Incorrect Instruction Cycle Time F28335

I enabled clock to show my how many clock cycles have passed and ran the instructions step by step.  This is how long they are taking to execute (shown to the left of the instruction):

for(;;)
	   {
	32	   GpioDataRegs.GPBSET.bit.GPIO32 = 1;
	16	   GpioDataRegs.GPBCLEAR.bit.GPIO32 = 1;
	16	   GpioDataRegs.GPBSET.bit.GPIO32 = 1;
	16	   GpioDataRegs.GPBCLEAR.bit.GPIO32 = 1;
	1	   asm (" NOT AH");
	16	   asm (" NOT AH");
	1	   asm (" NOT AH");
	16	   asm (" NOT AH");
	1	   asm (" NOT AH");
	16	   asm (" NOT AH");
	1	   asm (" NOT AH");
	16	   asm (" NOT AH");
	1	   asm (" NOT AH");
	16	   asm (" NOT AH");
	16	   GpioDataRegs.GPBSET.bit.GPIO32 = 1;
	16	   GpioDataRegs.GPBCLEAR.bit.GPIO32 = 1;
	   }

Why do those assembly instructions alternate between 1 & 16?

Alejandro,

Is this executing from Flash or RAM?  The code on Flash has wait states.

You should look at the disassembly.  You will see that there is not always a one-to-one mapping between C and assembly instructions.  Although that does not explain the clock cycles in the 'asm'  lines, it would explain why the first GPIO line is slower, because it includes the branch instruction for the loop.

I usually time algorithms with an oscilloscope, because it is usually easier to interpret the results.  Have you tried the function, DSP28x_usDelay?  It has a tight calibrated loop.  If you have code like:

#define CPU_CLOCK_HZ     150000000UL

#define wait_usec(us) DSP28x_usDelay((us)*(CPU_CLOCK_HZ/1000000UL/5)-2)

...

for (;;) {

    GpioDataRegs.GPBTOGGLE.bit.GPIO32 = 1;

    wait_usec(1000);

}

You should see a quite accurate toggle at 1ms.

Regards,

Bill

This is executing from Flash.

I tried running this code however it immediately takes me to:
interrupt void ILLEGAL_ISR(void)   // Illegal operation TRAP

Is that bit of code clean?  Also I still do not understand what in the world is going on in those 'ASM 'lines. Anyone?

Thanks.

I tried running from RAM and sure enough the cycle times make more sense, I am seeing those ASM instructions executing in <80nS.  So this leads me to the question, if I need this faster execution time for my project, I cannot program it each time and run it from the debugger,  would I need to get an external memory or something and load the program onto the uC on startup?  Is there another mechanism in the uC which I can utilize?

Thanks,

-Alex