Cosine function giving unexpected result

[NicRoberts]

NUCLEO F411RE, CubeIDE

As a learning exercise I'm trying to generate 100Hz cosine wave & then eventually do a DFT on it but I'm getting some odd results from the cos function using code below...

#include "arm_math.h"

#define NUM_SAMP 100
#define SIGNAL_FREQ 100.0
#define SAMPLING_FREQ 1000

static void fpu_enable(void);

typedef struct
{
  float real;
  float imag;
}complex_number;

complex_number samples[NUM_SAMP];

main()
{
  fpu_enable();
  /* Generate a test signal 100Hz */
  for(int n = 0; n < NUM_SAMP; n++ )
   {
     samples[n].real = cos(2 * PI* SIGNAL_FREQ * n / SAMPLING_FREQ);
     samples[n].imag = 0.0;
   }
}

static void fpu_enable(void)
{
  SCB->CPACR |= ((3UL << 10 * 2) | (3UL << 11 * 2));
}
 

When I check the values in samples in the debugger....

I was expecting a value of 1 for the first entry and then subsequent values in the interval [-1, 1]

Is this a formatting issue in Cube's debugger or am I missing something obvious here?

[mfgkw]

arm_math.h is for fixed point values or 32 bit floating point functions arm_sin_cos_q31() and arm_sin_cos_f32().

cos() is a double precision function from math.h.

(1) Do you really want to compute in double precision?

(2) Did you include math.h?

P.S.: The floating point unit on your STM32F411RET6 does not support double precision, only single precision (32 bit values).

[bmckenney]

It's a formatting error. 1065353216 is 0x3F800000 which is =1 (single-precision). The debugger should figure this out but sometimes it doesn't.

Right-click on the number and choose "Number Format".

[NicRoberts]

It was doing the same with cosf  which made me think it was a formatting thing

[NicRoberts]

Thanks that must be it as the rest of the code is working as expected.

Unfortunately the debugger doesn't give me that option when I right click the number, only the option to remove the variable.

[Ozone]

The debugger "knows" the variables are of type float as the screenshot proves, but nonetheless fails to display the value properly.
An issue with the CubeIDE debugger, obviously.

[NicRoberts]

Thanks, have now added bug-report tag.

[bmckenney]

> Unfortunately the debugger doesn't give me that option when I right click the number, only the option to remove the variable.

That's odd, I don't think I've encountered that, and I (have to) do this fairly often.

Are you sure your cursor was over the value (display)? You can only set "Number Format" on a particular element, not a structure nor even an array of scalars. (You can group-select elements in an array of scalars with shift-click.)

[NicRoberts]

No matter what I try I dont get the option.

[Ozone]

Perhaps the debugger has a problem with struct members.
Could you try with a "freestanding" float variable ?