Help with writing assembly..

One thing - you have two nested loops when one loop would be enough.
Two - loop unrolling 2 or maybe 4 pixels/iteration reduces the cost of the loop.
Three - the LPC23xx doesn't require you to first clear and then set pins values - you can do a direct assign of the 8 bits. Also remember that the LPC23xx supports a mask operation that allows a 32-bit wide write to the port registers to only affect your high 8 bits.
Four - make sure all your constants are in local variables and turn on max optimization.
Five - check what peripherial clock speed you have selected for GPIO. That affects how fast the processor can handle accesses to the GPIO subsystem.
Six - check that the LPC23xx port register can be direct-accessed as a pointer. Or consider using an explicit pointer.

Recompile and check the difference.

Hi Per,
Thanks for your reply,

One - The time change for the nested loops can be ignored.

Two - Can't be done in this hardware.

Three - the LPC23xx force me to first clear and then set pins values, there is no command to change bits. Mask don't help here..

Four - all my constants are in local variables, and i turn on max optimization (-Ot).

Five - Good advice - I Will check.

Six - No - it cant...

Thanks again,
Doron

1) A nested loop consumes more registers which just might matter.

2) You have magic hardware that doesn't allow you to duplicate the inner loop block 2 or 4 times - even when you after rewrite of the code has new code that is way smaller than your current code? Sorry - not sure I believe you.

3) You haven't read the processor user manual. Now would be a very good time to read chapter 10 of the manual. Especially section 5.4 about FIOxPIN registers.

Important sentences to think about:
"Writing to the IOPIN register stores the value in the port output register, bypassing the
need to use both the IOSET and IOCLR registers to obtain the entire written value."

"Access to a port pin via the FIOPIN register is conditioned by the corresponding bit of the FIOMASK register (see Section 10–5.5 “Fast GPIO port Mask register FIOMASK(FIO[0/1/2/3/4]MASK - 0x3FFF C0[1/3/5/7/9]0)”)."

4) You haven't shown us what exactly your constants look like. It is always (!) a good idea to supply the full information to let a reader actually be able to compile a problem function without having to guess.

We especially do not know what ILI_DAT_CLR, ILI_DAT_SET, ILI_WR_0, ILI_WR_1 actually looks like.

6) No you can't? Care to explain?

If you have clock bit on same port as the data:

set_chipsel();
set_port_mask_for_clock_and_data();
data1 = (data & 0xff00) << 16;
data1_b = data1 | clock_bit;
data2 = (data & 0xff) << 24;
data2_b = data2 | clock_bit;
fiopin = addr_fpin_port;
for (i = 0; i < 320*200; i++) {
    *fiopin = data1;
    *fiopin = data1_b;
    *fiopin = data1;
    *fiopin = data2;
    *fiopin = data2_b;
    *fiopin = data2;
}
clear_port_mask();
clear_chipsel();

Alternatively - note that the LPC23xx also allows 8-bit and 16-bit writes. So you might be able to use FIOxPIN3 to write just the high 8 bits without use of mask. Or use FIOxPINU to access top 16 bits and leave low 16 bits free from mask operation.

If the clock bit is on other port - or you use 8-bit or 16-bit operation to write the data then you could have an inner loop like:

    *fiopin_1 = data1;
    *fiopin_2 = 1;
    *fiopin_2 = 0;
    *fiopin_1 = data2;
    *fiopin_2 = 1;
    *fiopin_2 = 0;

or
<per> *fiopin_1 = data1; *fioset = clockbit; *fioclear = clockbit; *fiopin_1 = data2; *fioset = clockbit; *fioclear = clockbit;

Damn - last should have been:

    *fiopin_1 = data1;
    *fioset = clockbit;
    *fioclear = clockbit;
    *fiopin_1 = data2;
    *fioset = clockbit;
    *fioclear = clockbit;

And a footnote - depending on the operation of the clock bit, you might be able to shrink 6 writes into 4 writes inside the loop.

Hi Per,
You open my eyes to many new ideas...
Let me try all your suggestions - I'm sure it will improve my code,
Thanks,
Doron

Hi Per,
Many thanks for all the suggestion - now it looks great
Doron

Nice. How much did you manage to reduce the assembler output of the inner loop?

Hi Per,
The final code reduced to 6 lines - as you recomended - with pointer.
The WR bit is on the same port

                        *io_port=V_M_1;
                        *io_port=V_M_0;
                        *io_port=V_M_1;

                        *io_port=V_L_1;
                        *io_port=V_L_0;
                        *io_port=V_L_1;

The result is so fast.
Thanks,
Doron