GIGA R1 Wifi : Fast digital output handling

Projects Programming
41

Tests on M7 and M4 core.

M7 is the master core at 480MHz.
M4 is the secondary one, at 240MHz.

#ifdef CORE_CM7
#define _write(...)   { Serial.write(__VA_ARGS__); } 
#define _print(...)   { Serial.print(__VA_ARGS__); } 
#define _println(...) { Serial.println(__VA_ARGS__); }

#else

#define _print(...)    { RPC.print(__VA_ARGS__); }
#define _println(...)  { RPC.print(__VA_ARGS__); RPC.print("\n"); }

#endif

int           ledRGB[]={D86,D87,D88};
DigitalOutput led[3]; // Uses the class in post #1


void test()
{
 int i,t1,t2;

// LED control pins as output
 pinMode(ledRGB[0], OUTPUT); 
 pinMode(ledRGB[1], OUTPUT);
 pinMode(ledRGB[2], OUTPUT);

// LED control pins seen through class DigitalOutput
// as inverted pins
 led[0].init(GPIOI,12,true);
 led[1].init(GPIOJ,13,true);
 led[2].init(GPIOE, 3,true);

// test 1 : usage of standard digitalWrite()
// ------

 t1 = millis();
 for(i=0;i<1000000;i++)
 {
   digitalWrite(ledRGB[0], LOW);
   digitalWrite(ledRGB[0], HIGH);
 }
 t2 = millis();

 _println("delta1 = " + String(t2-t1));

// test 2 : usage of setHigh()/setLow() 
// ------

  t1 = millis();
  for(i=0;i<1000000;i++)
  {
    led[0].setHigh();
    led[0].setLow();
  }
  t2 = millis();

  _println("delta2 = " + String(t2-t1));

// Test for fun  (same price !) : 100000 square roots
 double f = 100000000000000.0;

 t1 = millis();
 for(i=0;i<100000;i++)
 {
    f = sqrt(f+0.1);
 }
 t2 = millis();

 _println("delta sqrt = " + String(t2-t1));
}

Results (in ms) :

M7 :

delta1 = 259
delta2 = 51
delta sqrt = 19


M4 :

delta1 = 645
delta2 = 23
delta sqrt = 505

The results are quite unexpected :

  • M7 is twice faster then M4 in clock speed, but twice slower in test 2 !?!

  • 100000 square roots take 19ms on M7 but 505ms on M4 (25x slower).

Conclusions :

  • M4 with class DigitalOutputs leaves the M7 far behind !
  • M7 is for calculations, seeming to be the only one with a FPU.