float and binary write to file

[dereach]

Hi
I'd like to write files with binary content (.vtk)

I need to put to file some data-points.
I know that inside file should I get (I know because I compared with some other correct written .vtk files):

3F 80 00 00 40 00 00 00

what means points = 1 , 2

but when I write myself :

          float zma;	
	  zma = 1;
			
	 fwrite(&zma, sizeof zma, 1, file_with_points);

i got: >> 00 00 80 3F 00 00 00 40

Thats big and Little Endian Byte Order problem.

So I wrote function to change byte order.

signed int LE_2_BI(signed int num)
{
	signed int wynik = 0x00000000, maska1= 0xFF, maska2= 0xFF00, 
        maska3= 0xFF0000, maska4= 0xFF000000, 
        liczba1, liczba2, liczba3, liczba4;
	
				liczba1 = num & maska1;
				liczba1 = liczba1 << 24;

				liczba2 = num & maska2;
				liczba2 = liczba2 << 8;
		
		 		liczba3 = num & maska3;
				liczba3 = liczba3 >> 8;
		
				liczba4 = num & maska4;
				liczba4 = liczba4 >> 24;
		
				wynik = liczba1 | liczba2 | liczba3 | liczba4;
		
				printf ( "num=%08x\n wynik=%08x\n", num, wynik);
	return wynik;
};

use function

float data1;
float zma;	

zma = 1;

data1 = LE_2_BI(*(int *)&zma);
fwrite(&data1, sizeof data1, 1, file_with_points);

result 00 3F 00 47 :-(

Why ??

How can I write correct byte order into the file?

Daniel

[Dino]

Here's some code I use to convert Big Endian ints to Little Endian ints. Perhaps you can use this to spark an idea for a new approach.

union  myunion {  // "myunion" is the type name. 
	char data[8] ; 
	int  myint ; 
} ; 


int fix_int(char * n, int len ) { 
	unsigned char temp ; 
	int i , j ; 
	myunion mydata ; 

	memcpy(mydata.data, n, len) ; // Copy data into my union 
	
	for (i = 0 , j = len-1 ; i < j ; ++i, --j ) {  // works for any length value 
	
		// To get little endian frm Big endian, swap this: 0X01020304 into this: 0X04030201 
	
		temp = mydata.data[i] ; 
		mydata.data[i] = mydata.data[j] ; 
		mydata.data[j] = temp ; 
	} 
		
	return mydata.myint ;  
}

[matsp]

Does your "wynik" output show you the correct value?

--
Mats

[dereach]

Yes.
num=3f800000
wynik=0000803f

[matsp]

I've got it:

data1 = LE_2_BI(*(int *)&zma);

fwrite(&data1, sizeof data1, 1, file_with_points);

This assigns a float value from an integer value.
You will need to store the int in a temporary value, then use a dereference-cast-to-float-pointer-address-of. Like you do to make the float an integer, but with float instead of int.

--
Mats

[Dave_Sinkula]

Another possibility:

void bin_rev_write(const void *value, size_t size, FILE *file)
{
   const unsigned char *byte = value;
   for ( byte += size - 1; size--; --byte )
   {
      fwrite(byte, sizeof *byte, 1, file);
   }
}

void foo(FILE *file)
{
   float a = 1, b = 2;
   bin_rev_write(&a, sizeof a, file);
   bin_rev_write(&b, sizeof b, file);
}

[matsp]

Dave,

That would indeed work. However, calling fread/fwrite many times to load and store a single float is not good for performance, compared to a few shuffle operations needed to swap byte order. [Of course, taking addresses and casting the resulting pointer is slightly outside the spec, but very few environments will have a problem with it].

--
Mats

[dereach]

Thanks for help.
However I don't understand how to use Your tip.
Piece of code would help how to convert int to float and use "dereference-cast-to-float-pointer-address-of".

Why I need to store temporary value ??

>>You will need to store the int in a temporary value, then use a dereference-cast-to-float-pointer-address-of. Like you do to make the float an integer, but with float instead of int.

[swoopy]

>Why I need to store temporary value ??
You don't really. Just be sure to declare data1 as an int (or better an unsigned int):

int data1;
float zma;	

zma = 1;

data1 = LE_2_BI(*(int *)&zma);
fwrite(&data1, sizeof data1, 1, file_with_points);

[matsp]

If you don't NEED the value as a float at all, then no, you don't need to store a temporary value. If you NEED the converted value to be a float, then you need to have an integer value at some point, then convert that by a method that makes sure that the compiler doesn't realize that the value is actually an integer, which means you have to store it in a location you can take the address of.

--
Mats