroValue	;used for setting odd terms to
Zero
MVK 2048,FFTPnts	;numOfFFTDataPoints 48 .no_mdep	;no loop: .trip 1024, 1024
;assume between 2048/2 and 2048/2 iterations
SHL	jCnt,1,memOffset	;shift left (multiply jCnt by two-->memOffset)	ADD
1,jCnt,jCnt ;jCnt++
LDW	*y_addr++,y_value;	;load a 32byte word into y_value, postincreamnet ;the
y_addr pointer	STW y_value,*x_addr[memOffset]	;store the value into x_addr
offset by memOffset ADD 1,memOffset,memOffset	;memOffset++. memOffset is now an
odd term	STW	zeroValue,*x_addr[memOffset];store the Zero value into x_addr
offset by memOffset
;which is now an odd term
SUB	FFTPnts,2,FFTPnts ;loop decreamnet by 2
[FFTPnts]	B loop	;continue until all points computed .endproc
Assembly:
.def _load_array .text ;text section	;.ref _numOfDataPoints _load_array: ;	MVK
_numOfDataPoints,A1 MVK 1024,A1 ||	ZERO A8	;Offset register, used for pointer
aritmetic ZERO A9 ;(j)
||	ZERO B9 LOOP: SHL .S1 A9,1,A8	;shift left (multiply A9 by two-->A8)	||	ADD
.L1 1,A9,A9 ;A9++
LDW .D2	*B4++,B2;
NOP	4	STW .D1 B2,*A4[A8]	;A8 is an even term
NOP	3	ADD .L1 1,A8,A8	;A8++. A8 is now an odd term	STW .D1 B9,*A4[A8]
NOP	3	SUB .S1 A1,2,A1 ;
[A1]	B .S2	LOOP NOP	5	B	.S2	B3 NOP 5 .end C:
j=0;
for (i=0;i<2048;i+=2){
data[i]=Input_Signal[j];
j=j+1;
data[i+1]=0;
}
Thanks,
Shlomo. --- In c...@yahoogroups.com, Andrew Elder <andrew_elder@b...>
wrote:

Shlomo,
Can you post the C code here for the group to look at (and comment 

on) ?

What processor are you using ?
Consider ASM a last resort. You should be able to get the C 

compiler to generate VERY close to optimal code if you have
everything organized correctly. You should be able to get VERY VERY VERY close
to optimal with linear assembly.

- Andrew
shlomo_kashani wrote:

Hi,
1-I am implementing a loop in assembly for a code that was 

originally 

written in C. When I use the C code with the -k, (keep generated
assembly files) -mw and -o3 compiler parameter on the compiler generates the
"SOFTWARE PIPELINE INFORMATION". I know the case is 

the 

same when using Linear assembly. However, what happens if I am
writing directly in assembly? why isn't this information 

available? 

How can I pipeline the assembly code without the compilers

tips ... ?

2-I wish to reference a variable that was defined in C as:
#define numOfDataPoints 1024
In my assembly code as: .ref _numOfDataPoints
The compiler generates an error that the symbol is undefined. Declaring the
variable as: const numOfDataPoints24; solved the issue however it generated a
different error since I am using the value for defining the size of an array and
array size 

must 

be a const so the compiler did not allow it. Any ideas ?
Thanks,
Shlomo.