CCS/TMS570LS3137: SDSCM00038278 - ARM floating point conversion routines do not support NaN

Hi George,

Thank you so much for your response.

We tried to reproduce the issue as per your inputs (by float to double conversion), however we failed to do so. Could you please let us know how to simulate the problem (provide some code snippet).

Pseudo code 1:

double y;

y = 0.0f / 0.0

In debug window:

 y = NaN

Memory browser:

0x08001330   7FF80000

Pseudo code 2:

float x;

double y;

x = 0.0f / 0.0f

y  = (double) x;

In debug window:

 x = NaN

 y = NaN

Memory browser:

(x) 0x08001328   7FC00000

(y) 0x08001330   7FF80000

Note: Note that different numeric memory representation for NaN for x and y;

We have chosen Compiler version as 5.1.6 in the project plan , hence we can't change the compiler version intermediately. As per your inputs, the issue would be applicable for this compiler version (5.1.6),  we would like to incorporate only updated CRTL library function instead of adapting different compiler version. could you please let us know where we can get updated CRTL library function.

Thanks,

Subash

0/0 is undefined and the compiler can return anything it wants, either NaN or infinity. You are free to check for 0/0 and supply whatever answer you think appropriate.

You are describing CODEGEN-831, also known as SDSCM00049912.  Here is a test case demonstrating the incorrect behavior:

/* armcl --abi=eabi CQ49912.c -z -llnk.cmd */
#include <stdio.h>
#include <math.h>
#include <assert.h>

int main()
{
    volatile float y = NAN;

    assert(isnan((double)y));
}

Still I unable to reproduce the issue, 

I have tested following scenarios, nevertheless both yield same results, in other words in both scenarios program control jumps to next instruction which is  x= x + 1;

/* Scenario -1 */

/* armcl --abi=eabi CQ49912.c -z -llnk.cmd */
#include <stdio.h>
#include <math.h>
#include <assert.h>

int main()
{
    volatile float y = NAN;

    int x ; 

    assert(isnan((double)y));

   x = x + 1;
}

/* Scenario -2 */

/* armcl --abi=eabi CQ49912.c -z -llnk.cmd */
#include <stdio.h>
#include <math.h>
#include <assert.h>

int main()
{
    volatile double y = NAN;

    int x ; 

    assert(isnan(y));

   x = x + 1;
}

PS: we would want to incorporate updated compiler CRTL  function in our old compiler version which is 5.1.6. Could you please provide us latest compiler CRTL library.