Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
16 years ago

How to write floating-point value to custom logic???

Hi all,

I worte a component which is using altfp_sqrt of megafunction, then I checked the simulation with avalon interface signal, it's match the timing and need 17 cycles. All configurations are set in sopc. BUT in nios, when I wrote the floating-point data using IOWR, the sqrt component did not compute. My code is under below

# include <stdio.h># include <float.h># include <unistd.h># include "system.h"

int main()

{

float data_in = 2.0;

float ans;

IOWR(SQRT_BASE, 0, data_in); //regnum = 0 is data_in

usleep(1);//because 17 cycle delay

ans = IORD(SQRT_BASE, 1); //regnum = 1 is calculate result

printf("%f\n", ans);

return 0;

}

The value will show 0.000000

If I change the data format such like this:

unsigned in data_in = 0x40000000; //IEEE-754 format, equivalence 2.0

unsigned int ans;

So, I can get the correct value, but the data type is NOT floating-point

If I want to use custom logic to calculate floating-point in NIOS, how to solve this problem???

thanks all

3 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    Hello,

    You can use a function to copy hexadecimal values using the address like that : 

    
unsigned long F32ToHex (float valueF)
{
    unsigned long* valueL = (unsigned long*) &valueF;
    return *valueL;
}
    And you call it as following : 

    
float value = 678.482;
unsigned long hexValue;
hexValue = F32ToHex(value);
IOWR(SQRT_BASE, 0, hexValue);
    Jérôme
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    Except that such code breaks the 'strict aliasing' rules of C, and gcc might optimise away the assignment.

    Better would be a cast on the LHS, eg:

    *(volatile float *)SQRT_BASE = data_in;

    An alternative is to define a C structure, and variable of a type that matches the IO device, and then get the linker to assign a fixed address (or use a pointer). eg: 

    
extern struct {
    volatile float  sqrt_in;
    volatile float  sqrt_out;
} *sqrt_device = (void *)(1u << 31 | SQRT_BASE);
    ...
    sqrt_device->sqrt_in = data_in;
    delay();
    printf("%f\n", sqrt_device->sqrt_out);

    IMHO using structures to access hardware devices is much less error prone than trying to use constants and access macros.

    Although not 100% portable, it is portable enough.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    THANKS LETS and dsl!!!

    My problem SOLVED TOTALLY!!!

    The function work out fine!!!

    Thank you very much!!!