For a NXP LPC2103

#define FIO_PIN  ((unsigned char *) &FIOPIN)
        unsigned short us;
	unsigned char buf[257];
	for (us = 1; us < 257; )
	    FIO_PIN[2] = buf[us++];     // byte addressing bits 16..23

 

This simulates as 41.04us. I have not tried it in real life, and I suspect that the hardware would not respond at that speed.

Looking at the ASM generated, the code could be made a lot more efficient. ARM compiles do not like byte-addressing the Fast i/o ports as a general rule.

   164:         for (us = 1; us < 257; ) 
0x00000600  E3A04001  MOV       R4,#0x00000001
0x00000604  EA000006  B         0x00000624
   165:             FIO_PIN[2] = buf[us++]; 
0x00000608  E1A00004  MOV       R0,R4
0x0000060C  E2841001  ADD       R1,R4,#0x00000001
0x00000610  E3C14801  BIC       R4,R1,#0x00010000
0x00000614  E28D1000  ADD       R1,R13,#0x00000000
0x00000618  E7D10000  LDRB      R0,[R1,R0]
0x0000061C  E51F1048  LDR       R1,[PC,#-0x0048]
0x00000620  E5C10016  STRB      R0,[R1,#0x0016]
0x00000624  E3540C01  CMP       R4,#0x00000100
0x00000628  DAFFFFF6  BLE       0x00000608

 

You could equally well do the same operation with the FIOSET, FIOCLR registers or the IOSET, IOCLR . In which case you can write to the whole 32 bit GPIO port in one go.

David.

p.s. I am NOT experienced with ARMs. There MAY be a deliberate error here.