Convert an array of 4 chars to float32

Hi  Danny,

If you just want to append the array elements into a single float32, then cant you use Bit-shifting operator (<< 8, << 16) and OR them?

Regards,

Sid

Hi Sid,

Thanks for the hint, I tried it with no success. 

After wasting 6 hours I found a hint in the TMS320C28x Optimizing C/C++ Compiler v5.0.0 User's Guide 

about using      __byte()  and    __mov_byte() intrinsics and this actually solved my problem.

The following code does the trick :)  

char fArray[4];

float32 f1 = 3.141592;

float32 f2 = 0;

// From a float variable (f1) to a chars array (fArray):

fArray[0] =  __mov_byte((int *)&f1,0);

fArray[1] =  __mov_byte((int *)&f1,1);

fArray[2] =  __mov_byte((int *)&f1,2);

fArray[3] =  __mov_byte((int *)&f1,3);

// From a chars array (fArray) to a float variable (f2):

__byte((int *)&f2,0) = fArray[0];

__byte((int *)&f2,1) = fArray[1];

__byte((int *)&f2,2) = fArray[2];

__byte((int *)&f2,3) = fArray[3];

Thanks,

Danny

The relevant detail is that char is a 16-bit type on C2000.  See section 6.3, Data Types, in spru514c.

The easiest way I've found is to create a union between a bit-field breakout in an int32 struct and a float.  For example:

typedef struct {

int32 byte0: 8;

int32 byte1: 8;

int32 byte2: 8;

int32 byte3: 8;

} CVT_BYTES;

typedef union {

float fp32;

CVT_BYTES bytes;

} CVT_32BIT;

then in your function...

CVT_32BIT chartofloat;

char bvals[4];

float fpval;

bvals[0] = ...

bvals[1] = ...

bvals[2] = ...

bvals[3] = ...

chartofloat.bytes.byte0 = (int32) bvals[0];

chartofloat.bytes.byte1 = (int32) bvals[1];

chartofloat.bytes.byte2 = (int32) bvals[2];

chartofloat.bytes.byte3 = (int32) bvals[3];

fpval = chartofloat.fp32;

Keep in mind that you need to account for little endianness (low-word | high-word in memory), too.  Doing a simple union with chars won't work properly because, as noted above, a char occupies 16 bits since the minimum addressable unit is 16 bits on a C2000.  Generally speaking, I also add a 16-bit (CVT_WORDS) struct type as well as an int32 and uint32 types in order to "cast" between any format.

Mark