big execution time

[Ahmed Zahran]

Posted on July 23, 2017 at 10:57

Hi!

when I tried to run the code below at an interrupt it took 123u sec! (I measured that by toggle PGIOB7) which is too big for me . any one can advice?

I am using STM32F429 running at 70MHz (I can't increase the clk frequency for some hardware limitations)

HAL_GPIO_WritePin (GPIOB, GPIO_PIN_7,GPIO_PIN_SET);

//HAL_GPIO_TogglePin (GPIOB, GPIO_PIN_7);

D101Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*CURRENT_SENSOR_SCALE;

D102Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*CURRENT_SENSOR_SCALE; D103Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*CURRENT_SENSOR_SCALE; D104Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*VOLTAGE_SENSOR_SCALE; D105Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*TEMP_SENSOR_SCALE1; D106Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*TEMP_SENSOR_SCALE2; D107Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*PRESSURE_SENSOR_SCALE; HAL_GPIO_WritePin (GPIOB, GPIO_PIN_7,GPIO_PIN_RESET);

I made another trial by replacing 

ADC1_ConvertedValue[0] with constant value (eg 50) to be sure if the problem from the calculation time or what. I wondered from the result where execution of (

D101Value=((50*3.3/ADC1_Full_Scale)-0)*CURRENT_SENSOR_SCALE;) took only 0.25u sec instead of 20u sec previously !!! so the long time should be spend in calling variable  

ADC1_ConvertedValue[0]

  from the memory and really I don't have explanation for that

[Vangelis Fortounas]

Posted on July 23, 2017 at 11:15

hello!

D101Value=((ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)-0)*CURRENT_SENSOR_SCALE;

is equal to D101Value=(ADC1_ConvertedValue[0]*3.3/ADC1_Full_Scale)*CURRENT_SENSOR_SCALE;

is equal to D101Value=ADC1_ConvertedValue[0] * CURRENT_SENSOR_SCALE*3.3/ADC1_Full_Scale;

CURRENT_SENSOR_SCALE*3.3/ADC1_Full_Scale, is a constant.  Use it as a constant K

So D101Value=ADC1_ConvertedValue[0] *K;      (k is calculated outside the loop)

Other methods

1. Don't  make the conversions  inside the interrupt .

2. Make the conversions outside the loop

3.Don't use Floating Point variables (especialy when your FP HW is disabled).

[Ahmed Zahran]

Posted on July 23, 2017 at 11:38

thanks a lot for your quick replay

'0' in the equation will be replaced by offset value later so I can't remove it. Also I need to make the conversion inside the interrupt for another purposes related to the code functionality.

I made another trial by replacing 

ADC1_ConvertedValue[0] with constant value (eg 50) to be sure if the problem from the calculation time or what. I wondered from the result where execution of (

D101Value=((50*3.3/ADC1_Full_Scale)-0)*CURRENT_SENSOR_SCALE;) took only 0.25u sec instead of 20u sec previously !!! so the long time should be spend in calling variable  

ADC1_ConvertedValue[0]

  from the memory and really I don't have explanation for that

[Tesla DeLorean]

Posted on July 23, 2017 at 15:52

So look at the generated code assembler.

Constants might be folded by the compiler.

The math may also be folded by repetitive use of 

ADC1_ConvertedValue[0]

 instead of values unique to each ADC sample.

Consider using integer math.

Avoid division, multiply by a scaling reciprocal.