TMS320F280049C: Calculation Time of Arithmetic in the CLA

Hi experts,

We would like to reduce the calculation time of arithmetic in the CLA and we found the floating data type is faster than the integer from the data sheet. 

But we found an interesting thing which is shown below (we use the oscilloscope to measure the elapsed time):

It is reasonable that scenario 2 is faster than scenario 1 and scenario 4 is faster than scenario 3. But we would like to know why scenario 3 takes much more time than scenario 1?

If it is caused by the different variables, scenarios 4 and 5 should also take much more time than scenario 2.

kindly help clarify our doubts.

Scenario 1: Data types of a and b are unsigned integers, it takes 1.07us.

uint16_t a, b; // a is a random variable

GpioDataRegs.GPBSET.bit.GPIO34 = 1; // pull high
b = a * (a + 1) - (a * a - 1);
b += a * (a + 2) - (a * a - 2);
b += a * (a + 3) - (a * a - 3);
b += a * (a + 4) - (a * a - 4);
b += a * (a + 5) - (a * a - 5);
b += a * (a + 6) - (a * a - 6);

GpioDataRegs.GPBCLEAR.bit.GPIO34 = 1; // pull low

Scenario 2: Data types of a and b are floating, it takes 0.42us.

float a, b; // a is a random variable

GpioDataRegs.GPBSET.bit.GPIO34 = 1; // pull high
b = a * (a + (float)1) - (a * a - (float)1);
b += a * (a + (float)2) - (a * a - (float)2);
b += a * (a + (float)3) - (a * a - (float)3);
b += a * (a + (float)4) - (a * a - (float)4);
b += a * (a + (float)5) - (a * a - (float)5);
b += a * (a + (float)6) - (a * a - (float)6);

GpioDataRegs.GPBCLEAR.bit.GPIO34 = 1; // pull low

Scenario 3: Data types of (a, c, d, e, f, g) and b are unsigned integers, it takes 34us.

uint16_t a, b, c, d, e, f, g; // a, c, d, e, f, g is a random variable

GpioDataRegs.GPBSET.bit.GPIO34 = 1; // pull high
b = a * (g + 1) - (e * f - 1);
b += g * (a + 2) - (f * e - 2);
b += c * (d + 3) - (g * d - 3);
b += d * (e + 4) - (c * c - 4);
b += e * (f + 5) - (a * g - 5);
b += f * (a + 6) - (d * a - 6);

GpioDataRegs.GPBCLEAR.bit.GPIO34 = 1; // pull low

Scenario 4: Data types of  (a, c, d, e, f, g) and b are floating, it takes 0.6us.

float a, b, c, d, e, f, g; // a, c, d, e, f, g is a random variable

GpioDataRegs.GPBSET.bit.GPIO34 = 1; // pull high
b = a * (g + (float)1) - (e * f - (float)1);
b += g * (a + (float)2) - (f * e - (float)2);
b += c * (d + (float)3) - (g * d - (float)3);
b += d * (e + (float)4) - (c * c - (float)4);
b += e * (f + (float)5) - (a * g - (float)5);
b += f * (a + (float)6) - (d * a - (float)6);

GpioDataRegs.GPBCLEAR.bit.GPIO34 = 1; // pull low

Scenario 5: Data types of  (a, c, d, e, f, g) and b are unsigned integers and floating, respectively, it takes 0.86us.

uint16_t a, c, d, e, f, g; // a, c, d, e, f, g is a random variable

float b;

GpioDataRegs.GPBSET.bit.GPIO34 = 1; // pull high
b = (float)a * ((float)g + (float)1) - ((float)e * (float)f - (float)1);
b += (float)g * ((float)a + (float)2) - ((float)f * (float)e - (float)2);
b += (float)c * ((float)d + (float)3) - ((float)g * (float)d - (float)3);
b += (float)d * ((float)e + (float)4) - ((float)c * (float)c - (float)4);
b += (float)e * ((float)f + (float)5) - ((float)a * (float)g - (float)5);
b += (float)f * ((float)a + (float)6) - ((float)d * (float)a - (float)6);

GpioDataRegs.GPBCLEAR.bit.GPIO34 = 1; // pull low

Best Regards,

C.C, Liu